Protein Protein Interactions Methods and

Methods in
Molecular Biology 1278
Cheryl L. Meyerkord
Haian Fu
Editors
Protein-Protein
Interactions
Methods and Applications
Second Edition
METHODS IN MOLECULAR BIOLOGY
Series Editor
John M. Walker
School of Life and Medical Sciences
University of Hertfordshire
Hatfield, Hertfordshire, AL10 9AB, UK
For further volumes:

http://www.springer.com/series/7651
Protein-Protein Interactions
Methods and Applications
Second Edition
Edited by
Cheryl L. Meyerkord
Department of Pharmacology and Emory Chemical Biology Discovery Center,
Emory University School of Medicine, Atlanta, GA, USA
Haian Fu
Department of Pharmacology, Department of Hematology & Medical
Oncology, and Emory Chemical Biology Discovery Center, Emory University
School of Medicine, Atlanta, GA, USA
Editors
Cheryl L. Meyerkord Haian Fu
Department of Pharmacology Department of Pharmacology
and Emory Chemical Biology Department of Hematology & Medical
Discovery Center, Emory University Oncology, and Emory Chemical Biology
School of Medicine Discovery Center, Emory University
Atlanta, GA, USA School of Medicine
Atlanta, GA, USA
ISSN 1064-3745 ISSN 1940-6029 (electronic)

Methods in Molecular Biology
ISBN 978-1-4939-2424-0 ISBN 978-1-4939-2425-7 (eBook)
DOI 10.1007/978-1-4939-2425-7
Library of Congress Control Number: 2015933374
Springer New York Heidelberg Dordrecht London

# Springer Science+Business Media New York 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction
on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation,
computer software, or by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to
be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty,
express or implied, with respect to the material contained herein or for any errors or omissions that may have been
made.
Printed on acid-free paper
Humana Press is a brand of Springer

Springer Science+Business Media LLC New York is part of Springer Science+Business Media (www.springer.com)
Preface
From regulation of DNA replication, RNA transcription, and protein translation to

posttranslational modifications and signal transduction, protein-protein interactions
are critical for most biological functions. In recent years, our understanding of the
protein-protein interaction landscape has significantly increased due to advances in geno-
mics, proteomics, and functional and network biology. It is also recognized that dysregu-
lated protein-protein interactions or aggregations form the molecular basis for a large
number of human diseases ranging from cancer to neurodegenerative disorders. Thus,
protein-protein interactions may represent a promising class of targets for therapeutic
intervention. The emerging prominence of therapeutic discovery has ignited a rising
interest in the analysis of protein-protein interactions, the development of various assays
for monitoring protein-protein interactions, and the identification of small molecule pro-
tein-protein interaction modulators. Targeting protein-protein interactions can be chal-
lenging and demands innovative technological platforms. Such techniques promise to
accelerate future advances in our understanding of protein-protein interactions, will help
to better define the role of these interactions in complex biological systems, and will aid in
the identification of small molecule modulators for targeting protein-protein interactions.
The growing interest in interrogating protein-protein interactions for mechanistic
studies and potential therapeutic discovery, coupled with new technological advances,
prompted us to update and expand the scope of Protein-Protein Interactions: Methods
and Applications. In this new edition, we have retained the core technological platforms
that are classical cornerstones used to study protein-protein interactions and added new
chapters for cutting-edge technologies that reflect recent scientific advances and the
emerging focus on therapeutic discovery with miniaturized protein-protein interaction
detection platforms. The second edition of Protein-Protein Interactions: Methods and
Applications covers a wide range of protein-protein interaction detection topics that are
separated into five parts. The book begins with overview chapters that describe the
fundamental principles of protein-protein interactions, available protein-protein interac-
tion databases, methods for quantitative and computational analysis of protein-protein
interactions, and targeting protein-protein interactions for small molecule modulator
identification and drug discovery. Part II details methodologies involving label-free bio-
sensors and techniques, which allow researchers to analyze protein-protein interactions
without the requirement for traditional radioisotope or fluorophore labeling of test
proteins. Part III includes tag- and antibody-based methods that offer the benefit of
specific and sensitive detection of a target protein and its interacting partners. Part IV is
a compilation of cell-based biomolecular interaction reporter assays, which provide a
physiologically relevant context and in some cases allow for the analysis of protein-protein
interactions in live cells. The final section consists of case studies describing protein-protein
interaction analysis platforms that are applicable in a miniaturized format for high-
throughput screening. This section provides examples of methods that have been effec-
tively used to analyze the modulation of protein-protein interactions by various small
molecules as well as examples of biophysical methods that can be used for identification
of fragment-based inhibitors of protein-protein interactions.
v
vi Preface
Our hope is that this book will serve as a valuable resource for all readers, from students
to experienced scientists. In the chapters of this book, experienced researchers describe in
detail the underlying theory and practical application of widely used methods for monitor-
ing protein-protein interactions. In keeping with the series tradition, the Notes section of
the chapters contains valuable explanations to sensitive procedures and potential pitfalls,
with tips on how to avoid them. This book is expected to empower readers in their quest to
elucidate the mechanisms of protein-protein interactions and the role of these interactions
in diverse biological processes, and to target protein-protein interactions for therapeutic
discovery.
We are grateful to the contributors of each chapter for their effort, commitment, and
willingness to share their knowledge and valuable experience with the scientific commu-
nity. We sincerely thank the series editor, John Walker, for all of his guidance. Together, we
present an extensively updated and expanded version of this valuable resource to all those
who study protein-protein interactions.
Atlanta, GA, USA Cheryl L. Meyerkord

Atlanta, GA, USA Haian Fu
Contents
Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
PART I OVERVIEWS
1 Structural Basis of Protein-Protein Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Robert C. Liddington
2 Quantitative Analysis of Protein-Protein Interactions . . . . . . . . . . . . . . . . . . . . . . 23
Ziad M. Eletr and Keith D. Wilkinson
3 Protein-Protein Interaction Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Damian Szklarczyk and Lars Juhl Jensen
4 Computational Prediction of Protein-Protein Interactions. . . . . . . . . . . . . . . . . . 57
Tobias Ehrenberger, Lewis C. Cantley, and Michael B. Yaffe
5 Structure-Based Computational Approaches for Small-Molecule
Modulation of Protein-Protein Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
David Xu, Bo Wang, and Samy O. Meroueh
6 Targeting Protein-Protein Interactions for Drug Discovery . . . . . . . . . . . . . . . . . 93
David C. Fry
PART II LABEL–FREE BIOSENSORS AND TECHNIQUES
7 Studying Protein-Protein Interactions Using Surface Plasmon Resonance . . . . 109

Zaneta Nikolovska-Coleska
8 Resonant Waveguide Grating for Monitoring Biomolecular Interactions . . . . . 139
Meng Wu and Min Li
9 Quartz Microbalance Technology for Probing Biomolecular Interactions . . . . 153
Gabriella T. Heller, Alison R. Mercer-Smith, and Malkiat S. Johal
10 Label-Free Kinetic Analysis of an Antibody–Antigen Interaction
Using Biolayer Interferometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Sriram Kumaraswamy and Renee Tobias
11 Characterization of Protein-Protein Interactions by Isothermal
Titration Calorimetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Adrian Velazquez-Campoy, Stephanie A. Leavitt, and Ernesto Freire
12 Sedimentation Equilibrium Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Ian A. Taylor, Katrin Rittinger, and John F. Eccleston
13 Detecting Protein-Protein Interactions by Gel Filtration Chromatography . . . 223
Yan Bai
14 Using Light Scattering to Determine the Stoichiometry
of Protein Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Jeremy Mogridge
15 Circular Dichroism (CD) Analyses of Protein-Protein Interactions . . . . . . . . . . 239
Norma J. Greenfield
vii
viii Contents
16 Protein-Protein Interaction Analysis by Nuclear Magnetic

Resonance Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Peter M. Thompson, Moriah R. Beck, and Sharon L. Campbell
17 Quantitative Protein Analysis by Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . 281
Vishwajeeth R. Pagala, Anthony A. High, Xusheng Wang,
Haiyan Tan, Kiran Kodali, Ashutosh Mishra, Kanisha Kavdia,
Yanji Xu, Zhiping Wu, and Junmin Peng
18 Using Peptide Arrays Created by the SPOT Method
for Defining Protein-Protein Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Yun Young Yim, Katherine Betke, and Heidi Hamm
PART III TAG/AFFINITY–BASED METHODS

19 Fluorescence Polarization Assay to Quantify Protein-Protein
Interactions: An Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
Ronald T. Raines
20 Förster Resonance Energy Transfer (FRET) Microscopy
for Monitoring Biomolecular Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
Alexa L. Mattheyses and Adam I. Marcus
21 Utilizing ELISA to Monitor Protein-Protein Interaction . . . . . . . . . . . . . . . . . . . 341
Zusen Weng and Qinjian Zhao
22 Glutathione-S-Transferase (GST)-Fusion Based Assays
for Studying Protein-Protein Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
Haris G. Vikis and Kun-Liang Guan
23 Hexahistidine (6xHis) Fusion-Based Assays for Protein-Protein
Interactions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Mary C. Puckett
24 Studying Protein-Protein Interactions via Blot Overlay/Far
Western Blot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
Randy A. Hall
25 Co-immunoprecipitation from Transfected Cells . . . . . . . . . . . . . . . . . . . . . . . . . . 381
Yoshinori Takahashi
26 In Vivo Protein Cross-Linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Fabrice Agou and Michel Véron
PART IV CELL–BASED BIMOLECULAR INTERACTION REPORTER ASSAYS
27 Identification of Protein-Protein Interactions by Standard

Gal4p-Based Yeast Two-Hybrid Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
Jeroen Wagemans and Rob Lavigne
28 Reverse Two-Hybrid Techniques in the Yeast Saccharomyces cerevisiae . . . . . . . 433
Matthew A. Bennett, Jack F. Shern, and Richard A. Kahn
29 MAPPIT, a Mammalian Two-Hybrid Method for In-Cell Detection
of Protein-Protein Interactions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
Irma Lemmens, Sam Lievens, and Jan Tavernier
Contents ix
30 Bioluminescence Resonance Energy Transfer to Detect

Protein-Protein Interactions in Live Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
Nicole E. Brown, Joe B. Blumer, and John R. Hepler
31 Mapping Biochemical Networks with Protein Fragment
Complementation Assays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
Ingrid Remy and Stephen W. Michnick
32 Detection of Protein-Protein Interaction Using Bimolecular
Fluorescence Complementation Assay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
Cau D. Pham
33 Split-Luciferase Complementation Assay to Detect Channel–Protein
Interactions in Live Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
Alexander S. Shavkunov, Syed R. Ali, Neli I. Panova-Elektronova,
and Fernanda Laezza
34 Confocal Microscopy for Intracellular Co-localization of Proteins . . . . . . . . . . . 515
Toshiyuki Miyashita
PART V HIGH THROUGHPUT SCREENING ASSAYS FOR PROTEIN-PROTEIN

INTERACTIONS: CASE STUDIES
35 Fluorescence Polarization Assay to Quantify Protein-Protein

Interactions in an HTS Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
Yuhong Du
36 Estrogen Receptor Alpha/Co-activator Interaction Assay: TR-FRET . . . . . . . . 545
Terry W. Moore, Jillian R. Gunther, and John A. Katzenellenbogen
37 High Content Screening Biosensor Assay to Identify Disruptors
of p53–hDM2 Protein-Protein Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555
Yun Hua, Christopher J. Strock, and Paul A. Johnston
38 Case Study: Discovery of Inhibitors of the MDM2–p53
Protein-Protein Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
Liu Liu, Denzil Bernard, and Shaomeng Wang
39 Biophysical Methods for Identifying Fragment-Based Inhibitors
of Protein-Protein Interactions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587
Samuel J. Pfaff, Michael S. Chimenti, Mark J.S. Kelly,
and Michelle R. Arkin
Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615
Contributors
FABRICE AGOU Département de Biologie Cellulaire et Infection, Institut Pasteur, Unité de

Signalisation et Pathogenèse, Paris, France
SYED R. ALI Department of Pharmacology and Toxicology, Pharmacology and Toxicology
Graduate Program, The University of Texas Medical Branch, Galveston, TX, USA
MICHELLE R. ARKIN Small Molecule Discovery Center, Department of Pharmaceutical
Chemistry, University of California San Francisco, San Francisco, CA, USA; UCSF, San
Francisco, CA, USA
YAN BAI Department of Pharmacology, Emory University School of Medicine, Atlanta,
GA, USA
MORIAH R. BECK Chemistry Department, Wichita State University, Wichita, KS, USA
MATTHEW A. BENNETT Department of Biochemistry, Emory University School of Medicine,
Atlanta, GA, USA
DENZIL BERNARD Comprehensive Cancer Center and Departments of Internal Medicine,
Pharmacology and Medicinal Chemistry, University of Michigan, Ann Arbor, MI, USA
KATHERINE BETKE Department of Pharmacology, Vanderbilt University Medical Center,
Nashville, TN, USA
JOE B. BLUMER Department of Cell and Molecular Pharmacology and Experimental
Therapeutics, Medical University of South Carolina, Charleston, SC, USA
NICOLE E. BROWN Department of Pharmacology, Emory University School of Medicine,
Atlanta, GA, USA
SHARON L. CAMPBELL Department of Biochemistry and Biophysics, Lineberger Cancer
Center, University of North Carolina, Chapel Hill, NC, USA
LEWIS C. CANTLEY Division of Signal Transduction, Department of Medicine, Beth Israel
Deaconess Medical Center, Harvard Medical School, Boston, MA, USA; Weill Cornell
Cancer Center, Weill Cornell Medical College, New York, NY, USA
MICHAEL S. CHIMENTI Department of Pharmaceutical Chemistry, University of
California San Francisco, San Francisco, CA, USA
YUHONG DU Department of Pharmacology, Emory Chemical Biology Discover Center,
Emory University School of Medicine, Atlanta, GA, USA
JOHN F. ECCLESTON Division of Physical Chemistry, MRC National Institute for Medical
Research, London, UK
TOBIAS EHRENBERGER Department of Biological, Massachusetts Institute of Technology,
Cambridge, MA, USA
ZIAD M. ELETR Department of Biochemistry, Emory University School of Medicine,
Atlanta, GA, USA
ERNESTO FREIERE Department of Biology, Johns Hopkins University, Baltimore, MD, USA
DAVID C. FRY Roche Research Center, Nutley, NJ, USA
HAIAN FU Departments of Pharmacology, Hematology and Medical Oncology,
Emory University School of Medicine, Atlanta, GA, USA; Emory Chemical Biology
Discovery Center, Emory University, Atlanta, GA, USA
NORMA J. GREENFIELD Associate Professor (retired), Department of Neuroscience and Cell
Biology, Robert Wood Johnson Medical School, Rutgers University, Piscataway, NJ, USA
xi
xii Contributors
KUN-LIANG GUAN Department of Pharmacology and Moores Cancer Center, University of

California San Diego, La Jolla, CA, USA
JILLIAN R. GUNTHER Department of Chemistry, University of Illinois at Urbana-
Champaign, Urbana, IL, USA; Division of Radiation Oncology, University of Texas MD
Anderson Cancer Center, Houston, TX, USA
RANDY A. HALL Department of Pharmacology, Emory University School of Medicine,
Atlanta, GA, USA
HEIDI HAMM Department of Pharmacology, Vanderbilt University Medical Center,
Nashville, TN, USA
GABRIELLA T. HELLER Department of Chemistry, Pomona College, Claremont, CA, USA
JOHN R. HEPLER Department of Pharmacology, Emory University School of Medicine,
Atlanta, GA, USA
ANTHONY A. HIGH St. Jude Proteomics Facility, St. Jude Children’s Research Hospital,
Memphis, TN, USA
YUN HUA Department of Pharmaceutical Sciences, School of Pharmacy, University of
Pittsburgh, Pittsburg, PA, USA
LARS JUHL JENSEN Faculty of Health Sciences, Novo Nordisk Foundation Centre for
Protein Research, University of Copenhagen, Copenhagen, Denmark
MALKIAT S. JOHAL Department of Chemistry, Pomona College, Claremont, CA, USA
PAUL A. JOHNSTON Department of Pharmaceutical Sciences, School of Pharmacy,
University of Pittsburgh Cancer Institute, Pittsburg, PA, USA
RICHARD A. KAHN Department of Biochemistry, Emory University School of Medicine,
Atlanta, GA, USA
JOHN A. KATZENELLENBOGEN Department of Chemistry, University of Illinois at Urbana-
Champaign, Urbana, IL, USA
KANISHA KAVDIA St. Jude Proteomics Facility, St. Jude Children’s Research Hospital,
Memphis, TN, USA
MARK J.S. KELLY Department of Pharmaceutical Chemistry, University of California San
Francisco, San Francisco, CA, USA
KIRAN KODALI St. Jude Proteomics Facility, St. Jude Children’s Research Hospital,
Memphis, TN, USA
SRIRAM KUMARASWAMY ForteBio Inc.—A Division of Pall Life Sciences, Menlo Park, CA,
USA
FERNANDA LAEZZA Department of Pharmacology and Toxicology, Center for Addiction
Research, Mitchell Center for Neurodegenerative Diseases, Center for Biomedical
Engineering, The University of Texas Medical Branch, Galveston, TX, USA
ROB LAVIGNE Laboratory of Gene Technology, Katholieke Uniersiteit Leuven, Leuven,
Belgium
STEPHANIE A. LEAVITT Gilead Sciences, Inc., Foster City, CA, USA
IRMA LEMMENS Department of Medical Protein Research, VIB, Ghent, Belgium;
Department of Biochemistry, Faculty of Medicine and Health Sciences, Ghent University,
Ghent, Belgium
MIN LI The Solomon H. Snyder Department of Neuroscience, Johns Hopkins Ion Channel
Center and High Throughput Biology Center, School of Medicine, Johns Hopkins
University, Baltimore, MD, USA; GSK, King of Prussia, PA, USA
ROBERT LIDDINGTON Sanford-Burnham Medical Research Institute, La Jolla, CA, USA
Contributors xiii
SAM LIEVENS Department of Medical Protein Research, VIB, Ghent, Belgium; Department
of Biochemistry, Faculty of Medicine and Health Sciences, Ghent University, Ghent,
Belgium
LIU LIU Comprehensive Cancer Center and Departments of Internal Medicine,
ADAM I. MARCUS Department of Hematology and Medical Oncology, Winship Cancer
Institute of Emory University, Atlanta, GA, USA
ALEXA L. MATTHEYSES Department of Cell Biology, Emory University School of Medicine,
Atlanta, GA, USA
ALISON R. MERCER-SMITH Department of Chemistry, Pomona College, Claremont,
CA, USA
SAMY O. MEROUEH Department of Biochemistry and Molecular Biology, Center for
Computational Biology and Bioinformatics, Indiana University School of Medicine,
Indianapolis, IN, USA; Department of Chemistry and Chemical Biology, Indiana
University Purdue University Indianapolis, Indianapolis, IN, USA
CHERYL L. MEYERKORD Department of Pharmacology, Emory University School
of Medicine, Atlanta, GA, USA; Emory Chemical Biology Discovery Center,
Emory University, Atlanta, GA, USA
STEPHEN W. MICHNICK Département de Biochimie, Université de Montréal, Montréal,
QC, Canada
ASHUTOSH MISHRA St. Jude Proteomics Facility, St. Jude Children’s Research Hospital,
Memphis, TN, USA
TOSHIYUKI MIYASHITA Department of Molecular Genetics, Kitasato University School
of Medicine, Sagamihara, Japan
JEREMY MOGRIDGE Department of Laboratory Medicine and Pathology, University
of Toronto, Toronto, ON, Canada
TERRY W. MOORE Department of Chemistry, University of Illinois at Urbana-
Champaign, Urbana, IL, USA; Department of Medicinal Chemistry and
Pharmacognosy, University of Illinois Hospital and Health Sciences System Cancer Center,
Chicago, IL, USA
ZANETA NIKOLOVSKA-COLESKA Department of Pathology, University of Michigan
Medical School, Ann Arbor, MI, USA
VISHWAJEETH R. PAGALA St. Jude Proteomics Facility, St. Jude Children’s Research
Hospital, Memphis, TN, USA
NELI I. PANOVA-ELEKTRONOVA Department of Pharmacology and Toxicology, The
University of Texas Medical Branch, Galveston, TX, USA
JUNMIN PENG St. Jude Proteomics Facility, Department of Structural Biology,
Department of Neurodevelopmental Biology, St. Jude Children’s Research Hospital,
Memphis, TN, USA
SAMUEL J. PFAFF Small Molecule Discovery Center, Department of Pharmaceutical
Chemistry, University of California San Francisco, San Francisco, CA, USA
CAU D. PHAM Department of Pharmacology, Emory University School of Medicine,
Atlanta, GA, USA
MARY C. PUCKETT Department of Pharmacology, Emory University School of Medicine,
Atlanta, GA, USA
RONALD T. RAINES Department of Biochemistry, University of Wisconsin-Madison,
Madison, WI, USA; Department of Chemistry, University of Wisconsin-Madison,
Madison, WI, USA
xiv Contributors
INGRID REMY Département de Biochimie, Université de Montréal, Montréal, QC, Canada

KATRIN RITTINGER Division of Molecular Structure, MRC National Institute for Medical
ALEXANDER S. SHAVKUNOV Department of Pharmacology and Toxicology, The University
of Texas Medical Branch, Galveston, TX, USA
JACK F. SHERN Department of Biochemistry, Emory University School of Medicine,
Atlanta, GA, USA
CHRISTOPHER J. STROCK Cyprotex US, Watertown, MA, USA
DAMIAN SZKLARCZYK Faculty of Health Sciences, Novo Nordisk Foundation Centre for
Protein Research, University of Copenhagen, Copenhagen, Denmark
YOSHINORI TAKAHASHI Department of Pediatrics, The Pennsylvania State University
College of Medicine, Hershey, PA, USA
HAIYAN TAN St. Jude Proteomics Facility, St. Jude Children’s Research Hospital, Memphis,
TN, USA
JAN TAVERNIER Department of Medical Protein Research, VIB, Ghent, Belgium;
Department of Biochemistry, Faculty of Medicine and Health Sciences, Ghent University,
Ghent, Belgium
IAN A. TAYLOR Division of Molecular Structure, MRC National Institute for Medical
PETER M. THOMPSON Department of Biochemistry and Biophysics, Program in Molecular
and Cellular Biophysics, University of North Carolina, Chapel Hill, NC, USA
RENEE TOBIAS Forte Bio—A Division of Pall Life Sciences, Menlo Park, CA, USA
ADRIAN VELAZQUEZ-CAMPOY Institute of Biocomputation and Physics of Complex Systems
(BIFI), Joint Unit IQFR-CSIC-BIFI, Department of Biochemistry and Molecular and
Cell Biology, Universidad de Zaragoza, Zaragoza, Spain; Fundacion ARAID,
Government of Aragon, Zaragoza, Spain
MICHEL VÉRON Département de Biologie Structurale et Chimie, Institut Pasteur, Unité de
Biochimie Structurale et Cellulaire, Paris, France
HARIS G. VIKIS Department of Pharmacology and Toxicology, MCW Cancer center,
Medical College of Wisconsin, Milwaukee, WI, USA
JEROEN WAGEMANS Laboratory of Gene Technology, Katholieke Uniersiteit Leuven, Leuven,
Belgium
BO WANG Center for Computational Biology and Bioinformatics, Indiana University
School of Medicine, Indianapolis, IN, USA; Department of Chemistry and Chemical
Biology, Indiana University Purdue University Indianapolis, Indianapolis, IN, USA
SHAOMENG WANG Comprehensive Cancer Center and Departments of Internal Medicine,
XUSHENG WANG St. Jude Proteomics Facility, St. Jude Children’s Research Hospital,
Memphis, TN, USA
ZUSEN WENG State Key Laboratory of Molecular Vaccinology and Molecular Diagnostics,
National Institute of Diagnostics and Vaccine Development in Infectious Diseases, School
of Public Health, Xiamen University, Xiamen, P. R. China
KEITH D. WILKINSON Department of Biochemistry, Emory University School of Medicine,
Atlanta, GA, USA
MENG WU High Throughput Screening Facility at Univ. of Iowa (UIHTS), Division
of Medicinal and Natural Products Chemistry, Department of Pharmaceutical Sciences
and Experimental Therapeutics, College of Pharmacy, Department of Biochemistry,
Carver College of Medicine, The University of Iowa, Iowa City, IA, USA
Contributors xv
ZHIPING WU Department of Structural Biology, Department of Neurodevelopmental

Biology, St. Jude Children’s Research Hospital, Memphis, TN, USA
DAVID XU Center for Computational Biology and Bioinformatics, Indiana University
School of Medicine, Department of BioHealth Informatics, Indiana University School of
Informatics, Indianapolis, IN, USA
YANJI XU St. Jude Proteomics Facility, St. Jude Children’s Research Hospital, Memphis,
TN, USA
MICHAEL B. YAFFE Department of Biology, Department of Biological Engineering, Koch
Institute for Integrative Cancer Biology, Massachusetts Institute of Technology,
Cambridge, MA, USA; Department of Surgery, Beth Israel Deaconess Medical Center,
Harvard Medical School, Boston, MA, USA
YUN YOUNG YIM Department of Pharmacology, Vanderbilt University Medical Center,
Nashville, TN, USA
QINJIAN ZHAO State Key Laboratory of Molecular Vaccinology and Molecular Diagnostics,
National Institute of Diagnostics and Vaccine Development in Infectious Diseases, School
of Public Health, Xiamen University, Xiamen, P. R. China
Part I
Overviews
Chapter 1
Structural Basis of Protein-Protein Interactions

Robert C. Liddington
Abstract
Regulated interactions between proteins govern signaling pathways within and between cells. Structural
studies on protein complexes formed reversibly and/or transiently illustrate the remarkable diversity of
interactions, both in terms of interfacial size and nature. In recent years, “domain–peptide” interactions
have gained much greater recognition and may be viewed as both pre-translational and posttranslational-
dependent functional switches. Our understanding of the multistep regulation of auto-inhibited multi-
domain proteins has also grown. Their activity may be understood as the “combinatorial” output of
multiple input signals, including phosphorylation, location, and mechanical force. The prospects for
bridging the gap between the new “systems biology” data and the traditional “reductionist” data are also
discussed.
Key words Structure, Complex, Energetics, Crystallography, Allostery, Regulation, Force,

Domain–peptide
1 Introduction
Proteins live, work, and die in a highly crowded environment and

must find their cognate partner(s) in a vast sea of non-partners. But
all soluble proteins tend to look rather similar: their surfaces are
covered by a mix of hydrophobic and hydrophilic residues, the
latter in greater abundance, as well as numerous backbone atoms
with unsatisfied H-bonding potential. Recent studies have consid-
ered how the remarkable feats of co-localization, recognition, and
specificity are achieved [1, 2].
But first I review what we know and do not know about pro-
ductive protein-protein interactions. The obvious starting point is
the weekly updated library of crystal structures contained within
the Protein Data Bank (PDB) (http://www.rcsb.org/pdb/),
which now number in the tens of thousands. In addition, many
searchable databases of protein complexes are available. For exam-
ple, two well-curated sites reported ~9,000 distinct classes of inter-
faces involving ~5,000 Pfam [3] domains, as of mid-2013 [4].
These numbers may be impressive, but they still represent only a
Cheryl L. Meyerkord and Haian Fu (eds.), Protein-Protein Interactions: Methods and Applications, Methods in Molecular Biology,
vol. 1278, DOI 10.1007/978-1-4939-2425-7_1, © Springer Science+Business Media New York 2015
3
4 Robert C. Liddington
Fig. 1 Atomic structure from single-particle electron microscopy. Structure of the

bacteriophage epsilon15 determined directly from cryo-EM images at a
resolution of 3–5 Å [92]. There are 14 protein chains (3,122 residues)
comprising the asymmetric unit particles (color-coded). The other subunits
(total of 420 asymmetric units) are generated by the T ¼ 7 icosahedral
symmetry. View is down one of the threefold axes of the icosahedral particle.
Subunits on fivefold axes are colored red (labeled white)
small subset of the complete human binary protein-protein interac-

tions (the “interactome”) which has been estimated at ~500,000 [5,
6].
Crystallography remains the major tool for determining the
binary interactome: protein complexes at atomic resolution, with
the upper limit defined by rigidity rather than size: the structure of
the ribosome may be the crowning achievement for a relatively
static, asymmetric, structure [7], while the organization of viral
capsids captures some of the complexity and much of the beauty
of protein-protein interactions. Advances in electron microscopy
have recently enabled atomic models to be built directly into EM
maps, but only in the case of highly symmetric viruses [8] (Fig. 1).
NMR has typically been limited to protein of 30 kDa, but it plays
an important role in defining interacting surfaces, especially at the
membrane (see Chapter 16).
Automated crystallization pipelines developed over the past
15 years have greatly increased the database of known structures,
extending the coverage of protein “fold-space,” as well as pointing
to its boundaries [9]. But our knowledge of protein “interaction-
space” has not kept pace: crystallizing complexes has not lent itself
Structural Basis 5
well to automation, neither has the crystallization of large

multidomain proteins that are typical of eukaryotic signaling path-
ways. The genome-wide studies have given us the first “wiring
diagrams” of the vast protein interaction landscape; but by them-
selves, they tell us little about how a pathway (or cell) works. We
still need to understand, at the atomic, energetic, and kinetic levels,
the how, where, and when of protein interactions.
So what are the prospects for bridging the gap between the
“systems biology” data (comprehensive but lacking mechanism)
and the “reductionist” data (atomistic/mechanistic but lacking in
quantity)? There are reasons to be optimistic. First, the growth in
the number of homologous protein superfamilies in the CATH
database [10] has nearly plateaued in recent years and is not
expected to exceed 3,000 [11], suggesting that protein “fold-
space” is finite and of a manageable size. And there are also indica-
tions that “interaction-space” may be similarly restricted.
How far can we extrapolate from what we do know, i.e., by
building homology models based on known complexes? Some
reports have suggested that suitable “docking templates” (derived
from diverse organisms) can be found for modeling all binary
protein complexes in the human interactome [12–15]. However,
by the authors’ own admission, only ~30 % of these models (rising
to ~50 % for homodimers [11]) are of “good” quality, i.e., partly
correct, partly wrong, but good enough to warrant experimental
verification/refinement. This is not surprising, as earlier studies
showed that when sequence identity with the template falls below
~30 %, predictive power becomes very low [16].
And if we are given the structures of two proteins known to
interact, but no template, can we predict how they will do so with
atomic precision? Well, sometimes yes, sometimes no. While much
effort has been expended over the past 10 years into testing and
evaluating ab initio docking algorithms [17], there is a problem. In
short, proteins are never truly rigid bodies, and they do not operate
in a vacuum. In the minimalist case, main chains make minor
adjustments (1–2 Å), interfacial residues alter their side-chain tor-
sion angles, and surface-bound water is released or reorganized
upon complex formation. These are considered “rigid-body” com-
plexes; and, in fact, modern algorithms and computing power can
now predict these reasonably well [18]; see Chapter 4. In contrast,
ab initio predictions nearly always fail when complex formation
triggers large conformational changes in either or both proteins
[17, 19, 20]; but these are the complexes that provide the most
biological insight into signaling pathways.
And moving beyond the binary interactome, how can we tackle
those large, often transient, multiprotein signaling hubs? They do
not crystallize, so in order to build atomic models, we need to
combine crystallography with other methods. Principal among
these complementary methods is electron microscopy, which is
actually optimal for larger particles (>500 kDa) and can provide
“sub-molecular” (5–15 Å resolution) maps or envelopes [21], into
which atomic models can be fitted with high overall precision in
the case of rigid-body docking, or when quaternary reorganization
of domains occur [22]. Another major technique is Small-Angle
X-ray Scattering (SAXS), which is performed in solution and can
discriminate (with high precision) between competing models of
quaternary organization [23]. Many other complementary techni-
ques are discussed in Parts II, III, and IV of this book.
In this chapter, I begin by classifying and describing the proper-
ties of binary complexes, and then review the regulatory mechan-
isms of binary interactions as well as large multidomain signaling
proteins. While the diversity of interactions may be finite, we should
continue to expect surprises. For example, over the past 10 years,
the profound role of “domain–peptide” interactions in higher
eukaryotes (arising from “pre-translational modification”) has
been established [24]; and giant steps have been made in defining
the roles of mechanical force, which can be considered as an addi-
tional (and often decisive) allosteric effector in many signaling
pathways [25]. Clearly, much remains to be learned.
2 Classification of Binary Protein Complexes
Interacting domain-pairs may be classified as reversible and irrevers-

ible (a very similar classification found in the literature is “transient”
and “obligate”). Irreversible complexes are typified by large
machines such as the proteasome or ribosome. They have high
pairwise and/or collective affinities, and remain associated until
they are damaged, or cells have no further use for them. By con-
trast, reversible complexes display a spectrum of affinities and life-
times: (1) Some complexes come together weakly and assemble
transiently; and some of these may proceed to form parts of a larger
multiprotein complex that localizes to a specific cellular location
and is stabilized by multiple interactions (2) Other reversible signal-
ing complexes have a high pairwise affinity, but only for a limited
time. For example, kinase-mediated phosphorylation on tyrosine
may generate a strong bond with a partner; but, sooner or later, a
cognate phosphatase will remove the phosphate group, and the
complex will disassemble [26]. (3) Some complexes are strong
but have built-in timers, such as the G protein complexes; their
active lifetime is determined by the rates of binding and hydrolysis
of GTP [27]. (4) At the far end of the reversible spectrum are
antibody–antigen and protease–inhibitor complexes, which may
have lifetimes of hours to days; some of these may be classified as
irreversible, depending on the timescale of the biological process
being studied [28].
Structural Basis 7
An independent classification distinguishes “domain–domain”

and “domain–peptide” complexes [29, 30]. In the former case,
both partners are pre-folded; in the latter, one component is a
pre-folded domain, while the “peptide” is a short linear motif
(appended to a different folded domain) that folds only when it
binds to its cognate domain. These types of interaction are a vital
element of reversible signaling in higher eukaryotes (see below).
Short motifs may also be observed as flexible, dynamic recognition
elements within the context of a larger irreversible complex such as
the ribosome [31].
There is also an intriguing class of protein that lacks any inher-
ent three-dimensional structure, at least when purified to homoge-
neity in a test tube, and yet functions as a signaling molecule in cells
by acting as a flexible linker, or wrapping around other proteins in
order to display its recognition motifs (reviewed in refs. [32–34]).
3 The Architecture and Energetics of Domain–Domain Interfaces
A number of general principles have arisen from the study of known

complexes, which have been refined and extended in recent years
(reviewed in refs. [35–37]).
3.1 The Standard Most domain–domain interfaces have a typical layout, which
Patch defines a standard “patch” (Fig. 2a). A central solvent-excluded
region (“core”) is surrounded by a partly buried outer ring (“rim”)
that includes water-mediated interactions. The core, on average,
Fig. 2 The standard patch and interface size. (a) Slice through an ideal domain–domain interfacial “patch”
(side view). All the atoms that lose full or partial accessibility to solvent when ligand binds are interface atoms.
Dark-shaded atoms, which were solvent-exposed but become fully buried constitute the core, which
is surrounded by a partly buried rim that forms a water-tight “O-ring.” “Hot spot” residues typically comprise
a subset of the core (see text). (b) Distribution of buried surface area (Å2) among a selection of mostly
reversible heterodimeric complexes, including transient redox complexes, signaling, enzyme–inhibitor, and
antibody–antigen complexes. Interfaces larger than 2,000 Å generally comprise more than one patch. Adapted
from ref. 35
contributes ~75 % of the buried surface area and the majority of the
binding energy. It is typically more hydrophobic than the surface in
general but less than the interior of the protein. The rim has a
similar composition to the protein surface; it contributes ~25 % of
the buried surface, and its major role is thought to be the exclusion
of water molecules from the core, analogous to the “O-ring” model
proposed by Bogan and Thorn [38]. Good shape and charge
complementarity is important, such that the packing density is
not very different from the interior of the protein. Most interfaces
are rather flat, with a typical RMS deviation from planarity of ~2 Å.
The interfacial “patch” varies in size, but most are in the range of
1,200–2,000 Å2 (Fig. 2b). Interfaces larger than this typically
comprise two or more patches.
Bringing two hydrophobic surfaces into close apposition in
water is generally favorable (the “hydrophobic effect”), although
its physical basis is still debated (see Chapters 2, 4, and 7). An
entropic origin, arising from the release of water molecules into
bulk solvent, is often cited; but careful thermodynamic measure-
ments of mutated interfaces point to an enthalpic driving force;
moreover, the free energy change (ΔΔG) scales linearly with the
change in buried surface, ~20 cal/Å2 at the rim and ~45 cal/Å2 at
the center of the core [39]. A minimal interface of 1,200 Å2 would
therefore equate to a free energy of ~4 kcal/mol or a Kd ~ 1 μM.
For a 2,000 Å2 interface, this estimate rises to Kd ~ 10 nM. Both
values are commensurate with the local concentrations of interact-
ing proteins in vivo.
A priori, the energetics of polar interactions at interfaces are
much harder to assess, because complex formation involves the
displacement of bound water molecules from both sides of the
interface (gain in entropy, loss in enthalpy), which must be com-
pensated by a new set of ionic, polar, and van der Waals contacts
between the protein partners (gain in enthalpy, loss in entropy), so
that the net energetic effect of polar interactions may be positive,
negative, or negligible. However, a simplifying feature may be the
presence of energetic “hot spots” (see below).
4 Residue Usage at Interfaces
As the number and type of interfaces found in the PDB have

increased over the past 20 years, it has been recognized that
amino acid (residue) composition is systematically different in
different types of complex, e.g., between irreversible and reversible
complexes; and between homodimeric and heterodimeric com-
plexes [40]. This probably explains many apparently conflicting
accounts, especially in the earlier literature, which were based on
smaller “mixed” data sets. And to further confound comparative
studies, there are at least four distinct ways of describing
Structural Basis 9
interfaces, as enumerated below. Note that while this discussion

focuses on side chains, it should be borne in mind that about
one-third of all interfacial H-bonds are mediated by backbone
atoms.
4.1 Residue Count Perhaps the simplest method of describing an interface is to count
the number of residues of a given type, and express them as a
percentage of the total. Sometimes, residues counts are weighted
by their size (i.e., their Å2 contribution to the buried surface).
Recently, Levy and colleagues collated a comprehensive “mixed”
set of protein-protein interfaces from human, yeast, and E. coli
[41]. Figure 3a plots, for human complexes, the relative incidence
Fig. 3 Residue usage at interfaces. (a) Residues are color-coded by type: hydrophobic (black); polar (orange);
charged (positive ¼ blue, negative ¼ red); or aromatic (green), and sorted within their groups by contribution
to interface (blue bars) and surface (gray bars (stacked)). Average interface incidence by type is given as %
below residues. Values above bars are total contributions for each type. Plot derived from data on human
complexes [41]. (b) The difference in residue incidence between reversible and irreversible heterodimeric
complexes. Bars above zero indicate an increase in residue usage at reversible interfaces. An enrichment in
charged and aromatic residues at the expense of hydrophobic and small polar residues is evident. Plot derived
from data in ref. [40]. (c) Residues ordered by increasing “Residue Propensity,” p (red bars; scale on right
ordinate), where p ¼ ln (fi/fs) (see main text). Blue bars (scale on left ordinate) show the actual incidence (%)
of residues at the core. Plot derived from data in ref. [41]. (d) Residues ranked in decreasing order of their
energetic contribution to hot spots (vertical bars), based on a data set of >2,000 alanine mutations [38]. Also
indicated, as connecting brackets (numbers refer to boxed inset), are the most common pairwise interactions
reported for a large set of reversible heterodimeric interfaces, ranked by incidence at the core [44]. Refer to
Table 1 for further information
of each residue, grouped by type in increasing order of their

appearance at the interface as well as their overall occurrence on
the surface.
Leu is the most prevalent residue at interfaces in general,
followed, perhaps surprisingly, by Arg: its special role is described
in more detail below. Charged residues are more prevalent than
polar residues, and both are more abundant on the surface in
general (except for Arg and His). The aromatics have a low abun-
dance on surfaces, but a high (except for Trp) abundance at inter-
faces. Hydrophobic residues have a generally high incidence at
interfaces, and a lower incidence on surfaces. Cys is particularly
rare on surfaces and interfaces.
Figure 3b illustrates the difference in residue incidence
between reversible and irreversible heterodimeric complexes, based
on data from Ofran and Rost [40]. The differences are modest, but
a clear and logical pattern emerges: reversible interfaces are enriched
in all four charged residues, as well as the polar Gln and the
aromatics, especially Tyr, while they are depleted in all hydrophobic
and smaller polar residues. A similar comparison for homodimers
shows additional changes in the aromatic content (not shown); but
reports from different authors vary, presumably because the data
sets and criteria for defining reversibility differ [40, 42].
4.2 Interface A popular measure of interface composition is the “interface pro-

Propensity pensity,” p, which answers the question, “Given the presence of a
particular residue on the protein surface, what is its a priori proba-
bility of being found at an interface?” (Fig. 3c). If fi and fs are the
probabilities of finding a given residue at the interface versus else-
where on the surface, then p ¼ ln(fi/fs). For example, aromatic
residues have high p values: they are poorly represented on surfaces
in general; but if you observe a Phe, Tyr, or Trp on the surface, then
there is a high a priori probability that it will be involved in a
protein-protein interaction. Note that the ratio fi/fs defines only
the relative probability for a given residue and does not consider its
absolute abundance on surfaces. Thus, charged residues have low
p values, but are abundant at both surfaces and interfaces. So that
while interface cores are highly “enriched” in aromatic residues, you
are still more likely to find one of Asp, Glu, or Lys in the core as one
of Phe, Tyr, or Trp.
4.3 Residue Pairs A third method focuses on the frequency with which specific pairs
of residues interact across interfaces [43]. In a recent study of a
diverse set of reversible heterodimeric complexes [44], with inter-
faces ranging from 700 to 8,500 Å2 and an average size of
2,100 Å2, the results are quite striking. The authors also divided
residue distribution into categories of total interface and core,
making the results more directly comparable with the approaches
above. The ten most abundant interactions are listed in Table 1 and
Structural Basis 11
Table 1
Common pairwise interactions at interfaces
Rank
Residue
Core Interface pair Interaction type(s)
1 1 Arg–Glu Mostly salt bridges
2 3 Lys–Asp Mostly salt bridges
3 9 Trp–Arg Π–cation (75 %)
4 5 Tyr–Arg Π–cation (40 %); s/c H-bond 40 %
5 – Phe–Leu Hydrophobic s/c packing
6 7 Tyr–Lys s/c H-bond (55 %); s/c-m/c (33 %)
7 6 Tyr–Asn Π–cation (15 %); s/c H-bond (55 %)
8 – Trp–Ile Hydrophobic s/c packing
9 2 Arg–Asp Mostly salt bridges
– 4 Lys–Glu Mostly salt bridges
The table lists the top ten most common pairwise interactions in a large collection
of reversible heterodimeric complexes [44]. They are ranked according to their incidence
at the core. Incidence at the interface is also shown. Residues are color-coded as in Fig. 3.
s/c side chain, m/c main chain, H-bond hydrogen bond
Fig. 3d. All four charge-charge pairs are present; two of them
(Arg–Glu and Lys–Asp) are highly abundant at both the rim and
core, while the other two are abundant at the rim but not the core.
The other six pairs involve aromatic residues: two of them involve
purely hydrophobic side-chain packing (Phe–Leu and Trp–Leu),
and are restricted to the core; the other four are polar interactions,
three of which are often mediated by Π–cation bonding (see
below).
4.4 Hot Spots and A fourth descriptor of interfaces focuses on the energetics of the
the Special Role of interaction and relies on the results of alanine scanning mutagene-
Π–Cation Interactions sis. Thus, while many mutations at the interface have a limited
effect on binding (ΔΔG), certain clusters, called “hot spots” [38],
have a large/dominant effect. Hot spots are nearly always buried
near the center of the core and well shielded from solvent. They
pack against hot spots on their cognate domains, forming an intri-
cate cooperative network of interactions. Sequence conservation
(between orthologous complexes) is highest at hot spots [45].
Residue preferences at hot spots are a subset of those described
in the previous section (Fig. 3d). Trp, Arg, and Tyr are the most
common, accounting for more than half of the total. These three
residues are versatile in being able to form hydrophobic, aromatic,
and polar interactions, all of which are required to bury comple-

mentary surfaces and satisfy unmet hydrogen-bonding needs.
Moreover, the polar “Π–cation” bond between Arg and either
Trp or Tyr is found at more than 50 % of hot spots [46]. The
guanidinium group of Arg may lie either coplanar or orthogonal to
the aromatic ring. Tyr can also make Π–cation interactions with
Arg, but conventional side-chain interactions are more common.
By contrast, the most common residue at interfaces, Leu, is rarely
found at hot spots (as judged by alanine substitution), while,
curiously, Ile is abundant.
When interfaces are viewed this way, it is not surprising, per-
haps, to find that algorithms can be trained rather well (using
known structures, their experimentally defined hot spots, as well
as evolutionary information) to recognize and predict hot spots
[45]; and significant success has even been reported based on
sequence information alone [47].
5 Complex Formation and Conformational Changes
A very useful starting point for probing the linkage between com-
plex formation and conformational changes is the “benchmark”
collection of high-resolution heterodimeric reversible complexes
last updated in 2010 [48]. The set is restricted to those complexes
for which structures of both isolated components are also known. A
plot of the RMS change in structure of the component domains
upon complex formation versus interface size points first to a tight
cluster with differences of ~1 0.5 Å; these approximate to “rigid-
body” docking, and account for about half of the total set (colored
red in Fig. 4a). Most of these interfaces conform to the standard
“patch” model described above. Large RMS differences (3–7 Å) are
almost exclusively associated with reversible signaling complexes, in
which conformational changes, including disorder–order transi-
tions and quaternary changes, are commonplace.
A particularly well-studied example is the interactions between
the small G-proteins (GTPases) and their effectors, activators, and
inhibitors (reviewed in ref. [49]). For example, guanine nucleotide
exchange factors (“GEFs”), which activate small G-proteins by
recognizing the inactive GDP-bound form and promoting GDP
release and GTP binding, show a great variety in their structures
and modes of interaction, and several distinct nucleotide exchange
mechanisms (Fig. 4b). And while some GEFs are specific for a
single G-protein, others are quite promiscuous. Although they all
utilize at least part of one or both of the “switch” regions (regions
that undergo conformational changes upon hydrolysis of GTP),
different GEFs bind to different parts of the G-protein surface and
display a large range of interface sizes that extend well beyond the
“rigid-body” set.
Structural Basis 13
Fig. 4 Conformational changes and complex formation. (a) RMS change in protein structure following complex
formation, as a function of buried surface area. Red diamonds approximate to “rigid-body” docking. Black
diamonds are typically signaling complexes. See main text for details. Based on data from ref. 48. (b) The
remarkable variety of interactions made between small G proteins (shown in gray, in approximately the same
orientation) and their GTP Exchange Factors (GEFs). All GEFs recognize the GDP-bound form by binding to at
least one of the switch regions of the G protein (highlighted in red). The complexes shown are: 1 Ras:SOS,
2 Sec4p:Sec2p, 3 Cdc42:Dock9, 4 Rab21:Rabex5, 5 Cdc42:Dbs, 6 Rab35:DENN1B, 7 Rab8:MSS4, 8 Rop4:
RopGEF8, 9 Ypt1p:TRAPP1, 10 Arf1:Gea2, 11 Ran:RCC1. Adapted from ref. 49, with permission
The integrin family of αβ heterodimeric plasma membrane

receptors illustrates a distinct type of protein-protein interaction,
one that is mediated in part by a metal ion (Mg2+ or Ca2+) that
forms a bridge between integrin and ligand (Fig. 5). The α-subunit
“I-domain” mediates binding to extracellular matrix and counter-
receptors on other cells, via a “metal-ion dependent adhesion site”
(MIDAS) [50]. The structure of a complex of the α2 I-domain with
a fragment of triple helical collagen shows how a glutamate side
chain from the collagen completes the coordination sphere of the
Mg2+ ion [51] (Figs. 5 and 6).
The total buried surface area—1,200 Å—is near the lower limit
of stable interfaces, which is even more surprising given that a
substantial part of the binding energy must be used to drive con-
formational changes (including a 10 Å shift of the C-terminal
helix), which are ultimately transduced through the plasma mem-
brane (“outside-in” signaling [52]). Presumably, the strong bond
formed between the metal ion and the ligand glutamate represents
a high-energy hot spot. This mode of interaction is conserved in the
structure of the αL I-domain in complex with ICAM-1 [53], and is
likely conserved across the integrin family. The hot spot provides
the energy for a general mode of binding, while specificity arises
from additional contacts with the surrounding residues. A similar
principle, called “dual recognition,” rationalizes the structural basis
of binding specificity between bacterial endonuclease colicins and a
family of immunity proteins [54].
Fig. 5 Conformational changes in the α2 I domain upon binding collagen. Upper

panel shows the major changes in the I domain, notably a large (10 Å) shift of the
C-terminal helix (α7) that promotes the “open” conformation of the integrin
head. Lower panel compares the MIDAS motif in the absence (left) and presence
(right) of ligand. Ligand binding is dominated by the metal–glutamate bond,
which changes the metal coordination (T221 directly coordinates; D254 loses
coordination), triggering the tertiary changes seen in the Upper panel that
underlie signal transduction (from PDB code 1DZI [51])
The β-subunit of integrins also contains an I-domain. Although

it lacks detectable sequence homology with the α-I domain, it does
contain a DxSxS sequence that is characteristic of the MIDAS
motif, and it recognizes proteins containing the linear motif,
RGD. Based on the α-I domain observation, the β-I domain fold
was correctly predicted [51, 55], including the role of the RGD Asp
in mediating binding to the MIDAS metal ion; and even
Structural Basis 15
Fig. 6 Homologous domains may bind the same ligand but at different locations. Both domains are members
of the same Pfam family (vWF A domain). The five central β-strands superpose with an RMS difference of
0.69 Å for 35 Cα positions. At right, the integrin α2 “I” domain engages triple-helical collagen via the MIDAS
motif (labeled “M”) at the “top” of the domain (see Fig. 5). At left, the von Willebrand Factor A3 domain lacks
one of the metal-coordinating residues that comprise the MIDAS motif, and does not bind metal (PDB code
4DMU [58]). It also binds collagen, but on the “front” face of the A domain, and no conformational changes are
observed
conformational changes within the C-terminal helix linked to

signaling [56].
As a caveat, the von Willebrand Factor A1 and A3 domains have
detectable (>20 %) sequence identity, as well as high structural
similarity with integrin α-I domains. However, they lack one or
more of the MIDAS residues required for binding metal, and
ligands bind at different surfaces [57, 58] and do not trigger
conformational changes (see Fig. 6).
6 Protein-Peptide Interactions: A Key Mediator of Complexity in Higher Eukaryotes
As noted above, recent years have witnessed an explosion in the

structural characterization and prediction of protein–peptide inter-
actions, and a comprehensive analysis has recently been performed
[24]. The “peptides,” typically 4–11 residues long, are often
referred to as “SLiMs” (Short Linear Motifs). SLiMs are widely
used by higher eukaryotes to perform a variety of targeting and
signaling functions [59]. The most common “receiver” domains
are PDZ, PTB, SH2, and WW [60]. SLiMs adopt a variety of
structures when bound to their receiver domain: an extended
structure that lies in a groove (14-3-3 [61]; Rb/E7 [62]); a
β-strand that augments a β-sheet (PTB [63] and PDZ [64]
domains); α-helices that augment a helical bundle (FAK-paxillin
[65]; β-/α-catenin [66]); or even a polyproline helix (SH3 [67]).
SLiMs most often arise by alternative gene-splicing (also by
alternative promoter usage and RNA editing), typically as part of
a larger intrinsically disordered region (IDR) that is encoded by a

single (alternative) exon [68]. Several curated websites collect and
classify these interactions. The Eukaryotic Linear Motif (ELM)
database is a major repository: it currently contains 2,070 validated
SLiMs [69], while the “3DID” identified 462 distinct classes of
domain–SLiM complexes [4].
The small interfaces (average ~350 Å2) imply that
domain–SLiM complexes are inherently weak (10 μM) and revers-
ible; but SLiMs and their surrounding (IDRs) are also enriched in
phosphorylation sites, which make them prime candidates for
regulated, dynamic assembly [70]. Moreover, SLiMs are not con-
strained by a preexisting 3D fold, facilitating rapid evolution [71].
Indeed, a dedicated server provides a collection of more than 700
validated examples of SLiM-based motif switch mechanisms that
have been categorized into ten different types [72]. Despite their
low pairwise binding energy, SLiMs can be highly specific in vivo.
Auxiliary mechanisms presumably account for this, including avid-
ity effects, augmentation by (phosphorylated) IDRs, and compart-
mentalization [73].
The interaction between the NPxY motif of integrin tails with
PTB domains provides an example of motif-switching. The motif
forms a helical turn preceded by a short β-strand that augments a
β-sheet. The PTB domain of talin prefers the non-phosphorylated
motif, but the “classical” PTB domain of Shc requires tyrosine
phosphorylation [74]. The talin–integrin linkage promotes cell
adhesion, but subsequent phosphorylation of the NPxY motif
leads to a switch in recognition to Shc, resulting in a shift from
cell adhesion to cell migration [75].
Translation of alternative mRNAs greatly increases the effective
size of the proteome of higher eukaryotes, and facilitates the fine-
tuning and diversification of signaling pathways [33]. In humans, it
has been estimated that ~86 % of genes are alternatively spliced
[68], often with multiple variants selected in a tissue- or
developmental-dependent fashion. One example is the γ-isoform
of the lipid kinase, PIPKIγ, which has at least five splice variants.
One of them has a unique 28-residue C-terminal segment that
binds to the talin PTB domain, targeting it to Focal Adhesions
(see below) [76], while two recently described variants cause them
to localize either to the nucleus or to intracellular vesicles [77].
This phenomenon of “Pre-translational modification” may
resolve the conundrum of how a complex organism, such as
Homo sapiens, can function with a “basis-set” of ~30,000 protein-
encoding genes, only ~4 times the size of the genome of the
unicellular budding yeast [68]. A recent study of the transcriptomes
of six primate species underlines the importance of alternative
splicing [78]. Thus, although a good deal of alternative exon
usage is species-dependent, and probably arises from neutral drift
or noisy splicing, the authors found that a “sizeable minority” of
Structural Basis 17
alternative exon usage, which is enriched in SLiMs and IDRs,

exhibited strong tissue-dependence that is conserved from human
to macaque.
Finally, the linear nature of SLiMs lends themselves to
structure-based predictions. Combined structural and sequence
information was used to predict models for 46 new domain–SLiM
interaction classes [24]. And turning the concept around, a
structure-based predictor of novel SLiM targets for PDZ domains
was reported [79].
7 Intramolecular Protein-Protein Interactions and Signal Transduction
Signal transduction involves much more than the regulated,

sequential association of proteins, one to the next. Multidomain
proteins typically adopt a default auto-inhibited globular confor-
mation, stabilized by head–tail interactions. They respond to
multiple upstream activating signals (such as phosphorylation, co-
localization of binding partners, and mechanical force); and, once a
threshold has been reached, a series of increasingly activated states
output distinct combinatorial signals [80–82].
Studies on Focal Adhesion Kinase (FAK), a scaffolding and
signaling protein, illustrate many of these features. Focal Adhesions
[83] are dynamic, multicomponent complexes that assemble
around the cytoplasmic face of integrins, controlling the linkage
between the Extracellular Matrix (ECM) and the actin cytoskele-
ton. FAK comprises FERM, kinase, and FAT (Focal Adhesion
Targeting) domains, connected by flexible linker regions rich in
phosphorylation sites. In the auto-inhibited state, the FERM and
kinase domain pack tightly together, inhibiting the functions of the
FERM domain [84], and the C-terminal FAT domain inhibits
kinase activity. Step 1: the FAT domain recognizes paxillin (via a
pair of α-helical SLiMs), close to the plasma membrane. Step 2: this
releases some restraints on the kinase domain, which can now
phosphorylate another FAK molecule at position Tyr397, but
only if one is nearby (i.e., “coincidence” detection). Step 3: A Src-
family kinase binds to pTyr397 (via its SH2 domain), and can then
further phosphorylate FAK at its kinase domain. Step 4: This dis-
rupts the FERM-kinase interaction, fully activating the kinase, and
exposing new binding sites on the FERM domain for integrin,
phospholipid, and talin that stabilize attachment to the FA.
Step 5: Subsequent phosphorylation of the FAT domain disrupts
paxillin-binding (which is no longer required for FA-targeting),
enabling Grb2 binding that provides a hub for Ras/MAPK signal-
ing (reviewed in ref. [85]).
Thus, weak binding of FAK to paxillin suggests that FAK first
samples the membrane environment until it finds clusters of paxillin
molecules (which occur at nascent FAs), where other FAK
molecules are likely to be bound, at least transiently.

Auto-phosphorylation in trans thus results from “coincidence-
detection.” Further phosphorylation of FAK then proceeds in an
orderly temporal progression, in which new binding surfaces are
exposed, new bonds to the membrane are made, and domains
switch their function. Interestingly, in neurons, an alternatively
spliced variant of FAK is found with a 7-residue insertion adjacent
to Tyr 397. This enables FAK to autophosphorylate in cis, elim-
inating the need for coincidence detection and implying a different
kind of regulation and function [86].
8 Force and the Single Molecule
Many intracellular signaling/scaffolding molecules switch from a

compact auto-inhibited form to an extended, open or “activated”
form. Now, if the molecule is anchored at two points—one fixed,
the other mobile—then mechanical force applied in the appropriate
direction should, in principle, (help to) open up and activate the
molecule. This force–function relationship has been demonstrated
in spectacular fashion for talin, which unfolds, reversibly, by up to
3,000 Å, in order to maintain its linkage between integrin and
actin. Here, the fixed linkage is integrin (bound to the ECM),
and the dynamic linkage is the actin cytoskeleton, which undergoes
a “retrograde” flow during cell migration [87, 88]. Unfolding of
talin exposes multiple binding sites for other cytoskeletal proteins,
including vinculin.
Indeed, it is now recognized that mechanical force, either of
intracellular (myosin-like motors) or extracellular (hydrodynamic
shear) origin, plays a major role in defining or fine-tuning the
biological activity of many proteins [89]. Interestingly, there is
also recent evidence for force-dependent and force-independent
mechanisms among related receptors. For example, integrin α5β1
requires fibronectin-mediated tension to activate FAK, while
collagen-bound integrin α2β1 activation of FAK is tension-
independent [25]. Understanding the structural basis of this selec-
tive interplay between mechanical and chemical forces is a
fascinating field for future study.
9 Concluding Remarks
The complexity of conformational switches in signal transduction

complexes provides perhaps the clearest imperative for further
careful biochemical and structural studies at atomic resolution,
using a combination of “direct” and hybrid methods. Modern
proteomics approaches produce “wiring diagrams” of signaling
pathways of whole cells, but these do not by themselves explain
Structural Basis 19
how cells work. We must learn the nature of each switch (e.g.,
phosphorylation, phospholipid-binding, mechanical stress), and
precisely how each of the binding functions or catalytic activity of
each protein is turned on/off by each switch, both singly and in
combination.
But this is only the end of the beginning. Next, we need to
observe multiprotein complexes in their native environment (see
Part IV). Recent advances in EM (“Tomography”) have enabled
direct visualization of protein complexes at molecular resolution
(20–60 Å) in flash-frozen (fixed, but unstained/undamaged) thin
cell sections [90]. And finally, cells are most definitely not static
structures: we must observe single molecules as they go about
their business in living cells. For example, during cell migration,
multiprotein complexes and higher subsystems (e.g., the cytoske-
leton–membrane–matrix linkage) assemble, work, and then disas-
semble, on a rapid timescale. Ongoing developments that combine
light microcopy of living cells (using fluorescently labeled protein)
with electron tomography of the same (subsequently frozen) cell
(termed “correlative microscopy”) provide a glimpse into a future
in which we may truly begin to understand how protein-protein
interactions drive the organization and dynamics of living
cells [91].
References
1. Schreiber G, Keating AE (2011) Protein bind- viruses from electron cryo-microscopy. Curr
ing specificity versus promiscuity. Curr Opin Opin Struct Biol 21:265–273
Struct Biol 21:50–61 9. Almo SC, Garforth SJ, Hillerich BS et al
2. Levy ED, De S, Teichmann SA (2012) Cellular (2013) Protein production from the structural
crowding imposes global constraints on the genomics perspective: achievements and future
chemistry and evolution of proteomes. Proc needs. Curr Opin Struct Biol 23:335–344
Natl Acad Sci U S A 109:20461–20466 10. Cuff AL, Sillitoe I, Lewis T et al (2011)
3. Finn RD, Tate J, Mistry J et al (2008) The Extending CATH: increasing coverage of the
Pfam protein families database. Nucleic Acids protein structure universe and linking structure
Res 36:D281–D288 with function. Nucleic Acids Res 39:
4. Mosca R, Ceol A, Stein A et al (2014) 3did: a D420–D426
catalog of domain-based interactions of known 11. Kundrotas PJ, Vakser IA, Janin J (2013) Struc-
three-dimensional structure. Nucleic Acids Res tural templates for modeling homodimers.
42:D374–D379 Protein Sci 22:1655–1663
5. Stumpf MP, Thorne T, de Silva E et al (2008) 12. Kundrotas PJ, Zhu Z, Janin J et al (2012)
Estimating the size of the human interactome. Templates are available to model nearly all
Proc Natl Acad Sci U S A 105:6959–6964 complexes of structurally characterized pro-
6. Venkatesan K, Rual JF, Vazquez A et al (2009) teins. Proc Natl Acad Sci U S A 109:
An empirical framework for binary interactome 9438–9441
mapping. Nat Methods 6:83–90 13. Zhang QC, Petrey D, Norel R et al (2010)
7. Voorhees RM, Weixlbaumer A, Loakes D et al Protein interface conservation across
(2009) Insights into substrate stabilization structure space. Proc Natl Acad Sci U S A
from snapshots of the peptidyl transferase cen- 107:10896–10901
ter of the intact 70S ribosome. Nat Struct Mol 14. Gao M, Skolnick J (2010) Structural space of
Biol 16:528–533 protein-protein interfaces is degenerate, close
8. Grigorieff N, Harrison SC (2011) Near-atomic to complete, and highly connected. Proc Natl
resolution reconstructions of icosahedral Acad Sci U S A 107:22517–22522
15. Zhang QC, Petrey D, Garzon JI et al (2013) 31. Tourigny DS, Fernandez IS, Kelley AC et al
PrePPI: a structure-informed database of (2013) Elongation factor G bound to the ribo-
protein–protein interactions. Nucleic Acids some in an intermediate state of translocation.
Res 41:D828–D833 Science 340:1235490
16. Aloy P, Ceulemans H, Stark A et al (2003) The 32. Brown CJ, Johnson AK, Dunker AK et al
relationship between sequence and interaction (2011) Evolution and disorder. Curr Opin
divergence in proteins. J Mol Biol Struct Biol 21:441–446
332:989–998 33. Dunker AK, Silman I, Uversky VN et al (2008)
17. Janin J (2013) The targets of CAPRI rounds Function and structure of inherently disor-
20–27. Proteins 81:2075–2081 dered proteins. Curr Opin Struct Biol
18. Janin J, Rodier F, Chakrabarti P et al (2007) 18:756–764
Macromolecular recognition in the Protein 34. Janin J, Sternberg MJ (2013) Protein flexibil-
Data Bank. Acta Crystallogr D Biol Crystallogr ity, not disorder, is intrinsic to molecular rec-
63:1–8 ognition. F1000 Biol Rep 5:2
19. Janin J (2010) The targets of CAPRI Rounds 35. Lo CL, Chothia C, Janin J (1999) The atomic
13–19. Proteins 78:3067–3072 structure of protein-protein recognition sites. J
20. Wass MN, David A, Sternberg MJ (2011) Mol Biol 285:2177–2198
Challenges for the prediction of macromolecu- 36. Wodak SJ, Janin J (2003) Structural basis of
lar interactions. Curr Opin Struct Biol macromolecular recognition. Adv Protein
21:382–390 Chem 61:9–73
21. Lander GC, Saibil HR, Nogales E (2012) Go 37. Levy ED, Teichmann S (2013) Structural, evo-
hybrid: EM, crystallography, and beyond. Curr lutionary, and assembly principles of protein
Opin Struct Biol 22:627–635 oligomerization. Prog Mol Biol Transl Sci
22. Rouiller I, Xu XP, Amann KJ et al (2008) The 117:25–51
structural basis of actin filament branching 38. Bogan AA, Thorn KS (1998) Anatomy of hot
by the Arp2/3 complex. J Cell Biol spots in protein interfaces. J Mol Biol 280:1–9
180:887–895 39. Li Y, Huang Y, Swaminathan CP et al (2005)
23. Rambo RP, Tainer JA (2013) Super-resolution Magnitude of the hydrophobic effect at central
in solution X-ray scattering and its applications versus peripheral sites in protein-protein inter-
to structural systems biology. Annu Rev Bio- faces. Structure 13:297–307
phys 42:415–441 40. Ofran Y, Rost B (2003) Analysing six types of
24. Stein A, Aloy P (2010) Novel peptide- protein-protein interfaces. J Mol Biol
mediated interactions derived from high- 325:377–387
resolution 3-dimensional structures. PLoS 41. Levy ED (2010) A simple definition of struc-
Comput Biol 6:e1000789 tural regions in proteins and its use in analyzing
25. Seong J, Tajik A, Sun J et al (2013) Distinct interface evolution. J Mol Biol 403:660–670
biophysical mechanisms of focal adhesion 42. Dey S, Pal A, Chakrabarti P et al (2010) The
kinase mechanoactivation by different extracel- subunit interfaces of weakly associated homo-
lular matrix proteins. Proc Natl Acad Sci U S A dimeric proteins. J Mol Biol 398:146–160
110:19372–19377 43. Glaser F, Steinberg DM, Vakser IA et al (2001)
26. Hunter T (2012) Why nature chose phosphate Residue frequencies and pairing preferences at
to modify proteins. Philos Trans R Soc Lond B protein-protein interfaces. Proteins 43:89–102
Biol Sci 367:2513–2516 44. Headd JJ, Ban YE, Brown P et al (2007)
27. Wolfenson H, Lavelin I, Geiger B (2013) Protein–protein interfaces: properties, prefer-
Dynamic regulation of the structure and func- ences, and projections. J Proteome Res
tions of integrin adhesions. Dev Cell 6:2576–2586
24:447–458 45. Keskin O, Ma B, Nussinov R (2005) Hot
28. Rawlings ND, Tolle DP, Barrett AJ (2004) regions in protein–protein interactions: the
Evolutionary families of peptidase inhibitors. organization and contribution of structurally
Biochem J 378:705–716 conserved hot spot residues. J Mol Biol
29. Dice JF (1990) Peptide sequences that target 345:1281–1294
cytosolic proteins for lysosomal proteolysis. 46. Crowley PB, Golovin A (2005) Cation-Π inter-
Trends Biochem Sci 15:305–309 actions in protein-protein interfaces. Proteins
30. Pawson T, Nash P (2003) Assembly of cell 59:231–239
regulatory systems through protein interaction 47. Chen P, Li J, Wong L et al (2013) Accurate
domains. Science 300:445–452 prediction of hot spot residues through
Structural Basis 21
physicochemical characteristics of amino acid 63. Eck MJ, Shoelson SE, Harrison SC (1993)
sequences. Proteins 81:1351–1362 Recognition of a high-affinity phosphotyrosyl
48. Hwang H, Vreven T, Janin J et al (2010) peptide by the Src homology-2 domain of
Protein-protein docking benchmark version p56lck. Nature 362:87–91
4.0. Proteins 78:3111–3114 64. Doyle DA, Lee A, Lewis J et al (1996) Crystal
49. Cherfils J, Zeghouf M (2013) Regulation of structures of a complexed and peptide-free
small GTPases by GEFs, GAPs, and GDIs. membrane protein-binding domain: molecular
Physiol Rev 93:269–309 basis of peptide recognition by PDZ. Cell
50. Lee JO, Rieu P, Arnaout MA et al (1995) 85:1067–1076
Crystal structure of the A domain from the 65. Hayashi I, Vuori K, Liddington RC (2002)
alpha subunit of integrin CR3 (CD11b/ The focal adhesion targeting (FAT) region of
CD18). Cell 80:631–638 focal adhesion kinase is a four-helix bundle that
51. Emsley J, Knight CG, Farndale RW et al binds paxillin. Nat Struct Biol 9:101–106
(2000) Structural basis of collagen recognition 66. Pokutta S, Weis WI (2000) Structure of the
by integrin α2β1. Cell 101:47–56 dimerization and beta-catenin-binding region
52. Hogg N, Harvey J, Cabanas C et al (1993) of alpha-catenin. Mol Cell 5:533–543
Control of leukocyte integrin activation. Am 67. Musacchio A, Saraste M, Wilmanns M (1994)
Rev Respir Dis 148:S55–S59 High-resolution crystal structures of tyrosine
53. Shimaoka M, Xiao T, Liu JH et al (2003) kinase SH3 domains complexed with proline-
Structures of the αL I domain and its complex rich peptides. Nat Struct Biol 1:546–551
with ICAM-1 reveal a shape-shifting pathway 68. Nilsen TW, Graveley BR (2010) Expansion of
for integrin regulation. Cell 112:99–111 the eukaryotic proteome by alternative splic-
54. Kuhlmann UC, Pommer AJ, Moore GR et al ing. Nature 463:457–463
(2000) Specificity in protein–protein interac- 69. Weatheritt RJ, Jehl P, Dinkel H et al (2012)
tions: the structural basis for dual recognition iELM – a web server to explore short linear
in endonuclease colicin-immunity protein motif-mediated interactions. Nucleic Acids
complexes. J Mol Biol 301:1163–1178 Res 40:W364–W369
55. Xiong JP, Stehle T, Zhang R et al (2002) Crys- 70. Van Roey K, Gibson TJ, Davey NE (2012)
tal structure of the extracellular segment of Motif switches: decision-making in cell regula-
integrin αVβ3 in complex with an Arg-Gly- tion. Curr Opin Struct Biol 22:378–385
Asp ligand. Science 296:151–155 71. Harrison SC (1996) Peptide-surface associa-
56. Luo BH, Carman CV, Springer TA (2007) tion: the case of PDZ and PTB domains. Cell
Structural basis of integrin regulation and sig- 86:341–343
naling. Annu Rev Immunol 25:619–647 72. Van Roey K, Dinkel H, Weatheritt RJ et al
57. Huizinga EG, Tsuji S, Romijn RA et al (2002) (2013) The switches ELM resource: a compen-
Structures of glycoprotein Ibα and its complex dium of conditional regulatory interaction
with von Willebrand factor A1 domain. Science interfaces. Sci Signal 6:rs7
297:1176–1179 73. Stein A, Aloy P (2008) Contextual specificity in
58. Brondijk TH, Bihan D, Farndale RW et al peptide-mediated protein interactions. PLoS
(2012) Implications for collagen I chain regis- One 3:e2524
try from the structure of the collagen von Will- 74. Garcia-Alvarez B, de Pereda JM, Calderwood
ebrand factor A3 domain complex. Proc Natl DA et al (2003) Structural determinants
Acad Sci U S A 109:5253–5258 of integrin recognition by talin. Mol Cell
59. Weatheritt RJ, Gibson TJ (2012) Linear 11:49–58
motifs: lost in (pre)translation. Trends Bio- 75. Cowan KJ, Law DA, Phillips DR (2000) Iden-
chem Sci 37:333–341 tification of Shc as the primary protein binding
60. Weatheritt RJ, Davey NE, Gibson TJ (2012) to the tyrosine-phosphorylated β3 subunit of
Linear motifs confer functional diversity onto αIIbβ3 during outside-in integrin platelet sig-
splice variants. Nucleic Acids Res 40:7123–7131 naling. J Biol Chem 275:36423–36429
61. Yaffe MB, Rittinger K, Volinia S et al (1997) 76. Di Paolo G, Pellegrini L, Letinic K et al (2002)
The structural basis for 14-3-3:phosphopep- Recruitment and regulation of phosphatidyli-
tide binding specificity. Cell 91:961–971 nositol phosphate kinase type 1-γ by the FERM
62. Lee JO, Russo AA, Pavletich NP (1998) Struc- domain of talin. Nature 420:85–89
ture of the retinoblastoma tumour-suppressor 77. Schill NJ, Anderson RA (2009) Two novel
pocket domain bound to a peptide from HPV phosphatidylinositol-4-phosphate 5-kinase
E7. Nature 391:859–865 type Igamma splice variants expressed in
human cells display distinctive cellular target- localization and action. Curr Opin Struct Biol
ing. Biochem J 422:473–482 21:808–813
78. Reyes A, Anders S, Weatheritt RJ et al (2013) 86. Toutant M, Costa A, Studler JM et al (2002)
Drift and conservation of differential exon Alternative splicing controls the mechanisms of
usage across tissues in primate species. Proc FAK autophosphorylation. Mol Cell Biol
Natl Acad Sci U S A 110:15377–15382 22:7731–7743
79. Hui S, Xing X, Bader GD (2013) 87. Margadant F, Chew LL, Hu X et al (2011)
Predicting PDZ domain mediated protein Mechanotransduction in vivo by repeated talin
interactions from structure. BMC Bioinfor- stretch-relaxation events depends upon vincu-
matics 14:27 lin. PLoS Biol 9:e1001223
80. Bakolitsa C, Cohen DM, Bankston LA et al 88. Hu K, Ji L, Applegate KT et al (2007) Differ-
(2004) Structural basis for vinculin activation ential transmission of actin motion within focal
at sites of cell adhesion. Nature 430:583–586 adhesions. Science 315:111–115
81. Balla T (2005) Inositol-lipid binding motifs: 89. Seifert C, Grater F (2013) Protein mechanics:
signal integrators through protein-lipid and how force regulates molecular function. Bio-
protein–protein interactions. J Cell Sci chim Biophys Acta 1830:4762–4768
118:2093–2104 90. Yahav T, Maimon T, Grossman E et al (2011)
82. Carlton JG, Cullen PJ (2005) Coincidence Cryo-electron tomography: gaining insight
detection in phosphoinositide signaling. into cellular processes by structural approaches.
Trends Cell Biol 15:540–547 Curr Opin Struct Biol 21:670–677
83. Schiller HB, Fassler R (2013) Mechanosensi- 91. Zhang P (2013) Correlative cryo-electron
tivity and compositional dynamics of cell- tomography and optical microscopy of cells.
matrix adhesions. EMBO Rep 14:509–519 Curr Opin Struct Biol 23:763–770
84. Lietha D, Cai X, Ceccarelli DF et al (2007) 92. Baker ML, Hryc CF, Zhang Q et al (2013)
Structural basis for the autoinhibition of focal Validated near-atomic resolution structure of
adhesion kinase. Cell 129:1177–1187 bacteriophage epsilon15 derived from cryo-
85. Arold ST (2011) How focal adhesion kinase EM and modeling. Proc Natl Acad Sci U S A
achieves regulation by linking ligand binding, 110:12301–12306
Chapter 2
Quantitative Analysis of Protein-Protein Interactions

Ziad M. Eletr and Keith D. Wilkinson
Abstract
Numerous authors, including contributors to this volume, have described methods to detect protein-protein
interactions. Many of these approaches are now accessible to the inexperienced investigator thanks to core
facilities and/or affordable instrumentation. This chapter discusses some common design considerations that
are necessary to obtain valid measurements, as well as the assumptions and analytical methods that are relevant
to the quantitation of these interactions.
Key words Ligand binding, Protein-protein interaction, Fluorescence, Binding equations, Binding
equilibria
1 Introduction
In the post-genomic era, the importance of protein-protein inter-

actions is becoming even more apparent [1]. We are coming to
recognize that most, if not all, catalytic and regulatory pathways
operate as networks, with frequent and extensive input from
signaling pathways, feedback, and cross talk. Replication, tran-
scription, translation, signal transduction, protein trafficking, and
protein degradation are all accomplished by protein complexes,
often temporally assembled and disassembled to accomplish vec-
toral processes. Often these interactions are driven by interaction
of recognized domains in the constituent proteins. We must
identify and understand these domain interactions in order to
discern the patterns and logic of cellular regulation [2].
1.1 Assumptions There are several assumptions inherent to any analysis of a simple
ligand–receptor interaction (https://tools.lifetechnologies.com/
downloads/FP7.pdf).
1. The interactions are assumed to be reversible. In the simplest
case, the association reaction is bimolecular while the dissocia-
tion reaction in unimolecular.
23
24 Ziad M. Eletr and Keith D. Wilkinson
2. All receptor molecules are equivalent and independent.

3. The measured response is proportional to the number of
occupied receptor sites.
4. The interactions are measured at equilibrium.
5. The components do not undergo any other chemical reactions
and exist only in the free or bound states.
Any or all of these assumptions may prove to be unfounded in a
more complex case. In fact, it is the deviation from simple behavior
that is often the first indication of a more complex binding event and
each assumption should be explored to explain deviations from
simple behavior. Outlined below are treatments for simple cases.
A general method to deriving binding formulas to more complex
cases has been derived from statistical thermodynamic principles [3].
1.2 Binding The receptor–ligand terminology is useful, even if artificial, in the

to One Site case of protein-protein interactions. Either protein could be con-
sidered the receptor or the ligand.
For the purposes of this chapter we refer to the protein present
in fixed and limiting amounts as the receptor and the component
that is varied as the ligand.
Thus, for one molecule of L binding to one molecule of R:
k1
Rf þ L f ----! RL;
---- ð1Þ
k2
where Rf is the concentration of free receptor, L f is the concentra-
tion of free ligand, RL is the concentration of the complex, k1 is the
association rate constant, and k2 is the dissociation rate constant. At
equilibrium:
½R f ½L f k 2
¼ ¼ K d; ð2Þ
½RL k1
where Kd is the dissociation constant.
In most binding titrations, the concentrations of free ligand
and receptor are difficult to quantify and one typically measures the
fractional saturation, [RL]/[Rt] as function of total ligand [L t].
Determining the dissociation constant Kd can be performed in two
ways, depending on the experimental design. If [L f] can be
measured, or if [L t] [Rt] and we can assume [L t] ¼ [L f], a
simpler derivation of fractional saturation vs. [L t] can be applied.
In more general cases where [L f] is not measured yet [L t] is known,
one must solve a quadratic equation to determine the Kd. These
derivations are shown below.
Rearranging the fractional saturation term ([RL]/[Rt]) by
substituting for [RL] in terms of [Rf], [L f] and Kd (Eq. 2) and
applying the conservation of mass assumption [Rf] ¼ [Rt] [RL]
gives:
Quantitation of Binding 25
½R f ½L f
½RL Kd
¼ ; ð3Þ
½R t ½Rf þ ½RL
which can be simplified to Eq. 4:
½RL ½L f
¼ : ð4Þ
½R t K d þ ½L f
Thus, a plot of fraction saturation [RL]/[Rt] vs. [L f] will give the

familiar rectangular hyperbola if only one type of binding site is
present (Fig. 1a). This equation is valid when one can directly
measure [L f], or when [L t] [Rt] and it can be assumed that
[L f] ¼ [L t]. This assumption is valid when 5–10 % of the ligand is
bound [3]. Alternatively, a plot of fractional saturation vs. log[L t]
can be used. If free concentrations are actually measured (instead of
calculated), we can use the Klotz plot [4], a plot of fractional
saturation vs. log[L f] (Fig. 1b); or the Scatchard plot, a plot of
ligand bound/free vs. ligand free (Fig. 1c).
In the second scenario, when [L f] is not measured or when [L t]
is not much greater than [Rt], we must apply the conservation of
mass assumption to [L f] as well. Substituting into Eq. 2:
½Rf ½L f ð½Rt ½RL Þð½L t ½RL Þ

Kd ¼ ¼ : ð5Þ
½RL ½RL
Equation 5 can then be rearranged to the quadratic form

ax2 + bx + c ¼ 0:
a b c
1 1 nRt 1
-1/Kd
0.8 0.8 0.8
0.6
[Lb]/[Lf]
0.6 0.6
[Lb]/[Rt]
[Lb]/[Rt]
0.4 0.4
0.4
0.2 0.2
0.2
Kd
0
0 0
0 2 4 6 8 10 0.01 0.1 1 10 100 0 0.2 0.4 0.6 0.8 1

[Lf] μM Log [Lf] μM [Lb] μM nRt /Kd
Rectangular Klotz plot Scatchard plot

hyperbola
Fig. 1 Plots of simulated data for simple binding. In all cases n ¼ 1 and Kd ¼ 1 μM. (a) Direct plot of
fractional saturation vs. free ligand; (b) Klotz plot of the same data. Note the log scale; (c) Scatchard plot of the
same data. The parameters nRt and nRt/Kd are estimated from the intercepts
0 ¼ ½RL 2 ð½Rt þ ½L t þ K d Þ þ ½Rt ½L t ; ð6Þ
where x ¼ [RL], a ¼ 1, b ¼ ([Rt] + [L t] + Kd) and c ¼ [Rt][L t].

One can then fit the fractional saturation vs. [L t] binding curve to
the quadratic equation solution (one root is positive) to determine
the unknown Kd using nonlinear least squares regression (see
Eq. 8). An example of this method which has been used to deter-
mine the Kd of a protein-protein interaction monitored by fluores-
cence anisotropy is given [5].
1.3 Binding to It should be noted that if more than one ligand molecule binds to R
Multiple Sites then the behavior may be more complex. For n multiple binding
sites we get:
½RL ¼ ½RL 1 þ ½RL 2 . . . ½RL n
½Rt ½L f ½R t ½L f ½R t ½L f
¼ þ ... ; ð7Þ
K d1 þ ½L f K d2 þ ½L f K dn þ ½L f
where n different sites can be occupied by ligand with the
corresponding binding constants.
1.3.1 Identical, Non- If all binding sites are identical and non-interacting (i.e., all bind
interacting Binding Site(s) with the same Kd) then Eq. 7 reduces to:
½L b ½L f
¼ ; ð8Þ
n ½R t K d þ ½L f
where n[Rt] ¼ [Rf] + [Lb].

Note that this equation is similar to Eq. 4 except for the
inclusion of the stoichiometry n. A Klotz plot of fractional satura-
tion vs. log[L f] will be sigmoidal and symmetrical about the mid-
point. The curve is nearly linear from 0.1 to 10 Kd and 99 %
saturation is achieved when [L t] is two orders of magnitude above
Kd. A complete description of binding and accurate estimation of
the plateau values requires that [L f] vary from two log units below
to two log units above Kd. A steeper curve is indicative of positive
cooperativity, while a flatter curve could be due to negative coop-
erativity or the presence of an additional binding site. The stoichi-
ometry is calculated from the plateau value and [Rt] while the Kd is
calculated from the midpoint [6], or more accurately using a non-
linear least squares fit to Eq. 8.
If free ligand is not measured then we must use a plot of
fractional saturation vs. log[L t] and the curve will deviate from
sigmoidal by the difference between log[L f] and log[L t]. This
condition is often referred to as ligand depletion [7, 8]. It should
be recognized, however, that it may not be possible to cover such a
large range of concentrations with proteins. At the low end we are
often limited by the sensitivity of the technique and at the high end
limited solubility or sample amounts that may prevent us from

attaining concentrations necessary to reach the plateau.
An alternative way to plot the data is with a Scatchard plot. For
the last 40 years this has been the traditional method for the analysis
of binding data where [L f] is measured. The Scatchard plot is
described by:
½L b ½L b n½Rt
¼ þ : ð9Þ
½L f Kd Kd
In the simple model, a plot of ligand bound vs. ligand bound/

ligand free gives a straight line with the x-intercept ¼ n[Rt], a
y-intercept of n[Rt]/Kd, and a slope of –1/Kd (Fig. 1c) [6].
Before the advent of computers, estimates of Kd and n were
obtained by any of a number of transformations of the relevant
equations to give linear plots. These include the double reciprocal
plot, and the Scatchard plot. These linearizations are notoriously
difficult to fit and generally fraught with problems. The preferred
method of obtaining Kd and n from binding data is direct fitting of
the data using a nonlinear least squares fitting algorithm. Many
commercial packages for doing such fits are available today. As
discussed in Subheading 1.2, if we do not explicitly measure the
concentration of ligand free, an appropriate solution of the binding
equation to obtain the dissociation constant requires that we deter-
mine and fit the fractional saturation as a function of the concentra-
tion of total L added. The solution of the equation for [RL]/[Rt]
as a function of [L t] is a quadratic equation with the following real
solution:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
½L b ð½ L t þ n ½ R t þ K d Þ ½L t n½Rt K d 2 4½L t n Rt
¼ :
½R t 2n½Rt
ð10Þ
1.3.2 Non-identical While the most common reason for observing multiple non-
Binding Sites identical binding sites in a protein-protein interaction is likely to
be nonspecific binding (see below), it is always possible that there
are two independent and non-interacting sites with different affi-
nities. Either case will manifest itself as a deviation from the
expected behavior for a simple binding model. The Scatchard plot
is a useful diagnostic tool to point out such deviations (Fig. 2). A
Scatchard plot that is concave upward is indicative of nonspecific
binding, negative cooperativity, or multiple classes of binding sites.
A concave downward plot suggests either positive cooperativity or
instability of the ligand. In any case, proper analysis of this behavior
requires other information (for instance stoichiometry, stability)
and the data are best fitted using nonlinear least squares fitting of
the data according to an appropriate model.
a b c
1 1 1
0.8 0.8 0.8
0.6 0.6 0.6

[Lb]/[Lf]
[Lb]/[Lf]
[Lb]/[Lf]
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
[Lb] μM [Lb] μM [Lb] μM
Simple binding Two different sites, Positive cooperativity
Negative cooperativity, or
or Unstable ligand
Non-specific binding
Fig. 2 Effects of complexities on the appearance of the Scatchard plot. (a) Represents the expected behavior in
the simple case; (b) a concave upward deviation as shown in this panel could be caused by the presence
of two different sites, the presence of negative cooperativity or a significant nonspecific binding component;
(c) Positive cooperativity or ligand instability would lead to the curvature shown in this panel
Most deviations from simple binding are expected to be due to

either multiple sites or nonspecific binding which, as discussed
below, may be difficult to distinguish. Either case can be fitted
with appropriate modifications of the simple binding expressions.
Note that a satisfactory analysis of such complicated binding will
require measurement of [L f].
1.4 Cooperativity Cooperativity is the term used to describe the situation where occu-
pancy of one site changes the affinity for ligand at another site. There
have been many treatments of cooperative binding interactions,
including analysis by Scatchard and Hill plots, but these are beyond
the scope of this discussion. In general, models explaining coopera-
tivity invoke subunit–subunit interactions in oligomeric protein
structures and may well be important in cases where multiple pro-
teins are being assembled into a multimeric complex. The reader is
referred to any of several other treatments of such binding if com-
plications of this sort are indicated [9–11]. However, it may be
simpler to restrict the measurements to conditions where individual
subcomplexes are assembled at saturating concentrations before
measuring the binding of a subsequent protein.
2 Materials
The only materials relevant to this chapter are a computer and a

program to mathematically fit the data. Many commercial and
shareware packages capable of nonlinear fitting of equations are
available for all platforms, i.e., Prism (GraphPad Software, Inc., San
Diego, CA), SigmaPlot (SPSS Science, Chicago, IL), Mathematica
(Wolfram Research Inc., Champaign, IL), DynaFit (BioKin, Ltd.,
Pullman, WA), MATLAB (MathWorks, Natick, MA), and others.
There are also published solutions using the popular spreadsheet
Microsoft Excel [12, 13]. The choice is largely up to personal
preference.
3 Methods
Several chapters in this book describe techniques for determining

fractional saturation and/or binding parameters. These basically fall
into two categories, direct methods that measure the actual con-
centration of bound or free ligand, and indirect methods that infer
the concentrations from some measured signal. The choice of
which technique to use may be limited by the strength of the
interactions and the inherent sensitivity of the technique. For
instance, NMR may be a poor choice to monitor binding constants
tighter than micromolar since one commonly needs mM concen-
trations of protein to see a signal. Thus, [Rt] may be Kd and we
would be restricted to measuring only the stoichiometry under
these conditions (see below). Similarly, with an interaction of milli-
molar affinity it may be difficult to determine the stoichiometry
since it may not be possible to attain a concentration of [Rt] Kd.
See below for a discussion of the relationships between Kd and [Rt].
3.1 Direct Direct methods require that we accurately determine the concen-
Measurement of Free trations of free and bound ligand. Examples of techniques that yield
Ligand such information include gel filtration, ultracentrifugation, ultrafil-
tration, or equilibrium dialysis. For binding with slow dissociation
rates pull-downs, band shift or electrophoresis techniques may be
appropriate. If the process of separating the bound and free ligand
is fast compared to the rate of dissociation of the complex such
methods can yield directly the concentrations of bound and free
ligand. If dissociation and separation of bound and free reactants
occur on similar time scales such methods are not appropriate for
quantitation as the equilibrium will be disturbed by the separation
of the reactants. For the same reasons techniques such as cross-
linking may overestimate the concentration of RL since the equi-
librium will be disturbed by the removal of RL from the
equilibrium.
3.2 Indirect More commonly an indirect measure of saturation is used to moni-

Measurements of tor binding. These include optical methods such as fluorescence,
Bound Ligand absorbance, and resonance techniques. These methods all assume
that the output signal is directly proportional to the concentration
of RL present. For instance, if a fluorescence change is being
monitored it is assumed that there are only two states, the bound
and the free, and that each has a characteristic value. If So is the
signal in the absence of binding, SL the signal in the presence of
total ligand concentration L, and S1 is the value at saturation, then:
ðS L S o Þ
fraction saturation ¼ : ð11Þ
ðS 1 S o Þ
The concentration of free ligand can be calculated by assuming a
stoichiometry n and using the expression [L f] ¼ [L t] n[Rt] ¼
(SL So)/(S1 So). Note that if n is incorrect, then the calcu-
lated [L f] will be incorrect also and this will be apparent in the
deviation of the data from the theoretical rectangular hyperbola.
This is one reason why the determination of n is an important
exercise in most binding studies. Alternatively, and preferably,
data are fitted using nonlinear least squares methods and n is
determined directly from this analysis.
3.3 Competition Direct methods measure either bound ligand [RL] or free ligand
Methods [L f] as a function of [L t] and indirect methods usually involve
measuring fractional saturation [RL]/n[Rt] as a function of [L t].
However, one of the most useful variations of the binding experi-
ment is the use of competitive binding assays where a single labeled
indicator ligand can be bound and subsequently displaced by any of
a variety of competitive inhibitors [14–19]. Such experiments are
particularly useful if the affinity of a series of inhibitors is to be
determined. Methods such as fluorescence polarization or flores-
cence resonance energy transfer are particularly well suited for such
measurements. A small amount of the labeled ligand is first bound
to the receptor and subsequently displaced by titrating with unla-
beled inhibitor. The Ki of the unlabeled inhibitor is then calculated.
The labeled ligand does not have to be physiological or bound with
a physiological affinity since we are always comparing the Ki of the
unlabeled inhibitor. Thus, any adverse effects of labeling the indi-
cator ligand will be unimportant.
The IC50 is the concentration of inhibitor necessary to displace
half the labeled ligand. If [Rt] Kd, IC50 is related to Ki, the
affinity of the unlabeled ligand by:
IC50
Ki ¼ ; ð12Þ
1 þ Lt =K d
where [L t] is the concentration of labeled ligand and Kd is its’

dissociation constant. If only relative affinities are to be measured
then comparing IC50 directly is sufficient. If absolute affinities are
desired, then we must also determine the concentration and affinity
of the labeled ligand in the assay.
If [Rt] is similar to or greater than Kd and/or Ki, it follows that
the concentrations of free ligand and inhibitor are not equal to their
respective total concentrations. For this reason, it is simplest to

work at conditions where [Rt] ~ 0.1 Kd so that less than 10 %
of the labeled ligand is bound to the receptor at the start of the
experiment.
If higher concentrations of receptor are necessary or if inhibitor
binds much tighter than ligand, then one has to fit with a more
complex equation [7, 15, 17–19]. The following treatment was
first published by Wang in 1995 [19] and is suitable for fitting the
data from competitive displacement experiments where absor-
bance, fluorescence, or fluorescent anisotropy are measured using
commercially available fitting programs. Consider, for example, the
binding of a fluorescent probe A to a non-fluorescent protein P in
the presence or absence of a competitive inhibitor B that prevents
binding of A.
Given:
K a ¼ ½A f ½P f =½PA
½Af þ ½PA ¼ ½At
½P f þ ½PA þ ½PB ¼ ½P t
K b ¼ ½B f ½P f =½PB
½B f þ ½PB ¼ ½B t ;
then Eq. 13 describes the fractional saturation:

n pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi o
2 ð a 2 3b Þ cos ðθ=3Þ a
ðS S o Þ
¼ n pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi o; ð13Þ
ðS 1 S o Þ 3K a þ 2 ða 2 3b Þ cos ðθ=3Þ a
2a 3 þ 9ab 27c

where : θ ¼ arccos qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
2 ða 2 3b Þ3
a ¼ K a þ K b þ ½A t þ ½B t ½P t
b ¼ K b ð½A t ½P t Þ þ K a ð½B t ½P t Þ þ K a K b
c ¼ K a K b ½P t :
The experiment requires the measurement of the fractional satura-
tion at various concentrations of At, Bt, and Pt. Only a small range
of measurements are useful: the ones where fractional saturation is
>0.05 and <0.95. Fractional saturation of P with the probe A is
determined by indirect measurements where it is the fluorescence
or the anisotropy of AP which gives rise to the signal. The usual
experiment is to measure the full binding curve, i.e., (S So)/
(S1 So) as a function of Pt. This experiment should then be
repeated at three or more concentrations of Bt to calculate Kb.
Although this may seem like its only giving you three data points,
if the curve is fitted, the actual number of useful data points is equal
to the total measurements made where fractional saturation is in a
useful range.
3.4 Parameters of Quantitation of binding often requires accurate estimates of the

Reversible Binding binding stoichiometry n. Many methods are appropriate for this
purpose including cross-linking, pull-downs, and electrophoretic
3.4.1 Stoichiometry methods (when off rates are slow). If association and dissociation
rates are fast these techniques will perturb the equilibrium and give
erroneous results. In these cases stoichiometry must be determined
from more conventional titrations measuring the equilibrium
amounts of RL. To determine stoichiometry an excess of ligand is
present and one of the components must be present at concentra-
tion well above the Kd in order to assure saturation. Often this is
the first experiment that is done as it helps greatly in fitting the data
to more complete titrations.
3.4.2 Kinetics The analysis of binding requires that we conduct the measurements
after binding has reached equilibrium or that we measure individu-
ally the rate constants involved. The binding constant can then be
calculated from the relationship Kd ¼ k2/k1. From a practical
standpoint, assuring that the reaction has reached equilibrium
often involves measuring a time course for binding at low ligand
concentrations and making all measurements after sufficient time to
allow attainment of equilibrium. Several examples of each type of
analysis are given in later chapters.
In any case it is instructive to consider the magnitudes of
association and dissociation rates. The association rate constants
expected for protein-protein interactions are limited by diffusion. If
we assume reasonable numbers for the diffusion rate of an average
protein, the diffusion limit in aqueous solution is around
108–109 M1 s1. There are also additional steric constraints as
only a fraction of the collisions occurring at this rate are oriented
properly, and it is commonly assumed that the rate limiting associa-
tion rate (k1) for two proteins binding to each other is around
108 M1 s1.
It can be shown that the rate of approach to equilibrium is
determined by the sum of the association rate and the dissociation
rate constants. Further, the concentrations of reagents must be at
or near the binding constant for accurate determination of both
stoichiometry and affinity in the same experiment (see below). If
the dissociation constant (Kd) for such an interaction is moderate
(106 M) then the dissociation rate for such a complex will be
k2 ¼ k1 Kd ¼ 102 s1. Thus, binding will be complete in
seconds and the half-life of the bound state will be tens of milli-
seconds. If, however, the binding constant is very tight, as may
occur in antibody–antigen interactions, the overall equilibrium may
take some time. Consider a binding interaction with a free energy of
16 kcal/mol, an affinity exhibited by many antibodies and other
protein-protein interactions [20]. This represents a dissociation
constant of 1013 M. Here, binding may take as long as hours and

the half-life of the bound state could be as long as 20 h. The latter
fact is the reason that tight binding can be detected using techni-
ques like immunoprecipitation and pull-down experiments, but
tight binding complicates the determination of accurate binding
constants.
3.5 Concentrations Equation 8 is the equation for the familiar rectangular hyperbola
of Components to Use with a horizontal asymptote corresponding to 100 % saturation and
half-maximal saturation occurring at L f ¼ Kd. This equation points
3.5.1 Ligand out that the concentrations of free ligand present must be similar to
Concentration the dissociation constant in order to vary the fractional saturation of
receptor, i.e., to measure the strength of binding. The most com-
mon form of the experiment then is to titrate a fixed amount of
receptor with variable amounts of ligand and to fit the experimental
data to the appropriate binding equation to determine the stoichi-
ometry n and the binding constant Kd.
3.5.2 Receptor If we consider the concentration of the fixed protein in this binding
Concentration equation, i.e., [Rt], we can define three limiting conditions: [Rt]
Kd, [Rt] Kd, and [Rt] ~ Kd. Figure 3 illustrates the interre-
lationships between [Kd] and [Rt] in such experiments.
[Rt] Kd Under these conditions saturation is achieved by varying [L] at

concentrations from 0.1 to 10 times Kd. Since [L t] is always
much greater than [RL] under these conditions, then [L f] ~ [L t].
Thus, Eq. 8 can be simplified to give:
a b
1 Kd=.01μM
1 Kd=.01 μM
=1μM
=1 μM
0.8 0.8
=2 μM
=5 μM
0.6 =2μM
0.6
[Lb]/[Rt]
[Lb]/[Rt]
=10 μM
Rt=1 μM Rt=1 μM
=5μM
0.4 =10μM
0.4
0.2 0.2
0 0
0 2 4 6 8 10 0.01 0.1 1 10 100

[Lt] μM log[Lt] μM
Fig. 3 Binding isotherms for a simple binding equilibrium where n ¼ 1 and the total concentration of receptor
is 1 μM. (a) Direct plot of fractional saturation vs. total ligand added and (b) the same data plotted on a log
scale. Note that as the Kd approaches [Rt] there is a significant deviation from the rectangular hyperbolic
behavior
½L b ½L t
¼ : ð14Þ
n ½R t K d þ ½L t
If we only measure the fractional saturation (i.e., the ratio [Lb]/n

[Rt]) as a function of L t then we cannot calculate [Lb] since [Lb]
¼ [L t] [L f] and we have not measured [L f]. Note that even if we
use direct methods and measure free ligand concentration, the
calculation of bound ligand is subject to large errors since the
bound is the difference between total and free and under these
conditions they are about equal [7, 8]. Thus, under these conditions,
we can accurately determine Kd but not n. Determination of accu-
rate values for n requires that the concentration of Rt be similar to
or larger than Kd.
[Rt] Kd If the concentration of Rt is much greater than Kd then Eq. 8 can

be rearranged to give:
½L b n ½R t
¼ : ð15Þ
½L f K d þ ½L f
In the first part of the titration curve, when [L f] is less than Kd (and
much less than n[Rt] in this example), the ratio of bound/free
ligand is determined solely by the ratio of n[Rt]/[Kd].
If we only measure [L t] the limiting slope for a plot of satura-
tion vs. [L t] is n[Rt]/[Kd]. For example, if n[Rt]/[Kd] ¼ 100,
then only about 1 % of the added ligand is free at low ligand
concentrations. In order to saturate binding [L t] must exceed
100 Kd. When [Rt] Kd, the saturation curve is really an end-
point determination consisting of two lines (first a slope of ~n and
then 0 intersecting at [L t] ¼ n[Rt]) with little curvature (Fig. 3).
Under these conditions we can accurately determine n, but not Kd.
If direct methods to measure free ligand are used, we can, in
theory, calculate Kd, but in practical terms the curve will only
deviate from its biphasic nature near L t ¼ Rt, and generally there
will not be enough data in this region to obtain accurate estimates
of Kd.
[Rt] ~ Kd The most useful conditions for determining both Kd and n are
when [Rt] ~ Kd. The binding curve still resembles a rectangular
hyperbola but with small deviations due to the fact that [L t] ¼
[L f] + n[RL]. Since [L f] is similar in magnitude to [Lb] each can
be measured (or calculated) with good accuracy. Under these con-
ditions we can determine both Kd and n with a good degree of
accuracy from the same experiment.
4 Notes
Nonspecific Binding: Specificity vs. Affinity

Almost any real-life binding experiment will show some low-affinity
binding that is often attributed to “nonspecific binding”. If indirect
methods are used to monitor binding, one may or may not see this
binding step and one must evaluate if the technique being used will
reveal nonspecific binding (i.e., does the detection of binding
require occupancy of a specific site such as in fluorescence resonance
energy transfer techniques). Nonspecific binding usually presents as
an additional slope added to the familiar rectangular hyperbola
apparent at high ligand concentrations and the temptation is to
simply subtract the linear phase from the observed binding to
obtain the specific binding profile. The ambiguity as to whether
this binding is “specific” (but just low affinity) or whether this is
“nonspecific” has, and will, bedevil many studies [7, 16, 21].
Numerous hydrophobic and ionic interactions can lead to nonspe-
cific binding but these may be saturable and show a defined n value
when two large proteins are involved. Because the binding may well
be saturable the linear subtraction of nonspecific binding may not
be appropriate. If we restrict ourselves to consider only two classes
of sites, one tight site binding n1 molecules with affinity Kd1 and a
second weaker site (either due to another specific site or nonspecific
binding) binding n2 molecules with affinity Kd2, then we can
modify Eq. 7 to give:
n1 ½Rt ½L f n2 ½Rt ½L f
½L b ¼ þ : ð16Þ
K d1 þ ½L f K d2 þ ½L f
Direct fitting of the data to this expression will allow assessment of

both classes of sites. If Kd2 [L f] then the second term is approx-
imately linear with [L f] and this is similar to the usual case of
nonspecific binding. But if Kd2 ~ [L f], then the second term will
not be linear. Thus, it is preferable to simply fit the binding as
though there are two different but independent binding sites.
After the data are analyzed with no assumptions one can question
if this interaction occurs at a defined site and in a physiological
range of concentrations and is therefore relevant.
Curve Fitting and Adequacy of the Models
Deviations from the simple binding expressions indicate complexity
such as multiple sites or cooperativity. However, the simplest model
that explains the data is to be preferred. If the data fit a model with
two independent binding sites no better than that with one, the
one site model should be chosen unless there is independent evi-
dence to suggest two sites. Methods of evaluating the goodness of
fitting are beyond the scope of this chapter, but are often available
with available fitting programs and should be evaluated before

proposing a more complicated expression [8].
Procedures and Problems
To summarize, the determination of Kd and n for a protein-protein
interaction requires that we select a technique appropriate for the
binding affinity to be measured. The best concentration of receptor
is near the Kd and the concentration of ligand should be varied
from two orders of magnitude below to two orders of magnitude
above the Kd. The concentration of bound ligand should be deter-
mined as a function of the free ligand and the data should be fit to
the simplest appropriate model. Generally, n can be determined
with a precision of 20 % and Kd within a factor of 2.
Several experimental limitations and errors can limit the accu-
racy and correctness of the observed fits. Common problems
(https://tools.lifetechnologies.com/downloads/FP7.pdf) are:
1. Incorrect correction for nonspecific binding or additional loose
binding sites. The suggested solution is to fit to Eq. 16.
2. Pooling data from experiments with different receptor concen-
trations. This will be a problem if the receptor concentrations
are near Kd. To avoid this, collect enough data from each
titration to do an independent fit and compare the fitted para-
meters from independent determinations.
3. Presence of a non-binding contaminant in the receptor or
labeled ligand. This may be relevant when labeling the ligand
damages the protein, when recombinant proteins are used and
there is undetected heterogeneity due to misfolded protein.
4. Use of a labeling method for the ligand that alters the binding
behavior of that ligand. Use of truncated constructs or incor-
poration of epitope tags or fluorescent labels may be particu-
larly troublesome. Such problems may be revealed if one
compares the apparent affinity from direct experiments using
titration with labeled ligand to experiments where unlabeled
ligand is used to displace labeled ligand.
5. Inadequate number of data points or range of ligand concen-
trations. This is avoided by collecting enough data points,
especially at high ligand concentrations.
References
1. Auerbach D, Thaminy S, Hottiger MO et al. 3. Johnson ML, Straume M (2000) Deriving

(2002) The post-genomic era of interactive complex ligand-binding formulas. Methods
proteomics: facts and perspectives. Proteomics Enzymol 323:155–167
2:611–623 4. Klotz IM (1985) Ligand–receptor interactions:
2. Pawson T, Raina M, Nash P (2002) Interaction facts and fantasies. Q Rev Biophys 18:227–259
domains: from simple binding events to com- 5. Eletr ZM, Huang DT, Duda DM et al. (2005)
plex cellular behavior. FEBS Lett 513:2–10 E2 conjugating enzymes must disengage from
their E1 enzymes before E3-dependent ubiqui- signal originating from a reference macromole-
tin and ubiquitin-like transfer. Nat Struct Mol cule. Application to Escherichia coli replicative
Biol. 12:933–934 helicase DnaB protein nucleic acid interactions.
6. Munson PJ, Rodbard D (1983) Number of Biochemistry 35:2117–2128
receptor sites from Scatchard and Klotz 15. Schwarz G (2000) A universal thermodynamic
graphs: a constructive critique. Science approach to analyze biomolecular binding
220:979–981 experiments. Biophys Chem 86:119–129
7. Swillens S (1995) Interpretation of binding 16. van Zoelen EJ (1992) Analysis of receptor
curves obtained with high receptor concentra- binding displacement curves by a nonhomol-
tions: practical aid for computer analysis. Mol ogous ligand, on the basis of an equivalent
Pharmacol 47:1197–1203 competition principle. Anal Biochem
8. Motulsky HJ, Ransnas LA (1987) Fitting 200:393–399
curves to data using nonlinear regression: a 17. van Zoelen EJ, Kramer RH, van Moerkerk HT
practical and nonmathematical review. FASEB et al (1998) The use of nonhomologous
J 1:365–374 Scatchard analysis in the evaluation of ligand-
9. Tuk B, van Oostenbruggen MF (1996) Solving protein interactions. Trends Pharmacol Sci
inconsistencies in the analysis of receptor- 19:487–490
ligand interactions. Trends Pharmacol Sci 18. van Zoelen EJ, Kramer RH, van Reen MM et al
17:403–409 (1993) An exact general analysis of ligand
10. Koshland DE Jr (1996) The structural basis of binding displacement and saturation curves.
negative cooperativity: receptors and enzymes. Biochemistry 32:6275–6280
Curr Opin Struct Biol 6:757–761 19. Wang ZX (1995) An exact mathematical
11. Forsen S, Linse S (1995) Cooperativity: over expression for describing competitive binding
the Hill. Trends Biochem Sci 20:495–497 of two different ligands to a protein molecule.
12. Hedlund PB, von Euler G (1999) EasyBound – FEBS Lett 360:111–114
a user-friendly approach to nonlinear regres- 20. Brooijmans N, Sharp KA, Kuntz ID (2002)
sion analysis of binding data. Comput Methods Stability of macromolecular complexes. Pro-
Programs Biomed 58:245–249 teins 48:645–653
13. Brown AM (2001) A step-by-step guide to 21. Rovati GE, Rodbard D, Munson PJ (1988)
non-linear regression analysis of experimental DESIGN: computerized optimization of
data using a Microsoft Excel spreadsheet. Com- experimental design for estimating Kd and
put Methods Programs Biomed 65:191–200 Bmax in ligand binding experiments. I.
14. Jezewska MJ, Bujalowski W (1996) A general Homologous and heterologous binding to
method of analysis of ligand binding to com- one or two classes of sites. Anal Biochem
peting macromolecules using the spectroscopic 174:636–649
Chapter 3
Protein-Protein Interaction Databases

Damian Szklarczyk and Lars Juhl Jensen
Abstract
Years of meticulous curation of scientific literature and increasingly reliable computational predictions have
resulted in creation of vast databases of protein interaction data. Over the years, these repositories have
become a basic framework in which experiments are analyzed and new directions of research are explored.
Here we present an overview of the most widely used protein-protein interaction databases and the
methods they employ to gather, combine, and predict interactions. We also point out the trade-off between
comprehensiveness and accuracy and the main pitfall scientists have to be aware before adopting protein
interaction databases in any single-gene or genome-wide analysis.
Key words Protein-protein interactions, Functional associations, Protein-protein interaction

databases, Pathways, Protein-protein interaction prediction, Biochemical pathways, Selection bias
1 Introduction
The continual cost decrease of high-throughput experiments and

the development of computational prediction methods have pro-
duced vast numbers of protein-protein interactions (PPIs). This
ability to provide fairly comprehensive and reliable sets of PPIs
prompted the development of many databases aiming to gather
and unify the available data, each with a different focus and differ-
ent strengths.
PPI databases can be categorized into three broad types: path-
way databases like Reactome [1] and KEGG [2] in which expert
curators collect consensus knowledge, databases of experimentally
verified PPIs like IntAct [3] and BioGRID [4] that collect primary
experimental data, and databases like STRING [5] and GeneMA-
NIA [6] that also include computationally predicted interactions
and text mining but perform no manual curation. Although
databases from the latter category may contain many false positives,
each interaction generally has a confidence score associated with it,
which allows the user to filter out the most likely errors.
39
40 Damian Szklarczyk and Lars Juhl Jensen
Some protein interaction databases are focused on a particular

organism [7–9], disease [10, 11], or type of interaction, e.g.,
kinase–substrate interactions [12]. Other databases, especially
resources with detailed pathway data [1, 2, 13], in addition to
PPIs, include interactions with other macromolecules (mainly
RNA and DNA) and small molecules such as drugs and
metabolites.
Another differentiating feature of PPI databases is the ability to
graphically visualize interaction networks instead of just showing a
list of interaction partners of a query protein. This is not only a
visual gimmick, as the network view gives the user an overview of
the interactions between first (or more) degree neighbors, which in
turn allows for quick visual recognition of highly interconnected
functional modules of proteins. The significance of network visual-
ization has been widely recognized and most databases provide
some kind of network view; if not as a native application working
in a browser, then by providing the network as a downloadable file
that can easily be imported into visualization tools such as Cytos-
cape [14] or NAViGaTOR [15]. These tools and some databases
also allow users to lay out networks, annotate nodes, and perform
various types of network analysis such as clustering and term enrich-
ment analysis (e.g., for Gene Ontology terms [16]). For different
PPI visualization methods, see Fig. 1.
Because of the versatility and diverse set of features, PPI
resources are now commonly used for data analysis, data interpre-
tation, and hypothesis testing. A comprehensive list of more than
300 pathways and interaction databases is available from Pathguide
(www.pathguide.org). However, the extent to which any of the
published PPI datasets reflect the biological interactome is
unknown, and it is thus essential to carefully evaluate the advan-
tages and drawbacks of each interaction data source before using
them. Undoubtedly till now, none can capture the full complexity
of biological systems with different protein variants, modifications,
and spatial and temporal dependencies.
2 The Many Faces of Interactions
Much like the term “function” does not only encompass molecular
functions such as enzymatic catalysis, the term “interaction” in
addition to direct physical binding covers a variety of indirect
links such as complex co-membership, regulatory relationships,
and genetic interactions. These are collectively referred to as “func-
tional associations”; however, the terms “interaction” and
“functional association” are often used interchangeably in the liter-
ature and in databases too.
Protein-Protein Interaction Networks 41
Fig. 1 Default visualizations of JAK1 interactions by different databases. (a) BioGRID HTML table showing
proteins that physically interact with JAK1 (indicated by the yellow color of “experimental evidence”). (b) JAK1
interactions shown in the IntAct “graph viewer” with small molecules being depicted as triangles. The network
view shows only the interactions with JAK1 and not between any other two nodes. (c) JAK1 STRING network.
Each different colored line that connects proteins indicates a separate evidence channel for the particular
interaction, such as text mining (green), experiments (magenta), and databases (blue). (d) InnateDB’s
Cytoscape with Cerebral plug-in view uses “cellular component” GO annotation to lay out proteins based
on their localization inside the cell with the horizontal lines indicating boundaries between different
components
This conceptual mixing of interactions and functions is in part

due to data limitations and in part because the two concepts are
closely related. Many proteins carry out their functions as parts of
complexes. Deciphering the exact topology of a complex requires
data on direct physical binding (e.g., yeast two-hybrid assays).
By contrast, many of the experimental (most notably Tandem
Affinity Purification) and computational methods identify highly
interconnected clusters of proteins that represent complexes or
other functional modules of unknown topology [17].
3 Pathway Databases
Reactome [1] is an open resource of, primarily human, curated

pathway data. Contrary to other interaction databases, Reactome
focuses on the accuracy of rather than the comprehensiveness of
interactomes. Interactions are curated by Ph.D. level curators and
reviewed by an expert from the relevant field. Each of the interac-
tions is annotated in depth; this includes the directionality
(if applicable) of interaction, type, substrates, stoichiometry of the
reaction, its localization, and any known disease associations. In the
default view, the interactive interface gives the user an ability to
select specific pathway, or its part, using a pathway hierarchy
browser, which then highlights its corresponding genes on the
interaction map. Each pathway in Reactome has an in-depth
description, relevant references, and should reflect current expert
consensus. Reactome gives users the ability to conduct pathway
enrichment analyses for an uploaded set of genes, which then can be
browsed in the hierarchical pathway viewer, with pathways color-
coded according to the found enrichment. The database also allows
users to upload expression data for given set of genes, which are
then mapped on the pathways with genes color-coded according to
the user-supplied data.
One of the most widely used pathway map resources is KEGG
(Kyoto Encyclopedia of Genes and Genomes) PATHWAY [2].
With over 1,500 different genomes, it is the most comprehensive
species-wise. Its pathways span different cellular processes ranging
from metabolism to genetic and environmental information pro-
cessing, and each pathway aims to represent the complete knowl-
edge about all existing reactions within it. Each unified manually
curated map is referred to as a reference pathway and is not made
with respect to any one specific species; instead each protein node
represents a group of genes from different species. To create a
species-specific pathway, the genes from each organism are mapped
onto the reference pathway via semiautomatically inferred orthol-
ogy relationships, and the map is color-coded to reflect which parts
of the pathway are present in the particular species. KEGG PATH-
WAY is tightly coupled with other KEGG resources on genes and
genomes (KEGG GENES), functional units (KEGG MODULE),
genetic and environmental perturbations (KEGG DISEASES), and
drugs (KEGG DRUG).
4 Databases of Experimentally Verified PPIs
It is very hard to assess how much knowledge about molecular

interactions there is as most of the information on interactions
exists only in the form of tables, supplementary data files, or free
text in separate research articles, thus in practice rendering it

inaccessible or hard to parse reliably by automatic means. Till this
day manual curation of research articles is the only reliable way to
accurately extract primary interaction data from the literature.
The BioGRID database [4] with its 370,000 unique PPIs is one
of the largest repositories of experimentally verified PPIs, curated
from 35,000 low- and high-throughput experiments covering
40 species. This includes full coverage of the literature for Saccha-
romyces cerevisiae, Schizosaccharomyces pombe, and Arabidopsis thali-
ana, with special focus on conserved networks and pathways. The
database makes it easy to visually differentiate physical from genetic
interactions as well as from high to low-throughput experiments.
Except standard searches by gene name and publication, BioGRID
allows user to construct complex Boolean queries with wildcards.
BioGRID also allows user to search for keywords, sentences, and
authors of abstracts associated with the interactions. Each interac-
tion is annotated with easily discernible annotations such as organ-
ism (color-coded), BAIT/HIT directionality, experimental method
used, and publication. In addition, all high-throughput interac-
tions are annotated by the confidence score used by the author in
the publication, with low-confidence interactions not being
imported into the database. BioGRID does not include a native
network visualization tool, but it allows the user to export data in a
format compatible with Cytoscape [14]. There are also two Cytos-
cape plug-ins that facilitate import of BioGRID data handling both
redundancy and annotations. It is also possible to access BioGRID
via REST API, which allows for building web sites and scripts that
directly communicate with the BioGRID database.
The DIP database [18], like BioGRID, is a manually curated set
of experimentally determined PPIs retrieved from research articles.
It does not focus on a particular organism, though more than 30 %
of its interactions are from S. cerevisiae. The interactions stored in
DIP can be viewed using Cytoscape’s MiSink plug-in [19], which
provides an interactive platform to analyze and visualize DIP data.
Additionally, the web interface of DIP provides services for other
external data analysis tools that allow the user to assess the reliabil-
ity of interactions, for example based on expression profiles [20],
comparison with interactions of paralogous proteins [20] and
protein domain composition of the interacting proteins [21]. DIP
also provides a subset of its database that consists of only signal
transducing ligand–receptor pairs (DLRP) [22].
IntAct [3], a part of the EMBL-EBI database ecosystem, is
another important repository of validated PPIs, both derived from
literature and through user submissions. The database extends the
list of binary interaction through “spoke” automatic co-complexes
expansion, when an accurate list of binary interactions could not be
derived from the experiment. The IntAct database allows the user to
browse and search PPI data based on various categories including
GO terms [16], taxonomy, ChEBI ontology [23], Reactome

pathway associations, mRNA expression, chromosomal location,
and protein domains. The results of search queries can be visualized
through a basic cytoscape plug-in working in a web browser or
opened as a java application on the user’s PC. Each experimentally
inferred binary interaction stored in the IntAct database has an
associated confidence score (MIscore, compliant with PSI-MI
standard), which is cumulative and weighted based on available
evidence. IntAct also includes interactions of proteins with small
molecules associated with the ChEBI dictionary.
The MINT database [24] is a resource of experimentally
validated molecular interactions extracted from peer-reviewed pub-
lications. In addition to protein-protein interactions, it also covers
interactions of proteins with mRNA and genes’ promoter regions.
The MINT database, like IntAct, provides universal confidence
scores, which range from 0 to 1 and are based on the number and
reliability of the interaction’s evidences. The results of the user
query can be viewed in the web browser (MINT Viewer) as a list
or as a modifiable network. The MINT Viewer allows user to
arrange nodes, expand a network, filter links based on confidence,
and show the links between all nodes in the network. The MINT
Viewer gives the ability to view additional information about the
protein, such as synonyms, proteins domains, and diseases associa-
tions. In addition to main database, MINT provides access to sister
databases: HomoMINT [25] that covers human proteins extended
by interactions from orthologous proteins from different species
(the interactions stored in homoMINT are automatically imported
from the main MINT database as soon as relevant interactions are
uploaded); VirusMINT [26], a collection of human–viral protein
interactions integrated with the human protein network; and
DOMINO [27], which is a database of protein interactions
mediated by a wide range of 200 protein-interaction domains.
5 Databases with Distinct Focus
The Human Protein Reference Database [7] is a resource of

literature-mined interactions involving human proteins, small
molecules, and nucleic acids. The HPRD, in addition to interac-
tions, also includes information about protein posttranslational
modifications, domain structures, localization, expression sites
(differentiating between cell lines, normal and disease tissues),
and OMIM [28] disease associations. The database features com-
prehensive browsing capabilities including searching by protein
length, weight, name, chromosome locus, molecular class, post-
translational modifications (PTM), cellular component, domain,
motif, and expression site. The PTM information contains
residue-specific information, such as location, type, and upstream
enzyme. The interactions are divided into direct (binary) and

complex (when the topology of interaction is unknown) types and
each is annotated with experiment type and link to the source
publication. Each protein sequence can be visualized in the build-
in viewer with the domains, motifs, modifications, and
corresponding genomic region highlighted. If available, HPRD
can also provide isoform-specific information for all of the above
annotations. All the proteins are cross-referenced with NetPath
pathway database [29] and Human Proteinpedia with which
HPRD is tightly integrated. HPRD also includes PhosphoMotif
Finder, built upon a literature-derived set of motifs, which reports
found motifs in a submitted protein sequence alongside the
annotations such as matched sequence, matched motif, upstream
kinase, and link to the original publication from which the motif is
derived.
The InnateDB [30] database is a resource for manually curated
PPI and pathway data that focuses primarily on interactions asso-
ciated with innate immunity. The database covers three organisms:
Homo sapiens, Mus musculus, and Bos Taurus. In addition to
18,000 interactions extracted by the InnateDB curators, the data-
base integrates various other resources including MINT, BioGRID,
IntAct, DIP, and BIND [31]. The InnateDB also incorporates
interolog transfer generated by in-house pipeline. Further it
cross-references various pathway databases including but not lim-
ited to KEGG, Reactome, and NetPath. The web interface allows a
user to query, apart from gene and protein names, for interactions
in specific pathways and for interactions according to various
criteria, among others: host system, cell type, tissue type, interac-
tion type, molecule type, and interaction detection method.
InnateDB provides data analysis tools for submitted protein/gene
lists including gene ontology overrepresentation analysis, transcrip-
tion factor biding sites overrepresentation analysis, and pathway
enrichment analysis. The PPIs can be viewed as HTML or
visualized, along with user submitted annotations (i.e., p-values,
expression values), and analyzed using versatile tools including two
Cytoscape plug-ins, CyOog [32] and Cerebral [33].
The Human Immunodeficiency Virus Type 1 (HIV-1), Human
Protein Interaction Database [34] is a resource focused on catalo-
ging all known interactions between HIV-1 and human proteins.
It consists of 2,589 unique interactions targeting 1,448 human
proteins. All these interactions are referenced and annotated with
one of the 42 different types of associations including binds, inhi-
bits, complexes with, upregulates, cleaves, and co-localizes. Although
the primary interface shows only the rudimentary list of interac-
tions along with their type, the user can download a subset or all the
interactions as a tab-separated value file, which contains both the
references to research articles from which the interaction
was curated from and a brief description of each interaction.
Alternatively, the tight coupling of the database with NCBI [35]

allows a user to view each HIV-1 or human protein in NCBI gene
viewer along with its genome location, intron-exon structure,
PubMed references, pathway associations, gene ontology terms,
human–HIV and human–human PPIs from other databases along
with their references to relevant research articles.
6 Interolog Prediction
Homology is evidence of functional similarity. Orthology and

paralogy form the two major types of homologous relationships
between genes and it is widely believed that, from the two, ortho-
logs have greater ability to retain the same function [36]. As a
consequence, most interolog prediction methods solely rely on
ortholog mapping by transferring interaction found between
orthologous genes in different species. Interolog predictions
depend principally on ortholog mapping therefore the quality of
the genome annotation, distance between species and number
of paralogs in both species have a major impact on the confidence
of these predictions.
As interolog prediction can be a reliable source of interaction
information, especially between closely related genomes where
orthology mapping could be accurately resolved, some of the data-
bases of manually curated interaction incorporate these predictions
as an additional source of interaction. The most obvious case is
KEGG PATHWAY where the concept of using reference pathways
depends solemnly on ortholog mapping. Reactome on the other
hand does not try to predict interactions but instead allows a user to
compare and visualize (only on human maps) pathway coverage
from 20 selected genomes.
Some of the databases of experimentally validated interactions
such as homoMINT and InnateDB also utilize interolog predic-
tions, but I2D (Interologous Interaction Database) [37] stands out
as it incorporates and transfers several major repositories of interac-
tion data. The data stored in the I2D database can be categorized
into two sets: the combined literature-derived human interaction
from DIP, MINT, HPRD, and BIND databases; and computa-
tional prediction of interlogs between Homo Sapiens, Saccharomyces
cerevisiae, Caenorhabditis elegans, Drosophila melanogaster, and
Mus musculus. The interologs are inferred using a custom built
pipeline using a best-hit approach and evaluated based on
co-expression, Gene Ontology terms, and domain pairs
co-occurrence. The results of the queries can be viewed either as
an HTML table, custom graph viewer, or exported to NAViGa-
TOR software [15]. In table view, each pair of interactions is
annotated with additional supporting information, including
domain co-occurrence, protein co-localization, co-mentioning in
abstracts, expression correlation, and GO similarity. The java-based

custom graph viewer allows for visual modification of the network,
including changing the size, spread, opacity, and shape of the
nodes. NAViGaTOR software is a comprehensive network analysis
and visualization tool developed in the same lab as I2D, but it has
the ability to import data of various formats including BioPAX
(www.biopax.org) and PSI [38].
7 Predicted PPIs
It has been estimated that human proteins could give rise to as

many as 200,000–300,000 direct interactions [39] and the most
comprehensive databases to date, HPRD and BioGRID, both list a
little more than 30,000 direct unique binary human protein
interactions (see Table 1). Although the total number of experimen-
tally validated associations (not only physical interactions) in all
databases specialized in manual curation of literature is fivefold
Table 1
Comparison of different interaction databases
Number of
Total number interactions Number Built in
of interactions in human of Computational network Unified
Database (thousands) (thousands) species predictions view scoring IMEx
BioGRID 372 75,9 40 No No No Yes (C)
DIP 68 3,4 485 No Yes No Yes
GeneMANIA 96208 24220,1 7 Yes Yes Yes No
HPRD 37 37,1 1 No No No Yes
I2D 666 152,4 6 Yes (A) No No Yes
InnateDB 104 84,0 3 Yes (A) Yes No Yes
IntAct 2331 44,0 397 No Yes Yes Yes
MINT 95 21,3 434 Yes (A, B) Yes Yes Yes
PIPs 78 78,4 1 Yes No Yes No
STRING 224346 1540,7 1133 Yes Yes Yes No
Reactome 2161 121,9 28 No Yes No No
KEGG (D) 5880 63,3 1509 Yes (B) Yes No No
Binary protein interaction count does not include self-associations. For databases focused on computational predictions,
the total number of interactions stored is not a valid predictor of quality of comprehensiveness of the database as most of
the interactions are low-scoring and not relevant for noncomputational analysis. A only interologs prediction, B only
homoMINT, C observing member. D the version of KEGG PATHWAY included in this breakdown is from July 2011,
which was the last set available from KEGG under free license
higher, the overall number of existing associations inside the

human cell could be at least a magnitude larger. In order to get a
more complete picture of any interactome, it will be necessary to
augment the existing knowledge with computational prediction.
In the late 1990s, access to the quickly expanding fields of
genomics, transcriptomics, and proteomics gave researchers data-
sets needed for the development of new computational methods
for prediction of functional associations. Some of the developed
methods, most notably interolog predictions, can infer interac-
tomes for species for which the only existing data is a sequenced
genome. These methods also gather more and more focus as it is
becoming effectively impossible to acquire experimentally inferred
interactomes with the pace new genomes are being sequenced.
Although computational predictions could contain many false posi-
tives, it is common that each computationally predicted interaction
is annotated with a confidence score. Filtering to include only the
highest scoring associations can thus yield interaction networks
with a very low false-positive rate.
Access to fully sequenced and annotated genomes allows for
the identification of protein fusion events. If such an event
occurred, in one or more lineages, different parts (domains) of
one fused protein could be found as being separate full-length
proteins in other species [40]. Therefore, the existence of a fusion
event is a strong predictor that, prior to it, the proteins were also
functionally associated [41].
Another method of predicting protein-protein interaction from
genomic data is based on phylogenetic profiles. This method relies
on the assumption that genes that interact with each other or
function together tend to be inherited together during speciation
events. Therefore, genes that are functionally associated should
exhibit similar patterns of absence and presence across many
sequenced genomes [42].
If, in addition to the gene sequence data, we have knowledge
about location and synteny of genes, we can utilize this as another
source for PPI predictions. Both in prokaryotes and in eukaryotes,
genes located near each other in the genome often form operons—
a cluster of genes which is under a single regulatory signal.
The conservation of gene order across several genomes [43] and
the concurrent directionality of transcription [44] of genes located
within these clusters point to their functional associations.
In order to computationally predict protein interactions,
researchers can also utilize transcriptomics data from various
RNA-seq or microarray experiments. A single experiment has the
ability to cover a considerable part of the genome, which provides
researchers an unparalleled look into the regulation of proteins
across various species and conditions. Similar to gene neighborhood
conservation, the co-expression of two proteins is considered to be
an indication of co-regulation and, consequently, functional
association [45]. Leveraging, for example, the vast library of over

30,000 different experiments stored in the GEO database [46]
researchers can explore co-expression across numerous experimen-
tal conditions.
Only 0.0015 % of all research articles in the MEDLINE
database have been covered by manual curation; therefore research-
ers have started to look into automated means of extracting PPIs
from literature. Automatic text mining methods can be divided into
two major categories: co-occurrence/frequency-based methods
and natural language processing (NLP). Co-occurrence text
mining purely relies on matching exact elements (tokens) from
the specified dictionary without taking into account the sequence
in which the tokens appear or—to some extent—if the tokens
are located in the same sentence. On the other hand, for NLP
these characteristics of the text are essential and the interaction
can be mined only when the two proteins are mentioned in the
same sentence. Although NLP does not require repeated
co-occurrence of same links in order to mine high-confidence
interactions, this advantage is counterbalanced by very low sensi-
tivity of NLP methods due to the sentence constrain [47].
8 Databases of Predicted PPIs
GeneMANIA [6] is a database of known and predicted functional

associations between proteins that covers seven major
organisms (D. melanogaster, R. norvegicus, S. Cerevisiae, C. elegans,
M. Musculus, A. Thaliana, and H. Sapiens). The physical and
genetic interactions included in GeneMANIA are assembled from
datasets stored in Pathway Commons (which consists of various
other resources including IMEx consortium databases and Bio-
GRID). The predicted associations include: interologs predictions
(mostly from I2D), predictions based expression profiles from
experiments stored in GEO [46], and predictions based on shared
domain composition. The distinctive feature of GeneMANIA is
that every dataset incorporated in GeneMANIA could be regarded
as a separate interaction network. The confidence scores (weights)
of links from each network are based on the raw score from the
particular set. The resulting final network consists of these scores
combined and weighted accordingly to the confidence associated
with each data source. The weights from all the links for each of the
data sources sum to 100 %, and the user has an ability to choose one
of seven different weighting methods including query- and
annotation-dependent methods. Using a friendly interface, the
user also has full control over which dataset or prediction method
should contribute to the network and can turn them on/off
accordingly. GeneMANIA automatically finds all the Gene Ontol-
ogy annotations of proteins in the network and assesses their false
discovery rate and their network coverage. It also has a function to

color-code the network node according to one or more GO terms.
The user has the ability to query GeneMANIA with more than one
protein or to upload its own interaction sets, along with confidence
scores, that will be incorporated with the GeneMANIA network
and framework.
Search Tool for the Retrieval of Interacting Genes/Proteins
(STRING) [5] is another database that incorporates both known
and predicted functional associations between proteins. It covers
1,133 fully sequenced genomes across all three domains of life.
It integrates several different kinds of sources: heavily curated
interactions stored in pathway databases such as Reactome and
KEGG; manually curated low and high-throughput experiments
from IMEx consortium databases and BioGRID; interactions
retrieved using the co-occurrence method from text mining of
20,000,000 PubMed abstracts; computational prediction of PPI
by correlating mRNA expression profiles across various experi-
ments stored in the GEO database; interactions predicted from
genomic context, such as, gene-fusion, gene neighborhood, and
phylogenetic profiles, which does not depend on prior knowledge
and could be generated for any species with a fully sequenced
genome; and interolog predictions for all evidence sources and all
species in the STRING database. All known and predicted links are
annotated with a confidence score (from 0 to 1), and scores origi-
nating from different sources are added in a probabilistic manner in
order to create a final network. The protein view contains basic
information about a protein including sequence, domain composi-
tion, 3D structure, and homologs in a selected species. The
STRING network can be recomputed based on user-chosen set
sources and cut-offs as well as expended based on user-selected
proteins or globally with the most confident links connected to
the nodes in a visible network. The interactive STRING network
view gives the user basic network analysis tools, such as network
clustering, protein-protein interaction enrichment, pathway
enrichment, protein domains enrichment, and Gene Ontology
terms enrichment. The STRING database is closely coupled with
its sister databases: STITCH [48], which, in addition to proteins,
covers small molecules and eggNOG [49], the database of ortho-
logous relationships between STRING’s proteins.
The Human protein-protein Interaction Prediction [50]
(PIPs) database is a resource containing human protein–protein
functional associations derived by purely computational methods.
Each link in the database is annotated with a score representing a
likelihood ratio of the particular interaction occurring. This ratio
is evaluated based on HPRD interactions as a positive set and a
100 times larger negative set constructed from random associations
between HPRD proteins. The interactions were calculated using
six different methods: co-expression; interacting domain
co-occurrence; interologs mapping; posttranslational modifications

co-occurrence; cell compartment protein co-localization; and a
network topology method based on the hypothesis that the more
interaction partner proteins share, the more likely that they interact
with each other. The confidence scores from different prediction
algorithms are combined using the Bayesian method, with a score
greater than 1 denoting that the interaction is more likely to
occur than not. The results of a user query are presented as an
HTML table with annotation on fractional scores and if the partic-
ular interaction was seen in other databases (BIND, DIP, HPRD,
and I2D).
9 PPI Database Federation
Database federation, which is the transparent interoperation of

several databases, can be viewed as an alternative to resources like
GeneMANIA and STRING that aim to integrate everything in one
place. The early collaboration between major PPI database provi-
ders, which resulted in the development of the MIMIx curation
standard and common data structure (PSI-MI XML and MITAB),
grew to become the International Molecular Exchange (IMEx)
consortium. As of 2012 the IMEx consortium comprised of the
following members and partners: DIP [18], IntAct [3], MINT
[24], MatrixDB [51], MPIDB [52], Molecular Connections
(http://www.molecularconnections.com), I2D [37], InnateDB
[53], SwissProt group [54], and as an observing member BioGRID
[4]. In addition to common curation guidelines and data struc-
tures, the member databases now use common controlled vocabul-
aries (PSI-MI ontologies) to describe, for example, the type of
experiment. All of these efforts make the datasets from different
IMEx databases relatively easy to compare in an automated way.
There is also an ongoing endeavor between IMEx members to
release the full, nonredundant, set of guideline compliant protein
interactions. Though, as of 2012, most of the participating data-
bases still have data to be released to the IMEx consortium dataset,
and integrating different databases is still necessary to get all the
available interactions (see Fig. 2). There is no central repository of
the IMEx interaction set, and the associated records exist only in
their source databases tagged respectively to indicate classification
as being a part of the IMEx dataset. To address this, the IMEx
consortium developed the PSI Common Query Interface
(PSICQUIC) [55], which gives users an ability to query all data-
bases simultaneously and presents the result as a list with the option
to cluster the results in order to create a nonredundant set of
interactions. PSICQUIC View incorporated within EBI web
services allows a user to query databases that are outside the
IMEx consortium, such as STRING and Reactome.
Fig. 2 Overlap of binary human protein interactions between selected IMEx

databases. The total count of interactions includes only the interactions between
proteins for which identifiers could be mapped onto a common set of human
ENSEMBL identifiers
10 Data Integration
The individual interaction sets derived through high-throughput

techniques are not only incomplete but also have relatively high
false-positive rates [56]. In addition, it has been estimated that the
nonfunctional direct biophysical interactions that are found
through Yeast 2-Hybrid assays could constitute to as much as
19 % of all interactions from these experiments [57]. These effects
alone can explain the often poor overlap between different interac-
tion studies. Not one method could create a complete picture of an
interactome; therefore, in order to obtain the closest approxima-
tion of it from existing data, we need to integrate different data
sources from experimental and computational methods.
The simplest way to integrate different high- and low-
throughput datasets is to combine all mined interactions in a list.
Especially in the case of manually curated high-throughput
interaction sets, this is often done with the accompanying reliability
annotation of a particular set either derived from the experiment
itself or annotated based on the method used. This method,
although simple, does have the advantage of not omitting any
supposedly relevant data and is most suitable in situations when a
researcher is interested in a very limited set of proteins and where all
interaction evidence can thus be assessed manually. The drawback
of such an approach is the more limited utility for high-throughput
network analysis without extensive postprocessing.
Another common way to integrate different data, employed by
several databases, is to compound confidence scores for links for
which there are more than one line of evidence. As a result, the final
compounded link would have a higher confidence score than any of
its contributing evidences individually. This could be done by

assigning one fixed confidence score for each link from a particular
data source—this is predominantly done for sources for which the
links cannot be sorted in a meaningful way that would correlate well
with their confidence, e.g., the combined list of interactions from
many small-scale experiments. Alternatively, one can calibrate con-
fidence scores based on raw scores from experiments or predictions;
these raw scores can, for example, be correlation coefficients
between the expression levels of different genes or text mining
co-occurrence frequencies.
This compounded confidence integration is predominately
done for combining different computational predictions. This is
due to the fact that the majority of links have low confidence levels
and only when sources are compounded can a more applicable
network be formed. If the biases of the prediction methods are
not correlated, it is beneficial to have several sources as it would
dilute the individual biases of any single source.
11 Bias
The story that is emerging from protein-protein interaction

networks forms a consistent view of scale-free networks [58], with
essential, well-conserved proteins being the hubs that give rise to
robustness in the networks [59]. Indeed, this is undoubtedly true
for the data, but it does not necessarily reflect the underlying
biological system. This discrepancy happens due to biases connected
to the data, as any experiment or prediction method is subject to
different unavoidable biases either linked to the experimental
method [60] or sampling bias caused by research interest [61].
The majority of research articles report an interaction only
when it has been observed (either by high- or low-throughput
methods) and do not report when an interaction has been exam-
ined but the result was negative. Due to this reporting bias, the
integration of the subsequent data can only add new interactions
rather than refine the network by eliminating false positives. As a
consequence of inherent false-positive rates and imperfect repro-
ducibility of data, it is thus to be expected that the compounded
false-positive rate should be higher for well-studied pathways or
proteins. Indeed, this appears to be the case as it has been shown
that the popularity of a field correlates with a higher false-positive
rate of interactions [61]. There is little doubt that to some degree
the same effect of selection bias applies, among others, to disease-
associated proteins [62], evolutionary conserved proteins [63], and
highly expressed proteins [64]. Of course a single high-throughput
experiment may not be subjected to selection bias; nonetheless the
experiment itself can contain different sorts of biases, i.e., the yeast
two-hybrid method forces the two proteins to localize to nucleus,
which for some types of proteins (in particular membrane proteins)

may cause them to misfold and aggregate, resulting in a lower
detection rate [65]. Furthermore, differences in the expression
level of proteins may influence the detection rate as well, produc-
ing, already mentioned, bias towards higher expressed proteins.
As for now, there is no agreement of how accurate protein‐
protein interaction networks represent true biological interactomes.
There are possibilities to increase reliability of known methods, i.e.,
by releasing raw, unprocessed data in order to refine networks by
incorporating true negatives. Till then, researchers have to rely on
their own assessment of biases and take them into account when
inferring any knowledge based on protein interaction networks.
References
1. Croft D, O’Kelly G, Wu G, Haw R et al (2011) 10. Goodman N, McCormick K, Goldowitz D,
Reactome: a database of reactions, pathways Hockly E et al (2003) Plans for HDBase—a
and biological processes. Nucleic Acids Res research community website for Huntington’s
39:D691–D697 Disease. Clin Neurosci Res 3:197–217
2. Kanehisa M, Goto S, Furumichi M, Tanabe M 11. Lechner M, Höhn V, Brauner B, Dunger I et al
et al (2010) KEGG for representation and anal- (2012) CIDeR: multifactorial interaction net-
ysis of molecular networks involving diseases works in human diseases. Genome Biol 13:R62
and drugs. Nucleic Acids Res 38:D355–D360 12. Dinkel H, Chica C, Via A, Gould CM et al
3. Kerrien S, Aranda B, Breuza L, Bridge A et al (2011) Phospho.ELM: a database of phos-
(2012) The IntAct molecular interaction data- phorylation sites–update 2011. Nucleic Acids
base in 2012. Nucleic Acids Res 40: Res 39:D261–D267
D841–D846 13. Caspi R, Foerster H, Fulcher CA, Kaipa P et al
4. Stark C, Breitkreutz B-J, Chatr-Aryamontri A, (2008) The MetaCyc Database of metabolic
Boucher L et al (2011) The BioGRID Interac- pathways and enzymes and the BioCyc collec-
tion Database: 2011 update. Nucleic Acids Res tion of Pathway/Genome Databases. Nucleic
39:D698–D704 Acids Res 36:D623–D631
5. Szklarczyk D, Franceschini A, Kuhn M, Simo- 14. Smoot ME, Ono K, Ruscheinski J, Wang P-L
novic M et al (2011) The STRING database in et al (2011) Cytoscape 2.8: new features for
2011: functional interaction networks of pro- data integration and network visualization.
teins, globally integrated and scored. Nucleic Bioinformatics 27:431–432
Acids Res 39:D561–D568 15. Brown KR, Otasek D, Ali M, McGuffin MJ
6. Warde-Farley D, Donaldson SL, Comes O, et al (2009) NAViGaTOR: Network Analysis,
Zuberi K et al (2010) The GeneMANIA pre- Visualization and Graphing Toronto. Bioinfor-
diction server: biological network integration matics 25:3327–3329
for gene prioritization and predicting gene 16. Gene T., Consortium O. (2010) The Gene
function. Nucleic Acids Res 38:W214–W220 Ontology Consortium in 2010: extensions and
7. Goel R, Harsha HC, Pandey A, Prasad TSK refinements. Nucleic Acids Res 38:D331–D335
(2012) Human Protein Reference Database 17. Hakes L, Robertson DL, Oliver SG (2005)
and Human Proteinpedia as resources for phos- Effect of dataset selection on the topological
phoproteome analysis. Mol Biosyst 8:453–463 interpretation of protein interaction networks.
8. Cherry JM, Hong EL, Amundsen C, Balakrish- BMC Genomics 6:131
nan R et al (2012) Saccharomyces Genome 18. Salwinski L, Miller CS, Smith AJ, Pettit FK et al
Database: the genomics resource of budding (2004) The Database of Interacting Proteins:
yeast. Nucleic Acids Res 40:D700–D705 2004 update. Nucleic Acids Res 32:
9. Murali T, Pacifico S, Yu J, Guest S et al (2011) D449–D451
DroID 2011: a comprehensive, integrated 19. Salwinski L, Eisenberg D (2007) The MiSink
resource for protein, transcription factor, Plugin: cytoscape as a graphical interface to the
RNA and gene interactions for Drosophila. Database of Interacting Proteins. Bioinformat-
Nucleic Acids Res 39:D736–D743 ics 23:2193–2195
20. Deane CM, Salwiński Ł, Xenarios I, Eisenberg 33. Barsky A, Gardy JL, Hancock REW, Munzner
D (2002) Protein interactions: two methods T (2007) Cerebral: a Cytoscape plugin for lay-
for assessment of the reliability of high out of and interaction with biological networks
throughput observations. Mol Cell Proteomics using subcellular localization annotation. Bio-
1:349–356 informatics 23:1040–1042
21. Deng M, Mehta S, Sun F, Chen T (2002) 34. Fu W, Sanders-Beer BE, Katz KS, Maglott DR
Inferring domain-domain interactions from et al (2009) Human immunodeficiency virus
protein-protein interactions. Genome Res type 1, human protein interaction database at
12:1540–1548 NCBI. Nucleic Acids Res 37:D417–D422
22. Graeber TG, Eisenberg D (2001) Bioinfor- 35. Resource Coordinators NCBI (2013) Data-
matic identification of potential autocrine sig- base resources of the National Center for Bio-
naling loops in cancers from gene expression technology Information. Nucleic Acids Res 41:
profiles. Nat Genet 29:295–300 D8–D20
23. Hastings J, de Matos P, Dekker A, Ennis M 36. Chen R, Jeong SS (2000) Functional predic-
et al (2013) The ChEBI reference database tion: identification of protein orthologs and
and ontology for biologically relevant chemis- paralogs. Protein Sci 9:2344–2353
try: enhancements for 2013. Nucleic Acids Res 37. Niu Y, Otasek D, Jurisica I (2010) Evaluation
41:D456–D463 of linguistic features useful in extraction of
24. Ceol A, Chatr Aryamontri A, Licata L, Peluso interactions from PubMed; application to
D et al (2010) MINT, the molecular interac- annotating known, high-throughput and pre-
tion database: 2009 update. Nucleic Acids Res dicted interactions in I2D. Bioinformatics
38:D532–D539 26:111–119
25. Persico M, Ceol A, Gavrila C, Hoffmann R et al 38. Hermjakob H, Montecchi-Palazzi L, Bader G,
(2005) HomoMINT: an inferred human net- Wojcik J et al (2004) The HUPO PSI’s molec-
work based on orthology mapping of protein ular interaction format–a community standard
interactions discovered in model organisms. for the representation of protein interaction
BMC Bioinformatics 6(Suppl 4):S21 data. Nat Biotechnol 22:177–183
26. Chatr-aryamontri A, Ceol A, Peluso D, Nar- 39. Hart GT, Ramani AK, Marcotte EM (2006)
dozza A et al (2009) VirusMINT: a viral pro- How complete are current yeast and human
tein interaction database. Nucleic Acids Res 37: protein-interaction networks? Genome Biol
D669–D673 7:120
27. Ceol A, Chatr-aryamontri A, Santonico E, 40. Burns DM, Horn V, Paluh J, Yanofsky C
Sacco R et al (2007) DOMINO: a database of (1990) Evolution of the tryptophan synthetase
domain-peptide interactions. Nucleic Acids of fungi. Analysis of experimentally fused
Res 35:D557–D560 Escherichia coli tryptophan synthetase alpha
28. Amberger J, Bocchini C, Hamosh A (2011) A and beta chains. J Biol Chem 265:2060–2069
new face and new challenges for Online Men- 41. Enright AJ, Iliopoulos I, Kyrpides NC, Ouzou-
delian Inheritance in Man (OMIM®). Hum nis CA (1999) Protein interaction maps for
Mutat 32:564–567 complete genomes based on gene fusion
29. Kandasamy K, Mohan SS, Raju R, Keerthiku- events. Nature 402:86–90
mar S et al (2010) NetPath: a public resource 42. Marcotte EM, Pellegrini M, Ng HL, Rice DW
of curated signal transduction pathways. et al (1999) Detecting protein function and
Genome Biol 11:R3 protein-protein interactions from genome
30. Breuer K, Foroushani AK, Laird MR, Chen C sequences. Science 285:751–753
et al (2013) InnateDB: systems biology of 43. Dandekar T, Snel B, Huynen M, Bork P (1998)
innate immunity and beyond–recent updates Conservation of gene order: a fingerprint of
and continuing curation. Nucleic Acids Res proteins that physically interact. Trends Bio-
41:D1228–D1233 chem Sci 23:324–328
31. Bader GD, Donaldson I, Wolting C, Ouellette 44. Overbeek R, Fonstein M, D’Souza M, Pusch
BF et al (2001) BIND–The Biomolecular GD et al (1999) Use of contiguity on the chro-
Interaction Network Database. Nucleic Acids mosome to predict functional coupling. In
Res 29:242–245 Silico Biol 1:93–108
32. Royer L, Reimann M, Andreopoulos B, 45. Eisen MB, Spellman PT, Brown PO, Botstein
Schroeder M (2008) Unraveling protein net- D (1998) Cluster analysis and display of
works with power graph analysis. PLoS Com- genome-wide expression patterns. Proc Natl
put Biol 4:e1000108 Acad Sci U S A 95:14863–14868
46. Barrett T, Troup DB, Wilhite SE, Ledoux P PSISCORE: accessing and scoring molecular
et al (2011) NCBI GEO: archive for functional interactions. Nat Methods 8:528–529
genomics data sets–10 years on. Nucleic Acids 56. Sambourg L, Thierry-Mieg N (2010) New
Res 39:D1005–D1010 insights into protein-protein interaction data
47. Hirschman L, Park JC, Tsujii J, Wong L et al lead to increased estimates of the S cerevisiae
(2002) Accomplishments and challenges in lit- interactome size. BMC Bioinformatics 11:605
erature data mining for biology. Bioinformatics 57. Nakayama M, Kikuno R, Ohara O (2002)
18:1553–1561 Protein-protein interactions between large
48. Kuhn M, Szklarczyk D, Franceschini A, von proteins: two-hybrid screening using a func-
Mering C et al (2012) STITCH 3: zooming tionally classified library composed of long
in on protein-chemical interactions. Nucleic cDNAs. Genome Res 12:1773–1784
Acids Res 40:D876–D880 58. Jeong H, Tombor B, Albert R, Oltvai ZN et al
49. Powell S, Szklarczyk D, Trachana K, Roth A (2000) The large-scale organization of meta-
et al (2012) eggNOG v3.0: orthologous bolic networks. Nature 407:651–654
groups covering 1133 organisms at 41 differ- 59. Wuchty S, Oltvai ZN, Barabási A-L (2003)
ent taxonomic ranges. Nucleic Acids Res 40: Evolutionary conservation of motif constitu-
D284–D289 ents in the yeast protein interaction network.
50. McDowall MD, Scott MS, Barton GJ (2009) Nat Genet 35:176–179
PIPs: human protein-protein interaction pre- 60. Von Mering C, Krause R, Snel B, Cornell M
diction database. Nucleic Acids Res 37: et al (2002) Comparative assessment of large-
D651–D656 scale data sets of protein-protein interactions.
51. Chautard E, Fatoux-Ardore M, Ballut L, Nature 417:399–403
Thierry-Mieg N et al (2011) MatrixDB, the 61. Ioannidis JPA (2005) Why most published
extracellular matrix interaction database. research findings are false. PLoS Med 2:e124
Nucleic Acids Res 39:D235–D240 62. Tang JL (2005) Selection bias in meta-analyses
52. Goll J, Rajagopala SV, Shiau SC, Wu H et al of gene-disease associations. PLoS Med 2:e409
(2008) MPIDB: the microbial protein interac- 63. Pál C, Papp B, Hurst LD (2003) Genomic
tion database. Bioinformatics 24:1743–1744 function: rate of evolution and gene dispens-
53. Lynn DJ, Winsor GL, Chan C, Richard N et al ability. Nature 421:496–497, discussion 497–8
(2008) InnateDB: facilitating systems-level 64. Bloom JD, Adami C (2003) Apparent depen-
analyses of the mammalian innate immune dence of protein evolutionary rate on number
response. Mol Syst Biol 4:218 of interactions is linked to biases in protein-
54. The UniProt Consortium (2011) Ongoing and protein interactions data sets. BMC Evol Biol
future developments at the Universal Protein 3:21
Resource. Nucleic Acids Res 39:D214–D219 65. Brito GC, Andrews DW (2011) Removing bias
55. Aranda B, Blankenburg H, Kerrien S, Brink- against membrane proteins in interaction net-
man FSL et al (2011) PSICQUIC and works. BMC Syst Biol 5:169
Chapter 4
Computational Prediction of Protein-Protein Interactions

Tobias Ehrenberger, Lewis C. Cantley, and Michael B. Yaffe
Abstract
The prediction of protein-protein interactions and kinase-specific phosphorylation sites on individual
proteins is critical for correctly placing proteins within signaling pathways and networks. The importance
of this type of annotation continues to increase with the continued explosion of genomic and proteomic
data, particularly with emerging data categorizing posttranslational modifications on a large scale. A variety
of computational tools are available for this purpose. In this chapter, we review the general methodologies
for these types of computational predictions and present a detailed user-focused tutorial of one such
method and computational tool, Scansite, which is freely available to the entire scientific community over
the Internet.
Key words Scansite, Protein-protein interaction prediction, Sequence motif, PSSM, Binding motif,
Phosphorylation sites, Bioinformatics
1 Introduction
Decades of research in molecular biology have resulted in the

availability of vast amounts of data, including genomic sequences,
protein sequences, structural data, and protein metadata including
functional domain information and interaction data. Unfortu-
nately, the availability of these data types does not necessarily result
in a clear understanding of what all the data means in a broader
context. The bulk of the available data is single molecule-centric,
limiting our ability to understand how molecules are integrated
into pathways and networks. With the advent of new experimental
techniques, it is possible to enrich these pieces of data with addi-
tional information related to protein-protein interactions and enzy-
me–substrate relationships. One of the most important
breakthroughs in this context was the rise of experimental techni-
ques that allowed the rapid and large-scale detection of protein-
protein interactions [1]. Since the molecular apparatus of a cell is
mainly controlled by protein–protein and protein–nucleic acid
interactions, detecting and understanding such events, particularly
57
58 Tobias Ehrenberger et al.
direct interactions, is the first step to a broader view of biological

systems. Interaction information has been collected in a number of
different public databases [2, 3], and the information stored in
these databases mostly contains experimentally verified informa-
tion, i.e., data from in vivo or in vitro experiments. Unfortunately,
given the current trend towards large-scale proteome wide analyses
and the fact that these databases are far from complete, this infor-
mation often proves insufficient for many analyses. The missing
pieces in the puzzle that elucidates a more complete view of the
cell interactome can be provided by interaction prediction tools.
These tools create in silico predictions of protein-protein interac-
tions and kinase–substrate relationships, are typically inexpensive
and fast compared to conventional time- and resource-intensive
experimental methods, and can provide a focused list of predictions
that can then be verified or refuted by further focused experimental
testing.
Over the past years, a number of different computational
approaches for predicting protein-protein interactions have been
developed. These techniques can be divided into those that are
based on a single biological feature and those that attempt to use
a range of different features and data types and can therefore be
categorized based on the types of data that they use. A detailed
overview of how each of these approaches works, and what the
shortcomings of these methods are, can be found elsewhere [4, 5],
but a short summary is provided here, focusing on which general
features are used to predict protein-protein interactions.
At one extreme are methods based on a protein’s three-
dimensional structure, generally referred to as “protein docking”
techniques. Given a 3D model (usually based on high-resolution
data from X-ray crystallography or NMR experiments, deposited in
the Protein Data Bank [1]) for two potentially interacting proteins,
the best fit for each potential interaction interface on the surface of
these models can be searched for and scored [6]. However, finding
low-energy fits is very challenging, often due to the static nature of
the PDB structures and the dynamic plasticity that can occur at
protein-protein interfaces. Thus, conformational changes, the
arrangements of side chains, and the energy levels of a potential
conformation combination and interaction, potential posttransla-
tional modifications that may or may not be included in a model,
and a number of other factors have to be considered. Because of the
large variance in the quality of the prediction of different methods
in this field, CAPRI (Critical Assessment of Protein Interactions), a
community-based program that regularly evaluates the algorithms
to predict protein-protein interactions based on structures in a
double-blind manner, has been initiated [7]. Alternatively, machine
learning methods can be developed, usually based on databases of
experimentally verified interactions and a number of additional
biological properties. These points of data are then used to train a
Protein-Protein Interactions using Scansite 3 59
prediction engine based on known data [8]. The problem with this
approach—as with any other machine learning approach used in
this manner—is that the resulting predictor does not provide
easily decipherable information about why the proteins are likely
to interact. This means that, although it may yield useful results, it
is hard to reconstruct and understand exactly why a prediction is
made by this black-box predictor. A very specific type of machine
learning method tries to find a pattern based on features at the
interaction interface of the proteins involved. A method closely
related to this approach is described later in this chapter. Other
prediction methods are based on genomics. Gene fusion methods
predict that discrete proteins are likely to interact if their homo-
logues are fused into single genomic entities in other species.
Other techniques based on gene neighborhood conservation are
built on the hypothesis that gene pairs within such neighborhoods
that are evolutionary conserved across different species are likely to
interact.
No matter which method is used, it is important to keep the
caveats of the method in mind. First, no prediction can guarantee
either biological correctness or relevance. This is especially impor-
tant if prediction tools are used to design and plan further experi-
ments without first confirming the initial prediction. Failure of a
method to predict an interaction may not reflect a fundamental
problem with the method but may instead reflect limitations of the
data that the method is based on. The data, be it experimentally
verified interaction sites, 3D data of proteins, or other information,
originates from experiments which are all error-prone, though in
some cases the extent of the error may be difficult to estimate. This
also applies to methods that use machine learning to train a predic-
tor, as these methods are highly dependent on the quality of the
underlying training dataset. Obviously, a large set of training data is
necessary to create a good predictor and, indeed, large databases of
experimentally verified protein-protein interactions are now avail-
able. However, training a predictor also requires a negative dataset
that provides information of what interactions are very unlikely to
happen. Experimental data of this type are typically not published, at
least in part due to difficulties in distinguishing whether the lack of an
observed interaction is the result of a technically failed experiment
or because there is no biologically relevant interaction [9]. The
end result is a lack of reliable negative training data for computational
method development. The quality and nature of the training data
should therefore be one important consideration in the user’s choice
of whether to trust a predictor trained on these kinds of data types,
including the species of the proteins in the training dataset, the type of
experiments used to verify the sites, and of course the number of sites
and proteins included. Thus, it is very important for prediction tools
to explicitly (1) give information about how the method works
and what information it uses, (2) provide some type of quantitative
measure that allows users to compare different results and distinguish

between good and not-as-good predictions, and (3) provide any
additional information that helps the user decide whether to trust
the predictions. This information could be incorporated into the
prediction method itself but is also very helpful if this type of metadata
is simply presented for the user to examine independently of whether
this information is explicitly used in the prediction algorithm.
One of the most important features in describing protein-protein
interactions is elucidating the exact sites on the proteins where
the interaction occurs, either at the detailed atomic/structural level
or at the level of specific amino acid sequences. That is, on the surface
of the protein, which part of the amino acid sequence directly con-
tacts or indirectly influences the interaction partner? By focusing
solely on sequence information, the complexities required for inter-
action prediction by docking-type simulations (conformational states
of side chains, energetic contributions, etc.) are radically reduced.
Since protein-protein interactions are mediated by attractive
forces based on the physicochemical properties of amino acids, in
many cases it is sufficient to describe potential interaction partners
by amino acid sequence patterns alone. This can be clearly shown
by considering kinase–substrate interactions: kinases generally
only phosphorylate serine (S), threonine (T), or tyrosine (Y)
residues based on the ability of their phosphate acceptor hydroxyl
groups to nucleophilically attack the γ-phosphate of ATP. How-
ever, more than 500 different kinases are known alone in humans,
each of which targets a different set of substrates [10]. The specific
site of phosphorylation is therefore not the only amino acid that
plays a role in substrate recognition. Instead, 4–12 amino acid
residues on the substrate flanking the phospho-acceptor likely
physically contact a kinase’s active site [11] and help to position
the substrate for an in-line attack on the phosphate while simulta-
neously optimizing the geometry of the kinase’s catalytic machin-
ery to facilitate stabilization of the resulting transition state. This
indicates that this part of the substrate’s primary structure may be
sufficient to determine whether an acceptor residue is likely to be
phosphorylated by a given kinase. Although sequence patterns are
only one piece of information in a puzzle of many (secondary
structure, tertiary structure, surface accessibility, etc.), an abun-
dance of data suggests that this is one of the most distinguishing
factors in describing a kinase–substrate interaction and in many
cases it is a sufficient predictive feature [12, 13]. Obviously, this
idea does not apply only to kinases but can be used to describe other
protein-protein interactions mediated by other types of modular
protein domains that recognize short linear sequence motifs on
their binding partners in a phospho-dependent or -independent
manner, such as SH2 and SH3 domains, FHA and BRCT domains,
14-3-3 proteins, etc.
Specific amino acid preferences can be described in two ways.
One is to describe them in a strict combinatorial regular expression-
like pattern (Boolean matching model). This approach was origi-
nally used in PROSITE [14] to search for patterns in a sequence
database. However, these patterns are very inflexible and do not
allow for including differently weighted preferences for amino
acids. A more flexible and powerful approach is the use of
position-specific scoring matrices (PSSMs) to describe patterns/
motifs in this form. This approach was implemented in Scansite
[15, 16], an application to predict short linear sequence motif sites.
A PSSM matrix like this contains a probability value for each amino
acid (columns) at each position of a sequence window of certain
size (rows), where each value in a column and row of the matrix
describes the binding partner’s preference for that amino acid at
that position in the motif. Scansite is a web application that uses
PSSMs to predict interaction sites that are important in cellular
signaling and includes more than 120 kinases and proteins that
recognize specific short linear binding motifs. It can be used to
show all potential sites in a given protein or all proteins in a database
that contain sites for one or more motifs. Directions for both uses
are provided in the following sections.
2 Materials
Scansite 3 (http://scansite3.mit.edu/) requires nothing more than

a computer with an Internet connection and a modern web
browser. Although it works with all popular web browsers, the
recommended options are Mozilla Firefox, Google Chrome, and
Opera. On some pages that display search results, Scansite will show
content in pop-up windows so that results can be viewed side by
side. Therefore it is recommended that you allow pop-ups in your
browser for these pages to work properly. Wherever Scansite allows
you to choose a sequence database (e.g., when selecting proteins or
for searching a database), you can choose from these resources:
SwissProt [17], SGD (yeast) [18], Ensembl (human and mouse)
[19], NCBI Protein (GenPept) [20], and TrEmbl [21]. Scansite
uses local mirrors of these databases in order to allow fast queries.
Over the past years Scansite has also become popular for analyses of
whole proteomes or subsets thereof. These are generally not done
using the web interface, but computationally. If you are interested
in using Scansite for this purpose, please see Note 1 for information
about Scansite’s web service.
To perform Protein Scans you need either a protein sequence or a

protein identifier (accession number or ID) for the protein you are
2.1 Scanning a interested in from one of Scansite’s protein sequence database

Protein for Motifs mirrors.
2.2 Searching a Database Searches only require information about the motif that is
Sequence Database for searched for in a particular sequence database. All of the standard
Motifs Scansite matrices for kinases and modular binding domains are
available. In addition, you may enter more specific information to
restrict the search to a smaller number of proteins.
3 Methods
Scansite’s two most important interaction prediction searches will

be described in detail in this section: Protein Scans that search for
motif matches in a given protein and Database Searches that find
proteins that contain one or more motifs in a protein sequence
database. A short overview of Scansite’s other features is given in
Note 2. In the following, you will be guided through the steps
necessary to use these features properly. Furthermore, some guid-
ance on how to interpret these searches’ results will be given.
3.1 Scanning a The key feature of Scansite is the prediction of motif-relevant sites
Protein for Motifs in a given protein. This feature is referred to as Protein Scan or Scan
Proteins for Motifs and allows a range of different inputs.
1. Navigate to Input Page. To get to the Protein Scan input screen
from anywhere in Scansite, click the “Scan Proteins for Motifs”
button in the navigation section on the left-hand side of the
web page.
2. Choose the Protein to Scan. There are two different ways of
choosing proteins in Scansite: by protein identifier (default
option) and by sequence.
To choose a protein by accession number, select “Protein
Accession” from the “Choose Protein by. . .” drop-down list.
Below, select a protein sequence database and enter a protein
ID. Links on the right-hand side of the text boxes refer to the
different sequence databases that Scansite currently supports
and where you can search for protein identifiers. After entering
at least three characters in the text box entitled “Protein Acces-
sion”, Scansite searches for protein IDs that start with these
characters and presents a list of options below the text box. The
same happens when the “Check!” button next to the text box is
clicked or the Enter key is pressed. You can either continue
typing or select an ID from the list. The text box turns green
for valid and red for invalid protein identifiers.
In order to enter a peptide sequence, select “Input
Sequence” from the drop-down list. The area below this menu
will change accordingly. Then, enter or paste a name and an
amino acid sequence. Invalid characters (punctuation marks,

white space, digits, etc.) are stripped from the sequence auto-
matically. This means that you can just paste a sequence that
is formatted with spaces and line breaks or annotated with
numbers. If you paste a FASTA-formatted sequence, make
sure not to copy the FASTA header (“>. . .”) with the sequence.
Otherwise all possible amino acid one letter codes in the header
will also become part of the sequence.
3. Choose Motifs to Consider. It is possible to search for all motifs of
a motif class, for only a selected subset of motifs or motif
groups or both, or for a user-defined motif (instructions on
how to create your own motif can be found in Note 3). You can
choose from these options in the drop-down menu entitled
“Look for”. Again, the area below this menu will change
accordingly dependent on your choice, offering a number of
additional choices. Begin by selecting a motif class (mammalian
or yeast). By default, Scansite’s mammalian motifs are dis-
played. To select more than one motif or motif group, hold
down the control key on your keyboard and make selections
using your mouse. If you are not sure which motifs belong to
which groups, you can either click the link below the list of
groups (“Show Group Definitions”) or follow the instructions
in Note 4. When using your own motif, select the motif file
from your computer. After the file is uploaded (this happens
automatically after you selected a file), you get a chance to
make changes to affinity values if you wish to do so.
4. Select a Stringency Level. This measure defines how high sites
have to score in order to be displayed as results. The setting
high only displays the very best sites, i.e., the top 0.2 % of sites
(sites that have a score less than or equal to the top 0.2 % of
motif-specific scores in the reference proteome). Medium strin-
gency displays the top 1 %, low the top 5 %, and minimum
displays the top 15 %. These settings apply only for motifs
from the Scansite database. Since no precompiled reference
proteome score distribution (see Note 5) is available for user-
defined motifs, these always display all sites with a score 5.
5. Additional Options. The two additional options that users are
given are to decide whether to show predicted domains in the
result as supporting information (see Note 6) and whether to
use an alternative reference proteome. At the moment users can
use either SwissProt’s Vertebrate proteins as a reference
(default) or all of SGD’s proteins (default for scans using
yeast motifs). Domains can also be requested later on from
the result page.
6. Click the Submit Button.
Fig. 1 The results of a high stringency protein scan for all mammalian motifs using the SwissProt protein
P53_HUMAN and the default reference proteome are shown. The section entitled “Scan Overview” which
summarizes the parameters of the scan of the page is collapsed to better fit this figure on the page
As an example for a Protein Scan result page, the results of a

high stringency protein scan are shown in Fig. 1. The result page is
split in seven sections (divided by grey bars): Protein Overview,
Scan Overview, Protein Plot, Predicted Motif Sites (Table), Repeat
Scan, Download Results, and Additional Analyses. Each of these
sections is collapsible by clicking on the grey title areas. This allows
the user to quickly get to the bottom of the page if a long list of
predicted sites is displayed.
In the “Protein Overview” section, some information about
the input protein is listed, including alternative identifiers and key-
words (only for proteins from Scansite’s databases), and the pro-
tein’s molecular weight and isoelectric point (calculated according
to ref. 22). The “Scan Overview” summarizes the input parameters
of the search and displays the number of sites that have been
detected using these settings. In the next part of the page (“Protein
Plot”), a plot of the protein gives a visual overview of the search
results displaying the protein sequence as a straight line annotated
with some additional information. If domain information about the
query protein was requested to be displayed, the predicted domains
are listed above the image. The plot displays the predicted sites
(annotated with the position and motif group), the protein’s
domains (if requested) along with their names and positions, and
a surface accessibility plot that shows which parts of the protein are
likely to be exposed to the surface and which ones are likely to be
buried. If domains have not been requested earlier, a button will be
displayed below the image that allows the user to request domain
prediction at this point. The links in the list of displayed domains
refer to these domains’ PFAM pages (see Note 6).
The sites that are outlined in the protein plot are listed in more
detail in the table view below (“Predicted Motif Sites”). Most
columns can be sorted by clicking on the label in the table’s header.
Here, each site that was found is displayed along with some motif
information (motif, motif group, hyperlink to motif’s gene infor-
mation page), its score and percentile, and the surrounding
sequence. In addition, Scansite-3 offers hyperlinks to PhosphoSite
[23], PhosphoELM [24], and Phosida [25] if a site was reported in
one of these databases before (for more information about “Previ-
ously Mapped Sites” see Note 7). The other links displayed in the
table, more specifically the columns “Score” and “Sequence,” refer
to a histogram view of a site in the reference proteome and to a view
that shows a site’s sequence highlighted in the protein’s sequence,
respectively. The latter view also offers a link that directly submits
the site’s sequence (15 amino acids) to NCBI’s basic local align-
ment search tool (BLAST) [26]. More information on BLASTing
sites in Scansite can be found in Note 8.
In the “Repeat Scan” section of the result page, it is possible to
either directly rerun the scan with a different stringency setting or to
go back to the input page to change other search parameters. This is
especially helpful if your search did not return any results. The next
part in the page (“Download Results”) offers a link to a download-
able version of the table shown above (tabulator-separated file).
At the bottom of the result page (“Additional Analyses”) users can
directly submit the current protein’s sequence to DisPhos [27], a
Disorder-Enhanced Phosphorylation Site Predictor (see Note 9).
3.2 Searching a The Scansite feature Search Sequence Database for Motifs or short
Sequence Database for Database Search performs a broader search than single protein scans.
Motifs Given a motif (or a set of motifs) and a sequence database, it searches
the database for sequences that contain motif-relevant sites. One of
the most powerful parts of this tool is the option of targeting a
search to specific experimental requirements by restricting searches
to proteins of a specific organism class, species, molecular weight
and isoelectric point range, annotation, and sequence property. For
example, this tool can be used to help identify unknown bands in
two-dimensional (2D) gel electrophoresis experiments.
1. Navigate to Input Page. To get to the Database Search input
screen from anywhere in Scansite, click the button “Search a
Sequence Database for Motifs” in the navigation section on the
left-hand side of the web page.
2. Choose the Search Method. The area below this drop-down list
will change dependent on what you select. Searches for single
“Database motifs” from the Scansite database are the easiest
option to choose. Alternatively, you can search for your own
motifs (see Note 3) or so-called “Quick Motifs” (see Note 10).
It is also possible to search for sequences that match up to five
motifs. These searches can include either database motifs, user-
defined motifs, or a combination of both. The score of a multi-
motif site is the mean (average) of all the scores of the sites
involved. Co-occurrences of different motif sites in proteins
can be filtered in different ways. First of all, it is possible to
penalize gaps between sites of different motifs. Gap penalty
settings are either high, medium, low, or none. Penalties p are
then added to the score according to the maximum distance
dmax between the involved sites (i.e., position of site closest to
C-terminus minus position of site closest to N-terminus). The
penalty values are calculated as follows: plow ¼ 0.001 dmax;
pmedium ¼ 0.01 dmax; phigh ¼ 0.1 dmax. Secondly, it is
possible to define up to three strict minimum and maximum
distance bounds between motif-specific sites. This can be used
if you know which motifs to expect and how far apart you
expect them to be in the protein sequence. If you just want to
get an overview of peptides that have multiple motif sites, it is
recommended to use a gap penalty. Using distance bounds is
the better option for very specific searches.
3. Select Database to Search from the drop-down menu.

4. Restrict Search. It is recommended that you exclude as many
proteins from the search as possible to both target your search
as much as possible to what you are looking for and to decrease
the runtime of the search. Database searches can take several
minutes and the runtime of a search mostly depends on the
number of proteins that are searched. You will find useful hints
on what kinds of restrictions you can apply in Note 11.
5. Select Number of On-Screen Results. Since Database Searches
may find a very high number of results and visual exploration of
a table of thousands of results generally is avoided, the number
of sites that are displayed in the web browser is limited. By
default, the size of the output list is limited to 50, but users
can also choose sizes 100, 200, 500, 1,000, and 2,000. Please
note that these are just the numbers of sites that are displayed in
the table on the result page. A file containing all the hits that
were found in this search can be downloaded from the result
page as well.
6. Click the Submit Button.
A result page of a Database Search is displayed in Fig. 2. Four
sections can be distinguished within the Database Search result
page. The “Search Input” section at the top of the page summarizes
the preferences defined in the input page. “Search Results” gives an
overview of the number of proteins in the entire sequence database,
the number of proteins found that matched the given restrictions,
and the number of sites found in these proteins. In addition, the
median and MAD (median absolute deviation) of these sites’ scores
is displayed. This part is followed by a table view of the sites found
(“Predicted Motif Sites”). The table shows the (combined) site
score, some information about the protein that was found (includ-
ing MW and pI), and displays some site-specific information (site
and surrounding sequence). For multi-motif searches a site and
sequence column for each motif in the motif’s site is given. The
first column in the table allows to directly scan the protein for other
motifs. This is useful if you want to know what other motifs are
found in that protein, if a site has been reported before (previously
mapped) in another database, and how the protein is generally
composed (domains, surface accessibility). The link in the column
labeled “Accession” takes the user to the protein’s page in its
primary database. The score column links to a histogram that
shows the site’s score in comparison to all scores found in that
search. At the bottom of the page, options for downloading the
entire result set and for repeating the search are given.
Fig. 2 The results of a Database Search for ATM in human proteins of SwissProt that are annotated with “cell
cycle” and contain the sequence “ARATT”. Here, only one protein matched the given restrictions and this
protein also contains the motif that was searched for
4 Notes
1. Accessing Scansite Computationally. The current era of geno-

mics and proteomics often requires analyses of large numbers of
proteins. To make tasks like this easier it is now possible to
access Scansite computationally using a web service. The para-
meters of protein scans, database searches, and other utility
functions are sent to Scansite using a URI. The results are
then returned in XML format. Detailed instructions and exam-
ples are available online at http://scansite3.mit.edu/
Scansite3Webservice/. This link can also be found in Scansite’s
FAQ online.
2. Getting the Most out of Scansite. In addition to the features

described in detail above, Scansite offers some more useful
tools. To start with, you can search Scansite’s sequence data-
bases for simple wildcard-based sequence patterns or regular
expressions. Another tool calculates a sequences molecular
weight and isoelectric point for a given number of putative
phosphorylations. Last, a tool called “Calculate Amino Acid
Composition” visualizes a protein sequence’s amino acid com-
position by highlighting selected sites and displaying the rela-
tive abundance of sites (e.g., all tyrosines in a sequence that are
followed by leucines two residues downstream). In addition,
this tool displays a protein’s domain information as calculated
by InterProScan [28]. One can also use one of these tools to
analyze a protein sequence, copy/paste it to make changes
(e.g., introduce mutations), and then use it as an input for
protein scans.
3. Creating Scansite Motifs. Both main search options in Scansite
allow the use of user-defined motifs. These motifs have to be in
a Scansite-specific tabulator-separated file format. All user-
defined motifs that are uploaded to Scansite are only used for
the user’s searches and are deleted as soon as the user leaves the
site. If you have a clear idea of what motif you want to look for,
use the information below to specify your own Scansite-specific
motif file.
PSSMs in Scansite describe amino acid-specific affinity values
for a sequence window of 15 residues. Lines correspond to
positions in the sequence window, columns (separated by tabu-
lators) to amino acids. It is not necessary to define values for
every single amino acid—default values are used for omitted
residues. The first line (row 1, header) defines the residue-to-
column assignments using amino acid one letter codes. Those
amino acids can be in any order. The following lines (rows
2–16) define affinity values for the respective residues; rows
2–8 and 10–16 define the N- and the C-terminal side of the
motif, respectively. Scansite’s search for sites in a peptide
sequence highly depends on the PSSM’s central residue
(row 9). At least one site in this position needs to be invariant
in the motif sequence. For example, the fixed residue should be
a Y for motifs recognized by tyrosine-kinases and S and T for
serine-/threonine-kinases. To mark a position as invariant, the
value 21 has to be used.
In addition to columns of standard amino acids (default
values of 1), it is also possible to incorporate special require-
ments. A motif’s preference for a protein sequence’s N- or C-
terminus can be incorporated by using a column labeled “$”
(dollar sign) or “*” (asterisk), respectively. These positions are
assigned values of 0 by default. Scansite 3 also recognizes the
rarely occurring amino acids selenocysteine (U) and pyrrolysine

(O), which can be added by their one letter code as well. Due to
their similar chemical structure, the default numbers for these
residues are the values of cysteines and lysines, respectively.
Lastly, some wildcard values can be used for very special cases:
B (aspartate/asparagine), Z (glutamate/glutamine), J (leu-
cine/isoleucine), and X (any residue). These symbols are
included because they occur rarely in public protein databases.
Generally speaking, they have no relevance for actual research
purposes. The default values for these wildcards are the mean
values of the amino acids they encode.
Now that the general structure and default values of motif
files were defined, you may wonder what values to use to define
affinities. Scansite’s scoring system ranges from 0 to roughly 21.
Giving an individual amino acid a score of 1 at one position in
the motif indicates that no preference exists, positive or nega-
tive, for that particular amino acid in that position. Giving
all amino acids in one position of the motif a score of 1 (i.e.,
making all values in a single row of the matrix equal to one)
indicates no preference exists for any particular residue type at
that position in the motif. The value 21 defines that the amino
acid that is given this value in a position is required in this
position for the motif to find a match. Values higher than 21
are permitted to indicate very strong affinities. However, nega-
tive values are not permitted for defining a strong disfavoring of
amino acids. Instead, values between zero and one should be
used for that purpose. Beware that the scoring function uses
logarithms, so values less than 1, particularly those less than
0.5, strongly penalize for that particular residue in a motif.
Here is a short checklist to avoid the most common pitfalls of
creating motifs:
l Is there at least one amino acid with value 21 in the central
position?
l Is there a header line defining the columns using amino
acid one letter codes?
l Are there 16 lines (1 header and 15 lines with values)
in the file?
l Are all column separators in the file tabulators (and not
spaces or other characters)?
4. Learn more about Scansite’s Data. In a section called “Data-
bases and Motifs” in the navigation section of the web page
(left-hand side), an overview of Scansite’s database mirrors
(release dates and sizes), motifs, and motif group definitions is
presented. In the motifs section you can select a motif and click
“Get Info!.” Clicking this button will visualize the motif as a
sequence logo [29] and display a link that takes you to a web
page that gives information about the gene that recognizes this
motif. Mammalian motifs and yeast-specific motifs are sup-
ported by information from GeneCards [30] and SGD,
respectively.
5. Interpreting Scansite Scores. Scansite’s scores range from 0 to
(theoretically) 1. However, you will never see scores higher
than 5 because sites with scores that high are discarded in the
scoring process. Please be aware that scores in Scansite are
always motif-dependent. This means that scores for different
motifs should not be directly compared to each other. For
example, knowing that one motif’s optimal score is 0.001 and
another motif’s best score is 0.4 it is easy to say that these are
the best possible scores, so hits with these scores are equally
good. However, the only way to extend this knowledge to
slightly poorer scores is to know how likely other scores are to
occur. To make this possible and allow a comparison among
motifs, Scansite offers percentile values. The percentiles used in
Scansite are calculated from the so-called reference proteomes
which are proteomes that are commonly used in research. In
the process of adding a motif to Scansite, it is scored against
every single peptide in the reference proteome and the scores
are stored to create a score distribution. This distribution is
then used to calculate percentile values from scores calculated
when users run certain searches. Using these values it is possible
to rank sites from different motifs.
6. Domains in Scansite 3. Scansite uses InterProScan [28] to
predict a protein’s PFAM domains [31]. Therefore the domain
positions displayed in Scansite may vary by a few amino acids
from the positional assignments seen on the PFAM homepage.
This is mentioned because these variations may cause confusion
but do not pose a problem since all these positions are predic-
tions and there is no way to tell which numbers are more
correct in the absence of clear structural data from crystallo-
graphic or NMR experiments.
7. Previously Mapped Sites in Scansite. Displaying previously
mapped sites in Scansite is only possible for proteins from public
protein databases and works best with proteins from SwissProt.
Please note that these references are only site-specific but not
motif-specific. This means that if a previously mapped site
shows up in the list, the site is reported in the linked databases;
however, this does not imply that the Scansite motif that was
found at this site is related to the site reported in the database.
It could be that a completely different gene is responsible for
this site. Wherever possible, the hyperlinks refer directly to the
external databases page about this site. If a database does not
support direct linking, the link just takes you to the database’s
homepage.
8. BLASTing of Sites. Scansite allows to directly submit the

15-mers around identified sites to NCBI’s BLAST. This is a
simple approach to see if a site is conserved in organisms
that are expected to be physiologically similar to the one at
hand. If the site is also found in similar proteins in other species,
the site is more likely to be biologically relevant.
9. Intrinsically Disordered Proteins. Disordered regions in pro-
teins are stretches of amino acids that do not have a rigid
tertiary structure and are therefore enabled to change confor-
mation. Disordered Proteins are proteins with disordered
regions. It has been shown [32] that many posttranslational
modifications and binding sites occur in disordered regions
because these regions make a protein more flexible, which
facilitates binding and interaction processes. DisPhos is a
disorder-prediction engine that focuses on potential phosphor-
ylation sites. The results of DisPhos searches can therefore be
used as supporting information for phosphorylation sites pre-
dicted by Scansite.
10. Using Quick Motifs. Creating a custom motif only makes sense
if enough information about the affinities of the kinase or
binding domain is known. This, however, requires a very spe-
cific idea about the motif. Often, only very little detail about a
motif is known. In cases like these, creating a “Quick Motif” to
search a database is the best option. For defining a quick motif,
the user can enter a set of primary and secondary preferences
for each position of a 15-mer. These preferences are then used
to calculate a simple Scansite motif. As for actual Scansite
motifs, the center position needs to be fixed, so it is not possible
to enter secondary preferences there. The web page describes a
number of wildcards that can be used in this process to easily
describe amino acid subsets by their physicochemical properties
(e.g., hydrophobic or positive residues). A simplified regular
expression-like version of the motif that is entered is displayed
below the text boxes (with resolved wildcards) as soon as values
in the text boxes are changed.
11. Restricting Searches. Searches of protein databases can be
restricted in a number of ways to allow better more targeted
searches. At the same time applying restrictions reduces the
number of proteins that have to be scanned and therefore
may significantly reduce the time a query takes. Consequently,
users are encouraged to restrict their searches as rigorously as
possible. For some on-site information, a short help text about
each restriction can be displayed by clicking on the links next to
the text boxes.
l For many searches, you may only be interested in matches
from humans or a particular model organism. Searches can
be restricted this way by entering the species’ name in the

text box labeled “Single Species.” This feature supports
many MySQL-style wildcards (regular expressions) to
match species names. For example, if you are tired of
writing out “Caenorhabditis elegans”, you can use “C.*
elegans” instead. In a regular expression, the period (.)
matches any single character, and the asterisk extends that
match to multiple characters (or even zero characters). This
also allows for genus-wide searches, by entering just “Rat-
tus” for example. However, this may yield unexpected
results when trying to search for all kinds of mice with
“Mus.” This expression will accidentally match “Thermus
aquaticus” as well, but you can avoid that by entering
“^Mus.” The caret symbol (^) requires the text to match
at the beginning of the entered name. One of the most
common pitfalls is when the species entered does not
match the organism class specified above (e.g., a search
for “yeast” when “Mammals” is selected). Please note
that Scansite’s organism classes are not taxonomic “classes”
in the conventional sense (except for Mammals) but groups
of species frequently used for research purposes.
l The molecular weight, isoelectric point, and phosphoryla-
tion options are intended for use in conjunction with 2D
gel electrophoresis experiments. When you find a few spots
appearing reproducibly on a 2D gel under a particular test
condition and not under the control, you could use Scan-
site to find what proteins are expected to be in that region
of the 2D gel by putting in ranges for molecular weights
and isoelectric points. You could simultaneously constrain
the species to match the cell line you used in the experi-
ment. If it is an experiment involving possible phosphory-
lation events, you can see how much a putative
phosphorylation would move the peptides on the gel.
l Matches for “Keywords” are searched in a protein’s anno-
tations and are therefore primarily useful for searching
well-annotated databases like SwissProt. For example, pro-
teins involved in the cell cycle can be easily identified by
entering “cell cycle,” novel proteins in GenPept by search-
ing for “hypothetical.”
l The “Sequence Contains” text field is a quick way to restrict
your search to proteins containing a consensus sequence. It
is important to note that the consensus sequence entered
here is not required to be part of the motif being searched
for. It is merely required to show up somewhere in the
sequence. Also, note that regular expressions have to be
used here instead of the protein wildcard signs (“.” instead
of “X”, “[ND]” instead of “B”, etc.). For example, the
sequence “PXXP” is represented as “P..P” in regular expres-

sion syntax. More information on regular expressions and
how they can be used in Scansite is available in Scansite’s
frequently asked questions (FAQ) section online.
References
1. Berman HM, Westbrook J, Feng Z et al (2000) 13. Pinna LA, Maria Ruzzene M (1996) How do
The Protein Data Bank. Nucleic Acids Res protein kinases recognize their substrates? Bio-
28:235–242 chim Biophys Acta 13143:191–225
2. Mathivanan S, Periaswamy B, Gandhi T et al 14. Bairoch A (1992) PROSITE: a dictionary of
(2006) An evaluation of human protein- sites and patterns in proteins. Nucleic Acids
protein interaction data in the public domain. Res 20(Suppl):2013–2018
BMC Bioinformatics 7:S19 15. Yaffe M, Leparc G, Lai J et al (2001) A motif-
3. Turinsky A, Razick S, Turner B, et al. (2010) based profile scanning approach for genome-
Literature curation of protein interactions: wide prediction of signaling pathways. Nat
measuring agreement across major public data- Biotechnol 19:348–353
bases. Database (Oxford) 2010: baq026 16. Obenauer J, Cantley L, Yaffe M (2003) Scan-
4. Shoemaker B, Panchenko A (2007) Decipher- site 2.0: proteome-wide prediction of cell
ing protein-protein interactions. Part II. signaling interactions using short sequence
Computational methods to predict protein motifs. Nucleic Acids Res 31:3635–3641
and domain interaction partners. PLoS Com- 17. M. M, UniProt-consortium (2011) UniProt
put Biol 3:e43 Knowledgebase: a hub of integrated protein
5. Pitre S, Alamgir M, Green J, et al. (2008) data. Database: bar009
Computational methods for predicting 18. Cherry J, Hong E, Amundsen C et al (2011)
protein–protein interactions. In: Advances in Saccharomyces Genome Database: the geno-
biochemical engineering/biotechnology: mics resource of budding yeast. Nucleic Acids
protein-protein interaction. Springer, Res 40(Database issue):D700–D705
Heidelberg, pp 247–267 19. Flicek P, Amode M, Barrell D et al (2011)
6. Andrusier N, Mashiach E, Nussinov R et al Ensembl 2011. Nucleic Acids Res 39(Suppl
(2008) Principles of flexible protein–protein 1):D800–D806
docking. Proteins 73:271–289 20. Burks C, Cassidy M, Cinkosky MJ et al (1991)
7. Janin J (2002) Welcome to CAPRI: a Critical GenBank. Nucleic Acids Res 19:221–225
Assessment of PRedicted Interactions. Proteins 21. Boeckmann B, Bairoch A, Apweiler R et al
Struct Funct Genet 47:257 (2003) The SWISS-PROT protein knowledge-
8. Rhodes DR, Tomlins SA, Varambally S et al base and its supplement TrEMBL in 2003.
(2005) Probabilistic model of the human Nucleic Acids Res 31:365–370
protein-protein interaction network. Nat Bio- 22. Bjellqvist B, Hughes G, Pasquali C et al (1993)
technol 23:951–959 The focusing positions of polypeptides in
9. Trost B, Kusalik A (2011) Computational pre- immobilized pH gradients can be predicted
diction of eukaryotic phosphorylation sites. from their amino acid sequences. Electropho-
Bioinformatics 27:2927–2935 resis 14:1023–1031
10. Hutti J, Jarrell E, Chang J et al (2004) A rapid 23. Hornbeck P, Kornhauser J, Tkachev S et al
method for determining protein kinase (2012) PhosphoSitePlus: a comprehensive
phosphorylation specificity. Nat Methods resource for investigating the structure and
1:27–29 function of experimentally determined
11. Songyang Z, Blechner S, Hoagland N et al post-translational modifications in man and
(1994) Use of an oriented peptide library to mouse. Nucleic Acids Res 40:D261–D270
determine the optimal substrates of protein 24. Dinkel H, Chica C, Via A et al (2011) (2011)
kinases. Curr Biol 4:973–982 Phospho.ELM: a database of phosphorylation
12. Kemp BE, Pearson RB (1990) Protein kinase sites—update 2011. Nucleic Acids Res 39:
recognition sequence motifs. Trends Biochem D261–D267
Sci 15:342–346 25. Gnad F, Ren S, Cox J et al (2007) PHOSIDA
(phosphorylation site database): management,
structural and evolutionary investigation, and
prediction of phosphosites. Genome Biol 8: 29. Schneider T, Stephens R (1990) Sequence

R250 logos: a new way to display consensus
26. Altschul SF, Madden TL, Sch€affer AA et al sequences. Nucleic Acids Res 18:6097–6100
(1997) Gapped BLAST and PSI-BLAST: a 30. Stelzer G, Dalah I, Stein T et al (2011) In-silico
new generation of protein database search pro- human genomics with GeneCards. Hum
grams. Nucleic Acids Res 25:3389–3402 Genomics 5:709–717
27. Iakoucheva L, Radivojac P, Brown C et al 31. Punta M, Coggill P, Eberhardt R et al (2012)
(2004) The importance of intrinsic disorder The Pfam protein families database. Nucleic
for protein phosphorylation. Nucleic Acids Acids Res 40:D290–D301
Res 32:1037–1049 32. Uversky V, Dunker A (2010) Understanding
28. Hunter S, Apweiler R, Attwood T et al protein non-folding. Biochim Biophys Acta
(2009) InterPro: the integrative protein signa- 1804:1231–1264
ture database. Nucleic Acids Res 37:
D211–D215
Chapter 5
Structure-Based Computational Approaches

for Small-Molecule Modulation of Protein-Protein
Interactions
David Xu, Bo Wang, and Samy O. Meroueh
Abstract
Three-dimensional structures of proteins offer an opportunity for the rational design of small molecules to
modulate protein-protein interactions. The presence of a well-defined binding pocket on the surface of
protein complexes, particularly at their interface, can be used for docking-based virtual screening of
chemical libraries. Several approaches have been developed to identify binding pockets that are implemen-
ted in programs such as SiteMap, fpocket, and FTSite. These programs enable the scoring of these pockets
to determine whether they are suitable to accommodate high-affinity small molecules. Virtual screening of
commercial or combinatorial libraries can be carried out to enrich these libraries and select compounds for
further experimental validation. In virtual screening, a compound library is docked to the target protein.
The resulting structures are scored and ranked for the selection and experimental validation of top
candidates. Molecular docking has been implemented in a number of computer programs such as Auto-
Dock Vina. We select a set of protein-protein interactions that have been successfully inhibited with small
molecules in the past. Several computer programs are applied to identify pockets on the surface, and
molecular docking is conducted in an attempt to reproduce the binding pose of the inhibitors. The results
highlight the strengths and limitations of computational methods for the design of PPI inhibitors.
Key words Protein-protein interactions, Molecular docking, Structure-based drug design, Virtual
screening, Small molecules, Inhibitors
1 Introduction
Protein-protein interactions (PPIs) control nearly every aspect of

normal cellular function. These interactions can also promote a
diverse set of cellular processes that lead to pathological processes
such as cancer [1]. Protein interactions are typically identified
through affinity purification and other pull-down techniques, but
modern approaches like yeast-two-hybrid have enabled large-scale
mapping of protein interaction networks [2]. Genomic data for
cancer cells can be mapped onto these networks to uncover new
PPIs. Three-dimensional structures can facilitate the rational design
77
78 David Xu et al.
of small molecules that bind and modulate these interactions [3].

Protein structures often harbor well-defined binding pockets that
may be located at the protein-protein interface, at distal sites out-
side the interface, or at enzyme active sites. It is expected that small
molecules that bind to these pockets will modulate the formation of
the PPIs either by directly interfering with binding or through
allosteric effects.
Structure-based computational methods have significantly
matured over the past two decades. These methods provide tools
to identify pockets and sites that are suitable for small-molecule
binding. Pockets enable the screening of large chemical libraries to
generate hit compounds that can be further optimized to modulate
the protein interactions. Here we describe how to search for protein
structures and scan their surface for binding pockets. We discuss the
use of molecular docking of chemical libraries to these pockets to
generate protein–ligand structures that can be scored and ranked.
We discuss the process of scoring in virtual screening. We end with a
case study of six PPIs that have been successfully targeted with small
molecules.
2 Methods
2.1 Identifying Starting with the sequence of a gene that encodes for PPI proteins,
Three-Dimensional the sequence of one of the binding partners can be used in a BLAST
Structures of Protein search using the advanced search tool at the Protein Data Bank
Interactions (PDB) [4] (see Note 1). The FASTA format [5] for the protein
sequence can be used in the search. A threshold value known as the
E-value limits the search to proteins that possess significant
sequence identity and coverage of the query sequence. Based on
previous studies [6, 7], a value of 10-6 is expected to identify
protein domains or very close homologs of the query protein
(see Note 2). In addition, the search parameter “Number of Enti-
ties” can be used to filter for protein complexes by setting the Entity
Type to “Protein” with a minimum bound of 2. Different structures
that overlap the same protein sequence can also be removed by
checking the “Remove similar sequences at 90 % identity” box.
Similarly, structures of protein complexes can be identified by
using the Web server Interactome3D [8]. This tool provides struc-
tural annotations of individual proteins as well as protein complexes
in a variety of model organisms, which can be queried by using
UniProt accession IDs. If the protein complex is known, then the
interaction pair can be used in the search. Otherwise, searching for
individual proteins results in a network with all the interactions of
that protein. Clicking on any of the edges in this network yields a
ranked list showing structures from PDB which contain this partic-
ular interaction, as well as other information such as: (1) resolution
of the structure; (2) chain of each of the binding partners in the
Structure-Based Computational Design 79
protein complex; (3) sequence identity of each of the structures;

and (4) how much of the protein sequence is covered by the
structure.
2.2 Identifying The presence of a well-defined binding cavity at the surface of a

Pockets on Protein protein can significantly facilitate the design of high-affinity small
Structures molecules. Pockets reduce compound exposure to solvent and
through van der Waals and electrostatic forces provide additional
stability to a ligand [9]. The location of a pocket on the structure of
a protein–protein complex determines the impact of a ligand on the
interaction. Compounds targeting an interface pocket will likely
disrupt a protein interaction. But compounds that bind outside the
interface may either stabilize or destabilize the complex through
allosteric effects.
Several methods have been developed to identify pockets on
protein surfaces. The most common approach uses a three-
dimensional grid around the entire protein to determine the van
der Waals energies at each grid point as implemented in SiteMap, a
module available in the Schrödinger software suite [10]. The van
der Waals radius of each grid point is compared to the distance from
the grid point to nearby protein atoms to determine whether the
point is outside the protein. Points outside the protein are kept if
the van der Waals interaction energy is less than a given cutoff.
These points are then clustered to define the binding pocket.
SiteMap typically returns the top 5 detected binding sites, but a
larger number of sites may be necessary on larger proteins. Another
method implemented in the freely available program fpocket [11]
uses a grid-free approach. Voronoi tessellation, a method for spatial
division, is used to determine the location of α spheres on the
protein. In this approach, α spheres are placed at the Voronoi
intersects formed by the tessellations. Each of these spheres forms
contacts with four atoms at its boundary, such that the sphere radii
determine whether or not it is on the protein’s surface. Pockets are
detected by clustering α spheres within a specific radius; a range of
3–6 Å is the default. A third approach implemented in the program
Cavbase [12] follows a similar strategy to what is implemented
in SiteMap. A Cartesian grid around the protein is defined by the
calculation of the van der Waals energies at each grid point. How-
ever, CavBase is mainly used for comparing different binding sites
based on the definition of pseudocenters [12]. Finally, FTSite [13]
is a Web server for pocket detection that uses fast Fourier trans-
forms to calculate the energy of chemical probes (functional
groups) on a grid. Thus, identifying the clusters formed by of
each of the probes separately and also the overlap of these clusters
between probes is used to identify likely binding sites.
In each case, the protein structure is preprocessed to be used
for the pocket identification program. The coordinates of the pro-
tein must be separated from the protein–protein or protein–ligand
80 David Xu et al.
complex as a separate file to properly detect binding sites.

In addition, noncovalently bound atoms and molecules are also
removed; these typically include water molecules, or ions used
during the crystallization process. It is often the case that
structures lack electron density within specific regions of the pro-
tein (see Note 3). Most commercial packages such as the Prime
module in the Schrödinger package can incorporate missing side
chains or even missing loops. Once the structure processing is
completed, the pocket identification programs are used to identify
various pockets on the structure.
2.3 Scoring Pockets Various metrics can aid in determining whether a pocket can
accommodate a small molecule. Two metrics have been developed
and implemented in SiteMap for this purpose: SiteScore and Drug-
Score [14]. SiteScore is a measure of whether a pocket can accom-
modate a small molecule, while DrugScore provides information
about the suitability of the pocket for the development of thera-
peutics. Both SiteScore and DrugScore use the weighted sums of
the same parameters, namely (1) the number of site points in the
binding pocket; (2) enclosure score that is a measure of how
accessible the pocket is to solvents; and (3) hydrophilic character
of the binding pocket (hydrophilic score). Unlike DrugScore,
SiteScore limits the impact of hydrophilicity in charged and highly
polar sites. On the other hand, fpocket provides only one metric
known as the Druggability Score [15]. The Druggability Score is a
general logistical model based on the local hydrophobic density of
the binding site, the hydrophobicity score, and the normalized
polarity score. Generally, a SiteMap SiteScore above 0.8, corre-
sponds to a pocket that can accommodate a high-affinity small
molecule. A SiteMap DrugScore above 0.9 [14], or an fpocket
Druggability Score above 0.7 [15] correspond to druggable pock-
ets. FTSite identifies and ranks the detected pockets by the number
of amino acids that are in contact with the probes, but does not give
a quantitative assessment on the druggability of a pocket.
It is often the case that pockets will be identified outside a
protein-protein interface or on the surface of a protein whose
complex with its binding partner has yet to be solved by crystallog-
raphy. In many cases, the binding partners may not even be known.
Potential binding partners can be identified from various datasets
that report PPIs such as BioGRID [16] or MINT [17]. These
datasets typically provide experimental data from yeast two-hybrid
studies or other techniques in tab-delimited or PSI-MI XML for-
mats, or through their web interface. The data can be readily
converted to protein sequence by first retrieving a common identi-
fier, such as the UniProt ID, and then retrieving the FASTA format
sequence from UniProt [18]. Pockets may also potentially be an
enzyme active site. Binding of a small molecule to an enzyme active
site located on a protein–protein complex may potentially act in an
allosteric manner and modulate the protein interaction. The

location of these enzymatic binding residues can be retrieved
from UniProt [18] or Catalytic Site Atlas [19] for the associated
protein and compared to the location of the identified binding
pockets.
2.4 Virtual Screening Structure-based virtual screening is widely used to enrich large
of Chemical Libraries chemical databases to generate focused libraries that can be experi-
to Target Pockets mentally validated [20–22]. To date, only a handful of cases have
been reported whereby compounds that inhibit PPIs emerged
directly from virtual screening [23–26]. In most cases, virtual
screening provides moderate to low affinity compounds that can
serve as a starting point for the development of PPI inhibitors.
Structure-based virtual screening consists of two steps, namely
docking and scoring. Docking corresponds to the series of compu-
tational steps to predict the binding mode of a ligand to a target
protein. The resulting protein–ligand complex provides coordi-
nates that can be used to score and rank-order the interactions.
In virtual screening, it is not uncommon that hundreds of
thousands of compounds are docked to a target, followed by
scoring for the selection of the top ~100 compounds.
Several computer programs have been developed for docking,
such as DOCK [27, 28], AutoDock [29], GOLD [30, 31], FlexX
[32], Glide [33, 34] among others [35–39]. AutoDock is one of
the most cited open-source molecular modeling simulation dock-
ing programs. It was designed, implemented, and maintained by
The Scripps Research Institute. Its latest version, Vina [40], was
released in 2010. Vina inherits major ideas and approaches from
AutoDock. By using a new source code, scoring function, and
algorithms, Vina significantly improves the average accuracy of
the binding mode prediction compared to AutoDock.
As described above, docking requires a three-dimensional
structure of a protein as well as a pocket on the surface. Typically,
a simple visualization of the structure can identify the location of
the pocket of interest. For PPIs, pockets located at the interface are
the most desirable since compounds that bind to these pockets are
expected to disrupt the interaction. Pockets that include hot-spot
residues are particularly attractive. The hot-spots can be located
within the binding partner that contains the pocket or on the
partner that occupies the pocket [25, 41–44]. If the location of
the pocket is unknown, it can be identified using programs such as
SiteMap or fpocket (see above) if necessary. Once a pocket is iden-
tified, a grid box based at the center of the pocket is created. The
pre-calculated grid maps obviate the need to calculate the interac-
tion energies at each step of the docking process. This results in a
significantly faster docking run (see Note 4). Vina generates the
map and carries out the energy calculations.
82 David Xu et al.
Typically, virtual screening is carried out on a single crystal

structure that is kept rigid during the docking. But in solution,
proteins sample multiple conformations. Different compounds will
often bind to distinct conformations of the protein. Hence, using a
single structure during virtual screening is likely to miss com-
pounds that would otherwise bind to alternative conformations.
There are several ways to introduce protein flexibility, such as
induced-fit docking [45–47], docking with multiple crystal struc-
tures [48–50], NMR structures [51, 52], and structures from
molecular dynamics (MD) simulations [23, 53]. AutoDock Vina
offers the option of predefining flexible residues in the pocket.
The selected side chains in the receptor are treated explicitly in a
separate file. The grid maps and interaction energies are also gen-
erated and calculated for flexible side chain atoms. In other words,
the atoms of flexible side chains are treated like atoms in the ligand.
The grid box has to be enlarged when using flexible residues to
offer enough space for the movement of the atoms in the flexible
side chains.
AutoDock Tools (ADT) in MGLTools package is the interac-
tive UI used to prepare a structure for docking with Vina. There are
Python scripts in MGLTools that can convert the files in batch
mode. All protein and ligand structure files have to be converted
to PDBQT before docking. PDBQT is an extended PDB format
coordinate file, which includes atomic partial charges, atom types,
and information on the torsional degrees of freedom. The PDBQT
files for ligand and receptor, center coordinates, and X, Y, Z dimen-
sions of the search space grid box are the required input parameters
for Vina docking. The exhaustiveness parameter controls the length
of the docking run. The default value of exhaustiveness is 8, which is
usually suitable for docking in a cube with an 18 Å edge. While a
larger exhaustiveness parameter for Vina will increase the length of
the docking run, it will not necessarily lead to improved docking
results [54]. It is recommended that the exhaustiveness is increased
when using a search space grid box larger than 30 30 30 Å.
ZINC [55] is a free database of commercial and annotated com-
pounds that are processed for virtual screening. Over 34 million
unique compounds are loaded from 134 commercial supplier cata-
logs and 36 annotated catalogs. About 40 % of the compounds are
“drug-like”; 13 % are “lead-like”; and 1.5 % are “fragment-like.”
Structure files in SDF, SMILES, and Mol2 format can be accessed
from the ZINC Website (http://zinc.docking.org/). The proton-
ation state of compounds is generated using Epik (version 2.1209)
[56] in the Schrödinger Suite at four different pH ranges: reference
range (pH ¼ 7.1); middle range (pH of 6–8); high range (pH of
7–9.5); and low range (pH of 4.5–6). Atomic charges and desolvation
of the compounds are calculated using AMSOL [57, 58]. OEChem
software [59] from OpenEye is applied to convert original 2D SDF
files into isomeric SMILES. Molecular Networks Corina program
[60] is used to generate the initial 3D conformation of the small

molecule. ZINC provides easy access for users to search the biological
activity of molecules, or search compounds that are active against a
particular target.
Other databases that provide chemical structures of small mole-
cules are also available. These include ChEMBL [61], PubChem
[62], DrugBank [63], and BindingDB [64]. ChEMBL is an data-
base that contains 5.4 million bioactivity measurements for more
than 1 million compounds and 5,300 protein targets. PubChem is
one of the most comprehensive databases that includes over 25
million unique chemical structures and 90 million bioactivity out-
comes associated with several thousand macromolecular targets,
but many of the compounds are not commercially available. Drug-
Bank is a richly annotated database of drug and drug target infor-
mation. BindingDB provides approximately 20,000 experimentally
determined binding affinities of protein–ligand complexes for 110
protein targets and 11,000 small molecule ligands.
It is important to note that docking programs use a fitness
function to guide the docking process towards the optimum bind-
ing mode. But these functions may not perform well for library
enrichment during virtual screening. Re-scoring of protein-
compound complexes from the docking step is often carried out
for better enrichment performance. There are several types of
scoring functions that have been developed over the years that
include; (1) force field, such as GBSA [65] and PBSA [66]; (2)
empirical, such as ChemScore [67], GlideScore [33], and SVRKB
[68]; (3) knowledge-based scoring functions, such as PMF [69],
DrugScore [70]; and (4) machine learning algorithms such as our
own SVMSP approach [71]. Force field-based scoring functions
use potential energy functions that describe various bonded and
nonbonded interactions in a molecule; examples include stretch,
bend, dihedral, and van der Waals interactions. Knowledge-based
scoring functions are developed from statistical pairwise potentials
derived from a large number of three-dimensional structures.
SVMSP and SVRKB are two newly developed scoring functions
that use a support vector machine trained from features obtained
from three-dimensional structures [71].
3 Case Studies
A set of protein–protein complexes that have been successfully

inhibited with small molecules and described in a previous review
article [72] is used to illustrate the above methods. To identify the
structure of these proteins, a keyword search was conducted at the
PDB. For example, a search for interleukin-2 led to 268 structures.
Structures with IL-2 in complex with other proteins were identified
by limiting the protein stoichiometry to display only heteromers.
84 David Xu et al.
Another approach used the sequence of IL-2 that was obtained

from UniProt (UniProt ID: P60568). A search using the FASTA
sequence and a more relaxed E-value of 10-4 led to 17 structures of
IL-2 without any additional filters. Of these 17 structures, one
structure contained the IL-2/IL-2RA complex (PDB: 1Z92),
another contained the ternary complex of IL-2 with its beta and
gamma subunits (PDB: 3QAZ), and two contained the quaternary
complex of IL-2 with all three subunits (PDB: 2ERJ and 2B5I).
Similarly, the structures of the other five protein–protein complexes
can be identified in this manner.
The structure of the binding partners can be used to identify
binding pockets and assess whether these pockets are suitable to
accommodate small molecules. To identify binding cavities on the
six PPIs in Table 1, the monomer structures were first separated
from the complex and preprocessed using the Protein Preparation
Wizard in the Schrödinger package. The structure of the cytokine
IL-2, an essential regulator in T cell growth and proliferation [73],
forms an interaction interface with its alpha receptor (PDB ID:
1Z92) that is inhibited by a small molecule (PDB: 1PY2). To
prepare the IL-2 monomer structure, the Protein Preparation
Wizard tool in Maestro was used to remove chain B and add explicit
hydrogen atoms to the structure. While SiteMap was unable to
identify the binding pocket of the ligand on the monomer struc-
ture, both fpocket and FTSite identified the binding pocket on the
IL-2/IL-2RA complex with a Druggability Score of 0.655.
Similarly, the BCL-XL-BAD complex (PDB: 2BZW) is inhib-
ited by the small molecule ABT-737 (PDB: 2YXJ). In addition to
the preparation needed for the IL-2 structure, the structure of
BCL-XL, an apoptosis regulator [74], contains additional water
molecules that need to be removed. The Arg-102 residue is also
missing heavy atoms on its side chain, which need to be
incorporated. All three pocket detection algorithms were able
to identify the binding pocket occupied by the ligand in the
Table 1
Pockets identified by the various detection programs
PPI Protein- SiteScore DrugScore Druggability

Protein Target structure compound (SiteMap) (SiteMap) (fpocket) FTSite
IL-2 IL-2RA 1Z92 1PY2 Undetected 0.655 Site 2
BCL-XL BAD 2BZW 2YXJ 0.793 0.824 0.365, 0.897 Site 1, 2, 3
HDM2 p53 1YCR 1RV1 0.871 0.923 0.773 Site 1, 2
HPV E2 HPV E1 1TUE 1R6N Undetected 0.084 Site 1
ZipA FtsZ 1F47 1Y2F Undetected Undetected Site 3
TNF TNF 1TNF 2AZ5 Undetected Undetected Site 2
protein–protein complex, with SiteScore and DrugScore scores of

0.793 and 0.824, respectively. The fpocket Druggability scores are
0.365 and 0.897, respectively. While SiteMap was able to enclose
the real binding site in one pocket, both of the two other algo-
rithms found that the ligand spanned more than one pocket as
evidenced by the two fpocket Druggability scores. In addition to
the pockets at the interaction interface, SiteMap identified a drug-
gable allosteric pocket near the interaction interface, with a Site-
Score of 0.831 and a DrugScore of 0.776.
The mouse homolog of HDM2 was shown to bind and block
the tumor-suppressor protein p53 [75]. Crystal structures show
that this complex (PDB: 1YCR) have been inhibited by both an
imidazoline (PDB: 1RV1) and a benzodiazepine (PDB: 1T4E)
inhibitor. Similar to the IL-2 monomer, HDM2 did not have any
water molecules or missing side chains. Once again, all three algo-
rithms identified the binding site at the PPI interface. While both
SiteMap and fpocket represented the binding site as only one
pocket, with a SiteScore of 0.871, DrugScore of 0.923, and Drugg-
ability Score of 0.773, FTSite required two pockets to properly
enclose the bound ligand.
The HPV E2 protein regulates the transcription and replication
of the viral genome with the E1 protein [76, 77]. The HPV E2-E1
complex (PDB: 1TUE) is inhibited by a small molecule at the
N-terminal transactivation domain of HPV E2 (PDB: 1R6N).
Water molecules were removed from the structure and pocket
detection algorithms were applied to the monomer structure.
While SiteMap failed to identify the specific binding pocket, fpocket
identified the binding site as an unlikely druggable target, with a
Druggability Score of only 0.084. FTSite, on the other hand,
identified the proper binding site as its highest ranking pocket.
The ZipA-FtsZ complex (PDB: 1F47), which is essential in cell
division of bacteria [78], is inhibited by a small molecule (PDB:
1Y2F). To prepare the ZipA structure, water molecules were
removed from the ZipA monomer of the protein–protein complex.
Both SiteMap and fpocket failed to identify the binding pocket of
the ligand, but fpocket did identify two additional druggable pock-
ets on ZipA, with Druggability Scores of 0.631 and 0.721. FTSite
also identified the proper binding site, but only as its third ranking
pocket.
Finally, the cytokine TNFα forms a homotrimer complex
(PDB: 1TNF) that is involved in the inflammatory response [79].
This complex was shown to be disrupted by small molecules (PDB:
2AZ5), thereby inhibiting TNFα activity. Similar to the ZipA-FtsZ
complex, both SiteMap and fpocket failed to detect the binding
pocket of the protein-protein interface occupied by the small
molecule. FTSite successfully identified the pocket as its second
ranking site.
86 David Xu et al.
In addition to the pockets at the interaction interface, all three

programs identified additional pockets outside of the interface,
albeit they are unlikely to be druggable pockets based on the pocket
scoring metrics. SiteMap identified only one noninterface pocket
with a SiteScore or DrugScore greater than 0.7 among all six
protein monomers, a pocket near the interaction interface on
BCL-XL with a SiteScore of 0.831 and a DrugScore of 0.776.
FPocket on the other hand only identified two additional drug-
gable pockets on ZipA with Druggability Scores of 0.631 and
0.721. FTSite consistently identified three pockets for each of
these six proteins, but offered no measurement of its druggability.
Once a pocket is identified, molecular docking can be used to
predict the binding mode of a large number of compounds for
subsequent scoring and ranking. The reliability of molecular dock-
ing depends on the properties of the binding pocket. To illustrate
this, all six PPI inhibitors in Table 1 were extracted and re-docked
using AutoDock Vina. Default parameters were used. Typically, a
docking program will generate multiple binding poses along with a
score associated with each pose. The predicted binding pose with
the lowest Vina binding energy is selected (Table 2). The root-
mean-square deviation (RMSD) comparing the position of het-
eroatoms of the docked inhibitors to the crystal structure are listed
in Table 2. For the first three targets with detectable pockets,
AutoDock Vina reliably reproduced the binding pose of the com-
pounds. This is evidenced by an RMSD that was less than 1.5 Å in
each case. In contrast, molecular docking of the compounds that
bind to HPV E2, ZipA, and TNF resulted in binding modes that
did not agree with the crystal structure (RSMD >5 Å). It is of
interest to note that the binding mode of the compounds that bind
to well-defined pockets (IL-2, BCL-XL, and HDM2) was reliably
predicted by molecular docking. The absence of a well-defined
pocket for HPV E2, ZipA, and TNF posed a significant challenge
to the docking program. These results highlight the strengths and
limitations of using molecular docking for the search of small
molecules that bind at PPIs. New scoring functions are needed to
guide molecular docking and scoring of small molecules that bind
to protein interfaces with shallow pockets.
4 Notes
1. If a large number of protein sequences are being queried, it is

often easier to install a local copy of BLAST+ and the PDB
amino acid (pdbaa) database and go through the sequences in
an iterative manner. The pdbaa database returns amino acid
sequence queries as PDB chains, whose metadata can be
retrieved using the REST API in PDB through the use of
XML queries. This metadata can then be filtered to identify
structures that contain dimerization.
Table 2
AutoDock Vina results and RMSD compared to crystal structure
PDB Compound structure Compound name Binding energy (kcal/mol) RMSD (Å)
1PY2 SP4206 7.7 1.06
2YXJ ABT-737 9.5 1.28
Structure-Based Computational Design

1RV1 IMIDAZOLINE 6.6 0.89
87
(continued)
Table 2
88
(continued)
David Xu et al.
PDB Compound structure Compound name Binding energy (kcal/mol) RMSD (Å)
1R6N Compound 23 7.5 6.65
1Y2F Compound 1 6.4 8.21
2AZ5 SP304 8.3 5.68

2. The cutoff used in the BLAST search will only identify PDB
entries that are nearly identical to the initial sequence query. If
structures that you expect to see are not returned, lowering the
cutoff will result in a larger number of potential structures.
However, a cutoff that is too low will often yield structures
that have very low sequence identity to the original.
3. Sometimes there are no atomic coordinates for residues within
a PDB entry. Often, these missing residues occur at the begin-
ning or end of the PDB sequence, or in intrinsically disordered
regions, which are difficult to crystallize. If these missing resi-
dues are within the proximity of a known binding cavity, using
an alternative crystal structure from the BLAST search can yield
a PDB structure that has these coordinates available. Other-
wise, homology modeling can be used to thread the missing
residues into a homologous structure. This can be accom-
plished with software packages such as MODELLER or Prime.
4. During docking, the ligands can dock outside of the specified
binding pocket if the grid is too large. In this case, either visual
inspection or a computational approach can determine which
ligands, if any, are outside the binding pocket. In the computa-
tional approach, the distance between the center of mass of the
binding pocket and the center of mass of the docked ligand can
be used to determine which ligands are outside the binding
cavity.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
References
1. Vidal M, Cusick ME, Barabasi AL (2011) 6. Li L, Bum-Erdene K, Baenziger PH et al

Interactome networks and human disease. (2010) BioDrugScreen: a computational drug
Cell 144:986–998 design resource for ranking molecules docked
2. Ngounou Wetie AG, Sokolowska I, Woods AG to the human proteome. Nucleic Acids Res 38:
et al (2013) Protein–protein interactions: D765–D773
switch from classical methods to proteomics 7. Huang YJ, Hang D, Lu LJ et al (2008) Target-
and bioinformatics-based approaches. Cell ing the human cancer pathway protein interac-
Mol Life Sci 71:205–228 tion network by structural genomics. Mol Cell
3. White AW, Westwell AD, Brahemi G (2008) Proteomics 7(10):2048–2060
Protein–protein interactions as targets for 8. Mosca R, Ceol A, Aloy P (2013) Interac-
small-molecule therapeutics in cancer. Expert tome3D: adding structural details to protein
Rev Mol Med 10:e8 networks. Nat Methods 10:47–53
4. Berman HM, Westbrook J, Feng Z et al (2000) 9. Li L, Meroueh SO (2008) Receptor-ligand
The protein data bank. Nucleic Acids Res interactions in biological systems. In: Encyclo-
28:235–242 pedia for the life sciences. Wiley, London, p. 19.
5. Lipman DJ, Pearson WR (1985) Rapid and http://onlinelibrary.wiley.com/book/10.
sensitive protein similarity searches. Science 1002/9780470048672/homepage/
227:1435–1441 EditorsContributors.html
90 David Xu et al.
10. Halgren T (2007) New method for fast 26. Geppert T, Bauer S, Hiss JA et al (2012)
and accurate binding-site identification and Immunosuppressive small molecule discovered
analysis. Chem Biol Drug Des 69:146–148 by structure-based virtual screening for inhibi-
11. Le Guilloux V, Schmidtke P, Tuffery P (2009) tors of protein–protein interactions. Angew
Fpocket: an open source platform for ligand Chem Int Edit 51:258–261
pocket detection. BMC Bioinformatics 10:168 27. Kuntz ID, Blaney JM, Oatley SJ et al (1982) A
12. Kuhn D, Weskamp N, Hullermeier E et al geometric approach to macromolecule-ligand
(2007) Functional classification of protein interactions. J Mol Biol 161:269–288
kinase binding sites using cavbase. ChemMed- 28. Makino S, Kuntz ID (1997) Automated flexi-
Chem 2:1432–1447 ble ligand docking method and its application
13. Ngan CH, Hall DR, Zerbe B et al (2012) for database search. J Comput Chem
FTSite: high accuracy detection of ligand bind- 18:1812–1825
ing sites on unbound protein structures. Bioin- 29. Goodsell DS, Olson AJ (1990) Automated
formatics 28:286–287 docking of substrates to proteins by simulated
14. Halgren TA (2009) Identifying and character- annealing. Proteins 8:195–202
izing binding sites and assessing druggability. J 30. Jones G, Willett P, Glen RC et al (1997) Devel-
Chem Inf Model 49:377–389 opment and validation of a genetic algorithm
15. Schmidtke P, Barril X (2010) Understanding for flexible docking. J Mol Biol 267:727–748
and predicting druggability. A high- 31. Jones G, Willett P, Glen RC (1995) Molecular
throughput method for detection of drug recognition of receptor sites using a genetic
binding sites. J Med Chem 53:5858–5867 algorithm with a description of desolvation.
16. Stark C, Breitkreutz BJ, Reguly T et al (2006) J Mol Biol 245:43–53
BioGRID: a general repository for interaction 32. Rarey M, Kramer B, Lengauer T et al (1996) A
datasets. Nucleic Acids Res 34:D535–D539 fast flexible docking method using an incre-
17. Licata L, Briganti L, Peluso D et al (2012) mental construction algorithm. J Mol Biol
MINT, the molecular interaction database: 261:470–489
2012 update. Nucleic Acids Res 40:D857–D861 33. Friesner RA, Banks JL, Murphy RB et al (2004)
18. UniProt C (2012) Reorganizing the protein Glide: a new approach for rapid, accurate dock-
space at the Universal Protein Resource (Uni- ing and scoring. 1. Method and assessment of
Prot). Nucleic Acids Res 40:D71–D75 docking accuracy. J Med Chem 47:1739–1749
19. Porter CT, Bartlett GJ, Thornton JM (2004) 34. Halgren TA, Murphy RB, Friesner RA et al
The catalytic site atlas: a resource of catalytic (2004) Glide: a new approach for rapid, accu-
sites and residues identified in enzymes using rate docking and scoring. 2. Enrichment fac-
structural data. Nucleic Acids Res 32: tors in database screening. J Med Chem
D129–D133 47:1750–1759
20. Leach AR, Gillet VJ, Lewis RA et al (2009) 35. Pierce BG, Hourai Y, Weng Z (2011) Acceler-
Three-dimensional pharmacophore methods ating protein docking in ZDOCK using an
in drug discovery. J Med Chem 53:539–558 advanced 3D convolution library. PLoS One
21. Hubbard RE (2011) Structure-based drug dis- 6:e24657
covery and protein targets in the CNS. Neuro- 36. McGann M (2011) FRED pose prediction and
pharmacology 60:7–23 virtual screening accuracy. J Chem Inf Model
22. Cheng T, Li Q, Zhou Z et al (2012) Structure- 51:578–596
based virtual screening for drug discovery: a 37. Pedretti A, Villa L, Vistoli G (2004) VEGA –
problem-centric review. AAPS J 14:133–141 an open platform to develop chemo-bio-infor-
23. Khanna M, Wang F, Jo I et al (2011) Targeting matics applications, using plug-in architecture
multiple conformations leads to small molecule and script programming. J Comput-Aided Mol
inhibitors of the uPAR·uPA protein–protein Des 18:167–173
interaction that block cancer cell invasion. 38. Thomsen R, Christensen MH (2006) MolDock:
ACS Chem Biol 6:1232–1243 a new technique for high-accuracy molecular
24. Scheper J, Guerra-Rebollo M, Sanclimens G docking. J Med Chem 49:3315–3321
et al (2010) Protein–protein interaction 39. Abagyan R, Totrov M, Kuznetsov D (1994)
antagonists as novel inhibitors of non-canonical ICM – a new method for protein modeling
polyubiquitylation. PLoS One 5:e11403 and design: applications to docking and
25. Koes D, Khoury K, Huang Y et al (2012) structure prediction from the distorted native
Enabling large-scale design, synthesis and vali- conformation. J Comp Chem 15:488–506
dation of small molecule protein–protein 40. Trott O, Olson AJ (2010) AutoDock Vina:
antagonists. PLoS One 7:e32839 improving the speed and accuracy of docking
with a new scoring function, efficient optimiza- model for HIV-1 integrase. J Med Chem
tion, and multithreading. J Comput Chem 43:2100–2114
31:455–461 54. Kukol A (2011) Consensus virtual screening
41. Obiol-Pardo C, Alcarraz-Vizán G, Cascante M approaches to predict protein ligands. Eur J
et al (2012) Diphenyl urea derivatives as inhi- Med Chem 46:4661–4664
bitors of transketolase: a structure-based virtual 55. Irwin JJ, Sterling T, Mysinger MM et al (2012)
screening. PLoS One 7:e32276 ZINC: a free tool to discover chemistry for
42. Dessal AL, Prades R, Giralt E et al (2011) biology. J Chem Inf Model 52:1757–1768
Rational design of a selective covalent modifier 56. Greenwood JR, Calkins D, Sullivan AP et al
of G protein βγ subunits. Mol Pharm 79:24–33 (2010) Towards the comprehensive, rapid,
43. Trosset J-Y, Dalvit C, Knapp S et al (2006) and accurate prediction of the favorable tauto-
Inhibition of protein–protein interactions: the meric states of drug-like molecules in aqueous
discovery of druglike β-catenin inhibitors by solution. J Comput Aid Mol Des 24:591–604
combining virtual and biophysical screening. 57. Cramer CJ, Truhlar DG (1992) An SCF solva-
Proteins 64:60–67 tion model for the hydrophobic effect and
44. Gr€uneberg S, Stubbs MT, Klebe G (2002) absolute free energies of aqueous solvation.
Successful virtual screening for novel inhibitors Science 256:213–217
of human carbonic anhydrase: strategy and 58. Cramer CJ, Truhlar DG (1992) AM1-SM2 and
experimental confirmation. J Med Chem PM3-SM3 parameterized SCF solvation mod-
45:3588–3602 els for free energies in aqueous solution.
45. Elokely KM, Doerksen RJ (2013) Docking J Comput Aided Mol Des 6:629–666
Challenge: Protein Sampling and Molecular 59. Hawkins PCD, Skillman AG, Nicholls A
Docking Performance. J Chem Inf Model (2006) Comparison of shape-matching and
53:1934–1945 docking as virtual screening tools. J Med
46. Lill MA, Winiger F, Vedani A et al (2005) Chem 50:74–82
Impact of Induced Fit on Ligand Binding to 60. Tetko IV, Gasteiger J, Todeschini R et al
the Androgen Receptor: A Multidimensional (2005) Virtual computational chemistry
QSAR Study To Predict Endocrine-Disrupting laboratory-design and description. J Comput-
Effects of Environmental Chemicals. J Med Aided Mol Des 19:453–463
Chem 48:5666–5674 61. Gaulton A, Bellis LJ, Bento AP et al (2012)
47. Sherman W, Day T, Jacobson MP et al (2005) ChEMBL: a large-scale bioactivity database for
Novel procedure for modeling ligand/receptor drug discovery. Nucleic Acids Res 40:
induced fit effects. J Med Chem 49:534–553 D1100–D1107
48. Arooj M, Sakkiah S, Kim S et al (2013) A com- 62. Li Q, Cheng T, Wang Y et al (2010) PubChem
bination of receptor-based pharmacophore as a public resource for drug discovery. Drug
modeling & QM techniques for identification Discov Today 15:1052–1057
of human chymase inhibitors. PLoS One 8: 63. Knox C, Law V, Jewison T et al (2011) Drug-
e63030 Bank 3.0: a comprehensive resource for
49. Zhou S, Li Y, Hou T (2013) Feasibility of “omics” research on drugs. Nucleic Acids Res
using molecular docking-based virtual screen- 39:D1035–D1041
ing for searching dual target kinase inhibitors. J 64. Liu T, Lin Y, Wen X et al (2007) BindingDB: a
Chem Inf Model 53:982–996 web-accessible database of experimentally
50. Li Y, Kim DJ, Ma W et al (2011) Discovery of determined protein–ligand binding affinities.
novel checkpoint kinase 1 inhibitors by virtual Nucleic Acids Res 35:D198–D201
screening based on multiple crystal structures. 65. Still WC, Tempczyk A, Hawley RC et al (1990)
J Chem Inf Model 51:2904–2914 Semianalytical treatment of solvation for
51. Isvoran A, Badel A, Craescu C et al (2011) molecular mechanics and dynamics. J Am
Exploring NMR ensembles of calcium binding Chem Soc 112:6127–6129
proteins: perspectives to design inhibitors of 66. Luo R, David L, Gilson MK (2002) Acceler-
protein–protein interactions. BMC Struct Biol ated Poisson–Boltzmann calculations for static
11:24 and dynamic systems. J Comput Chem
52. Knegtel RMA, Kuntz ID, Oshiro CM (1997) 23:1244–1253
Molecular docking to ensembles of protein 67. Eldridge MD, Murray CW, Auton TR et al
structures. J Mol Biol 266:424–440 (1997) Empirical scoring functions: I. The
53. Carlson HA, Masukawa KM, Rubins K et al development of a fast empirical scoring func-
(2000) Developing a dynamic pharmacophore tion to estimate the binding affinity of ligands
92 David Xu et al.
in receptor complexes. J Comput Aided Mol 73. Malek TR (2003) The main function of IL-2 is
Des 11:425–445 to promote the development of T regulatory
68. Li L, Wang B, Meroueh SO (2011) Support cells. J Leukoc Biol 74:961–965
vector regression scoring of receptor–ligand 74. Willis S, Day CL, Hinds MG et al (2003) The
complexes for rank-ordering and virtual Bcl-2-regulated apoptotic pathway. J Cell Sci
screening of chemical libraries. J Chem Inf 116:4053–4056
Model 51:2132–2138 75. Moll UM, Petrenko O (2003) The MDM2-
69. Muegge I, Martin YC (1999) A general and p53 interaction. Mol Cancer Res 1:1001–1008
fast scoring function for protein ligand 76. Muller M, Demeret C (2012) The HPV E2-
interactions: a simplified potential approach. host protein–protein interactions: a complex
J Med Chem 42:791–804 hijacking of the cellular network. Open Virol J
70. Gohlke H, Hendlich M, Klebe G (2000) 6:173–189
Knowledge-based scoring function to predict pro- 77. Hughes FJ, Romanos MA (1993) E1 protein
tein–ligand interactions. J Mol Biol 295:337–356 of human papillomavirus is a DNA helicase/
71. Li L, Khanna M, Jo I et al (2011) Target- ATPase. Nucleic Acids Res 21:5817–5823
specific support vector machine scoring in 78. Pazos M, Natale P, Vicente M (2013) A specific
structure-based virtual screening: computa- role for the ZipA protein in cell division: stabi-
tional validation, in vitro testing in kinases, lization of the FtsZ protein. J Biol Chem
and effects on lung cancer cell proliferation. 288:3219–3226
J Chem Inf Model 51:755–759 79. Locksley RM, Killeen N, Lenardo MJ (2001)
72. Wells JA, McClendon CL (2007) Reaching for The TNF and TNF receptor superfamilies:
high-hanging fruit in drug discovery at protein– integrating mammalian biology. Cell 104:
protein interfaces. Nature 450:1001–1009 487–501
Chapter 6
Targeting Protein-Protein Interactions for Drug Discovery

David C. Fry
Abstract
Protein-protein interactions are associated with key activities and pathways in the cell, and in that regard are
promising targets for drug discovery. However, in terms of small molecule drugs, this promise has not been
realized. The physical nature of many protein-protein interaction surfaces renders them unable to support
binding of small drug-like molecules. In addition, there are other unique hurdles presented by this class that
make the drug development process difficult and risky. Nevertheless, success stories have begun to steadily
appear in this field. These experiences are starting to provide general strategies and tools to help overcome
the problems inherent in pursuing protein‐protein interaction targets. These lessons should improve the
rate of success as these systems are pursued in the future.
Key words Protein-protein interactions, Drug discovery, MDM2, Nutlins, Druggability, Protein
NMR, NMR screening, Screening library, Fragment-based drug discovery
1 Framing the Issue
Proteins commonly interact with other proteins. Some act as

“hubs” and have many partners. In fact, it has been estimated
that there are on the order of 600,000 protein‐protein interactions
that occur in the cell [1]. These interactions serve various purposes,
including regulation of function, step-wise transmission of a signal,
or attraction of a partner to a particular location. Protein-protein
interactions are associated with a wide variety of critical physiologi-
cal activities and pathways. Accordingly, one would expect that
malfunctions with these systems could be linked to a number of
important disease states. Beneficial modulation of protein‐protein
interactions could, therefore, be the mode of action of effective
drugs. However, historically, this situation has not been exploited
[2]. If one surveys the mechanisms of conventional (i.e., small
organic molecule) drugs, it is found that very few of them act by
affecting a protein‐protein interaction. Why is this so?
The physical nature of most protein‐protein interaction sur-
faces lowers the probability that a small organic compound can
93
94 David C. Fry
bind there [3–5]. These surfaces are typically large and flat, and
devoid of significant subpockets. Binding affinity is achieved by
summing up a large number of weak interactions. The interactions
are so widely spaced that a small molecule cannot adequately dupli-
cate them. However, this situation is a generalization, and not all
protein‐protein systems adopt such a strategy. Pioneering work by
Jim Wells and colleagues demonstrated that, in some cases, a lim-
ited number of amino acids mediate nearly all of the key interac-
tions that produce binding affinity. These sub-regions have been
referred to as “hot spots,” and their dimensions can be comparable
to the size of a small organic molecule [6].
Therefore, protein‐protein interaction systems that are candi-
dates for drug discovery must be judged for druggability on a
case-by-case basis. The approaches and techniques for making
such a judgment are still being refined and are not fully reliable
yet. As a consequence, embarking on a drug discovery program
targeting a protein‐protein interaction system is still a high-risk
undertaking. An unwillingness to take on such risk has meant that
protein‐protein targets have not received the same degree of active
pursuit as other target classes. This is one of the major reasons for
the paucity of marketed drugs with a mode of action that involves
modulation of a protein‐protein interaction.
Another reason for lack of success against this target class is the
inability to overcome the unique hurdles that are presented by this
class during the drug discovery and development process. The field
is gaining an awareness of, and an appreciation for, these hurdles,
and is acquiring an understanding of how to overcome them, as
success stories steadily appear in this area. It has been helpful to
examine, at a molecular level, cases in which a drug-like small
molecule has been successfully developed against a protein‐protein
interaction target, and a collective consideration of these examples
has led to important lessons that should improve the means by
which these targets are addressed in the future.
2 Recent Success Stories
One of the more recent examples, and a case in which a small

molecule inhibitor of a protein‐protein interaction has progressed
fully to clinical trials, is the development of the “Nutlin” series of
compounds, which bind to MDM2 and inhibit its interaction with
p53 [7]. Overexpression of MDM2 is one way in which certain
cancer cells evade the apoptosis-triggering activity of p53, and the
Nutlins have been shown capable of activating cell cycle arrest and
apoptosis, and of inhibiting growth of human tumor xenografts.
The Nutlin case exemplifies several of the key issues that need to be
considered when approaching a protein‐protein interaction target,
and reveals important attributes of the system that, if understood
and exploited correctly, can enhance the chance of success.
PPIs in Drug Discovery 95
Fig. 1 Series of one single and four superimposed X-ray structures showing how the p53 peptide binds to
MDM2 [8] (a), and how various small molecule inhibitors [7, 34–36] are able to mimic its binding strategy
(b–e). MDM2 is depicted as a tan surface. The p53 peptide backbone is shown as a green ribbon, and key side
chains (Phe19, Trp23, and Leu26) as green sticks. The inhibitors are shown in stick form. All structural figures
in this chapter were prepared using the PyMol Molecular Graphics System (Schroedinger, LLC)
An initial structure of a complex between MDM2 and a peptide

representing p53 [8] indicated that the interaction occurs at a
single binding cleft on the MDM2 protein. The p53 peptide adopts
an alpha helical conformation, and inserts three hydrophobic
side chains (Phe19, Trp23, and Leu26) into subpockets of
MDM2 (Fig. 1a). An assessment of the druggability of this site,
using early computational tools, suggested that the target would be
challenging but worth pursuing [9].
A high-throughput screen of MDM2 was conducted, using a
library of diverse drug-like small molecule compounds, and numer-
ous active hits were obtained. Protein-observe NMR, using an
isotopically labeled version of a stabilized construct of MDM2,
was employed to assess the authenticity of the hits [10]. This was
a critical step, because all but two of the hits were found to be false
96 David C. Fry
positives. Extensive synthetic chemistry was applied, and one of the

lead classes (the Nutlin series) was able to be optimized into more
potent derivatives, and ultimately into a clinical candidate [11].
During this process, an NMR structure [12], and X-ray structures
[7], of Nutlins bound to MDM2 were obtained (Fig. 1b), and they
revealed the means by which the small molecules were able to
replicate the binding strategy of the crucial segment of p53.
The imidazoline core scaffold, which is the primary feature of
the Nutlins, is able to direct substituents into the three subpockets
of MDM2 that are normally occupied by hydrophobic side chains
of p53. Trp23 and Leu26 are each mimicked by a halogenated
phenyl group, and Phe19 is mimicked by an ethoxy group. It is
important to note that there is no exact matching of chemotype—
for example, an aliphatic group can substitute for an aromatic side
chain, and an aromatic group can fulfill the role of an aliphatic side
chain. Also, the trajectories by which the small molecule substitu-
ents enter the subpockets are completely different from those
implemented by the side chains.
The backbone of p53 is not directly replicated at all by the small
molecule. While the protein must make use of a series of hydrogen
bonds to attain rigidity, in this case along an alpha helix, the
imidazoline scaffold possesses inherent rigidity. Further, in terms
of geometry, this scaffold is able to economically span a segment of
alpha helix that is eight residues in length.
The Nutlins feature an appendage emanating from the N1
atom of the imidazoline core that appears to project out beyond
contact with the binding cleft, into the solvent. The chemical
composition of this appendage was found to be crucial with respect
to influencing potency, yet its precise role cannot be gleaned from
the structure. One can speculate that it helps as a general shield
keeping solvent away from the binding cleft, or that it may sterically
direct a key substituent into the Phe subpocket. A third alternative,
that it contributes to a local dipole and in that way influences
orientation and affinity, has not been studied.
In the period since the discovery of the Nutlins, MDM2 inhi-
bitors representing other chemical classes have been reported [13].
It is instructive to compare how these various scaffolds bind to the
same site, to appreciate the variety of binding interactions that can
be made there, and to observe what kinds of substituent groups can
participate in these interactions (Fig. 1c–e).
The core scaffolds of the various inhibitors are quite different
chemically and with regard to shape. Nevertheless, the compounds
share certain binding principles. In every case, the p53 backbone is
not directly replicated, and in every case the small molecule inhibi-
tor fills all three subpockets of MDM2 with hydrophobic moieties.
Into the Trp subpocket, the inhibitors insert a chloro-indole or
halogenated phenyl group, and these rings all attain the same
orientation as the parent Trp side chain. The Leu subpocket is filled
by all inhibitors in a similar manner, through use of a chlorophenyl,

bromophenyl, or chlorobenzyl group. None of these groups corre-
spond to the branched aliphatic nature of the parent Leu side chain.
Also, the trajectories of these inhibitor rings into the subpocket,
while similar to each other, are basically perpendicular to that of the
Leu side chain. Occupancy of the Phe subpocket is the most varied
of all, and is achieved by the following variety of chemotypes:
ethoxy, iodophenyl, benzyl edge from a four-ring system, and
phenyl. Similar to the situation with the Leu subpocket, the inhibi-
tor substituents enter with different trajectories from that of the
parent Phe side chain, and they end up oriented perpendicular to it.
Recently, a potent MDM2 inhibitor from yet another chemical
class has been discovered, and this series binds in a completely
unique manner [14]. This class emerged from a screen against a
protein called MDMX, which is highly similar to MDM2, binds to
p53 in a comparable manner, and participates in regulation of p53
function in the cell. The inhibitor binds in an equipotent manner to
both proteins, and with a similar strategy. As exemplified by MDM2
(Fig. 2), the inhibitor induces the protein to dimerize, and the result
is a symmetric complex consisting of two MDM2 molecules and two
inhibitor molecules. Each inhibitor molecule spans the two proxi-
mal MDM2 binding sites, by inserting its difluorophenyl group into
the Trp subpocket on one side, and positioning its indolyl-
hydantoin group into the Phe subpocket of the other MDM2
molecule. This orientation allows a stacking interaction between
Fig. 2 MDM2/MDMX dual inhibitor that causes dimerization. Depicted is the

X-ray structure of the complex of MDM2 with the p53 peptide [7] superimposed
with the structure of the inhibitor [14]. The peptide is visualized as described for
Fig. 1, and the inhibitor is in stick form with its carbons colored white. Both
inhibitor molecules from the dimer are shown, but only one of the MDM2
molecules is shown, as a tan surface
98 David C. Fry
the two indolyl groups, and this energetically favorable situation is

reinforced by further stacking with a Tyr63 ring from each protein
molecule. The Leu subpocket is basically ignored by this inhibitor.
While it is not unusual to encounter a ligand that bridges two
binding sites across protein molecules that normally dimerize, the
present case is a heretofore unobserved instance of an inhibitor
pulling together two molecules of a protein domain that is normally
monomeric, and forming a ligand/ligand dimer in the process.
In summary, although at the beginning of the program it was
wondered whether any small molecule could be found that would
bind tightly to MDM2 and effectively inhibit its interaction with
p53, there are now several different classes of molecule which have
achieved this goal.
With regard to other protein‐protein interaction systems, there
have been a number of reports describing successful discovery of
small drug-like modulators. These cases have been collected and
described in a series of review articles [15–20]. There are also newer
reports that have appeared since the latest survey [21, 22]. Some of
these modulators bind to a remote site and act allosterically, and
some are directly competitive with respect to the partner protein. If
one collectively considers all of the successful cases, some general
principles emerge:
1. For instances involving direct competition, successful cases
usually feature early identification of a peptide that can ade-
quately replicate the binding of the entire partner protein.
2. The protein‐protein modulator is usually higher in molecular
weight than a typical drug. This is likely due to the need to
participate in multiple interactions that are spaced relatively far
apart, due to the large surface area typically observed at a
protein‐protein interface.
3. The modulator is also usually more three-dimensional than a
typical drug [23], and often more rigid. The three-
dimensionality can be attained by stacking or by the inherent
geometric constraints of the central scaffold. As a consequence
of having a three-dimensional molecule seated in a shallow
cleft, parts of the modulator may stick out beyond the perime-
ter of the binding cleft. For flexible protein targets, rigidity may
help allosteric modulators induce subpocket formation at their
binding site.
4. For the modulators that act competitively, there is usually little
need to replicate the backbone of the partner protein, just the
critical side-chains. An organic scaffold can deliver these sub-
stituents in a much more economical way than can a peptide
backbone.
5. There is substantial leeway with respect to the chemotypes that
can serve as effective core scaffolds, and with respect to
substituents that can mimic the key interacting side chains of

the natural partner protein. Also, entry vectors of these sub-
stituents into subpockets at the binding site can be completely
different from, and even perpendicular to, those displayed by
the side chains of the natural partner.
3 Current Best Approaches
Given the experience gained from many attempts over the years at
drug discovery involving protein‐protein systems, and the
recent availability of numerous successful examples in this area, a
generalized optimal approach can be laid out for an attack on
a protein‐protein target:
3.1 Choosing The target will be chosen based on biology considerations, but then
Targets/Assessing must be quickly put under scrutiny at a molecular level [24]. If a
Druggability drug-like small molecule modulator is already known for the system
in question, then by definition it is druggable. Otherwise, if struc-
tural information is available for the target protein, a druggability
assessment can be made by computational methods [25–28]. These
methods have not yet proven to be completely accurate, but should
be a reasonable guide at the extremes—for example, a target with a
complete lack of subpockets at the interface is almost certain to be
highly challenging. Given the unreliability of the computational
methods, an inexpensive experimental method for assessing drugg-
ability is advisable. The most established approach is to perform a
fragment screen. It has been shown in two retrospective studies that
there is a correlation between the inability to find fragment hits for
a target and its eventual failure in a small molecule drug discovery
program.
3.2 The Role of A helpful first step is identification of a peptide that can serve as a
Peptides fully competent surrogate for the entire partner protein. This can
be accomplished via design, based on structural or mutagenesis
information; by phage display; or by a screen of a peptide library.
Once obtained, a surrogate peptide can serve as a useful tool for
assay development and calibration, and for structural studies. One
should even consider developing the peptide itself into a drug. The
methodologies for doing this represent an area of intense current
interest [29], but an exposition is beyond the scope of this article.
3.3 Computational Experimental screens are expensive, and any method that can effi-
Screening and De Novo ciently focus such a screen, or bypass it altogether, is worthwhile. If
Design sufficient structural information is available on the target, a virtual
screen can be carried out. This requires the preparation of a virtual
compound library, where choices must be made about the confor-
mations of the small molecules that will be sampled. Also, it is
100 David C. Fry
desirable, although costly CPU-wise, to allow for flexibility in the

target protein during the virtual screen, and to somehow incorpo-
rate the influence of solvent. Finally, robust and reliable docking
and scoring functions must be available to predict the affinity of
compounds during the screening process. The resulting hits can be
followed up in experimental assays, or can be the springboard for a
focused screen limited to related chemotypes. The current state of
affairs is that virtual screening has had sporadic successes, but few
would depend solely on it for a high priority target. However, the
reliability of virtual screening should improve steadily, and because
it is so much more cost efficient, the hope is that it will ultimately be
able to fully replace experimental random screening.
Structural information on the target complexed with the part-
ner protein, a peptide, or some other lead molecule, should allow
the de novo design of alternate scaffolds that could be more drug-
like and possibly more potent. The ability of experts in the field to
design an active small organic molecule directly from a bound
peptide structure is still not very advanced. More success has been
realized by designing alternate scaffolds from bound small mole-
cules. This is another case where improvements in methodologies
are expected, and ultimately will lead to major cost savings, but at
the moment the de novo design approaches are not robust enough
to allow the abandonment of experimental screening.
3.4 Experimental Screening involves utilization of a compound library and a method

Screening for detecting when interaction of a compound with the target
protein has occurred. There are a variety of methods available for
detection. Some are indirect, observing an activity that is triggered
by the protein‐protein interaction and monitoring whether added
compounds reduce this activity. Others measure binding directly. In
this category, labels on the proteins can be used to report that they
are bound together, and to monitor whether addition of com-
pound has disrupted the complex. Alternatively, there are label-
free methods that employ just the member of the protein pair that is
the target of interest, measure an inherent property of the target
protein, and detect perturbations caused by binding of a com-
pound. Such methods include SPR, Tm-shift, and NMR. In these
cases, hit compounds must be checked in a follow-up assay to see if
their binding is consequential—that is, whether they are actually
modulating the protein‐protein interaction. Another approach is to
utilize a labeled peptide ligand and monitor competition by added
compounds.
Most of the screening techniques involve some sort of manipu-
lation of the native protein, such as immobilization on a matrix, or
attachment of a label or a fusion partner, and any of these can
cause falsification of signal. Screening campaigns involving a
protein‐protein interaction target are notorious for producing
a high rate of false positives. Screening results are expressed as an
activity curve, and for well-behaved targets, such as kinases and

other enzymes, the compounds in the high-activity tail of the curve
are usually real—while in the case of protein‐protein interactions,
these are usually false, and the real hits are buried in the main part of
the curve. This high false positive rate probably occurs because
protein‐protein interactions involve exposed hydrophobic surfaces,
and these can allow opportunistic binding—that is, weak transient
binding that is insufficient for further chemical optimization, or for
formation of co-crystals, but that nonetheless produces a detectable
signal. It is essential, therefore, that following a screen of a protein‐
protein target, some means of sorting the reals from the false
positives is applied. This could consist of performing multiple
screens using different techniques and looking for hits that are
agreed upon by all the methods. An even better approach, if possi-
ble, would be to apply one or both of the two “gold standard” hit
verification techniques that are essentially free of false positive read-
outs—protein-observe NMR and X-ray crystallography. One extra
advantage of these two techniques is that they reveal the location of
binding, allowing a distinction to be made between allosteric effects
and a directly competitive mechanism.
Deciding on a screening strategy requires an analysis of numer-
ous parameters, and ultimately some level of compromise on the
issues they present. In general, methods that have the advantages of
high speed, high sensitivity, and small volumes leading to low
protein consumption usually have the disadvantages of high cost
of robots and consumables, and the need to format the compound
library in a specific way. Methods that are more accurate are usually
slower, less sensitive, require more protein, and are more demand-
ing with regard to the purity and physical state of the protein. It
should be added that, when considering overall cost and speed for
the less accurate “high-throughput” methods, one must add the
resources and time needed to sort out the true positives. It has not
been unusual for optimization chemistry to begin on a false lead,
and for the inability to obtain meaningful SAR to be the basis of
making the final judgment, which is a costly mode of operating.
3.5 Protein-Protein It has already been noted that modulators of protein‐protein inter-
Focused Screening actions have unique properties, in particular that they tend to be
Library larger and more three-dimensional than typical drugs. We have
observed in screens of protein‐protein targets that the hits are
statistically more three-dimensional than the hits from screens
involving other target classes [23]. These extremes of size and
shape tend to be underrepresented in typical corporate compound
libraries. To remedy this, we have built a “PPI Library,” which is a
collection of compounds for screening that have substantial three-
dimensionality. This library has already shown promise by produc-
ing authenticated hits against protein‐protein interaction targets
that had yielded very low hits rates in earlier screens using our
conventional library [30].
102 David C. Fry
3.6 Fragment Screening libraries have traditionally consisted of compounds in the

Approach molecular weight range 200–500 Da. However, a strategy
employing libraries comprised only of small compounds, the “frag-
ment-based approach” [31], has been gaining in popularity. Practi-
tioners seem to have settled on a similar set of characteristics for the
fragments composing their libraries, featuring a molecular weight
range of 100–300 Da. Protein-protein interaction systems, how-
ever, represent a unique class of drug target, and it is an open
question whether fragments meant to serve as potential leads for
protein‐protein interaction targets should also have properties dis-
tinct from those of conventional fragments. One step toward
answering that question would be to deconstruct known successful
protein‐protein inhibitors into successively smaller fragments, until
the smallest active fragment for each is identified, and then derive
the key properties that are shared by the successful fragment leads.
Such a deconstruction has already been applied to ABT-737, a
potent inhibitor of the Bcl-2 family, which has a molecular weight
of 813 Da. In a retrospective study, compounds comprising por-
tions of ABT-737 were gathered and were checked for activity, and
the smallest piece that still exhibited binding was identified and
found to have a molecular weight of 293 Da [32]. We have system-
atically deconstructed a Nutlin (MW ¼ 728 Da) into successively
smaller fragments, and measured the ability of these fragments to
bind to MDM2 [33] (Fig. 3). Binding activity has been verified and
examined using protein-observe NMR and X-ray crystallography. In
this system, the smallest active fragment has a molecular weight of
305 Da. In both cases, where the parent had a high molecular
weight typical of protein‐protein modulators, the smallest active
fragment was found to be at the highest limit of the molecular
weight range that normally defines fragments. This suggests that a
“PPI Fragment Library” should be a unique entity that contains
relatively larger fragments, and possibly other specialized proper-
ties. With regard to three-dimensionality, we have observed in frag-
ment screens against protein‐protein interaction targets that the hit
sets were no more three-dimensional than those obtained against
other target classes. So, unlike the final optimized compounds in a
protein‐protein program, at the fragment level the leads are not
exceptionally three-dimensional.
3.7 Optimization Leads must undergo chemical optimization to acquire greater

potency, and to take on the properties that will allow performance
as a drug—bioavailability, stability, and lack of toxicity. These
requirements are not different when the target is a protein‐protein
system. However, the unique properties of the compounds that
tend to be selected for this target class may present heightened
challenges.
Structure-guided design during lead optimization is now a
fairly common approach, but may be particularly important in the
Fig. 3 Deconstruction of a Nutlin into fragments. The parent molecule is shown at the top as a chemical
structure and, to the immediate right, as a simplified schematic depiction. The various fragments are shown
below in schematic form, and their binding behavior to MDM2 as assessed by protein-observe NMR is given,
expressed as “Yes” or “No”
case of a protein‐protein target. These binding sites present fewer

clear opportunities for making additional interactions—such as
hydrogen bond partners to engage with—and fewer obvious
nearby subpockets to try to fill. Further, the protein site is likely
to be more mobile, presenting a “moving target.” So optimization
steps are more speculative and exploratory, and it is more critical for
the project chemists to have a steady source of feedback to under-
stand what is being accomplished and what next modifications
appear promising. Also, the structural work can verify that increas-
ing activity is still grounded in authentic binding, and not due to
the introduction of an unwanted mechanism such as higher stoi-
chiometry or unfolding of the protein.
It has been stated that successful compounds in this target class
are larger than what is typical. Also, we have observed that in
protein‐protein projects, increases in potency tend to be accompa-
nied by increases in three-dimensionality. At first glance, higher
molecular weight and greater chemical complexity would suggest
that problems may be encountered with candidates in these
104 David C. Fry
projects, with regard to solubility and intracellular availability.

While this expectation would be consistent with historic trends,
some protein‐protein modulators have been found to be clear
exceptions to these trends, so a negative judgment should not be
made a priori. Also, formulation methodology has advanced signif-
icantly and may be able to deal quite successfully with unusual
molecules.
Specificity is always an issue, since a lack thereof may lead to
toxicity. For protein‐protein targets it is hard to measure, or
even predict, the level of specificity of a compound. Although
“protein‐protein” is a clearly defined class, it does not carry with
it a common active site architecture, as does a class like “kinases.”
There is no pre-established panel for doing assays that adequately
represents coverage of the protein‐protein universe. One source of
worry is that proteins can employ “anti-binding” elements to
achieve specificity. That is, the interaction site can feature protru-
sions that will keep unwanted protein partners away, while the
desired partner will have a complimentary surface that can safely
avoid or accommodate these protrusions. A small molecule that is
already mimicking the interactions that mediate affinity is too small
to also mimic anti-binding elements. In the end, there are many
potential off-target protein‐protein interaction surfaces that a com-
pound could inadvertently bind to, and these cannot be pre-
identified, so it is impossible to design against them. All one can
do is take a clinical candidate into the stage of in vivo studies and
hope that toxicity due to nonspecificity does not appear.
4 Future Outlook
The field of targeting protein‐protein interactions for drug discov-

ery holds considerable promise. The numbers alone indicate that
many viable targets must exist in this class. The lack of progress in
this area has been due largely to an unwillingness to gamble
resources on these high-risk targets. However, emerging success
stories are starting to indicate ways to reduce the risk.
One overriding lesson is that candidate protein‐protein inter-
action targets must be examined on a case-by-case basis. While
many protein‐protein surfaces may be too large and featureless to
be druggable by a small molecule, this is a blanket characterization
that does not apply to all. As the reliability for predicting and
measuring druggability improves, the trepidation toward pursuing
these targets will be alleviated. Also, as important diseases continue
to resist drug discovery, the failures on more traditional target
classes will necessitate a move toward more difficult targets.
Following a commitment to invest in protein‐protein
programs, improved experimental approaches and methodologies
will increase the rate of success. Strategies will be acquired for
overcoming the special hurdles presented by this class. From

reducing the false positive rate in initial screening, to finding ways
of formulating optimized candidates for in vivo work, the discovery
process will become much more efficient. More reliable computa-
tional methods will allow virtual steps to replace experimental steps,
resulting in significant cost savings.
As more drug discovery chemistry is performed on protein‐
protein targets, compound libraries will begin to better match this
target class, and cover more extensively the compound space that
fits these binding sites, just as occurred historically with other
popular target classes. Better-matched libraries will lead to higher
success rates in screening, and faster expansion around hits.
In conclusion, the field of protein‐protein interactions presents
an attractive opportunity for drug discovery, and high optimism is
justified for future engagements with this challenging, but captivat-
ing, target class.
References
1. Stumpf MPH, Thorne T, de Silva E, Stewart R, Adams PD (eds) Protein‐protein interactions:
An HJ, Lappe M, Wiuf C (2008) Estimating a molecular cloning manual. Cold Spring
the size of the human interactome. Proc Nat Harbor Laboratory, Cold Spring Harbor, NY,
Acad Sci 105:6959–6964 pp 893–906
2. Overington JP, Al-Lazikani B, Hoopkins AL 10. Fry DC, Graves BJ, Vassilev LT (2005) Devel-
(2006) How many drug targets are there? Nat opment of E3-substrate (MDM2-p53) binding
Rev Drug Disc 5:993–996 inhibitors: structural aspects. Methods Enzy-
3. Fry DC (2006) Protein‐protein interactions as mol 399C:622–633
targets for small molecule drug discovery. Bio- 11. Vu BT, Vassilev L (2011) Small-molecule inhi-
polymers 84:535–552 bitors of the p53-MDM2 interaction. In: Vas-
4. Wells JA, McClendon CL (2007) Reaching for silev L, Fry D (eds) Small-molecule inhibitors
high-hanging fruit in drug discovery at protein‐ of protein‐protein interactions. Springer, Ber-
protein interfaces. Nature 450:1001–1009 lin, pp 151–172
5. Doemling A (2008) Small molecular weight 12. Fry DC, Emerson SD, Palme S, Vu BT, Liu
protein‐protein interaction antagonists – an CM, Podlaski F (2004) NMR structure of a
insurmountable challenge? Curr Opin Chem complex between MDM2 and a small molecule
Biol 12:281–291 inhibitor. J Biomol NMR 30:163–173
6. Clackson T, Wells JA (1995) A hot spot of 13. Wang S, Zhao Y, Bernard D, Aguilar A,
binding energy in a hormone-receptor inter- Kumar S (2012) Targeting the MDM2-p53
face. Science 267:383–386 protein‐protein interaction for new cancer
7. Vassilev LT, Vu BT, Graves B, Carvajal D, therapies. Top Med Chem 8:57–80
Podlaski F, Filipovic Z, Kong N, Kammlott 14. Graves B, Thompson T, Xia M, Janson C,
U, Lukacs C, Klein C, Fotouhi N, Liu E Lukacs C, Deo D, DiLello P, Fry D, Garvie
(2004) In vivo activation of the p53 pathway C, Huang K, Gao L, Tovar C, Lovey A, Wanner
by small-molecule antagonists of MDM2. Sci- J, Vassilev L (2012) Activation of the p53 path-
ence 303:844–848 way by small-molecule induced MDM2 and
8. Kussie PH, Gorina S, Marechal V, Elenbaas B, MDMX dimerization. Proc Nat Acad Sci
Moreau J, Levine AJ, Pavletich NP (1996) 109:11788–11793
Structure of the MDM2 oncoprotein bound 15. Arkin MR, Wells JA (2004) Small-molecule
to the p53 tumor suppressor transactivation inhibitors of protein‐protein interactions:
domain. Science 274:948–953 progressing towards the dream. Nat Rev Drug
9. Fry DC, Graves BJ, Vassilev LT (2005) Disc 3:301–317
Exploiting protein‐protein interactions to 16. Fry DC (2008) Drug-like inhibitors of protein‐
design an activator of p53. In: Golemis EA, protein interactions: a structural examination
106 David C. Fry
of effective protein mimicry. Curr Prot Pep Sci binding pockets for drug discovery. Nucleic
9:240–247 Acids Res 38:W407–W411
17. Sperandio O, Reynes CH, Camproux AC, 28. Sugaya N, Kanai S, Furuya T (2012) Dr. PIAS
Villoutreix BO (2010) Rationalizing the chem- 2.0: an update of a database of predicted drug-
ical space of protein‐protein interaction inhibi- gable protein‐protein interactions. Database
tors. Drug Disc Today 15:220–229 2012:bas034
18. Higueruelo AP, Schreyer A, Bickerton GRJ, 29. Vlieghe P, Lisowski V, Martinez J, Khrestcha-
Pitt WR, Groom CR, Blundell TL (2009) tisky M (2009) Synthetic therapeutic peptides:
Atomic interactions and profile of small mole- science and market. Drug Disc Today 15:40–56
cules disrupting protein‐protein interfaces: the 30. Fry D, Huang K-S, Di Lello P, Mohr P, Mueller
TIMBAL database. Chem Biol Drug Des 74: K, So S-S, Harada T, Stahl M, Vu B, Mauser H
457–467 (2013) Design of libraries targeting protein‐pro-
19. Morelli X, Bourgeas R, Roche P (2011) Chem- tein interfaces. Chem Med Chem 8:726–732
ical and structural lessons from recent successes 31. Murray CW, Rees DC (2009) The rise of
in protein‐protein interaction inhibition fragment-based drug discovery. Nat Chem 1:
(2P2I). Curr Opin Chem Biol 15:1–7 187–192
20. Fry DC (2012) Small-molecule inhibitors of 32. Hajduk PJ (2006) Fragment-based drug
protein‐protein interactions: how to mimic a design: how big is too big? J Med Chem
protein partner. Curr Pharm Des 18: 49:6972–6976
4679–4684 33. Fry DC, Wartchow C, Graves B, Janson C,
21. Cerchietti LC, Ghetu AF, Zhu X, DaSilva GF, Lukacs C, Kammlott U, Beluins C, Palme S,
Zhong S, Matthews M, Bunting KL, Plol JM, Klein C, Vu B (2013) Deconstruction of a
Fares C, Arrowsmith CH, Yang SN, Garcia M, nutlin: dissecting the binding determinants of
Coop A, MacKerell AD, Prive GG, Melnick A a potent protein‐protein interaction inhibitor.
(2010) A small-molecule inhibitor of BCL6 ACS Med Chem Lett 4(7):660–665
kills DLBCL cells in vitro and in vivo. Cancer 34. Grasberger BL, Lu T, Schubert C, Parks DJ,
Cell 17:400–411 Carver TE, Koblish HK, Cummings MD,
22. Buckley DL, Van Molle I, Gareiss PC, Tae HS, LaFrance LV, Milkiewicz KL, Calvo RR,
Michel J, Noblin DJ, Jorgensen WL, Ciulli A, Maguire D, Lattanze J, Franks CF, Zhao S,
Crews CM (2012) Targeting the von Hippel- Ramachandren K, Bylebyi GR, Zhang M, Man-
Lindau E3 ubiquitin ligase using small mole- they CL, Petrella EC, Pantoliano MW, Deck-
cules to disrupt the VHL/HIF-1alpha interac- man IC, Spurlino JC, Maroney AC, Tomczuk
tion. J Am Chem Soc 134:4465–4468 BE, Molloy CJ, Bone RF (2005) Discovery and
23. Fry DC, So S-S (2012) Modulators of protein‐ cocrystal structure of benzodiazepinedione
protein interactions: the importance of three- HDM2 antagonists that activate p53 in cells.
dimensionality. In: Doemling A (ed) Protein‐ J Med Chem 48:909–912
protein interactions in drug discovery. Wiley- 35. Allen JG, Bourbeau MP, Wohlhieter GE, Bart-
VCH, Weinhelm, pp 55–62 berger MD, Michelsen K, Hungate R, Gad-
24. Wanner J, Fry DC, Peng Z, Roberts J (2011) wood RC, Gaston RD, Evans B, Mann LW,
Druggability assessment of protein‐protein Matison ME, Schneider S, Huang X, Yu D,
interfaces. Future Med Chem 3:2021–2038 Andrews PS, Reichelt A, Long AM, Yakowec
25. Fuller JC, Burgoyne NJ, Jackson RM (2008) P, Yang E, Lee TA, Oliner JD (2009) Discovery
Predicting druggable binding sites at the pro- and optimization of chromenotriazolopyrimi-
tein‐protein interface. Drug Disc Today 14: dines as potent inhibitors of the mouse double
155–161 minute 2 – tumor protein 53 protein interac-
26. Bourgeas R, Basse MJ, Morelli X, Roche P tion. J Med Chem 52:7044–7053
(2010) Atomic analysis of protein‐protein 36. Popowicz GM, Czarna A, Wolf S, Wang K,
interfaces with known inhibitors: the 2P2I Wang W, Doemling A, Holak T (2010) Struc-
database. PLoS One 5:e9598 tures of low molecular weight inhibitors bound
27. Meireles LMC, Doemling AS, Camacho CJ to MDMX and MDM2 reveal new approaches
(2010) ANCHOR: a web server and database for p53-MDMX/MDM2 antagonist drug dis-
for analysis of protein‐protein interaction covery. Cell Cycle 9:1–8
Part II
Label-Free Biosensors and Techniques

Chapter 7
Studying Protein-Protein Interactions Using Surface

Plasmon Resonance
Zaneta Nikolovska-Coleska
Abstract
Protein-protein interactions regulate many important cellular processes, including carbohydrate and lipid
metabolism, cell cycle and cell death regulation, protein and nucleic acid metabolism, signal transduction,
and cellular architecture. A complete understanding of cellular function depends on full characterization of
the complex network of cellular protein-protein interactions, including measurements of their kinetic and
binding properties. Surface plasmon resonance (SPR) is one of the commonly used technologies for
detailed and quantitative studies of protein-protein interactions and determination of their equilibrium
and kinetic parameters. SPR provides excellent instrumentation for a label-free, real-time investigation of
protein-protein interactions. This chapter details the experimental design and proper use of the instrumen-
tation for a kinetic experiment. It will provide readers with basic theory, assay setup, and the proper way of
reporting this type of results with practical tips useful for SPR-based studies. A generic protocol for
immobilizing ligands using amino coupling chemistry, also useful if an antibody affinity capture approach
is used, performing kinetic studies, and collecting and analyzing data is described.
Key words Protein-protein interactions, Surface plasmon resonance, Real-time and label-free
biosensor applications, Kinetics, Affinity, Association, Dissociation
1 Introduction
Protein-protein interactions (PPIs) are fundamental in regulation

of biological processes such as cellular signaling pathways. The
“omic” technologies (genomics, transcriptomics, proteomics)
have provided important insights into PPIs in biological systems
demonstrating that these interactions control the assembly of
multi-protein complexes that mediate biological functions. System-
atic mapping of interactome networks and identification of the
“central nodes” essential for homeostasis has become a major goal
of current biomedical research [1, 2]. Changes in PPIs are involved
in many diseases, including cancer, neurodegenerative diseases,
viral and bacterial infections [3, 4], and such alterations are increas-
ingly attracting attention as a new frontier in drug development.
109
110 Zaneta Nikolovska-Coleska
Characterization of the protein-protein binding interactions and

understanding their function and role in biological processes and
diseases is important and necessary for validation of potential ther-
apeutic targets [3, 5–7].
The ultimate goal of studying protein-protein interactions is
recognition of the consequences of the interaction for cell function,
which depend on the strength of the interaction. Thus, characteri-
zation of the PPIs through determination of the interaction affinity,
selectivity and binding kinetic parameters, has become an impor-
tant part of defining and understanding in vivo bimolecular
interactions.
One of the most used technologies for studying, analyzing and
measuring the binding parameters of PPIs is surface plasmon reso-
nance (SPR). SPR is a powerful and sensitive experimental tool,
based on an optical detection technique, which allows quantitative
kinetic, equilibrium, and thermodynamic analysis of biomolecular
interactions [8, 9]. Since its initial development, SPR has been
widely used in molecular biology due to its many advantages
which include: (1) high sensitivity, (2) label-free detection, (3)
lack of dependence on spectroscopic properties, (4) real-time mon-
itoring, (5) quantitative evaluation, (6) determination of kinetic
rate constants, (7) low volume sample consumption, and (8) high
degree of automation and throughput. SPR can be used to assess
and quantify the kinetic and binding parameters of protein-protein,
protein–DNA, and protein–ligand interactions, and in combination
with other methods, to identify specific and selective interactions.
There are a variety of commercially available SPR systems that can
be categorized as laboratory, portable, and imaging SPR. The first
SPR system was developed in the early 1990s, by Biacore,
subsequently acquired by GE Healthcare, currently the main sup-
plier of SPR instruments [10]. These instruments typically are
able to measure association or on-rates (ka) in the range of
103–107 M1 s1, dissociation or off-rates (kd) of 105–1 s1,
affinity at equilibrium (KD) in the range of 100 μM to 200 pM,
and analyte concentration in the range of 1 mM to 10 pM.
This chapter provides an overview of the detection principles,
assay design, immobilization techniques and data analysis of SPR,
followed by a detailed description of the materials and methods,
and guidance for analysis of protein-protein interactions.
1.1 Surface Plasmon SPR is an optical-based real-time detection method for monitoring
Resonance: Principle specific binding events between two or more biomolecules, without
the use of labels, and provides quantitative information on the
specificity, kinetics and affinity of the biomolecular interactions
[11–13]. SPR can be also used for thermodynamic analysis, epitope
mapping, and to determine analyte concentration [14–16].
A unique approach to protein investigations is enabled by the
combination of SPR and mass spectrometry, which increases limits
Studying Protein-Protein Interactions 111
Flow
Analyte
Flow channel
Ligand
Carboxymethylated Sensor chip
dextran
surface
Gold surface
dq
Resonance Units (RU)

Steady State
Dissociation
Association
Time (sec)
Injection of analyte Running buffer
Fig. 1 Experimental setup of a SPR experiment. Protein “ligand” is immobilized

on the dextran-coated sensor chip surface. Protein “analyte” is passed over the
chip surface. Protein-protein interactions between the “ligand” and the “ana-
lyte” increase the density at the sensor chip surface and change the angle of
refracted light (δθ) generating a real-time binding curve known as a sensorgam.
Schematic representation of a sensorgram: Association phase (two or more
molecules bind to each other); Steady state-equilibrium (the amount of the
molecules that are binding is equal to the amount of molecules that dissociate);
Dissociation phase (the bonds between the molecules are breaking and the
molecules are dissociating)
of detection, ligand fishing, multi-protein analysis, and protein-

complex delineation [17, 18].
In SPR systems, interactions and detection occur in multichan-
nel flow cells on a sensor surface, which is part of the flow system.
The SPR-based binding method involves immobilization of a
ligand on the surface of a sensor chip which has a monolayer of
carboxymethylated dextran covalently attached to a gold surface.
The ligand of interest is immobilized on the surface of the sensor
chip using well-defined chemistry allowing solutions with different
concentrations of an analyte to flow over it and to characterize its
interactions to the immobilized ligand (Fig. 1). The SPR signal
originates from changes in the refractive index at the surface of the
gold sensor chip. The increase in mass associated with a binding
event causes a proportional increase in the refractive index, which is
observed as a change in response. These changes are measured as
changes in the resonance angle (δθ) of refracted light when the

analyte, flowing in a microfluidic channel, binds to the immobilized
ligand and increases in density at the sensor chip. Importantly, for
protein-protein interactions the change in refractive index on the
surface is linearly related to the number of molecules bound [19].
The response signal is quantified in resonance units (RU) and
represents a shift in the resonance angle. Monitoring the change
in the SPR signal over time produces a sensorgram, a plot of the
binding response (RU) versus time which allows different stages of
a binding event to be visualized and evaluated (Fig. 1). During the
injection of an analyte, the binding response increase is due to the
formation of analyte–ligand complexes at the surface and the sen-
sorgram is dominated by the association phase. After a certain time
of injection, a steady state is reached, in which binding and dis-
sociating molecules are in equilibrium. The decrease in response
after analyte injection is terminated is due to dissociation of the
complexes, defining the dissociation phase. Depending on the
dissociation rate of the tested ligand, some assays may require a
regeneration step in order to reach the baseline again. Fitting the
sensorgram data to an appropriate kinetic binding model allows
calculation of kinetic parameters such as the association (ka) and
dissociation (kd) rate constants, and the binding affinity of the
tested interactions.
1.2 Sensor Surface The first step in the interaction analysis is the surface preparation
Preparation/ and immobilization of one of the binding partners (the “ligand”)
Immobilization on the sensor chip surface. Several factors should be considered for
this procedure and include selecting (1) the binding partner for the
immobilization, (2) the appropriate immobilization level, and (3)
the method to be used for immobilization, and consequently, the
sensor chip that will be selected for use.
1.2.1 Selection of Ligand Several properties can influence the decision as to which ligand will
be immobilized on the chip. These include the molecular weight of
binding partners, their purity, the number of binding sites, the
functionality of the immobilized ligand, the isoelectric point (pI)
of the protein(s), the amount of ligand available, and the assay
conditions.
When direct coupling is used as an immobilization method, it is
important that the purity of the ligand be >95 %. The immobilized
ligand should be stable and the binding activity must survive the
coupling procedure as well as the regeneration protocol. During
SPR experiments, binding is measured as a change in the refractive
index on the surface of the sensor chip caused by accumulation of
mass within the surface dextran layer and the measured response is
related to the mass of the bound analyte. Therefore, when possible,
the smallest and most stable molecule in the system should be used
as the immobilized component. The larger molecule of the pair
should be the analyte thus providing high sensitivity to the assay.

Proteins with more than one binding site should be immobilized
on the chip. If they are used as the analyte, avidity issues can arise,
complicating the kinetic analysis. Less amount of sample is required
for coupling and consequently the protein that is less available
should be immobilized.
1.2.2 Immobilization The binding capacity of the sensor chip surface depends on the
Levels immobilization level and the optimal amount of the immobilized
ligand depends on the experimental goal. For example, higher
levels of immobilized ligand are needed in experiments designed
to determine the concentration of analytes in solution, while for
kinetic analysis the immobilization level of the ligand should be low.
The maximum binding capacity (Rmax) of the immobilized ligand
recommended in the literature is typically in the range of 50–150
resonance units (RU), depending on the relative molecular weights
and sizes of ligand and analyte. An overloaded surface causes
problems of steric hindrance, aggregation, rebinding and diffu-
sional limitations such as mass transport. Mass transport limitations
can result in calculated binding kinetics that are slower than the
true binding parameters. Problems due to mass transport effects are
complex and they are best avoided by use of a low density surface to
provide a maximum binding response level (Rmax) of no more than
50–100 RU, and high flow rates, preferably 100 μL/min.
The maximum response (Rmax) that can be obtained in experi-
ments depends directly on the molecular weight of the protein-protein
complex that will be formed on the surface after flowing the analyte
molecule. The theoretical Rmax describes the maximum binding capac-
ity of the surface, which is usually higher than the experimental Rmax.
The theoretical Rmax can be calculated using the following equation:
MW analyte
Rmax ¼ RL Sm
MW ligand
Rmax—maximum capacity of surface

(MWanalyte/MWligand)—ratio of the mass of the analyte and ligand
RL—amount of ligand on surface
Sm—stoichiometry
The following rearranged equation can be used to determine an
appropriate immobilization level that will generate an Rmax in the
desired range of 50–150 RU for measurement of the binding
kinetics of studied PPIs:

MW ligand

1
RL ¼ Rmax
Sm MW analyte
This calculation assumes 100 % active analyte and ligand. The

capture technique typically retains 100 % activity of the bound
ligand whereas the direct coupling technique has 50–80 % of the

activity of the bound ligand, although in some cases this can be
lower. The quality of the binding data depends on the ligand
activity, which is based on the determined experimental Rmax and
should be calculated using the following equation:
Experimental Rmax
% ligand activity ¼ 100 %
Theoretical Rmax
Poor surface activity can result from denaturation of the ligand as a
result of the pre-concentration solution, blocking and inactivating
the binding site by immobilization chemistry, or impurities present
in the ligand which can also be immobilized, lowering the surface
activity. In these cases, changes in the coupling chemistry or ligand
capture approach should be considered in order to improve the
surface activity.
1.2.3 Immobilization Direct covalent coupling and affinity capturing techniques are two
Method general methods for immobilization of the ligand. Covalent cou-
pling is the most commonly used means of immobilizing the
ligand. In this technique, the ligand is directly immobilized on
the sensor surface using established coupling chemistry depending
on the available reactive groups, for example amine, thiol, or alde-
hyde groups in the protein. In such cases, direct immobilization
does not require any modification of the ligand. The sensor chip,
CM5, is the most versatile available chip for covalent coupling of
ligands with high binding capacity (Table 1). It gives high response
and excellent chemical stability providing accuracy and allowing
repeated analyses on the same surface. The immobilization level is
easily controlled and ligand consumption is low. The disadvantage
of this immobilization method is that it can be associated with
multiple attachment sites leading to randomized coupling and
heterogeneous orientation of the ligand.
In contrast to direct immobilization, the affinity capturing
method offers steric orientation of the immobilized interaction
partner for optimum site exposure and assurance that all immobi-
lized molecules will be in the same orientation, resulting in a highly
active and homogeneous surface (Fig. 2a). The method relies on
non-covalent protein-protein interactions and provides oriented
coupling, because binding occurs at a well-defined site on the
target. Another advantage of this approach is that it is fully regen-
erable and allows immobilization from crude protein mixtures,
such as cell lysates. A disadvantage of the capturing method is
that it may not produce a stable surface, thus complicating the
analysis of analyte binding.
Three major coupling classes can be used for the affinity cap-
turing method: artificially introduced affinity tags (e.g., biotin or
hexahistidine), antibody–antigen systems (e.g., antibodies against
Table 1
Available sensor chips, their characteristics and applications
Sensor
chip Surface Characteristics and applications
CM5 Standard surface—Carboxymethylated dextran Excellent chemical stability; Versatile chip
matrix suitable for most applications
CM4 Low carboxylation—Carboxymethylated Reduces nonspecific binding of highly
dextran matrix with lower degree of positively charged molecules that may be
carboxylation than CM5, i.e., less negatively found, for example, in crude samples
charged Convenient for measurement of low Rmax
needed in kinetic applications
CM3 Short dextran—Carboxymethylated dextran For low immobilization levels and work with
matrix, with the same level of carboxylation cells, viruses, and multicomponent
as CM5, but shorter matrix complexes with high molecular weight
analytes
CM7 High Carboxylation—Comparable with the Gives higher immobilization capacity; Ideal
CM5 chip but has a higher density of for applications that use small molecules
carboxymethylated dextran and fragments
C1 Flat carboxymethylated surface For work with particles such as cells and
viruses, and in applications where dextran
matrix is undesirable
SA Carboxymethylated dextran matrix pre- Captures biotinylated ligands such as
immobilized with streptavidin carbohydrates, peptides, proteins, and
DNA fragments
NTA Carboxymethylated dextran matrix pre- Designed to bind histidine-tagged ligands via
immobilized with NTA (nitrilotriacetic acid) metal chelation; Allows control of steric
orientation of ligand component for
optimal site exposure
HPA Flat hydrophobic surface For studying lipid monolayers interacting
with membrane binding biomolecules and
membrane-associated interactions
L1 Carboxymethylated dextran matrix modified For rapid and reproducible capture of
with lipophilic substances liposomes with retention of lipid bilayer
structure
different tags as capture molecules), and interactions between pro-

teins and naturally occurring sites (e.g., Protein A/IgG). An exam-
ple of the antibody–antigen system is using anti-Flag antibody,
which can be covalently linked to the sensor chip surface. Proteins
fused with Flag are readily immobilized on a sensor chip surface
through the Flag/anti-Flag antibody interaction. The advantage of
this approach is that the antibody serves as a universal adaptor,
permitting the capture of different Flag-fusion proteins on the
sensor chip surface. As the antibody is covalently bound to the
sensor chip, the anti-Flag surface may be regenerated many times
a
Analyte
Tagged protein
Anti-tag
antibody
CM5 Sensor chip
b
1250
1000
750
RU
pCDNA3-FLAG-DOT1L (2.4µg/µl)
500
pCDNA3-FLAG-DOT1L (1.2µg/µl)
250 pCDNA3-FLAG-DOT1L (0.6µg/µl)
0 pCDNA3 (2.4µg/µl)
0 100 200 300 400
time (s)
Fig. 2 Affinity based immobilization method. (a) Illustration of the capture affinity immobilization method using
an antibody–antigen interaction. (b) Dose-dependent and specific immobilization of Flag-tagged full length
histone methyltransferase, DOT1L, obtained from crude cell lysate of 293 cells transfected with pCDNA3-
FLAG-DOT1L, using anti-flag antibody as a capture molecule20 (pCDNA3—empty vector)
and coated with different Flag-fusion constructs. This approach

also allows selective ligand capture when crude samples are used,
for example immobilization of proteins that are transiently
expressed in cells and can be hard to purify. Using this strategy,
the Flag-tagged full length of histone methyltransferase DOT1L,
obtained from crude cell lysate, was specifically and dose depen-
dently captured on a CM5 sensor chip through immobilized anti-
Flag antibody (Fig. 2b), and used for determination of the kinetic
binding parameters of PPIs between DOT1L and the MLL-fusion
proteins, AF9 and ENL [20].
Additional affinity capturing strategies involve using biotin
and hexahistidine tags for which different sensor chips should
be used. The sensor chip SA has carboxymethylated dextran
pre-immobilized with streptavidin and is suitable for capture of
biotinylated ligands such as proteins, peptides, nucleic acids, or
carbohydrates, where controlled biotinylation enables oriented
immobilization. The sensor chip NTA is another surface that con-
trols steric orientation for immobilization of histidine-tagged
molecules. The carboxymethylated dextran is pre-immobilized
with nitrilotriacetic acid (NTA) and His-tagged molecules are
immobilized via Ni2+/NTA chelation. A list of the available sensor

chips for the Biacore Systems with their surface characteristics as
well as applications are presented in Table 1. Recently, it was
reported that the negative charge of the carboxymethyldextran
matrix on the biosensor surface can influence the binding kinetics
and KD determination of PPIs, suggesting that it is important that
these types of matrix-mediated artifacts to be identified early in the
measurement process and the most appropriate sensor chip be
selected [21].
The choice of immobilization strategy depends on ligand prop-
erties. For example, covalent coupling might on occasion result in
loss of ligand activity due to direct modification of residues in the
binding site or steric hindrance. In such cases, alternative chemistry,
such as thiol coupling, or an affinity capturing technique for immo-
bilization should be used. If the ligand is unstable, impure, or the
regeneration of the surface is difficult then the capture method is
preferable. In order to confirm that the immobilized ligand is in its
proper confirmation and its activity is retained, the surface should
be tested with a positive control, if available. The time period for
which the chips are reusable depends upon the stability and main-
tenance of the functionality of the attached ligand.
1.2.4 Coupling Chemistry Different covalent coupling chemistries have been developed for
immobilization of ligands: coupling through primary amines, thiol
groups, and aldehyde coupling (Fig. 3). Choosing an immobiliza-
tion method depends mostly on the nature of the ligand.
Amine coupling is the most generally applicable coupling
chemistry because of its universality, stability, speed and most macro-
molecules contain amine groups. This coupling uses primary amine
groups (the N-terminus and lysine residues), which are often
solvent-exposed due to their hydrophilicity. They react directly
with active esters generated by N-hydroxysuccinimide–1-ethyl-3-
(3-dimethylaminopropyl)carbodiimide (NHS/EDC) activation
(Fig. 3). The second type of covalent coupling chemistry, thiol
coupling, provides an alternative approach when amine coupling
cannot be performed or the immobilization level will not be suffi-
cient, such as immobilization of acidic proteins. Thiol coupling uti-
lizes exchange reactions between thiols and active disulphide groups
which can be introduced either on the dextran matrix to exchange
with a thiol group on the ligand (ligand thiol approach) or on the
ligand molecule to exchange with a thiol group introduced on the
dextran matrix (surface thiol approach). A recommended reagent for
introducing active disulphide groups is 2-(2-pyridinyldithio)ethanea-
mine (PDEA). Thiol coupling chemistry is not suitable for experi-
ments where the surface is exposed to reducing agents or high pH,
since the coupling bond is unstable under such conditions.
Ligands containing aldehyde groups (either native or introduced by
oxidation of cis-diols) can be immobilized after activating the surface
NHS/EDC
a)
b)
Sensor chip CM5 Amine functionalized
ligands
PDEA
c) d)
Cystamine/DTE
c)
Aldehyde
functionalized ligands
Ligand thiol
coupling Surface thiol
coupling
Fig. 3 Steps in coupling chemistry for covalently attaching biomolecules to the sensor surface: (a ) Activation
of the matrix-based carboxyl groups by NHS (N-hydroxysuccinimide) and EDC (1-ethyl-3-(3-dimethylamino-
propyl)-carbodiimide); (b) Direct immobilization of amine functionalized ligands; (c ) Thiol coupling utilizing
disulphide exchange (ligand and surface thiol coupling); (d ) Immobilization of aldehyde functionalized ligands
using reductive ammination. (PDEA Thiol coupling reagent: 2-(2-pyridinyldithio)ethaneamine hydrochloride;
H2N-NH2 Hydrazine; NaCNBH4 Cyanoborohydride)
with hydrazine or carbohydrazide. Aldehyde coupling provides an

alternative approach for immobilizing glycoproteins and other
glycoconjugates.
1.2.5 Determining of The immobilization process is based upon the principle of electro-
Optimal Immobilization pH static interaction of the ligand and the activated surface molecules,
(pH Scouting) which results in a covalent binding. In order to reach the highest
efficiency of the amine coupling reaction, proteins have to be pre-
concentrated on the sensor chip surface. A ligand can be concen-
trated at the sensor surface by electrostatic attraction. The pKa of
the dextran matrix is 3.5 and at pH > 3.5 the dextran matrix will
have a net negative charge. Efficient pre-concentration requires that
the pH be between the pKa of the sensor surface and the isoelectric
point (pI) of the ligand (the pH at which there is no net charge on
the protein). Such surface attraction will be reached by using
immobilization buffer with pH > 3.5, but lower than the pI of
the ligand. As a general rule, the pH of the buffer should be at least
20000
pH 4.0

pH 4.5
15000
pH 5.0
pH 5.5
10000
5000 Injection of NaOH

wash solution
0
0 100 200
Time (sec)
Fig. 4 Pre-concentration profile of DOT1L protein onto the dextran surface of a

CM5 sensor chip. Solutions of 50 μg/mL of Mocr-DOT1L (826–1095) protein in
10 mM sodium acetate buffer with various pHs were injected over an untreated
chip surface at 10 μL/min. After each injection, 50 mM NaOH was used to
completely remove the protein from the surface. Electrostatic interactions of
DOT1L protein occur for each tested pH and binding increases as the pH is
increased from 4.0 to 5.0. The sodium acetate buffer with pH 5.0 was chosen as
optimum to minimize possible denaturation
0.5 units below the pI of the ligand. In addition, the lower the ionic
strength of the coupling buffer, the more ligand will be pre-
concentrated and immobilized. This will allow strong electrostatic
attractions between the negatively charged carboxyl group of the
matrix and the positively charged amine groups of the ligand.
Therefore, acidic proteins with pI values <3.5 cannot be immobi-
lized by amine coupling. Buffers such as Tris buffer and reagents
bearing primary amines should not be used due to possible compe-
tition with the amino groups of the protein. To determine the
appropriate pH for immobilization, an experimental procedure
known as pre-concentration pH-scouting, should be performed.
The pre-concentration is driven by electrostatic interactions, which
are most pronounced in low ionic strength buffers. For this pur-
pose 10 mM sodium acetate buffer with different pH values, in a
range from 4.0 to 5.5, is tested in order to determine the most
optimal pH (Fig. 4). The magnitude and slope of the response
generated due to the increased mass on the surface provide infor-
mation about the ligand density. By comparing the pre-
concentration responses of each immobilization buffer, the mildest
pH solution that can achieve the targeted level for immobilization
should be selected in order to minimize possible denaturation. In
general, for many proteins, 10 mM sodium acetate buffer (pH 4.5)
is optimal for their preconcentration.
1.2.6 Reference Surface During an SPR experiment the choice of an appropriate reference
surface is crucial. Usually the surface on the chip is divided into
several flow cells, depending on the instrument, which can be used
individually or in a number of combinations, and at least one flow
cell of the chip should be used as a reference surface. The use of a

reference surface is important because it is a means to: (1) demon-
strate that the observed response is not due to nonspecific binding,
(2) correct for bulk refractive index change, (3) correct for baseline
drift, and (4) in combination with “double referencing”, correct for
instrument noise due to injections.
Three types of reference surfaces can be used. An unmodified
surface is used to check for nonspecific binding to the dextran
matrix. A surface that is treated with the same coupling chemistry
used to immobilize the ligand is the most commonly used control
as a reference surface. This activated–deactivated surface is prepared
by treating the surface with the immobilization procedure but
omitting the ligand. In this way the negative charge is decreased
and therefore nonspecific binding is reduced. The third type of
control surface is similar to that used for immobilization but
using a different ligand, which can be a nonspecific molecule or
an inactive form of the ligand that does not bind the analyte. This
inactive protein should be immobilized to approximately the same
level as the ligand on the active surface and it should mimic the
physical properties of the active surface ligand in terms of, for
example, size and charge. If the capture-based assay is used it
might be necessary to have an additional control surface onto
which the capture ligand will be immobilized at a level similar to
that on the active test surface.
Before starting the kinetic measurements, it is very important
to demonstrate that the analyte does not show significant binding
responses on the control surface as a result of electrostatic or
hydrophobic interactions. One of the possibilities for a significant
degree of nonspecific binding is a highly positive charge of the
analyte in the sample buffer and a high electrostatic attraction to
the chip surface. Electrostatic nonspecific binding can be mini-
mized by the addition of NaCl to the sample and running buffers,
or using a CM4 chip that has a carboxymethylated dextran matrix
with a lower degree of carboxylation, therefore a less negatively
charged ligand than CM5 (Table 1), or the analyte could be immo-
bilized on the chip surface. In this way, nonspecific binding of
highly positively charged molecules will be reduced. Hydrophobic
nonspecific binding usually can be minimized by the addition of a
detergent, such as 0.05 % polysorbate 20 or 10 mM CHAPS in the
buffer solution.
Each analyte and blank sample should be injected over the
ligand and the reference surfaces simultaneously. Differences in
the refractive index between the running and sample buffers will
give rise to changes in responses that are known as bulk refractive
index changes. This is a common phenomenon and in order to
eliminate its effect on the measurement of the binding interactions,
in particular when the association phase is used for quantification of

Reference surface Active surface
Binding + Bulk
Reference
subtraction True binding
Time (sec) Time(sec) Time(sec)
Fig. 5 Bulk effects and reference surface subtraction. Bulk effects are due to differences in the refractive index
of the running buffer and sample solution. The bulk contribution should be subtracted using a reference
surface
the binding interaction, it is important that the bulk contribution

be subtracted using the reference surface responses (Fig. 5).
Furthermore, SPR experiments should be designed in a way
that double referencing of the sensorgram data can be performed.
Double referencing data helps correct for artifacts such as bulk
refractive index changes, nonspecific binding, systematic instru-
ment noise, and baseline drift, all of which are common in almost
every SPR experiment. To double reference sensorgram data, both
the reference surface responses and the blank responses are sub-
tracted from the analyte sensorgrams during sensorgram proces-
sing. The former corrects for refractive index shifts and nonspecific
binding, while the latter corrects for systematic instrument noise
and baseline drift. Often, the quality of sensorgram data sets cannot
be assessed without double referencing, and normally, data cannot
be reliably fit for kinetic rate constants without double referencing.
2 Materials
The materials and step-by-step procedures provided in this chapter

are universally applicable and can be used for any biomolecular
interaction analysis and most Biacore systems, GE Healthcare.
2.1 Instrument 1. Maintenance sensor chip (GE Healthcare).

Cleaning 2. Running buffer:
(a) HBS-P: 10 mM HEPES, pH 7.4, 150 mM NaCl, 0.005 %
(v/v) Surfactant P20 or Tween 20.
(b) HBS-EP is an additional commonly used buffer where the
HEPES buffer is supplemented with 3 mM EDTA.
The buffer must be filtered (0.22 μm), degassed, and stored at
4 C. It should be allowed to equilibrate to room temperature
prior to use.
3. Desorb solution 1: Sodium dodecyl sulfate (0.5 % w/v SDS).
4. Desorb solution 2: 50 mM Glycine, adjusted to pH 9.5 with

5 N NaOH.
5. 1 % AcOH.
6. 0.2 M NaHCO3.
7. 6 M guanidine-HCl.
8. 10 mM HCl.
9. Ultrapure water.
2.2 Surface 1. Sensor chip CM5 (GE Healthcare).

Preparation, Ligand 2. Running buffer HBS-P.
Pre-concentration,
3. 10 mM HCl.
and Amine Coupling
4. 50 mM NaOH.
5. 0.1 % SDS.
6. Sodium acetate (NaOAc) buffers for immobilization: 10 mM
NaOAc, pH 4.0, 4.5, 5.0 and 5.5, adjusted with 10 % (v/v)
AcOH and filtered (0.22 μm). The immobilization buffer
should contain no primary amines such as Tris buffer.
7. Ligand solution (for direct immobilization) or antibody (for
capture coupling) diluted with immobilization buffer.
8. NHS amine coupling reagent: 100 mM N-Hydroxysuccini-
mide in water.
9. EDC amine coupling reagent: 400 mM 1-Ethyl-3-(3-dimethy-
laminopropyl) carbodiimide hydrochloride in water.
10. Ethanolamine solution: 1 M Ethanolamine hydrochloride,
pH 8.5.
11. 20 mM NaOH.
12. Polypropylene tubes with rubber caps (GE Healthcare).
2.3 Regeneration 1. Full range of regeneration solutions:

(a) 10 mM glycine buffers, pH 1.5–3.0.
(b) 10–100 mM phosphoric acid, HCl and NaOH.
(c) 1–5 M NaCl and 2–4 M MgCl2.
(d) 8 M urea and 6 M guanidine hydrochloride.
2. Analyte solutions prepared in a range of different concentra-
tions including the highest concentration or a concentration
above the range that will be used in the application.
2.4 Kinetic Studies 1. Stock solution of the analyte (100 μg/mL).

2. Running buffer (HBS-P is the most commonly used).
3. Regeneration solution, as previously determined and
optimized.
3 Methods
3.1 Amine Coupling The coupling procedure consists, in general, of three steps: (1)
activation resulting in an active group that can be further modified
or used to couple the ligand; (2) coupling, when the ligand is
injected over the activated surface until sufficient ligand is bound;
and (3) deactivation to quench remaining activated sites and obtain
the final level of immobilized ligand (Fig. 6). The covalently immo-
bilized ligand either participates directly in the interaction under
study or is used for affinity capture of one of the interacting mole-
cules. The methods presented below are the same if an interacting
ligand is immobilized or a corresponding antibody will be used for
affinity capturing of the ligand.
3.1.1 Instrument Cleaning and routine instrument maintenance is a key contributor

Cleaning to good quality SPR data. Besides the “desorb” and “super clean”
protocols which are described below, it is also important to sanitize
the instrument fluidics frequently, especially if the instrument is
exposed to crude biological samples.
1. Use a maintenance chip and running buffer to run desorb from
working tools (20 min) with Desorb solutions 1 (0.5 % w/v
SDS) and 2 (50 mM Glycine-NaOH, pH 9.5) (see Note 1).
2. Run “Super Clean”—a comprehensive cleaning procedure to
clean the system including the needle and tubing. The whole
procedure requires about 2 h. Run desorb method with the
listed reagents below, followed by priming with water heated to
50 C:
(a) 2 5 mL vials of 1 % AcOH (use desorb method).
(b) Prime.
Ethanolamine
40000
Deactivation
Ligand
Resonance Unit (RU)
30000 Immobilization
NHS/EDC
Activation
20000 x
RU of
immobilization
x
10000
0
0 500 1000 1500 2000
Time(sec)
Fig. 6 Sensorgram from a typical amine coupling illustrating the immobilized

amount of the ligand
(c) 2 5 mL vials of 0.2 M NaHCO3 (use desorb method).

(d) Prime.
(e) 2 5 mL vials of 6 M guanidine-HCl (use desorb
method).
(f) Prime.
(g) 2 5 mL vials of 10 mM HCl (use desorb method).
(h) Prime three times and then put the instrument on standby
or continue with preparation of the sensor chip.
3.1.2 Preconditioning of 1. Allow the sealed sensor chip pouch to equilibrate at room
the Sensor Chip CM5 temperature for 30 min in order to prevent condensation on
the chip surface. Insert Sensor Chip CM5 in the instrument
and dock the chip.
2. Run three times Prime with running buffer (HBS-P) to fill the
syringes and compartments.
3. Set the flow rate to 100 μL/min and sequentially using
“Quickinject” command inject two times 10 mM HCl,
50 mM NaOH and 0.1 % SDS to precondition the chip surface.
This procedure allows washing and hydrating the surface
simultaneously, before starting the immobilization procedure.
4. “Rinse,” “Flush,” and “Prime” to wash the microfluidic system
to be ready for the next step of experiments.
3.1.3 Ligand 1. Start a sensorgram, choose the flow cell where the immobiliza-
Pre-concentration tion will take place and decide on a control, reference surface
for blank subtraction. Usually, flow cell 1 is used as the refer-
ence surface.
2. Set flow rate to 10 μL/mL.
3. Dilute and prepare the ligand solution (protein sample for
immobilization) (see Note 2) to a constant concentration in a
range from 5 to 20 μg/mL using a variety of 10 mM NaOAc
immobilization buffers with different pH values (e.g., 4.0, 4.5,
5.0, and 5.5 where the pH is adjusted using 10 % (v/v) acetic
acid) (see Note 3). Place the solutions into propylene tubes and
cap the tubes with rubber caps.
4. Inject 10 μL of the prepared protein in NaOAc for 60 s contact
time.
5. Inject 50 mM NaOH for 15 s after each pre-concentration as a
wash solution to remove the ligand from the surface, if the
response does not return to baseline.
6. Analyze the level of RU and the slope of the pre-concentration
response to determine the ligand density that can be achieved
(Fig. 4). Compare all responses to identify the mildest pH
solution that can achieve the targeted level of immobilization,

which depends on the application (see Note 4).
3.1.4 Ligand 1. Start new sensorgram with a 10 μL/min flow rate and select the
Immobilization surface upon which the ligand protein will be immobilized
(Fc2, Fc3, and/or Fc4).
2. Prepare fresh solution for activation of the CM5 surface by
mixing equal volumes of 100 mM NHS and 400 mM EDC
and inject 70 μL (7 min contact time) (see Note 5).
3. Prepare the ligand in the NaOAc immobilization buffer (with
pH determined in the pre-concentration procedure) at a con-
centration between 10 and 50 μg/mL and inject it over the
activated surface until the target levels for ligand immobiliza-
tion are obtained (see Note 6).
4. Inject 70 μL of 1 M ethanolamine hydrochloride, pH 8.5, to
deactivate excess reactive groups (see Note 7).
5. Repeat steps 1–4 until each flow cell has been treated and the
ligand has been immobilized.
6. Prepare the reference surface on flow cell 1 (Fc1) to be used for
reference-subtracted data necessary for kinetic and affinity mea-
surements. Repeat steps 1–4, replacing the ligand for immobi-
lization with an inactive protein or omit immobilization of the
control protein so that the control surface is treated only with
the activation and deactivation reagents.
7. Wash the surface(s) with a washing solution (i.e., a regenera-
tion solution that is tolerated by the ligand, such as two times
15 s injections of 20 mM NaOH). This allows determination of
stably immobilized ligand and the net ligand immobilization
level.
8. The sensor chip is now ready to be used or stored. To ensure a
stable baseline for the interaction analysis, running a sensor
chip overnight in running buffer is recommended (see Note 8).
3.2 Sample Preparation of the samples for testing and injection over a prepared
Preparation immobilized surface is a critical factor for obtaining high-quality
results and accurate kinetic measurements. Similar to the ligand,
the analyte should also be pure, homogenous and unaggregated
since differently sized binding species will give different responses.
The buffer used in experiments can make a significant difference in
the binding of the analyte to the ligand and additions such as
detergents, chelating agents or denaturing chemicals can influence
the binding characteristics and the stability of the complex. The
standard running buffer, HBS-P, contains: 10 mM HEPES, pH
7.4, 150 mM NaCl, and 0.005 % P20 (polyoxyethylene sorbitan),
with or without EDTA (3 mM). The NaCl is necessary to reduce
nonspecific binding, P20 is a non-ionic detergent used to avoid
adsorption of the analyte by the flow channels and should be

included in the running buffer if possible. Phosphate buffer
(10.1 mM Na2PO4, 1.8 mM KH2PO4, pH 7.4, 137 mM NaCl,
2.7 mM KCl) and Tris buffer (50 mM Tris-HCl, pH 7.4, 150 mM
NaCl) are also commonly used buffers. In certain cases if the
analyte requires specific additives, for example, when analytes bind
in a metal ion-dependent manner, supplementation of the buffers
with compatible metal salts will be necessary. During preparation
the samples, in order to prevent bulk effects due to the differences
in the refractive index of sample solution and running buffer, it is
important that the sample buffer match the running buffer as
closely as possible. Usually, samples are prepared by serial dilution
with running buffer, especially if the dilution factor is sufficiently
high so that residual solution components will be negligible, or by
dialysis. Samples that contain high refractive index components,
such as high-salt solutions, glycerol, or DMSO, should be buffer-
exchanged into the running buffer or the components should be
added to the running buffer to match the composition of the
samples. For example, when small-molecules are tested for their
binding to immobilized target proteins, the same concentration of
DMSO is added to the running buffer, since DMSO is a commonly
used solvent for small molecules. Use of similar buffers will ensure
that association (during sample injection) and dissociation (after
the end of the injection) occur in the same environment.
Accurate analyte concentration is another important factor
since determination of the association rate and affinity constants
are dependent on concentration. Thus, it is important that appro-
priate methods be used to measure the analyte concentration [22].
The ideal value is the concentration of analyte that is active and can
bind to the immobilized ligand, which is not necessarily the same as
the total analyte concentration.
The concentrations tested in the kinetic experiment should be
determined based on the dissociation constant (KD) of the interac-
tion. Ideally, the analyte samples should be injected at several con-
centrations ranging from tenfold higher and tenfold less than the
expected KD. If the KD is unknown, preliminary experiments over a
wide range of analyte concentrations should be performed in order
to generate complete binding curves and obtain information about
appropriate concentrations.
3.3 Regeneration Regeneration is a process that completely removes the bound

analyte from the sensor chip surface after analysis of a sample.
Determination of the optimal conditions for efficient surface
regeneration is fundamental to efficient and robust protein-protein
interaction kinetic studies. It is necessary that the optimal
conditions for regeneration be determined empirically, and these
conditions will be specific for each studied PPIs. The most impor-
tant part in this process is that the activity of the surface remains
unaffected. Determining suitable regeneration conditions is a two-

step process including scouting for the best conditions followed by
verification of the suitability of the chosen conditions.
There are several conditions that are generally effective with a
wide range of interactants, and these should be used as starting
points. These conditions include low pH (10 mM glycine-HCl, pH
3–1.5) [23]; ethylene glycol (50, 75, and 100 %); high pH
(1–100 mM NaOH), high ionic strength (up to 5 M NaCl or
4 M MgCl2), detergent (low concentrations of SDS, up to 0.5 %),
and denaturant (8 M urea, 6 M guanidine hydrochloride). Addi-
tional conditions that have been identified as useful for regenera-
tion of some ligand–analyte interactions include: 10–100 mM HCl,
0.1 % trifluoroacetic acid, 1 M formic acid, 1 M ethanolamine–HCl,
pH 9 or higher. In some cases, it is necessary to use regeneration
cocktails composed of mixtures of different types of solutions [24].
Regeneration conditions have to be tested experimentally, and
whether the regeneration is efficient and removes all bound analyte
should be determined. Regeneration reagents of either low pH
(e.g., phosphoric acid, glycine–HCl) or high pH (e.g., NaOH)
should be injected in relatively short pulses of 10–30 s to minimize
exposure time at the ligand surface. Often multiple pulses of a
regeneration reagent may be needed. A wash step and an injection
of sample buffer should always be performed after the regeneration
injections to wash out the microfluidics system, as the regeneration
solution can be carried over between binding cycles and affect the
next binding reaction and kinetic analysis. The amount of analyte
bound to the surface can affect the conditions required for optimal
regeneration. Therefore, high analyte concentrations (the highest
or a concentration above) that will be tested in the application
should be used for the scouting procedure. It is also important
that the regeneration solution is fully compatible with the running
buffer, so that precipitation will not occur at interfaces between the
two solutions.
Once regeneration scouting has identified suitable conditions,
it is important to verify the performance of regeneration over a
larger number of repeated cycles of analyte injection and regenera-
tion to fully assess the selected conditions. As a general recommen-
dation, good regeneration conditions should give a consistent
analyte response, within 10 %, over multiple cycles demonstrating
that the ligand is still fully active. If regeneration conditions are too
mild, the analyte response will decrease and the baseline response
will increase, indicating that the analyte has not been completely
removed. On the other hand, conditions that are too harsh might
lead to decreases in the response and a constant or decreasing
baseline, suggesting that the surface is not fully active and/or has
been overstripped. A successful regeneration procedure has been
achieved when multiple, properly referenced sensorgrams of iden-
tical analyte concentrations can be reproduced. If the off rate is fast,
then the regeneration step will not be necessary. In this case, one
should just wash the surface until all of the analyte dissociates from
it. Recently, a detailed method for a systematic, seven-step experi-
mental approach to efficiently determine the optimal regeneration
conditions for SPR surfaces with covalently coupled proteins was
reported [25]. Successful selection of a regeneration solution is a
very important step ensuring that the surface chip will have intact
immobilized ligand and no build-up of bound analyte protein. This
will allow multiple uses of the chip and accurate KD determinations.
3.3.1 Identification of 1. Run sensorgrams using the prepared analyte solutions and
Suitable Regeneration various regeneration solutions, under constant contact time
Solution and flow rate (see Note 9).
2. Insert report points and analyze the results through obtained
responses to determine that regeneration is within 10 % of the
first injection (see Note 10).
3.4 Measurement of When designing a kinetic experiment, several experimental para-

Binding Kinetics and meters should be known or determined as discussed above: immo-
Data Analysis bilization level, injection time and flow rate, dissociation time, and
optimal analyte concentration range [26].
Kinetic characterization of interactions can be performed using
several different assay formats, which depend on the ligand–analyte
interaction kinetics. The most commonly used assay format is the
classical format involving multi-cycle kinetics in which each injec-
tion of analyte or blank is done in a separate cycle. In this format,
the dissociation phase is monitored for the same length of time with
each analyte concentration. If the dissociation is very slow, a more
efficient approach is the “short and long” experiment in which
short and long dissociation times are combined. Usually, the high-
est tested concentration is used to obtain a long dissociation time.
This allows an adequate decrease in response and collection of more
dissociation information for calculation of the dissociation rate
constant (kd). Both of these assay formats require regeneration of
the ligand surface after each analyte injection in order to remove the
bound analyte and prepare the surface for the next concentration to
be tested. A kinetic titration or single-cycle kinetics format is useful
for studying interactions that are difficult to regenerate or when
regeneration is detrimental to the ligand [27]. The analyte is
injected from a low to high concentration separated by short disso-
ciation times concluding with a long dissociation time. All the
injections are analyzed in one sensorgram with a special equation
for kinetic titration. A steady state or equilibrium experiment is
possible when the dissociation rate is fast, usually when the kd is
greater than 103 s1. This assay format is often used with small
compounds, which have a fast dissociation and consequently reach
steady state quickly. Generally, determination of the association and
dissociation rate using this assay format is difficult because the
steady state is reached rapidly. When performing equilibrium anal-

ysis to determine the equilibrium constant KD, it is important to
use data in which the responses of all analyte concentrations have
reached equilibrium (Rmax) and plot at a given range of concentra-
tions of analyte.
The order of analyte injection can be important for generation
of accurate data. It is better to inject the analyte concentrations in a
randomized manner, because injecting the analyte from low to high
concentration can hide problems with regeneration and baseline
drift. Normally, a single experiment contains several injections of
the same analyte at various concentrations to show the stability and
reproducibility of the system. Typically, analyte injections are per-
formed at a high flow rate between 50 and 100 μL/min which helps
to reduce mass transport effects. In order to calculate accurate
binding constants, the association and dissociation phase should
be long enough. In particular, higher analyte concentrations
should show enough curvature in the association phase of the
sensorgrams to allow the fitting model to estimate ka reliably, and
show enough signal decay in the dissociation phase to support a
reliable estimate of kd.
The obtained sensorgrams should be overlaid and used for
curve fitting and determination of the kinetic rate constants and
binding affinity of the interaction (Fig. 7). Robust curve fitting
100
RU
50
0 100 200 300

Time (s)
ka (1/Ms) kd (1/s) KD (M) Rmax (RU) Chi2 (RU)

6.96 x 104 0.0116 1.67 x 10-7 87.4 1.09
Fig. 7 Plots illustrating the experimental curve-fitting methodology for a simple

binding model (1:1 Langmuir). Association and dissociation phases can be seen in
each plot. Black curves indicate the experimental binding data for each tested
concentration (2,000, 500, 250, 125, and 65 nM) of the analyte (ENL protein
489–559) flowing over immobilized Flag-tagged full length DOT1L protein using
the capture affinity immobilization method. Red traces show the corresponding
binding model curves. Calculated binding parameters of the tested protein‐protein
interactions using global fitting are presented
starts with high quality data from several independent runs or

experiments. Many fitting programs (BIAevaluation, Scrubber
2 software) can calculate values for single curves (local fitting) or
calculate a single value for all the fitted curves (global fitting). A
global value is a more robust outcome because it averages all curves.
The most important step is choosing a mathematical model for the
fitting, which should be based on an understanding the chemistry
and physiology of the studied PPIs.
There are a number of kinetic models that can be applied to the
measured data. In general, the simplest model should be used to
determine the kinetic constants such as the Langmuir model, which
describes a 1:1 interaction between two interactants. This model
assumes that the binding reaction is the same and independent at all
binding sites and the reaction rate is not limited by mass transport.
In this case, analysis of the sensorgram curve in the association
phase, in which the binding is measured while the analyte solution
flows over the ligand surface, defines the rate of the complex
formation and allows calculation of the association constant, ka
(M1 s1). In the dissociation phase where the injection of the
analyte is stopped, the rate of the complex decrease is defined by
the dissociation constant, kd (s1). In a kinetic analysis the equilib-
rium constant, KD (M), is calculated as a ratio from these two
kinetic constants: KD ¼ kd/ka (see Note 11).
After the fit is made, the fitted curves should be analyzed with
the following questions: (1) how well do the fitted curves follow
the measured data; (2) is the dissociation fitted correctly; and (3)
are the calculated association constant and Rmax within the
expected range? From the initial 1:1 fitting, a decision should be
made regarding whether there is a need for using two other Lang-
muir kinetic models, which consider the baseline drift and mass
transport in order to improve the fit. Langmuir with drift is com-
monly used in experiments that use a capture surface, where the
capture ligand may dissociate from the capture reagent on the chip
surface, leading to baseline drift. Adding drift to the fitting can
make the fit better but also can skew the results [28]. Therefore, it is
very important that the experimental conditions first be optimized
before use of different fitting models to give better fit of the data
(see Note 12).
When the best fit is chosen, validation of the fitting results is
essential and evaluation of the parameters should be performed.
For example different zones of the experimental curves should be
used for fitting purposes and different fitting should be performed
and compared (local versus globally fitted data). If kinetic para-
meters are consistent throughout all these fits, the chosen kinetic
model is probably correct and the determined kinetic parameters
are confirmed and reliable. The fitting and experimental curves
should be visually inspected and deviations of the fitting from the
actual data can be easily identified as either random deviations,
which should have a normal distribution, or systematic deviations,

which arise because the model is not an adequate description of the
experimental data [29]. Chi2-values and the residual plots for the
fitting should be also examined. Residual plots should form a
random scattering of the same order of magnitude as the noise
level and can reveal systematic differences. Chi2 is the average of
the squared residuals (differences between the measured data
points and the corresponding fitted values) and it is an indicator
of the fitting confidence. Empirically, this value should be less than
10 % of Rmax. Calculated values (Rmax, Chi2, ka, kd) should be
inspected and should be within an expected and reasonable range
(Fig. 7). For example, in order to achieve a valid calculation of kd
the dissociation should be at least 5 % of the starting value. The
value of Rmax, which is also calculated during the fit and reflects the
maximal response when all ligand sites are occupied, should be used
to judge if the curves follow a 1:1 kinetic relationship. If Rmax is
very high compared to the response of the curves, this can be an
indication that the fit is in error.
In summary, the following factors should be considered in
order to obtain robust results, to avoid systematic errors and to
validate them: (1) prepare a wide (100-fold) concentration range of
the analyte, (2) run replicate experiments, (3) inject the samples in a
random order, (4) include buffer injections, (5) use double refer-
encing, (6) apply global analysis, (7) use sensor surfaces with
different ligand concentrations or different immobilization
techniques, and (8) reverse the analyte and ligand to confirm the
results [10, 30].
Most potential sources for generating artifactual data using
SPR technology can be avoided with the proper experimental
design and data processing, using high-quality protein reagents
and proper reporting of the results including all necessary experi-
mental conditions in order to be reproducible. Rich and Myszka
have done an excellent work reviewing and analyzing the biosensor
literature, and recognizing the lack of proper communication of
the use of SPR technology [10, 31–38]. In order to improve
the quality, reliability and reproducibility of reported biosensor
data, they proposed a list of The Bare Minimum Requirements
For An Article Describing Optical Biosensor Experiments
(TBMRFAADOBE) [39] (see Note 13).
3.4.1 Kinetic Analysis 1. Run Desorb (step 1, Subheading 3.1), then dock the CM5
of PPIs chip with ligand and prime with running buffer.
2. Prepare a twofold serial dilution of the analyte into running
buffer based on the optimum concentration range. The recom-
mended range is 10–0.1 times the expected KD value. How-
ever, it is important to avoid analyte concentrations that can
introduce artifacts such as nonspecific binding, steric hindrance
or aggregation. Prepare at least one set of duplicate analyte
sample concentrations as well as a buffer sample (zero analyte

sample), which will be used for double-reference subtraction
during data evaluation prior to curve fitting. Load the dilutions
into polypropylene tubes with rubber caps and place them in
the instrument.
3. Acceptable flow rates for binding studies are 30 μL/min with
contact times (i.e., the association phase) ranging from 2 to
4 min. The dissociation period depends on the dissociation rate
of the analyte and is usually from 10 to 30 min (see Note 14).
4. Inject the sample through the control flow cell followed by the
flow cell(s) immobilized with the ligand. Using the Kinject
command will allow definition of the dissociation period (see
Note 15).
5. Regenerate the surface using two 30 s pulses of optimum
regeneration solution determined previously, followed by
1–2 min stabilization period after regeneration.
6. Repeat steps 4 and 5 with all analyte dilutions, including
“zero” analyte concentration, and determine the binding
curves. The various dilutions of analyte should be tested in
“random order” (see Note 16).
7. Prepare the obtained sensorgrams for evaluation and curve fit-
ting by aligning all sensorgrams based on the inject start point
and set zero on the x-axis. Set the baseline at zero on the y-axis.
8. Subtract the buffer sample from all sensorgrams. If the subtrac-
tion was carried out correctly, the control run will be a flat line.
Delete any spikes that may occur at the start or the end of the
injection, as well as air spikes.
9. Use the command “Kinetics Simultaneous kon/koff” in the
BIAevaluation software to calculate kon, koff, and KD. This
command requires definition of the injection start and end
points, checking the alignment of the sensorgrams and select-
ing the association and dissociation phases on the sensorgram,
which will be used for calculation of the kinetic parameters (see
Note 17). The “Kinetics Simultaneous kon/koff” command fits
the curves globally and gives the most robust data. It is also
possible to perform local fitting, which is usually adopted when
the maximum response, Rmax, during the experiment
decreases, allowing the Rmax parameter to be treated as a local
variable across the data set.
10. Insert the tested analyte concentrations and their units (see
Note 18).
11. Selection of the proper binding model for fitting is important.
The 1:1 binding model and global fit module should be used
unless there is a strong reason to do otherwise (see Note 19).
12. Identify inappropriately fitted curves or aberrant binding pro-
files by visual inspection of the fitted curves. The residual
spread and Chi2 values should fall within the defined accep-
tance criteria.
4 Summary
SPR technology is a versatile and reliable platform to assess and

quantify the kinetics and equilibrium constants of different
biological molecules including protein-protein, protein–DNA,
protein–RNA, and protein–small molecule interactions and thus
separate selective and non-selective interactions. Improved meth-
ods for data collection and processing significantly improve the
quality of biosensor data. The ability to describe binding data
with simple interaction models, together with the option of
providing thermodynamic information about the binding reaction,
contribute to elucidating and understanding of biomolecular rec-
ognition and interaction events. The SPR platform continues to
evolve and provide valuable information on binding kinetics, and is
expanding beyond these studies towards diagnostic, therapeutic,
and drug discovery settings.
5 Notes
1. A Desorb routine should be performed after each use or at least

once per week to keep the needle, tubing and microfluidics
clean. The method for Desorb is under the Tools tab, Working
tools, and the reagents and protocol are described by the
software. This procedure is performed using a docked mainte-
nance chip.
2. For immobilization, either direct or via capture, a very small
amount of material (2–10 μg) is required. By calculating the
Rmax, one can determine whether the ratio of the molecular
masses of the binding partners could limit the response of the
interaction and verify which protein should be immobilized. If
an antibody–antigen interaction is studied, the antibody should
be immobilized as the ligand to avoid binding avidity effects.
3. The pHs of tests are chosen based on the pI of the ligand
protein and have to be 1–3 units below the pI, when the net
charge of the protein will be positive. There are available bioin-
formatics programs that can assess the pI of proteins, for exam-
ple ProtParam available at the SIB Bioinformatics Resource
Portal ExPASy. Many proteins have limited stability in the low
ionic strength, low pH solutions used for pre-concentration, so
dilution of the ligand should be performed just prior to injec-
tion. A pH scouting experiment should be performed on the
flow cell that will be used for the immobilization, not on the
reference surface.
4. As a general rule, pH values down to about 3.5 can be used and
for pH 3.5–4.0 apply citrate buffer. If pre-concentration is
inadequate even at pH 3.5, the ligand may be too acidic and a
different immobilization approach should be considered. Some
ligands can be immobilized at pH values above 5.5; maleate
buffers are suitable for immobilization at pH values in the range
5–6. Furthermore, low ionic strength buffer should be used,
and the ligand should be sufficiently diluted or desalted from
salt-containing stock solutions. The total ion concentration
should be 10 mM or less.
5. After 7 min of activation of the sensor chip surface with EDC/
NHS, approximately 40 % of the carboxyl groups will be acti-
vated leaving a net negative charge during the immobilization
procedure, which enables pre-concentration. Ligand contact
should be completed within 15 min after surface activation to
ensure coupling before the reactive esters on the surface can be
hydrolyzed. If high-density surfaces are required, the injection
time of EDC/NHS can be increased.
6. If the immobilization level is low, increase the contact time if
the immobilization sensorgram indicates that more ligand can
bind. Also, increase the ligand concentration. Make sure that
the EDC and NHS are fresh solutions. Immobilization and
running buffers should not contain primary amines, for exam-
ple Tris or sodium azide, since they can compete with the ligand
for reactive groups on the surface.
7. Immobilization can be performed on each flow cell of the
sensor chip simultaneously or only on the flow cell intended
for use. Experience shows that the activation level of each flow
cell might not be the same if the activation solution is run over
all flow cells at the same time, however, deactivation of all flow
cells together is satisfactory.
8. Once the immobilization is finished, the sensor chip with
bound ligand can be undocked and kept in a 50 mL conical
tube at 4 C. Before using the sensor chip again and docking it
in the instrument, allow the sensor chip to equilibrate to room
temperature.
9. Initial regeneration steps should be performed with mild regen-
eration solutions for a minimal contact time to ensure that the
immobilized ligand will not be affected or otherwise damaged.
If necessary, the concentration of the regeneration solution can
be increased during the procedure. NaOH regeneration solu-
tion should be kept in glass vials and should be freshly prepared
because its efficiency as a regeneration solution decreases over
the time.
10. A larger number of repeated cycles of analyte injection and

regeneration (minimum of 20 cycles) should be repeated to
verify the regeneration. Usually there is a slight reduction in
activity in the first few cycles, thus the assessment should be
calculated using the later cycles, for example 6–25. For some
particular interactions, it is not always possible to find satisfac-
tory regeneration conditions. Two alternative approaches
should be applied: (1) reverse the roles of the ligand and
analyte, since regeneration should preserve the activity of the
ligand; (2) use a capturing approach instead of directly attach-
ing the ligand to the sensor surface. In this case, regeneration is
directed at removing the ligand from the capturing molecule,
and any damage to the ligand does not matter.
11. The association rate constant ka describes the rate of complex
formation, i.e., the number of ligand–analyte (LA) complexes
formed per second in a one molar solution of L and A. The
units of ka are M1 s1 and its values are typically between 103
and 107 in biological systems. The dissociation rate constant kd
describes the stability of the complex, i.e., the fraction of com-
plexes that decays per second. The unit of kd is s1 and is
typically between 101 and 106 in biological systems. A kd
of 1.102 s1 (0.01 s1) means that 1 % of the complexes decay
per second. KD describes the system at equilibrium but not the
dynamics. For the dynamics, refer to the association and disso-
ciation rate constants. The best parameter to compare the
strength of the binding is the dissociation rate constant (kd)
because this parameter gives the time an interaction exists.
12. Baseline drift is usually a sign of a suboptimally equilibrated
sensor surface. Adding sufficient wash steps after regeneration,
longer equilibration times and blank injections in the measure-
ments (double referencing) can improve or eliminate drift.
Several buffer injections before the actual experiment can min-
imize drift during analyte injection. To determine whether a
particular interaction is limited by mass transport and whether
the Langmuir with mass transport model should be used, the
analyte sample should be injected at different flow rates. If the
association curves are different, than this interaction is mass-
transport limited. In contrast, if the association curves are
independent of the flow rate (all binding curves overlap),
then diffusion is not the rate-limiting factor, and the simple
Langmuir model can be applied.
13. A list of the experimental conditions that should be included in
publications in order to facilitate reproducibility by other inves-
tigators include: (1) instrument used in analysis; (2) identity,
source, and molecular weight of the ligand and analyte; (3)
surface type; (4) immobilization condition; (5) ligand density;
(6) experimental buffers; (7) experimental temperatures; (8)
analyte concentrations; (9) regeneration conditions; (10) fig-

ure of binding responses with fit; (11) overlay of replicate
analyses; (12) model used to fit the data; and (13) binding
constants with standard errors.
14. Strong binding (kd > 104 s1) is difficult to analyze when the
dissociation curve is too short. As a rough rule, the dissociation
curve should decrease by at least 5 % before analysis is
attempted. For a dissociation constant of 104 s1 this will
result in a dissociation time of at least 12 min. In addition,
sufficient blank injections with the same long “dissociation”
times are needed to compensate for possible baseline drift [40].
Long dissociation times are not practical when obtained with
every injection. Instead, long dissociation time should be
applied for the highest tested analyte concentration, while
other concentrations should be analyzed using short dissocia-
tion time (200–300 s).
15. Inspect the analyte response on the reference flow cell to iden-
tify nonspecific binding. The bulk refractive index has a square-
shaped response, while nonspecific binding will typically have
an increasing response on the reference, control surface. Pro-
blems with nonspecific binding can, in general, be addressed in
several ways: (1) experimental conditions: purifying the sample
to remove components that interfere with the assay, optimizing
the composition of the running buffer (150 mM or higher salt
concentration will help to suppress nonspecific electrostatic
interactions); (2) sensor surface: different sensor chip types
have different characteristics with respect to nonspecific bind-
ing (Table 1); (3) sample additives such as soluble
carboxymethyl-dextran which can compete for molecules that
bind to the dextran on the sensor surface without interfering
with the analyte–ligand interaction.
16. Before data evaluation, it is important to analyze and compare
the first and last run checking the baseline level, as changes in
this level will affect data evaluation. For example, a rise in the
baseline level indicates accumulation of the analyte, while a
decrease in the baseline level is an indicator of excessive removal
of ligand.
17. When a set of binding curves are being evaluated, the region
that will be selected for implementing the fit is important. The
BIAevaluation software offers a split view function where the
original curves and their derivative functions can be examined,
which can help in the evaluation of whether the model and the
parts of the sensorgram selected are appropriate for data evalu-
ation. The ln(Y0/Y) tool, for the dissociation phase, and ln[abs
(dY/dX0)] tool, for the association phase, enable observation
of binding curves exhibiting single-binding kinetics which are

linear, while they are curved for more complex systems.
18. It is very important that the analyte is well characterized, with
accurate known concentration, high purity (>95 %), func-
tional, no aggregates and monovalent for 1:1 interaction stud-
ies, because regardless of the experimental design, the quality of
the SPR data directly depends on the quality of the reagents
used. In addition, it is important that glycerol is not present in
the analyte solution, as glycerol has high refractive index and
might interfere with the SPR response.
19. Avoid model shopping until a decent fit has been obtained.
Instead, before using fitting models other than the 1:1 Lang-
muir model be sure that: (1) the system is clean and equili-
brated; (2) the ligand is pure and homogenous; (3) the amount
of ligand is low in order to minimize mass transport; (4) the
analyte is pure and homogenous; (5) the analyte concentration
range is wide enough (zero to saturation); (6) the analyte
buffer matches the running buffer, minimizing bulk shift; (7)
the injection time is long enough to give the association curva-
ture; (8) the dissociation time is long enough to give a reason-
able signal decay; (9) blank injections are used for double
referencing; and (10) replicate analyte injections are performed
to demonstrate system stability.
References
1. Cusick ME, Klitgord N, Vidal M et al (2005) 7. Ivanov AA, Khuri FR, Fu H (2013) Targeting
Interactome: gateway into systems biology. protein-protein interactions as an anticancer
Hum Mol Genet 14(Spec No. 2):R171–R181 strategy. Trends Pharmacol Sci 34:393–400
2. Ge H, Walhout AJ, Vidal M (2003) Integrating 8. Fagerstam LG, Frostell-Karlsson A, Karlsson R
‘omic’ information: a bridge between geno- et al (1992) Biospecific interaction analysis
mics and systems biology. Trends Genet using surface plasmon resonance detection
19:551–560 applied to kinetic, binding site and concentra-
3. Wells JA, McClendon CL (2007) Reaching for tion analysis. J Chromatogr 597:397–410
high-hanging fruit in drug discovery at protein- 9. Myszka DG (2000) Kinetic, equilibrium, and
protein interfaces. Nature 450:1001–1009 thermodynamic analysis of macromolecular
4. Blazer LL, Neubig RR (2009) Small molecule interactions with BIACORE. Methods Enzy-
protein-protein interaction inhibitors as mol 323:325–340
CNS therapeutic agents: current progress and 10. Rich RL, Myszka DG (2008) Survey of the year
future hurdles. Neuropsychopharmacology 2007 commercial optical biosensor literature. J
34:126–141 Mol Recognit 21:355–400
5. Ryan DP, Matthews JM (2005) Protein- 11. Hunter MC, O’Hagan KL, Kenyon A et al
protein interactions in human disease. Curr (2014) Hsp90 binds directly to fibronectin
Opin Struct Biol 15:441–446 (FN) and inhibition reduces the extracellular
6. Gerrard JA, Hutton CA, Perugini MA (2007) fibronectin matrix in breast cancer cells. PLoS
Inhibiting protein-protein interactions as an One 9:e86842
emerging paradigm for drug discovery. Mini 12. Chow CR, Suzuki N, Kawamura T et al (2013)
Rev Med Chem 7:151–157 Modification of p115RhoGEF Ser(330)
regulates its RhoGEF activity. Cell Signal biosensor assays. A multivariate cocktail
25:2085–2092 approach. Anal Chem 71:2475–2481
13. Pal A, Huang W, Li X et al (2012) CCN6 25. Drake AW, Klakamp SL (2011) A strategic and
modulates BMP signaling via the Smad- systematic approach for the determination of
independent TAK1/p38 pathway, acting to biosensor regeneration conditions. J Immunol
suppress metastasis of breast cancer. Cancer Methods 371:165–169
Res 72:4818–4828 26. Karlsson R, Larsson A (2004) Affinity measure-
14. Day YS, Baird CL, Rich RL et al (2002) Direct ment using surface plasmon resonance. Meth-
comparison of binding equilibrium, thermody- ods Mol Biol 248:389–415
namic, and rate constants determined by sur- 27. Karlsson R, Katsamba PS, Nordin H et al (2006)
face- and solution-based biophysical methods. Analyzing a kinetic titration series using affinity
Protein Sci 11:1017–1025 biosensors. Anal Biochem 349:136–147
15. Mehand MS, Srinivasan B, De Crescenzo G 28. Rich RL, Papalia GA, Flynn PJ et al (2009) A
(2011) Estimation of analyte concentration by global benchmark study using affinity-based
surface plasmon resonance-based biosensing biosensors. Anal Biochem 386:194–216
using parameter identification techniques. 29. Cornish-Bowden A (2001) Detection of errors
Anal Biochem 419:140–144 of interpretation in experiments in enzyme
16. Towne V, Zhao Q, Brown M et al (2013) kinetics. Methods 24:181–190
Pairwise antibody footprinting using surface 30. Cannon MJ, Papalia GA, Navratilova I et al
plasmon resonance technology to characterize (2004) Comparative analyses of a small mole-
human papillomavirus type 16 virus-like parti- cule/enzyme interaction by multiple users of
cles with direct anti-HPV antibody immobili- Biacore technology. Anal Biochem 330:98–113
zation. J Immunol Methods 388:1–7
31. Rich RL, Myszka DG (2000) Survey of the
17. Nedelkov D, Nelson RW (2003) Surface plas- 1999 surface plasmon resonance biosensor lit-
mon resonance mass spectrometry: recent erature. J Mol Recognit 13:388–407
progress and outlooks. Trends Biotechnol
21:301–305 32. Rich RL, Myszka DG (2001) Survey of the year
2000 commercial optical biosensor literature.
18. Williams C, Addona TA (2000) The integra- J Mol Recognit 14:273–294
tion of SPR biosensors with mass spectrometry:
possible applications for proteome analysis. 33. Rich RL, Myszka DG (2002) Survey of the year
Trends Biotechnol 18:45–48 2001 commercial optical biosensor literature.
J Mol Recognit 15:352–376
19. Davis TM, Wilson WD (2000) Determination
of the refractive index increments of small 34. Rich RL, Myszka DG (2003) A survey of the
molecules for correction of surface plasmon year 2002 commercial optical biosensor litera-
resonance data. Anal Biochem 284:348–353 ture. J Mol Recognit 16:351–382
20. Shen C, Jo SY, Liao C et al (2013) Targeting 35. Rich RL, Myszka DG (2005) Survey of the year
recruitment of disruptor of telomeric silencing 2004 commercial optical biosensor literature.
1-like (DOT1L): characterizing the interac- J Mol Recognit 18:431–478
tions between DOT1L and mixed lineage leu- 36. Rich RL, Myszka DG (2005) Survey of the year
kemia (MLL) fusion proteins. J Biol Chem 2003 commercial optical biosensor literature.
288:30585–30596 J Mol Recognit 18:1–39
21. Drake AW, Tang ML, Papalia GA et al (2012) 37. Rich RL, Myszka DG (2006) Survey of the year
Biacore surface matrix effects on the binding 2005 commercial optical biosensor literature.
kinetics and affinity of an antigen/antibody J Mol Recognit 19:478–534
complex. Anal Biochem 429:58–69 38. Rich RL, Myszka DG (2007) Survey of the year
22. Pace CN, Vajdos F, Fee L et al (1995) How 2006 commercial optical biosensor literature.
to measure and predict the molar absorption J Mol Recognit 20:300–366
coefficient of a protein. Protein Sci 39. Rich RL, Myszka DG (2010) Grading the
4:2411–2423 commercial optical biosensor literature-Class
23. Andersson K, Areskoug D, Hardenborg E of 2008: ‘The Mighty Binders’. J Mol Recognit
(1999) Exploring buffer space for molecular 23:1–64
interactions. J Mol Recognit 12:310–315 40. Katsamba PS, Navratilova I, Calderon-Cacia M
24. Andersson K, Hamalainen M, Malmqvist M et al (2006) Kinetic analysis of a high-affinity
(1999) Identification and optimization of antibody/antigen interaction performed by mul-
regeneration conditions for affinity-based tiple Biacore users. Anal Biochem 352:208–221
Chapter 8
Resonant Waveguide Grating for Monitoring

Biomolecular Interactions
Meng Wu and Min Li
Abstract
Label-free detection technologies have been widely used to characterize biomolecular interactions without
having to label the target molecules. These technologies exhibit considerable potential in facilitating assay
development and enabling new integrated readouts. When combined with high-throughput capability,
label-free detection may be applied to small molecule screens for drug candidates. Based on the resonant
waveguide grating biosensors, a label-free high-throughput detection system, the Epic® System, has been
applied to monitor molecular interactions. Here we describe a generic label-free assay to quantitatively
measure phospho-specific interactions between a trafficking signal—phosphorylated SWTY peptide and
14-3-3 proteins or anti-phosphopeptide antibodies. Compared with the solution-based fluorescence
anisotropy assay, our results support that the high-throughput resonant waveguide grating biosensor
system has shown the capability not only for high-throughput characterization of binding rank and affinity
but also for the exploration of potential interacting kinases for the substrates. Hence, it provides a new
generic HTS platform for phospho-detection.
Key words Resonant waveguide grating, Label-free, Epic, High throughput, Phospho-specific
interactions, SWTY peptide, 14-3-3 proteins, Kinase
1 Introduction
Label-free detection technologies have been widely used to charac-

terize biomolecular interactions without having to label the target
molecules [1, 2]. Different transduction techniques such as surface
plasmon resonance (SPR), optical ellipsometry, quartz crystal micro-
balance, Raman scattering, and calorimetry have been developed. In
spite of their diverse applications, most of their compatibility with
high-throughput screening remains a challenge, although recent
developments, i.e., SPR imaging [3], or incidence-angle dependence
of optical reflectivity difference imaging [4], have shown potential for
increasing the throughput. Resonant waveguide grating (RWG) sen-
sors have been developed into a high-throughput microplate-based
biosensor system, the Epic® System, combining the features of
139
140 Meng Wu and Min Li
label-free detection with high-throughput capability [5]. Among the

many applications label-free optical biosensors can be used for, the
Epic system offers the possibility of developing a generic assay for
evaluating protein‐protein interactions specific to phosphorylated
target sequences.
1.1 Assay Principle The Epic® system is based on a RWG biosensor, which exploits the
of Biosensors evanescent wave generated by the resonant coupling of light into a
waveguide via a diffraction grating. The guided light can be viewed
as one or more mode(s) of light that all have directions of propaga-
tion parallel with the waveguide, due to the confinement by total
internal reflection at the substrate–film and medium–film interfaces.
The waveguide has a higher refractive index value than its surround-
ing medium. Because the guided light mode has a transversal ampli-
tude profile that covers all layers, the effective refractive index N of
each mode is a weighted sum of the refractive indices of all layers:
N ¼ f N ðnF ; nS ; nC ; nad ; d F ; d ad ; λ; m; σ Þ ð1Þ
Here, nF, nS, nC, and nad is a refractive index of the waveguide, the
substrate, the cover medium, and the protein adlayer, respectively.
dF and dad are the effective thicknesses of the film and the protein
adlayer, respectively. λ is the vacuum wavelength of the light used.
m ¼ 0, 1, 2, is the mode number; and σ is the mode type number,
which equals 1 for TE (transverse electric or s-polarized) and 0 for
TM modes (transverse magnetic or p-polarized). Because of its
higher sensitivity, generally the TM (σ ¼ 0) is used for measuring
the binding of biomolecules to the probe proteins immobilized on
the surface of the waveguide substrate.
When a laser illuminates the waveguide at varying angles or
wavelengths, light is coupled into the waveguide only at
corresponding specific angles or wavelengths. This coupling is deter-
mined by the effective refractive index of the guided mode, denoted
as N. The value of N can be calculated numerically from the mode
equation for a given mode of a four-layer waveguide configuration:
" #σ !
2 2 0:5
n2A n2C ðN =nC Þ2 þ ðN =nA Þ2 1
0 ﬃ πm k nF N dF þ dA 2
nF n2C ðN =nC Þ2 þ ðN =nF Þ2 1
2 !0:5 3
2 2 2
n F N n S
þarctan4 5
nS n2F N 2
2 !0:5 3
2 2 2
n F N n C
þarctan4 5 ð2Þ
nC n2F N 2
Here, k ¼ 2π/λ.
Since the laser light is coupled to, and propagates parallel to the
surface in the plane of a waveguide film, this creates an
Label-free Detection of Biomolecular Interactions 141
electromagnetic field (i.e., an evanescent wave) in the liquid adja-

cent to the interface. The amplitude (Em) of the evanescent wave
decays exponentially with increasing distance d from the interface:

d
E m ðd Þ ¼ E m ð0Þexp ð3Þ
ΔZ C
with:
h i1
2 2
1σ σ ðN =n F Þ þ ðN =n C Þ 1
ΔZ C ¼ 0:5
þ 0:5
ð4Þ
k N 2 n2C k N 2 n2C

ΔZC is the penetration depth (also termed as sensing volume;

typically ~150 nm) of the evanescent tail of the waveguide mode
that extends into the cover medium. This means that a target or
complex of a certain mass contributes more to the overall response
when the target or complex is closer to the sensor surface, as
compared to when it is further from the sensor surface.
1.2 Resonant A beta version of the Corning® Epic® System is comprised of three
Waveguide Grating major components for bioassay applications: an Epic® sensor
Biosensor Detection microplate, an RWG detector, and a liquid handling system. The
sensor microplate consists of a glass bottom plate attached to a
holey plastic 384-well plate, which enables high-throughput
screening. Each well in the 384-well Epic® microplate contains an
RWG sensor, which consists of an optical grating and a high index
of refraction waveguide coating [6, 7]. When illuminated with
broadband light at a fixed angle of incidence, these sensors reflect
only a narrow band of wavelengths that is a sensitive function of the
effective index of refraction of the waveguide. The sensors are
coated with a surface chemistry layer that enables covalent attach-
ment (via peptide bond formation) of peptides/proteins or other
biomolecules. Binding of molecular recognition partners to the
immobilized target induces a change in the effective index of
refraction of the waveguide, and this is manifest as a shift in the
wavelength of light that is reflected from the sensor. The magnitude
of this wavelength shift is proportional to the amount of analyte
that binds to the immobilized target. Unlike most commercial SPR
biosensors, which generally use a continuous flow system for the
determination of the kinetics of binding, the Epic® System does not
utilize flow channels. RWG sensors are evanescent in nature which
means that the magnitude of the electric field in the medium
adjacent to the sensor surface decays exponentially from the sensor
surface. The distance from the sensor surface at which the electric
field strength has decreased to 1/e of its initial value is the penetra-
tion depth. For the Epic® System, the penetration depth is
~150 nm. Thus, the system is selective and sensitive to binding
events that take place within this penetration. For most of the
Chemistry
Waveguide
Substrate
White light Reflected

wavelength
Fig. 1 Detection scheme of RWG biosensor for detecting the binding of

biomolecules. The detection scheme of RWG biosensor for detecting the
binding of target molecules (l) in a sample to the probe molecules (Y)
immobilized onto the surface of a waveguide substrate. The specific binding
event is manifested the shift in the wavelength of the reflected light. The
waveguide substrate consists of a region within which a grating structure is
embedded. The probe molecules are coupled to the derivatized waveguide
substrate. From: Wu, M Long S, Frutos AG, Eichelberger M, Li M, Fang Y.,
Journal of Receptors and Signal Transduction, 2009; 29 (3, 4): 202–210,
copyright © 2009, Informa Healthcare. Reproduced with permission from Informa
experiments described in this chapter, a well adjacent to the sample

well is used as a reference. The most recent version of the Epic®
microplate utilizes a self-referencing scheme in which each well of
the plate has its own reference region. This is enabled by a plate
design that provides protein binding chemistry on only half of the
sensor surface so that when a solution of peptide/protein target is
added to the well, it only binds to half of the sensor surface, leaving
the other half as an in-well or self-reference.
Figure 1 is a schematic drawing for detecting the binding of
biomolecules in a sample to the probe molecules immobilized onto
the surface of the waveguide substrate. In a particular assay, the probe
molecules are pre-coupled to the surface, primarily through covalent-
coupling or bio-specific interaction (e.g., biotin–avidin interaction).
Alternatively, the immobilization of probe molecules can be moni-
tored in real-time to ensure the efficiency and quality of the coupling.
Nonetheless, after the immobilization of probe molecules, the bind-
ing of target molecules, in the absence and presence of a modulator,
can be directly monitored by the label-free RWG biosensor, as man-
ifested by the shift in wavelengths or angles of the reflected light.
1.3 RWG Biosensor Phosphorylation is a key posttranslational process that confers

Detection of diverse regulation in biological systems involving specific
Phosphorylated protein‐protein interactions recognizing the phosphorylated
Protein-Protein motifs. Of great interest for ion channels, which represent an
Interactions important, but underdeveloped class of drug targets, is the surface
expression of these membrane proteins modulated by the
posttranslational phosphorylation. A common effect of phosphor-

ylation is a change in protein‐protein interactions. 14-3-3 proteins
were the first protein modules to be identified as binding specifically
to phosphorylated substrates. Evidence from structural studies and
sequence analyses indicates that the primary function of 14-3-3
proteins lies in their preferential binding to phosphorylated sub-
strates, through their antiparallel bivalent binding sites. In addition
to known canonical binding motifs, several earlier reports have
identified interactions between 14-3-3 proteins and the C termini
of target proteins. This characteristic binding (i.e., SWpTY motif)
has high binding affinity that is comparable to that of the canonical
binding motifs. Recently, studies have suggested that 14-3-3 pro-
teins, through binding to inducible phosphorylated motifs, regu-
late protein expression on the cell surface.
Currently, multiple methods are available for detecting phos-
phorylation, including antibody-based fluorescence detection,
radioactive-ATP, and fluorescent-labeled peptide substrates for
FRET or fluorescence polarization detection. Although several
choices are available for characterization of phosphorylation and
potentially for the development of high-throughput screening
assays, such label-based methods are prone to artifacts and other
detrimental effects (i.e., label effect on antibody interactions). Here
we present a protocol for using a high-throughput label-free optical
biosensor system–Epic®—for the interrogation of an example
phospho-specific interaction, 14-3-3 with SWpTY motif.
2 Materials
2.1 Reagents 1. Inorganic salts of analytical purity.

2. PEG-amine (O,O0 -Bis(2-aminopropyl)polyethylene glycol
1900).
3. Ethanolamine.
4. Boric acid.
5. Dimethylsulfoxide (DMSO).
6. 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfo-
nate (CHAPS).
7. Octylphenolpoly (ethyleneglycolether) (NP40).
8. Trition X-100.
9. Dithiothreitol (DTT).
10. PBS buffer: a phosphate buffered saline solution with a phos-
phate buffer concentration of 0.01 M and a sodium chloride
concentration of 0.154 M. The solution pH will be 7.4.
11. Anti-SWpTY antibody produced from rabbit.
12. Peroxidase labeled anti-Rabbit IgG (H + G, made in goat).

13. Anti-14-3-3 antibody (made in rabbit).
14. Supersignal ELISA Femto Maximum Sensitivity substrate kit.
All solutions were prepared in a 10 mM 4-(2-Hydroxyethyl)
piperazine-1-ethanesulfonic acid (HEPES) buffer (pH 7.3) unless
indicated.
2.2 Preparation and 1. 14-3-3ζ protein expressed as a GST-tagged fusion and purified
Purification of from E. coli strain BL21-SI as previously described [8].
Recombinant 14-3-3 2. Recombinant GST-tagged 14-3-3 proteins purified using Glu-
Proteins tathione Sepharose 4B beads.
Fluorescence anisotropy measurements were used to determine
the binding affinity of 14-3-3ζ for SWpTY [8].
2.3 Synthesis and 1. NH2-SWpTY (NH2AhxAhxFRGRSWpTY-COOH, Ahx:

Preparation of 6-aminohexyl-) peptide and non-phosphorylated NH2-SWTY
Peptides peptide synthesized by Biomer Technology.
2. SWpTY, RGRSWpTY-COOH; SWTY, RGRSWTY-COOH;
SWpTD, RGRSWpTD-COOH; SWpTP, RGRSWpTP-
COOH; and SWpTP, RGRSWpTP-COOH peptides from
New England Peptides (or another commercial source).
The peptides were prepared as stock solutions by dissolving in
water, and if necessary with the addition of a minimal amount of
acetonitrile.
3 Methods
3.1 14-3-3 Detection As a common protocol for 14-3-3 detection [9] (Fig. 2), the first
Protocol on the Epic® step is to immobilize NH2-SWpTY peptide (50 μg/ml, pH 7.5), as
System a 14-3-3 binding motif, on the Epic® plate. This was followed by
the addition of ethanolamine to quench the remaining active sites
on the surface. Reference wells were made by the addition of PEG-
amine (50 μg/ml, pH 9.2) or the non-phosphorylated peptide
NH2-SWTY (50 μg/ml, pH 7.5). The second step is the addition
of 14-3-3 protein solutions into both the sample wells and the
reference wells. As seen in Fig. 2c, the wavelength shift after the
subtraction of the signals from the reference wells is termed
the referenced signals for the 14-3-3 binding event. The signal
without subtraction of the reference is termed the unreferenced
signal. Usually six sample wells and two reference wells were used
for each binding reaction unless otherwise described.
To evaluate the specificity of the interaction, 14-3-3 was added
into wells modified with either NH2-SWpTY or non-
phosphorylated NH2-SWTY (Fig. 3a). Only wells immobilized
a b
O O O O O
X
X
=
=
|
|
C C C C C GST-14-3-3ζ
| | | | |
NH-SWpTY
Sample
well
NH2-Peptide, pH 7.5
or
PEG-amine, pH 9.0
GST-14-3-3ζ
NH-Peptide
NH-Peptide
NH-Peptide
NH-Peptide
PEG-amine
Reference
O O O O O well
X
=
=
|
C C C C |
C
| | | | |
OH
GST-14-3-3ζ NH-SWpTY PEG-amine
NH 2 pH 9.0
160 Buffer GST-14-3-3ζ

NH-Peptide
NH-Peptide
NH-Peptide
NH-Peptide
on SWpTY
120
OH
Net Signal (pm)
80
S3
O O O NH O O 40 S1
=
=
|
C C C C C
| | | | | 0
on PEG-amine
-40
S2
0 1000 2000 3000 4000
Time (s)
Fig. 2 Scheme of Epic® detection. (a) Surface reaction scheme for immobilization of N-terminal amino group
modified peptides. X, reactive group; NH2-Peptide, N-terminal amino group modified peptides. (b) Binding
scheme for 14-3-3. Left two panels show the immobilized substrate (NH2-SWpTY) and reference (PEG-amine).
Right two panels show GST-14-3-3ζ binding. In the sample well, where GST-14-3-3ζ is added, GST-14-3-3ζ
binds to the immobilized NH2-SWpTY peptide. In the reference well containing immobilized PEG-amine, there
is no binding of GST-14-3-3ζ. (c) The time-course charts of the respective responses from sample wells and
reference wells, with S1 as an unreferenced signal from GST-14-3-3ζ on NH2-SWpTY (50 μg/ml at pH 5.5.
GST14-3-3 at 1 μM) S2 as an unreferenced signal from GST-14-3-3ζ on PEG-amine (50 μg/ml, pH 9.0 GST14-
3-3 at 1 μM), and S3 as a referenced signal obtained by subtraction of S1 with S2. From: Wu, M Long S, Frutos
AG, Eichelberger M, Li M, Fang Y., Journal of Receptors and Signal Transduction, 2009; 29 (3–4): 202–210,
copyright © 2009, Informa Healthcare. Reproduced with permission from Informa
with the phosphorylated peptide NH2-SWpTY gave detectable

signals. Furthermore, upon addition of competitive 14-3-3 binding
peptides, SWpTY or a nonhomologous 14-3-3 binding peptide
R18, the binding signals were significantly reduced (Fig. 3b). In
contrast, the non-phosphorylated SWTY peptide did not affect the
Fig. 3 Specificity of detection. (a) The referenced signals for the binding of 14-3-3 to immobilized SWpTY, and
SWTY. SWpTY, and SWTY were immobilized at 50 μg/ml, pH 5.5, and GST-14-3-3ζ was added at 0.4 μM. (b)
Competition assay with competitors and control. 1 μM of competitors (SWpTY and R18) and control peptide
(SWTY) were pre-incubated with 0.4 μM GST-14-3-3ζ and applied on the immobilized NH2-SWpTY (50 μg/ml).
(c) Comparison of the referenced signals from the GST-14-3-3ζ and its validated binding mutant GST-14-3-3ζ
(K49E). GST-14-3-3 was used at 0.4 μM
binding of 14-3-3. To be more definitive, the SWpTY peptide

binding was tested with wild type 14-3-3 and 14-3-3 K49E,
which has a mutation at the binding site [14]. Consistently, the
binding signal was obtained only in wild type 14-3-3 proteins
(Fig. 3c). Therefore, these experimental results provide the evi-
dence for the specific detection of 14-3-3 interactions with sub-
strate peptides.
3.2 Validation of 14- To further validate the 14-3-3 binding event on the Epic® plates,
3-3 Detection by In the same Epic® plates above were then subjected to an ELISA assay.
Situ ELISA After washing five times with PBS buffer, 50 μl of the anti-14-3-3
antibody (1:200 dilution) was added and incubated at 4 C for
0.5 h. After washing with PBS buffer with 0.1 % Tween, 50 μl of
Fig. 4 Binding detection of 14-3-3 to immobilized NH2-SWpTY peptide. NH2-SWpTY in different concentra-
tions (pH 5.5) was immobilized on the Epic™ plate. After addition of GST-14-3-3ζ (0.4 μM), the referenced
signals were plotted on the left axis against the concentrations of NH2-SWpTY. The same plate was used for
ELISA detection as described in Subheading 3 and the luminescence signals were plotted on the right axis
Peroxidase labeled Anti-Rabbit IgG (1:1,000 dilution) was added

and incubated at 4 C for 0.5 h. After washing with PBS buffer with
0.1 % Tween again, 50 μl of luminescence substrate from Super-
signal ELISA Femto Maximum Sensitivity substrate kit was added
to each well and subjected to luminescence detection on a multi-
functional plate reader.
The ELISA signals proportionally matched those of the Epic™
System (Fig. 4, filled circles). In contrast, reference wells with PEG-
amine did not produce any ELISA signal (Fig. 4, open circles).
Hence, immobilization of NH2-SWpTY peptide at 50 μg/ml
allows for approximately 80 % of the maximal signal and therefore
was used for the subsequent experiments.
3.3 Affinity Ranking The rank order of affinity of the peptide motifs was determined by
by Competitive competition assays. After immobilization of NH2-SWpTY on the
Replacement Epic® plate, a pre-mixed solution of 14-3-3 and varying concentra-
tions of the competitive peptides were added into the Epic® wells.
The referenced signals were applied in Eq. 5 to determine the
relative affinities of the corresponding competitive peptides.
( !
K D NH‐SWpTY
y¼B x þ K D NH‐SWpTY þ D 0 þ P 0
K D competitor
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2
u KD
NH‐SWpTY
t x þ K D NH‐SWpTY þ D 0 þ P 0 4 D 0 P 0 g
K D competitor
ð5Þ
The fitting was done with Eq. 5 using Origin 7.0 (OriginLab), with
y, the referenced binding signal; B as a constant; D0, the concentra-
tion of NH2-SWpTY that conferred 90 % of saturated GST-14-3-3ζ
response; P0, the concentration of GST-14-3-3; K D NH‐SWpTY , the KD
(¼2.1 μM) for NH2-SWpTY binding to GST-14-3-3ζ as deter-
mined by the concentration response of GST-14-3-3ζ with 50 μg/ml
of NH2-SWpTY immobilized at pH 7.5; K D competitor as KD for competi-
tive peptides, and x as the concentration of competitor peptide.
In contrast to the direct detection using the immobilized bind-
ing partner, the competition assay uses the binding competitors to
measure binding affinity in solution. Affinity detection through
competitive replacement has been a routine method [10, 11].
The distinct signals by competitor peptides can be observed with
considerably different affinities using the Epic® System. The fitting
results for KD values of the competitors SWpTY, SWTY, SWpTD,
and SWpTP are 0.12 0.05 μM, >50 μM, 1.75 0.14 μM, and
6 1.67 μM, respectively. Direct comparison of this method with
the previous detection of SWTY-14-3-3 interaction with fluores-
cence anisotropy suggests a similar rank order of the affinity
(Table 1).
Table 1
Comparison of the 14-3-3–SWpTY interaction detection from fluorescence anisotropy and Epic®
label-free detection
Fluorescence Polarization Epic® System

Sensitivity (nM) 16 38
Linear range (nM) 16–700 38–2,000
Probe KD (μM) 1.7 0.3 2.1 0.4
(GST-14-3-3ζ)
KD (μM, competitive) SWpTY SWTY SWpTD SWpTP SWpTY SWTY SWpTD SWpTP
0.17 >100 2.2 45 0.12 >50 1.75 6
Z factor >0.5 >0.5
S/N ratio ~8.4 ~15
3.4 Characterization To characterize the capability of the Epic® System for high-
for High-Throughput throughput screening of 14-3-3 phospho-specific interactions,
Screening 50 μg/ml of NH2-SWpTY at pH 7.5 was immobilized in one set
of rows and PEG-amine was immobilized in a second set of rows on
the Epic® plate with PEG-amine immobilized wells used as a refer-
ence. GST-14-3-3ζ at 1 μM with varying DMSO amounts was
added to test the compatibility of adding small molecule library in
DMSO solution. GST-14-3-3ζ at 1 μM with or without pre-
incubated 20 μM SWpTY (as a positive inhibition control) was
added into the wells of one Epic® plate with intrawell self-
referencing for the statistic data. The Z factor, Z0 factor, and S/N
ratio [12] were calculated from Eqs. 6 and 7, respectively, where Av
is the average and SD is the standard deviation of the referenced
signals.
Z ¼ 1 3ðSDWith 14‐3‐3 þ SDWithout 14‐3‐3 Þ=ðAvWith 14‐3‐3 AvWithout 14‐3‐3 Þ
ð6Þ

S=N ¼ AvWith 14‐3‐3 AvWith 14‐3‐3 and SWpTY =
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
ð7Þ
ðSDWith 14‐3‐3 Þ2 þ SDWith 14‐3‐3 and SWpTY
The statistic data of responses of GST-14-3-3ζ, with or without

pre-incubated SWpTY, have shown a good signal to noise ratio of
~15. The addition of 20 μM competitor SWpTY peptide in the
solution resulted in a 94 % decrease of the referenced signals, with a
Z factor of 0.8, indicating the robustness and compatibility of the
assay for high-throughput screening.
4 Notes
Common to most of label-free systems is the issue of system prone

to nonspecific interactions. This justifies the thorough investigation
of the specificity of the current assay by including the following
components.
1. Non-phosphorylated SWTY peptide: Only the immobilized
phosphorylated peptide SWpTY gave detectable signals, while
immobilized non-phosphorylated peptide SWTY gave only
minimal background signals.
2. Competitive 14-3-3 binding peptides: Only the competitive
peptides added in the solution gave the reduction of RWG
signals due to the competitive binding. SWpTY peptide or a
nonhomologous 14-3-3 binding peptide R18, the binding
signals were significantly reduced, where non-phosphorylated
peptide SWTY has shown no such effect.
3. Wild type and mutant 14-3-3 proteins: The assay was tested
with wild type 14-3-3 and 14-3-3 K49E, which has a mutation
at the binding site. Consistently, the binding signal was
obtained only in wild type 14-3-3 proteins.
4. Two different SWpTY binding proteins: Different phosphor-
recognizing modules, 14-3-3 protein and an anti-SWpTY anti-
body, have been applied using the same detection format. Both
gave detectable signals, where BSA gave only minimal back-
ground signals.
5. Validation with alternative assays: Direct and indirect schemes
of the Epic assay have been validated by the alternative assays,
by in situ ELISA and fluorescent polarization assay respectively.
The ranking and affinity data have further quantitatively ver-
ified the specificity of the current Epic detection method.
In contrast to peptides, proteins, and nucleic acids, small mole-
cules are quite challenging for label-free detection, because of the
small molecular weight and theoretically much smaller signals. In
addition, the nonspecific interactions by small molecules, especially
those promiscuous hydrophobic ones, require additional control
tests to discriminate artifacts from real signals. To overcome this
issue, a self-reference Epic® plate has been designed, as shown in
Fig. 5. One half of the sensor was coated with active chemical, good
for immobilization; while the other half was not coated, and resis-
tant to the immobilization, and consequently serves as the control.
The Epic® system can detect two areas and normalize the
Fig. 5 Schematic representation of the self-referenced biosensor in the well of

Epic® plates. The pink area denotes the chemical active area good for immobi-
lization; while the grey area denotes the area resistant to the immobilization, and
consequently serves as the control. The Epic® system can detect two areas and
normalize the nonspecific interaction with the surface from the promiscuous
small molecules/aggregates. From: Wu, M Long S, Frutos AG, Eichelberger M, Li
M, Fang Y., Journal of Receptors and Signal Transduction, 2009; 29 (3–4):
202–210, copyright © 2009, Informa Healthcare. Reproduced with permission
from Informa
nonspecific interaction with the surface from the promiscuous small

molecules/aggregates. For example, as expected, the presence of
DMSO did result in a change of bulk refractive index, and, hence, a
change in the unreferenced signals of both test wells and reference
wells. However, proper use of referencing eliminates this effect, and
thus there is no significant difference for the referenced signals
between 0 % DMSO up to 10 % DMSO. This is very helpful in
the case of competitive format for the detection for screening
protein‐protein interaction modulators.
5 Conclusion
A label-free assay for the phospho-specific interactions of 14-3-3

proteins has been developed using the resonant waveguide grating
Epic® System. When the SWpTY (NH2AhxAhxFRGRSWpTY-
COOH) peptide is covalently immobilized to the surface, binding
of 14-3-3 proteins can be detected at as low as 38 nM, with adjus-
table linear ranges depending on the density of immobilized SWpTY.
The specificity of detection was validated using competition experi-
ments with a non-phosphorylated SWTY peptide and a binding
mutant of 14-3-3 protein. Furthermore, competition assays were
performed to determine the rank order of binding affinities of the
different peptide motifs. In addition, the assay is compatible with
high-throughput screening with a Z factor larger than 0.5 and up to
10 % DMSO tolerance. Therefore the reported assay offers a label-
free screening system for phospho-specific interactions applicable at
the quantitative level. With minor modifications, the reported
assay can be applied for the screening of modulators for 14-3-3
protein‐protein interactions. Since the kinase/phosphatase
reactions can introduce the phosphorylation and dephosphorylation
of the SWTY/SWpTY sequences immobilized on the surface of
the sensor plate. The current assay can be further developed into a
high-throughput label-free protocol for modulators of kinases
and phosphatases. In addition, the assay can be applied into other
peptide–protein interactions, i.e., PDZ domains and other extracel-
lular loop sequences of ion channels [13].
The reported assay conditions are tuned for 14-3-3 interactions
with the SWpTY peptide. Many aspects of the experimental condi-
tions may be transferable to other interaction systems, especially
protein–peptide interactions that are exemplified by phospho-
specific antibody to the SWTY peptide. Comparison with
solution-based assays such as fluorescence polarization suggests
that at least for the interaction between SWpTY and 14-3-3, the
Epic® System gave comparable characteristics in terms of sensitivity
and dynamic range. Because Epic® allows for quick estimation of
binding affinity, it provides an attractive means for quantitative
assessment of interactions between different ligands and one
protein or between one ligand and different interacting proteins.

This could be applicable to a number of assays such as antibody
evaluation and detection of interacting components in cell lysates.
References
1. Halai R, Cooper MA (2012) Using label-free 8. Wu M, Coblitz B, Shikano S, Long S, Spieker
screening technology to improve efficiency in M, Frutos AG, Mukhopadhyay S, Li M (2006)
drug discovery. Expert Opin Drug Discov Phospho-specific recognition by 14-3-3 pro-
7:123–131 teins and antibodies monitored by a high
2. Filiou MD, Martins-de-Souza D, Guest throughput label-free optical biosensor. FEBS
PC, Bahn S, Turck CW (2012) To label Lett 580:5681–5689
or not to label: applications of quantitative 9. Wu M, Long S, Frutos AG, Eichelberger M, Li
proteomics in neuroscience research. Proteo- M, Fang Y (2009) Interrogation of phosphor-
mics 12:736–747 specific interaction on a high-throughput label-
3. Saito A, Kawai K, Takayama H, Sudo T, Osada free optical biosensor system-Epic system.
H (2008) Improvement of photoaffinity SPR J Recept Signal Transduct Res 29:202–210
imaging platform and determination of the 10. Huang X (2003) Fluorescence polarization
binding site of p62/SQSTM1 to p38 MAP competition assay: the range of resolvable inhib-
kinase. Chem Asian J 3:1607–1612 itor potency is limited by the affinity of the
4. Landry JP, Gray J, O’Toole MK, Zhu XD fluorescent ligand. J Biomol Screen 8:34–38
(2006) Incidence-angle dependence of optical 11. Dai JG, Murakami K (2003) Constitutively and
reflectivity difference from an ultrathin film on autonomously active protein kinase C asso-
solid surface. Opt Lett 31:531–533 ciated with 14-3-3 zeta in the rodent brain. J
5. Li G, Lai F, Fang Y (2012) Modulating cell-cell Neurochem 84:23–34
communication with a high-throughput label- 12. Zhang L, Wang H, Masters SC, Wang B,
free cell assay. J Lab Autom 17:6–15 Barbieri JT, Fu H (1999) Residues of 14-3-
6. Fang Y, Ferrie AM, Fontaine NH, Yuen PK 3 zeta required for activation of exoenzyme
(2005) Characteristics of dynamic mass S of Pseudomonas aeruginosa. Biochemistry
redistribution of epidermal growth factor 38:12159–12164
receptor signaling in living cells measured 13. Sun H, Li M (2013) Antibody therapeutics
with label-free optical biosensors. Anal Chem targeting ion channels: are we there yet? Acta
77:5720–5725 Pharmacol Sin 34:199–204
7. Fang Y, Ferrie AM, Fontaine NH, Mauro J, 14. Zhang L, Wang H, Liu D, Liddington R, Fu H
Balakrishnan J (2006) Resonant waveguide (1997) Raf-1 kinase and exoenzyme S interact
grating biosensor for living cell sensing. Bio- with 14-3-3zeta through a common site involv-
phys J 91:1925–1940 ing Lysine 49. J Biol Chem 272:13717–13724
Chapter 9
Quartz Microbalance Technology for Probing Biomolecular

Interactions
Gabriella T. Heller, Alison R. Mercer-Smith, and Malkiat S. Johal
Abstract
Quartz crystal microbalance with dissipation monitoring (QCM-D) is a useful technique for observing the
adsorption of molecules onto a protein-functionalized surface in real time. This technique is based on
relating changes in the frequency of a piezoelectric sensor chip, onto which molecules are adsorbing, to
changes in mass using the Sauerbrey equation. Here, we outline the cleaning, preparation, and analysis
involved in a typical QCM-D experiment, from which one can obtain mass adsorption and kinetic binding
information.
Key words QCM-D, Surface, Piezoelectric, Sauerbrey, Frequency, Mass, Adsorption, Deposition,
Kinetics
1 Introduction
Quartz crystal microbalance with dissipation monitoring

(QCM-D), unlike traditional gravimetric techniques, can measure
the binding of molecules to a surface with nanogram sensitivity
(Fig. 1). This is achieved by exploiting the piezoelectric properties
of crystalline quartz, namely, that applying a precise alternating
current to a quartz crystal will cause the crystal to expand and
contract, resulting in oscillatory motion. This oscillatory motion
is a function of the size and mass of the quartz crystal. Upon
altering the mass of the crystal (for example, upon protein binding
to the surface) the resonant frequency of the quartz will shift.
QCM-D technology works by applying an AC potential to a quartz
crystal, which induces the quartz to oscillate at its resonant fre-
quency. By monitoring this resonant frequency over time as mole-
cules are adsorbing to the surface, QCM-D is able to detect these
subtle changes in frequency. Because these changes in frequency are
related to changes in mass, QCM-D can be thought of as an
extremely sensitive balance that measures changes in mass at the
surface over time [1].
153
154 Gabriella T. Heller et al.
Fig. 1 QCM-D flow cell schematic
Under the assumptions of a rigid surface, the decrease in fre-

quency can then be used to estimate changes in mass using the
Sauerbrey equation. Furthermore, information about the viscoelas-
tic properties of the adsorbed film may be obtained using dissipa-
tion values, which may be measured by breaking the electrical
circuit supplying voltage to the sensor chip. The resulting damping
oscillations of the sensor chip are then used to determine dissipa-
tion values. If the deposited film is rigid, the energy from the sensor
chip oscillations dissipates slowly, and a low dissipation value is
reported [1]. Under these assumptions, the Sauerbrey relation
may be used. Otherwise, a viscoelastic model must be used to
accurately relate frequency changes to changes in mass on the
surface. Without examination of dissipation, changes in frequency
cannot be accurately correlated to changes in mass.
QCM-D is a useful technique that is comparable to surface plas-
mon resonance (SPR) and dual polarization interferometry (DPI).
While SPR and DPI are optically based methods, QCM-D measure-
ments are based on mechanical frequency changes. As a result, QCM-
D measurements account for mass due to trapped solvent within the
adsorbent, whereas DPI and SPR do not. Consequently, comparing
results from the two methods allows one to decouple the mass effects
due to solvent hydration at the surface. Like SPR and DPI, QCM-D
can be used to make binding affinity measurements. However, as is the
case with any surface technique, a true affinity will not be obtained
because the protein’s degrees of freedom are constrained due to its
immobilization state on a surface [2].
QCM-D experiments are relatively inexpensive and simple to
perform. There are three major steps to a typical QCM-D proce-
dure: (1) cleaning, (2) functionalization and data collection, and
(3) analysis. As is the case in any surface measurement, a clean
QCM-D Methods 155
system is crucial for reproducible results, and so the sensor chips

and flow cells must be thoroughly decontaminated before use. In
some cases after cleaning, the sensor chips must be functionalized
outside the flow cells; in other cases, this can be accomplished with
the sensor chips mounted in the flow cells before the interacting
protein is exposed to the sensor chips. In either case, buffer is
flowed through the flow cell before taking measurements in order
to achieve a stable baseline. Once the measured frequency and
dissipation values are stable, the interacting protein or ligand is
introduced over the sensor chips. The data collected can then be
analyzed to obtain mass deposition values and kinetic information.
Overall, the steps involved in the QCM-D procedure are straight-
forward, making this an approachable and efficient technique for
measuring protein–ligand interactions.
2 Materials
When preparing solutions, use ultrapure water (resistivity

>18 MΩ cm) and analytical grade reagents. Prepare solutions at
room temperature.
2.1 QCM-D Setup 1. O-rings.

(See Note 1) 2. SiO2 sensor chips.
3. Gold sensor chips.
4. Tubing.
5. Gaskets.
6. Tweezers.
7. Teflon sensor holder.
8. Lint-free tissues.
9. Beakers.
2.2 UV/Ozone 1. UV lamp.

Treatment Equipment 2. Safety chamber.
(See Note 2)
3. Stand.
2.3 Cleaning 1. Alkaline liquid detergent solution, specifically designed for

Chemicals sensitive lab equipment cleaning: 2 % (v/v) alkaline liquid
concentrate diluted with water.
2. Ethanol, 200 Proof.
3. Compressed nitrogen gas.
4. Oxidizing solution: 1:1:5 volume solution of ammonium
hydroxide–hydrogen peroxide–water.
5. Disposable, phosphate-free liquid detergent.
6. Water–ethanol mixture: 1:1 volume solution of water and
ethanol.
2.4 Sensor Chip 1. Phosphate-buffered saline (PBS): Weigh 8 g NaCl, 0.2 g KCl,
Preparation Materials 1.44 g Na2HPO4, and 0.24 g KH2PO4. Transfer to a 1,000 mL
volumetric flask. Fill the flask with ultrapure water to a few
milliliters below the 1,000 mL mark and transfer to a bottled
1 L container. Adjust the pH to 7.4 with HCl or NaOH. Add
ultrapure water to obtain a final volume of 1 L as needed.
2. 10 mM salicylic acid solution in water.
3. 10 mM mercaptoundecanoic acid (MUA) solution in methanol.
4. 5 mM 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide/
N-Hydroxysuccinimide (EDC/NHS) solution in water.
5. 2 mg/mL metmyoglobin in PBS.
6. 50 mM ethanolamine in ultrapure water.
7. Borate buffer: Weigh 6.18 g boric acid and 1.3 g NaOH.
Transfer to a 500 mL graduated cylinder and fill with water to
a few milliliters below the 500 mL mark. Transfer to a bottled
500 mL container. Adjust the pH to 9.5 with NaOH and HCl,
as needed. Shake to mix thoroughly. Add ultrapure water to
obtain a final volume of 500 mL.
8. 0.2, 0.5, 0.7, 1 M sodium azide.
9. 1 mM Nα-(tert-Butoxycarbonyl)-L-asparagine, Boc-L-asparagine
(Boc-Asn-OH) solution in 200 proof ethanol.
10. 200 mM Nickel (NiSO4) solution in water.
11. 2 mg/mL bovine serum albumin (BSA) in PBS.
3 Methods
Due to QCM-D’s high sensitivity, a common source of poor mea-

surements is contamination of the instrument. Thus, it is critical to
carefully clean the sensor chips, flow cells, tubing, and handling
equipment (tweezers, etc.) before any data is collected. The decon-
tamination procedures outlined below (Subheadings 3.1 and 3.2)
should be performed before the experiment (Subheading 3.3) is
started to ensure contamination does not confound results
obtained using QCM-D.
3.1 Sensor 1. Assemble UV/Ozone treatment setup (see Note 2).

Decontamination 2. Expose SiO2 sensor chips surface side up to UV/Ozone treat-
3.1.1 Cleaning SiO2 ment for 10 min.
Sensor Chips 3. Soak sensor chips in alkaline liquid detergent solution for
(See Note 3) [3] 10 min.
QCM-D Methods 157
4. Rinse sensor chips with water and then rinse with ethanol.
5. Dry sensor chips with nitrogen gas (see Note 4).
6. Expose sensor chips to UV/Ozone treatment for another
10 min.
7. Place sensor chips in decontaminated QCM-D flow cells
immediately.
3.1.2 Cleaning Gold 1. Prepare a 75 C water bath.

Sensor Chips [3] 2. Expose gold sensor chips, active surface side up, to UV/Ozone
treatment for 10 min.
3. Prepare an oxidizing solution in a beaker (see Note 5).
4. Place sensor chips in a Teflon sensor holder and submerge
holder in the oxidizing solution.
5. Place the beaker in the hot water bath for 5 min.
6. Take the beaker out and let cool.
7. Submerge the Teflon holder with the sensor chips in a beaker
containing water.
8. Rinse sensor chips individually with ethanol.
9. Dry sensor chips individually with nitrogen gas.
3.2 Flow Cell 1. Mount a “cleaning sensor chip” (see Note 6) in each flow cell,
Decontamination ensuring that a good seal is formed between the sensor chip and
the O-ring.
3.2.1 Regular Cleaning
(To Be Performed Before 2. Flow approximately 10 mL of alkaline liquid detergent solution
Each Use) [3] (see Note 7) through the measurement chamber.
3. Flow 20 mL of water through the flow cells.
4. Allow air to flow through the flow cells for 5 min.
5. Remove cleaning sensor chips.
6. Use nitrogen gas to blow extra liquid out of the tubing and dry
the flow cells.
3.2.2 Deep Cleaning (To 1. Disassemble tubing and measurement chambers. Remove
Be Performed as Needed) O-rings.
(See Note 8) [3] 2. Place any component that comes directly into contact with
liquid into a beaker filled with disposable, phosphate-free liquid
detergent. Sonicate for 1 h.
3. Rinse components with water and place in a beaker containing
water–ethanol mixture. Sonicate for at least 15 min.
4. Rinse components with water and place them in a beaker con-
taining water. Sonicate for at least 10 min.
5. Remove all components and let air-dry. Lint-free tissues and

nitrogen gas may be used to dry components, but this is not
essential.
3.3 Attachment When selecting a method for attaching proteins to a surface, it is

Procedures important to consider orientation of the protein, minimization of
secondary interactions, and packing density. Some of these meth-
ods of attachment are performed in the QCM-D flow cell, while
others are performed outside of the flow cell. In this section we
outline three popular methods of attaching proteins to surfaces and
discuss the advantages and limitations of each method (see Note 9).
3.3.1 Electrostatic It is often possible to form a densely packed layer of protein on the
Protein Layer Formation surface due to electrostatic interactions between the protein and a
charged surface. This method is very simple as it can be carried out
entirely within the QCM-D flow cell. It should be noted, however,
that as a layer of protein is formed, the tertiary structure of the
protein becomes compromised, making this method of attachment
less desirable than the others discussed in this chapter. Below, we
outline a procedure for creating a layer of charged protein on a SiO2
sensor chip (see Note 10) and measuring the deposition of salicylic
acid onto the surface.
1. Mount clean SiO2 sensors chips in decontaminated flow cells.
2. Flow PBS until the frequency stabilizes (see Note 11).
3. Flow BSA solution for 10 min.
4. Rinse with PBS until stable.
5. Flow over solution of interacting salicylic acid until stable.
3.3.2 Cross-Linking Cross-linking is the process of covalently bonding two or more

Protein Attachment molecules, and it is commonly used for QCM-D studies. Cross-
linking reagents will contain two reactive ends, which chemically
attach to specific functional groups on other molecules or proteins.
There are several cross-linking reagents that can be selected based
on chemical specificity, length between conjugated molecules, and
solubility (see Note 12).
In most of our studies, we employ EDC/NHS cross-linking
chemistry (Fig. 2). Unlike procedure 3.3.1, this procedure is rela-
tively time-intensive, and sensor preparation is performed outside
of the QCM-D flow cell. Below we summarize the procedure for
attaching metmyoglobin to the surface using EDC/NHS cross-
linking and measuring its binding interaction with azide.
1. Place clean gold sensor chips in a solution of MUA overnight.
2. Rinse sensor chips with ethanol and dry with nitrogen gas.
QCM-D Methods 159
Fig. 2 EDC/NHS functionalization schematic for attaching proteins to a gold surface
3. Submerge in a solution of EDC/NHS for 2 h at 4 C

4. Rinse with PBS and place in a solution of metmyoglobin for 2 h
at 4 C.
5. Rinse with ultrapure water and place in a solution of ethanol-
amine for 2 h at 4 C.
6. Rinse crystals with ultrapure water and dry with nitrogen gas.
Mount in a decontaminated liquid flow cell.
7. Obtain a stable baseline by flushing the QCM-D flow cell with
borate buffer.
8. Flow 0.2 M sodium azide in borate buffer (pH 9.5) over the
surface.
9. Rinse with PBS until stable (see Note 11).

10. Repeat steps 1–9 for other concentrations of sodium azide
(0.5, 0.7, 1 M).
3.3.3 Histidine Tag When used appropriately, the His-tag capture method of protein-
Capture immobilization is perhaps the most desirable method because of its
ability to orient proteins on the surface. When a histidine tag is on
the opposite side of a protein than the binding site, this procedure
allows the binding site to be exposed during immobilization, thus
preventing results from becoming confounded by its inaccessibility.
Unfortunately, this method does not apply to all proteins, although
some analogs do exist.
1. Rinse Au crystal with ethanol (see Note 13) until frequency
stabilizes (see Note 11).
2. Flow Boc-Asn-OH solution through the systems for 15 min at
0.1 mL/min (see Note 14).
3. Flow ethanol until the frequency stabilizes.
4. Flow water until the frequency stabilizes.
5. Charge with nickel solution by flowing nickel solution through
the system (see Note 15).
6. Rinse with water until stable.
8. Flow protein with an exposed histidine residue through the
system.
10. Flow ligand through the system.
3.4 Data Analysis Changes in frequency obtained from QCM-D can be converted to
changes in mass using either the Sauerbrey relation or viscoelastic
modeling, depending on the viscoelastic properties of the surface.
The Sauerbrey equation, Eq. 1, describes a linear relationship
between frequency shifts and mass changes in thin films under the
assumption that the film is rigidly attached to the sensor.
CΔf
Δm ¼ ð1Þ
n
In this equation Δf is the frequency shift (Hz), Δm is the change

of mass per area (ng cm2), C is a constant (17.7 ng Hz1 cm2
for a 4.95 MHz quartz crystal), and n is the overtone number
(1, 3, 5, or 7) [3]. For simplicity, here we assume that that dissipa-
tion values are minimal, and it is therefore acceptable to proceed
using the Sauerbrey relation to convert changes in mass to changes
in frequency (see Note 16). Experimental frequency data and the
QCM-D Methods 161
Fig. 3 Sample QCM-D protein-protein binding data. After a stable baseline was achieved for the functionalized
sensor chip, interacting protein was introduced to the system at 150 s. Association was terminated at 425 s,
and the sensor chip was rinsed with buffer to measure dissociation rates until equilibrium was achieved
corresponding mass data, calculated by the Sauerbrey equation, can

be found in Fig. 3.
As is the case with any surface measurement, data can become
confounded by unintended interactions between the ligand and the
functionalized sensor surface. The method of data analysis
presented here assumes that unintended interactions on the surface

are minimal.
Procedure for Analyzing Data [2]:
1. Divide data into the following sections (Fig. 3):
Baseline: a stable frequency before exposure (drift less that 2 Hz
per 10 min).
Association: time during which the sample flows over the sensor
and association occurs.
Dissociation: time during which buffer flows over the sensor
and dissociation occurs.
Equilibrium: steady state is reached.
2. Fit the Dissociation segment of the data according to Eq. 2.
f t ¼ F 0 ekoff ðtt 0 Þ ð2Þ
In this case ft denotes change in frequency at time t (seconds) in

Hz, t0 denotes the start of the dissociation in seconds, F0
denotes the change in frequency at t0 in Hz, and koff denotes
the dissociation rate constant in inverse seconds (see Note 17).
Because ft, F0, and t0 are all known constants, they can be
inserted into Eq. 2 to obtain koff.
3. Fit the Association segment of the data according to Eq. 3.
kon C ΔF max ðkon Cþkoff Þðtt 0 Þ

ft ¼ e 1 þ F0 ð3Þ
kon C þ kon
Here ft denotes change in frequency at time t (seconds) in Hz, kon

is the association rate constant in M1 s1, koff is the dissociation
constant determined in step 2, C is the ligand concentration in
mol/l, ΔFmax is the frequency shift for a fully saturated surface, t0
is the time at the start of the association phase, and F0 is the
change in frequency at t0 in Hz. By inputting all known constants,
kon can be obtained.
4. The dissociation constant, KD, can be obtained using Eq. 4
koff
KD ¼ ð4Þ
kon
where KD is in M1.
5. Repeat steps 1–4 for each ligand concentration and average KD
values.
QCM-D Methods 163
4 Notes
1. Follow manufacturer’s assembly instructions. Be sure to check

chemical compatibility between solvents and all liquid
handling equipment including tubing, O-rings, and gaskets.
For simplicity, it is helpful to ensure that all tubing is cut to
the same length.
2. UV/Ozone treatment should be performed in a chamber in
which sensor surfaces are approximately 5 mm from a lamp,
which generates light at 185 and 254 nm. The UV light with
the ozone, produced in the breakage of the O-O bond by the
185 nm light, volatilize organic contaminants on the surface
and slightly oxidize the surface [4].
3. The same procedure may be followed to treat any tweezers used
for handling sensor chips.
4. The sensor chips are dried with nitrogen gas from a pressurized
source.
5. Handle the oxidizing solution carefully. Wear appropriate
gloves when using this solution. Dispose of the solution
down the sink with copious amounts of water.
6. It is common for older, worn out sensor chips that are no
longer sensitive enough for measurements to be recycled as
“cleaning sensor chips.” The presence of the sensor creates a
seal within the measurement chamber, so that the pressure
within the system can draw up cleaning fluid.
7. Other detergents may also be used such as sodium dodecyl
sulfate. Before use, it is crucial to ensure that all liquids are
compatible with tubing, O-rings, and gaskets.
8. We typically perform this after 5–10 experiments or if it takes
particularly long for the system to achieve a stable baseline (see
Note 11).
9. In all procedures, we generally use flow rates between 100 and
300 μL/min. When switching between solutions, it is crucial to
ensure that the pump is not drawing up air, which can cause air
bubbles and disrupt measurements. This can be avoided by
temporarily stopping the pump before removing it from one
solution and only restarting it once the open-end of the tubing
is submerged in another solution. The QCM-D sensor used
consisted of an AT-cut piezoelectric quartz crystal disk coated
with a gold electrode (100 nm thick) on the underside and an
active surface layer of gold or SiO2 (this varies by experiment).
All QCM-D sensors described have been optically polished by
the vendor with a root-mean-square roughness less than 3 nm.
10. Local positive regions interact with the negatively charged SiO2
surface, thus allowing for layer formation.
11. We define stable frequency as a change no larger than 2 Hz over
10 min. Sometimes this can take 15 min, but sometimes it can
take as long as 3 h.
12. This website provides a great review of cross-linking:
http://www.piercenet.com/browse.cfm?fldID¼CE4D6C5C-
5946-4814-9904-C46E01232683.
13. Be sure to use tubing that is compatible with ethanol.
14. A decrease in frequency of 12–13 Hz should be observed.
15. A small change in frequency should be observed.
16. Theoretically, as soon as dissipation values are larger than zero,
the Voigt model or another viscoelastic model should be used.
When a surface displays viscoelastic behavior, the linear nature
of the frequency-mass relation fails, and true mass on the surface
will be underestimated using the Sauerbrey equation. Data
generally require viscoelastic modeling if the dissipation values
(in 1E6) are greater than 5 % of the frequency shifts (in Hz).
17. For best results, this segment should consist of at least 50 points.
Acknowledgements
We would like to acknowledge the Pomona College Chemistry

Department and Pomona College Summer Undergraduate
Research Program for their continual support.
References
1. Johal MS (2011) Quartz crystal microbalance. interactions: a molecular cloning manual. Cold
In: Press C (ed) Understanding nanomaterials, Spring Harbor Laboratory Press, Cold Spring
1st edn. CRC Press, Boca Raton, FL, pp 101–108 Harbor, NY, pp 273–284
2. Hauck S, Drost S, Prohaska E, Wolf H, D€ ubel S 3. www.qsense.com
(2005) Analysis of protein interactions using a 4. Vig JR (1985) UV/Ozone cleaning of surfaces.
quartz crystal microbalance biosensor. In: Gole- J Vac Sci Technol A 3:1027
mis EA, Adams PD (eds) Protein-protein
Chapter 10
Label-Free Kinetic Analysis of an Antibody–Antigen

Interaction Using Biolayer Interferometry
Sriram Kumaraswamy and Renee Tobias
Abstract
Biolayer Interferometry (BLI) is a powerful technique that enables direct measurement of biomolecular
interactions in real time without the need for labeled reagents. Here we describe the analysis of a high-
affinity binding interaction between a monoclonal antibody and purified antigen using BLI. A simple Dip-
and-Read™ format in which biosensors are dipped into microplate wells containing purified or complex
samples provides a highly parallel, user-friendly technique to study molecular interactions. A rapid rise in
publications citing the use of BLI technology in a wide range of applications, from biopharmaceutical
discovery to infectious diseases monitoring, suggests broad utility of this technology in the life sciences.
Key words Label-free, Antibody, Antigen, PSA, Biolayer interferometry, Affinity constant, Kinetic
analysis, Ligand, Analyte, Biosensor, Association, Dissociation
1 Introduction
1.1 Label-Free Label-free biosensors are in routine use for the analysis of the
Technology for kinetics of binding interactions between two biomolecules [1]. In
Analysis of Molecular biosensor-based analysis, one of the binding partners is immobi-
Interactions lized on the solid surface of the biosensor (ligand) while the other
molecule is present in solution (analyte). While traditional techni-
ques for measuring binding activity and affinity rely on a similar
format, they typically require enzymatic or fluorescent molecular
labeling. Generating labeled biomolecules not only consumes time
and material, but can lead to altered protein activity or steric
blocking of binding sites. Unlike standard endpoint assays such as
ELISA, label-free biosensor technology enables monitoring of
binding interactions in real time. Real-time kinetic measurements
provide more information on mechanisms of interaction, including
association rates (ka), dissociation rates (kd), and affinity constants
(KD) [2]. Label-free technology has greatly advanced in recent
years, enabling rapid, sensitive and accurate measurement of bind-
ing kinetics, affinity and activity of biomolecular complex formation
165
166 Sriram Kumaraswamy and Renee Tobias
that minimizes artifacts or issues associated with traditional end-

point techniques [3].
1.2 BioLayer BLI is an optical technique that utilizes disposable fiber-optic bio-
Interferometry sensors for measurement of biomolecular interactions. In BLI,
white light is directed down the length of the biosensor fiber
toward two interfaces separated by a thin layer at the tip: the
biocompatible surface of the tip, and an internal reference layer.
Light is reflected back to the detector from each of the two layers,
and the two reflected beams interfere constructively or destruc-
tively at different wavelengths in the spectrum. When the tip of a
biosensor is dipped into a sample, analyte binds to immobilized
ligand on the biosensor surface. This binding forms a molecular
layer which increases in thickness as more analyte molecules bind to
the surface. As the thickness at the biosensor tip increases, the
surface layer effectively moves away from the internal reference
layer and the detector, creating a shift in the interference pattern
of the reflected light (Fig. 1a). The spectral pattern changes as a
function of the optical thickness of the molecular layer, i.e., the
number of molecules bound to the biosensor surface. This shift is
monitored at the detector, and reported on a sensorgram as a
change in wavelength (nm shift) (Fig. 1b). Multiple layers can be
bound sequentially to the surface, providing accurate, comprehen-
sive kinetic data on the interactions between molecules [4].
1.3 Biosensors BLI biosensors are made of glass, and coated with a proprietary
biocompatible matrix which minimizes nonspecific binding to the
surface. This matrix is pre-coated with one of a wide selection of
capture chemistries for specific binding of analyte molecules in a
sample. Using a dip-and-read format, biosensors are moved
a b
Reflected Beams
Incident
white light
nm shift
Biocompatible
surface
Bound
molecule
Un-bound molecules
have no effect
Time
Fig. 1 BLI diagram. (a) As more molecules bind to the surface of the biosensor, the interference pattern of the
reflected light changes, creating a wavelength shift that is reported in real time on a sensorgram (b)
Kinetic Analysis Using Biolayer Interferometry 167
between samples held in standard 96- or 384-well microplates or in

a micro-volume drop holder, eliminating the need for microflui-
dics. Ease of use combined with the ability to measure multiple
samples simultaneously allows for higher throughput measurement
of binding kinetics and affinity, rapid determination of analyte
concentration, and screening of biomolecular interactions. Only
molecules binding to or dissociating from the surface of the bio-
sensor shift the spectral interference pattern and generate a
response. Unbound molecules in the surrounding solution do not
affect the interference pattern, enabling measurements in crude
samples such as cell lysates or culture supernatants. Changes in
refractive index only minimally affect BLI signals, enabling analysis
in solutions containing high refractive index components such as
glycerol or DMSO. BLI biosensors are cost-effective and can be
disposed of after a single use, or regenerated and reused using
optimized conditions suitable for the binding pair under study.
Samples are not consumed or destroyed in BLI analysis, and can
be recovered when the assay is complete.
1.4 Binding Kinetics An example of a binding kinetics experiment on a BLI biosensor is

with BLI shown in Fig. 2a. The experiment begins with immobilization
(loading) of a ligand molecule, such as an antibody, on the surface
of the biosensor. The ligand-loaded biosensor is dipped into a
solution containing the analyte (association) followed by dipping
into buffer (dissociation). Measurements are plotted in real time on
a sensorgram (Fig. 2b). Binding association and dissociation rate
constants (ka and kd, respectively) and affinity constant (KD) are
calculated by fitting the binding and dissociation curves using
mathematical equations. The simplest model used to describe an
interaction between two biomolecules is
ka
A þ B ⇌ AB ð1Þ
kd
where A represents the ligand, and B is the analyte [5]. This binding
model assumes a 1:1 interaction, where one ligand molecule inter-
acts with one analyte molecule, and binding is independent and of
equal strength for all binding sites. The rate of complex (AB)
formation during the association step of an assay is a function of
the association constant, ka, which is expressed in M1 s1. When
the biosensor is dipped into buffer in the dissociation step, the
complex dissociates back to A and B. The dissociation constant,
kd, identifies the fraction of complexes decaying per second, and is
expressed in units of s1. The dissociation constant is a measure of
the stability of the interaction; the smaller the kd, the more stable
the complex. The affinity constant, KD, is a measure of how tightly
a ligand binds to its analyte. It is the ratio of the on-rate and
off-rate:
a
Baseline Loading Baseline Association Dissociation
Streptavidin Biotin Antibody Analyte
b 4.0
3.6
Dissociation
3.2
Association
2.8
2.4
Binding, nm
2.0
Loading
1.6
1.2
0.8
0.4
Baseline
0
0 400 800 1,200 1,600 2,000 2,400
Time (seconds)
Fig. 2 Binding kinetics experiment on Streptavidin biosensors. (a) Biotinylated ligand is immobilized on the
streptavidin-coated surface. After a baseline step, an association step is performed where analyte binding
occurs, followed by a dissociation step in buffer. (b) A typical sensorgram trace showing loading, baseline,
association, and dissociation assay steps for replicate samples
½A ½B kd
KD ¼ ¼ ð2Þ
½AB ka
KD is represented in molar units (M).

Knowing the concentration of analyte in a sample is required
for calculating certain kinetic parameters. To calculate KD and ka,
the concentration of analyte must be known. By contrast, kd is
concentration-independent, and can therefore be calculated with-
out knowing how much analyte is present. The dissociation rate
constant, kd, is useful for ranking sets of analytes of unknown
concentration, such as unpurified proteins, based on off-rate.
1.5 Considerations Maintaining the structure and activity of the immobilized ligand is
for Designing a the most important consideration for biosensor selection. Direct
Successful Assay immobilization of a target protein to a biosensor can be accom-
plished by (1) covalent linkage to free lysine residues in the target
1.5.1 Choosing an protein via a carboxyl group on the biosensor or by (2) biotinylated
Immobilization Strategy ligand binding to streptavidin-coated biosensors. These methods
are compatible with most proteins and enable creation of “custom”
biosensor surfaces with virtually any protein. However, these
approaches require purified protein and creation of a covalent
bond, either directly to the surface or to biotin. Site-directed or
capture-based biosensors provide an alternative immobilization
strategy that can be used to maximize activity of the ligand on the
surface without covalent bonding. Capture biosensors are pre-
immobilized with a high affinity capture antibody or protein
which binds to the protein ligand via a known motif or tag,
enabling favorable orientation of the ligand and improved homo-
geneity on the surface. Because of the high specificity of these
interactions, ligand protein can be captured directly from crude
samples such as culture media without need for purification.
1.5.2 Assay Orientation The choice of which molecule in a binding pair to use as ligand or
analyte depends on several factors [6]. Primary factors include (1)
binding valency (use the lower valency molecule as analyte to
reduce avidity effects and simplify curve fitting; Fig. 3), (2) avail-
ability of purified protein to use as ligand, (3) stability of the protein
when immobilized (if unstable, use the other molecule as ligand)
and (4) sensitivity of detection (BLI signal is a function of molecu-
lar size and packing density, so use the larger molecule among the
binding pair as analyte to maximize detection sensitivity). Regard-
less of orientation or assay format, proper assay development is a
necessity for obtaining reliable kinetic data with any system.
Biosensor
Surface
Immobilized
Ligand
Bivalent
Analyte
Fig. 3 Avidity effects with multivalent molecules. When a multivalent molecule,

such as an antibody, is used as analyte in solution, there is potential for a single
analyte molecule to bind multiple immobilized ligand molecules causing avidity
effects
1.5.3 Assay Optimization The amount of ligand to immobilize needs to be optimized for each
assay and reagent lot. Typically, increasing the loading density will
lead to increased signal in the analyte association step. However, it
is a common observation that loading biosensors with very high
density of ligand may lead to artifacts and secondary binding
effects, such as analyte rebinding, that alter observed binding
rates [7]. Hence, a “scouting” experiment must be performed to
find the balance between maximizing sensitivity for analyte detec-
tion and minimizing secondary binding effects. As a rule, the lowest
concentration of immobilized ligand that yields an acceptable signal
in the analyte association step should be used.
Analyte concentration is another important consideration for
obtaining accurate kinetic and affinity constants. A dilution series of
at least four analyte concentrations should be measured in a kinetic
assay. The analyte dilution series ideally should range from a con-
centration of about tenfold higher to tenfold lower than the
expected affinity constant. For screening purposes or qualitative
analyses, a single concentration is often sufficient. If the approxi-
mate KD of the interaction is not known, performing an analyte
concentration scouting step is recommended. In this case, choose a
few concentrations that span a wide range to obtain an approxima-
tion of the affinity constant [8].
1.5.4 Nonspecific Biological molecules typically have many points of hydrophobicity

Binding and charge that can lead to nonspecific interactions with solid
surfaces. Nonspecific binding is commonly observed on micro-
plates and other sample containers, as well as the biosensor tips
[9]. Label-free assay technologies generate signal upon binding of
biomolecules, so nonspecific binding will contribute to signal. It is
thus important to optimize assay conditions to avoid such binding.
The biocompatible layer on BLI biosensors greatly mitigates
nonspecific binding; however, a reference sample must be included
with every experiment. The reference sample typically consists of
the solution medium used for the assay minus the analyte. Double
referencing with both a reference sample and a reference biosensor
can be performed when background signal due to nonspecific
binding is an issue, or in assays where the signal is very small in
relation to background. A reference biosensor should be loaded
with a non-active protein similar to the active ligand.
1.5.5 Biosensor In many cases, biosensors can be regenerated by removing

Regeneration bound analyte under conditions that do not irreversibly disrupt
the binding capacity of the ligand. Regeneration allows for reuse
of the same ligand-loaded biosensor to measure binding to multiple
analytes or various analyte concentrations [10, 11]. However, reuse

of biosensors does require that significant time, cost, and effort be
spent on optimizing the conditions of regeneration. While biosen-
sor regeneration is an option with BLI technology, using a fresh
biosensor for every sample is convenient and cost-effective. BLI
biosensors are 20–40 times less expensive than those used in other
techniques and are unique among label-free biosensors for enabling
single use.
1.6 BLI to Study an Here we describe how to perform kinetic characterization of the
Antibody–Antigen interaction between an antibody and antigen using the BLI-based
Interaction Octet analytical system. The reagents used in this example are
purified human prostate-specific antigen (PSA) and purified anti-
PSA monoclonal IgG. This is a previously characterized binding
pair of low nanomolar affinity [12]. Anti-PSA antibody is first
biotinylated for loading onto streptavidin biosensors. The PSA
antigen in solution then is bound to the antibody in an association
step, followed by a dissociation step in buffer. The kinetics of the
binding interaction are monitored in real time. Data processing and
curve fitting parameters are described.
2 Materials
2.1 Reagents 1. Biotin-PEG4-NHS reagent. NHS-PEG12-Biotin or Sulfo-

NHS-LC-LC-Biotin can alternatively be used (see Note 1).
2. Distilled water.
3. Phosphate buffered saline (1 PBS): 138 mM NaCl, 2.7 mM
KCl, 10 mM Na2HPO4, 2 mM KH2PO4, pH 7.4.
4. Purified mouse IgG monoclonal anti-PSA antibody (e.g., Fitz-
gerald Industries, clone M612166), 100 μg minimum at a
concentration of at least 100 μg/mL (see Note 2).
5. PSA purified protein antigen (can also be purchased from
Fitzgerald Industries).
6. Kinetics Buffer: (Pall ForteBio Part No. 18-5032) 1 PBS,
0.1 % BSA, 0.2 % Tween 20 (see Note 3).
2.2 Materials and 1. Desalting spin columns or dialysis membrane/cartridge.

Instrumentation 2. Streptavidin (SA) biosensors (Pall ForteBio, Part No. 18-5020).
3. 96-well flat-bottom polypropylene microplates, black
(see Note 4).
4. Pall ForteBio’s BLI-based Octet RED96, Octet RED384,
Octet QKe, or Octet QK384 instrument and software [13].
3 Methods
3.1 Biotinylation of 1. Pipet a minimum of 100 μg purified antibody into a microcen-

Anti-PSA Monoclonal trifuge tube.
Antibody Ligand [14] 2. Follow instructions from the manufacturer to prepare a fresh
concentrated stock of biotin-PEG4-NHS reagent. Dilute this
stock in distilled water to make a 1 mM solution of biotin
reagent (see Note 5).
3. Calculate the volume of 1 mM biotin reagent to add to your
antibody to achieve a 1:1 molar coupling ratio (MCR) of biotin
to protein (see Note 6).
4. Add the appropriate volume of biotin reagent as calculated in
step 3. Mix immediately.
5. Incubate the biotinylation mixture for 30 min at room
temperature.
6. Remove excess biotin reagent by desalting with a size-exclusion
spin column or by dialysis (see Notes 7 and 8).
7. For dialysis: Dialyze sample 1:1,000 in 1 PBS. Allow to stir
gently for a minimum of 3 h before changing PBS buffer.
Perform at least four exchanges of buffer before extracting
biotinylated protein.
3.2 Optimization of 1. Pipet 200 μl per well of 1 Kinetics buffer into wells of a 96-well
Antibody Loading black flat-bottom microplate corresponding to the number and
Concentration on position of biosensors to be used. The buffer used for hydration
Biosensors should be the same as that used throughout the assay.
3.2.1 Hydrate Biosensors
2. Insert the hydration plate into the biosensor tray. Align the
biosensor rack over the hydration plate and lower the biosen-
sors into the wells, taking care not to scrape or touch the tips of
the biosensors. Allow biosensors to hydrate for at least 10 min.
3.2.2 Prepare Ligand and 1. Dilute biotinylated anti-PSA antibody to 50 μg/mL in 1

Analyte Reagents and Kinetics Buffer. Equilibrate reagents and samples to room tem-
Assay Plate perature prior to sample preparation. For frozen samples, thaw
and mix thoroughly prior to use.
2. Perform dilutions of biotin–anti-PSA in 1 Kinetics buffer to
yield antibody concentrations of 50, 25, 10, 5, and 1 μg/mL
(see Notes 9 and 10).
3. Dilute PSA antigen to a concentration of 200 nM in 1
Kinetics buffer (see Note 11).
4. Pipet 200 μL of each biotin–anti-PSA dilution into Column 2,
Rows A–E of a black 96-well polypropylene microplate (see
Note 12).
5. Pipet 200 μL 1 Kinetics buffer into Column 2, Row F.
6. Pipet 200 μL diluted PSA antigen into Column 4, Rows A–F.
7. Pipet 200 μL 1 Kinetics buffer into Column 1, Rows A–F, and

Column 3, Rows A–F. The plate layout will be as shown in Fig. 4.
3.2.3 Load Sample Plate 1. Ensure that the Octet instrument lamp is warmed up for at least
and Biosensors onto Octet 40 min prior to starting the assay.
Instrument and Run Assay 2. Set the sample plate temperature to 30 C in the Octet software
(see Note 13).
3. Place the sample plate on the sample plate stage inside the Octet
system with well A1 toward the back right corner. Place the
biosensor hydration assembly on the biosensor stage. Ensure
that both the tray and the sample plate are securely in place.
4. Equilibrate the plates in the instrument for 10 min prior to
starting the experiment. The delay timer can be used to auto-
matically start the assay after 10 min (600 s).
5. Set up the assay in the instrument software. For details, refer to
the Octet Data Acquisition User Guide. Table 1 shows an
example of the settings for a kinetic assay, which can be used
for the ligand loading experiment. Steps consist of equilibra-
tion, ligand loading, baseline, association, and dissociation
steps (see Note 14).
Assay Step
1 2 3 4 5 6 7 8 9 10 11 12
A 50 Buffer
B 25 Biotin-anti-PSA (μg/mL)
C 10 PSA antigen
D 5
E 2.5
F
G
H
Fig. 4 Plate map diagram for loading optimization experiment
Table 1
Recommended settings for kinetic assay
Sample plate
Step# Step name Time (s) Flow (RPM) column
1 Equilibration 60 1,000 1
2 Loading 300–600 1,000 2
3 Baseline 180–600 1,000 3
4 Association 300–600 1,000 4
5 Dissociation 300–3,600 1,000 3
100 μg/mL
3.0 50 μg/mL
25 μg/mL
2.5
2.0
Binding, nm
1.5
10 μg/mL
1.0
5 μg/mL
0.5
2.5 μg/mL
0 0 μg/mL
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
Time (sec)
Fig. 5 Sensorgram output for ligand loading optimization experiment
6. Run the assay. An example of the sensorgram output for the ligand
loading optimization is shown in Fig. 5. The shape of individual
binding curves can be observed. Note that the initial slope of the
binding curve corresponds to concentration of ligand.
7. In Octet Data Analysis software, process the data so that the
association step is aligned to the baseline. In this view, the
relative signal of the analyte binding at each corresponding
ligand concentration can be clearly observed (Fig. 6). The
optimal concentration of biotin–anti-PSA for a binding kinetics
experiment can be selected based on these data. In this exam-
ple, the optimal loading concentration is estimated to be
25 μg/mL (see Note 15).
8. Check for nonspecific binding of analyte to the biosensor. The
non-ligand loaded sample should show a flat response even in
the presence of analyte.
3.3 Run Kinetics Hydrate biosensors as described in Subheading 3.2.1.

Experiment
3.3.1 Hydrate Biosensors
3.3.2 Prepare Assay 1. Equilibrate reagents and samples to room temperature and mix
Samples thoroughly.
2. Dilute the biotin–anti-PSA in 1 Kinetics Buffer to a working
concentration of 25 μg/mL, based on results from the ligand
loading experiment above.
0.5 25 μg/mL
50 μg/mL
100 μg/mL
0.4
Binding, nm
0.3
10 μg/mL
0.2
5 μg/mL
0.1 2.5 μg/mL
0 0 μg/mL
0 100 200 300 400 500 600
Time (sec)
Fig. 6 Aligned sensorgram traces showing association and dissociation steps
3. Dilute the PSA analyte sample to 300 nM in 1 Kinetics

Buffer. Perform serial threefold dilutions in 1 Kinetics Buffer
to generate 300 nM, 100 nM, 33 nM, 11 nM, and 3.7 nM
samples (see Note 16).
4. Transfer 200 μL of 1 Kinetics Buffer into Columns 1 and 3 of a
black polypropylene 96-well microplate (see Notes 17 and 18).
5. Pipet 200 μL of biotin–anti-PSA ligand into wells in Rows A–F,
Column 2.
6. Pipet 200 μL of each dilution of PSA antigen into Column 4,
Rows A–E.
7. Pipet 200 μL of 1 Kinetics Buffer into Column 4, Row F
(see Note 19). An example plate setup for a kinetic assay is
shown in Fig. 7.
3.3.3 Load Sample Plate 1. Ensure that the instrument lamp is warmed up for at least
and Biosensors onto Octet 40 min prior to starting the assay.
Instrument and Set Up 2. Set the sample plate temperature in the Octet software.
Kinetic Assay in Octet
3. Open instrument door and place the sample plate on the
Software
sample plate stage with well A1 toward the back right corner.
Place the biosensor hydration assembly on the biosensor stage.
Ensure that both the tray and the sample plate are securely in
place. Close the instrument door.
4. Equilibrate the plates in the instrument for 10 min prior to
starting the experiment. The delay timer can be used to auto-
matically start the assay after 10 min (600 s).
Assay Step
1 2 3 4 5 6 7 8 9 10 11 12
A 300 Buffer
B 100 Biotin-anti-PSA
C 33 PSA antigen (nM)
D 11
E 3.7
F
G
H
Fig. 7 Plate map diagram for PSA antibody–antigen kinetics
3.5
3.0
2.5
Binding, nm
2.0
1.5
1.0
0.5
0
0 200 400 600 800 1000 1200 1400 1600 1800
Time (sec)
Fig. 8 Raw sensorgram output for kinetic assay
5. Set up the assay in the instrument software. For details, refer to

the Octet Data Acquisition User Guide. Refer to Table 1 for an
example of the settings for the binding kinetics experiment
using the plate described in Fig. 7, consisting of equilibration,
ligand loading, baseline, association, and dissociation steps (see
Notes 20 and 21).
6. Run the binding assay (see Note 22).
3.4 Data Analysis 1. Load data into the Octet Data Analysis software. A sample of
the raw sensorgram output is shown in Fig. 8 (see Note 23).
2. Select parameters for data processing. Recommended para-
meters for a standard protein-protein interaction are:
(a) Y-axis alignment (select last 5–10 s)
0.7
300 nM
0.6
100 nM
0.5
Binding, nm
0.4
0.3
33 nM
0.2
0.1 11 nM
3.7 nM
0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
Time (sec)
Fig. 9 Processed data for anti-PSA–PSA binding, showing aligned association and dissociation steps
(b) Use inter-step correction, as long as baseline and associa-

tion steps were performed in the same well (see Note 24)
(c) Reference subtraction using Rows G and H as reference
samples
(d) Savitzky–Golay filtering
3. Process data. Processed data output is shown in Fig. 9, with
X and Y axes aligned.
4. Analyze data. In the analysis tab, select parameters for curve
fitting. Several curve fitting models are available (see Note 25),
to represent different types of binding interactions:
(a) 1:1 binding
(b) 2:1 heterogeneous ligand
(c) Mass transport
(d) 1:2 bivalent analyte
The anti-PSA–PSA interaction shows well-behaved single-
analyte to single-ligand molecule binding stoichiometry, so
select 1:1 under Model. Fitting should be global, with sensors
grouped by color, and Rmax unlinked by sensor (see Note 26).
Use entire step times. For details on analysis parameters, refer
to the Octet Data Analysis User Guide.
5. Fit curves (see Note 27). The ideal fit traces will appear on the
graph (Fig. 10), and data table will update with kinetic and
affinity constants, rates, errors, and statistics. For guidelines on
determining accuracy of fit, see Note 28.
0.7 300 nM
0.6
100 nM
0.5
Binding, nm
0.4
0.3
33 nM
0.2
0.1 11 nM
3.7 nM
0
0 200 400 600 800 1000 1200
Time (sec)
ka (1/Ms) kd (1/s) KD (M)
4.20E+04 8.29E-05 1.98E-09
Fig. 10 Analyzed data for anti-PSA–PSA binding using 1:1 binding model. Curve fit overlays are shown as thin
lines over traces
4 Notes
1. Biotin-PEG4-NHS is a biotinylation reagent that reacts specif-

ically with primary amine groups, such as the side chains of
lysine residues in a polypeptide. The PEG spacer arm confers
increased solubility and prevents aggregation of labeled mole-
cules. The spacer also provides a long and flexible linker to
minimize steric hindrance when bound via streptavidin to the
solid surface of the biosensor.
2. Ensure purified antibody is carrier free and is not in a buffer
containing primary amines, such as Tris or glycine. If the pro-
tein is suspended in a buffer containing primary amines, per-
form buffer exchange into 1 PBS either by dialysis or
desalting spin columns. It is recommended that the starting
concentration be at least 1 mg/mL if buffer exchange is
required.
3. Streptavidin biosensors are compatible with a wide range of
buffers. Kinetics Buffer is provided from ForteBio as a 10
stock to be diluted 1:10 with PBS, pH 7.4. Other buffers can
be used for the assay. Best results are obtained when all buffers/
matrices used in the assay are closely matched.
4. For Octet RED384 and QK384 instruments, a 384-well
microplate can be used for increased throughput and smaller
sample volumes (down to 40 μL).
5. Prepare the biotinylation reagent fresh and use immediately to

ensure efficient biotin incorporation. The NHS reagent is
moisture-sensitive and typically loses activity in 15–30 min in
aqueous solution.
6. We recommend using a 1:1 molar coupling ratio when bioti-
nylating proteins for immobilization on streptavidin biosen-
sors. Over-biotinylation does not improve biosensor loading
and has the potential to reduce protein activity.
7. Biotinylation reactions must be desalted to remove excess unin-
corporated biotin reagent, which will compete for binding sites
on the streptavidin surface. Desalting spin columns filled with
size-exclusion resin provide a convenient and effective method
for reaction buffer exchange and desalting. Dialysis into PBS
buffer with an appropriate molecular weight cutoff membrane
or cartridge can be used for gentle buffer exchange of more
sensitive proteins (100 kDa MW cutoff for antibodies).
8. When using desalting spin columns for buffer exchange, be sure
to use the appropriate molecular weight cutoff and select a
column size that is suited to the sample volume being applied.
This will prevent sample loss.
9. A dilution series of biotinylated ligand is recommended, in
general starting around 50 μg/mL and titrating down, to
optimize loading concentration. The higher concentrations
will be expected to quickly saturate the sensor, while low con-
centrations may require longer loading times to reach equilib-
rium. A typical immobilization concentration for a biotinylated
antibody lies between 5 and 25 μg/mL.
10. If the ligand concentration is low, e.g., below 5 μg/mL, a
longer loading time may be required for sufficient immobiliza-
tion signal. Overnight incubation in ligand solution may also
be performed at 4 C. Overnight incubation is beneficial in
cases with capture biosensors where a ligand molecule is being
captured from a dilute supernatant or cell culture sample and
can greatly improve results.
11. Bind the analyte at a high concentration during the ligand
optimization procedure. Use of a high analyte concentration
will enable selection of the lowest ligand concentration that
gives an acceptable signal during the association step. It is
useful to run a zero-ligand biosensor as a control in this step
to assess whether there is nonspecific binding of the analyte to
the biosensor.
12. In a standard 384-well plate, use 80–100 μL sample volume per
well. In a 384-well Tilted-Bottom plate (Pall ForteBio Part No.
18-5080), use 40 μL or more.
13. We recommend that assays be run at 30 C for optimal results.

Binding to the biosensor is sensitive to fluctuations in temper-
ature. By working at a few degrees above ambient, a consistent
temperature can be maintained through the course of the assay.
14. Octet software allows the user to extend a step or skip to the
next step, if more or less time is needed. This feature may be
used during assay development. When measuring binding
kinetics in a run involving multiple concentrations of analyte,
use of this feature is not recommended.
15. For best results in a kinetic binding experiment, select the
ligand concentration that does not saturate the biosensor, but
still provides a strong analyte signal. Ideally, for a 150 kDa
antibody, the signal in the loading step should reach about
1.0 nm after 5–10 min loading.
16. For accurate kinetic analysis, it is recommended to run a two-
fold or threefold dilution series of at least four concentrations
of the analyte. The selected dilutions should ideally range from
tenfold above to tenfold below the KD. If the KD is not known
to even an approximation, a scouting experiment to approxi-
mate the KD is recommended before performing final kinetic
analysis.
17. The same sample wells should be used for the baseline and
dissociation steps (column 4 in the example above). This will
enable use of the inter-step correction feature in the data
analysis software, which corrects for steps in data associated
with minor changes in microplate well artifacts that occur
between sample wells.
18. The buffer used for prehydration, baseline, and dissociation
must match the matrix of the analyte sample in the association
step.
19. The buffer-alone sample in the analyte column (Column 4,
Row F) will serve as a reference sample, or negative control,
for subtracting background signal.
20. Shaking speed and assay step lengths can be optimized,
depending upon the strength and speed of the interaction.
Increasing the shake speed will increase the sensitivity of the
assay, and is recommended for weaker binders or lower con-
centrations of reagent. The association step time can be
increased for a slow interaction or a weak binder, to enhance
binding signal. The dissociation step should be longer for
higher affinity binders. In general, step lengths of 1 min for
baseline, 5 min for association and 10 min for dissociation are
good starting points, however some optimization may be
required.
21. For a high affinity binding pair (KD < 1 nM), a dissociation
time of 30 min or more may be required. For accurate calcula-
tion of binding and affinity constants, at least 5 % of the
complex must dissociate. Do not run the assay longer than
3 h, as sample evaporation from the microplate wells may
begin to impact results.
22. When running the assay, be sure the baseline is stable before
proceeding to association step. If needed, use the Extend Step
function in the software to run the baseline for longer periods.
With biotin-streptavidin interactions, there should be minimal
signal drift.
23. Some variation in loading levels between individual streptavidin
biosensors during the loading step is normal. This variability
will not affect calculation of kinetic constants in a properly
designed experiment. As long as several concentrations of ana-
lyte are run and global curve fitting is performed, small differ-
ences in ligand loading level are compensated for.
24. Avoid using inter-step correction for binding pairs with very
fast association and dissociation rates.
25. “Model surfing”, or determining the type of interaction based
on the curve-fitting model that best fits your data, is not
recommended. Unless the interaction is known to be more
complex, use 1:1 binding model. If the interaction being stud-
ied is predicted to follow 1:1 binding, and the data do not fit
well, this indicates that further assay development is required.
26. Rmax may be linked if the same biosensor is used for every
sample concentration in the series. This selection is typically
made in small molecule analyses, where dissociation is rapid and
complete and allows for reuse of the biosensor in a new sample.
27. Note that steady state, or equilibrium analysis can also be
performed, but should be used only if the response at the
association step for each concentration has reached
equilibrium.
28. To determine the quality of the fit and accuracy of the calcu-
lated constants, consider the following general guidelines:
(a) Visually inspect the fit: do the fit lines conform well to the
data traces?
(b) Look at residuals, which are plotted below the data traces.
Residual values should not be greater than 1 % Rmax.
(c) The ka error and kd error values should be no greater than
one order of magnitude below the reported constant.
(d) R2 should be above 0.95 to indicate goodness of fit.
(e) Chi-squared value should be below 3.
References
1. Rich RL, Myszka DG (2010) Grading the kinetics assays. (2012). http://www.fortebio.
commercial optical biosensor literature - Class com/literature.html
of 2008: ‘The Mighty Binders’. J Mol Recognit 9. Karlsson R, Falt A (1997) Experimental
23:1–64 design for kinetic analysis of protein-
2. Rich RL, Myszka DG (2007) Higher- protein interactions with surface plasmon
throughput, label-free, real-time molecular resonance biosensors. J Immunol Methods
interaction analysis. Anal Biochem 361:1–6 200:121–133
3. Cooper MA (2006) Optical biosensors: where 10. Pall ForteBio Technical Note 8: regeneration
next and how soon? Drug Discov Today strategies for amine reactive biosensors on the
11:1061–1067 octet system. (2007). http://www.fortebio.
4. Concepcion J et al (2009) Label-free detection of com/literature.html
biomolecular interactions using BioLayer Inter- 11. Pall ForteBio Technical Note 14: regeneration
ferometry for kinetic characterization. Comb strategies for streptavidin biosensors on the
Chem High Throughput Screen 12:791–800 octet platform. (2009). http://www.fortebio.
5. Elwing H (1998) Protein absorption and ellip- com/literature.html
sometry in biomaterial research. Biomaterials 12. Katsamba P et al (2006) Kinetic analysis of a
19:397–406 high-affinity antibody/antigen interaction per-
6. Markey F (2009) Macromolecular interactions. formed by multiple Biacore users. Anal Bio-
In: Cooper MA (ed) Label-free biosensors. chem 352:208–211
Cambridge University Press, Cambridge, 13. More information may be found at www.
pp 143–158 fortebio.com
7. Myszka DG (1999) Improving biosensor anal- 14. Pall ForteBio Technical Note 28: biotinylation
ysis. J Mol Recognit 12:279–284 of protein for immobilization onto streptavidin
8. Pall ForteBio Application Note 8: optimizing biosensors. (2011). http://www.fortebio.
protein-protein and protein-small molecule com/literature.html
Chapter 11
Characterization of Protein-Protein Interactions

by Isothermal Titration Calorimetry
Adrian Velazquez-Campoy, Stephanie A. Leavitt, and Ernesto Freire
Abstract
The analysis of protein-protein interactions has attracted the attention of many researchers from both a
fundamental point of view and a practical point of view. From a fundamental point of view, the development
of an understanding of the signaling events triggered by the interaction of two or more proteins provides
key information to elucidate the functioning of many cell processes. From a practical point of view,
understanding protein-protein interactions at a quantitative level provides the foundation for the develop-
ment of antagonists or agonists of those interactions. Isothermal Titration Calorimetry (ITC) is the only
technique with the capability of measuring not only binding affinity but the enthalpic and entropic
components that define affinity. Over the years, isothermal titration calorimeters have evolved in sensitivity
and accuracy. Today, TA Instruments and MicroCal market instruments with the performance required to
evaluate protein-protein interactions. In this methods paper, we describe general procedures to analyze
heterodimeric (porcine pancreatic trypsin binding to soybean trypsin inhibitor) and homodimeric (bovine
pancreatic α-chymotrypsin) protein associations by ITC.
Key words Protein-protein interaction, Thermodynamics, Calorimetry, Titration, Binding, Dimer-

ization, Dissociation
1 Introduction
1.1 Protein-Protein Protein-protein interactions play a critical role in biological signaling.

Interactions Many pathological conditions including cancer, inflammation, auto-
immune diseases, diabetes, osteoporosis, infection, etc. are asso-
ciated with specific protein-protein interactions and consequently
have defined targets for drug development. The number of targets of
interest is continuously increasing and range from a vast number of
cell surface receptors, such as EGFR, TNFR, and IGFR, to other
proteins involved in signaling and regulation [1, 2]. Biologics, i.e.,
monoclonal antibodies or recombinant versions of ligand proteins
and/or soluble regions of the receptors, define the therapeutic
arsenal aimed at targeting those interactions. In fact, biologics have
become the fastest growing segment of the pharmaceutical industry.
183
184 Adrian Velazquez-Campoy et al.
The development of new biologics requires a precise characteriza-

tion of the interaction of the target proteins as well as the biologic
itself with the selected target.
A survey of protein-protein interactions indicates that their
binding affinities span a wide range, from the micromolar to the
high picomolar level [3–5], and references therein). In particular,
protein ligand–receptor interactions may bind with affinities in the
nanomolar and high picomolar level [6–16]. Examples of sub-
nanomolar interactions are the binding of IL-4 and erythropoietin
to their respective receptors with Kd values of 0.2 nM [8, 10]. The
binding affinity is not the only value that shows significant varia-
tion. The enthalpy and entropy changes associated with binding
also exhibit a significant spread, reflecting the magnitude and
nature of the conformational changes coupled to binding. For
example, the binding of proteins characterized by intrinsically dis-
ordered domains is usually associated with large favorable binding
enthalpies and equally large unfavorable binding entropies, as the
binding process results in the folding or structuring of the disor-
dered domains [17–19].
ITC provides a unique opportunity to measure ΔG, ΔH, and
ΔS simultaneously and, therefore, to develop a complete character-
ization of the binding process [20–23]. In addition, performing
experiments at different temperatures provides access to the change
in heat capacity, ΔCP . The experimental guidelines presented below
can be extended to most situations.
1.2 ITC of Protein- Two situations are possible when characterizing intermacromole-
Protein Interactions cular interactions in binding reactions: the binding partners are (a)
different (heterodimeric complex) or (b) identical (homodimeric
complex). Even though experimentally they require different meth-
odologies, the underlying principles are the same in both cases, that
is, they follow the same chemical scheme based on a reversible
association equilibrium:
M1 þ M2 ↔M1 M2 ð1Þ
where M1 and M2 are the interacting macromolecules. The

strength of the interaction is described by the association constant,
Ka, or the dissociation constant, Kd:
½M 1 M2 1
Ka ¼ ¼ ð2Þ
½M1 ½M2 K d
where [M1] and [M2] are the concentrations of the free reactants
and [M1M2] is the concentration of the complex. These constants
are related to the Gibbs energy of association, ΔGa, and
Isothermal Titration Calorimetry 185
dissociation, ΔGd, and can be expressed in terms of the enthalpy,

ΔH, and entropy, ΔS, changes in the process:
ΔG a ¼ RT ln K a ¼ ΔH a T ΔS a
ð3Þ
ΔG d ¼ RT ln K d ¼ ΔH d T ΔS d
where R is the gas constant (1.9872 cal/K mol) and T is the

absolute temperature (kelvins).
Both enthalpic and entropic contributions to the Gibbs energy
reflect different types of interactions underlying the overall process.
Accordingly, complexes predominantly stabilized enthalpically or
entropically will be preferentially stabilized by specific (hydrogen
bonds, electrostatics, van der Waals) or unspecific (hydrophobicity)
interactions and will respond differently to environmental changes
or mutations in the binding species. A number of reports have
shown the importance and consequences of the distribution of
the binding affinity into the enthalpic and entropic contributions
regarding the identification of binding mechanisms, the optimiza-
tion of ligand affinity and selectivity, the minimization of ligand
susceptibility to target mutations in drug design, as well as the
assessment of conformational changes coupled to binding and
protein activation and signaling in drug design [17–19, 24–40].
1.2.1 Heterodimeric Typical ITC experiments allow for the characterization of the inter-
Interactions action between two different binding partners. A solution of one of
the reactants is placed in a syringe and a solution with the other
interacting macromolecule is located in a calorimetric cell (Fig. 1).
The stepwise addition of the macromolecule from the injection
syringe solution triggers the binding reaction, leading the system
through a sequence of equilibrium states, the composition of each
one being dictated by the association constant. Given the total
concentrations of reactants in the calorimetric cell, [M1]T and
[M2]T, the association constant, Ka, determines the partition
between the different chemical species:
½M1 M2
Ka ¼ ð4Þ
½M1 ½M2
ITC directly measures the heat, qi, associated with each change
of state after each injection, which is proportional to the increment
in the concentration of complex in the calorimetric cell after the
injection i:

q i ¼ V ΔH a ½M1 M2 i ½M1 M2 i1 ð5Þ
where ΔHa is the enthalpy of binding and V is the calorimetric

cell volume. The sequence of injections proceeds until no signifi-
cant heat is detected, that is, the macromolecule in the cell is
saturated and the concentration of complex reaches its maximum.
a b time (min)
0 30 60 90 120
0.0
dQ/dt (mcal/s)
-1.0
-2.0
-3.0
c
0.0
Q (kcal/mol of injectant)
-2.0
-4.0
DH Ka
-6.0
-8.0 n
-10.0
0.0 1.0 2.0 3.0
[M1]T/[M2]T
Fig. 1 (a) Illustration of the configuration of an ITC reaction cell. The cell volume
is 1.4 mL and initially is filled with the macromolecule solution (gray). The
injection syringe is filled with the ligand solution (black). At specified time
intervals (400 s), a small volume (10 μL) of the ligand solution is injected into
the cell triggering the binding reaction and producing the characteristic peak
sequence in the recorded signal (b). After saturating the macromolecule, the
residual heat effects (the so-called dilution peaks), if any, are due to mechanical
and dilution phenomena. After integration of the area under each peak (and
subtraction of the dilution heat effects and normalization per mol of injected
ligand) the individual heats are plotted against the molar ratio (Panel c) from
which, through nonlinear regression, it is possible to estimate the thermody-
namic parameters n, Ka, and ΔH
Throughout the experiment the total concentrations of reactants,

[M1]T and [M2]T, are the known independent variables. Nonlinear
regression analysis of qi, the dependent variable, allows estimation
of the thermodynamic parameters (Ka and ΔHa and, therefore,
ΔSa) (see Subheading 3.3).
1.2.2 Homodimeric The self-association of a protein leading to the formation of homo-

Interactions dimers has been scarcely studied by ITC. From a practical point of
view, it is impossible to isolate individual partners and perform a
standard mixing assay. However, the strength of the interaction can
be measured in dilution experiments [41–43]. A solution of reac-
tant is placed in the syringe and a buffer solution is located in the
calorimetric cell. The stepwise addition of the solution in the

injection syringe, with the subsequent dilution of the macromole-
cule, triggers the dissociation reaction, leading the system through
a sequence of equilibrium states, the composition of each one being
dictated by the dissociation constant, Kd. Given the total macro-
molecule concentration in the calorimetric cell, [M]T, the dissocia-
tion constant determines the partition between the different
chemical species, monomer [M] and dimer [M2]:
½M2
Kd ¼ ð6Þ
½M2
Again, the heat, qi, associated with each injection is propor-

tional to the increment in the concentration of monomer, [M], in
the calorimetric cell after the injection i:

v
q i ¼ V ΔH d ½Mi ½Mi1 F 0 ½M0 ð7Þ
V
where ΔHd is the enthalpy of dissociation (per monomer), V is the

calorimetric cell volume, v is the injection volume, [M]0 is the total
concentration of macromolecule (per monomer) in the syringe,
and F0 is the fraction of monomer in the concentrated solution
placed in the syringe. The last term in the parenthesis is a correction
that accounts for the increment of monomer concentration in the
cell due to the injection of monomers from the syringe and, there-
fore, not contributing to the heat. As the protein concentration in
the calorimetric cell progressively increases, the dissociation process
is less favored and the sequence of injections proceeds until no
significant heat is detected. Throughout the experiment the total
concentration of reactant, [M]T, is the known independent vari-
able. Nonlinear regression analysis of qi, the dependent variable,
allows the estimation of the thermodynamic parameters (Kd and
ΔHd and, therefore, ΔSd) (see Subheading 3.3).
1.3 Information Every equilibrium binding technique requires the reactant concen-
Available by ITC and trations to be in an appropriate range in order to obtain reliable
Experimental Design estimations of the association constant. A practical rule of thumb
for ITC is given by the parameter c ¼ Ka [M2]T ¼ [M2]T/Kd,
1.3.1 Simultaneous which must lie between 0.1 and 1,000, thus, imposing a limit to the
Determination of the lowest and largest association constant measurable at a given mac-
Association Constant, romolecule concentration [44]. This phenomenon is illustrated in
the Enthalpy of Binding Fig. 2. As the c value increases, the transition from low to high total
titrant concentration is more abrupt. In the case of very high
association constants (macromolecule concentration much higher
than the dissociation constant), all the titrant added in any injection
will bind to the macromolecule until saturation occurs and, there-
fore, all the peaks, except the last ones after saturation, exhibit the
same heat effect. For low association constants (macromolecule
time (min)
0 30 60 90 120 0 30 60 90 120 0 30 60 90 120
3.0 a b c
dQ/dt (mcal/s)
2.0
1.0
0.0
10.0
Ka = 104 M-1 Ka = 106 M-1 Ka = 108 M-1
8.0
Ka[M]T = 70 Ka[M]T = 7000
Ka[M]T= 0.7
6.0
4.0
2.0
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
[M1]T/[M2]T
Fig. 2 Heterodimer formation. Illustration of the effect of the association constant value on the shape of a
titration curve. The plots represent three titrations simulated using the same parameters (concentrations of
reactants and association enthalpy), but different association constants. Low (a), moderate (b), and high
affinity (c) binding processes are shown. In order to obtain accurate estimates of the association constant, an
intermediate case is desirable (1 < c ¼ [M2]T Ka < 1,000)
concentration far below the dissociation constant), from the very

first injection only a fraction of the titrant added will bind, produc-
ing a less steep titration in which saturation is hardly reached and
often lacking the inflection point. In order to obtain accurate
estimates of the association constant, an intermediate case is
desirable.
To obtain a satisfactory titration curve, the concentration of
titrant should be enough to exceed the stoichiometric binding after
completion of the injection sequence (i.e., for a cell volume and an
injection volume of ~1.5 mL and ~10 μL, respectively, the reactant
concentration in the syringe should be 10–20 times the concentra-
tion of macromolecule in the cell). In case of poor solubility, the
reactant with lower solubility should be placed in the cell.
The constraints dictated by the parameter c impose an experi-
mental limitation, i.e., for very high association constants optimal
concentrations are too small to be practical (low signal-to-noise
ratio), and for very low association constants the concentrations
may be prohibitively high (possibility of aggregation or economic
consideration). While very low c values allow determining the
association constant and the binding enthalpy as long as the con-
centrations of reactants is precisely known, very high c values only
allow determining the binding enthalpy.
Sometimes it is possible to change the experimental conditions

(temperature or pH, without compromising stability against aggre-
gation or unfolding) in order to modify the association constant
toward accessible experimental values [45–47]. To extrapolate to
the original conditions appropriate equations will need to be
applied.
However, there is an extension of the ITC protocol aimed to
overcome such drawbacks. Without changing experimental condi-
tions at all, displacement experiments implemented in ITC extend
the range for the association constant determination [27, 48–50].
Basically, a displacement experiment consists of a titration of the
high-affinity ligand into a solution of the macromolecule prebound
to a weaker ligand, therefore decreasing the apparent affinity of the
potent ligand. The thermodynamic parameters for the binding of
the high-affinity ligand are calculated from the apparent binding
parameters of the displacement titration and the known binding
parameters for the weak ligand. This approach can be used to
measure extremely high affinity binding processes, as well as very
low affinity binding reactions [51, 52].
1.3.2 Simultaneous For self-associating systems that can be measured by ITC dissocia-
Determination of the tion experiments, another dimensionless parameter can be used to
Dissociation Constant and define a practical limit for the accurate measurement of the dissoci-
the Enthalpy of Dissociation ation constant for a dimeric system. Now, c ¼ [M]T/Kd, which
must lie between 10 and 10,000, imposes a limit to the lowest and
largest dissociation constant measurable at a given macromolecule
concentration. This phenomenon is illustrated in Fig. 3.
As the c value increases, the transition from low to high total
protein concentration in the cell is more abrupt. In the case of very
low dissociation constants (macromolecule concentration much
higher than the dissociation constant) most of the macromolecule
in the syringe is forming dimers and, when diluted in the cell, the
strength of the interactions determines that only a small fraction of
the dimers will dissociate. For that reason the observed heats are
very small. For very high dissociation constants (macromolecule
concentration far below the dissociation constant), the weak inter-
action within the dimer makes it possible for a large fraction of the
dimers to dissociate upon dilution; however, because the mono-
mer–monomer interaction is weak, the population of dimers in the
syringe will be small. In order to obtain accurate estimates of the
dissociation constant, an intermediate case is desirable: the curva-
ture of the plot should be enough for a reliable estimation of the
dissociation constant and the size of the peaks should give an
acceptable signal-to-noise ratio.
1.3.3 Determination of Binding enthalpy is usually determined from titrations in the opti-
the Enthalpy of Binding mal range of reactant concentrations. However, a greater accuracy
can be achieved if enthalpy is measured by performing injections of
time (min)
0 30 60 90 120 0 30 60 90 120 0 30 60 90 120
3.0 a b c
dQ/dt (mcal/s)
2.0
1.0
0.0
10.0 Kd = 10-5 M Kd = 10-7 M

8.0
[M]T/Kd= 100 [M]T/Kd= 10000
6.0
4.0 Kd = 10-3 M
2.0
[M]T/Kd= 1
0.0
0 40 80 120 160 0 40 80 120 160 0 40 80 120 160
[M]T (mM monomer)
Fig. 3 Homodimer dissociation. Illustration of the effect of the dissociation constant value on the shape of a
dissociation curve. The plots represent three titrations simulated using the same parameters (concentration of
reactant and dissociation enthalpy), but different dissociation constants. Low (a), moderate (b), and high (c)
dissociation constant processes are shown. In order to obtain accurate estimates of the association constant,
an intermediate case is desirable (10 < c ¼ [M]T/Kd < 10,000)
titrant into a large excess of macromolecule (far above the dissocia-

tion constant). Therefore, although there will be no saturation, all
the titrant injected will be fully associated giving a good measure-
ment of the binding reaction heat. Blank experiments (see below)
must be considered in order to eliminate the contribution of differ-
ent phenomena (dilution of reactant, solution mixing, and other
nonspecific effects) occurring simultaneously during the injections
and to obtain the heat effect associated with the binding process.
1.3.4 Blank Experiment In order to estimate the heat effect associated with dilution, mix-
ing, and other phenomena different from binding, blank experi-
ments can be performed. In this case, the experiment is similar to
the one described in Subheading 1.3.3: titrant is injected into a
buffer solution without macromolecule. However, this may be not
the ideal way to estimate such heat effects, since the degree of
solvation of the titrant and the chemical composition of the cell
solution would be different compared to the real titration with the
other interacting macromolecule present. For that reason, some
researchers usually consider the average effect of the last injection
peaks as the reference heat effect. If during the experiment it is not
possible to reach complete saturation, precluding such averaging, it
is possible to include in the fitting function a term accounting for
such dilution heat effects. Blank experiments, also called “heat

of dilution experiments,” are required to measure binding
enthalpies under excess macromolecule concentration, as men-
tioned previously.
1.3.5 Determination The temperature derivative (at constant pressure) of the enthalpy,
of Heat Capacity Change i.e., the heat capacity change in the process (association or dissocia-
of Binding tion), ΔCP, is defined as
∂ΔH

ΔC P ¼ ð8Þ
∂T P
and can be determined by performing the same experiment at
several temperatures. The slope of the enthalpy vs. temperature
plot gives the heat capacity of the reaction.
ΔCP has been shown to originate from surface desolvation upon
binding or solvation upon dissociation and, to a less extent, from the
difference in vibrational modes between the complex and the free
species [53, 54]. It provides information about the nature of the
interactions driving the binding, in addition to the main thermody-
namic functions (Gibbs energy, enthalpy, and entropy of binding).
1.3.6 Determination Strong dependency of the thermodynamic parameters of the asso-

of Coupled Proton ciation (dissociation) process on pH is an indication of proton
Transfer Process exchange between the binary complex and the bulk solution upon
complex formation (dissociation). This is due to changes in the pKa
of some ionizable groups located in any of the binding partners: the
microenvironment of these groups is altered upon binding or dis-
sociation, and so do their pKa and their proton saturation fraction.
ITC is one of the more suitable techniques for the assessment
of protonation/deprotonation processes coupled to binding.
When a binding process is coupled to proton transfer between the
bulk solution and the bound complex, the enthalpy of binding will
depend on the ionization enthalpy of the buffer molecule, ΔHion:
ΔH ¼ ΔH 0 þ nH ΔH ion ð9Þ
where ΔH0 is the buffer-independent enthalpy of binding and nH is

the number of protons being exchanged. Therefore, by repeating
the titration under the same conditions, but using several buffers
with different ionization enthalpies it is possible to estimate nH
(slope) and ΔH0 (intercept with the y-axis) from linear regression
analysis [55–57]. If nH is zero, there is no net proton transfer; if nH
is positive, there is a protonation, i.e., a proton transfer from the
solution to the complex; if nH is negative, there is a deprotonation.
After determination of these two parameters, nH and ΔH0, linkage
equations can be used to couple the binding or dissociation process
to the proton transfer process and allow the estimation of the
thermodynamic parameters for the proton exchange event

(enthalpy of ionization and pKa for each ionizable group involved)
[46, 58, 59]. Analogous linkage equations couple the binding or
dissociation reaction to the transfer of other ions (e.g., metals or
salts [60, 61]).
2 Materials
2.1 Reagents 1. Porcine pancreatic trypsin.

and Supplies 2. Soybean trypsin inhibitor.
3. Bovine pancreatic α-chymotrypsin.
4. Dialysis membrane of 10,000 molecular weight cutoff
(MWCO).
5. Filters (0.22 μm) with low protein binding properties.
6. Pyrex tubes for ITC syringe: 6 50 mm.
7. Potassium acetate–calcium chloride buffer: 25 mM potassium
acetate, pH 4.5, 10 mM calcium chloride.
8. Sodium acetate–sodium chloride buffer: 20 mM sodium ace-
tate, pH 3.9, 180 mM sodium chloride.
2.2 Isothermal Modern ITC instruments are simple to use, compact, and
Titration Calorimeter computer-controlled. The calorimeter unit consists of two cells,
the reference cell and the sample cell embedded in an adiabatic
chamber. The system holds the reference cell at a constant temper-
ature. Initially, constant power is applied to the sample cell in order
to activate a feedback control mechanism whose purpose is to
maintain the temperature difference between the two cells as close
to zero as possible. As the reaction, initiated with the injection of
titrant, occurs in the sample cell, the system adjusts the power
applied to the sample cell up or down depending on whether an
endothermic or exothermic reaction is taking place. This power is
recorded by the computer and corresponds to the signal observed
in the characteristic form of a peak. The area under each peak
corresponds to the heat released or absorbed during the reaction
after each injection.
2.3 Biological Two different systems, but functionally and structurally closely
Systems related, have been selected to illustrate the two main types of
association reactions that can be studied by ITC: formation of
heterocomplexes (trypsin–trypsin inhibitor) and homocomplexes
(chymotrypsin dimer). These complexes have been characterized
enzymatically, energetically, and structurally.
1. Porcine pancreatic trypsin, PPT, (EC. 3.4.21.4) is a 23.8 kDa

protein consisting of a single polypeptide chain with six disul-
fide bonds. It is a serine protease that hydrolyzes peptide bonds
with basic side chains (Arg or Lys) on the carboxyl end of the
bond.
2. Soybean trypsin inhibitor, STI, is a 20.0 kDa protein consisting
of a single polypeptide chain with two disulfide bonds. It
inhibits competitively the peptidase activity of trypsin.
3. Bovine pancreatic α-chymotrypsin, BP-α-CT, (EC. 3.4.21.1) is
a 25.2 kDa protein consisting of three polypeptide chains
interconnected and linked by two of the existing five disulfide
bonds. It is a serine protease that hydrolyzes peptide bonds
with aromatic or large hydrophobic side chains (Tyr, Trp, Phe,
Met, Leu) on the carboxyl end of the bond. This enzyme
exhibits a monomer–dimer equilibrium in solution.
The experiments were performed under slightly acidic condi-
tions to minimize the autocatalysis of both enzymes. In addition, at
high pH (approx 8) the interaction between trypsin and its inhibi-
tor is so strong that the affinity is above the practical limits
(Ka ~ 1010–1011 M1).
3 Methods
3.1 Sample 1. PPT: Prepare PPT at 400 μM in potassium acetate–calcium

Preparation chloride buffer (pH 4.5) dissolving the protein in buffer
(approx 1 mL) and dialyze overnight at 4 C against 4 L of
the same buffer. Filter the solution after dialysis. Measure
concentration after dilution (1:20) in the same buffer using
an extinction coefficient of 35,700 M1 cm1 at 280 nm.
2. STI: Prepare STI at 30 μM in potassium acetate–calcium chlo-
ride buffer (pH 4.5) dissolving the protein in buffer (approx
5 mL) and dialyze overnight at 4 C against 4 L of the same
buffer. Filter the solution after dialysis. Measure concentration
without dilution using an extinction coefficient of
18,200 M1 cm1 at 280 nm.
3. BP-α-CT: Prepare BP-α-CT at 200 μM in sodium acetate–sodium
chloride buffer (pH 3.9) dissolving the protein in buffer (approx
5 mL) and dialyze overnight at 4 C against 4 L of the same buffer.
Filter the solution after dialysis. Immediately prior to the experi-
ment concentrate the sample to about 1 mM. Measure concen-
tration after dilution (1:50) in the same buffer using an extinction
coefficient of 50,652 M1 cm1 at 280 nm.
3.2 Experimental This general protocol was designed for most experiments. Depend-
Procedure ing on the calorimeter or the particular biological system, the
protocol may require some adjustments.
What follows is a step-by-step protocol for running an ITC
experiment to study heterodimer formation. When studying
homodimer dissociation the only differences are in step 5 (cell is
filled with buffer solution only), step 6 (syringe is filled with BP-α-
CT solution) and step 12 (data analysis (see Subheading 3.3)).
1. For typical experiments, allow the equipment to equilibrate one
degree below the experimental temperature.
2. Prepare 2.2 mL of STI solution, 0.5 mL of PPT (in a glass tube,
6 50 mm) and 10 mL of buffer solution. All solutions should
be degassed for 10 min with a vacuum pump (see Note 1).
3. Meanwhile, thoroughly clean the calorimetric reaction cell. The
cell can be washed with a 5 % Contrad 70™ solution or 0.5 M
NaOH if necessary, and rinsed thoroughly with water.
4. Fill the reference cell with water. Usually, water should be
replaced every week.
5. Rinse the reaction cell with buffer. Slowly load the STI solution
into the reaction cell, and carefully remove bubbles. The con-
centration of the sample should be determined again after
loading because some dilution can take place due to residual
buffer in the cell.
6. Fill the 250-μL injection syringe with PPT solution (see Note
2). Rinse the syringe tip with buffer or water and dry.
7. Carefully insert the injection syringe into the reaction cell.
Avoid bending the needle or touching any surface with the
needle tip.
8. Equilibrate the calorimeter at the experimental temperature.
9. Set the running parameters for the experiment: number of
injections (28), temperature (e.g., 25 C), reference power
(10 μcal/s), initial delay (180 s), concentration in syringe
(approx 300 μM), concentration in cell (approx 20 μM), stir-
ring speed (490 rpm), file name (*.itc), feedback mode (high),
equilibration options (fast), comments, injection volume (first
injection 3 μL, the rest 10 μL), duration (automatically set
according to the injection volume), spacing between injections
(400 s), filter (2 s) (see Notes 3–7).
10. Start the experiment (see Note 8). There will be thermal
and mechanical equilibration stages. Initiate the injection
sequence after a stable no-drift noise-free baseline (as seen in
the 1 μcal/s scale) (see Note 9).
11. At the end of the experiment the system should be cleaned
thoroughly with water and the syringe rinsed and dried.
time (min) time (min)

0 30 60 90 120 150 180 210 0 30 60 90 120 150 180 210
1.0 1.0
dQ/dt (mcal/s)
dQ/dt (mcal/s)
25°C 30°C
0.5 0.5
0.0 0.0
10.0 10.0
8.0 8.0
6.0 6.0
4.0 4.0
2.0 2.0
0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
[PPT]T/[STI]T [PPT]T/[STI]T
Fig. 4 Titrations of STI with PPT. The experiments were performed in 25 mM potassium acetate, pH 4.25,
10 mM calcium chloride, at 25 C (left) and 30 C (right). The concentrations of reactants are 21 μM STI (in
cell) and 312 μM PPT (in syringe). The inhibitor was placed in the calorimetric cell due to its low solubility. The
solid lines correspond to theoretical curves with n ¼ 1.26, Ka ¼ 1.5 106 M1, and ΔHa ¼ 8.4 kcal/mol
(left) and n ¼ 1.28, Ka ¼ 2.2 106 M1, and ΔHa ¼ 7.6 kcal/mol (right)
12. Use the software provided by the calorimeter manufacturer for

data analysis according to the equations described previously
(see Notes 10 and 11). Results should be consistent with the
information shown in Fig. 4.
3.3 Data Analysis Equilibrium equations are very simple when using free concentra-
tions of reactants (Eqs. 4 and 6). However, they become more
complex when expressed in terms of total concentrations of reac-
tants, the known independent variables in ITC. Data processing
through transformation of the experimental data in order to linear-
ize the equilibrium equations must be avoided, since it often pro-
vokes a systematic propagation of errors and uneven distribution of
statistical weights over the binding curve. Nonlinear fitting proce-
dures should be implemented to directly analyze the raw calorimet-
ric data.
The analysis should be performed with the individual or differ-
ence heat plot, not with the cumulative or total heat plot. The
analysis in terms of the individual heat (heat associated with each
injection) is more convenient since it eliminates error propagation
associated with cumulative data. See Notes 12–16 for additional
considerations.
3.3.1 Heterodimeric The heat, qi, released or absorbed during injection i is proportional
Interaction to the change in concentration of binary complex after that injec-
tion and is given by Eq. 5. The 1:1 stoichiometric model permits an
explicit analytical solution for the concentration of complex (1, 13):
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
1 þ ½M2 T, i K a þ K a ½M1 T, i 1 þ ½M2 T, i K a þ K a ½M1 T, i 4½M2 T, i K 2a ½M1 T, i
½M1 M2 i ¼
2K a
ð10Þ
where [M1]T,i and [M2]T,i are the total concentrations of macro-
molecule M1 and M2 in the cell after the injection i.
Even when assuming a model corresponding to 1:1 stoichiom-
etry, it is useful to introduce a parameter n representing the number
(or fraction) of binding sites, as in the general 1:n model. When
estimating the parameters, n should be equal to unity, within the
experimental error, otherwise the following statements would hold:
n > 1: There is an error in the determination of the reactants
concentration (actual titrating molecule M1 concentration is
lower and/or actual titrated macromolecule M2 concentration
is higher).
There is more than one binding site (specific binding or
not) per macromolecule M2.
n < 1: There is an error in the determination of the reactants
concentration (actual titrating macromolecule M1 concentra-
tion is higher and/or actual titrated macromolecule M2 con-
centration is lower).
There is less than one binding site per macromolecule M2, that
is, the sample is not chemically or conformationally homogeneous,
or there is more than one binding site (specific binding or not) per
macromolecule M1.
The 1:n stoichiometric model (n identical and independent
binding sites) also permits an explicit analytical solution for the
concentration of complex (1):
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
1 þ n½M2 T, i K a þ K a ½M1 T, i 1 þ n½M2 T, i K a þ K a ½M1 T, i 4n½M2 T, i K 2a ½M1 T, i
½M1 M2 i ¼
2K a
ð11Þ
Therefore, knowing the total concentrations [M1]T and [M2]T

in the cell after the injections i and i 1 it is possible to evaluate the
heat qi according to Eq. 5.
The dilution effect on the concentrations, due to the addition
of the syringe solution, must be considered. The experiment pro-
ceeds at constant cell volume, i.e., when injecting a certain volume
v, the same volume of liquid is expelled from the cell. Therefore, the
total concentrations of reactants after the injection i are given by:
i
½M2 T, i ¼ ½M2 0 1 Vv

i ð12Þ
½M1 T, i ¼ ½M1 0 1 1 Vv

where [M2]0 is the initial concentration of macromolecule M2 in

the cell, [M1]0 is the concentration of macromolecule M1 in the
syringe, v is the injection volume, and V is the cell volume. Because
the dilution of reactants lowers the effective concentration of
macromolecules in the cell, it will also affect the heat signal
measured in each injection. Therefore, a correction to Eq. 5 is used:

v
q i ¼ V ΔH a ½M1 M2 i ½M1 M2 i1 1 ð13Þ
V
Nonlinear fitting of qi as a function of [M1]T,i or [M1]T,i/[M2]T,i

(the so-called molar ratio) provides n, Ka, and ΔHa as adjustable
parameters. It is possible to avoid performing blank experiments to
estimate the dilution heat effect by including in Eq. 13 a constant
(but floating in the fitting procedure) term, qd, representing such
contribution.
In general, models with different sets of binding sites require a
numerical approach in which the model parameters (nj, Ka,j, and
ΔHj, where j stands for each set of binding sites) are determined
iteratively by numerical solution of the binding equations.
For the experiment of porcine pancreatic trypsin and soybean
trypsin inhibitor, the data obtained at two temperatures were ana-
lyzed using a model that allows the heat of dilution to be fitted
simultaneously (Fig. 4). The enthalpy was plotted against tempera-
ture and the Gibbs energy dependence on temperature is shown in
Fig. 5.
3.3.2 Homodimeric The heat, qi, released or absorbed during injection i is proportional
Interaction to the change in concentration of monomer after that injection and
is given by Eq. 7. As already stated, the third term in the parenthesis
accounts for a contribution to the increase in monomer concentra-
tion in the cell that does not contribute to the heat measured.
The monomer–dimer equilibrium model permits an explicit
analytical solution for the monomer concentration (12):
0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
K d@ 8 ½ M T, i
½M i ¼ 1þ 1A ð14Þ
4 Kd
where [M]T,i is the total concentration of macromolecule M (per

monomer) in the cell after the injection i. Therefore, knowing the
total concentration [M]T in the cell after the injections i and i 1 it
is possible to evaluate the heat qi according to Eq. 7.
8.4 -7.2
3x106
DHa (kcal/mol)
DGa (kcal/mol)
8.2 -7.6
Slope = DCp
Ka(M-1)
6
2x10
8.0 -8.0
7.8 1x106 -8.4
7.6 -8.8
0
25 26 27 28 29 30 0 20 40
T (°C) T (°C)
Fig. 5 (Left panel) Estimation of the change of heat capacity upon association. Once the enthalpy of
association is determined at different temperatures (see Fig. 4), these values can be plotted vs. temperature.
The slope of this plot corresponds to the heat capacity change. In the case of the PPT/STI interaction, the heat
capacity was determined to be 160 cal/K mol. This negative value reflects the desolvation of molecular
surfaces upon binding. (Right panel) Outline of the dependency of the association constant and the Gibbs
energy of association on the temperature according to Eq. 3 and ΔGa(T) ¼ ΔHa(T0) + ΔCP,a(T T0)-T
[ΔSa(T0) + ΔCP,aln(T/T0)], where T0 is a given reference temperature at which the enthalpy and the affinity
have been simultaneously measured. The curves were calculated using the thermodynamic parameters
obtained at 25 C (dashed) and 30 C (continuous). The estimation of the heat capacity change allows the
determination of the thermodynamic parameters (affinity, Gibbs energy, enthalpy, and entropy) at any
temperature
Considering the dilution effect on the concentration, the total

concentration of reactant after the injection i is given by:
!
v i

½MT, i ¼ ½M0 1 1 ð15Þ
V
where [M]0 is the concentration of macromolecule M in the
syringe, v is the injection volume, and V is the cell volume. Again,
a correction due to the dilution of reactants has to be included in
Eq. 7:

v v
q i ¼ V ΔH d ½Mi ½Mi1 1 F 0 ½M0 ð16Þ
V V
where F0 has been defined above and can be calculated as:

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
Kd 8 ½M 0
F0 ¼ 1þ 1 ð17Þ
4½M0 Kd
Nonlinear fitting of qi as a function of [M]T,i provides Kd and ΔHd

as adjustable parameters. Blank experiments to estimate the dilu-
tion heat effect are obviously not possible; therefore, a constant
(but floating in the fitting procedure) term, qd, representing such
contribution must be included in Eq. 16.
time (min)
0 30 60 90 120 150 180 210
0.9
dQ/dt (mcal/s)
0.6
0.3
0.0
4.0
3.0
2.0
1.0
0.0
0 20 40 60 80 100 120
[a-Chymotrypsin]T (mM)
Fig. 6 Dissociation experiment of BP-α-CT. The experiment was performed in

20 mM sodium acetate, pH 3.9, 180 mM sodium chloride, at 25 C. The
concentration of BP-α-CT in the syringe is 608 μM. The solid line corresponds
to theoretical curve with Kd ¼ 53 106 M1 and ΔHd ¼ 5.5 kcal/mol of
monomer. The value of the dissociation constant is in agreement with that
determined by ultracentrifugation [62]. However, in the ITC experiment the
possibility of autocatalysis is minimized since it takes only 2 h and the
experiment by ultracentrifugation lasted for 2 days. The short time required to
do an experiment is one of the advantages of ITC compared to other binding
techniques (e.g., dialysis or ultracentrifugation). Another experiment performed
at 30 C (not shown) yielded a dissociation constant and a dissociation enthalpy
of 63 106 M1 and 6.2 kcal/mol. The estimated heat capacity of
dissociation is 140 kcal/K mol. A positive value reflects the solvent exposure
of molecular surfaces upon dissociation
For the experiment of bovine pancreatic α-chymotrypsin the

data were analyzed using a dimer dissociation model that allows the
heat of dilution to be fitted simultaneously (Fig. 6).
4 Notes
1. When performing experiments below room temperature, sam-

ples should be kept on ice or one degree below the experimen-
tal temperature while degassing.
2. The available total injection volume may be larger than the
nominal volume of the syringe. Filling the syringe incompletely
would generate nonuniform injections.
3. If the reaction is expected to be so exothermic and/or the

concentration of ligand is so high that the heat effect in the
first injections exceeds 200 μcal, the reference power should
be raised or the injection volume lowered in order to prevent
the signal to go below zero. Alternatively, the concentrations
could be lowered accordingly.
4. In the equilibration options, fast equilibration is more conve-
nient, compared to the other choices, because it permits simul-
taneous thermal and mechanical equilibration and it allows the
user to manually start the injection sequence whenever the
baseline is good.
5. The first injection is usually erroneous because there is some
mixing of the solutions inside and outside of the needle when
inserting the injection syringe and/or during the time required
to reach thermal and mechanical equilibration. Therefore, the
first injection is usually set smaller in order to waste the minimal
amount of ligand possible. In the data analysis, the first injec-
tion will be taken into consideration for the calculation of the
total concentrations in the cell. However, its associated heat
effect will not be included in the fitting procedure.
6. The feedback should be set to high feedback gain. Otherwise a
performance drop will occur: given an injection heat effect, the
corresponding peak would be smaller and broader (sensitivity
loss and increase of the time for running an experiment).
7. If possible, the stirring speed should be set at a sufficient level in
order to ensure proper mixing. Occasionally, lower stirring
speeds are required to avoid mechanical denaturation or desta-
bilization of the macromolecules.
8. The experimental conditions and buffer solutions should be
carefully determined for each system, depending on the partic-
ular characteristics (thermal stability, isoelectric point, solubil-
ity, propensity to interact with organic molecules, etc.) of the
interacting macromolecules.
9. The time in between injections may need to be adjusted if the
system is slow to reach equilibrium.
10. The data analysis software provided by the calorimeter manu-
facturer is straightforward and sufficient for simple systems
(e.g., 1:1 binding reaction). However, for the analysis of
more complex systems, such as the dissociation reaction pre-
sented here, or for including dilution heat effects, it might be
necessary to create additional fitting routines.
11. In data analysis, the heat associated to each injection is usually
normalized per mol of macromolecule injected. Therefore,
Eqs. 13 and 16 are modified including a normalizing factor:
Q i ¼ qi/(v [M1]0) and Q i ¼ qi/(v [M]0)i, respectively.
12. Given the values of enthalpy and association (or dissociation)

constant, it is possible to determine the optimal range of con-
centrations for reactants in an experiment. It is encouraged to
carry out simulations in the experimental design stage.
13. The user may find different expressions for the necessary cor-
rections in the concentrations and the heat effect due to dilu-
tion after injection. However, all those expressions are
equivalent as long as the injection volume is small compared
to the cell volume.
14. The user should be aware that a correct integration of the heat
effect is instrumental for successful data analysis. The baseline
may need to be adjusted independently for each injection.
15. Special care must be taken when determining the concentration
of the reactants. Uncertainties in the concentration of any
protein (e.g., due to contaminants or defective folding, which
would decrease the fraction of binding-competent protein) will
cause errors in the parameters. Active site titrations and/or
reliable spectrophotometric methods should be employed.
Colorimetric methods very often provide invalid results.
16. Performing direct and reverse titrations is the best criterion for
discriminating between 1:1 and more complex stoichiometries.
When normalizing the heat signal per mol of reactant injected,
both titrations will provide similar results if the stoichiometry is
1:1 (that is, there is a single binding site). Otherwise, the two
titrations will lead to different results if the system exhibits
higher stoichiometries [63].
Acknowledgments
This work was supported by grants from the National Science

Foundation (MCB0641252) and the National Institutes of Health
(GM56550 and GM57144).
References
1. Wells JA, McClendon CL (2007) Reaching for 4. Falconer RJ, Collins BM (2011) Survey of the
high-hanging fruit in drug discovery at protein- year 2009: Applications of isothermal titration
protein interfaces. Nature 450:1001–1009 calorimetry. J Mol Recognit 24:1–16
2. Zinzalla G, Thurston DE (2009) Targeting 5. Ghai R, Falconer RJ, Collins BM (2012)
protein-protein interactions for therapeutic Applications of isothermal titration calorimetry
intervention: a challenge for future. Future in pure and applied research - Survey of the
Med Chem 1:65–93 literature from 2010. J Mol Recognit 25:32–52
3. Falconer RJ, Penkova A, Jelesarov I, Collins 6. Banner DW, D’Arcy A, Janes W, Gentz R,
BM (2010) Survey of the year 2008: Applica- Schoenfeld HJ, Broger C, Loetscher H, Les-
tions of isothermal titration calorimetry. J Mol slauer W (1993) Crystal structure of the solu-
Recognit 23:395–413 ble human 55 kd TNF receptor-human TNF
beta complex: implications for TNF receptor helicase in complex with its molecular match-
activation. Cell 73:431–445 maker E2. Genes Dev 18:1981–1996
7. Shibata H, Yoshioka Y, Ohkawa A, Minowa K, 17. Myszka DG, Sweet RW, Hensley P, Brigham-
Mukai Y, Abe Y, Taniai M, Nomura T, Kaya- Burke M, Kwong PD, Hendrickson WA, Wyatt
muro H, Nabeshi H, Sugita T, Imai S, Nagano R, Sodroski J, Doyle ML (1997) Energetics of
K, Yoshikawa T, Fujita T, Nakagawa S, Yama- the HIV gp120-CD4 binding reaction. Proc
moto A, Ohta T, Hayakawa T, Mayumi T, Natl Acad Sci U S A 97:9026–9031
Vandenabeele P, Aggarwal BB, Nakamura T, 18. Gift SK, Zentner IJ, Schön A, McFadden K,
Yamagata Y, Tsunoda S, Kamada H, Tsutsumi Umashankara M, Rajagopal S, Contarino M,
Y (2008) Creation and x-ray structure analysis Duffy C, Courter JR, Zhang MY, Gershoni
of the tumor necrosis factor receptor-1-selec- JM, Cocklin S, Dimitrov DS, Smith AB 3rd,
tive mutant of a tumor necrosis factor-alpha Freire E, Chaiken IM (2001) Conformational
antagonist. J Biol Chem 283:998–1007 and structural features of HIV-1 gp120 under-
8. Hage T, Sebald W, Reinemer P (1999) Crystal lying the dual receptor antagonism by cross-
structure of the interleukin-4/receptor alpha reactive neutralizing antibody m18. Biochem-
chain complex reveals a mosaic binding inter- istry 50:2756–2768
face. Cell 97:271–281 19. Kwong PD, Doyle ML, Casper DJ, Cicala C,
9. Rickert M, Wang X, Boulanger MJ, Goriatch- Leavitt SA, Majeed S, Steenbeke TD, Venturi
eva N, Garcia KC (2005) The structure of M, Chaiken I, Fung M, Katinger H, Parren
interleukin-2 complexed with its alpha recep- PW, Robinson J, Van Ryk D, Wang L, Burton
tor. Science 308:1477–1480 DR, Freire E, Wyatt R, Sodroski J, Hendrick-
10. Philo JS, Aoki KH, Arakawa T, Narhi LO, Wen son WA, Arthos J (2002) HIV-1 evades
J (1996) Dimerization of the extracellular antibody-mediated neutralization through
domain of the erythropoietin (EPO) receptor conformational masking of receptor-binding
by EPO: one high-affinity and one low-affinity sites. Nature 420:678–682
interaction. Biochemistry 35:1681–1691 20. Freire E, Mayorga OL, Straume M (1990) Iso-
11. Syed RS, Reid SW, Li C, Cheetham JC, Aoki thermal titration calorimetry. Anal Chem
KH, Liu B, Zhan H, Osslund TD, Chirino AJ, 62:950A–959A
Zhang J, Finer-Moore J, Elliott S, Sitney K, 21. Doyle ML (1997) Characterization of binding
Katz BA, Matthews DJ, Wendoloski JJ, Egrie interactions by isothermal titration. Curr Opin
J, Stroud RM (1998) Efficiency of signalling Biotechnol 8:31–35
through cytokine receptors depends critically 22. Jelessarov I, Bosshard HR (1999) Isothermal
on receptor orientation. Nature 395:511–516 titration calorimetry and differential scanning
12. Kelekar A, Chang BS, Harlan JE, Fesik SW, calorimetry as complementary tools to investi-
Thompson CB (1997) Bad is a BH3 domain- gate the energetics of biomolecular recogni-
containing protein that forms an inactivating tion. J Mol Recognit 12:3–18
dimer with Bcl-XL. Mol Cell Biol 23. Leavitt S, Freire E (2001) Direct measurement
17:7040–7046 of protein binding energetics by isothermal
13. Kussie PH, Gorina S, Marechal V, Elenbaas B, titration calorimetry. Curr Opin Struct Biol
Moreau J, Levine AJ, Pavletich NP (1996) 11:560–566
Structure of the MDM2 oncoprotein bound 24. Parker MH, Lunney EA, Ortwine DF, Pav-
to the p53 tumor suppressor transactivation lovsky AG, Humblet C, Brouillette CG
domain. Science 274:948–953 (1999) Analysis of the binding of hydroxamic
14. Sundstrom M, Lundqvist T, Rodin J, Giebel acid and carboxylic acid inhibitors to the
LB, Milligan D, Norstedt G (1996) Crystal stromelysin-1 (matrix metalloproteinase-3)
structure of an antagonist mutant of human catalytic domain by isothermal titration calo-
growth hormone, G120R, in complex with its rimetry. Biochemistry 38:13592–13601
receptor at 2.9 Å resolution. J Biol Chem 25. Velazquez-Campoy A, Todd MJ, Freire E
271:32197–32203 (2000) HIV-1 protease inhibitors: enthalpic
15. Walsh ST, Jevitts LM, Sylvester JE, Kossiakoff versus entropic optimization of the binding
AA (2003) Site2 binding energetics of the reg- affinity. Biochemistry 39:2201–2207
ulatory step of growth hormone-induced 26. Todd MJ, Luque I, Velazquez-Campoy A,
receptor homodimerization. Protein Sci Freire E (2000) Thermodynamic basis of resis-
12:1960–1970 tance to HIV-1 protease inhibition: calorimet-
16. Abbate EA, Berger JM, Botchan MR (2004) ric analysis of the V82F/I84V active site
The X-ray structure of the papillomavirus resistant mutant. Biochemistry
39:11876–11883
27. Velazquez-Campoy A, Kiso Y, Freire E (2001) properties are dependent on thermodynamic

The binding energetics of first- and second- signature. Chem Biol Drug Des 77:161–165
generation HIV-1 protease inhibitors: Implica- 41. Burrows SD, Doyle ML, Murphy KP, Franklin
tions for drug design. Arch Biochem Biophys SG, White JR, Brooks I, McNulty DE, Scott
390:169–175 MO, Knutson JR, Porter D, Young PR, Hens-
28. Velazquez-Campoy A, Freire E (2001) Incor- ley P (1994) Determination of the monomer-
porating target heterogeneity in drug design. J dimer equilibrium of interleukin-8 reveals it is a
Cell Biochem S37:82–88 monomer at physiological concentrations. Bio-
29. Ward WH, Holdgate GA (2001) Isothermal chemistry 33:12741–12745
titration calorimetry in drug discovery. Prog 42. Czypionka A, de los Paños OR, Mateu MG,
Med Chem 38:309–376 Barrera FN, Hurtado-Gomez E, Gomez J,
30. Velazquez-Campoy A, Muzammil S, Ohtaka Vidal M, Neira JL (2007) The isolated C-
H, Schön A, Vega S, Freire E (2003) Structural terminal domain of Ring1B is a dimer made
and thermodynamic basis of resistance to HIV- of stable, well-structured monomers. Biochem-
1 protease inhibition: implications for inhibitor istry 46:12764–12776
design. Curr Drug Targets Infect Disord 43. Bello M, Perez-Hernandez G, Fernandez-
3:311–328 Velasco DA, Arreguin-Espinosa R, Garcia-
31. Vega S, Kang LW, Velazquez-Campoy A, Kiso Hernandez E (2008) Energetics of protein
Y, Amzel LM, Freire E (2004) A structural and homodimerization: effects of water sequester-
thermodynamic escape mechanism from a drug ing on the formation of beta-lactoglobulin
resistant mutation of the HIV-1 protease. Pro- dimer. Proteins 70:1475–1487
teins 55:594–602 44. Wiseman T, Williston S, Brandts JF, Nin LN
32. Ohtaka H, Muzammil S, Schön A, Velazquez- (1989) Rapid measurement of binding con-
Campoy A, Vega S, Freire E (2004) Thermo- stants and heats of binding using a new titra-
dynamic rules for the design of high affinity tion calorimeter. Anal Biochem 179:131–137
HIV-1 protease inhibitors with adaptability to 45. Doyle ML, Louie GL, Dal Monte PR, Soko-
mutations and high selectivity towards loski TD (1995) Tight binding affinities deter-
unwanted targets. Int J Biochem Cell Biol mined from linkage to protons by titration
36:1787–1799 calorimetry. Methods Enzymol 259:183–194
33. Ohtaka H, Freire E (2005) Adaptive inhibitors 46. Baker BM, Murphy KP (1996) Evaluation of
of the HIV-1 protease. Prog Biophys Mol Biol linked protonation effects in protein binding
88:193–208 using isothermal titration calorimetry. Biophys
34. Ruben AJ, Kiso Y, Freire E (2006) Overcom- J 71:2049–2055
ing roadblocks in lead optimization: a thermo- 47. Doyle ML, Hensley P (1998) Tight ligand
dynamic perspective. Chem Biol Drug Des binding affinities determined from thermody-
67:2–4 namic linkage to temperature by titration calo-
35. Lafont V, Armstrong AA, Ohtaka H, Kiso Y, rimetry. Methods Enzymol 295:88–99
Amzel LM, Freire E (2007) Compensating 48. Sigurskjold BW (2000) Exact analysis of com-
enthalpic and entropic changes hinder binding petition ligand binding by displacement iso-
affinity optimization. Chem Biol Drug Des thermal titration calorimetry. Anal Biochem
69:413–422 277:260–266
36. Freire E (2008) Do enthalpy and entropy dis- 49. Velazquez-Campoy A, Freire E (2005) ITC in
tinguish first in class from best in class? Drug the post-genomic era. . .? Priceless. Biophys
Discov Today 13:869–874 Chem 115:115–124
37. Freire E (2009) A thermodynamic approach to 50. Velazquez-Campoy A, Freire E (2006) Isother-
the affinity optimization of drug candidates. mal titration calorimetry to determine associa-
Chem Biol Drug Des 74:468–472 tion constants for high-affinity ligands. Nat
38. Ladbury JE, Klebe G, Freire E (2010) Adding Protoc 1:186–191
calorimetric data to decision making in lead 51. Zhang Y-L, Zhang Z-Y (1998) Low-affinity
discovery: a hot tip. Nat Rev Drug Discov binding determined by titration calorimetry
9:23–27 using a high-affinity coupling ligand: a thermo-
39. Kawasaki Y, Freire E (2011) Finding a better dynamic study of ligand binding to protein
path to drug selectivity. Drug Discov Today tyrosine phosphatase 1B. Anal Biochem
16:985–990 261:139–148
40. Schön A, Madani N, Smith AB, Lalonde JM, 52. Velazquez-Campoy A, Ohtaka H, Nezami A,
Freire E (2011) Some binding-related drug Muzammil S, Freire E (2004) Isothermal
titration calorimetry. Curr Protoc Cell Biol 59. Velazquez-Campoy A, Luque I, Todd MJ,
Chapter 17, Unit 17.8 Milutinovich M, Kiso Y, Freire E (2000) Ther-
53. Murphy KP, Freire E (1992) Thermodynamics modynamic dissection of the binding energet-
of structural stability and cooperative folding ics of KNI-272, a potent HIV-1 protease
behavior in proteins. Adv Protein Chem inhibitor. Protein Sci 9:1801–1809
43:313–361 60. Wyman J, Gill SJ (1990) Binding and linkage:
54. Gomez J, Hilser VJ, Freire E (1995) The heat functional chemistry of biological macromole-
capacity of proteins. Proteins 22:404–412 cules. University Science, Mill Valley, CA
55. Hinz HJ, Shiao DDF, Sturtevant JM (1971) 61. Edgcomb SP, Baker BM, Murphy KP (2000)
Calorimetric investigation of inhibitor binding The energetics of phosphate binding to a pro-
to rabbit muscle aldolase. Biochemistry tein complex. Protein Sci 9:927–933
10:1347–1352 62. Patel CN, Noble SM, Weatherly GT, Tripathy
56. Biltonen RL, Langerman N (1979) Microcalo- A, Winzor DJ, Pielak GJ (2002) Effects of
rimetry for biological chemistry: experimental molecular crowding by saccharides on α-chy-
design, data analysis and interpretation. Meth- motrypsin dimerization. Protein Sci
ods Enzymol 61:287–319 11:997–1003
57. Gomez J, Freire E (1995) Thermodynamic 63. Freire E, Kawasaki Y, Velazquez-Campoy A,
mapping of the inhibitor site of the aspartic pro- Schön A (2011) Characterisation of ligand
tease endothiapepsin. J Mol Biol 252:337–350 binding by calorimetry. In: Podjarny A, Dejae-
58. Baker BM, Murphy KP (1997) Dissecting the gere A, Kieffer B (eds) Biophysical approaches
energetics of a protein-protein interaction: the determining ligand binding to biomolecular
binding of ovomucoid third domain to elas- targets. Detection, measurement and model-
tase. J Mol Biol 268:557–569 ing. RSC Publishing, Cambridge
Chapter 12
Sedimentation Equilibrium Studies

Ian A. Taylor, Katrin Rittinger, and John F. Eccleston{
Abstract
The reversible formation of protein-protein interactions plays a crucial role in many biological processes.
In order to carry out a thorough quantitative characterization of these interactions it is essential to establish
the oligomerization state of the individual components first. The sedimentation equilibrium method is
ideally suited to perform these studies because it allows a reliable, accurate, and absolute value of the
solution molecular weight of a macromolecule to be obtained. This technique is independent of the shape
of the macromolecule under investigation and allows the determination of equilibrium constants for a
monomer–multimer self-associating system.
Key words Analytical ultracentrifugation, Sedimentation equilibrium, Solution interaction,

Quaternary protein structure, Protein-protein interaction
1 Introduction
Analysis of macromolecular interactions, in particular protein-

protein interactions, is an increasingly common goal in modern
biological research. Methods vary widely, from qualitative techni-
ques aimed at the detection of interactions in vivo to quantitative
in vitro analyses that attempt to produce a detailed thermodynamic
description of an interacting system. A necessary part of any quan-
titative in vitro method is that an initial characterization of the
oligomerization state of the systems components be undertaken.
This characterization is essential in order to correctly interpret
binding data produced from subsequent titration experiments
monitored, for example, by optical and magnetic resonance spec-
troscopy or isothermal titration calorimetry. Sedimentation equi-
librium studies are unique in that they can be used at all stages of
these quantitative studies. The technique allows a reliable,
accurate and absolute value of the solution molecular weight of a
macromolecule to be obtained making it invaluable in an initial
{
(Deceased)
205
206 Ian A. Taylor et al.
characterization of a system. Furthermore, the technique can be

extended to look at self-associating systems and heterologous
equilibria making it complementary to spectroscopic and calori-
metric methods (some good reviews about this technique include
refs. [1–7].
1.1 Determination Popular methods for the determination of molecular weights are
of Solution Molecular size exclusion chromatography and dynamic laser light scattering.
Weight In these methods, estimates of the molecular weight of unknowns
are obtained by interpolation, using a curve generated by plotting
an experimentally determinable parameter against the molecular
weight of a set of standards. In size exclusion chromatography
Kav, the partition coefficient between a porous matrix and free
solution, is often used. In dynamic laser light scattering, a transla-
tional diffusion coefficient DT, derived from the measured autocor-
relation function, is often the parameter of choice. Along with the
fact that using these methods molecular weights have to be
obtained by interpolation, the main drawback is that no account
is taken of molecular shape. Consequently, there are often large
errors in the values obtained. The effects of molecular shape are not
accounted for because the experimentally measured parameter
reported by both methods is a function of DT rather than molecular
weight. Although DT is correlated with molecular weight, it is
directly related, through Eq. 1, to an important molecular parame-
ter, the frictional coefficient, f [8].

RT
DT ¼ ð1Þ
N 0f
The value of f can be regarded as what determines inherent capacity

of a molecule to undergo translational motion, and both molecular
size and shape contribute to its value. Taking this into account it is
easy to see how measurements of molecular weights are so shape
dependent, since it is the frictional coefficient that directly deter-
mines both the DT measured in dynamic light scattering and the
partition coefficient observed in size exclusion chromatography.
The complications due to molecular shape are serious enough in
single component systems. When dealing with protein complexes
and interacting systems it should be apparent that these problems
are further exacerbated because of the need to account for the
effects of equilibrium constants.
1.2 Sedimentation To overcome problems associated with solution molecular weight

Equilibrium Theory determinations it is best to examine a phenomenon where the shape
of the molecule does not contribute to the measurement. This
occurs when a concentration gradient is established by a macromo-
lecular species sedimenting in a gravitational field, referred to as
“sedimentation equilibrium.” This method enables accurate,
shape-independent, and absolute measurements of molecular
Sedimentation Equilibrium Studies 207
weights to be obtained. Furthermore, sedimentation equilibrium

studies can be employed to analyze interacting systems, allowing
the determination of equilibrium constants and stoichiometries
alongside molecular weights.
In order to understand how this works it is necessary to intro-
duce the concept of flux. Equation 2 is a general expression for flow
and can be likened to Ohm’s law. Where the flux (Ji), or the flow of
material, is related to a term for a generalized conductivity (Li) and
(δUi/δr) a generalized gradient of potential.

δU i
J i ¼ L i ð2Þ
δr
In analytical ultracentrifugation, we are concerned with the flow of

mass. The total potential in this case is comprised from a compo-
nent due to the applied field, the centrifugal potential energy,
together with a component from the chemical potential gradient
generated from the solute. The conductivity term (Li) in this case is
the manifestation of the frictional coefficient and contributes to
both flow due to sedimentation and flow due to diffusion.

M ð1 νρÞ 2 RT δC
J ¼ ω rC ð3Þ
N 0f N 0f δr
M, molecular weight; v, partial specific volume; ρ, solute density;

N0, Avogadro’s number; ω, angular velocity; r, radial distance from
center of rotation; R, gas constant; T, absolute temperature).
Equation 3 provides a full description of the transport process
occurring in the ultracentrifuge cell. The terms [M(1 vρ)/N0f ]
and [RT/N0f ] correspond to the sedimentation coefficient (s) and
translational diffusion coefficient (DT), respectively, and chemical
potential, Ui has been replaced by concentration C. It should be
noted that the term for the strength of the applied gravitational
field (ω2r) will dominate at high rotor speeds, diffusion then only
manifests itself as the boundary spreading observed in sedimenta-
tion velocity experiments. At lower rotor speeds, back diffusion due
to the establishment of the chemical potential gradient counteracts
transport from the applied gravitational field. Under these condi-
tions equilibrium is established where the net transport in the
system vanishes to zero at all points. Equation 3 can then be written
as Eq. 4

M ð1 νpÞ 2 RT dC
ω rC ¼ ð4Þ
N 0f N 0f dr
An important result of this is that the shape term (N0f ) cancels,

meaning that whilst the concentration gradient established by the
macromolecular species is dependent on the molecular weight, it is
completely independent of shape. The useful form of the expression
involves rearrangement to give Eq. 5 and then integration with

respect to C and r, between the limits C0 and Cx and r0 and rx to
give Eqs. 6a and 6b.
ðC
1 M ω2 ð1 νρÞ r
ð
dC ¼ rdr ð5Þ
C0 C RT r0
M ω2 ð1 νpÞ 2

Cx
r x r 20

In ¼ ð6aÞ
C0 2RT
or
M ω2 ð1vρÞ
C x ¼ C 0e 2RT ðr 2x r 20 Þ ð6bÞ
1.3 Analysis There are currently two models of analytical ultracentrifuges pro-
of Molecular duced by Beckman-Coulter: the XL-A and the XL-I. Both contain
Weight Data an absorbance optical system able to measure the absorbance, at a
chosen wavelength, at many points in the cell between r0 and rx [9].
Providing data is collected within the usable linear range of the
optical system, absorbance can replace concentration in Eq. 6b
simply using the Beer–Lambert law, Eq. 7.
A xλ ¼ ελ c x l ð7Þ
In addition, the XL-I also has Rayleigh interference optics that

measures the displacement of fringes (Δj) in an interference
pattern produced from combining two monochromatic laser light
beams, one passed through the sample channel and the other
through the reference. The number of fringes displaced in
crossing from a point x1 to a point x2 in the sample cell is then
proportional to the difference in weight concentration between
these points, Eq. 8. Knowledge of the refractive index increment
(dn/dc ¼ 0/186 g ml1 for proteins) and laser wavelength
(λ ¼ 670 nm) then allows Δj to replace c in Eq. 6b.

dn
c l
dc
Δj ¼ ð8Þ
λ
Molecular weights are then simply extracted by direct nonlinear
least squares fitting of the data to Eq. 6b between r0 and rx with
substitution of c by either A or Δj. An offset is also included in the
fitting procedure for contributions to the profile from non-exact
matching of the cells, but should be treated with care. In the
simplest case, a single sedimenting species, the data should fit to a
single exponential curve in terms of r2. Deviations in the data from
a single exponential are indicative of sample heterogeneity, nonide-
ality or an associative system. Often this is revealed by inspection of
a plot of the fit residuals. A more complex model to account these
effects can then be applied.
1.4 Chapter Outline The instrumentation necessary to carry out analytical

ultracentrifugation studies is described in the Materials section
followed by a description of sample requirements and potential
practical limitations of this technique. The Subheading 3 describes
general considerations concerning experimental conditions for
equilibrium runs and contains a detailed protocol for data collec-
tion and analysis (see Subheading 3.3 and 3.4).
2 Materials and Equipment
2.1 Reagents 1. Baseline buffer: 10 mM Tris–HCl, 100 mM NaCl, 1 mM

TCEP, pH 7.5.
2. Dialysis units (such as a D-Tube Dialyzer, Novagen).
3. Purified proteins (~1–2 mgs).
4. Fluorocarbon oil FC-42.
2.2 Equipment 1. Optima XL-A or XL-I analytical ultracentrifuge (Beckman-

Coulter).
2. Six channel centerpieces for analytical ultracentrifuge cells.
3. Scanning (UV/Visible) spectrophotometer.
2.3 Description Although there are two optical systems available for Beckman-
of the Instrument Coulter analytical ultracentrifuges, in this chapter we will concen-
trate mainly on the use of absorbance optics. There are some
advantages to the use of Rayleigh interference optics, but these
largely surround the speed, the quality, and the density of data
that can be collected and are more relevant to sedimentation veloc-
ity rather than equilibrium experiments. A good comparison of
the advantages and disadvantages of the two optical systems can
found in [9].
The basic feature of the absorbance optics is to allow absor-
bance, as a function of radial distance, to be measured while the
solution is being centrifuged. This is achieved by the use of cells in
which the solution is contained within quartz windows. A mono-
chromator placed above the cell allows it to be illuminated with
monochromatic light between 190 nm and 800 nm, while a slit
mechanism below the cell scans radially at 0.001 cm or greater step
sizes (Fig. 1).
Two types of centrifuge rotors that hold the cells are available,
the An-50 Ti rotor (maximum speed 50,000 rpm) and the An-60
Ti rotor (maximum speed 60,000 rpm). The An-50 rotor has eight
rotor holes, seven for sample cells and one for a counterbalance,
whereas the An-60 has four rotor holes, three for sample cells and
one for a counterbalance. The counterbalance is required in all
centrifuge runs in order to coordinate the flash of the lamp with
the cell position when it is above the detector.
Fig. 1 A Schematic diagram of the optical system of the Beckman Optima XL-A analytical ultracentrifuge.
Figure courtesy of Beckman-Coulter
2.4 Software The Beckman Optima XL-A/XL-I is supplied with analysis soft-
and Data Analysis ware based on the Origin® software package (MicroCal). This
software has been used to fit all the data described in this chapter.
The calculations of buffer density (ρ) and the partial specific
volumes (ν) have been carried out using SEDNTERP [10] (see
Note 1). This and several other useful programs for the analysis
of equilibrium data have been developed by experts in the field and
are available by download over the Internet, in most cases as
Freeware. The Reversible Associations in Structural and Molecular
Biology group (RASMB) maintains a website that provides links to
most of these programs, some of which are available for different
platforms (http://www.bbri.org/RASMB/rasmb.html). Simula-
tions to determine optimal run conditions can be carried out
using the Beckman XL-A software and the freely available Sedfit
[11] and Sedphat [1] while the relative amounts of monomer and
n-mers for self-associating systems can be calculated using the
Ultrascan software (http://www.ultrascan.uthscsa.edu/).
2.5 Sample When preparing a sample for sedimentation equilibrium analysis

Preparation various points should be taken into account concerning buffer
composition and protein concentration.
2.5.1 Sample Buffer The sample buffer should not absorb more than 0.3 OD,
referenced against water at the wavelength chosen for the experi-
ment as this may otherwise interfere with the range and linearity of
absorbance measurements (see Note 2). Charge repulsion between
macromolecules will lead to problems with non-ideality, which is
best addressed by using a buffer with ionic strength >0.1 M and by
carrying out the experiments close to the pI of the protein in order
to minimize the charge on the surface of the molecule. Further-
more, it should be borne in mind that samples might be in the
centrifuge for long periods of time (>60 h), and therefore, buffer
components that have significant time-dependent changes in their
absorbance spectrum, for instance DTT, should be avoided.
2.5.2 Protein The sample concentration should not exceed 0.7 OD when
Concentration measured against the sample buffer, otherwise the linear range of
the detector (approximately 1.5 OD) will be exceeded when the
absorbance rises towards the bottom of the cell during the run (see
Fig. 1). If no information about the association state of the protein
is available, 0.5 OD is a good starting point. If self-association is
present, the loading concentration should be chosen such that
there are detectable amounts of monomers and multimers (see
Subheading 2.4). The protein solution and sample buffer have to
be in equilibrium, this is best achieved by gel filtration (using for
example NAP-5 columns, GE Healthcare) or exhaustive dialysis
(2–3 buffer changes). We routinely use D-Tube Dialyzer, Novagen
which is available for a range of different volumes starting at 10 μl.
Before every ultracentrifugation run, the sample should be
checked for aggregation. This might be done by a separate sedi-
mentation velocity experiment or by static and/or dynamic light-
scattering analysis if an instrument is available. Otherwise, a simple
test is to spin the sample in a microcentrifuge for 15–20 min at
maximum speed and to recheck the absorbance (see Note 3). If the
sample OD has significantly decreased, the sample might not be
suitable for equilibrium analysis at this time.
2.6 Limitations The range of molecular weights tractable by sedimentation equilib-

of Measurement rium is extremely large (5 102–107), governed only by the rotor
for Protein–Protein speed attainable. However, the analysis of an interacting system is
Complexes limited by the absorbance range where reliable concentration mea-
surements can be made within the confines of the equilibrium
constant under investigation. Figure 2 shows the fraction of pro-
tein monomers expected for a set of monomer–dimer equilibria
(Ka 103–109 M1) over a concentration range of 1 nM to 1 mM.
Superimposed on this figure, in grey, is the concentration range
over which reliable absorbance measurements can be made using
the XL-A/XL-I optical system.
Fig. 2 Equilibrium binding isotherms for a homodimeric interaction. The fraction

of monomer, [m]/[mt] is plotted as a function of the total monomer concentration
[mt]. Curves from left to right are log order decreases in the association constant
ranging from Ka ¼ 109 M1 to Ka ¼ 103 M1. The grey shaded area represents
the concentration range over which reliable measurements can be made. The
dark grey area represents measurements made at 280 nm using standard
12 mm cells. The lighter region is where data has to be collected off peak, at
shorter wavelength or using short pathlength cells
It should be apparent that for interactions with Ka > 108 M1

the sedimentation experiment will give an accurate value for the
molecular weight of the complex and therefore the stoichiometry.
For interactions much weaker than 104 M1 or where a heteroge-
neous mixture is present, sedimentation equilibrium will provide
the weight averaged molecular weight of the mixture showing no
concentration dependency. For interactions between these two
limits Ka > 104 M1 and <108 M1 sedimentation experiments
will also provide a weight averaged molecular weight, but in this
case with a concentration dependence allowing the equilibrium con-
stant for the system to be evaluated. Figure 2 is representative of a
50 kD protein with a molar extinction coefficient of 50,000 at
280 nm. Using a cell with a 1.2 cm optical path length, a concentration
of 13.3 μM gives a starting absorbance of 0.8 and a concentration of
1.66 μM has an absorbance of 0.1. These represent the limits in
absorbance for which reliable data can be collected. However, there
are a number of ways to extend this range. One example is by collect-
ing at a shorter wavelength where peptide bond absorbance will
contribute to the extinction coefficient. The absorbance optics can
measure at 230 nm and this will usually increase the extinction coeffi-
cient by about three to four times, extending the absorbance range
down to 0.5 μM (see Note 2). To extend the range upwards shorter
pathlength cells (3 mm) are available extending the range up to
50 μM. Finally, data can be collected away from the peak of the
absorbance in the 250 nm trough or at 295 nm to give around a two
to threefold decrease in sensitivity, expanding the range up to 150 μM.
3 Methods
3.1 Cell Assembly The cells themselves are assembled for each centrifugation run.
There are several different types of cell but all consist of the follow-
ing main components: a cylindrical cell housing that fits into the
rotor, two window assemblies consisting of a window holder, a
quartz window, a white Vinylite window gasket, and a Bakelite
window liner. These two window assemblies form a sandwich
around a centerpiece that contains the samples. This sandwich is
assembled in the cell housing and a screw ring and gasket is then
screwed into one or both ends of the cell housing and tightened
with a torque wrench (see Note 4). Full details of the cell assembly
are provided in the manufacturer’s instructions and should be
followed very carefully (https://www.beckmancoulter.com/
wsrportal/techdocs?docname¼LXLA-TB-003).
There are many types of centerpieces constructed of different
materials and dimensions. The simplest centerpieces are the double
sector centerpieces, which allow for a sample solution and a buffer
blank. These have a path length of 1.2 cm and are made in alumi-
num, aluminum filled Epon or charcoal filled Epon. The aluminum
double sector centerpiece has the advantage of having a maximum
speed of 60,000 rpm while the others are limited to 42,000 rpm.
The charcoal filled Epon centerpiece should be used if there is any
possibility of aluminum ions interacting with the sample, otherwise
all of the centerpieces are compatible with aqueous solutions but
softening of the Epon can occur if strong acids or organic solvents
are used (see Note 4). A 3 mm path length charcoal filled Epon
centerpiece is also available and requires the use of two spacers.
Charcoal filled Epon centrerpieces that contain six channels (three
samples and three buffer blanks) have a maximum speed of
48,000 rpm whereas the eight channel cells (four samples and
four buffer blanks) can be operated up to 50,000 rpm. The double
sector centerpieces that hold 450 μl per sector are generally used for
sedimentation velocity measurements because of their longer col-
umn length. However, they can also be used for equilibrium runs if
the volume is reduced to 100 μl. This is particularly useful if the
3 mm path length centerpiece is required since this is not available
in six channel centerpieces. In this case, the volume is reduces to
33 μl. The volumes required to fill a 6-sector cell are 110 μl of
protein and 120 μl of sample buffer.
3.2 Speed and Before starting a sedimentation equilibrium experiment, two

Duration of the important parameters need to be determined. First, the optimal
Experiment speed of centrifugation needs to be known. Beckman provides
a chart relating optimal speed to expected molecular weight
(see https://www.beckmancoulter.com/wsrportal/bibliography?
docname¼362784.pdf). Alternatively, Eq. 6b can be solved for
rpm [rpm ¼ (30/π) ω] for a given value of Cx/C0. A good guide
is that at equilibrium, around a fivefold increase in concentration
over the column length will produce an exponential distribution
suitable to get a good fit to the data (Cx/C0 ¼ 5). Regardless of
this calculation, it is advisable to centrifuge at speeds lower and
higher than this optimal speed (allowing equilibrium to be reached
first at the lower speed, then optimal speed and finally at higher
speed). The lower and higher speeds can be calculated from Eq. 6b,
say for (Cx/C0) ¼ 3 and 7. Another way of finding optimal experi-
mental conditions in terms of protein concentration or rotor speed
is to simulate equilibrium data for a particular set of parameters (see
Note 5). This can be done with the Origin XL-A analysis software
under ‘Utilities/Data Simulator’.
The other important parameter that needs to be determined is
the time taken for equilibrium to be reached, Teqm. This can be
calculated from Eq. 9 where, Teqm (seconds) is directly proportional
to the square of the solution height, h (cm) and inversely propor-
tional to DT.
0:7ðh Þ2
T eqm ¼ ð9Þ
DT
From Eq. 9 it is apparent that smaller molecules reach equilib-
rium faster than larger ones. A 20 kD protein in about 16 h and a
100 kD protein in around 27 h. A value for DT can be determined
experimentally, but for these purposes it is reasonable to estimate a
value by assuming the macromolecule is spherical and applying
Eq. 10.
3 105
DT ¼ p
3
ffiffiffiffiffi ð10Þ
M
Whichever method is used, it is still necessary to show experi-
mentally that the system has reached equilibrium. The best way to
do this is to overlay or subtract successive scans, taken at two hourly
intervals. When equilibrium is established no further change in the
absorbance profile should occur, the only differences being due to
noise in the data.
3.3 Sedimentation 1. Prepare around 2 L of a suitable buffer for the sedimentation

Equilibrium Studies study to be carried out in. For instance, 10 mM Tris–HCl,
of a Monomeric 100 mM NaCl, 1 mM TCEP pH 7.5 (r/t) (see Note 2).
Species 2. Dialyze a stock protein solution against at least three changes of
500 ml of this baseline buffer. If necessary concentrate the
sample after dialysis. Ensure that there is enough of the stock

protein to be able to prepare around 700 μl of a solution with
an optical density of 0.7 at 280 nm, in a 1 cm pathlength cell
(see Note 6).
3. Using this working solution prepare a dilution series of protein
solutions, 120 μl in each, ranging from the highest optical
density down to about 0.1 in a 1 cm cell. Because of the setup
of the cells, it is convenient to use either three or nine samples.
If sample is limiting then three is adequate; however, nine
allows duplicates and to cover a larger concentration range.
4. Place 10 μl of Fluorocarbon oil FC-42 in each channel of the
pre-assembled bottom sections of the six channel centrifuge
cells and add 110 μl of each protein solution into the sample
channels and 120 μl of baseline buffer into the buffer channels
(see Note 7). Finally, insert the top window assembly and
tighten down the top section of each cell.
5. Balance the cells and place into the analytical rotor as described
in the user manual. Carefully install the monochromator,
set the run temperature, usually 20 C, and apply the vacuum
(see Note 8).
6. Set the rotor speed to an initial speed of 3,000 rpm. When the
centrifuge has reached the required speed, temperature and
vacuum, collect a single radial scan at λ ¼ 280 nm of each cell
using a step size of 0.001 cm and two averages per scan. Refer
to see Note 9 for expected appearance of this pre-scan.
7. Set the centrifuge speed to the lowest of the three speeds to be
collected during the run, for instance about 10,000 rpm for a
90 kDa protein. If association equilibria are expected or sus-
pected set the rotor speed to optimize the gradient in favor of
the higher molecular weight species (the slowest).
8. After 16 h collect radial scans every 2 h using a step size of
0.001 cm and five averages per scan. Assess whether equilib-
rium has been reached by overlaying or subtracting successive
scans. Do this until there is no further change in the absorbance
profile. The final scan in this dataset should be suitable for
molecular weight analysis.
9. Set the centrifuge to the next speed in the set of three, wait for
about 12 h before again collecting radial scans every 2 h using a
step size of 0.001 cm and five averages per scan. Again, assess
when equilibrium has been reached and collect a scan to use in
the molecular weight analysis.
10. Finally, set the centrifuge to the highest speed where data will
be collected. After 12 h, collect the radial scan data, assess for
equilibrium as before, and collect a final scan suitable for use in
data analysis.
11. When data has been collected at all the necessary speeds, set the
centrifuge to 42,000 rpm. After approximately 16 h, collect a
radial scan using a 0.001 cm step size with two replicates. This
“overspeed” scan gives a depleted boundary that should be
used to provide a reasonable estimate of the cell offset during
the data fitting procedure (see Note 10).
3.4 Analyzing All data analysis described in the following paragraph has been
the Data carried out using the Origin® software package. This program
allows the direct fitting of the equilibrium concentration gradients
to mathematical models for single species or associating systems
using nonlinear least squares algorithms [12, 13]. Furthermore, it
allows global fitting of multiple datasets covering a range of con-
centrations and rotor speeds. This is particularly important for self-
associating systems.
1. Once a whole dataset is collected, select the files that will be
used in the molecular weight analysis. Load the relevant “RA”
files into Origin, then cut out the data corresponding to the
sample absorbance profiles using the select subset command
(see Note 11).
2. First, use the Origin software to individually fit each “cut” file
assuming an ideal single species model. In this fitting proce-
dure, include the offset value obtained from the “overspeed”
scan along with the calculated values for ρ and ν (see Note 10).
3. Plot the molecular weights obtained against the initial protein
concentration. Is there any correlation between initial concen-
tration and apparent molecular weight? If so, this may be
indicative of an interacting system.
4. Plot ln(C) against r2 (under Utilities/Plot) (see Note 12).
5. If the fit of the individual data files provided any indication of
self-association carry out a simultaneous (global) fit, still
assuming a single species. Under these conditions there should
be clear systematic deviations in the residuals plot indicating
that the protein is not an ideal monomeric species.
6. Carry out global fits to monomer–dimer, monomer–trimer,
and monomer–tetramer equilibria and carefully inspect for
deviations between the fitted curve and the experimental data
as well as for systematic deviations in the residuals plot. At this
stage it is of great importance to have some previous knowl-
edge of the monomer molecular weight of the protein under
investigation as this allows the monomer molecular weight to
be fixed (see Note 13).
7. Convert the association constant expressed in absorbance units
into the commonly used association constant Ka (see Note 14).
3.5 Examples of Fits 1. The Swi6 transcriptional regulator is a 90 kD protein. Analysis

to Experimental Data of the molecular weight using size exclusion chromatography
and dynamic light scattering provided estimates of the molecu-
lar weights of 250 kD and 280 kD, respectively, suggesting that
the protein is either dimeric or trimeric. Subsequent analysis of
the Swi6 solution molecular weight using sedimentation equi-
librium, Figure 3, reveal the data are fit well with an ideal
monomer molecular mass of 90 kD [14] and that lnC vs. r2
plots and the application of associative models provide no
indication of oligomerization. The comparative analysis
demonstrates how erroneous estimates of molecular weight
arise when using techniques that do not account for the effect
of molecular shape.
2. Sedimentation equilibrium runs of bovine dynamin I, a 100 kDa
protein that oligomerizes in vivo and in vitro [15, 16]. The runs
Fig. 3 Analysis of the solution molecular weight of Swi6 by sedimentation equilibrium. The lower panel shows
the absorbance profile produced at equilibrium by Swi6 at a rotor speed of 12,000 rpm, T ¼ 293 K,
ν ¼ 0.727, ρ ¼ 1.003. The upper panel shows the residuals to the plot
Fig. 4 Sedimentation equilibrium data of dynamin I fitted to different models: (a) monomer alone; (b) dimer
alone; (c) tetramer alone; (d) monomer/tetramer. Reproduced with permission from Journal of Protein
Chemistry (1999) 18, 277–290
shown in Fig. 4 have been carried out at 4 C at a speed

of 7,000 rpm and protein concentration of 0.33 mg/ml.
Figure 4a–c show fits of the data, along with residuals, assuming
a monomeric, dimeric, and tetrameric species, respectively. The
fit to a single species gave a molecular weight of 285 kD, signifi-
cantly bigger than that of a monomeric protein. Furthermore,
the non-random distributions of the residuals of all three fits
clearly indicate that none of these models represents the correct
association behavior. Only a monomer–tetramer model gave a
satisfactory fit with a random distribution of residuals as shown
in Fig. 4d, allowing the calculation of an association constant of
K1,4 ¼ 1.67 1017 M3. Assuming a monomer–dimer–tetra-
mer model resulted in the same quality of fit with no improve-
ment over the monomer–tetramer model. The association
constant K1,2 for the monomer–dimer equilibrium calculated
from this fit is so small relative to K1,4, that there would be hardly
any dimer present under the experimental conditions, thereby
justifying the assumption of a monomer–tetramer model.
3.6 Heterologous Sedimentation equilibrium methods could be especially useful for

Systems studying the heterologous interactions that are important in almost
every biological process. However, the analysis of these data is
complicated by the fact that the absorbance along the cell has
contributions from both the free proteins and the complex(es). If
the two proteins are of similar molecular weight and extinction
coefficient, they can be treated as a homologous associating system
[17]. However, this is usually not the case and so more complex
strategies must be used. These are beyond the scope of this chapter,
and interested readers are referred to a review article by Minton and
colleagues [3] that discusses different methods for the quantitative
characterization of heterologous interactions and to Philo [18]
who has discussed how problems can be alleviated through the
use of numerical constraints during data analysis.
4 Notes
1. In fact, it is the buoyant molecular weight, M* that is directly

obtainable from the experimental data. The absolute molecular
weight has to be calculated from Eq. 11 using the solvent
density ρ and the partial specific volume, ν of the particle.
M
M ¼ ð11Þ
ð1 vρÞ
Good values for the solvent density ρ can be obtained from

tabulated values. An approximate value for the partial specific vol-
ume of a protein ν is 0.73 ml/g. In some cases, it may be determin-
able experimentally using densitometry. A reasonably accurate
alternative is to calculate partial specific volumes based on the
amino acid composition of the protein. The SEDNTERP program
does this using Eq. 12, where, ni, Mri, and vi are the number,
molecular mass, and partial specific volume of the ith residue.
∑ni M i vi
v¼ ð12Þ
∑ni M i
The need to obtain accurate values for these parameters stems

from the (1 νρ) relationship and because of this a 3 % error in ν
results in close to a 10 % error in molecular weight.
2. The optical system of the XL-A/XL-I is capable of measuring
data at wavelengths between 800 and 190 nm, although below
230 nm there is very little lamp intensity and so this region is
not generally useful. In all measurements it is important to try
to keep the buffer absorbance to a minimum because low
transmittance of the buffer blank will decrease the range, line-
arity, and sensitivity of any measurements. If measurements are
carried out at a wavelength in the ultraviolet region, a

non-absorbing buffer has to be selected. At short wavelengths
NaCl should be substituted by NaF to reduce buffer absor-
bance. 232 nm is a good wavelength because there is a peak in
the lamp spectrum. However, the intensity of the light source
of the instrument decreases with time, in particular in the UV
region, due to deposits of oil on the lamp and a scan of the light
intensity against wavelength should be done regularly to deter-
mine if the lamp needs cleaning. This problem has been par-
tially relieved on newer machines by incorporation of a window
at the base of monochromator stem that reduces the oil
buildup.
3. A UV spectrum taken from 230 to 350 nm can provide a good
indication of sample quality. A sloping baseline between 310
and 350 nm is indicative of protein aggregation.
4. Before assembling the cells make sure everything is clean because
salt deposits can lead to leakage or cracking of the windows.
Furthermore, it is necessary to check if the buffer components
are compatible with the cell components. Chemical compatibil-
ity tables are available at (https://www.beckmancoulter.com/
5. The ‘Simulator Setup’ window allows the user to define all neces-
sary run parameters. Note please that the loading concentration is
given in mg/ml and will be converted to absorbance based onto
the concentration conversion factor. A detailed description of the
program can be found at (https://www.beckmancoulter.com/
6. The optical pathlength of the ultracentrifuge cells is 12 mm,
rather than 10 mm used in most standard spectrophotometric
cuvettes. Bear this in mind when making up the sample dilution
series.
7. Load the reference column slightly higher than the sample so
the reference meniscus does not interfere with the sample.
Fluorocarbon oil FC-42 has the same refractive index as water
but is denser and forms a boundary at the bottom of the cell
and thereby reduces artifacts caused by light at the bottom of
the cell.
8. Do not forget to equilibrate the rotor at the experimental
temperature or best at a slightly lower temperature. Also, the
experimental parameters, ρ and ν are temperature dependent
and usually hydrodynamic data is referenced to 20 C. If the
experiment is not carried out at 20 C this needs to be
accounted for.
9. The 3,000 rpm pre-scan should have a constant absorbance

along the whole of the cell. It is worth writing these values
down for any subsequent analysis of concentration depen-
dency. The presence of a boundary or the appearance of
upward curvature in the absorbance profile towards the bot-
tom of the cell is indicative of high molecular weight or
aggregated material in the sample. This probably means the
sample is not suitable for further analysis. The lack of any
absorbance indicates cell leakage and requires the run to be
repeated.
10. The offset value should be used with caution and only if it can
be justified in some experimental way. It is highly correlated
with A0 and “adjustment” of its value will dramatically alter
the molecular weight values obtained. The use of the over-
speed at the end of a run is a reasonable estimation of the
value, but may not be correct. The best way to avoid offset
problems is to avoid buffers with absorbing components and
to take care when preparing the cells for a run, avoid scratches
and fingerprints.
11. If the equilibrium run has been carried out using six or eight
channel centerpieces, all the data for the sample channels will
be in a single file and have to be separated into individual files
before any fitting procedures can be carried out. This is done
via the “Select subset” option in the in the Utilities menu.
This feature allows the selection of a subset of data-points
from a whole profile to be saved as an individual file.
12. A simple, model-independent method to check for heteroge-
neity is a plot of ln(C) versus r2. A straight line for this plot
suggests the presence of a single species, with the slope equal-
ling M*, the buoyant molecular weight of the protein (see
Eq. 11 and Note 1). An upward curving line is indicative of
an associating system. However, this method can be rather
insensitive to heterogeneity and should only be taken as a
guide.
13. The molecular weight of the protein can easily be calculated
based on the amino acid composition if the amino acid sequence
of the protein is known, see for example (http://www.expasy.
ch/proteomics/protein_characterisation_and_function). Oth-
erwise, the molecular weight has to be experimentally deter-
mined, for example by mass spectrometry. Alternatively,
sedimentation equilibrium runs can be carried out under dena-
turing conditions.
14. Equilibrium constants determined in the Beckman origin
software have units expressed in absorbance units. Equa-
tions 13a, 13b, and 13c are used to convert them into con-
centration units for monomer–dimer, monomer–trimer, and
monomer– tetramer equilibria, respectively.
K aabs ðεm l Þ
Ka ¼ ð13aÞ
2
K aabs ðεm l Þ2
Ka ¼ ð13bÞ
3
K aabs ðεm l Þ3
Ka ¼ ð13cÞ
4
Acknowledgement
This work was supported by the Medical Research Council, UK.
References
1. Vistica J, Dam J, Balbo A et al (2004) Sedi- SE, Rowe AJ, Horton JC (eds) Analytical
mentation equilibrium analysis of protein inter- Ultracentrifugation in Biochemistry and Poly-
actions with global implicit mass conservation mer Science. The Royal Society of Chemistry,
constraints and systematic noise decomposi- Cambridge United Kingdom, pp 90–125
tion. Anal Biochem 326:234–256 11. Brown PH, Schuck P (2006) Macromolecular
2. Howlett GJ, Minton AP, Rivas G (2006) Ana- size-and-shape distributions by sedimentation
lytical ultracentrifugation for the study of pro- velocity analytical ultracentrifugation. Biophys
tein association and assembly. Curr Opin Chem J 90:4651–4661
Biol 10:430–436 12. Johnson ML, Faunt LM (1992) Parameter
3. Rivas G, Stafford W, Minton AP (1999) Char- estimation by least-squares methods. Methods
acterization of heterologous protein-protein Enzymol 210:1–37
interactions using analytical ultracentrifuga- 13. Johnson ML, Correia JJ, Yphantis DA et al
tion. Methods 19:194–212 (1981) Analysis of data from the analytical
4. Laue TM, Stafford WF 3rd (1999) Modern ultracentrifuge by nonlinear least-squares tech-
applications of analytical ultracentrifugation. niques. Biophys J 36:575–588
Annu Rev Biophys Biomol Struct 28:75–100 14. Sedgwick SG, Taylor IA, Adam AC et al (1998)
5. Cole JL, Lary JW, PMoody T et al (2008) Structural and functional architecture of the
Analytical ultracentrifugation: sedimentation yeast cell-cycle transcription factor swi6. J Mol
velocity and sedimentation equilibrium. Meth- Biol 281:763–775
ods Cell Biol 84:143–179 15. Binns DD, Helms MK, Barylko B et al (2000)
6. Cole JL, Hansen JC (1999) Analytical ultra- The mechanism of GTP hydrolysis by dynamin
centrifugation as a contemporary biomolecular II: a transient kinetic study. Biochemistry
research tool. J Biomol Tech 10:163–176 39:7188–7196
7. Liu J, Shire SJ (1999) Analytical ultracentrifu- 16. Binns DD, Barylko B, Grichine N et al (1999)
gation in the pharmaceutical industry. J Pharm Correlation between self-association modes
Sci 88:1237–1241 and GTPase activation of dynamin. J Protein
8. Cantor CR, Schimmel PR (1980) Ultracentri- Chem 18:277–290
fugation. In: biophysical chemistry, Part II. 17. Silkowski H, Davis SJ, Barclay AN et al (1997)
Freeman Characterisation of the low affinity interaction
9. Laue TM (1996) Choosing which optical sys- between rat cell adhesion molecules CD2 and
tem of the Optima XL-I analytical centrifuge to CD48 by analytical ultracentrifugation. Eur
use. Beckman-Coulter Technical Application Biophys J 25:455–462
Information Bulletin A-1821-A 18. Philo JS (2000) Sedimentation equilibrium
10. Laue TM, Shah BD, Ridgeway TM et al (1992) analysis of mixed associations using numerical
Computer-aided interpretation of analytical constraints to impose mass or signal conserva-
sedimentation data for proteins. In: Harding tion. Methods Enzymol 321:100–120
Chapter 13
Detecting Protein-Protein Interactions by Gel Filtration

Chromatography
Yan Bai
Abstract
Upon protein-protein interaction, the formed complex is subject to a change in size. A number of methods
can be utilized to detect such a change. Gel filtration technology is well recognized for its ability to monitor
and separate protein species of different sizes, and can greatly facilitate functional studies of protein
complexes. In addition, gel filtration can be performed in any buffer system that preserves the protein
complex formation and function. Therefore, it can be a significantly useful method for studying protein
interactions. In this chapter, a protocol for performing gel filtration is described in detail.
Key words Gel filtration, FPLC, Molecular weight estimation, Protein dimerization
1 Introduction
Gel filtration separates substances by their difference in size as they

pass through a medium packed in a column, and therefore this
method is also known as size-exclusion chromatography (SEC).
Gel filtration medium is a porous matrix made up of cross-linked
polymer spherical particles. There are many kinds of media that
differ in their chemical and physical properties, and are designed to
suit a variety of separation tasks. When a solution composed of
different sized molecules is applied to a column, the molecules
that are too large to diffuse into the pores will be eluted from the
column all together. These large molecules are said to be excluded.
The smaller molecules diffuse into the pores of the medium parti-
cles, take routes with different lengths based on their sizes, and
elute in the order of their sizes with the largest molecule first and
the smallest last (Fig. 1). The terminology for the size measurement
is “Stokes radii” [1], which is the radius of a hard sphere that
diffuses at the same rate as the molecule. Since a molecule’s Stokes
radius is often unavailable, as for globular molecules, their molecu-
lar weights are used to represent their size. Commercially available
gel filtration media cover a wide molecular weight range from 100
223
224 Yan Bai
Small
molecule
pores
large
molecule
Fig. 1 Illustration of the gel filtration principle. Protein molecules pass through
the gel medium via different routes based on their size. Small molecules take
longer to elute because they can get into small pores and big pores; big
molecules take less time because they are too large to travel through the pores
to 80,000,000 Da, from peptides to large proteins and protein

complexes. Any packed column, based on its physical properties,
has its own selectivity. Therefore, in practice, an effective separation
range is always the primary consideration when choosing an appro-
priate gel column for a specific task.
Compared with other chromatographic separation methods,
gel filtration has several advantages: (1) the solute recovery rate is
high, approaching 100 % [2], (2) high reproducibility is expected
for repeated runs, (3) various buffer conditions can be used for
many types of samples. Separation can be performed in the presence
of essential ions or cofactors, therefore proteins can be preserved in
their active conformation. The tolerance of high concentrations of
detergents and wide range of ionic strength of gel filtration are
greatly appreciated when working with less soluble proteins, such as
membrane proteins and transcription factors.
The fundamental usage of gel filtration is for protein purifica-
tion. However, because the technique provides size information for
biological substances, it has been used extensively for other
biological events that result in a molecule size change or molecule
complex formation, such as protein refolding [3], the generation of
amyloid fibrils [4], and, especially, functional proteomics by reveal-
ing protein-protein interactions under conditions that preserve
native protein complexes [5–7]. This chapter reviews important
concepts in gel filtration and describes, how to utilize gel filtration
technology to study protein-protein interactions, including sample
preparation, instrument operation, and data analysis. The proce-
dure for gel filtration is as follows: (1) Set up the gel filtration
system and a specific gel filtration method. (2) Equilibrate the
Gel Filtration 225
Intermediate MW
Injection point
High MW
Low MW
Vo Vt
Void volume Vo Total volume Vt Vs(Vt-Vo)
Ve
Vt-Vo
Fig. 2 Common terms in gel filtration. Vo (the void volume): space taken by solvent surrounding the gel beads,
usually about 30 % of the total bed volume. Vt: the total volume of a gel bed. Vs: space occupied by solvent
inside the medium particles
column and prepare protein standards and samples. (3) Run protein
standards to establish a calibration curve. (4) Run protein samples.
(5) Analyze collected protein fractions on an SDS-PAGE gel.
1.1 General Terms A set of terminology is employed to describe the gel filtration
and Concepts in Gel process. As shown in Fig. 2, the total volume of a gel bed, Vt,
Filtration includes space occupied by solvent inside the medium particles Vs,
and space taken by solvent surrounding the gel beads called Vo (the
void volume), which is about 30 % of the total bed volume.
Vt ¼ Vo þVs ð1Þ
The elution volume, Ve, is the volume of mobile phase entering
the column between the start of the elution and the emergence of
the peak maximum [8].
1.2 Gel Filtration Detection of protein-protein interaction by size change requires

Resolution high-resolution gel filtration. Several factors influence resolution:
column dimensions, particle size and distribution, pore size of the
particles, medium packing density, sample volume, flow rate, and
viscosity of the sample and buffer. These factors contribute to two
effects that determine the final resolution, the selectivity and the
efficiency of the medium. The selectivity of the medium, which
solely depends on its pore size distribution, refers to the molecular
weight range over which a gel filtration medium can effectively
separate molecules. The medium selectivity is described by a selec-
tivity curve, in which a partition coefficient, Kav, is plotted against
the logarithm of the molecular weight for a set of standard proteins.
K av ¼ ðV e V o Þ=ðV t V o Þ ð2Þ
K av ¼ alogðMW Þ þ b ð3Þ
The steeper the curve, the higher the resolution reached.

226 Yan Bai
The efficiency of a packed column defines its ability to produce

narrow symmetrical peeks. The uniformity of the column packing
and particle size contributes the most to the efficiency. Increasing
medium bedding height and decreasing the medium particle size
can increase the resolution; however, this slows down the flow rate.
Besides the selectivity of the medium, the sample volume and
column dimensions are also crucial for the resolution. Usually, the
recommended sample volume is within 0.5–4 % of the column bed
volume, and for high-resolution results, the sample volume should
not exceed 2 % of the total bed volume.
For studying protein-protein interaction by gel filtration, the
running buffer is another critical matter. The buffer should be able
to maintain the interactions of interest, such as appropriate pH, ion
strength, detergent concentration, and optimal buffering range
should be carefully chosen.
2 Materials
1. Chromatography system (e.g., AKTApurifier FPLC system

(see Note 1) and AKTA design fraction collector, Frac-920).
2. Gel filtration columns. Superdex 75 (Superdex 75 10/300)
(see Note 2).
3. Gel filtration molecular weight markers. 3.5 mg albumin
bovine serum (66 kDa), 1 mg carbonic Anhydrase from bovine
erythrocytes (29 kDa), 1 mg aprotinin (6,512 Da).
4. Purified protein sample (e.g., 2 mL His-14-3-3 zeta (~30 kDa)
(see Note 3). The protein was expressed in an E. coli system and
purified using Ni-NAT beads. The protein sample was filtered
through a 0.2 μm, 25 mm syringe filter (Nylon, sterile) to
remove any insoluble materials).
5. Distilled water (500 ml), filter through a 0.2 μm filter (non-
pyrogenic, sterile) and degas.
6. 20 % ethanol in distilled water (500 ml), filter through a 0.2 μm
filter (nonpyrogenic, sterile) and degas.
7. Running buffer (500 ml): 50 mM Tris–HCl, pH 7.5, 100 mM
NaCl (see Note 4), filter through a 0.2 μm filter (nonpyrogenic,
sterile) and degas.
3 Methods
3.1 Chromatography It is advisable to read the manufacture’s user manual for detailed
System Setup instrument operation and maintenance. A simplified outline for the
AKTApurifier FPLC system is provided here.
Gel Filtration 227
Fig. 3 Parameter settings to conduct a typical gel filtration run
The AKTApurifier is an automated liquid chromatography

system. It has two major modules: Pump P-900 and monitor
UPC-900, a highly precise online monitor to measure UV absorp-
tion at 280 nM, conductivity and pH. Each module has its own
power switch. The system is controlled and monitored by a soft-
ware package, UNICORN, which is installed on a PC and runs
under the Microsoft Windows operating system.
1. Turn on the chromatography system, turn on the computer
and launch UNICORN. Four interfaces will pop up: UNI-
CORN manager, Method Editor, System Control and Evalua-
tion (see Note 5).
2. Create methods. New methods are created through the
Method Wizard. All major parameters for a gel filtration run,
such as column type, pressure limit, flow rate, sample volume,
fraction volume and length of run can be entered step by step
(Fig. 3) (see Note 6). By clicking Finish in the last dialog, the
Run Setup Window appears, which allows the user to review all
information about the method. The user can save the method
by selecting File:Save and saving it with an appropriate name.
3. Set up the column and the sample loop. Check the tubing
connections and the sample loop. The tubing on the top of
the column is connected to port 1 of the injection valve, and
the bottom tubing goes to the UV cell port. The sample loop
228 Yan Bai
a Inject Mode b Load Mode
Column 1 7 Pump Column

1 7 Pump
Sample Sample
2 syringe 2 syringe
6 6
3 3
Sample Sample
loop loop
4 5 4 5
Waste Waste Waste Waste
Fig. 4 Illustration of the injection valve at inject mode and load mode. After being manually injected through
port 3, the sample will be retained in the sample loop between port 2 and port 6. (a) Upon sample loading on
the column, the system automatically switches to the “Inject” mode, in which the flow path is 7-6–2-1. (b)
When elution starts, the liquid flows directly from 7 to 1, bypassing the sample loop
connects between valve port 2 and port 6. The waste tubing is

connected to port 4 and 5 (Fig. 4).
4. Set up the fraction collector. Insert collection tubes into the
tube holder starting at position 1 and ending at position 70.
Gently lift the delivery arm up and slide it to the outer stop,
then place the tube rack over the central spindle and pull the
spring-loaded drive sleeve out and snap in the rack. Lift and
lower the delivery arm, and move it in so the tube sensor
touches the outer wall of the tube at position 1 and the eluent
tubing is above the center of the collection tube (see Note 7).
3.2 Sample Loading 1. Equilibrate the column (see Note 8). Wash the column using at
and Fractionating least 1CV (column volume) of filtered distilled water, and
equilibrate the column using at least 1CV of running buffer.
By the end of the equilibration, the conductivity trace should
show an “S” shape and level off in the end. The operation can
be set up and executed through Manual on the System control
module.
2. Wash the sample loop. The sample loop (0.5 ml) should be well
cleaned using running buffer. Use a needle syringe to inject 1 ml
of running buffer into the fill port three times (see Note 9).
3. Load the sample. Go to the UNICORN System Control inter-
face, under Manual/ flow path, set “Inject valve” to “Inject”
before taking out the syringe. Take out the syringe and fill the
syringe with 180–200 μl of protein sample (see Note 10).
Insert the syringe needle into the fill port. Set “Inject valve”
Gel Filtration 229
to “load”, and then slowly inject the sample into the loop.
Leave the syringe in position during the entire run. Go to the
System Control interface, click “End” on the operation tool
bar. Ensure that 1 mL of buffer volume is used to empty the
sample loop in order to load all of the sample onto the column.
4. Start a run. Right click the “Run” button on the top tool bar to
select a method for the run. Several dialog windows will be
displayed sequentially, on which the user can review/edit the
method settings and set up how the results file should be
generated. Click “Start” on the last page of the dialog window
to initiate the run.
5. Monitor the run. The gel filtration progress can be monitored
in the System Control module. The ongoing experiment can
be pulsed or ended at any time by clicking the “Pulse” or
“End” button on the top tool bar.
3.3 Create the Studies [9, 10] have shown that, for globular proteins, their parti-
Standard Curve and tion coefficients, Kav, are related to their corresponding molecular
Estimate the Molecular weights in a sigmoidal fashion on gel filtration columns. Therefore,
Weight a calibration curve, also called a selectivity curve, can be created by
plotting a group of known proteins’ partition coefficients against
the logarithm of their molecular weight. Other elution parameters,
such as Ve, Ve/Vo, Kd, have been used in the calibration curve,
however the use of Kav is recommended because it is insensitive to
errors that may be generated by column preparation and dimension
variations.
An unknown protein sample’s molecular weight can be esti-
mated according to its elution volume using a calibration curve. It
is important to make sure that the calibration curve’s linear range
covers the size of the protein of interest since only the linear region
of the plot is reliable [10]. When gel filtration is used to detect or
confirm protein complex formation, the protein complex’s molec-
ular weight is usually considered as the sum of each individual
protein. However, when the complex exhibits an irregular shape,
the molecular weight cannot be simply extrapolated from the cali-
bration curve. For example, an elongated protein complex will
behave like a bigger molecular on the gel chromatogram than it
actually is.
1. Determine the void volume (Vo). Make 2 mg/ml blue dextran
in running buffer (see Note 11).
2. Prepare the protein standards. Dissolve 3.5 mg of Albumin
bovine serum, 1 mg of Carbonic Anhydrase and 1 mg of
aprotinin in 500 μl of running buffer (see Note 12).
3. Equilibrate the column using running buffer and run the pro-
tein standards as described in Subheading 3.2.
4. Calculate Kav values for each protein maker.
230 Yan Bai
a b mAU
MW: KDa
Kav 200 14-3-3
0.5 75
160
0.4 37
25 14-3-3
120
0.3
80
0.2
y = -0.3185x + 0.6917 Void
0.1 R² = 0.9998 40
0 0
0.0 0.5 1.0 1.5 2.0 Log Mr 0 5 10 15 20 25 ml
Fig. 5 (a) A standard curve was created based on the experimentally determined Kav of the standard proteins.
(b) Chromatogram of the protein sample. The experimental Ve is 9.39 mL. Therefore, the estimated molecular
weight of the protein sample is 61.74 kDa, indicating a dimer form of the protein. The collected fractions were
pooled and run on 12.5 % SDS-PAGE, the protein monomer was observed on the gel at approximately 30 kDa
5. Plot the Kav values (Y axis) versus the logarithm of the

corresponding molecular weights (X axis). Use linear fitting
to generate the equation Kav ¼ alog(MW) + b, and obtain the
values for a and b (Fig. 5a).
6. Calculate the molecular weight of the protein sample based on
the calibration curve once its Kav has been obtained experimen-
tally. Plug the experimental values for Kav(sample), a and b in the
equation in Fig. 5a, and obtain the MW value for the protein
sample.
7. Analyze 1–2 μg of each fraction collected by SDS-PAGE to
confirm the size of the protein (Fig. 5b).
4 Notes
1. An HPLC or FPLC system is commonly used for gel filtration;

in this chapter we use an FPLC system as an example.
2. There are many commercially available gel filtration columns,
choose an appropriate column according to experimental
requirements. For high-resolution results, Superdex medium
is the first choice. All columns should be well sealed at both
ends and stored in 20 % ethanol at 4 C.
3. Usually 0.1–0.8 ml of protein sample (0.5–4 % column volume)
is used for one run. Use the right size sample loop according to
the protein sample volume.
4. Running buffers should contain at least 50–100 mM salt to
prevent protein aggregation. However, the maximum salt con-
centration should be empirically determined since high salt
concentration can break down protein complexes.
Gel Filtration 231
5. The user can control the UNICORN displays and manage files
through the UNICORN manager module. Method Editor is
for creating new methods. System Control is used to monitor
and control runs in real time. The results for each run will be
displayed and evaluated in Evaluation module.
6. Correct maximum pressure setup is the most critical parameter
to protect columns. Ensure that the back-pressure over the
column does not exceed the maximum recommended pressure
(1.8 M for a Superdex 75). The recommended maximum
pressure for a selected column can be found in manufacturer’s
instruction sheet. Pressure could gradually build up with
increasing system usage. One of the common reasons is that
impurities build up on the in-line filter. The filter can be cleaned
by soaking it in 0.1 N NaOH in a beaker with gentle agitation
overnight and rinsing it thoroughly with filtered deionized
water. In addition, the columns have to be washed completely
after each use. Once column clogging is observed, a stringent
cleaning operation has to be performed immediately according
to manufacturer’s instructions.
7. The process of setting up a fraction collector varies depending
on the type of collector used.
8. The column equilibration step can be incorporated into the
method through the Method Wizard so that the equilibration
will automatically end and switch to the elution step. However,
it may be more advantageous to have the equilibration run
separately through manual control, by which the user can
alter the equilibration volume as needed to make sure that the
column is perfectly equilibrated before starting a run.
9. If partial filling is used (see Note 10), at the end of the last loop
wash, do not remove the syringe from the fill port, otherwise
the buffer inside the sample loop will drain out.
10. There are two ways to load the sample, partial filling and
complete filling. When a high recovery rate is required, partial
filling is used. The maximum sample volume loaded cannot
exceed 50 % loop volume, 1/3 of the loop volume is usually
loaded. Make sure the sample loop is completely filled with
buffer before sample loading. Complete filling requires about
2–3 loop volumes of the sample to achieve 95 % maximum loop
volume. For example, in order to fill a 0.5 mL sample loop with
95 % protein sample without diluting it, 1–1.5 mL of protein
sample is required to be injected into the loop. When complete
filling is used, a buffer volume at least five times the sample loop
volume should be used to flush the loop during emptying the
loop.
11. 5 % glycerol can be contained in the running buffer to increase
the density of the solution, but it is optional.
232 Yan Bai
12. If more than three protein markers are used, dissolving all
markers together may cause unassignable peaks because BSA
may form a complex with other protein markers. It is advisable
to prepare multiple protein marker groups, 2–3 proteins for
each group.
References
1. Ackers GK (ed) (1975) The proteins, vol 1. assimilation in Haloferax mediterranei: inter-
Molecular sieve methods of analysis. Academic, action between glutamine synthetase and two
New York, NY GlnK proteins. Biochim Biophys Acta 1834:
2. Cooper TG (ed) (1977) Gel permeation chro- 16–23
matography. The tools of biochemistry. Wiley, 7. Kuo WY, Huang CH, Liu AC et al (2013)
New York CHAPERONIN 20 mediates iron superoxide
3. Freydell EJ, van der Wielen LA, Eppink MH dismutase (FeSOD) activity independent of its
et al (2010) Size-exclusion chromatographic co-chaperonin role in Arabidopsis chloroplasts.
protein refolding: fundamentals, modeling and New Phytol 197:99–110
operation. J Chromatogr A 1217:7723–7737 8. Wechsler DS, Dang CV (1992) Opposite
4. Hall D, Huang L (2012) On the use of size orientations of DNA bending by c-Myc and
exclusion chromatography for the resolution of Max. Proc Natl Acad Sci U S A 89:7635–7639
mixed amyloid aggregate distributions: I. 9. Porath J, Flodin P (1959) Gel filtration: a
Equilibrium partition models. Anal Biochem method for desalting and group separation.
426:69–85 Nature 183:1657–1659
5. Monti M, Cozzolino M, Cozzolino F et al 10. Abate C, Luk D, Gentz R et al (1990)
(2009) Puzzle of protein complexes in vivo: a Expression and purification of the leucine
present and future challenge for functional pro- zipper and DNA-binding domains of Fos
teomics. Expert Rev Proteomics 6:159–169 and Jun: both Fos and Jun contact DNA
6. Pedro-Roig L, Camacho M, Bonete MJ directly. Proc Natl Acad Sci U S A 87:
(2013) Regulation of ammonium 1032–1036
Chapter 14
Using Light Scattering to Determine the Stoichiometry

of Protein Complexes
Jeremy Mogridge
Abstract
The stoichiometry of a protein complex can be calculated from an accurate measurement of the complex’s
molecular weight. Multiangle laser light scattering in combination with size exclusion chromatography and
interferometric refractometry provides a powerful means for determining the molecular weights of proteins
and protein complexes. In contrast to conventional size exclusion chromatography and analytical centrifu-
gation, measurements do not rely on the use of molecular weight standards and are not affected by the
shape of the proteins. The technique is based on the direct relationship between the amount of light
scattered by a protein in solution, and the product of its concentration and molecular weight. A typical
experimental configuration includes a size exclusion column to fractionate the sample, a light scattering
detector to measure scattered light, and an interferometric refractometer to measure protein concentration.
The determination of the molecular weight of an anthrax toxin complex will be used to illustrate how
multiangle laser light scattering can be used to determine the stoichiometry of protein complexes.
Key words Light scattering, Stoichiometry, Molecular weight, Size exclusion chromatography,
Interferometric refractometer, Protein complex
1 Introduction
Elucidating the stoichiometry of a protein complex can provide

insights into its structure and its molecular mechanisms of action.
Stoichiometry can be calculated from the measurement of the
molecular weight of a complex, but this can be challenging if
there are several copies of more than one protein species. Techni-
ques such as conventional size exclusion chromatography and ana-
lytical centrifugation have been used to estimate molecular weights,
but these techniques are limited by their reliance on protein stan-
dards and by the influence of protein shape on the measurement. In
contrast, the combination of size exclusion chromatography, laser
light scattering, and interferometric refractometry is an absolute
method for the determination of molecular weight [1, 2]. It is
based on the theory that the amount of light scattered by a protein
233
234 Jeremy Mogridge
in solution is directly proportional to the product of the protein’s

concentration and molecular mass. Scattered light is measured by a
light scattering (LS) detector, protein concentration is measured by
an interferometric refractometer (IR), and the molecular weight is
calculated by computer software. The accuracy of this technique
allows for precise molecular weight measurements that are required
to calculate the stoichiometry of a multi-subunit protein complex.
The determination of the stoichiometry of the Bacillus anthracis
edema toxin complex will be used as an example to describe how
multiangle laser light scattering can be used to probe protein
structure [3]. Edema toxin consists of two proteins, edema factor
(EF) and protective antigen (PA), that are secreted from Bacillus
anthracis and assemble on the surface of mammalian cells [4]. After
PA is cleaved on the cell surface into two fragments, the remaining
cell-associated fragment, PA63, oligomerizes into heptamers that
can bind edema factor to form a toxic complex. The molecular
weight of the saturated complex measured in solution corre-
sponded to a stoichiometry of three molecules of EF per PA63
heptamer.
2 Materials
1. HPLC.
2. DAWN EOS 18-angle light scattering detector (Wyatt
Technology).
3. OPTILAB DSP interferometric refractometer (Wyatt
Technology).
4. ASTRA software (Wyatt Technology).
5. Superdex 200 HR 10/30 column.
6. Column buffer: 20 mM Tris–HCl, pH 8.2, 200 mM NaCl.
7. 0.2 μm filter units.
8. 0.02 μm Anotop 10 filters.
3 Methods
3.1 System Setup Before measurements are taken, the LS detector must be calibrated
with a pure solvent that has a known Rayleigh ratio, such as toluene.
Calibration relates the voltage detected by the 90 detector to the
measured intensity of scattered light. If the LS detector was cali-
brated by the manufacturer, it needs to be re-calibrated only if any
changes to the detector that affects the 90 signal, such as realign-
ment of the laser, have been made.
The HPLC is connected in series to the LS detector and the IR.
The IR should be placed after the LS detector to avoid subjecting
the IR to high back-pressure. Connections should be made with
Light Scattering 235
tubing of low inner diameter (0.0100 ) to minimize the dead volume

between the instruments.
A solution of 4 mg/mL of bovine serum albumin (an isotropic
scatterer) in column buffer can be injected onto the size exclusion
column and used to normalize the LS detector, a process that
relates the measured voltages at each detector to the 90 detector
in order to compensate for slight differences in electronic gain
among the detectors. Normalization must be performed each
time a different solvent is used.
A dn/dc (refractive index increment) value of 0.185 mL/g is
used for non-glycosylated proteins in the molecular weight calcula-
tions made by ASTRA software.
3.2 Preparation 1. Turn on the IR approximately 12 h before measurements are

taken to warm the instrument. Set the temperature to approxi-
mately 10 C above room temperature.
2. Turn on the LS detector 2 h before measurements are taken.
3. Connect the size exclusion column to the HPLC (see Note 1)
and run buffer through the column while the light scattering
detector is warming (see Note 2).
4. Once the column has pre-equilibrated with buffer (approxi-
mately 2 column volumes), connect the waste line from the
HPLC to the LS detector and the output from the LS detector
to the IR (see Note 3).
5. Equilibrate the LS detector and IR with column buffer (see
Note 4).
6. Flow buffer simultaneously through the reference and sample
cells of the IR and set the baseline to zero. After the reference
has been set, program the IR so that the column buffer
bypasses the reference cell.
3.3 Injecting Do not stop the HPLC pumps after the equilibration step or
a Sample between measurements because starting and stopping the pumps
may cause particles from the column to dislodge and interfere with
the measurements. Start data collection by ASTRA software and
inject each component of the protein complex (EF and PA63; see
Note 5) individually and then a mixture of the proteins (see Note
6). The results of a chromatography run in which a molar excess of
EF was mixed with PA63 are shown in Fig. 1.
3.4 Analyzing The results can be analyzed using a computer program, such as
the Results ASTRA.
1. Set the baseline voltages of the signals detected by the LS
detector and IR.
2. When selecting values for the light scattering and refractive
index signals, use data relating to the middle of the protein
236 Jeremy Mogridge
Fig. 1 Multiangle laser light scattering of EF:[PA63]7. A mixture (120 μL) of EF

(260 μg) and [PA63]7 (93 μg; approximately twofold molar excess of EF over PA63
monomer) was chromatographed over a Superdex 200 size exclusion column,
which was connected to a light scattering detector (lower panel ) and an
interferometric refractometer (upper panel ). The values of molecular mass
determined in volume increments across each peak are shown (arrows).
Reprinted with permission from Biochemistry 41: 1079–1082. Copyright 2002
American Chemical Society
Table 1
Molecular mass determinations using multiangle laser light scattering
Measured molecular Theoretical molecular

mass (kDa) mass (kDa)
EF 93 0.3 91
[PA63]7 460 3 444
EF:[PA63]7 720 20 717a
Reprinted with permission from Biochemistry 41: 1079–1082. Copyright 2002 Ameri-
can Chemical Society
a
Theoretical value for saturated complex is based on a stoichiometry of three molecules
of EF per PA63 heptamer
peak where the signal-to-noise ratio is the highest (signal-to-

noise ratio should be at least 2).
3. Use ASTRA software to calculate the molecular weights
(Table 1; see Note 7).
Light Scattering 237
4 Notes
1. The HPLC system should have a pulse dampener since small

fluctuations in flow rate can affect the measurements. The
Superdex 200 size exclusion column was chosen for this study
because EF and the EF:[PA63]7 complex are separated by this
column.
2. It is important to make fresh buffer with high quality reagents.
Filter the buffer through a 0.2 μm filter and degas it thor-
oughly before use. Pre-equilibrate the column at the flow-rate
(e.g., 0.5 mL/min) that will be used to take measurements to
avoid changes in flow-rate that might cause particles to dis-
lodge from the column.
3. New columns should be equilibrated for approximately 48 h
before measurements are taken. The LS detector and IR are
stored in methanol, which should be replaced with filtered
water before introducing buffer (solvents that are injected
directly into the LS detector or IR should be filtered through
a 0.02 μm filter).
4. Equilibrate the LS detector and IR until the signals have stabi-
lized. Baseline noise for the LS detector should be in the
10–20 mV range. The lower angle LS detectors (the Wyatt
EOS LS detector has 18 detectors) will have a higher level of
noise, but these detectors can be turned off without adversely
affecting the measurements. Detectors 8–18 were used in this
study.
5. EF and [PA63]7 were purified as described previously [5, 6].
Small amounts of aggregated protein in a sample can interfere
with the light scattering measurements because aggregates are
very efficient at scattering light. An aggregate will produce a
high signal from the LS detector and a very low signal from the
IR. Recently prepared protein samples that have not been
frozen may contain less aggregate than a sample that has been
frozen and thawed.
6. The maximum injected volume is determined by the column
specifications (250 μL for the Superdex 200 column), but
smaller volumes can be used. The amount of protein required
to make an accurate measurement depends on the molecular
weight of the protein—more sample is required for low molec-
ular weight proteins (~50 kDa) than for high molecular weight
proteins (>100 kDa) because high molecular weight proteins
scatter more light. Accurate measurements of the amino termi-
nal domain of Bacillus anthracis lethal factor (30 kDa) were
made with 200 μg of protein [3]. Accurate measurements of
transferrin (75 kDa) and ovalbumin (43 kDa) were made with
less than 10 μg of protein [7]. An additional consideration
238 Jeremy Mogridge
when measuring protein complexes is that the proteins should

be at a high enough concentration to ensure that they fully
associate. In general, the protein concentration should exceed
the dissociation constant of the interaction by approximately
tenfold (taking into account that the protein may be diluted by
severalfold on the size exclusion column). The ratios of the
proteins in the samples should be varied to ensure that satura-
tion has occurred. The amount of one protein is held constant
and increasing amounts of the other protein are added. Satura-
tion has been achieved when the addition of protein does not
change the molecular weight of the complex peak and a free
protein peak is observed.
7. ASTRA software calculates molecular weights at multiple
points across the selected section of an elution peak and uses
this data to calculate the average molecular weight of the
protein. These points can be displayed in a graph of molecular
weight versus elution volume to aid in the analysis of the
experiment (Fig. 1, lower panel). Calculated molecular weights
across a homogeneous peak will be similar at each point, but
these points may form a frowning pattern, which is indicative of
peak broadening between the LS detector and IR (see EF:
[PA63]7 complex peak in Fig. 1). Slight peak broadening will
not adversely affect the molecular weight measurements. If the
protein peak is not homogeneous, one might observe a linear
pattern with higher molecular weights at lower elution volumes
and lower molecular weights at higher elution volumes. If this
is the case, a size exclusion column that better separates the
species can be used.
References
1. Wyatt PJ (1993) Light scattering and the abso- 5. Zhao J, Milne JC, Collier RJ (1995) Effect of
lute characterization of macromolecules. Anal anthrax toxin’s lethal factor on ion channels
Chim Acta 272:1–40 formed by the protective antigen. J Biol Chem
2. Gell DA, Grant RP, Mackay JP (2012) The 270:18626–18630
detection and quantitation of protein oligomer- 6. Miller CJ, Elliott JL, Collier RJ (1999) Anthrax
ization. Adv Exp Med Biol 747:19–41 protective antigen: prepore-to-pore conversion.
3. Mogridge J, Cunningham K, Collier RJ (2002) Biochemistry 38:10432–10441
Stoichiometry of anthrax toxin complexes. Bio- 7. Folta-Stogniew E, Williams KR (1999) Deter-
chemistry 41:1079–1082 mination of molecular masses of proteins in
4. Young JA, Collier RJ (2007) Anthrax toxin: solution: implementation of an HPLC size
receptor binding, internalization, pore forma- exclusion chromatography and laser light scat-
tion, and translocation. Annu Rev Biochem tering service in a core laboratory. J Biomol Tech
76:243–265 10:51–63
Chapter 15
Circular Dichroism (CD) Analyses of Protein-Protein

Interactions
Norma J. Greenfield
Abstract
Circular dichroism (CD) spectroscopy is a useful technique for studying protein-protein interactions in
solution. CD in the far ultraviolet region (178–260 nm) arises from the amides of the protein backbone and
is sensitive to the conformation of the protein. Thus, CD can determine whether there are changes in the
conformation of proteins when they interact. Changes in the conformation of the protein complexes as a
function of temperature or added denaturants, compared to the individual proteins, can be used to
determine binding constants. CD bands in the near ultraviolet (350–260 nm) and visible regions arise
from aromatic amino acid side chains and prosthetic groups. There are often changes in these regions when
proteins bind to each other. Because CD is a quantitative technique, these changes are directly proportional
to the amount of the protein–protein complexes formed and thus also can be used to estimate binding
constants.
Key words Conformation, Secondary structure, Binding constants, Thermodynamics of folding
1 Introduction
Circular dichroism (CD) is a valuable spectroscopic technique for

studying protein-protein interactions in solution. There are many
review articles in the literature on the theory and general applica-
tions of CD [1–8], so this chapter will focus on how CD can be
used to follow protein-protein interactions. CD is best known as a
method of determining the secondary structure of proteins in
solution. It thus can be used to determine whether there are
changes in protein conformation when proteins interact with each
other. However, more importantly, CD can be used to estimate
binding affinities for protein interactions. Since CD is a quantitative
technique, changes in CD spectra are directly proportional to the
concentrations of the protein–protein complexes formed, and these
changes can be used to estimate binding constants. One can moni-
tor changes in the “intrinsic” CD arising from the conformational
transitions of the amide backbone and amino acid side chains or
239
240 Norma J. Greenfield
“extrinsic” CD of the proteins arising from bound ligands or

prosthetic groups. Changes in the thermodynamics of folding
upon protein-protein interaction can also be used to determine
binding constants.
1.1 Spectra The intrinsic CD spectrum of a protein, arising from its amide
of the Peptide backbone, is sensitive to the secondary structure of the protein.
Backbone The amide backbone bonds absorb in the far UV. Most commercial
CD machines can collect far UV spectra between 260 and 178 nm.
To a first approximation, increases in negative ellipticity at 222 and
208 nm and positive ellipticity at 193 nm usually indicate an
increase in α-helical content, while increases in a single negative
band near 218 nm and positive band at 195 nm indicates an
increase in β-structure. Figure 1a shows typical CD curves obtained
for a polypeptide, Poly-L-lysine in the 100 % α-helical, 100 %
β-sheet, and 100 % denatured states [9]. Figure 1b illustrates the
increase in α-helix that occurs when a peptide from titin, a Z-repeat,
interacts with a fragment of the C-terminal domain of α-actinin
[10]. Figure 1c illustrates the increase in β-structure when two cell
cycle control proteins, Arf and Hdm2, interact [11]. Both of these
proteins are mainly disordered in the absence of the interactions.
1.2 Side Chain and Non-backbone intrinsic CD bands are due to the chromophores of
Extrinsic CD Spectra the aromatic amino acids, tryptophan, tyrosine, and phenylalanine
and disulfide bonds. Extrinsic bands are due to noncovalently
bound chromophores such as hemes, aromatic cofactors, and col-
ored metal ions. Figure 2a illustrates a change in the aromatic
spectrum of the catalytic domain of a cAMP-dependent protein
kinase when it binds to an artificial peptide substrate [12], and
Fig. 2b illustrates the change in the heme spectrum of cytochrome
C when it binds to cytochrome C oxidase [13].
1.3 Binding There are four methods that can be used to extract binding (or
Constants dissociation) constants from CD data. These include:
1. Direct titration of one protein with another.
2. Serial dilution of a protein complex.
3. Changes in susceptibility to denaturants or osmolytes of the
protein complex compared to the unbound components.
4. Changes in thermal stability of the complex compared to the
unbound components.
If formation of the complex changes the protein secondary
structure or causes changes in the environment of optically active
aromatic or prosthetic groups of one or both of the interacting
proteins, one protein can be titrated by the other. The change in
CD upon binding is directly proportional to the amount of com-
plex formed, and thus the binding constant can be directly
CD Spectroscopy 241
Fig. 1 Examples of conformational changes upon protein-protein interactions.

(a) CD spectra of a polypeptide, poly-L-Lysine, showing spectra typical for the
100 % α-helical (solid line), 100 % antiparallel β-sheet (dotted line), and 100 %
denatured (short dashed line) conformations. Data from Greenfield and Fasman
[9] used with permission from the American Chemical Society. (b) Gain of signal
for the Act-EF34-Zr7 complex (complex of peptides of titin, Zr7, and actinin,
EF34) upon complex formation. The spectra of the isolated components, Zr7
(solid line), Act-EF34 (short dashed line), and the complex (dotted line), are
shown. The sum of the spectra of the two isolated components is also reported
for comparison (dotted dashed line). Data from Joseph et al. [10] used with
permission from the American Chemical Society. (c) The complex of two cell
cycle control proteins, Arf and Hdm2, are comprised of β-strands and are
thermally stable. CD spectra for Hdm2210-304 alone (solid line) and mArfN37
alone (dashed dotted line) and Hdm2210-304 mixed with 1 (short dashed line), 1.7
(dotted line), and 2 (long dashed line) molar equivalents of mArfN37. Data of
Bothner et al. [11] used with permission from Elsevier
determined. However, even if there are no conformational changes,

if the interactions cause a change in stability, one can determine
the thermodynamics of folding of the individual and complexed
proteins and determine the binding constant by assuming that all of
the change in free energy of folding is due to the binding. The
binding constant can then be determined using the relationship
Fig. 2 Examples of changes in the CD spectra arising from nonamide

chromophores upon protein-protein interactions. (a) Near UV CD of the
catalytic subunit of adenosine cyclic 50 monophosphate-dependent protein
kinase. (Inset) CD spectrum of the catalytic subunit from 275 to 300 nm (solid
line) enzyme alone; (dashed line) enzyme plus 250 μM Kemptide, a synthetic
peptide substrate (buffer plus Kemptide base line subtracted). Data from Reed
and Kinzel [12] used with permission from the American Chemical Society. (b)
CD difference spectra of cytochrome c upon binding to cytochrome c oxidase.
Data from Michel et al. [13] used with permission from the American Chemical
Society
k ¼ exp(ΔΔG/RT), where ΔΔG is the difference in free energy of

folding of the mixture versus the unmixed proteins, R is the gas
constant, T is the temperature, and k is the association constant.
2 Materials
2.1 Proteins CD can be performed on proteins ranging in concentration from

0.005 to 2 mg/ml depending on the path lengths of the cuvettes
used (see below). CD depends on concentration of chromophores,
rather than molar protein concentrations. CD is usually reported
either in units of mean residue ellipticity, [θ], in deg cm2/dmol, or
the difference between the absorption of left- and right-handed
circularly polarized light, ΔE. ΔE ¼ [θ]/3298. For quick calcula-
tions, [θ] ¼ θ MRW/c, where θ is the ellipticity in millidegrees,
MRW is the mean residue weight (the molecular weight of the
protein divided by the number of amino acids), and c is the protein
concentration in mg/ml. When one is studying the aromatic spec-
trum of a protein, the results are usually reported as molar ellipticity
(mean residue ellipticity times the number of residues).
2.2 Buffers For the most precise estimates of protein secondary structure from
CD spectra, it is important to collect data to low wavelengths;
178 nm has been recommended [3, 14, 15]. In this case one should
use buffers with very low absorption, such as 10 mM potassium
CD Spectroscopy 243
phosphate. If salt is necessary to stabilize the protein, 100 mM KF

or 50 mM K2SO4 is preferred as they are relatively transparent.
NaCl should be avoided. However, for most purposes an adequate
estimate of protein conformation can be obtained using data col-
lected between 260 and 200 nm [5]. For these measurements, one
can use buffers such as phosphate buffered saline or 20 mM
Tris–HCl, or 2 mM Hepes containing low concentrations of
EDTA or EGTA, e.g., 1 mM, and 1 mM Dithiothreitol and up to
500 mM NaCl if necessary.
For folding experiments, e.g., those monitored at 222 nm,
almost any buffers can be utilized provided that the total absorption
of the sample (protein plus buffer) does not exceed 1 at the wave-
length of interest.
2.3 Cuvettes CD measurements are usually performed in cuvettes with path-

lengths of 0.1 or 1 cm. Both standard rectangular and cylindrical
cuvettes are available for CD machines, and the choice depends on
the instrument utilized. For measurements between 260 and
200 nm for concentrations ranging from 0.005 to 0.02 mg/ml,
one needs 2–3 ml of solution, as the samples will be studied in 1-cm
cuvettes, with a total volume capacity of 3.5 ml. For measurements
between 178 and 260 nm at higher concentrations, e.g.,
0.05–1 mg/ml, one would use 0.1- to 0.2-cm cuvettes with vol-
ume capacities of 0.3 and 0.6 ml, respectively. If one needs to use
samples with high concentrations or buffers with high absorbance,
0.1-cm cuvettes are available.
For measurements in the near UV, for following changes in
aromatic groups of proteins, or visible regions, e.g., following the
CD of prosthetic groups such as flavins or hemes, higher concen-
trations of 1–2 mg/ml in a 1-cm cell are generally necessary,
depending on the number and asymmetry of the chromophores.
2.4 Instrumentation The following is a list of companies that currently produce com-
mercially available CD machines. All are adequate to perform ther-
mal denaturations and titration experiments. They are listed
alphabetically:
1. Applied Photophysics Ltd, 21 Mole Business Park, Leather-
head, Surrey KT22 7BA, UK.
2. Aviv Biomedical, 750 Vassar 750 Vassar Avenue—Suite 2 Lake-
wood, NJ 08701-6929, USA.
3. JASCO Inc., 28600 Mary’s Court, Easton, MD 21601, USA.
4. Olis, Inc., 130 Conway Drive Suites A, B & C, Bogart, GA
30622, USA.
3 Methods
3.1 Analyzing While changes at 222 or 218 nm can give an estimate of increases in
Changes in α-helical or β-structure content upon protein complex formation,
Conformation more precise estimates of changes in structure accompanying pro-
Accompanying tein-protein interactions can be made utilizing computer programs.
Protein-Protein Computer programs for analyzing CD spectra are not available
Interactions commercially, but many are available from their authors without
charge (see Note 1 for sources of software).
The use of the various methods to determine protein confor-
mation from CD spectra has been reviewed, with a detailed com-
parison of the advantages and disadvantages of each [5], so only a
brief description of each method is given below.
3.1.1 Constrained Constrained linear regression programs deconvolute CD spectra

Multilinear Regression into component spectra characteristic of specific secondary struc-
(LINCOMB) tures [9] by fitting the data to a set of reference spectra using the
methods of least squares. Standards include spectra of polypeptides
with known conformations [9, 16] or reference spectra deconvo-
luted from the spectra of proteins with known conformations by
the method of least squares [17, 18] or the convex constraint
algorithm [19, 20]. The sum of the fractions of each component
must equal 1. The LINCOMB program of Perczel et al. [19] uses
constrained linear regression to determine the percent α-helix, β-
sheet, and β-turns contributing to a spectrum. The method uses
invariant standards, so it is useful for quantifying changes in each
type of secondary structure when one protein is titrated with
another. The best method is to fit the mean residue ellipticities of
the individual proteins and the protein–protein complexes and
calculate the number of residues in each conformation by multi-
plying the fraction of each conformation by the number of residues.
One can then subtract the number of residues in each conformation
of the individual proteins from the number of residues in the
complex to estimate the total number of residues that undergo
conformational changes upon complex formation.
3.1.2 Nonconstrained Nonconstrained multilinear regression [16] analyzes the CD spec-

Multilinear Regression trum of a protein by fitting the spectrum to those of standards using
the method of least squares. The sum of the components is not
constrained to equal 100 %. The method is independent of the
intensity of the spectrum, so it is the only method that can be used
if the protein concentration isn’t known precisely. The method can
be used to analyze difference spectra, obtained when the sum of the
spectra of the unmixed components is subtracted from the spec-
trum of the mixture, because one doesn’t need to know the number
of residues that change conformation. The method is much less
CD Spectroscopy 245
Fig. 3 (a) Circular dichroism spectra of (open circles) E-Tmod1-130, (inverted filled triangles) AcTM1bZip,
(inverted open triangles) the sum of the spectra of E-Tmod1-130 and AcTM1bZip, and (filled circles) the
spectrum of the mixture E-Tmod1-130 and AcTM1bZip. Data are the average and standard error of four to
seven measurements. (b) The mean residue ellipticity of (open circles) Tmod11-130, (inverted filled triangles)
AcTM1bZip, and (filled circles) the mixture of the two peptides. The data are fit by (dashed dotted line)
constrained and (dotted line) nonconstrained least squares fits. (c) (Filled circles) Difference between the
spectrum of the mixture of the components and the sum of the unmixed components in (a), (Solid line)
nonconstrained least squares fit of the data. The difference data are the average of four measurements. The
data are from Greenfield and Fowler [21] used with permission from Elsevier
accurate than the constrained method, however, because one loses

the information of the intensities of each CD band.
Figure 3 illustrates the change in CD that occurs when a
peptide fragment of E-tropomodulin, now called tropomodulin 1
(E-Tmod1-130), binds to a chimeric two-chained coiled-coil peptide
containing the N-terminus of a short α-tropomyosin (AcTM1b-
Zip). Tmod1-130 contains residues 1–130 of chicken E-Tmod
plus 12 residues of a glutathione S-transferase linker [21]. The
conformations of the proteins were fit by both constrained and
nonconstrained multilinear regression using the spectrum of
α-tropomyosin [22] as the reference for a coiled-coil α-helix and
the data of Brahms and Brahms [16] for the spectra of peptides
in the β-pleated sheet, β-turn, and disordered conformations.
Figure 3a shows the raw spectra of the two unmixed peptides, the
sum of the spectra of the unmixed peptides, and the spectrum of
the complex. Figure 3b shows the mean residue ellipticities of the
individual proteins and the complex and the best fits using
constrained and nonconstrained multilinear regression. Figure 3c
shows the difference spectrum of the complex minus the sum of
unmixed peptides and the best fit to the curves using non-
multilinear regression with the same peptide references. The per-
centage of the conformations for the nonconstrained fits were
normalized to a sum of 100 %. The results are compared in Table 1.
The results for the TM1bZip peptide are also compared to the
structures determined by X-ray crystallography [24] and NMR
[25]. The number of residues of each protein fragment in each
conformation was calculated by multiplying the percentage of each
conformation by the number of residues of each protein fragment,
74 for AcTM1bZip, 142 for E-Tmod (Tmod11-130), and 216 for
the complex. When the tropomyosin peptide binds to the tropo-
modulin fragment, there is an increase in ellipticity of about 10 %
relative to the unmixed peptides. When the data in Fig. 3b is
analyzed using LINCOMB, the results suggested that most of the
change is due to an increase in helical content of about 22 residues,
~10 %, and a small increase in β-sheet residues, 3, ~1 %. When the
nonconstrained least squares fitting program MLR was used, the
results suggested that there was an almost equal increase in both
helical and β-sheet content of approximately 6 %, 14 and 13 resi-
dues, respectively. When the difference spectrum (Fig. 3c) was
analyzed directly, the results suggested that the increase in ellipticity
upon binding was 44 % due to an increase in helix and 55 % increase
in β-sheet. NMR studies of a similar E-Tmod peptide containing
residues 1–92 [23] show only that 12 residues [21–35] of tropo-
modulin are α-helical in the absence of tropomyosin peptides, but
residues 1–35 become ordered when tropomodulin binds to the
N-terminus of tropomyosin, an increase of 23 residues, in reason-
able agreement with the calculations from the CD data. Mutation
experiments suggest the TMod and tropomyosin peptides bind via
formation of a multichain α-helical coiled coil, so the constrained
least squares fits of the CD data are more accurate than the non-
constrained fits.
3.1.3 Ridge Regression CONTIN, developed by Provencher and Glöckner [26], fits the
(CONTIN) CD of unknown proteins by a linear combination of the spectra of a
large database of proteins with known conformations. In this
method, the contribution of each reference spectrum is kept
small, unless it contributes to a good agreement between the
theoretical best fit curve and the raw data. The method usually
gives very good estimates of β-sheets and turns and gives good
fits with polypeptides [5].
Table 1
Characterization of the secondary structural changes induced by AcTM1bZip-Tmod11-130 complex formation
Number of residues in each conformation
AcTm1bZip LINCOMB MLR X-ray NMR LINCOMB MLR X-raya NMRb,c

α-Helix 90.53 % 73.44 % 81.00 % 79.00 % 67.0 54.3 60 60
β-Sheet 9.47 % 26.56 % 0.00 % 0.00 % 7.0 19.7 0 0
β-Turn 7.36E-09 0.00 % 0.00 % 0.00 % 0.0 0.0 0 0
Remainder 8.08E-08 0.00 % 19.00 % 21.00 % 0.0 0.0 14 16
Total sum 100.00 % 100.00 % 100.00 % 100.00 % 74.0 74.0 74 76
E-Tmod1-130 LINCOMB MLR LINCOMB MLR
α-Helix 19.42 % 26.16 % 27.6 37.1
β-Sheet 18.66 % 8.50 % 26.5 12.1
β-turn 2.54E-10 0.00 % 0.0 0.0
Remainder 61.92 % 65.34 % 87.9 92.8
Total sum 100.00 % 100.00 % 142.0 142.0
Mixture LINCOMB MLR LINCOMB MLR
α-Helix 53.99 % 48.81 % 116.6 105.4
CD Spectroscopy
β-Sheet 16.72 % 20.56 % 36.1 44.4
β-turn 0.00 % 0.00 % 0.0 0.0
Remainder 29.29 % 30.63 % 63.3 66.2
Total sum 100.00 % 100.00 % 216.0 216.0
(continued)
247
Table 1
248
(continued)
Norma J. Greenfield
Number of residues in each conformation
AcTm1bZip LINCOMB MLR X-ray NMR LINCOMB MLR X-raya NMRb,c

Change in number of residues % Change in number of residues
LINCOMB MLR LINCOMB MLR
α-Helix 22.0 13.9 10.21 % 6.45 %
β-Sheet 2.6 12.7 1.21 % 5.88 %
β-Turn 0.0 0.0 0.00 % 0.00 %
Remainder 24.7 26.6 11.42 % 12.33 %
Total sum 0.0 0.0 0.00 % 0.00 %
MLR fit of difference spectra
Raw Normalized to 100 %
α-Helix 1.11E-04 44.04 %
β-Sheet 1.40E-04 55.96 %
β-Turn 5.64E-13 0.00 %
Remainder 4.77E-13 0.00 %
Total sum 2.51E-04 100.00 %
a
Meshcheryakov et al. [24]
b
Greenfield et al. [25]
c
The NMR studies were performed on a peptide were a glycine residue replaced the N-terminal acetyl group
CD Spectroscopy 249
3.1.4 Singular Value The application of singular value decomposition to determine the
Decomposition (SVD, secondary structure of proteins by comparing their spectra to the
VARSLC, CDSSTR, spectra of a large number of proteins with known structure was first
SELCON) developed by Hennessey and Johnson [14]. This method extracts
basis curves, with unique shapes, from a set of spectra of proteins with
known structures. Each basis curve is then related to a mixture of
secondary structures, which are then used to analyze the conforma-
tion of unknown proteins. The method is excellent for estimating the
α-helical content of proteins but is poor for sheets and turns unless
the data is collected to very low wavelengths, at least 184 nm [14].
Several newer programs have improved the method of singular
value decomposition by selecting references that have spectra that
closely match the protein of interest. They include VARSLC [15],
SELCON [27–31], and CDSSTR [32]. These give good estimates
of β-sheet and turns in proteins and work with data collected only
to 200 nm but give poor fits to polypeptides with high β-structure
content because reference sets for such compounds are not in their
database [5].
3.1.5 Neural Network A neural network is an artificial intelligence program which can
Programs (CDNN and K2D) detect patterns and correlations in data. Two widely used programs
are CDNN [33] and K2D [34]. A neural network is first trained
using a set of known proteins so that the input of the CD at each
wavelength results in the output of the correct secondary structure.
The trained network is then used to analyze unknown proteins. The
method works very well and the fits seem to be relatively indepen-
dent of the wavelength range that is analyzed [5].
3.1.6 Convex Constraint The CCA algorithm [19, 20, 35] deconvolutes a set of spectra into
Algorithm (CCA) basis spectra that when recombined generate the entire data set
with a minimum deviation between the original data set and recon-
structed curves. It is very useful for determining whether there are
intermediate states in thermal and denaturant induced unfolding.
The method has also been used to estimate protein conformation
but is poorer than least squares, SVD, or neural net analyses [5].
3.1.7 Recommendations 1. For determination of globular protein conformation in solu-

tion and evaluating the conformation of protein–protein com-
plexes: SELCON, CDSSTR, CDNN, CONTIN, and K2D.
2. For evaluating conformational changes upon protein-protein
interactions: LINCOMB
3. For deconvoluting sets of CD spectra, e.g., to follow the effects
of binding, denaturants, ligands, or changes in temperature on
protein and peptide conformation: the CCA algorithm.
Table 2 compares the results when the CONTIN, SELCON,
K2D, and MLR programs were used to analyze the spectra (Fig. 4)
Table 2
Secondary structure analysis of fragments and cleaved and uncleaved thioredoxin
Protein Method α-Helix β-Sheet Turns Other

b
N-terminal fragment K2D 6.0 29.0 65.0
CONTIN 4.0 14.0 10.0 72.0
MLRa 2.5 28.7 10.3 58.5
b
C-terminal fragment K2D 2.0 16.0 82.0
CONTIN 3.0 11.0 1.0 86.0
MLRa 0.0 15.7 0.0 84.3
b
Peptide complex K2D 30.0 20.0 50.0
CONTIN 20.0 40.0 7.0 34.0
MLRc,d 22.8 23.7 9.2 21.8, 22.5d
SELCON 24.66 21.5 16.9 22.5
b
Thioredoxin K2D 27.0 25.0 48.0
CONTIN 18.0 37.0 8.0 37.0
MLRc,d 31.8 22.9 7.2 16.9, 21.2d
SELCON 28.4 24.8 24.4 21.4
X-ray datae 40.0 28.3 14.0 17.8
X-ray dataf,g 33.3 27.8 9.3 26.8
X-ray datah 37.0 27.7 14.8 20.5
NMR datag,i 36.1 25.9 9.2 28.8
Data from Georgescu et al. [36] used with permission from the American Chemical Society
a
Polypeptide ref. [16]
b
Undetermined
c
Polypeptide ref. [19]
d
Estimate of aromatic and disulfide bond contributions
e
Structure calculations of Georgescu et al. [36] based on molecule A of the PDB file 2TRX [37]
f
Based on molecule A of the PDB file 2TRX [37]
g
Percentage of each structure calculated using MolMol [46]
h
Based on PDB file 1SRX [42]
i
Based on PDB file 1XOA [45]
Fig. 4 Circular dichroism of an artificial heterodimer of cleaved peptides of

thioredoxin from E. coli versus the intact protein. The protein concentration was
20 μM in KPi. Far-UV CD spectra of the isolated N- and C-fragments (open
circles) N; (filled circles), C; (open squares), cleaved; and (solid line) uncleaved
Trx. Data from Georgescu et al. [36] used with permission from the American
Chemical Society
CD Spectroscopy 251
of N- and C-terminal fragments of thioredoxin, the binary complex

of the two thioredoxin fragments, and the intact protein [36]. All of
the methods gave reasonable fits to the data. Note that the amount
of helix in the intact protein appeared to be underestimated by all of
the programs. This may have occurred because approximately 50 %
of the helix in thioredoxin is 3–10 helix rather than α-helix [37].
3.2 Determining When two proteins interact, there are often changes in either
Protein–Protein intrinsic or extrinsic circular dichroism. CD is a quantitative tech-
Association Constants nique and the change in CD, relative to the ellipticity of the
Using CD Data unmixed components, is directly proportional to the amount of
complex formed. The change can then be used to determine the
3.2.1 Determination of association or dissociation constant of the complex. Ideally, when
Binding Constants by Direct performing titrations, the concentration of the peptide being
Titration of One Protein by titrated, e.g., protein A, should remain constant. Thus, one should
Another at Constant titrate protein A with aliquots of a solution containing the starting
Temperature (Isothermal concentration of A and a large excess of protein B. In this case one
Titrations) does not have to correct for dilution of protein A when determining
the change in ellipticity upon complex formation.
When one titrates A with B to form a complex AB, the associa-
tion constant is K.
K ¼ ½AB=½A ½B ¼ ½AB=ðð½A o ½ABÞð½Bo ½ABÞÞ ð1Þ
where [Ao] and [Bo] are the initial concentrations of A and B and
[AB] is the amount of complex formed. The saturation fraction is s.
s ¼ ½AB=½Ao ð2Þ
If there is a change in the intrinsic or extrinsic CD accompanying

binding, the change, Δ[θ], is proportional to the amount of com-
plex formed.
Δ½θ ¼ ε½AB ð3Þ
where ε is the proportionality constant. When all of protein A is

bound by protein B,
Δ½θmax ¼ ε½Ao and s ¼ Δ½θ=Δ½θmax ¼ Δ½θ=ε½A o ð4Þ
Weak Binding Constants If the binding of one protein to the other is relatively weak, under
conditions where the dissociation constant, 1/k, of the complex is
more than 100-fold the concentration of A, one can assume that
the concentration of unbound B is approximately equal to the total
amount of B added to A. In this case it is easy to determine the
binding constant and one can use linear plots to determine Δ[θ]max
and KD.
For example one can fit the data to the Scatchard equation [38]
where
Δ½θ=½Ao ½B ¼ kðε Δ½θ=½Ao Þ ð5Þ
One plots Δ[θ]/[B] versus Δ[θ]. The y intercept ¼ kε[Ao] and the
slope ¼ k.
The Scatchard equation [38] described above works well for
studying interactions where there is one or multiple equivalent
binding sites of one protein for another. However, when there are
multiple sites, and the binding is cooperative, it is much more
difficult to estimate binding constants from spectroscopic data,
and usually only apparent constants can be estimated. The Hill
equation [39] may be used to fit the change in CD as a function
of titrant in a phenomenological fashion to give an estimate of the
apparent binding affinity. Here, the first protein A is titrated with a
second protein B, where [Bo] is the total concentration of added
protein giving an observed change in ellipticity Δ[θ].
Δ½θ ¼ Δ½θmax kh ½Bo h = 1 þ kh ½Bo h þ C ð6Þ
where k is the binding constant, [θ]max is the maximal changes in

ellipticity upon protein-protein interaction, h is a constant describ-
ing the apparent cooperativity of the interaction, and C is a constant
correcting for baseline offsets. When h ¼ 1, the binding is
noncooperative.
It should be noted that in the case of weak binding, titrations
are only feasible using data obtained at wavelengths where one
component has very low absorption and ellipticity, such as in the
case when an aromatic protein or a protein containing a heme is
titrated with a protein with low ellipticity in the aromatic (near UV)
or Soret band (visible) regions. These methods are very useful,
however, for measuring the interactions of small molecules, such
as chromophoric ligands or metals, to proteins.
Tight Binding Constants If the dissociation constant is close to the concentrations of the
proteins being studied, one cannot assume that the free concentra-
tion of the added protein, B, is the same as the total protein. One
must correct the concentration of the added protein B for the
amount that is bound. In this case [B] ¼ [Bo][AB]. From
Eq. 4, the fraction of protein A which is bound

½A bound ¼ ½A O Δ½θ=Δ½θmax ð7Þ
Therefore, the free concentration of B, [B] at any added concentra-

tion of Bo
CD Spectroscopy 253

½B ¼ ½Bo ½Ao Δ½θ=Δ½θmax ð8Þ
Solving for the change of circular dichroism, Δ[θ], at any added

concentration of protein B, [Bo],
n
Δ½θ ¼ Δ½θmax ðð1 þ k½Bo þ k½Ao Þ=2 k½A o Þ
o
ðð1 þ k½Bo þ k½A o Þ=2k½A o Þ2 ð½Bo =½Ao Þ1=2
ð9Þ
as described by Engel [40], who originally developed the equation

for determining binding constants of enzyme–ligand complexes
from fluorescence titrations.
Equation 9 can be fit by many available commercial curve
fitting programs, e.g., SigmaPlot, Microcal, and Origin among
others, that find the best fits to nonlinear equations. To use these
programs one inputs starting values of the parameters to be fit, here
ΔCDmax, and k, and has the program find the values of Amax and k
that best fit the data.
Figure 5 illustrates the binding of 130 residue fragments of
E- and Sk-Tropomodulins (now called tropomodulin 1 and 4,
respectively) to peptides containing the N-terminus of two tropo-
myosin isoforms [21], where the binding data were fit to Eq. 9.
Estimating Binding When two unfolded or partially unfolded proteins fold upon bind-
Constants When ing to each other, a simple method of determining the dissociation
Association of the Proteins constant is to follow the mean residue or molar ellipticity of a 1:1
Is Coupled with Folding mixture of the proteins as a function of concentration. In this case
h 1=2 i
kD þ k2D þ 4P t kD

½θobs ¼ ½θF þ ½θU ½θF =ð2P t Þ
ð10Þ
where kD is the dissociation constant of the two proteins, [θ]obs is

the ellipticity of the mixture of the proteins, at any total concentra-
tion, Pt, [θ]F is the ellipticity of the fully folded complex, and [θ]U is
the sum of the ellipticity of the fully unfolded individual proteins.
Figure 6 illustrates the use of this equation to determine the disso-
ciation constant of two fragments of the immunoglobulin binding
domain B1 of streptococcal protein G [41].
3.2.2 Estimating One can determine the thermodynamics of folding of

Protein–Protein Association protein–protein complexes by measuring the unfolding of the
Constants from the complex either by thermal denaturation or by chemical denatur-
Thermodynamics of ation using guanidine-HCl or urea. In some cases, the proteins are
Folding only folded when associated. In these cases, determining the ΔG of
unfolding directly gives the dissociation constants of the protein–
protein complexes. In other cases the proteins are both folded, but
Fig. 5 Increase in ellipticity at 222 nm, 30 C, when fragments of SK- and E-tropomodulin, (a) Sk-Tmod1-130
and (b) E-Tmod1-130 bind to chimeric peptides containing the N-termini of long and short α-tropomyosins,
(open circles) AcTM1aZip and (filled squares) AcTM1bZip, respectively [21]. 10 nmol of the TMZip peptides in
0.5 ml of 100 mM NaCl, 10 mM sodium phosphate, pH 6.5 in 2 mm pathlength cells were titrated with
concentrated solutions of the Tmod peptides. The ellipticity changes were corrected for the ellipticity
of the Tmod peptides alone and for dilution. The data were fit (solid line) to the equation
2 1=2
Δ½θ ¼ Δ½θmax 1 þ k ½Bo þ k ½Ao =2 k ½Ao 1 þ k ½Bo þ k ½Ao =2k ½Ao ½Bo =½Ao
using the Levenberg–Marquardt algorithm [43] implemented in the commercial program SigmaPlot to yield
dissociation constants for Sk-TMod1-130 for AcTM1aZip and AcTM1bZip of 1.2 0.9 and 0.3 0.3 μM,
respectively and for E-Tmod1-130 for AcTM1aZip and AcTM1bZip of 2.8 1.2 and 0.3 0.2 μM, respec-
tively. Data from Greenfield and Fowler [21] and used with permission from Elsevier
Fig. 6 Association of two complementary fragments of the immunoglobulin binding domain B1 of streptococcal
protein G, PGB1(1–40) and PGB1(41–56), evaluated by molecular ellipticity at 222 nm. Equimolar
concentrations of the fragments were dissolved in 50 mM phosphate buffer (pH 5.5) at 298 K. The filled
circles and solid line hshow observed data and besti curve fitted to the equation
0:5
k D þ k 2D þ 4P t k D

½θobs = ½θF þ ½θU ½θF =ð2P t Þ to determine the apparent dissociation
constant of the two fragments (Kapp ¼ 9 106 M). Data from Honda et al. [41]. Used with permission from the
American Chemical Society
CD Spectroscopy 255
complex formation increases their stability. In this case, the free

energy of binding can be determined by subtracting the free ener-
gies of unfolding of the individual proteins from the free energy of
unfolding of the complex. The difference, ΔΔG, can then be used
to calculate the dissociation constant of the complex.
Chemical Induced In the simplest case, two proteins A and B form a complex with 1:1
Unfolding stoichiometry. The isolated proteins are unfolded but fold to form
the complex. When this complex is treated with denaturants, the
complex (native form) dissociates to give the disordered proteins. At
any concentration of denaturant, [D], the association constant is:
k ¼ ½AB=½A ½B ¼ ½Pt α=ð½Pt ð1 αÞÞ2 ð11Þ
where [Pt] is the total concentration of complex and α is the

fraction folded. The fraction folded expressed in terms of the
protein concentration and association constant is:
h i1=2
2 2
2 4 2
= 2P 2t k

α ¼ 2P t k þ P t 2P t k P t 4P t k
ð12Þ
If one assumes that the binding constant is equal to the folding

constant, then ΔG can be calculated from the equation:
ΔG ¼ RT lnk ð13Þ
It has been shown for many systems that the ΔG of folding of a

protein is linearly dependent on the concentration of denaturant,
see, e.g., Santoro and Bolen [42].
ΔG D ¼ ΔG o þ m½D ð14Þ
where ΔGD is the free energy of folding in the presence of denaturant,

ΔGo is the free energy of folding in the absence of the denaturant, [D]
is the concentration of denaturant, and m is the slope.
To calculate the binding constant, one then studies the unfold-
ing of the protein complex by denaturants using several different
protein concentrations, by following the changes in molar or mean
residue ellipticity Δ[θ]obs as a function of denaturant concentration.
The fraction folded can be calculated at any concentration of dena-
turant using the equation

α ¼ ½θobs ½θU m U ½D = ½θF mF ½D ½θU m U ½D
ð15Þ
where [θ]U and [θ]F are the ellipticities of the fully unfolded and
folded complexes, respectively, and mU and mF are any necessary
linear baseline corrections for the change in ellipticity of the fully

unfolded and folded peptides with denaturant observed before and
after the folding transition.
The k at any denaturant concentration can be calculated using
Eqs. 11 and 15, and ΔG can be calculated from Eq. 13 and can be
plotted as a function of total denaturant concentration. By linear
extrapolation ΔG can be determined at zero denaturant, and the
dissociation constant KD for the protein–protein complex can then
be calculated by rearranging Eq. 13, KD ¼ exp(ΔGo/RT). One
also can fit the data directly using nonlinear regression techniques
[41]. Protocols for the direct fitting of data for the unfolding of a
heterodimeric peptide are given below (see Note 2). Georgescu et al.
[36] used this method to study the guanidine denaturation of the
complex of the two fragments of thioredoxin, shown in Fig. 7.
The free energy of folding of the protein complex was 10
0.4 kcal/mol, which agreed with the value determined by direct
titration of one protein with another of 9.8 0.2 kcal/mol.
Thermally Induced A change in either the intrinsic or extrinsic CD as a function of

Unfolding temperature can be used to determine the van’t Hoff enthalpy
of folding since changes in ellipticity are directly proportional to
the changes in concentration of the native and denatured forms.
Fig. 7 Chemical denaturation of cleaved Thioredoxin (Trx). Concentration depen-

dence of GnHCl-induced unfolding of cleaved Trx at (filled circles) 2.33 μM,
(open circles) 5 μM, (filled inverted triangles) 15 μM, and (open inverted
triangles) 25 μM in KPi at 20 C. The fraction of unfolded cleaved Trx was
calculated according to a two-state transition process using the intrinsic fluo-
rescence at 350 nm and ellipticity at 222 nm. The data represent the average of
three to five experiments. Redrawn from data in Georgescu et al. [36] and used
with permission from the American Chemical Society
CD Spectroscopy 257
When a protein is fully folded [θ]obs ¼ [θ]F and when it is fully

unfolded [θ]obs ¼ [θ]U. In the simplest case of a monomeric pro-
tein, the equilibrium constant of folding, k ¼ folded/unfolded. If
we define α as the fraction folded at a given temperature, T, then
k ¼ α=ð1 αÞ ð16Þ
ΔG ¼ RT lnk ð17Þ
where ΔG is the free energy of folding and R is the gas constant.

From the Gibbs–Helmholtz equation:

ΔG ¼ ΔH þ ðT T M ÞΔCp T ΔS þ ΔCpðlnðT =T M Þ ð18Þ
where ΔH is the van’t Hoff enthalpy of folding and ΔS is the

entropy of folding, TM is the observed midpoint of the thermal
transition, and ΔCp is the change in heat capacity for the transition.
At the TM, k ¼ 1, therefore ΔG ¼ 0 and ΔS ¼ ΔH/TM. Rearran-
ging these equations one obtains:
ΔG ¼ ΔH ð1 T =T M Þ ΔCp½ðT M T Þ þ T ðlnðT =T M ÞÞ

ð19Þ
k ¼ expðΔG=RT Þ ð20Þ
α ¼ k=ð1 þ kÞ ð21Þ

½θobs ¼ ½θF ½θU α þ ½θU ð22Þ
To calculate the values of ΔH and TM that best describe the folding

curve, initial values of ΔH, ΔCp, TM, [θ]F, and [θ]U are estimated,
and Eq. 22 is fitted to the experimentally observed values of the
change in ellipticity as a function of temperature, by a curve fitting
routine such as the Levenberg–Marquardt algorithm [43]. Since at
k ¼ 1, ΔG ¼ 0, the entropy of folding can therefore be calculated
using Eq. 18 where ΔS ¼ ΔH/TM. The free energy of folding at
any other temperature can then be calculated using Eq. 19. Similar
equations can be used to estimate the thermodynamics of folding of
proteins and peptides that undergo folded multimer to unfolded
monomer transitions (see Note 3). The association constant, k, of
protein–protein complexes can be obtained by subtracting the free
energies of folding of the sum of the uncomplexed proteins from
the free energy of folding of the complex and then k ¼ exp
(ΔΔG/nRT).
Figure 8 illustrates the increased stability when a fragment of
E-tropomodulin binds to a peptide containing the N-terminus of
Fig. 8 Thermal denaturation curves for (filled squares) E-Tmod1-130, (open

circles) AcTM1bZip, (open inverted triangles) the sum of the spectra of
E-Tmod1-130 and AcTM1bZip, and (filled diamonds) the spectrum of the
mixture E-Tmod1-130 and AcTM1bZip. The lines are the best fits of the data
assuming two-state transitions between folded and unfolded conformations. The
data were used to estimate dissociation constants for the complex of the
peptides of 0.2 0.2 μM. Data from Greenfield and Fowler [21] and used
with permission from Elsevier
an isoform of α-tropomyosin. The change in stability was used

to estimate a dissociation constant of 0.2 0.2 μM, which is in
experimental agreement with the direct titration method of
0.3 0.2 μM (duplicate measurements).
4 Notes
1. Sources of programs for determining secondary structure from

circular dichroism data:
Circular dichroism software can be downloaded from the
following sites:
(a) Variety of Microsoft DOS programs [5] including SEL-
CON, VARSLC, K2D, MLR, LINCOMB CONTIN, and
CCA algorithm are available from: http://www.nature.
com/nprot/journal/v1/n6/suppinfo/nprot.2006.202_
S1.html. These programs can be run under Windows 7 in a
DOS emulation program: http://www.dosbox.com/down
load.php?main¼1. In addition, equations for evaluating
constrained and nonconstrained least squares fits using
programs that use the Levenberg–Marquardt algorithm
[43] such as Sigmaplot are available at this site.
CD Spectroscopy 259
(b) DICHROPROT [47]: A variety of programs including

MLR, VARSLC, SELCON, and CONTIN in a Windows
environment: http://dicroprot-pbil.ibcp.fr/.
(c) CDPro [31]: Versions of CONTIN, SELCON, and
CDSSTR which calculate regular and distorted helices and
sheets: http://lamar.colostate.edu/~sreeram/CDPro/
main.html.
(d) Online Data Analysis is available at the Dichroweb web site:
http://dichroweb.cryst.bbk.ac.uk/html/home.shtml.
Programs include versions of CONTIN, SELCON,
CDSSTR, VARSLC, and K2D.
(e) The latest versions of the individual programs: CCA algo-
rithm [19, 20, 35]: A version that runs under 32 bit or
better versions of Windows is available from http://www.
chem.elte.hu/departments/jimre/.
CDNN [33]: CDNN is available from Applied Photophy-
sics on request. http://www.photophysics.com/tutorials/
cdnn-secondary-structure-analysis.
CDSSTR [32]: http://biochem.science.oregonstate.edu/
dr-johnsons-software-download-instructions (This pro-
gram must be run in a command box, CMD.exe, when
running Windows.)
CONTIN [26]: http://s-provencher.com/index.shtml
and http://lamar.colostate.edu/~sreeram/CD.
K2D [34]: http://www.embl-heidelberg.de/%7Eandrade/
k2d.html. Two updated versions named SOMCD [48] and
K2D3 [49] are available respectively from http://geneura.
ugr.es/cgi-bin/somcd/index.cgi and http://www.ogic.
ca/projects/k2d3/.
LINCOMB [19]: A version of Lincomb that runs under
most versions of Windows is available from http://www.
chem.elte.hu/departments/jimre/. (Note when using the
current version of the program, the reference set should be
in the form of a comma separated file with the wavelength
as the first file, separated by two commas from the columns
of the conformational references, e.g., nm, helix, beta
sheet, beta turn, random.)
SELCON [31]: http://lamar.colostate.edu/~sreeram/
CD.
CAPITO [50] is a new web server-based analysis and plot-
ting tool for circular dichroism data. It is available at http://
capito.nmr.fli-leibniz.de.
2. Protocol for fitting the unfolding of a heterodimeric protein
complex by denaturants from CD data.
One can fit the change in ellipticity, [θ]obs, as a function of
denaturant concentration, [D], to obtain the free energy of
folding of a protein complex and the binding affinity in the

absence of denaturant. The variables are [D] and [θ]obs. The
known constant is the protein concentration, [Pt]. The para-
meters to be fit are ΔGo, m, [θ]F, [θ]U, and mU and mF, where
ΔGo is the free energy of folding in the absence of denaturant,
m is the constant relating the linear change in free energy to the
concentration of denaturant, [θ]F and [θ]U are the ellipticities
of the fully folded and unfolded protein complexes, respec-
tively, and mF and mU are the slope describing any linear
changes in ellipticity of the folded and unfolded proteins before
and after the folding transition.
The equations that are used in the curve fitting procedure
for the unfolding of a dimer are:
ΔG ¼ ΔG o þ m½D
k ¼ expðΔG=1:987 T Þ
n 2 o
2 P 2t k þ P t sqrt 2P 2t k P t 4P t4 k2 = 2P 2t k

α ¼

½θobs ¼ α ½θF m F ½D ½θU mU ½D þ ½θU
In these equations, Pt is the total concentration of the pro-

tein, input as a constant, ΔG is the free energy of folding in the
presence of denaturant, T is the absolute temperature at which
the unfolding experiment is conducted, and 1.987 is the gas
constant in kcal/mol, k is the folding/association constant, and
α is the fraction folded. The difference between the calculated
change in ellipticity and the observed change in ellipticity is
minimized to give the best values of the fitting parameters
using a nonlinear least squares curve fitting programs. Such
curve fitting algorithms are usually included in most commer-
cially available graphics plotting programs, e.g., Kaleidagraph,
Origin, or Sigmaplot. If one knows the values of [θ]F, [θ]U, and
mU and mF, they may be input as constants rather than para-
meters to be determined. Once ΔGo is determined, the binding
constant is calculated using the equation k ¼ exp(ΔG/RT).
A curve fitting routine in SigmaPlot format is available: http://
www.nature.com/nprot/journal/v1/n6/suppinfo/nprot.
2006.229_S1.html.
3. Protocols for fitting circular dichroism data to determine the
thermodynamics of folding of monomers, dimers, heterodi-
mers, trimers, and heterotrimers from changes in CD as a
function of temperature using nonlinear least squares analysis
programs.
CD Spectroscopy 261
Unless otherwise noted, in all of these equations, ΔG is the

Gibbs free energy of folding, ΔH is the van’t Hoff enthalpy of
folding, ΔCp is the change in heat capacity going from the
folded to the unfolded form, T is the absolute temperature,
Kelvin, TM is the temperature where k ¼ 1, [θ]obs is the
observed ellipticity at any temperature T, α is the fraction
folded, [θ]F is the ellipticity of the fully folded protein and
[θ]U is the ellipticity of the unfolded protein, and p is the
concentration of the folded complex. The parameters to be fit
are ΔH, ΔCp, TM, [θ]F, and [θ]U. One inputs the protein
concentration, Pt, as a constant. For complicated fits, to reduce
the number of parameters, ΔCp may be set to 0 and the values
of [θ]F and [θ]U may be set as constants rather than parameters
to be minimized if they are known.
(a) Monomers
ΔG ¼ ΔH ð1 T =T M Þ ΔCp½ðT M T Þ þ T ðlnðT =T M ÞÞ
K ¼ expðΔG=RT Þ
α ¼ K =ð1 þ K Þ

½θobs ¼ ½θF ½θU α þ ½θU
(b) Homodimers
ΔG ¼ ΔH þ ðΔCpðT T m ÞÞ
T ððΔH =T m Þ þ ðΔCpðlnðT =T m ÞÞÞÞ
k ¼ expðΔG=ð1:987 T ÞÞ
a ¼ 4kP 2t
b ¼ 8kP 2t P t
c ¼ 4kP 2t
b sqrt b 2 4ac =2a

α ¼

½θobs ¼ α ½θF ½θU þ ½θU
(c) Heterodimers
Here the equations are set up in terms of unfolding and Pt is
the total concentration of protein in terms of monomers,
and the ΔH and ΔCp values are those of unfolding rather

than folding.

ΔG ¼ ΔH ð1 T =T m Þ þ Cp T T m T ðlnðT =T m ÞÞ
1:987T lnðP t =4Þ
kD ¼ expðΔG=1:987 T Þ
a ¼ 2k
b ¼ 2P t 2 k
c ¼ Pt
b sqrt b 2 4 ac =2a

α ¼

This expression is set up to evaluate the apparent Tm at

α ¼ 0.5. To evaluate it at k ¼ 1, the equation to calculate
the free energy of folding simplifies to:

ΔG ¼ ΔH ð1 T =T m Þ þ Cp T T m T ðlnðT =T m ÞÞ
Note that the equations give exactly the same TM and

enthalpy of folding if the unfolding data was analyzed as
if the complex was a homodimer, above.
(d) Homotrimers
k ¼ exp ΔH =ð1:987 T Þ½ðT =T m Þ 1Þ lnð:75ÞP t 2

z ¼ 3P 2t
q ¼ ðð3kz Þ þ 1Þ=ðkz Þ
e ¼ q3
d ¼ q3
n ð1=2Þ oð1=3Þ
½ð1Þðd=2Þ þ d 2 =4 þ e 3 =27

A ¼
n ð1=2Þ oð1=3Þ
B ¼ ð1Þ ðd=2Þ þ d 2 =4 þ e 3 =9

CD Spectroscopy 263
X ¼ Aþ B
α ¼ X þ 1

Note that in this treatment ΔCp is set at 0 and Tm is the

observed midpoint of the unfolding transition where
α ¼ 0.5.
(e) Heterotrimers
k ¼ exp ðΔH =ð1:987 T ÞðT =T m 1ÞÞ ln P t 2

Q ¼ 1= 12 P 2t k

R ¼ 1= 8 P 2t k

ð1=3Þ
A ¼ R þ sqrt R2 Q 3

B ¼ Q =A
y ¼ A þ B
α ¼ 1y

Note that in this treatment ΔCp is set at 0 and Tm is the

observed midpoint of the unfolding transition where
α ¼ 0.5. The equations give the same enthalpy and TM of
folding as the equations for analyzing the unfolding of a
homotrimer, above [21]. All of these routines are available
in SigmaPlot format from: http://www.nature.com/nprot/
journal/v1/n6/suppinfo/nprot.2006.204_S1.html.
References
1. Adler AJ, Greenfield NJ, Fasman GD (1973) 3. Johnson WC Jr (1990) Protein secondary
Circular dichroism and optical rotatory disper- structure and circular dichroism: a practical
sion of proteins and polypeptides. Methods guide. Proteins 7:205–214
Enzymol 27:675–735 4. Sreerama S, Woody RW (1995) Computation
2. Johnson WC Jr (1988) Secondary structure of and analysis of protein circular dichroism spec-
proteins through circular dichroism spectros- tra. Methods Enzymol 383(Part D):318–351
copy. Annu Rev Biophys Biophys Chem 5. Greenfield NJ (1996) Methods to estimate the
17:145–166 conformation of proteins and polypeptides
from circular dichroism data. Anal Biochem 19. Perczel A, Park K, Fasman GD (1992) Analysis
235:1–10 of the circular dichroism spectrum of proteins
6. Greenfield NJ (2006) Using circular dichroism using the convex constraint algorithm: a prac-
spectra to estimate protein secondary struc- tical guide. Anal Biochem 203:83–93
ture. Nat Protoc 1:2876–2890 20. Perczel A, Park K, Fasman GD (1992) Decon-
7. Greenfield NJ (2006) Using circular dichroism volution of the circular dichroism spectra of pro-
collected as a function of temperature to deter- teins: the circular dichroism spectra of the
mine the thermodynamics of protein unfolding antiparallel beta-sheet in proteins. Proteins
and binding interactions. Nat Protoc 13:57–69
1:2527–2535 21. Greenfield NJ, Fowler VM (2002) Tropomyo-
8. Greenfield NJ (2006) Determination of the sin requires an intact N-terminal coiled coil to
folding of proteins as a function of denaturants, interact with tropomodulin. Biophys J
osmolytes or ligands using circular dichroism. 82:2580–2591
Nat Protoc 1:2733–2741 22. Greenfield NJ, DeGregori H (1993) Confor-
9. Greenfield N, Fasman GD (1969) Computed mational intermediates in the folding of a
circular dichroism spectra for the evaluation of coiled-coil model peptide of the N-terminus
protein conformation. Biochemistry of tropomyosin and αα-tropomyosin. Protein
8:4108–4116 Sci 2:1263–1273
10. Joseph C, Stier G, O’Brien R, Politou AS, 23. Greenfield NJ, Kostyukova A, Hitchcock-
Atkinson RA, Bianco A, Ladbury JE, Martin DeGregori SE (2005) Structure and tropomy-
SR, Pastore A (2001) A structural characteriza- osin binding properties of the N-terminal cap-
tion of the interactions between titin Z- repeats ping domain of tropomodulin 1. Biophys J
and the alpha-actinin C-terminal domain. Bio- 88:372–383
chemistry 40:4957–4965 24. Meshcheryakov VA, Krieger I, Kostyukova AS,
11. Bothner B, Lewis WS, DiGiammarino EL, Samatey FA (2011) Structure of a tropomyosin
Weber JD, Bothner SJ, Kriwacki RW (2001) N-terminal fragment at 0.98 Å resolution.
Defining the molecular basis of Arf and Hdm2 Acta Crystallogr D Biol Crystallogr
interactions. J Mol Biol 314:263–277 67:822–825
12. Reed J, Kinzel V (1984) Near- and far-ultraviolet 25. Greenfield NJ, Huang YJ, Palm T, Swapna GV,
circular dichroism of the catalytic subunit of Monleon D, Montelione GT, Hitchcock-
adenosine cyclic 50 -monophosphate dependent DeGregori SE (2001) Solution NMR structure
protein kinase. Biochemistry 23:1357–1362 and folding dynamics of the N terminus of a rat
13. Michel B, Proudfoot AE, Wallace CJ, Bosshard non-muscle alpha-tropomyosin in an engi-
HR (1989) The cytochrome c oxidase- neered chimeric protein. J Mol Biol
cytochrome c complex: spectroscopic analysis 312:833–847
of conformational changes in the protein–pro- 26. Provencher SW, Glockner J (1981) Estimation
tein interaction domain. Biochemistry of globular protein secondary structure from
28:456–462 circular dichroism. Biochemistry 20:33–37
14. Hennessey JP Jr, Johnson WC Jr (1981) Infor- 27. Sreerama N, Woody RW (2000) Estimation of
mation content in the circular dichroism of protein secondary structure from circular
proteins. Biochemistry 20:1085–1094 dichroism spectra: comparison of CONTIN,
15. Manavalan P, Johnson WC Jr (1987) Variable SELCON, and CDSSTR methods with an
selection method improves the prediction of expanded reference set. Anal Biochem
protein secondary structure from circular 287:252–260
dichroism spectra. Anal Biochem 167:76–85 28. Sreerama N, Woody RW (1994) Poly(pro)II
16. Brahms S, Brahms J (1980) Determination of helices in globular proteins: identification and
protein secondary structure in solution by vac- circular dichroic analysis. Biochemistry
uum ultraviolet circular dichroism. J Mol Biol 33:10022–10025
138:149–178 29. Sreerama N, Woody RW (1994) Protein sec-
17. Saxena VP, Wetlaufer DB (1971) A new basis ondary structure from circular dichroism spec-
for interpreting the circular dichroic spectra of troscopy. Combining variable selection
proteins. Proc Natl Acad Sci U S A principle and cluster analysis with neural net-
68:969–972 work, ridge regression and self-consistent
methods. J Mol Biol 242:497–507
18. Chang CT, Wu C-SC, Yang JT (1978) Circular
dichroic analysis of protein conformation: 30. Sreerama N, Woody RW (1993) A self-
inclusion of β-turns. Anal Biochem 91:13–31 consistent method for the analysis of protein
CD Spectroscopy 265
secondary structure from circular dichroism. 41. Honda S, Kobayashi N, Munekata E, Uedaira H
Anal Biochem 209:32–44 (1999) Fragment reconstitution of a small pro-
31. Sreerama N, Woody RW (2000) Estimation of tein: folding energetics of the reconstituted
protein secondary structure from CD spectra: immunoglobulin binding domain B1 of strepto-
comparison of CONTIN, SELCON and coccal protein G. Biochemistry 38:1203–1213
CDSSTR methods with an expanded reference 42. Santoro MM, Bolen DW (1988) Unfolding free
set. Anal Biochem 282:252–260 energy changes determined by the linear extrap-
32. Johnson WC (1999) Analyzing protein circular olation method. 1. Unfolding of phenylmetha-
dichroism spectra for accurate secondary struc- nesulfonyl alpha-chymotrypsin using different
tures. Proteins 35:307–312 denaturants. Biochemistry 27:8063–8068
33. Bohm G, Muhr R, Jaenicke R (1992) Quanti- 43. Marquardt DW (1963) An algorithm for the
tative analysis of protein far UV circular dichro- estimation of non-linear parameters. J Soc
ism spectra by neural networks. Protein Eng Indust Appl Math 11:431–441
5:191–195 44. Holmgren A, Soderberg BO, Eklund H, Bran-
34. Andrade MA, Chacon P, Merelo JJ, Moran F den CI (1975) Three-dimensional structure of
(1993) Evaluation of secondary structure of Escherichia coli thioredoxin-S2 to 2.8 A reso-
proteins from UV circular dichroism spectra lution. Proc Natl Acad Sci U S A
using an unsupervised learning neural network. 72:2305–2309
Protein Eng 6:383–390 45. Jeng MF, Campbell AP, Begley T, Holmgren
35. Perczel A, Hollosi M, Tusnady G, Fasman GD A, Case DA, Wright PE, Dyson HJ (1994)
(1991) Convex constraint analysis: a natural High-resolution solution structures of oxi-
deconvolution of circular dichroism curves of dized and reduced Escherichia coli thiore-
proteins. Protein Eng 4:669–679 doxin. Structure 2:853–868
36. Georgescu RE, Braswell EH, Zhu D, Tasayco 46. Koradi R, Billeter M, W€ uthrich K (1996)
ML (1999) Energetics of assembling an artifi- MOLMOL: a program for display and analysis
cial heterodimer with an alpha/beta motif: of macromolecular structures. J Mol Graph 14
cleaved versus uncleaved Escherichia coli thior- (51–5):29–32
edoxin. Biochemistry 38:13355–13366 47. Deléage G, Geourjon C (1993) An interactive
37. Katti SK, LeMaster DM, Eklund H (1990) graphic program for calculating the secondary
Crystal structure of thioredoxin from Escher- structure content of proteins from circular
ichia coli at 1.68 A resolution. J Mol Biol dichroism spectrum. Comput Appl Biosci
212:167–184 9:197–199
38. Scatchard G (1949) The attractions of proteins 48. Unneberg P, Merelo JJ, Chacón P, Morán F,
for small molecules and ions. Ann NY Acad Sci SOMCD (2001) Method for evaluating pro-
51:660–672 tein secondary structure from UV circular
39. Hill AV (1910) The possible effects of the dichroism spectra. Proteins 42:460–470
aggregation of the molecules of haemoglobin 49. Louis-Jeune C, Andrade MA, Perez-Iratxetal C
on its dissociation curves. J Physiol (Lond) 40: (2012) Prediction of protein secondary struc-
iv–vii ture from circular dichroism using theoretically
40. Engel G (1974) Estimation of binding para- derived spectra. Proteins 80:374–381
meters of enzyme-ligand complex from fluoro- 50. Wiedemann C, Bellstedt P, Görlach M (2013)
metric data by a curve fitting procedure: seryl- CAPITO – a web server-based analysis and
tRNA synthetase-tRNA Ser complex. Anal Bio- plotting tool for circular dichroism data. Bioin-
chem 61:184–191 formatics 29:1750–1757
Chapter 16
Protein-Protein Interaction Analysis by Nuclear Magnetic

Resonance Spectroscopy
Peter M. Thompson, Moriah R. Beck, and Sharon L. Campbell
Abstract
Nuclear magnetic resonance (NMR) has continued to evolve as a powerful method, with an increase in the
number of pulse sequences and techniques available to study protein-protein interactions. In this chapter, a
straightforward method to map a protein–protein interface and design a structural model is described,
using chemical shift perturbation, paramagnetic relaxation enhancement, and data-driven docking.
Key words Nuclear magnetic resonance (NMR), Protein-protein interaction, Chemical-shift pertur-
bation (CSP), Paramagnetic relaxation enhancement (PRE), Docking
1 Introduction
Nuclear magnetic resonance (NMR) continues to be a valuable

technique for scientists wishing to study protein-protein interac-
tions. These noncovalent interactions drive much of biology
through their presence in signaling pathways, molecular complexes,
biomolecular processing, and protein degradation. NMR and X-ray
crystallography are the two most common techniques for deter-
mining the structure of proteins, RNA, and DNA at a high resolu-
tion. However, NMR can be used for more than structure
determination. NMR provides, in a site-specific manner, informa-
tion on conformational dynamics of proteins and nucleic acids in
solution, the protonation state of functional groups, and, of rele-
vance to this article, both the identity of atoms involved in pro-
tein–ligand interactions and the kinetics of said interaction.
NMR is a unique tool for the study of protein-protein interac-
tions primarily because it allows for the study of these interactions
in physiological or near-physiological conditions. Additionally,
NMR provides detailed, atomic-level information for even weak
interactions (Kd > 100 μM). Furthermore, NMR is a very flexible
technique, with many different pulse sequences and variations in
267
268 Peter M. Thompson et al.
sample preparation to obtain information. Techniques now exist to

study complexes in solution as large as 80–90 kDa. A complete
discussion of these techniques and pulse sequences is outside the
scope of this chapter, but the reader is directed to many reviews that
discuss these in greater detail [1–7]. The techniques discussed here
are chemical shift perturbation (CSP), paramagnetic relaxation
enhancement (PRE), and molecular docking.
NMR is based upon the ability of magnetically active nuclei,
when placed in a strong magnetic field, to absorb electromagnetic
radiation at distinct frequencies. These frequencies are determined
by the chemical environment of each nucleus and are affected by (1)
atoms bound to the nucleus detected, (2) atoms within 5 Å of the
nucleus (through space), and (3) the dynamics of the environment
(how fast and how drastically it changes). The diversity of chemical
environments within the protein results in dispersion of the fre-
quencies absorbed by the nuclei, giving rise to unique signals
(chemical shifts) for each nucleus. A variety of pulse sequences
exist to assign the nucleus associated with each signal [8–10].
Once the chemical shifts of each peak have been assigned, changes
in either location or intensity of a specific peak can be associated to a
change in the chemical environment of the corresponding nucleus.
This is the principle behind chemical shift perturbation (CSP),
where binding of a ligand to the observed protein can/may change
the chemical environment for nuclei at the interface. Thus, changes
in the chemical shift or intensity of a peak upon titration of the
binding partner suggest that the observed nucleus is near the site of
interaction. This does not always hold true, as some proteins
undergo a conformational change associated with binding. There-
fore changes in chemical shift or intensity may be the result of the
conformational change at a greater distance, instead of direct con-
tact with the ligand.
The specific changes observed during a CSP experiment
depend upon the rate of exchange between the free and bound
states. If the exchange rate is slow on the NMR time scale (k < δν,
where k is the rate of exchange and δν is the change in chemical shift
for the nucleus between the two states), then as the ligand is
titrated in, the signal for a specific nucleus will be split into two
separate peaks, one for the free state and one for the bound state.
The relative intensities of the two peaks provide information on the
relative populations of the protein. However, if the exchange rate is
fast on the NMR time scale (k > δν), only one peak will be
observed. In this case, the chemical shift for the complex will be
an average of the chemical shifts of the free and bound states,
weighted by their respective populations. The third type of
exchange is intermediate (k δν). In this case, as the ligand is
titrated, the signal broadens out and eventually disappears. These
exchange regimes, slow, intermediate, and fast, generally tend to
correspond with tight (Kd < 1 μM), intermediate (1 μM < Kd
Mapping Protein-Protein Interactions by NMR 269
<100 μM), or weak binding (Kd > 100 μM), assuming diffusion-
limited association rates. Therefore, CSP is most useful when the
binding interaction is either in slow or fast exchange.
PRE is another NMR technique that involves changing the
chemical environment of nuclei involved in an interaction. Para-
magnetic ions have a free, unpaired electron and therefore a net
spin, which generates a magnetic field [11]. When the paramag-
netic ion is allowed to rotate freely (isotropically), it exposes nearby
nuclei to a constantly changing magnetic field. This increases the
rate of relaxation of the signal from the nucleus and is observed as a
decreased intensity of the signal and a significantly increased line-
width. This relaxation enhancement is related to the distance
between the ion and the magnetically susceptible nucleus by a
factor of 1/r6, where r is the distance between the two atoms.
Thus, if a paramagnetic ion is attached to a ligand, then nuclei in
the protein near the ligand-binding site should have increased
relaxation rates. More complex experiments can be run to measure
the rate of relaxation and determine the distance from the paramag-
netic ion, but it is more common to use PRE to identify nuclei with
decreased signal intensity and define them as being close to the
paramagnetic ion when the ligand is bound [12].
The final method we discuss is NMR data-driven docking,
which couples computational methods with limited NMR
restraints. When NMR restraints are too limited for structural
determination or a high-resolution structure is not critical, docking
procedures can be used to develop a testable ligand-binding model.
There are many available docking programs, each with its own
distinct methods for performing docking and scoring its structures
[13–18]. We used the program HADDOCK, which is designed to
allow for straightforward input of restraints from NMR data, as well
as other methods [13]. HADDOCK refers to these inputs as
Ambiguous Interaction Restraints (AIRs) and relies on them to
generate docked structures from the structures of the individual
binding partners. The structures are then scored, with those scor-
ing the highest being selected for semi-flexible simulated annealing.
Next, a four-step, iterative annealing process that allows for the
movement of amino acid residues at the interaction interface is
performed. Finally, the structures are refined in explicit solvent
(water), clustered into groups, and scored. The scoring function
is based upon physical parameters and experimental restraints, and
can be customized for each project.
In this chapter, we show how a combination of CSP, PRE, and
docking were used to generate a model for an interaction between
an EF-hand domain of α-actinin-2 and a short region of the protein
palladin [19]. These two proteins are involved in regulation of the
actin cytoskeleton and have been implicated in cancer. A structural
model of the interaction between the two provides a starting point
for understanding the functional role of the interaction. While
other restraints were used for model generation (including residual

dipolar couplings (RDCs) and mutagenesis), we are presenting
these three techniques because they are some of the most common
and straightforward NMR methods to study protein-protein
interactions.
2 Materials
15
1. N-enriched human α-actinin 2 EF-hand residues 823–894
(referred to as Act-EF34, numbered 1–73 herein).
15
2. NH4Cl.
3. Unlabeled palladin peptide (residues 235–252 of human palla-
din, numbered 1–19 herein).
4. Palladin peptide with an N-terminal cysteine labeled with
S-(2,2,5,5-tetramethyl-2,5-dihydro-1H-pyrrol-3-yl)methyl
methanesulfonothioate (MTSL).
5. Varian Inova 600 or 700 MHz NMR spectrometer.
6. 5.0 mm Wilmad standard NMR sample tube.
7. NMRPipe/NMRDraw (Delaglio, NIH).
8. NMRViewJ (One Moon Scientific, Inc.).
9. NMR Buffer: 20 mM 3-(N-morpholino)propanesulfonic acid,
pH 6.6, 10 mM NaCl, 2 mM tris(2-carboxyethyl)phosphine,
0.01 % NaN3, and 10 % D2O.
10. 1 M ascorbic acid.
3 Methods
3.1 NMR Sample Act-EF34 was expressed as a Z-tagged, histidine-tagged

Preparation (Chapter 23) fusion protein and purified by affinity chromatogra-
phy [20]. Uniform 15N-enrichment was obtained by growing
Escherichia coli in M9 media (see Note 1) with 15NH4Cl as the
sole nitrogen source. NMR samples were prepared in NMR buffer
with an Act-EF34 concentration of 0.22 mM and a variable peptide
concentration described in Subheadings 3.3 and 3.4. The NMR
sample volume was 600 μL in a 5 mm OD precision tube, standard
wall (see Note 2).
3.2 Spectroscopy NMR experiments were performed on a Varian Inova 700-MHz

NMR spectrometer with a triple-resonance cryoprobe with z-axis
pulsed-field gradients at 27 C. All experiments described here are
two-dimensional 1H–15N heteronuclear single quantum coherence
spectroscopy (HSQC) experiments [21]. Data were acquired with
2,048 512 complex data points and a 1H spectral width of
10,000 Hz and a 15N spectral width of 2,000 Hz (see Note 3).

Processing and analysis of data were done with NMRPipe,
NMRDraw [22], and NMRViewJ.
3.3 Chemical Shift Proton and 15N resonance assignments of apo Act-EF34 were
Perturbation obtained from the Biological Magnetic Resonance Bank (BMRB),
entry 17627. Two-dimensional 1H–15N HSQC data were acquired
on the Act-EF34:palladin peptide (see Note 4) complex at molar
ratios of 1:0, 1:0.1, 1:0.5, 1:1, 1:2, 1:3, 1:4, and 1:5, with 220 μM
Act-EF34. At a ratio of 1:5, complete saturation of binding was
achieved as peaks no longer shifted. To prevent dilution effects,
samples were made by mixing the 1:0 and 1:5 samples in the
appropriate ratio. The Kd of the interaction, determined by iso-
thermal titration calorimetry (Part II Chapter 11) to be ~16 μM,
suggested that exchange between the free and bound state would
likely be intermediate or fast on the NMR time scale (see Note 5).
In this system, we see exchange on the fast time scale, which results
in a single peak for each nucleus with a chemical shift that is the
average of the weighted population of the free- and bound-state
chemical shifts.
3.4 Paramagnetic An NMR sample with 220 μM Act-EF34 and 1.1 mM MTSL-
Relaxation labeled palladin peptide was created with a final volume of
600 μL. MTSL-labeled peptide was generated by incubation of
MTSL and the palladin peptide with an N-terminal cysteine with
an MTSL:peptide ratio of 4:1 in a nonreducing buffer [19]. This
allows for formation of a disulfide bond between the MTSL label
and the thiol group on the cysteine. Then excess, unbound MTSL
label was removed by extensive dialysis into nonreducing NMR
buffer. A 2D 1H–15N HSQC was acquired on the sample, using
the same parameters as those given above. Then 2 μL of 1 M
ascorbic acid was added to the sample to reduce the MTSL label,
allowing it to diffuse into the solution, and a duplicate 1H–15N
HSQC was acquired on this sample.
3.5 NMR Data Both the chemical shift perturbation and paramagnetic relaxation
Analysis studies relied on changes in the chemical shifts or peak intensities,
respectively, as observed by 2D 1H–15N HSQC analyses. Addition-
ally, in both experiments, 15N-labeled Act-EF34 was used, meaning
that the signals observed come only from Act-EF34. The 2D
1
H–15N HSQC provides information on all 15N atoms bound to
a 1H. This includes one NH peak for the backbone amide, which
we expect for all residues except proline, giving a site-specific probe
for each residue. Side-chain amides also provide peaks, though their
assignment is more difficult. The signal from the proton is in the
first dimension, while the signal from the 15N nucleus is in the
second. The peaks in the resulting spectrum appear as spots with
contour lines to represent their height or intensity. Both the
Fig. 1 Expanded region of a 2D 1H–15N HSQC spectral overlay showing NH peaks

associated with 0.22 mM Act-EF34 in the presence of varied concentrations of
WT palladin peptide. The Act-EF34 is uniformly enriched with 15N, while the
peptide is unlabeled; therefore peaks correspond to changes in chemical shift of
Act-EF34 upon binding palladin peptide. Peptide concentrations are 0 (black), 1
(purple), 10 (blue), 50 (green), 100 (yellow ), 200 (orange), and 500 (red ) μM.
Residues S13, L33, and K43 show significant changes in chemical shift upon
titration with WT palladin peptide (yellow arrows), while residues L29 and R31 do
not (green arrows). Residue L67 shows a very significant change in chemical
shift (red arrow, greater than average CS + twice standard deviation)
amplitude and the chemical shift position of the signal provide

information on the corresponding residue within Act-EF34. For
the chemical shift perturbation titration, the peaks for the apo (or
free) Act-EF34 had previously been assigned to their respective
residues (BMRB ID:17626). The HSQC spectra collected during
the titration were superimposed on each other using NMRViewJ
(see Fig. 1). To select shifts that are significantly large for use in
docking, a weighted chemical shift that incorporates both dimen-
sions was calculated. The calculation was performed as follows:
sffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Nfree Nbound 2
CSresidue ¼ þ ðHfree Hbound Þ2
10
where Nfree is the chemical shift of a peak in the nitrogen dimension
in the free, or apo, state (see Note 6). The 15N dimension is
weighted at 1/10 of the proton dimension, because the absolute
value for the gyromagnetic ratio for 15N is approximately 1/10 that
of 1H. If CSresidue > 0.173 ppm (average CS + half the standard
deviation), then the residue was deemed to have shifted signifi-
cantly enough to be used as a docking restraint (see Fig. 2a).
Fig. 2 Structural model of Act-EF34 complexed with WT palladin peptide. (a) CSP and PRE data mapped onto
the primary structure of Act-EF34. CSP data are shown in black, with PRE data in gray. The secondary
structure (from PDB 1H8B) of Act-EF34 is given by black bars (α-helices) and white bars (β-sheets).
Figure modified from Beck et al. [19] (b) Model structure of Act-EF34 when bound to WT palladin peptide
as created by PyMOL [28]. Sidechains shown as sticks belong to residues selected as AIRs for docking with
HADDOCK. Residues colored in pink showed significant relaxation when bound to the MTSL-labeled palladin
peptide. (c) Model of the ligand-bound complex. The color scheme is maintained, and the palladin peptide is
shown in blue. The sidechains of L9 and L12 are represented as sticks. The N-terminus (where the MTSL label
was added) is shown as spheres
Additional considerations for docking restraints were made on the

basis of solvent exposure, chemical shift perturbations of the Cα
and Cβ atoms, and absence of peaks upon palladin peptide titration.
If a protein-protein interaction does not involve significant confor-
mational rearrangement, then atoms involved in the interaction
should be exposed to the solvent and available for mediating the

interaction. Solvent exposure was determined by running NAC-
CESS, a program that defines the solvent accessibility of atoms
based on the known protein structure in a PDB file [23]. Residues
chosen for AIRs had atoms with a solvent accessibility greater than
50 % [23]. Chemical shift perturbations for the Cα and Cβ atoms
were determined during assignment of peaks for the palladin-bound
Act-EF34; residues with a CSresidue > 1.2 ppm (average CS + half
the standard deviation) were chosen as AIRs. As with the CSP for
HN peaks described above, a CSresidue > 1.2 ppm suggested
involvement in the binding interaction. Additionally, some HN
peaks disappeared during titration, either due to line broadening
(exchange on the intermediate time scale) or CSP that could not be
easily tracked. Residues that show solvent accessibility and fit one or
more of the above criteria are Q9, I11, S13, F14, I16, L17, E32,
L33, D36, Q37, C41, Y68, and G69 (Fig. 2b). Additionally, many
of the peaks that showed significant Act-EF23 CSP upon binding to
the palladin fragment showed similar CSP upon binding of Act-
EF34 to a titin peptide, suggesting that Act-EF34 bound titin and
palladin in a similar fashion [19].
To analyze the paramagnetic relaxation data, the reduced (dia-
magnetic probe) and oxidized (paramagnetic probe) HSQCs were
superimposed. No changes in chemical shift were observed, which
suggests that the presence of the MTSL label did not affect the
binding of the peptide to Act-EF34. The intensities of the peaks
were calculated using NMRViewJ, and a difference in intensities
(heights) for each peak was determined:
PREresdiue ¼ Iresidue, oxidized Iresidue, reduced
The paramagnetic probe increases the rate of magnetic relaxation

for nearby atoms, therefore peaks that exhibit a significantly lower
RI are likely close to the paramagnetic probe. Peaks corresponding
to residues M1, A2, D3, T4, D5, S13, A26, K43, S71, D72, and
L73 all exhibited a significant decrease in intensity in the presence
of the paramagnetic probe, suggesting that these residues are near
the N-terminus of the palladin peptide when it is bound to Act-
EF34 (Fig. 2). These results were not used to generate AIRs for
HADDOCK, but were used as an independent confirmation of the
HADDOCK model generated (see below).
3.6 Docking The docking program HADDOCK [13] was used to develop a
model for the interaction between the palladin peptide and Act-
EF34. HADDOCK starts with known structures of the two ligands
and docks using experimental restraints and a traditional energy
function; it does allow for small conformational changes, a com-
mon occurrence in protein-protein interactions. The starting struc-
ture for Act-EF34 was taken from the solution structure of
Act-EF34 bound to the seventh Z-repeat of titin (PDB 1H8B).

This choice was justified by the similarity of CSP data in binding of
the titin peptide and the palladin peptide [19]. Further validation
for using this starting structure came from 1H–15N RDCs on the
peptide-bound Act-EF34, which agreed very well with those calcu-
lated for the Act-EF34:titin structure (see Note 7). The starting
structure for the palladin peptide was designed in MacPyMOL as an
alpha helix. Circular dichroism data and the PSIPRED protein
structure prediction server both suggested that the peptide was
helical when bound [19]. Ambiguous interaction restraints
(AIRs) for the docking interaction came from CSP and mutagenesis
of leucines L9 and L12 in the palladin peptide, which resulted in
loss of binding. Additional residues surrounding L9 and L12 on the
palladin peptide were also defined by HADDOCK as AIRs. Dock-
ing was performed on the HADDOCK server website using the
default parameters [24]. HADDOCK generated 1,000 initial struc-
tures, from which the 200 with the best score were selected. Scores
are generated using a weighted sum of energy values corresponding
to calculated van der Waals forces, electrostatic forces, solvation,
and the AIRs. The highest-scoring models underwent an annealing
stage with some flexibility and subsequent refinement in water. The
final models were then clustered into groups of no fewer than 4
with clustering cutoff of 7.5 Å. The clustering cutoff ensures that
two distinct clusters of models are grouped into one larger cluster.
One cluster of 12 model structures had a significantly better score
than all other clusters and a root mean square deviation in the Act-
EF34 structure from PDB 1H8B (Act-EF34 bound to titin) of only
0.76 Å. This model was selected as the best model for the interac-
tion between Act-EF34 and palladin (Fig. 2) and is further testable
by mutagenesis. Model-driven mutagenesis can lead to refinement
of the model and the development of specific tools (mutants or
variants) to study the functional relevance of a protein-protein
interaction in vivo.
4 Notes
1. Expression of proteins in M9 minimal medium can be much

more difficult than expression in LB. The reader is encouraged
to work test expression in M9 without isotopic enrichment
before expressing with isotopic enrichment. Common techni-
ques to improve growth and expression in M9 medium include
using an overnight growth that is 50 % H2O, 50 % LB, dou-
bling the concentration of glucose in the M9 medium, growing
in a baffled flask to increase aeration, and supplementing the
medium with a vitamin mixture [25]. Other techniques that are
also used to improve expression in LB, including expression at
lower temperatures or use of a fermentor, may be useful.
2. A number of considerations should be made when preparing an

NMR sample. As NMR is insensitive by nature, a large amount
of soluble protein is required. Most NMR experiments use
samples with protein concentrations between 0.1 and 2 mM,
with higher concentrations generally preferred. To improve the
ratio of signal-to-noise (S/N), NMR signals are frequently
averaged multiple times. The S/N increases by the square
root of the number of times the signal is averaged, but increases
linearly with the protein concentration. A doubling in concen-
tration therefore saves one four times the amount of acquisi-
tion time. Therefore, the NMR buffer should optimize protein
solubility while maintaining low ionic strength, especially if
using a magnet with a cryoprobe. Generally, a buffer with a
pH between 5.5 and 7 is optimal. Many pulse sequences rely on
the amide proton in the backbone as a starting point, and at
higher pH, these protons exhibit increased exchange and
become increasingly difficult to observe. Having sufficient sam-
ple volume is critical to obtaining quality spectra. An ideal
volume is 500 μL for the Wilmad tube (or 350 μL for a Shigemi
tube). Anything less makes shimming of the sample difficult
and leads to poor water suppression and a decrease in S/N.
Increasing sample volume is generally a poor idea, as the vol-
ume “seen” by the spectrometer is fixed.
3. For a 2D 1H–15N HSQC, determining the number of complex
points and spectral width in the 15N dimension is important. In
the 1H dimension, the signal is directly detected, so the num-
ber of complex points depends solely upon the number of times
the signal is collected, and the spectral width is usually set
sufficiently wide to capture all signals (backbone amide protons
generally fall between 11 and 6 ppm). However, as the 15N
dimension is indirectly detected, these values should be opti-
mized. A spectral width that is too large results in poorer
resolution in the 15N dimension, or requires more points
(and time) to achieve the same resolution. When starting
NMR experiments, one should first carry out an HSQC with
a large spectral window to see where the data fall in the 15N
dimension. Subsequent HSQC experiments should be per-
formed with the 15N offset centered in the middle of the data
and a smaller spectral width. If one or two peaks in the HSQC
lie sufficiently far away from other peaks in the 15N dimension,
these peaks can be “folded” to further decrease the spectral
width, though this is often reserved for 3D experiments. Once
an optimal spectral width is determined, the minimum number
of complex points required is equal to the spectral width
divided by the linewidth of the peaks in the 15N dimension.
The linewidth is related to the size of the protein (larger
proteins have larger linewidths) and can be determined from
an HSQC with a significantly large number of complex points

in the 15N dimension. A good estimate for the line width of a
protein is approximately 0.6 Hz per kDa molecular weight.
4. With many proteins, expression and purification of the protein
(or the domain of interest) at levels sufficient for NMR experi-
ments is very difficult. Additionally, some protein-protein
interactions are mediated primarily by a small part of one
protein. In these cases, a small peptide can be used to study
the interaction between proteins. Small peptides provide their
own challenges, especially in determination of their concentra-
tion [26]. Solutions of peptide at high concentration often
need pH adjustments.
5. Exchange between the bound and free states of a protein takes
place in one of three NMR time scales: fast, intermediate, or
slow. If the interaction is in intermediate exchange, the signal
from the bound state will shrink in amplitude and broaden in
linewidth, resulting in a disappearance of signal as the popula-
tion shifts to the bound state. This is the least optimal exchange
regime for studying protein-protein interactions by NMR, as it
makes quantification of the relative populations difficult.
Changing the temperature, buffer conditions, or the nature
of the ligand can push the interaction into the fast or slow
regimes.
6. The most commonly accepted formula for calculating weighted
chemical shift differences is shown below, as reported by
Mulder et al. [27].
sffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Nfree Nbound 2
CSresidue ¼ þ ðHfree Hbound Þ2
6:5
This formula uses a different weighting factor for 15N chemical

shift differences, which is based upon the disparity in the aver-
age dispersion of chemical shift values for 15N and 1H signals in
a 2D 1H–15N HSQC.
7. A complete description of RDCs and their use in studying
protein-protein interactions is outside the scope of this chapter.
The reader is referred to reviews on the subject for a more
detailed experimental and theoretical discussion [5, 6]. RDCs
can be very useful for the study of protein-protein interactions.
Simply put, they describe the relative orientation of bond
vectors within a protein and can be used to compare structures.
As described in Beck et al. [19], RDCs can be used to compare
a known structure to an unknown one to determine a degree of
similarity. Additionally, RDCs for the ligand-free and ligand-
bound states of a protein can be compared to test for large
conformational changes that sometimes occur upon binding.
References
1. Akke M (2002) NMR methods for characteriz- approach based on biochemical or biophysical
ing microsecond to millisecond dynamics in information. J Am Chem Soc 125:1731–1737
recognition and catalysis. Curr Opin Struct 14. Lyskov S, Gray JJ (2008) The RosettaDock
Biol 12:642–647 server for local protein–protein docking.
2. Clore G, Gronenborn A (1991) Two-, three-, Nucleic Acids Res 36:W233–W238
and four-dimensional NMR methods for 15. Claussen H, Buning C, Rarey M et al (2001)
obtaining larger and more precise three- FlexE: efficient molecular docking considering
dimensional structures of proteins in solution. protein structure variations. J Mol Biol 308:
Annu Rev Biophys Biophys Chem 20:29–63 377–395
3. Kay LE (1997) NMR methods for the study of 16. Tovchigrechko A, Vakser IA (2006) GRAMM-
protein structure and dynamics. Biochem Cell X public web server for protein–protein dock-
Biol 75:1–15 ing. Nucleic Acids Res 34:W310–W314
4. Vinogradova O, Qin J (2012) NMR as a 17. Pierce BG, Hourai Y, Weng Z (2011) Acceler-
unique tool in assessment and complex deter- ating protein docking in ZDOCK using an
mination of weak protein–protein interactions. advanced 3D convolution library. PLoS One
Top Curr Chem 326:35–45 6:e24657
5. O’Connell MR, Gamsjaeger R, Mackay JP 18. Morris GM, Huey R, Lindstrom W et al (2009)
(2009) The structural analysis of protein–pro- AutoDock4 and AutoDockTools4: automated
tein interactions by NMR spectroscopy. Prote- docking with selective receptor flexibility.
omics 9:5224–5232 J Comput Chem 30:2785–2791
6. Jensen MR, Ortega-Roldan JL, Salmon L et al 19. Beck MR, Otey CA, Campbell SL (2011)
(2011) Characterizing weak protein–protein Structural characterization of the interactions
complexes by NMR residual dipolar couplings. between palladin and alpha-actinin. J Mol Biol
Eur Biophys J 40:1371–1381 413:712–725
7. Marintchev A, Frueh D, Wagner G (2007) 20. Joseph C, Stier G, O’Brien R et al (2001) A
NMR methods for studying protein–protein structural characterization of the interactions
interactions involved in translation initiation. between titin Z-repeats and the alpha-actinin C-
Methods Enzymol 430:283–331 terminal domain. Biochemistry 40:4957–4965
8. Wittekind M, Mueller L (1993) Hncacb, a high- 21. Bodenhausen G, Ruben DJ (1980) Natural
sensitivity 3d Nmr experiment to correlate abundance N-15 Nmr by enhanced heteronuc-
amide-proton and nitrogen resonances with lear spectroscopy. Chem Phys Lett 69:185–189
the alpha-carbon and beta-carbon resonances 22. Delaglio F, Grzesiek S, Vuister GW et al (1995)
in proteins. J Magn Reson Ser B 101:201–205 NMRPipe: a multidimensional spectral proces-
9. Ikura M, Kay LE, Bax A (1990) A novel- sing system based on UNIX pipes. J Biomol
approach for sequential assignment of H-1, NMR 6:277–293
C-13, and N-15 spectra of larger proteins – 23. Hubbard SJ, Thornton JM (1993) NACCESS
heteronuclear triple-resonance 3-dimensional Computer Program, Department of Biochem-
Nmr-spectroscopy – application to calmodulin. istry and Molecular Biology, University
Biochemistry 29:4659–4667 College London www.bionf.manchester.ac.
10. Grzesiek S, Bax A (1992) Correlating back- uk/naccess/nac_intro.html
bone amide and side-chain resonances in larger 24. de Vries SJ, van Dijk M, Bonvin AM (2010)
proteins by multiple relayed triple resonance The HADDOCK web server for data-driven
NMR. J Am Chem Soc 114:6291–6293 biomolecular docking. Nat Protoc 5:883–897
11. Solomon I (1955) Relaxation processes in a 25. Venters RA, Calderone TL, Spicer LD et al
system of two spins. Phys Rev 99:559–565 (1991) Uniform 13C isotope labeling of pro-
12. Jahnke W (2002) Spin labels as a tool to teins with sodium acetate for NMR studies:
identify and characterize protein-ligand inter- application to human carbonic anhydrase II.
actions by NMR spectroscopy. Chembiochem Biochemistry 30:4491–4494
3:167–173 26. Kuipers BJ, Gruppen H (2007) Prediction
13. Dominguez C, Boelens R, Bonvin AM (2003) of molar extinction coefficients of proteins
HADDOCK: a protein–protein docking and peptides using UV absorption of the
constituent amino acids at 214 nm to enable of related native and engineered high-alkaline
quantitative reverse phase high-performance Bacillus subtilisins. J Mol Biol 292:111–123
liquid chromatography-mass spectrometry 28. DeLano WL (2006) MacPyMOL: The
analysis. J Agric Food Chem 55:5445–5451 PyMOL Molecular Graphics System, Version
27. Mulder FA, Schipper D, Bott R et al (1999) 1.5.0.4 Schrödinger, LLC www.pymol.org
Altered flexibility in the substrate-binding site
Chapter 17
Quantitative Protein Analysis by Mass Spectrometry

Vishwajeeth R. Pagala, Anthony A. High, Xusheng Wang,
Haiyan Tan, Kiran Kodali, Ashutosh Mishra, Kanisha Kavdia,
Yanji Xu, Zhiping Wu, and Junmin Peng
Abstract
Mass spectrometry is one of the most sensitive methods in analytical chemistry, and its application in
proteomics has been rapidly expanded after sequencing the human genome. Mass spectrometry is now the
mainstream approach for identification and quantification of proteins and posttranslational modifications,
either in small scale or in the entire proteome. Shotgun proteomics can analyze up to 10,000 proteins in a
comprehensive study, with detection sensitivity in the picogram range. In this chapter, we describe major
experimental steps in a shotgun proteomics platform, including sample preparation in the context of
studying protein-protein interaction, mass spectrometric data acquisition, and database search to identify
proteins and posttranslational modification analysis. Proteome quantification strategies and bioinformatics
analysis are also illustrated. Finally, we discuss the capabilities, limitations, and potential improvements of
current platforms.
Key words Proteomics, Mass spectrometry, Posttranslational modifications, Spectral counting,

Metabolic labeling, Isobaric labeling
1 Introduction
During the last two decades, mass spectrometry (MS)-based shot-

gun proteomics has emerged as the mainstream technology for
protein characterization with unprecedented specificity, sensitivity,
and throughput [1–3]. The technology is now capable of detecting
all components in the yeast proteome and the vast majority of the
human proteome [4]. Traditionally, protein-protein interactions
are discovered by yeast two-hybrid analysis [5] and affinity purifica-
tion followed by protein identification [6]. Rapid development in
mass spectrometry markedly accelerates protein identification and
expands the investigation of the protein interactome [7]. More
recently, a large-scale analysis of human stable protein complexes
has been carried out by chromatographic separation coupled with
281
282 Vishwajeeth R. Pagala et al.
quantitative mass spectrometry [8]. Here, we introduce the basics

of current proteomics technology, as well as its advances and
caveats.
A shotgun proteomics experiment typically consists of MS and
MS/MS analysis of peptides derived from proteolytically digested
proteins, followed by in silico matching of MS/MS spectra against a
database of theoretical peptide spectra generated from protein
sequences [9–11]. For complex protein mixtures, fractionation
technologies are implemented to separate proteins (e.g., by gel
electrophoresis) or peptides (e.g., by strong cation exchange or
reverse phase columns) before MS analysis [12]. Because of solvent
compatibility with ionization, reverse phase liquid chromatography
is usually coupled with online tandem mass spectrometry, termed
LC-MS/MS. Even with all the technological advances in the field,
there are still challenges that must be addressed. Sensitivity, repro-
ducibility, and comprehensiveness are interconnected factors that
must be considered in order to design a successful MS-based exper-
iment [13].
Current commercial mass spectrometers have limits of detec-
tion in the low femtomole or attomole range. This limit of detec-
tion appears to be sensitive enough to detect most proteins in their
pure state but the true sensitivity of mass spectrometry is deter-
mined by the type of protein sample. [13] Biological samples have a
wide range of protein abundance: more than six orders of magni-
tude in mammalian cells [4], namely, expression levels of proteins
may range from a few copies per cell to over ten million copies per
cell. Mass spectrometers have a difficult time dealing with this wide
dynamic range. When complex protein/peptide mixtures are ion-
ized together, ion suppression prevents the detection of species of
low abundance. Sample fractionation is routinely used to improve
the dynamic range.
During LC-MS/MS analysis, digestion of complex protein
mixtures produces a massive number of peptides beyond the ana-
lytical capacity of mass spectrometers. The scan speed of MS/MS is
not sufficient to analyze all peptides during the period of elution
time. Only a small fraction of peptides are selected for MS/MS and
the majority of peptides are ignored, often designated as a problem
of “undersampling” [13]. In the case of total cell lysate, less than
20 % of detectable peptides are selected for MS/MS [14]. There-
fore, lack of MS identification cannot be treated as evidence of
absence, especially for proteins of low abundance. In general, only
~70 % of proteins in a complex mixture are repeatedly identified
even in a replicate. This under-sampling issue presents a challenge
for quantifying known proteins of interest in a reproducible man-
ner. Undersampling can be partially corrected and reproducibility
improved by repeating analysis or by targeted proteomics
approaches [15, 16].
Protein Analysis by MS 283
Selection of mass spectrometers is critical for the purpose of a

proteomics study. A mass spectrometer consists of an ion source,
ion optics, and a mass analyzer [17]. Mass analyzers are the defining
analytical components that differentiate current MS platforms on
the market. Ion trap, Orbitrap, and ion cyclotron resonance mass
analyzers separate ions based on m/z resonance frequency, whereas
quadrupoles use m/z stability as a method of ion separation. A
time-of-flight (TOF) analyzer separates ions based on flight time.
Each mass analyzer has multiple properties that must be considered
before choosing the best fit for the experiment. Some of these
properties include accuracy and resolution, mass range, scan
speed, sensitivity, dynamic range, and the choices of MS/MS frag-
mentation methods (Table 1). Many current mass spectrometers on
the market, known as hybrid instruments, combine more than one
analyzer to achieve greater experimental flexibility.
The quantity of proteins in MS analysis can be evaluated by a
number of approaches, including label-free quantification based on
spectral counting (SC) or extracted ion current, as well as stable
isotope labeling methods. The spectral counts are the total num-
bers of MS/MS spectra identifying a protein, which increase almost
linearly with protein abundance after normalizing for protein size
[18]. This method works reasonably well for proteins with high
spectral counts, but the reliability decreases dramatically for pro-
teins with low spectral counts [19]. Alternatively, the abundance of
peptides in samples can be compared by extracted ion current of
corresponding ions [20]. Because ionization efficiency of the pep-
tides may vary among different LC runs, largely due to fluctuation
of the LC system and ion suppression, considerable variations may
be expected and need to be normalized in the label-free method.
To alleviate the problems associated with intrinsic LC-MS/MS
variations, a gold standard in quantitative MS is to use stable
isotope-labeled peptides as internal standards [1–3]. The internal
standards and their counterparts are eluted and ionized simulta-
neously during the LC-MS/MS runs, so that relative quantification
can be achieved by comparing peak areas of the peptide pairs. Stable
isotopes can be introduced into samples by in vitro labeling, such as
iTRAQ or TMT [21]. The iTRAQ/TMT reagents differentially
label the amine group at the N-termini and Lys residues of peptides
after protein digestion. Ten labeling reagents are now available to
allow multiplex comparisons in a single experiment. Moreover, the
strategy of stable isotope labeling with amino acids in cell culture
(SILAC) [22] has been developed as a highly accurate, in vivo
labeling method for large-scale proteomics. Given the simplicity
of label-free quantification, high accuracy of SILAC, and multiplex
comparison of iTRAQ/TMT, all of these methods are commonly
used by researchers. More recently, the Gygi group has combined
the SILAC and TMT strategies to analyze as many as 18 samples in
a single experiment [23].
284
Vishwajeeth R. Pagala et al.
Table 1
Common tandem mass spectrometer specifications (adapted from http://masspec.scripps.edu/mshistory/whatisms_details.php)
Triple quadrupole Linear ion trap Q-TOF LTQ orbitrap (ELITE) Q exactive
Accuracy 0.01 % (100 ppm) 0.01 % (100 ppm) 0.001 % (10 ppm) <2 ppm <2 ppm
Resolution 4,000 4,000 10,000 240,000 140,000
m/z Range 4,000 4,000 10,000 6,000 4,000
Scan speed (s) ~1 ~1 ~1 ~1 ~0.5
Comments Good accuracy, Good accuracy, Excellent accuracy, Excellent accuracy, resolution, Excellent accuracy, resolution,
good resolution, good resolution, good resolution, and sensitivity; flexible and sensitivity; only HCD
low-energy low-energy low-energy collisions, fragmentation (e.g., CID, available; bench top and
collisions, low cost collisions, low cost high sensitivity HCD, and ETD) easy to use
Here we describe a number of detailed protocols for routinely

analyzing proteins in our proteomics facility, including sample
preparation and protein digestion, protein identification by LC-
MS/MS, dissection of protein posttranslational modifications,
label-free protein quantification, protein profiling by SILAC or
iTRAQ/TMT, and bioinformatics data processing.
2 Materials
2.1 Sample Digestion 1. 20 mg/ml bovine serum albumin as an internal standard.

and LC-MS/MS 2. Acetonitrile.
Analysis for Protein
3. 5 mM dithiothreitol in 100 mM ammonium bicarbonate (in
Identification
HPLC grade water).
4. 10 mM iodoacetamide in 100 mM ammonium bicarbonate (see
Note 1).
5. 10 % formic acid.
6. 5 % formic acid in 50 % acetonitrile.
7. Trypsin (sequencing grade, 20 μg/50 μl aliquots in 50 mM
acetic acid): 2 μg/5 μl aliquots are prepared and stored at
80 C. An aliquot is thawed and diluted to 200 μl with
25 mM ammonium bicarbonate (working concentration is
0.01 μg/μl, pH ~8.0, see Note 2).
8. Solid urea (see Note 3).
9. 100 mM calcium chloride.
10. Lys-C (5 μg of lyophilized protein with salt in each vial, recon-
stituted in 50 μl of water, 0.1 μg/μl in 50 mM Tris, 10 mM
EDTA, pH 8.0).
11. C18 Zip tip (loading capacity: 10 μg peptide per μl resin, see
Note 4).
12. Trifluroacetic acid (TFA).
13. Equilibration buffer: 5 % acetonitrile and 0.1%TFA.
14. Elution buffer: 70 % acetonitrile and 0.1%TFA.
15. Standard peptide digest: tryptic BSA.
16. C18 column (e.g., 75 μm 10–15 cm, 15 μm tip orifice,
2.7 μm HALO beads, New Objective, (see Note 5) on column
selection).
17. LC loading buffer: 5 % formic acid and 0.1 % TFA.
18. Buffer A: 0.2 % formic acid.
19. Buffer B: 0.2 % formic acid, 70 % acetonitrile.
20. HPLC system (e.g., Waters nanoACQUITY UPLC or Thermo
EASY-nLC 1000).
21. Tandem MS instrument (e.g., Thermo Q Exactive or LTQ

Orbitrap Elite).
Additional reagents for phosphopeptide enrichment:
22. TiO2 beads.
23. Glutamic acid.
24. 15% ammonium hydroxide and 40 % acetonitrile.
2.2 Protein 1. A cell line for expressing a protein of interest.

Quantification 2. Base SILAC medium free of Arg and Lys amino acids.
Strategies
3. Dialyzed fetal calf serum (containing no free amino acids).
4. L-arginine and L-lysine.
5. Heavy stable isotope labeled L-type amino acids: [13C6 15 N4]
arginine (+10.0083 Da) and [13C6 15 N2] lysine
(+8.0142 Da).
6. SILAC light medium: mixing the base SILAC medium, dia-
lyzed fetal calf serum, and regular L-arginine and L-lysine.
7. SILAC heavy medium: similar to the light medium except
equal molar concentration of the heavy stable isotope labeled
L-arginine and L-lysine (see Note 6).
8. TMT 6-plex Isobaric Mass Tagging Kit.

9. 50 mM HEPES, pH 8.5.
10. 5 % Hydroxylamine.
2.3 Bioinformatics 1. Database downloaded from NCBI or UniProtKB/Swiss-Prot

Data Processing websites.
2. Database search engines, such as Sequest [24], and Mascot
[25].
3. In-house software suite for summarizing protein identification
and quantification.
4. A computer cluster for data processing.
3 Methods
3.1 Sample Digestion Protein samples are usually separated by sodium dodecyl sulfate
and LC-MS/MS (SDS) polyacrylamide gel electrophoresis followed by protein stain-
Analysis for Protein ing with Coomassie blue or SYPRO Ruby (see Note 7). The SDS
Identification gel not only resolves proteins based on protein size, but also
removes contaminants that interfere with mass spectrometry (e.g.,
3.1.1 Protein In-Gel salt and detergents, Fig. 1). Total protein level of the samples may
Digestion be also evaluated according to staining intensity if known amount
of bovine serum albumin is titrated on the same gel (e.g., 10, 100,
1,000 ng). As most protein quantification kits need microgram
Fig. 1 Major steps in spectral counting based protein quantification. Protein

samples with biological replicates are resolved on a gel, followed by in-gel
digestion, LC-MS/MS analysis, and computational data processing
levels of protein for accurate measurement, this gel-based assay

provides a simple approach for evaluating protein concentration
[26]. For gel excision, we determine the number of gel bands per
lane based on the complexity of samples and the total protein level
(e.g., at least 100 ng per gel band, Fig. 1). The in-gel digestion
protocol is modified from a previously reported version [27] as
listed below.
1. Cut protein bands from the gel (e.g., 50 μl of volume); chop
each gel band into pieces of ~1 cubic millimeter with a scalpel,
and transfer to a 0.5 ml microcentrifuge tube (see Note 8).
2. Wash the gel pieces twice with 400 μl of 50 % acetonitrile.

Vortex the tubes for 20 s, wait for 5 min, and then remove
the solution.
3. Reduce proteins with 150 μl of 5 mM DTT in 100 mM ammo-
nium bicarbonate. Vortex the tubes for 20 s, incubate for
30 min at 37 C, and then remove the solution.
4. Alkylate proteins with 150 μl of 10 mM iodoacetamide in
100 mM ammonium bicarbonate. Vortex the tubes for 20 s,
incubate for 30 min in dark at room temperature, and then
remove the solution.
5. Quench the reaction with 150 μl of 5 mM DTT in 100 mM
ammonium bicarbonate. Vortex the tubes for 20 s, incubate for
30 min at 37 C, and then remove the solution.
6. Wash twice with 400 μl of 50 % acetonitrile.
7. Wash once with 400 μl of 100 % acetonitrile.
8. Dry the gel pieces completely by speedvac for 10 min at low
temperature. The dehydrated gel pieces will have a paper-like
appearance.
9. Rehydrate the gel pieces with 40 μl of diluted trypsin in 25 mM
ammonium bicarbonate (0.01 μg/μl, pH ~ 8.0) on ice for
20 min. Top with 25 mM ammonium bicarbonate if needed
to cover the gel pieces.
10. Incubate at 37 C overnight (see Note 9).
11. Spin the tubes to collect any condensation during the incuba-
tion, add 20 μl of 10 % formic acid, vortex, and incubate for
20 min.
12. Spin, remove, and save the supernatant.
13. Extract peptides with 50 μl of 5 % formic acid in 50 %
acetonitrile.
14. Add 100 μl of 100 % acetonitrile to fully shrink the gel pieces to
recover peptides.
15. Pool all peptide-containing supernatant together (~200 μl).
16. Spin at 20,000 g for 10 min (see Note 10).
17. Transfer the supernatant to a 0.5 ml microcentrifuge tube, dry
in speedvac, and store at 20 C.
18. Dissolve the peptides right before LC-MS/MS analysis.
3.1.2 Protein In-Solution If protein samples contain no detergents, the samples may be
Digestion directly digested in solution to avoid the lengthy in-gel digestion
protocol. In-solution digestion efficiency might be lower than in-
gel digestion, which may be due to residual protein folding that
reduces trypsin accessibility. Thus, a two-step digestion protocol is
commonly used: (1) to cleave proteins under highly denaturing

conditions (e.g., 8 M urea) with Lys-C, and (2) to dilute urea to
2 M and further digest the proteins with trypsin. After in-solution
digestion, peptides are generally desalted and then analyzed by LC-
MS/MS.
1. Evaluate protein complexity and concentration in the samples
by running a SDS gel.
2. Add solid urea to the protein samples to final concentration of
8 M.
3. Add Lys-C at an enzyme-to-substrate ratio of 1: 100 (w/w),
and add CaCl2 to 10 mM.
4. Incubate at room temperature for 3 h.
5. Dilute the reaction fourfold with 25 mM ammonium bicarbon-
ate to reduce the urea concentration to 2 M.
6. Add trypsin at an enzyme-to-substrate ratio of 1: 50 (w/w).
7. Incubate at room temperature overnight.
8. Add TFA to 0.5 % to acidify the reaction.
9. Spin at 20,000 g for 10 min.
10. Desalt the supernatant by C18 Zip tip or another C18 column
as follows.
11. Wet the Zip tip (~1 μl bed volume) and wash three times with
10 μl of 70 % acetonitrile and 0.1%TFA.
12. Equilibrate with 10 μl of 5 % acetonitrile and 0.1 % TFA three
times.
13. Load the digest (<10 μg protein) on the Zip tip by aspirating
and dispensing ten times to maximize peptide binding. Wash
with 10 μl of 5 % acetonitrile and 0.1 % TFA three times.
14. Elute peptides twice with 10 μl of 70 % acetonitrile and 0.1%
TFA.
15. Dry the eluent in a speedvac, and store at 20 C.
3.1.3 Standard LC-MS/ Successful LC-MS/MS analysis relies on a robust and reproducible
MS Analysis LC-MS/MS system, including a nanoscale HPLC system with
autosampler, and an inline modern tandem mass spectrometer.
Peptide samples are loaded on a reverse-phase C18 column by the
HPLC system and eluted by a gradient. Eluted peptides are ionized
and detected by the mass spectrometer (e.g., Thermo Q Exactive or
LTQ Orbitrap Elite). MS spectra are collected first (in ~0.5 s), and
the top 20 abundant ions are sequentially isolated for MS/MS
analysis (each in ~0.1 s, totaling ~2 s). This process (~2.5 s) is
cycled over the entire liquid chromatography gradient, and more
than 14,000 MS/MS spectra are acquired during a 30-min elution

(Fig. 1).
1. Calibrate and tune the MS instrument regularly according to
the manufacturer’s guidelines.
2. Thoroughly wash the C18 column (75 μm 15 cm, bead
volume of 0.7 μl) with 95 % buffer B + 5 % buffer A, and
then equilibrate in 5 % buffer B + 95 % buffer A.
3. Run tryptic BSA (e.g., 0.2 μg) twice as a quality control step to
evaluate the sensitivity and reproducibility of the LC-MS/MS
system. Monitor signal intensity, peak width, and retention
time of the selected BSA peptide ions, as well as run-to-run
variability.
4. Dissolve dried samples in LC loading buffer before running
samples. Extended storage of samples in low concentration
results in significant peptide loss.
5. Load the sample on the LC-MS system (e.g., LTQ Orbitrap
Elite) using an optimized platform [26]. Briefly, the peptides
(~1 μg for the column above) are loaded, washed, and then
eluted during a gradient of 8–30 % acetonitrile in 30 min (flow
rate of 250 nl/min). The eluted peptides are detected by Orbi-
trap (400–1,600 m/z, 1,000,000 AGC target, 100 ms maxi-
mum ion time, resolution 240,000 FWHM) followed by 20
data-dependent MS/MS scans in LTQ (2 m/z isolation width,
35 % collision energy, 3,000 AGC target, 100 ms maximum ion
time, 30 s dynamic exclusion, only +2 and +3 ions selected, see
Note 11).
3.1.4 Analysis of Protein Approximately 300 forms of protein modifications are documen-
Posttranslational ted. In cells, the modifications range from the simple addition of
Modifications (PTM) small chemical groups (e.g., phosphorylation, methylation, and
acetylation), to more complex moieties (e.g., glycosylation, lipida-
tion, and ubiquitination). The modifications are reversible and
catalyzed by an array of enzymes, leading to regulation of protein
surfaces, structures, and interacting partners. Thus, protein mod-
ifications are central to signal transduction in cells.
Mapping modified residues by mass spectrometry is often chal-
lenging, which is highly different from protein identification analy-
sis that requires one unique peptide (matched to a single MS/MS
spectrum). For PTM mapping, approximately 100 % “sequence
coverage” by MS/MS is needed. As some digested peptides (with
extreme lengths or hydrophobicity) may not be compatible with
the standard LC-MS/MS method, custom methods may be
required, such as digestions with additional enzymes (e.g., Arg-C,
Glu-C, Asp-N) to improve peptide coverage. In addition, the stoi-
chiometry of PTM may be low and the modified sites may be
heterogeneous in the samples, further limiting the amount of mod-

ified species. For instance, one protein can be modified by one or
more modifications at a single site or multiple sites. Therefore, the
assignment of modification sites requires a much higher starting
sample amount (typically >100 ng of one purified protein excised
in a gel band). During isolation of modified proteins, specific
inhibitors are needed to prevent the loss of modifications, increas-
ing the yield of the modified forms.
For a complex protein mixture, modified peptides are usually
undetectable because of low abundance. Numerous strategies have
been developed to enrich for modified peptides. For instance,
phosphopeptides are isolated by immobilized metal-affinity chro-
matography (IMAC) enrichment incorporating metal oxides such
as Fe3+ ion [28], TiO2 [29], cation and anion exchange chroma-
tography [30], antibody capture [31], chemical derivation [32],
and a combination of these approaches. The TiO2 method is widely
used and the protocol for this method is described below:
1. Select the amount of TiO2 beads based on the level of peptide
digestion (0.4–0.6 mg beads for 0.1 mg digested protein, see
Note 12).
2. Wash with 200 μl of 50 % acetonitrile saturated with glutamic
acid.
3. Dissolve desalted and dried peptides with 100 μl of 50 % aceto-
nitrile saturated with glutamic acid.
4. Incubate with TiO2 for 1 h at room temperature.
5. Spin and remove supernatant.
6. Wash beads with 100 μl of 50 % acetonitrile with 0.1 % TFA
three times.
7. Elute the phosphopeptides with 20 μl of 15 % ammonium
hydroxide with 40 % acetonitrile, and acidify the samples (see
Note 13).
8. Dry the peptides by speedvac and store at 80 C until ready
for analysis by LC-MS/MS.
3.2 Protein SC-based quantification is straightforward and sensitive enough

Quantification to identify large changes among protein samples. Using co-
Strategies immunoprecipitation (IP) analysis as an example, the purpose
of the analysis is to identify novel interacting proteins of the
3.2.1 Protein Profiling by antigen. One expects that the antigen and associated proteins are
Spectral Counting (SC) abundant in the IPs but not in the controls. The LC-MS/MS
procedure is basically the same as the standard protocol above,
with the following exceptions (Fig. 1)
1. Biological replicates are highly recommended for quantitative

comparison. Batch-to-batch variation during IP analysis may
be large. It is critical to perform the analysis with at least one
negative control and at least one biological replicate to evaluate
the variation.
2. Examine the quality of IP samples. Make three aliquots (e.g.,
10, 20, and 70 %) of the samples for western blotting, SDS
PAGE, and MS analysis, respectively. The immunoblotting
confirms the antigen captured by antibodies, whereas the SDS
gel is stained with silver or SYPRO Ruby to assess the antigen
level of and total protein in the IPs, assisted by BSA loading on
the side of the gel (Fig. 1). Finally, the remaining 70 % of the
samples are equally loaded on a SDS gel, stained by Coomassie
blue or SYPRO Ruby, and subjected to in-gel digestion.
3. Extend the elution gradient from the standard 30 to 60 min to
increase scan number (see Note 14).
4. Finally, evaluate the dataset by several critical parameters: (1)
the antigen is expected to be one of the most abundant proteins
in the IPs (e.g., among the top 10. If not, the IP conditions
may need to be further refined to improve purity; (2) the SC of
the antigen should be high (e.g., >50). Otherwise, the likeli-
hood of finding its interacting proteins is low, because genuine
interacting partners are usually present at substoichiometric
level; (3) the variation between biological replicates should be
small.
3.2.2 Protein Profiling The SILAC approach can accurately compare the relative intensities
by SILAC of peptides from protein samples. First, cells are grown in medium
containing either light or heavy stable isotope labeled Lys and Arg.
Then the differentially labeled cells are treated under different
conditions, such as transfection with mock or a protein-expressing
plasmid in a co-IP experiment. The cells are harvested and lysed for
immunoprecipitation. Finally, the heavy and light IP samples are
pooled together for SDS PAGE and LC-MS/MS (Fig. 2, see Note
15). We continue to use the co-IP experiment to explain details in a
SILAC analysis.
1. Select appropriate cell lines and culture in light or heavy
medium to ensure full labeling.
2. Examine for cell viability in SILAC medium, as certain cell lines
are not able to thrive in SILAC medium.
3. Examine if the use of SILAC medium alters the biology of the
cells (e.g., expression of protein of interest).
4. Test for complete labeling of proteins. Certain cell types may
require extra passages to achieve full labeling of all proteins. It is
necessary to achieve >95 % labeling for accurate quantification.
Fig. 2 Protein profiling by the SILAC strategy. Protein samples are differentially
labeled by light and heavy SILAC media. The two samples are harvested and
subjected to immunoprecipitation. The IP samples are then pooled together,
subjected to gel separation, in gel digestion and reverse phase LC-MS/MS
analysis. Biological replicates can be performed using a forward and reverse
labeling strategy
5. Treat the cells under different conditions.

6. Perform a biological replicate by swapping the labeling order.
7. Harvest and lyse the cells for immunoprecipitation.
8. Resolve pooled IP samples on an SDS gel, followed by in-gel
digestion and LC-MS/MS analysis.
3.2.3 Protein Profiling The development of multiplex isobaric labeling methods (e.g.,
by iTRAQ/TMT iTRAQ and TMT) greatly improves the throughput of proteomics.
In this protocol, we describe how to perform a 6-plex TMT analysis
(Fig. 3). Every labeling reagent consists of three groups: an amine-
specific active ester group, a balance group, and a reporter group.
Six reagents can label six different samples by reacting with free
Fig. 3 Protein profiling by the TMT technology. Six samples can be differentially
labeled by 6-plex TMT reagents for a quantitative analysis. The samples are
first digested separately, labeled by TMT reagents, followed by pooling and
LC-MS/MS analysis. If necessary, two dimensional LC may be implemented,
including strong cation exchange (SCX) chromatography and reverse phase
chromatography
alpha N-termini of peptides and the epsilon amino groups of lysine

residues. The isobaric TMT tags, defined by different distribution
of isotopes (e.g., 13C and 15 N) between the reporter group and the
balance group, allow labeled peptides to coelute during chroma-
tography and ionize at the same mass-to-charge in MS scans. The
labeled samples are equally mixed, further fractionated, and ana-
lyzed by LC-MS/MS. During ion fragmentation, the TMT tags are
cleaved between the balance group and the reporter group, pro-
ducing reporter ions with different mass for quantification.
1. Digest six protein samples (~100 μg) using an in-gel or in-

solution protocol.
2. Desalt the peptides to remove salt and nonpeptide amine-
containing contaminants (e.g., ammonium bicarbonate).
3. Resuspend each sample in 20 μl of 50 mM HEPES buffer.
4. Reconstitute each TMT reagent in 41 μl of anhydrous acetoni-
trile, vortex and spin to collect all of the solution.
5. Mix each pair of peptide sample and TMT reagent; incubate for
1 h at room temperature.
6. Quench the reaction by adding 8 μl of 5 % hydroxylamine and
incubate for 15 min.
7. Make aliquots for the labeled samples and store at 80 C until
ready for analysis.
8. Perform a trial mix using a small aliquot of each sample (e.g.,
5 μl), analyze by LC-MS/MS to obtain the global peptide ratio
of all six samples.
9. Use the trial data to ensure equal mixing of the remaining
samples.
10. Desalt pooled TMT peptides by C18 cartridge, during which
the washing buffer is changed to 5 % acetonitrile with 0.2 %
formic acid to completely remove quenched TMT reagents.
11. The desalted peptides are ready for subsequent LC-MS/MS
analysis. If necessary, two dimensional LC separation may be
implemented to improve the throughput (Fig. 3, see Note 16).
3.3 Bioinformatics The pipeline of MS data processing is outlined in Fig. 4. First, a

Data Processing database search is used to convert MS raw data into peptide/
protein identification information. Then these proteins are quanti-
fied and subjected to statistical evaluation. Finally, proteins with
altered expression are reported for users to derive subsequent
hypotheses for targeted studies. However, any large-scale datasets
inevitably contains false positives, and proteomics results also have
certain pitfalls. Thus, it is important to validate MS results prior to
significant investment. Currently, there is no consensus on the best
software package in the proteomics field. We describe the pipeline
used in our proteomics facility, which integrates in-house software
and algorithms developed by colleagues.
3.3.1 Protein Numerous computation approaches have been developed to search

Identification databases for peptide identification [9, 10]. These approaches com-
pare one acquired MS/MS spectrum to all theoretical fragmenta-
tion spectra in databases, and the best matched peptide is selected as
putative identification. There are three fundamental steps during a
database search, including preprocessing, database search and post-
processing. While preprocessing improves data quality and removes
Fig. 4 The bioinformatics pipeline uses computer programs to match experimen-

tal MS/MS spectra to theoretical patterns in protein databases, resulting in the
identification of peptides/proteins. The target-decoy strategy is used to filter the
peptide matches by mass accuracy and matching scores until the protein false
discovery rate is lower than 1 %. For protein quantification analysis, additional
programs are used for peak extraction, sample comparison and normalization,
and statistical inference
low quality spectra to save time and reduce risk of false identifica-
tions, post-processing filters database search results to maintain the
false discovery rate at a tolerable level (e.g., 1 %).
Preprocessing 1. Extract MS/MS data from raw files. Mass spectrometers from
different companies generate raw data in different formats. For
instance, Thermo MS instruments produce files in raw format.
The raw files are first converted to mzXML format with
“ReAdW”, then to dta files by “MzXML2Search”. Both pro-
grams are available from the Institute for Systems Biology
(http://tools.proteomecenter.org). Each dta file contains mass

information for one MS/MS scan, including precursor ion,
charge state, product ions, and intensities.
2. Decharge and deisotope precursor ions in MS scans. Firstly, the
precursor ion recorded in MS/MS scans is searched in the
corresponding survey scan within a defined isolation window
(e.g., 2 m/z). Secondly, charge state and monoisotopic mass of
precursor ions are defined based on high resolution survey scan
data.
3. Correct mass in dta files. Systematic mass shift in the survey
scan is determined based on the polycyclodimethylsiloxane ion
(445.120025 m/z). On the basis of the mass shift, monoiso-
topic precursor ion and charge state, the dta files are edited to
increase accuracy.
4. Consolidate peaks in MS/MS scans to remove weak ions. The
top 10 peaks within every 100 m/z window are selected to
remove weak ions (see Note 17 for TMT data).
Database Search 1. Construct a target-decoy concatenated database. The target

database includes downloaded proteins and common contami-
nants, while the decoy database is created by reversing all target
sequences. The two databases are concatenated together for
MS/MS search. False discovery rate (FDR) is estimated by the
number of decoy matches (nd) and the total number of
assigned matches (nt). FDR¼2*nd/nt, assuming that mis-
matches in the target database have equal possibility of being
derived from the decoy database.
2. When using the Sequest search program (version 28), we use
the following search parameters: partially tryptic restriction,
five maximal missed cleavage sites, five maximal modification
sites, and 50 ppm mass tolerance for precursor ion. Product ion
mass tolerance is set to 0.02 Da for high resolution MS/MS
spectra from Q Exactive MS, or 0.5 Da for low resolution data
from LTQ Orbitrap Elite MS. Static modification on Cys is
carbamidomethylation (+57.02146 Da), and dynamic modifi-
cation includes Met oxidation (+15.99492 Da), and Ser/Thr/
Tyr phosphorylation (+79.96633) if needed. The search result
for each dta file is saved in an out file (see Note 18).
Post-processing 1. Prefilter the MS/MS matched peptides by minimal peptide

length (e.g., 7 for human database), and minimal matching
scores (e.g., XCorr and ΔCn in a Sequest search).
2. Filter the matched peptides by mass accuracy (e.g., 2 ppm).
3. Further filter the peptides by XCorr and ΔCn scores. The
peptides are divided into groups according to trypticity and
charge states, and then filtered by the two matching scores to

reduce peptide and protein FDR to approximately 1 %.
4. Finally, when peptides are shared among multiple members of a
protein family, all matched protein members are clustered
into a single group. On the basis of the parsimony principle,
the group is represented by one protein with the highest num-
ber of assigned peptides. Other proteins in the same group are
not accepted unless they are matched by additional unique
peptide(s).
3.3.2 Protein As described above, three protein quantification strategies, spectral

Quantification counting, SILAC, and iTRAQ/TMT, require different sample
preparation protocols. Their related data processing methods are
also distinct. After obtaining raw quantitative data for multiple
sample comparisons, statistical inference is implemented to evaluate
p-values, and a fraction of proteins are claimed to be altered. Two
commonly used methods are adjusted p-value (i.e., q-value) and
FDR derived from null experiments. The q-value is similar to p-
value, but it measures the proportion of false positives incurred
when the particular test is called significant. The latter is consider-
ably straightforward, simply by estimating false positives in null
comparisons.
Protein Profiling by Spectral Spectral count of a given protein is the summed number of its
Counting (SC) matching MS/MS spectra, similar to the number of reads in RNA-
seq analysis. Since peptide ions are selected for MS/MS based on
the rank of ion intensity, the spectral count of an identified protein
is correlated with its abundance. Some pitfalls are that the spectral
counting method is influenced by LC-MS/MS run-to-run varia-
tion, and small SC number (e.g., less than 5) markedly decreases
the reliability of quantitative comparison.
1. Acquire spectral counts for proteins assigned by database
search.
2. Normalize spectral counts to assume that the average SC per
protein should be the same in all samples. This assumption
works only when protein composition and concentration
among samples are highly similar. For IP samples, this assump-
tion may not apply well if the control and IP samples are vastly
different.
3. Use a G-test to determine statistical significance (p-value) of
protein abundance difference (see Note 19)
4. Calculate FDR based on null experiments (e.g., comparison
between biological replicates) and set up a threshold for filter-
ing the dataset.
Protein Profiling by SILAC The SILAC method offers many benefits compared to label-free
approaches. It can detect relatively small changes in protein expres-
sion levels or posttranslational modifications during comparison.
Whereas SILAC quantification may be carried out prior to protein
identification, as in the MaxQuant program [33], we usually derive
SILAC data after identifying proteins [34].
1. Extract ion currents in MS scans for identified peptides. Define
peak intensity, area intensity, and single-to-noise (S/N) ratio.
2. Match light and heavy isotope labeled peaks using predicted
m/z difference and a defined mass tolerance (e.g., 6 ppm for
high resolution MS data).
3. Compare peak intensity of matched peptides to obtain abun-
dance ratios of the peptides/proteins, and convert to log2
(ratio).
4. Summarize peptide ratios into protein ratios. Remove outliers
by Dixon’s Q test. The S/N ratio of a protein is represented by
that of the most abundant peptide ion.
5. Fit the data to Gaussian distribution to obtain mean and stan-
dard deviation (SD). If the vast majority of proteins are not
expected to change, the mean can be used to correct for sys-
tematic bias (e.g., experimental loading bias).
6. Select proteins with significantly changes in expression by the
cutoff of log2(ratio) that is outside of a 95 % confidence inter-
val (~2 SD) of the Gaussian distribution. FDR is evaluated by
null experiments of biological replicates. We also notice that
increasing the protein S/N cutoff leads to less false positives.
Protein Profiling Current isobaric mass tags (e.g., iTRAQ and TMT) provide simul-
by iTRAQ/TMT taneously analysis of 4, 6, 8, or 10 biological samples. Quantitative
information is extracted from report ions detected in MS/MS
spectra.
1. Acquire ion intensity of reporter ions in MS/MS scans for all
matched peptides, and define the S/N ratio of the ions.
2. For every binary comparison, compare the peak intensity of
reporter ions to obtain abundance ratios, and convert to log2
(ratio). The S/N ratio of a protein is represented by that of the
most abundant peptide ion.
3. The remaining protocol is the same as that in the SILAC
analysis.
3.3.3 Validation of MS Theoretically, a computer program could match almost any MS/
Results MS spectrum to a putative peptide sequence. Post-processing is
crucial to reduce false positives. The widely used target-decoy
strategy provides a general estimation of the FDR of an entire
dataset, but the reliability of individual peptides/proteins is not

well evaluated, especially for proteins matched by only one single
peptide, often termed one-hit-wonders. The same problem may
occur for PTM identification that is typically matched by one MS/
MS spectra. Errors in protein quantification also exist, because of
misidentification, mismatching of ion peaks, variation of weak
peaks, and peptide sharing among proteins. Thus, it is highly
recommended that extra validation strategies are used to affirm
the MS data.
1. Manually examine raw files for proteins of interest. Although
this is a tedious step, manual interpretation can eliminate most
errors in protein identification and quantification, as no current
computer programs consider all possibilities during data
processing.
2. Obtain both high resolution MS and MS/MS scans to improve
the fidelity of the dataset.
3. Perform additional targeted MS/MS analysis with different
fragmentation methods (see Note 20) for database search.
4. Chemically synthesize the peptides of interest, and then use
them as internal standards to confirm the identification of
native peptides. Their MS/MS patterns and retention time
during LC-MS/MS should be identical.
5. Verify protein changes by targeted MS strategy approaches
[16].
6. Verify protein changes by antibody-based approaches (e.g.,
Western blotting and immunohistochemistry).
4 Notes
1. Iodoacetamide (IAA) is used as a Cys-alkylation reagent and

should be freshly made. At high temperatures (e.g., heating in
SDS gel loading buffer), IAA modifies a fraction of Lys resi-
dues. Additional modification (i.e., two IAA molecules reacting
with one amino group) form a tag of 114.0429 Da, the same as
the mass of a tag generated by tryptic digestion of ubiquitin
[35]. At low temperature (i.e., room temperature or lower) and
concentration, this side reaction is essentially eliminated [34].
2. Unmodified trypsin undergoes auto-proteolysis to generate
peptides that may interfere with mass spectrometric analysis.
Also auto-proteolysis of trypsin can produce psuedotrypsin that
has chymotrypsin-like activity. Acetylation of the ε-amino
groups of lysine residues prevents autolysis. It is therefore
desirable to use modified trypsin for protein digestion.
3. Urea can be converted to ammonium cyanate in solution,

which is highly active and causes carbamylation of amine groups
in proteins. The side reaction is accelerated by heating, so we
generally use fresh urea solutions and perform digestion at
room temperature in the presence of urea. In addition, when
making 8 M urea solution, the dissolved urea occupies a signifi-
cant volume.
4. It is important to use an appropriate amount of C18 resin for
desalting. Either an excessive or limited amount of resin may
lead to low recovery of peptides.
5. The sample loading amount should vary based on the LC
system. In general, we load peptides digested from 1 μg of
protein onto a 10–15 cm 75 μm ID column. Increasing the
gradient elution time may increase the number of identified
proteins, but a plateau can be reached at a certain point due to
peak broadening. [26]. A higher loading level may be applied
for long columns (up to 1 m) installed on ultrahigh pressure
HPLC systems. The operation pressure of a LC system is
determined by column internal diameter, length, resin particle
size and porous feature, and viscosity of buffers.
6. To avoid heavy isotope-labeled Arg-Pro conversion, extra pro-
line is often included in the SILAC medium. Alternatively, Arg
may be omitted and only Lys labeling used. In this case, sam-
ples are digested with Lys-C to ensure quantification of identi-
fied peptides. In addition to culturing cells in SILAC medium,
flies [36] and mice [37] can also be fully labeled by feeding
SILAC food.
7. Coomassie Blue G-250 colloidal protein staining can detect
proteins as low as 10 ng and has little effect on in-gel digestion.
A barely visible protein band is sufficient for sensitive MS
identification. Sypro ruby stain has higher sensitivity and also
works for MS. An MS-compatible silver staining protocol is
recommended but could work with reduced sensitivity.
8. Special care needs to be taken to prevent keratin contamination
of samples. Wear gloves at all times, and rinse them occasionally
as they readily accumulate static charge and attract dust and
pieces of hair and wool. If possible, perform all operations
under laminar flow hood. Always visually check flasks, tubes,
and pipette tips for contaminating particles.
9. Keep a jar of water inside the incubator to avoid a temperature
gradient between the bottom and the lid of the tube. This
prevents condensation of water at the inner surface of the lid
and, consequently, premature dehydration of the gel pieces.
10. Centrifugation removes any residual gel pieces or other parti-
cles in extracted peptide solutions. This step is important to
avoid clogging of the nanoscale LC system, in particular, if

in-gel digested samples are directly injected onto nanoscale
LC columns.
11. MS instrumentation settings are optimized to achieve maximal
peptide identification during gradient elution. The LTQ Orbi-
trap Elite hybrid instrument is capable of acquiring data depen-
dent MS/MS in the ion trap while acquiring a high resolution
full scan MS spectrum in the Orbitrap without affecting the
duty cycle. The Q Exactive instrument can scan ions in the
Orbitrap and accumulate ions simultaneously. The duty cycle is
mainly affected by scan AGC, maximum ion injection time, and
resolution.
12. Selectivity of phosphopeptide enrichment methods is affected
by the ratio between beads, total level of peptides, the percent-
age of phosphopeptides, and the use of competitive reagent to
reduce background binding [38]. Numerous competitive
reagents have been compared and glutamic acid appears to
one of the best choices [39].
13. A high pH in the elution step tends to degrade phosphopep-
tides. Acidify the samples immediately after elution.
14. During SC-based quantitative analysis, we shorten the dynamic
exclusion time from 30 to 8 s in MS to increase the SC of
proteins on a LTQ Orbitrap Elite instrument. When dynamic
exclusion is set to 30 s, once one peptide ion is selected for
MS/MS, it will not be selected again in the following 30 s, as its
elution peak width is ~30 s. This function is designed to avoid
repetitive selection of highly abundant ions. Reducing the
dynamic exclusion may slightly affect the total number of pro-
teins identified, but greatly improves the accuracy of SC-based
quantification.
15. In SILAC analysis, one may think that mixing cell lysates before
IP would reduce experimental variation. However, after these
experiments failed, we recognized that noncovalently asso-
ciated proteins are exchangeable during IP, and the exchange
rate is rapid [40]. Thus it is important to perform IP first and
then mix the SILAC-labeled proteins for LC-MS analysis.
16. Isobaric labeling strategies suffer from quantification distor-
tion, a problem caused by co-eluted peptides that give rise to
the same reporter ions, raising noise level of reporter ions. For
example, a tenfold change may be detected as a fourfold
change. In most cases, the co-eluted peptides may be very
low in intensity but could be present in a large number,
which may explain the distortion that occurs even when the
precursor ions look highly abundant with no obvious co-eluted
peaks. The problem may be alleviated by reducing precursor
isolation window or extensive fractionation, but neither
method solves the issue. Alternatively, the problem can be

addressed by gas phase ion purification [41], isolation of MS2
ions for MS3 fragmentation, and measurement of C-terminal
peptide ions [42, 43].
17. The MS/MS consolidation step is inactivated during TMT
analysis, because this function may remove TMT reporter
ions (e.g., 6 ions between 126 and 131 m/z).
18. Although the Orbitrap allows the acquisition of high resolution
data with mass accuracy within a few ppm or even sub-ppm
dependent on the setting and intensity of ion signal, a wide
window (50 ppm) is used during the search and a much nar-
rower mass window is used later during data filtering to remove
false positives. However, the narrow mass window may be
applied during the search step, and then cannot be used for
filtering [26]. The same scenario can be applied to the settings
of tryptic restriction, maximal missed cleavage sites, and maxi-
mal modification sites.
19. The G-value of each protein is calculated by the following
equation [19]:

S1 S2
G¼2 S 1 ln þ S 2 ln
2 ðS 1 þ S 2 Þ 2 ðS 1 þ S 2 Þ
Where S1 and S2 are the detected spectral counts of a given

protein in any of the two samples for comparison, respectively.
Neither S1 nor S2 can be zero in the equation. In practice, we
use a relatively small number (e.g., 0.000001) to estimate G-
value for zero spectral counts in either of the two samples.
20. Peptide ions can be fragmented by multiple technologies, such
as collision-induced dissociation (CID), higher-energy colli-
sion dissociation (HCD), and electron transfer dissociation
(ETD), generating different product ion MS/MS patterns.
Combination of these MS/MS spectra can significantly
improve the confidence of protein identification [44].
Acknowledgements
This work was partially supported by the National Institutes of

Health (R21NS081571, R21AG039764, and P30CA021765),
the American Cancer Society (RSG-09-181), and ALSAC (Ameri-
can Lebanese Syrian Associated Charities).
References
1. Cravatt BF, Simon GM, Yates JR 3rd (2007) 16. Picotti P, Aebersold R (2012) Selected reaction
The biological impact of mass-spectrometry- monitoring-based proteomics: workflows,
based proteomics. Nature 450:991–1000 potential, pitfalls and future directions. Nat
2. Choudhary C, Mann M (2010) Decoding sig- Methods 9:555–566
nalling networks by mass spectrometry-based 17. Aebersold R, Mann M (2003) Mass
proteomics. Nat Rev Mol Cell Biol spectrometry-based proteomics. Nature
11:427–439 422:198–207
3. Gstaiger M, Aebersold R (2009) Applying 18. Liu H, Sadygov RG, Yates JR 3rd (2004) A
mass spectrometry-based proteomics to genet- model for random sampling and estimation of
ics, genomics and network biology. Nat Rev relative protein abundance in shotgun proteo-
Genet 10:617–627 mics. Anal Chem 76:4193–4201
4. Mann M, Kulak NA, Nagaraj N et al (2013) 19. Zhou JY, Afjehi-Sadat L, Asress S et al (2010)
The coming age of complete, accurate, and Galectin-3 is a candidate biomarker for amyo-
ubiquitous proteomes. Mol Cell 49:583–590 trophic lateral sclerosis: discovery by a proteo-
5. Chien CT, Bartel PL, Sternglanz R et al (1991) mics approach. J Proteome Res 9:5133–5141
The two-hybrid system: a method to identify 20. Wang W, Zhou H, Lin H et al (2003) Quanti-
and clone genes for proteins that interact with a fication of proteins and metabolites by mass
protein of interest. Proc Natl Acad Sci U S A spectrometry without isotopic labeling or
88:9578–9582 spiked standards. Anal Chem 75:4818–4826
6. Behrends C, Sowa ME, Gygi SP et al (2010) 21. Ross PL, Huang YN, Marchese JN et al (2004)
Network organization of the human autophagy Multiplexed protein quantitation in Saccharo-
system. Nature 466:68–76 myces cerevisiae using amine-reactive isobaric
7. Vidal M, Cusick ME, Barabasi AL (2011) tagging reagents. Mol Cell Proteomics
Interactome networks and human disease. 3:1154–1169
Cell 144:986–998 22. Ong SE, Blagoev B, Kratchmarova I et al
8. Havugimana PC, Hart GT, Nepusz T et al (2002) Stable isotope labeling by amino acids
(2012) A census of human soluble protein in cell culture, SILAC, as a simple and accurate
complexes. Cell 150:1068–1081 approach to expression proteomics. Mol Cell
9. Eng JK, Searle BC, Clauser KR et al (2011) A Proteomics 1:376–386
face in the crowd: recognizing peptides 23. Dephoure N, Gygi SP (2012) Hyperplexing: a
through database search. Mol Cell Proteomics method for higher-order multiplexed quantita-
10(R111):009522 tive proteomics provides a map of the dynamic
10. Nesvizhskii AI, Vitek O, Aebersold R (2007) response to rapamycin in yeast. Sci Signal 5:rs2
Analysis and validation of proteomic data gen- 24. Eng J, McCormack AL, Yates JR 3rd (1994)
erated by tandem mass spectrometry. Nat An approach to correlate tandem mass spectral
Methods 4:787–797 data of peptides with amino acid sequences in a
11. Peng J, Gygi SP (2001) Proteomics: the move protein database. J Am Soc Mass Spectrom
to mixtures. J Mass Spectrom 36:1083–1091 5:976–989
12. Xie F, Smith RD, Shen Y (2012) Advanced 25. Perkins DN, Pappin DJ, Creasy DM et al
proteomic liquid chromatography. J Chroma- (1999) Probability-based protein identification
togr A 1261:78–90 by searching sequence databases using mass
spectrometry data. Electrophoresis
13. Duncan MW, Aebersold R, Caprioli RM 20:3551–3567
(2010) The pros and cons of peptide-centric
proteomics. Nat Biotechnol 28:659–664 26. Xu P, Duong DM, Peng J (2009) Systematical
optimization of reverse-phase chromatography
14. Michalski A, Cox J, Mann M (2011) More than for shotgun proteomics. J Proteome Res
100,000 detectable peptide species elute in sin- 8:3944–3950
gle shotgun proteomics runs but the majority is
inaccessible to data-dependent LC-MS/MS. J 27. Shevchenko A, Wilm M, Vorm O et al (1996)
Proteome Res 10:1785–1793 Mass spectrometric sequencing of proteins
silver-stained polyacrylamide gels. Anal Chem
15. Gerber SA, Rush J, Stemman O et al (2003) 68:850–858
Absolute quantification of proteins and
phosphoproteins from cell lysates by 28. Ficarro SB, McCleland ML, Stukenberg PT
tandem MS. Proc Natl Acad Sci U S A et al (2002) Phosphoproteome analysis by
100:6940–6945 mass spectrometry and its application to
Saccharomyces cerevisiae. Nat Biotechnol 37. Kruger M, Moser M, Ussar S et al (2008)

20:301–305 SILAC mouse for quantitative proteomics
29. Larsen MR, Thingholm TE, Jensen ON et al uncovers kindlin-3 as an essential factor for
(2005) Highly selective enrichment of phos- red blood cell function. Cell 134:353–364
phorylated peptides from peptide mixtures 38. Li QR, Ning ZB, Tang JS et al (2009) Effect of
using titanium dioxide microcolumns. Mol peptide-to-TiO2 beads ratio on phosphopep-
Cell Proteomics 4:873–886 tide enrichment selectivity. J Proteome Res
30. Ballif BA, Villen J, Beausoleil SA et al (2004) 8:5375–5381
Phosphoproteomic analysis of the developing 39. Kettenbach AN, Gerber SA (2011) Rapid and
mouse brain. Mol Cell Proteomics 3:1093–1101 reproducible single-stage phosphopeptide
31. Rush J, Moritz A, Lee KA et al (2005) Immu- enrichment of complex peptide mixtures:
noaffinity profiling of tyrosine phosphorylation application to general and phosphotyrosine-
in cancer cells. Nat Biotechnol 23:94–101 specific phosphoproteomics experiments. Anal
32. Zhou H, Watts JD, Aebersold R (2001) A Chem 83:7635–7644
systematic approach to the analysis of 40. Wang X, Huang L (2008) Identifying dynamic
protein phosphorylation. Nat Biotechnol interactors of protein complexes by quantita-
19:375–378 tive mass spectrometry. Mol Cell Proteomics
33. Cox J, Mann M (2008) MaxQuant enables 7:46–57
high peptide identification rates, individualized 41. Wenger CD, Lee MV, Hebert AS et al (2011)
p.p.b.-range mass accuracies and proteome- Gas-phase purification enables accurate, multi-
wide protein quantification. Nat Biotechnol plexed proteome quantification with isobaric
26:1367–1372 tagging. Nat Methods 8:933–935
34. Xu P, Duong DM, Seyfried NT et al (2009) 42. Ting L, Rad R, Gygi SP et al (2011) MS3
Quantitative proteomics reveals the function of eliminates ratio distortion in isobaric multi-
unconventional ubiquitin chains in proteaso- plexed quantitative proteomics. Nat Methods
mal degradation. Cell 137:133–145 8:937–940
35. Nielsen ML, Vermeulen M, Bonaldi T et al 43. McAlister GC, Huttlin EL, Haas W et al (2012)
(2008) Iodoacetamide-induced artifact mimics Increasing the multiplexing capacity of TMTs
ubiquitination in mass spectrometry. Nat using reporter ion isotopologues with isobaric
Methods 5:459–460 masses. Anal Chem 84:7469–7478
36. Sury MD, Chen JX, Selbach M (2010) The 44. Guthals A, Bandeira N (2012) Peptide identifi-
SILAC fly allows for accurate protein quantification by tandem mass spectrometry with alter-
cation in vivo. Mol Cell Proteomics nate fragmentation modes. Mol Cell
9:2173–2183 Proteomics 11:550–557
Chapter 18
Using Peptide Arrays Created by the SPOT Method

for Defining Protein-Protein Interactions
Yun Young Yim, Katherine Betke, and Heidi Hamm
Abstract
Evaluating sites of protein-protein interactions can be an arduous task involving extensive mutagenesis
work and attempts to express and purify individual proteins in sufficient quantities. Peptide mapping is a
useful alternative to traditional methods as it allows rapid detection of regions and/or individual residues
important for binding, and it can be readily applied to numerous proteins at once. Here we describe the use
of the ResPep SL SPOT method to evaluate protein–protein binding interactions such as that between G-
protein βγ subunits and SNARE proteins, identifying both regions of interest and subsequently individual
residues which can then be manipulated in further biochemical assays to confirm their validity.
Key words ResPep SL, Fmoc chemistry, Peptide, Protein-protein interaction, Alanine screening
1 Introduction
1.1 History/Theory Automated peptide arrays such as that performed by the ResPep SL
enable protein-protein interactions to be easily studied. Using the
SPOT method, overlapping linear peptides are simultaneously
synthesized on cellulose membranes and used to predict the
biological activity of a protein. The SPOT method was first
described by Ronald Frank in 1992 [1] as a method for in situ
peptide synthesis. It creates short, immobilized peptides on a glass
plate or cellulose membrane, which are easy to handle and retain
some of the biological characteristics derived from the primary
sequence [2]. Cellulose membranes have become the preferred
medium for these arrays as the membranes are inexpensive, stable
in aqueous solutions, nontoxic to biological samples, and able to
withstand organic solvents and acids [3]. Further, for a given
membrane, the SPOT method allows the parallel synthesis of up
to 600 different peptides in a highly reliable manner. This is done
using Fmoc-chemistry whereby the amino group of each amino
acid is temporarily protected by a 9-fluorenylmethoxycarbonyl
group (Fmoc) to prevent polymerization and allow the addition
307
308 Yun Young Yim et al.
of only one amino acid to a growing peptide chain at each step. It

enables repetitive amino terminus deprotection by piperidine and
allows a mild-acid labile peptide–resin linkage [4]. Other reactive
functional side chain groups are also preserved by protection
groups such as tert. butyl-oxy(OtBu) and trityl (Trt), preventing
interaction between side chains and the amino terminal of activated
amino acids. As the synthesis proceeds, the machine cycles through
deprotection steps whereby the Fmoc groups are removed by
piperidine, the membranes washed, and activated amino acids cou-
pled to their respective partners according to the desired sequence
until the full peptide is assembled [5]. Using the ResPep SL SPOT
method, we have shown that such peptide arrays can be used to
identify biologically active protein-protein interaction sites. Despite
the fact that peptides only share some of the biological activities of
their full-length counterparts, this method enables the generation
of a large number of individual peptides at once to facilitate screen-
ing for biological activity with a diverse group of effectors.
1.2 The Applications Peptides arrays generated using the SPOT method have wide
of Peptide Arrays biological applications, including evaluating the actions of
enzymes, cell adhesion, and the binding of metals [2] but are
particularly useful for identifying sites of protein-protein inter-
actions and the development of inhibitors or activators of enzymes
[6]. To understand protein-protein interactions, investigators often
use co-immunoprecipitation, pull-down, or two hybrid assays, as
well as in vivo Förster resonance energy transfer (FRET) [7]. While
such assays can identify the presence of protein-protein interac-
tions, they cannot determine the interaction sites without conjunc-
tion to other methods such as X-ray crystallography. At present, X-
ray crystallography is widely used to evaluate sites of protein-
protein interaction; however, this process is both challenging and
time consuming, particularly with proteins that are difficult to
express, crystallize, or obtain diffraction. In such instances, peptide
arrays offer an easy alternative as they can rapidly highlight regions
of residues which may be involved in binding without the challenge
of expressing and crystallizing full-length proteins. While caution
must be taken as this method does not use full-length expressed
proteins, it offers a method to determine regions of interaction
which can then be followed up with alanine screening and other
biochemical methods to test identified residues in the context of the
full-length proteins for their functional role in the interaction. As
an example, Wells et al. [8] recently described this method to
characterize the interaction sites between G-protein βγ subunits
and a member of the SNARE complex, SNAP25. Following initial
peptide screens, they determined which specific amino acid residues
were important using Ala screening, before demonstrating that
those residues were important in the context of the whole protein
using biochemical and functional assays. Comparatively, peptide
Peptide Arrays 309
arrays generated using the SPOT method can also be used in a high
throughput manner to develop enzyme inhibitors, as well as active
substrates for various enzymes. As Thiele et al. shows, by generat-
ing peptide libraries on a glass plate or cellulose membrane, the
peptide-enzyme activity, such as phosphorylation of peptide, can be
detected using autoradiography, chemiluminescence, or enzymatic
color development [6]. By monitoring the phosphorylation of
peptides by a particular enzyme, researchers can select lead peptides
which have the potential to be developed as enzyme inhibitors or
activators. This is exemplified in Mukhija et al. [9], where the
authors successfully used this method to identify peptides inhibit-
ing enzyme I (EI) of the bacterial phosphotransferase system
(PTS). Using immobilized combinatorial peptide libraries and
phosphorimaging, they identified heptapeptides and octapeptides
which selectively inhibit EI in vitro. Similarly, Collet et al. [10] has
screened potential substrates of proteases by synthesizing peptides
on a fluorous-coated glass. For the first time, the authors describe
the use of fluorous to probe peptide sequences in protease screen-
ing. With the change in material where the peptide is immobilized
and the detection methods [5], the peptide array by SPOT method
was used as a screening method for the development of inhibitors
or activators of enzymes. Thus, peptide arrays generated by the
SPOT method have wide biological implications including signifi-
cantly enhancing protein-protein interaction studies, as well as
aiding in the development of reagents such as inhibitors and active
substrates of enzymes.
In this chapter, we demonstrate how to use peptide arrays
created by the SPOT method to understand protein-protein inter-
actions. We first describe how to prepare all the reagents and
program software for the SPOT method. With appropriate reagents
and correctly programmed software, peptides are synthesized on
cellulose membrane. Then, we show how to de-protect side chain
and use far-western technique to identify peptides involved in
protein-protein interaction. Lastly, we describe how to identify
amino acids involved in protein-protein interaction by alanine
screen.
2 Materials
Prepare all solutions using Milli-Q water and analytical grade

reagents. Reagents should be stored at room temperature unless
otherwise noted and materials disposed of in the appropriate man-
ner. Carry out all procedures at room temperature and inside a
fumehood unless otherwise specified.
1. Auto SPOT robot, such as the ResPep SL from Intavis loaded

with ResPep software.
2.1 SPOT Method 2. Spotter needle.

Preparation 3. 1 ml ResPep glass syringe.
4. Protein sequence: From UniProt (www.uniprot.org), find the
protein sequence of interest and save as a text file.
5. 1 l of anhydride dimethylformamide (DMF): Pour DMF into a
glass 1 l bottle and load into the Auto SPOT robot in the
dilutor 1 reservoir.
6. 4 l of anhydride dimethylformamide (DMF): Connect 4 l glass
bottle of DMF to ResPep SL equipment as the dilutor 2 solvent
1 (see Note 1).
7. 2 l of ethanol: Connect to ResPep SL equipment as the dilutor
2 solvent 2 (see Note 1).
2.2 Peptide Array 1. Amino acid derivatives: Remove amino acids derivatives from
Synthesis the 20 C freezer and thaw until they reach room tempera-
ture (approximately 2 h). Weigh out the appropriate amount of
each amino acid derivative into 2 ml microcentrifuge tubes and
dissolve in N-methyl pyrrolidone (NMP) using an oscillating
shaker (see Note 2).
2. 2 ml ResPep tubes with caps.
3. Membrane preparation: Pre-swell cellulose membrane from
INTAVIS in 30 ml of DMF in a black container covered with
a lid for 1 h. Agitate gently every 15 min.
4. 1.1 M hydroxybenzotriazole (HOBTlH2O): Weigh out the
appropriate amount of HOBTlH2O into a 30 ml conical tube
and dissolve in DMF to produce a 1.1 M solution (see Note 3).
5. 1.1 M diisopropylcarbodiimide (DIC): Using a glass pipette,
dilute the appropriate volume of DIC with DMF in a 50 ml
conical tube to produce a 1.1 M solution (see Note 3).
6. 20 % piperidine: In a 250 ml plastic bottle, add the appropriate
volume of piperidine to DMF to produce a 20 % solution (see
Note 3).
7. 5 % acetic anhydride: Using a glass pipette, transfer the appro-
priate volume of acetic anhydride to a 50 ml conical tube and
mix with DMF to produce a 5 % solution (see Note 3).
2.3 Side Chain 1. Deprotection solution (10 ml/membrane): Using a glass

Deprotection pipette, add 9.5 ml of trifluoroacetic acid (TFA) to a glass
beaker. Add 300 μl of triisopropylsilane and 200 μl of ddH2O
to the beaker and mix thoroughly.
2. Dichloromethane (DCM): Using a glass pipette, transfer 80 ml
of DCM into a glass beaker.
3. DMF: Using a glass pipette, transfer 80 ml of DMF into a glass
beaker.
Peptide Arrays 311
4. Ethanol: Using a glass pipette, transfer 40 ml of 100 % ethanol

into a glass beaker.
2.4 Far-Western 1. Plastic container for cellulose membrane.
Development 2. Ethanol.
3. Tris Buffered Saline (TBS): 50 mM Tris–HCl, pH 7.6, and
150 mM NaCl.
4. TBS containing 0.1 % Tween (TBST).
5. Blocking solution: Dissolve 5 % dried skim milk (w/v) in TBST
(0.1 % Tween). Store at 4 C until needed.
6. Primary antibody solution: Dissolve 5 % dried skim milk (w/v)
in TBST (0.2 % Tween) and 0.1 % n-octylglucoside (OG). Add
the primary antibody at the appropriate dilution and store at
4 C until needed (see Note 4).
7. Secondary antibody solution: Dissolve 5 % dried skim milk
(w/v) in TBST (0.2 % Tween) and 0.1 % n-octylglucoside
(OG). Add the secondary antibody at the appropriate dilution
and store at 4 C until needed (see Note 4).
8. Wash buffer: TBST (0.1 % Tween) and 0.1 % OG (see Note 4).
9. TBS with n-octylglucoside (OG): 50 mM of Tris–HCl, pH 7.6,
150 mM NaCl, and 0.1 % OG (see Note 4).
10. Protein binding buffer: 20 mM HEPES-KOH, pH 7.5, 2 mM
MgCl2, 0.1 % n-octylglucoside, and 5 % glycerol (see Note 5).
11. Binding partners: Purified protein(s) of interest to examine
interaction with immobilized peptides (see Note 6).
12. Antibodies: Primary and horseradish peroxidase (HRP) conju-
gated secondary antibodies (see Note 7).
13. Enhanced chemiluminescence (ECL) reagents.
2.5 Alanine Screen 1. This process requires the same materials listed in
Subheadings 2.1–2.4.
3 Methods
Carry out all procedures at room temperature unless otherwise

specified. Thoroughly read the manual of the auto spotter equip-
ment such as the ResPep SL manual to understand how to use the
equipment and programs.
3.1 SPOT Method 1. Turn on Auto SPOT robot such as ResPep SL and vacuum
Preparation (Fig. 1) pump by pressing the green ON/OFF switch on the side of the
AUTO spot robot.
2. Start the spotter software such as ResPep SL Spotter software
on the computer (see Note 8).
3. Change waste bottles as needed.
Fig. 1 ResPep SL and membrane design to study protein-protein interactions. (a) The ResPep SL is a peptide
synthesizer that needs to be connected to a computer, vacuum pump, and various solutions to synthesize
peptides on a membrane. The 4 L of anhydride Dimethylformamide and 2 L of ethanol are connected to the
equipment from the outside as dilutor 2 solvent 1 and solvent 2, respectively (permission from Intravis AG). (b)
Shown is a representative image of membrane design. White dots represent the location where peptides will
be synthesized. Grey dots represent negative controls which are left without peptide synthesis. On a
membrane, the positive control to the protein of interest and primary antibody are also synthesized. For
example, the sequences for the SIRK peptide, QEHA peptide, βARK peptide, the Gβγ binding domain of the
calcium channel CaV2.2, and the C-terminus of Gβ1 are used to synthesize the positive control peptides to
study the interaction of SNAP-25 and Gβ1γ2 (permission from Intravis AG)
4. Complete steps 5–12 below for each protein to be synthesized

on the membrane (see Note 9).
5. Under the “Edit sequence” tab, choose “open” and open text
file.
6. Convert sequence to proper form that can be read by ResPep.
7. Type the name of the protein in the format “;<name of
protein>” (e.g., human SNAP25).
8. On the next line, enter the length of the peptides to be synthe-
sized and the extent of overlap between subsequent peptides
(e.g., .seq,14,3 indicates peptides are to be 14 amino acids in
length and overlap by three residues).
9. To add a single space between peptide spots, enter the com-
mand: “.space.”
10. To clear the rest of the row after your last peptide for a particu-
lar protein, enter the command: “.newline.”
11. To leave a line completely blank between two proteins, enter
the command: “.newline”; Go to next line, type “.space” and
again go to a new line and type “.newline.”
12. Type “.end” where you want the machine to stop.
13. Once all sequences have been entered, press the “clean
sequence” button under the “edit sequence” tab to ensure
that all information is in the correct format (see Note 10).
Peptide Arrays 313
14. Press the “SHOW PEPTIDES” tab and click any peptide to
ensure it is located at the right position.
15. Load the method: Click the “EDIT METHODS” tab, press “file”
and open the method called “SPOTpreactiv.” (see Note 11).
16. Click the “REPORT” tab and check the amount of amino acid
and reagents needed for the prep (see Note 12).
17. Mount and calibrate the spotter needle and 1 ml syringe (see
Note 13).
18. Test the vacuum pump by clicking the ON and OFF buttons
under the “MANUAL” tab in the ResPep software.
19. Fill the DMF reservoir bottle in the ResPep machine and
ensure that the ethanol and DMF bottles on the outside are
attached to the Auto SPOT robot as appropriate (see Note 14).
20. Prime dilutor 1 and 2 to remove air bubbles (see Note 15).
21. Wash the membrane three times with ethanol using the wash
protocol in the method folder.
3.2 Peptide Array 1. In the fumehood, transfer the dissolved amino acid derivatives
Synthesis (Fig. 2) into 2 ml ResPep tubes with caps (see Note 16).
2. Place the pre-swelled membrane onto the SPOT synthesis
frame in fumehood (see Note 17).
3. Load the SPOT synthesis frame into the ResPep SL equipment
and connect the vacuum tube to bottom of the SPOT synthesis
frame.
4. Wash the membrane three times with ethanol using the wash
protocol in methods (see Note 18).
5. Reload the “SPOTpreactiv” method for peptide generation.
Fig. 2 Generating peptide arrays. (a) The cellulose membrane is pre-swelled by DMF. (b) Each peptide is
generated as 14mers, and peptides overlap as shown. The blue residues are part of three different peptides.
(c) Linear peptides are generated on a membrane as shown. Each color dot represents amino acids, which are
synthesized into peptides
6. Remove the caps from reagents (1.1 M HOBt H2O, 1.1 M

DIC, 20 % piperidine, and 5 % acetic anhydride solutions) and
cover with tin foil. Create a hole in the tin foil to allow the
spotter needle to pass into the reagent tubes unimpeded (see
Note 19).
7. Load all reagents (1.1 M HOBt H2O, 1.1 M DIC, 20 % piper-
idine, and 5 % acetic anhydride solutions) into the specified
positions on the reagent rack in the ResPep SL equipment.
8. Place the amino acid tubes into the appropriate locations on the
amino acid rack as directed by the report file (see Note 20).
9. Remove the lids from the 0.6 ml microcentrifuge tubes and
place into the small holes next to each amino acid tube (see
Note 21).
10. Ensure all reagents are in place and close the cabinet door.
11. Verify the protein sequences and peptides to be generated.
12. Verify the loaded method.
13. Under “Run Synthesis” tab, hit the START button.
14. Monitor the reagents over the course of the run as reagents
may need to be refilled (see Note 22).
15. Once the ResPep SL finishes the peptide synthesis, remove all
remaining reagents from the rack and discard in appropriate
waste containers (see Note 23).
3.3 Side Chain 1. At the end of the run, use a pencil to mark the top and bottom
Deprotection corners of the membrane and any other location required for
cutting the membrane (see Note 24).
2. In the fumehood, remove the membrane from the SPOT
synthesis frame and place it in a container labeled TFA.
3. Add 10 ml of deprotection solution to the container and let
react for 1 h in the fumehood (see Note 25). Following incu-
bation, discard in a TFA waste container.
4. Wash the membrane four times in 20 ml dichloromethane
(DCM) for 10 min per wash. Discard each wash in dichloro
waste container.
5. Wash the membrane four times in 20 ml DMF for 2 min per
wash. Discard waste in dichloro waste container.
6. Wash the membrane two times in 20 ml EtOH for 2 min per
wash. Discard in dichloro waste container.
7. Allow the membrane to dry in the fumehood until completely
dry. Wrap loosely in tin foil, and place in a labeled container in
the refrigerator until ready to perform the far western
experiment.
Peptide Arrays 315
3.4 Far-Western 1. In a gel box, soak the membrane in ethanol for 5 min and wash
Development (Fig. 3) twice with distilled water for 5 min on a shaker table at room
temperature.
2. Block the membrane for 1 h on a shaker in 20 ml of 5 % milk in
TBS/0.1 % Tween.
3. Wash the membrane five times for 5 min per wash in TBS/
0.1 % Tween on shaker table.
4. Incubate the membrane overnight with an appropriate concen-
tration of the protein of interest in protein binding buffer at
4 C (see Note 26).
5. Wash the membrane three times for 5 min per wash with an
appropriate wash solution on a shaker table
6. Incubate the membrane in 20 ml of primary antibody in pri-
mary antibody solution for 1 h (see Note 7).
7. Wash the membrane three times for 5 min/wash with an
appropriate wash solution on shaker table.
8. Incubate the membrane in 20 ml of horseradish peroxidase
(HRP) conjugated secondary antibody in secondary antibody
solution for 1 h (see Note 7).
9. Wash the membrane two times for 5 min/wash in an appropri-
ate wash solution on shaker.
10. Wash the membrane two times for 10 min/wash in TBS with
OG.
11. Visualize using HRP: in a container, combine 3 ml of each
enhanced chemiluminescence (ECL) reagent and mix well.
Place membrane in the container and cover with tin foil.
Swish container back and forth for 1 min. Transfer the mem-
brane to a plastic sheet cover and use a pipette to smooth away
bubbles. Set up the camera and place the membrane in the
machine to focus as per instructions. Set number of pictures,
exposures, etc. and take pictures (see Note 27).
12. Analyze data using a densitometry program (see Note 28).
3.5 Alanine Screen For the alanine screen, use the methods listed in Subhead-
(Fig. 4) ings 3.1–3.4. At Subheading 3.1, step 5, command “.replace, A”
is needed for alanine screening. This will generate overlapping
peptides as shown in Fig. 2b. The amount of amino acid derivatives
and reagents needed for the experiment will change according to
the peptide sequence, but the protocol is identical to that of general
peptide array synthesis.
Fig. 3 Far-Western technique to screen for interactions with a protein of interest. Peptides are synthesized on
a cellulose membrane, and binding interactions assessed using a far-western technique as stated in
Subheading 3. (a) Peptides on the membrane are washed with ethanol and water, and then the blot is
exposed to the protein of interest overnight. The next day, membranes are incubated with a primary antibody
for the binding partner of interest, and a HRP-conjugated secondary antibody. (b) Using enhanced chemilumi-
nescence (ECL) reagents that detect horseradish peroxidase (HRP) enzyme activity, peptides on membranes
are screened for interaction with potential binding partners. Selective peptides will bind to their binding
partners. Shown is a representative image of SNAP-25 peptides on a membrane exposed to Gβ1γ2. Using
image J and Prism, densitometry analysis is used to quantify the intensity of each dot, and values are plotted
via bar graph. Shown are representative densitometry results from the membrane examining SNAP-25 and
Gβ1γ2 interactions. The x-axis reflects both the peptide number and residue number of SNAP-25. Circled
regions, representing regions that interact with of the protein of interest, are used for subsequent alanine
screening (reproduced from Molecular Pharmacology December 2012 82:1136–1149 with permission of
American Society for Pharmacology and Experimental therapeutics)
Peptide Arrays 317
Fig. 4 Alanine Screening. Clusters of SNAP-25 peptides that are circled in red in Fig. 3b are tested via alanine
screen. By monitoring the change in the intensity of each dot, the alanine screening suggests residues that are
important for the protein-protein interaction. (a) Shown is a representative image of alanine screening for the
SNAP-25 130–143 peptide that was synthesized on a membrane. The first spot is a wild type peptide, while
the following 14 are mutated with a single alanine replacement of the residue at position 1–14 from wild type.
Boxes highlight residues which are shown to interrupt the SNAP25-Gβ1γ2 interaction by alanine mutation. (b)
The membrane is developed using the enhanced chemiluminescence (ECL) reagents. Densitometry is
quantified and plotted into a graph as shown above. When arginine residues in the SNAP25 130–143 peptide
are mutated to alanine, a significant decrease in the intensity of the dots is observed (*** p < 0.001)
(reproduced from Molecular Pharmacology December 2012 82:1136–1149 with permission of American
Society for Pharmacology and Experimental therapeutics)
4 Notes
1. ResPep SL equipment has two solvent connectors for dilutor 2.

DMF is connected to dilutor 2 solvent 1, and ethanol is
connected to dilutor2 solvent 2.
2. In the “REPORT” tab, the mass of each amino acid needed for
the prep is calculated. Calculations are for a 0.5 M solution of
each amino acid. Solutions are made according to the amount
calculated by dissolving the amino acid derivatives in N-methyl
pyrrolidone (NMP) overnight on a rotator in the fumehood.
3. In the “REPORT” tab, amounts of each reagent needed for the
prep are calculated. Reagents are made fresh on the day of
synthesis in DMF.
4. 0.1 % OG has to be added right before the solution is used.
5. Protein binding buffer depends on the protein you are using to
study the interaction and must be determine empirically. This
buffer cannot have high detergent levels as it will interfere with
the detection signal. Detergents have to be added right before

the solution is used.
6. The purified protein should be concentrated so it can be diluted
with the protein binding buffer when it is used in this
experiment.
7. Primary and secondary antibodies are chosen based on the
protein of interest. The amount of antibody needed for optimal
signal detection must be optimized.
8. The system will start in simulation mode if a computer is not
connected or the system is not switched ON. The following
procedure in SPOT method preparation section is specific to
ResPep SL from Intravis. If you are using different equipment,
please follow the manual of your equipment.
9. You can view the spaces or lines that you’ve introduced by
going to the “SHOW PEPTIDES” tab. This will show you
the position of each peptide for every protein, as well as the
number of spaces or lines that have been designated.
10. When pressing the “clean sequence” button, options such as
single letter format, delete spaces, etc. will show up. Ignore
these and press OK.
11. Peptide synthesis is a cyclic procedure with a number of steps
carried out for each amino acid addition. Step 1 is Fmoc
deprotection. Step 2 is DMF wash. Step 3 is ethanol wash.
Step 4 is air drying of the membrane. Step 5 is spotting of
activated amino acids and waiting for reaction time. Step 6 is
capping (optional). Step 7 is DMF wash. These steps are
repeated until the desired peptides have been assembled.
After peptide assembly, the side chain protection groups will
be removed using a side chain deprotection protocol.
12. Under the “DERIVATIVES” tab, the amount (in mg) of each
amino acid that you will need to weigh out for your synthesis,
the volume (in ml) of solvent needed to dissolve amino acids,
and the final volume you will end up with will be shown. It will
also have the amount of various reagents needed for the
experiment.
13. Check the robot arm is capable of moving to various locations
using the “go home,” “go to vial,” and “check vial positions”
buttons under the manual tab. Ensure the needle contacts the
membrane properly.
14. ResPep SL equipment is connected to dilutor 1 and dilutor 2.
Dilutor 1 represents the DMF bottle inside the equipment.
Dilutor 2 represents the DMF and ethanol bottles connected
outside of the equipment. During the peptide synthesis, DMF
is used to wash the membrane and ethanol is used to both wash
the membrane and start the drying process.
Peptide Arrays 319
15. For priming, allow depressing five times and emptying out
waste reservoir by turning on the vacuum pump and completely
draining into waste. For dilutor 2, there is solvent 1 (DMF) and
solvent 2 (EtOH), both solutions need to be primed. By click-
ing the dilutor 2 solvent 1 prime button, you can prime the
dilutor 2 solvent 1. Once it is finished, click the dilutor 2 solvent
2 prime button to prime the dilutor 2 solvent 2.
16. Only 2 ml ResPep tubes will fit onto the amino acid holder in
the ResPep equipment. Remember to label each cap and tube
appropriately.
17. Ensure that there are no bubbles or wrinkles on the membrane
once mounted on the SPOT membrane frame.
18. If there are bubbles or the membrane dries unevenly, remove
the membrane from the frame and repeat the DMF and wash
steps.
19. Put a hole on foil for the needle to go through.
20. Before placing the amino acid tubes onto the rack, uncap all
tubes.
21. These tubes will be used as mixing tubes for the activation of
amino acid derivatives.
22. During refilling, synthesis is paused by clicking the PAUSE key.
This will suspend the operation. Only do this when the needle
is away from the reagent rack and NOT pressing down on the
membrane or in an activation tube. A good time to pause is
when the software is counting down between piperidine depro-
tection steps.
23. Waste should be collected following the hazardous waste rules
at each institution. A carboy is connected to the ResPep SL
equipment during peptide array synthesis. There should be
separate waste containers for the deprotection solution and
washes in side-deprotection steps.
24. Use a pencil to mark the top and bottom corners of the mem-
brane. This will be important to identify the orientation of the
membrane later.
25. Ensure the lid is on for the container. Every 15 min or so, swish
solution around.
26. Protein and detergent concentrations have to be determined
empirically. Individual proteins may need different concentra-
tions to obtain optimal signals for analysis. The purity of
expressed proteins may also affect the amount of protein
needed for detection. For example, 0.4 μM Gβγ protein is
needed to identify Gβγ-SNARE interactions. Similarly, for
these experiments, the membrane was determined to tolerate
0.1 % OG.
27. Set number of pictures is 6; exposures are up to 60 s. Starting at

10 s, it takes a photo every 10 s. This depends on the strength
of signals.
28. Analysis programs, which have a densitometry analysis func-
tion, can be used to analyze the data. Image J can be used to
quantify the intensity of each dot. Each dot is normalized to the
most intense dot on the membrane and plotted into a bar graph
by prism.
References
1. Frank R (2002) The SPOT synthesis technique 6. Thiele A, Zerweck J, Schutkowski M (2009)
– synthetic peptide arrays on membrane sup- Peptide arrays for enzyme profiling. Methods
ports – principles and applications. J Immunol Mol Biol 570:19–65
Methods 267:13–26 7. Phizicky EM, Fields S (1995) Protein–protein
2. Min DH, Mrksich M (2004) Peptide arrays: interactions – methods for detection and anal-
towards routine implementation. Curr Opin ysis. Microbiol Rev 59:94–123
Chem Biol 8:554–558 8. Wells CA, Zurawski Z, Betke KM et al (2012)
3. Winkler DF, Hilpert K, Brandt O et al (2009) G beta gamma inhibits exocytosis via interac-
Synthesis of peptide arrays using SPOT- tion with critical residues on soluble N-ethyl-
technology and the CelluSpots-method. Meth- maleimide-sensitive factor attachment protein-
ods Mol Biol 570:157–174 25. Mol Pharmacol 82:1136–1149
4. Frank R, Doring R (1988) Simultaneous mul- 9. Mukhija S, Germeroth L, Schneider-Mergener J
tiple peptide-synthesis under continuous-flow et al (1998) Identification of peptides inhibiting
conditions on cellulose paper disks as segmen- enzyme I of the bacterial phosphotransferase
tal solid supports. Tetrahedron 44:6031–6040 system using combinatorial cellulose-bound
5. Dikmans A, Beutling U, Schmeisser E et al peptide libraries. Eur J Biochem 254:433–438
(2006) SC2: a novel process for manufacturing 10. Collet BY, Nagashima T, Yu MS et al (2009)
multipurpose high-density chemical microar- Fluorous-based peptide microarrays for prote-
rays. Qsar Comb Sci 25:1069–1080 ase screening. J Fluor Chem 130:1042–1048
Part III
Tag/Affinity-Based Methods
Chapter 19
Fluorescence Polarization Assay to Quantify

Protein-Protein Interactions: An Update
Ronald T. Raines
Abstract
A fluorescence polarization assay can be used to evaluate the strength of a protein-protein interaction. A
green fluorescent protein variant is fused to one of the protein partners. The formation of a complex is then
deduced from an increase in fluorescence polarization, and the equilibrium dissociation constant of the
complex is determined in a homogeneous aqueous environment. The assay is demonstrated by using the
interaction of the S-protein and S-peptide fragments of ribonuclease A as a case study.
Key words Fluorescence anisotropy, Fluorescence polarization, Fusion protein, Green fluorescent
protein, Protein-protein interaction
1 Introduction
Fluorescence polarization can be used to analyze macromolecular

interactions in which one of the reactants is labeled with a fluor-
ophore (see Note 1). In this assay, the formation of a complex is
deduced from an increase in fluorescence polarization, and the
equilibrium dissociation constant (Kd) of the complex is deter-
mined in a homogeneous aqueous environment (see Note 2).
Most fluorescence polarization assays have used a small molecule
such as fluorescein as a fluorophore [1–4].
Here, a variant (S65T) of green fluorescent protein (GFP) is
used as the fluorophore in a polarization assay [5–7]. The advan-
tages of using S65T GFP as the fluorophore are the ease with which
a protein can be fused to GFP by using recombinant DNA techni-
ques, the high integrity of the resulting chimera, and the broad
chemical and physical stability of GFP compared to small-molecule
fluorophores. To quantify the formation of a complex of two
proteins (X and Y), a GFP fusion protein (GFP–X) is produced by
using recombinant DNA technology (see Note 3).
Protein GFP–X is then titrated with protein Y, and the
equilibrium dissociation constant is obtained from the increase in
323
324 Ronald T. Raines
fluorescence polarization that accompanies complex formation.

Like a free fluorescein-labeled ligand, free GFP–X is likely to rotate
more rapidly and therefore to have a lower rotational correlation
time than does the GFP–X·Y complex. An increase in rotational
correlation time upon binding results in an increase in fluorescence
polarization, which can be used to assess complex formation [8].
In a fluorescence polarization assay, the interaction between the
two proteins is quantified in a homogeneous solution. The fluores-
cence polarization assay thereby allows for the determination of
accurate values of Kd in a wide range of solution conditions. GFP is
particularly well suited to this application because its fluorophore is
held rigidly within the protein, as revealed by the three-dimensional
structure of wild-type GFP and the S65T variant [9, 10]. Such a rigid
fluorophore minimizes local rotational motion, thereby ensuring
that changes in polarization report on changes to the global rota-
tional motion of GFP, as affected by a protein-protein interaction.
2 Materials
1. 20 mM Tris–HCl buffer (varying pH).

2. Solution of aqueous NaCl (varying concentration).
3. Purified GFP–X and purified protein Y.
4. Fluorometer equipped with polarization measurement
capability.
5. Graphics software capable of nonlinear regression analysis (e.g.,
DeltaGraph or SigmaPlot).
3 Methods
3.1 Fluorescence 1. Mix protein GFP–X (0.50–1.0 nM) with various concentrations
Polarization Assay of protein Y in 1.0 mL of 20 mM Tris–HCl buffer, pH 8.0, with
or without NaCl at 20 C (see Note 4). Conditions such as
buffer, pH, temperature, and salt can be varied as desired.
2. After mixing, take five to seven polarization measurements at
each concentration of protein Y (see Notes 5 and 6). For a
blank measurement, use a mixture that contains the same com-
ponents except for protein GFP–X.
3.2 Data Analysis 1. Fluorescence polarization (P) is defined as
I jj I ⊥
P¼ ð1Þ
I jj I ⊥
where I|| is the intensity of the emission light parallel to the

excitation light plane and I⊥ is the intensity of the emission
Fluorescence Polarization Assay 325
light perpendicular to the excitation light plane. P, the ratio of

light intensities, is a dimensionless number with a maximum
value of 0.5. Calculate values of Kd by fitting the data to the
equation:
ΔP F
P¼ þ P min ð2Þ
Kd þ F
In Eq. 2, P is the measured polarization, ΔP (¼Pmax – Pmin) is
the total change in polarization, and F is the concentration of
free protein Y (see Note 7).
2. Calculate the fraction of bound protein ( fB) by using the
equation
P P min F
fB ¼ ¼ ð3Þ
ΔP Kd þ F
Plot fB versus F to show the binding isotherms.
3.3 Case Study Fluorescence polarization was used to determine the effect of salt
concentration on the formation of a complex between the S15 and
S-protein fragments of ribonuclease A [11–15]. A GFP chimera of
S-peptide [S15–GFP(S65T)–His6] was produced from bacteria and
titrated with free S-protein. The value of Kd increased fourfold
when NaCl was added to a final concentration of 0.10 M (Fig. 1).
A similar salt dependence for the dissociation of RNase S had been
observed previously [16]. The added salt is likely to disturb the
Fig. 1 Fluorescence polarization assay of a protein-protein interaction: S15–GFP

(S65T)–His6 with S-protein. S-Protein is added to 20 mM Tris–HCl buffer, pH 8.0,
in a volume of 1.0 mL. Each data point is an average of 5–7 measurements.
Curves are obtained by fitting the data to Eq. 3. The values of Kd in the presence
of 0 and 0.10 M NaCl are 1.1 10–8 M and 4.2 10–8 M, respectively
326 Ronald T. Raines
water molecules hydrating the hydrophobic patch in the complex

between S15 and S-protein, resulting in a decrease in the binding
affinity [17]. Finally, the value of Kd ¼ 4.2 Å 10–8 M observed
in 20 mM Tris–HCl buffer (pH 8.0) containing NaCl (0.10 M) is
similar (i.e., threefold lower) to that obtained by titration calorim-
etry in 50 mM sodium acetate buffer (pH 6.0) containing NaCl
(0.10 mM) [18].
4 Notes
1. We used the term “polarization” instead of “anisotropy”

herein. Fluorescence polarization (P) and fluorescence anisot-
ropy (A) are related [A ¼ 2P/(3 – P)] and contain equivalent
physical information with respect to monitoring macromolec-
ular complex formation. Many instruments report on both
polarization and anisotropy and either parameter can be used
to evaluate Kd.
2. Polarization is proportional to the rotational correlation time,
which is defined as
3ηV
P /τ¼ ð4Þ
RT
In Eq. 4, rotational correlation time (τ) is the time taken for a
molecule to rotate 68.5 and is related to the solution viscosity
(η), molecular volume (V ), gas constant (R), and absolute
temperature (T ). Thus, under conditions of constant viscosity
and temperature, polarization is directly proportional to the
molecular volume, which increases upon complex formation.
3. Using the “superfolder” variant of GFP could facilitate the
creation of a GFP–X fusion protein [19].
4. In the assay solution, [GFP–X] should be significantly lower
than the value of Kd ([GFP–X] < < Kd) but still be high
enough to generate detectable fluorescence in the spectrome-
ter. In the case study, [GFP–X] ¼ 1 nM and Kd > 10 nM.
5. Data collection must be done at equilibrium. To estimate the
time to reach equilibrium, a pilot experiment can be performed
in which Y is added at [Y] ¼ Kd, and the polarization is moni-
tored until it reaches a stationary value.
6. At each [Y], the sample should be blanked with an identical
mixture that lacks GFP–X.
7. The change in polarization (ΔP) upon complex formation must
be detectable. For example, if the value of τ for GFP–X does
not change significantly upon formation of the GFP–X·Y com-
plex, then the value of ΔP is small and the data analysis is
difficult.
Fluorescence Polarization Assay 327
References
1. Jameson DM, Ross JA (2010) Fluorescence 11. Richards FM, Vithayathil PJ (1959) The
polarization/anisotropy in diagnostics and preparation of subtilisin modified ribonucle-
imaging. Chem Rev 110:2685–2708 ase and separation of the peptide and
2. Smith DS, Eremin SA (2008) Fluorescence protein components. J Biol Chem
polarization immunoassays and related methods 234:1459–1465
for simple, high-throughput screening of small 12. Watkins RW, Arnold U, Raines RT (2011)
molecules. Anal Bioanal Chem 391:1499–1507 Ribonuclease S redux. Chem Commun
3. Owicki JC (2000) Fluorescence polarization 47:973–975
and anisotropy in high throughput screening: 13. Richards FM (1958) On the enzymic activity of
perspectives and primer. J Biomol Screen subtilisin-modified ribonuclease. Proc Natl
5:297–306 Acad Sci U S A 44:162–166
4. Royer CA, Scarlata SF (2008) Fluorescence 14. Raines RT (1998) Ribonuclease A. Chem Rev
approaches to quantifying biomolecular inter- 98:1045–1065
actions. Methods Enzymol 450:79–106 15. Kim J-S, Raines RT (1993) Ribonuclease S-
5. Park S-H, Raines RT (1997) Green fluorescent peptide as a carrier in fusion proteins. Protein
protein as a signal for protein–protein interac- Sci 2:348–356
tions. Protein Sci 6:2344–2349 16. Schreier AA, Baldwin RL (1977) Mechanism of
6. Park SH, Raines RT (2000) Green fluorescent dissociation of S-peptide from ribonuclease S.
protein chimeras to probe protein–protein Biochemistry 16:4203–4209
interactions. Methods Enzymol 328:251–261 17. Baldwin RL (1996) How Hofmeister ion inter-
7. Park SH, Raines RT (2004) Fluorescence actions affect protein stability. Biophys J
polarization assay to quantify protein–protein 71:2056–2063
interactions. Methods Mol Biol 261:161–166 18. Connelly PR, Varadarajan R, Sturtevant JM
8. Jameson DM, Sawyer WH (1995) Fluores- et al (1990) Thermodynamics of protein–pep-
cence anisotropy applied to biomolecular inter- tide interactions in the ribonuclease S system
actions. Methods Enzymol 246:283–300 studied by titration calorimetry. Biochemistry
9. Ormö M, Cubitt AB, Kallio K et al (1996) Crys- 29:6108–6114
tal structure of the Aequorea victoria green fluo- 19. Pédelacq JD, Cabantous S, Tran T et al (2006)
rescent protein. Science 237:1392–1395 Engineering and characterization of a super-
10. Yang F, Moss LG, Phillips GN Jr (1996) The folder green fluorescent protein. Nat Biotech-
molecular structure of green fluorescent pro- nol 24:79–88
tein. Nat Biotechnol 14:1246–1251
Chapter 20
Förster Resonance Energy Transfer (FRET) Microscopy

for Monitoring Biomolecular Interactions
Alexa L. Mattheyses and Adam I. Marcus
Abstract
Förster or Fluorescence resonance energy transfer (FRET) can be used to detect protein-protein interactions.
When combined with microscopy, FRET has high temporal and spatial resolution, allowing the interaction
dynamics of proteins within specific subcellular compartments to be detected in cells. FRET microscopy
has become a powerful technique to assay the direct binding interaction of two proteins in vivo. Here, we
describe a sensitized emission method to determine the presence and dynamics of protein-protein interactions
in living cells.
Key words Fluorescence, FRET, Sensitized emission, Dynamics, Microscopy, Live cell imaging
1 Introduction
The spatial resolution of the light microscope is on the order of

several hundred nanometers. Therefore, standard fluorescence
microscopy techniques such as co-localization cannot establish
whether two proteins are directly interacting, only whether pro-
teins are in the same general area. Förster or Fluorescence reso-
nance energy transfer (FRET) can overcome this limitation and
directly detect dynamic protein-protein interactions in living sys-
tems. FRET only occurs over distances less than approximately
10 nm and when combined with microscopy provides the resolu-
tion necessary to observe protein-protein interactions with high
spatial and temporal resolution.
FRET is a non-radiative transfer of energy between two mole-
cules, in this case fluorophores [1]. Fluorophores are excited by
absorption of photons at one wavelength and emit photons at
a lower energy, longer wavelength. In FRET there are two fluoro-
phores, a donor and an acceptor, where an excited donor fluorophore
transfers energy to an acceptor fluorophore resulting in acceptor
emission. This non-radiative energy transfer does not involve emis-
sion of a photon from the donor and results from dipole coupling.
329
330 Alexa L. Mattheyses and Adam I. Marcus
a No FRET FRET
Spectral Overlap
Dex Dem Aex Aem Dex Dem Aex Aem
b
Distance
D A D A
>10nm <10nm
c
Orientation
A
A
D D
θ = 90˚ θ = 0˚
Fig. 1 Conditions for FRET include (a) A donor and acceptor pair where the emission spectra of the donor
overlaps with the excitation spectra of the acceptor (spectral overlap; grey shading). (b) The distance between
the donor and acceptor must be less than approximately 10 nm, at larger distances there is no FRET. (c) FRET
efficiency is highest when the excitation and emission dipoles are oriented parallel and there is no FRET if they
are perpendicular
There are three main conditions that need to be met for

efficient FRET (Fig. 1). First, there must be “spectral overlap” of
the donor’s emission and the acceptors excitation spectra (Fig. 1a).
Spectral overlap is a major consideration when the donor and
acceptor fluorophores (termed the FRET pair) are selected.
Second, the donor and acceptor fluorophores must be in close
proximity to one another. If more than 10 nm separates the
donor and acceptor, there is very little FRET; if they are within
10 nm, there is FRET (Fig. 1b). This distance dependence is the
critical factor that enables detection of protein-protein interactions.
Third, the dipoles of the donor and acceptor must be aligned. The
efficiency of transfer is modulated by the angle between the excita-
tion/emission dipoles, which is typically assumed to be random.
Orientation is critical in systems where the fluorophores are not
random but rigid with respect to one another: parallel dipoles have
the potential for energy transfer while perpendicular dipoles do not
(Fig. 1c).
FRET Microscopy 331
The efficiency of energy transfer (E) is modulated by the

distance between the two molecules (r):
1
E¼
1 þ ðr=R0 Þ6
where R0, also called the Förster radius, is the characteristic

distance between molecules where the FRET efficiency is 50 %.
The 1/r6 distance dependence allows FRET to be used as a molec-
ular ruler for distances close to R0 [2]. However, for most cell
biology applications detecting interactions between two proteins,
FRET is used as a binary readout: high FRET indicating the pres-
ence of an interaction and no FRET indicating the absence of
interaction.
The Förster radius can be calculated for any pair of fluorescent
molecules:
6
R0 ¼ 2:8 1017 κ 2 Q D E A J ðλÞ

where κ 2 is the orientation factor between the two fluorescent

dipoles, QD is the quantum yield of the donor, EA is the maximum
extinction coefficient of the acceptor, and J(λ) is the spectral overlap
integral between the donor and acceptor. For the fluorescent
protein FRET pairs R0 is typically 5 1 nm [3]. To assay
protein-protein interactions in cells, fluorescent proteins are most
commonly employed as the FRET donor and acceptor. Some com-
mon FRET pairs are listed in Table 1.
There are several basic approaches for measuring FRET in
microscopy, including acceptor photobleaching, fluorescence life-
time imaging microscopy (FLIM), spectral imaging, fluorescence
polarization, and sensitized emission [4–10]. This chapter
Table 1
Common FRET pairs
Donor Acceptor
Donor–acceptor Excitation Emission Excitation Emission

ECFP-EYFP 433 475 514 527
ECFP-Venus 433 475 515 528
Cerulean-Venus 433 475 515 528
mTFP-Venus 462 492 515 528
EGFP-mCherry 488 507 587 610
Venus-mCherry 515 528 587 610
Clover-mRuby2 505 515 559 600
describes a sensitized emission method for measuring FRET that

can be applied to dynamic systems. This basic technique can be
modified, for example to measure intramolecular FRET, the stoi-
chiometry of the FRET interaction, or protein conformation
changes [9, 11].
The sensitized emission method, also called 3-cube FRET, is
based on the detection of acceptor fluorescence after donor excita-
tion. Sensitized emission measurements can be achieved using a
standard confocal or widefield microscope with the appropriate
excitation sources and emission filters. Three images are acquired
for imaging donor, acceptor, and FRET channels (Fig. 2, Table 2).
In the donor image, the sample is excited at the donor excitation
wavelength and imaged with a donor emission filter. In the acceptor
image, the sample is excited at the acceptor excitation wavelength
and imaged with an acceptor emission filter. In the FRET image,
the sample is excited at the donor excitation wavelength and
imaged with an acceptor emission filter. Generally, it is not possible
to image the sensitized emission directly due to contamination of
the FRET image by both donor and acceptor. One source of
contamination in the FRET image is that the donor excitation
wavelength also excites the acceptor. Therefore, it is critical to select
an acceptor that has minimal excitation at the donor excitation
wavelength. Another source of contamination is donor fluores-
cence “bleed-through” that is detected in the acceptor emission
channel. Therefore, the “FRET” signal measured when the sample
is excited at the donor excitation wavelength and imaged at the
acceptor emission wavelength consists of three components: (1)
bleed-through from donor emission, (2) acceptor emission due to
direct excitation of the acceptor, and (3) sensitized emission due to
energy transfer. Thus, the measurement must be corrected for
bleed-through and direct excitation. These corrections require cal-
ibration factors that are measured with reference samples contain-
ing either donor or acceptor molecules alone, as described in detail
in Subheading 3.
The materials and methods are written to probe the interaction
of protein-A and protein-B with FRET using fusion proteins with
protein-A tagged with the donor CFP and protein-B tagged with
the acceptor Venus. Protein-A and protein-B can be any two
proteins of interest thought to interact based on biochemical or
other experiments. Other Cyan and Yellow variants (i.e., Cerulean,
Citrine) can be used in the constructs with no adjustments to the
microscope. Other suitable FRET pairs can be substituted with
appropriate adjustments to the constructs, filters, and lasers.
a Dex Dem Aex Aem
relative fluorescence
Idonor
Donor Excitation
Donor Emission
wavelength
b Dex Dem Aex Aem
Iacceptor
Acceptor Excitation
Acceptor Emission
wavelength
c Dex Dem Aex Aem

IFRET
Donor Excitation
Acceptor Emission
wavelength
Fig. 2 Microscope setup for FRET. Two lasers, 458 nm for donor excitation and
514 nm for acceptor excitation, are selected with an AOTF and directed into the
microscope by a polychroic mirror. After excitation of the sample, fluorescence is
collected by the objective lens and the fluorescence split based on wavelength
and imaged through a 480/20 filter (donor) or 535/30 filter (acceptor)
Table 2
Imaging parameters
Image Excitation Emission

Idonor Donor; 458 nm Donor; 480/20 nm
Iacceptor Acceptor; 514 nm Acceptor; 535/30 nm
IFRET Donor; 458 nm Acceptor; 535/30 nm
2 Materials
2.1 Sample There are three required and two optional control samples and
one experimental sample. Before transfection with fluorescently
tagged protein constructs cells should be plated in an appropriate
live cell imaging dish, such as MatTek #1.5 coverslip bottom dish
(MatTek Corporation, Ashland, MA). Fixed cell samples should
be plated on #1.5 coverslips, fixed and mounted for imaging (see
Note 1).
1. Required Control sample, donor only. Transfect cells with
cytosolic CFP (or Protein-A-CFP, see Note 2).
2. Required Control sample, acceptor only. Transfect cells with
cytosolic Venus (or Protein-B-Venus, see Note 2).
3. Required Control sample, medium only. Prepare a dish con-
taining no cells with the imaging medium.
4. Required Experimental sample. Transfect cells with protein-A-
CFP and protein-B-Venus. Ideally the expression level of the
proteins should be similar (see Note 3).
5. Suggested Control sample, FRET positive. Transfect cells with
CFP-Venus tandem protein.
6. Suggested Control sample, FRET negative. Transfect cells with
cytoplasmic CFP and cytoplasmic Venus.
2.2 Microscope This protocol describes data acquisition on a laser scanning confo-
cal microscope (Fig. 3). It can be easily adapted for widefield
microscopy (see Note 4).
1. Inverted laser scanning confocal fluorescence microscope with
excitation lasers 458 nm (CFP) and 514 (Venus) and emission
filters 480/40 (CFP) and 535/30 (Venus).
2. Objective lens: 60 1.4 NA oil immersion or similar high
resolution objective lens.
3. Environmental control: heated stage and optional CO2 to
maintain live cells on the microscope (not required for fixed
cell imaging).
2.3 Image Analysis 1. Image analysis software: Fiji (ImageJ pre-packaged with plu-
gins) (free download: fiji.sc/Fiji) or equivalent commercial
software.
FRET Microscopy 335
cells expressing
FRET probes
excitation
458 laser
lasers
514 laser
AOTF
polychroic
458 and 514
beamsplitter
PMT
514 LP
480/20
PMT
fluorescence
detection
535/30
Fig. 3 The three acquisition conditions for sensitized emission FRET. (a) Excite at
the donor wavelength and image at the donor wavelength (Idonor). (b) Excite at
the acceptor wavelength and image at the acceptor wavelength (Iacceptor). (c)
Excite at the donor wavelength and image at the acceptor wavelength (IFRET)
3 Methods
3.1 Optimize The imaging parameters must be held constant for acquisition of all
Imaging Conditions control and experimental images. First, you will optimize these
parameters for all three imaging channels. Use the experimental
sample expressing protein-A-CFP and protein-B-Venus.
1. Prepare the microscope. Select the 60 1.4 NA oil high-
resolution objective. Place a drop of immersion oil on the
objective and focus on the doubly labeled sample. Set the
temperature and CO2 environmental control (see Note 5).
2. Set the image dimensions. The zoom/magnification and num-
ber of pixels should be selected to achieve optimum resolution
as defined by Nyquist [12]. For our 60 1.4 NA objective this
is 90 nm/pixel. The ideal pixel size is determined by the optics
and the NA of the objective (see Note 6).
3. Set the scan speed and line averaging. For live cell imaging the
scan speed should be relatively fast and the line averaging low
to minimize sample movement artifacts.
4. Sequentially for each channel, set the acquisition parameters for
laser power, and detector gain and offset. The signal levels
for each channel should be roughly similar. There should be no

saturation in the image, and there should be some “room” to
accommodate an increase or decrease in fluorescence as the
FRET signal dynamically changes. Check for and minimize
photobleaching.
(a) Optimize the Idonor channel. Excite the sample at the donor
excitation wavelength 458 nm, and select the donor emis-
sion filter 480/20.
(b) Optimize the Iacceptor channel. Excite the sample at the
acceptor excitation wavelength 514 nm, and select the
acceptor emission filter 535/30.
(c) Optimize the IFRET channel. Excite the sample at the donor
excitation wavelength 458 nm, and select the acceptor
emission filter 535/30.
3.2 Data Collection All images are collected with the parameters established in
Subheading 3.1.
1. Image the required control for donor bleed-through. Imaging
the CFP only sample, collect Idonor and IFRET for 10–20 indi-
vidual cells.
2. Image the required control for acceptor bleed-through. Imag-
ing the Venus only sample, collect Iacceptor and IFRET for 10–20
individual cells.
3. Image the required control for background. Imaging the
medium only sample, collect Idonor, Iacceptor, and IFRET from
20 independent positions. Ensure the microscope is focused in
the medium (see Note 7).
4. Image the experimental sample. Imaging the protein-A-CFP
and protein-B-YFP sample, collect Idonor, Iacceptor, and IFRET.
A time lapse should be set up to record dynamics.
Suggested controls
5. Image the recommended FRET positive control. From the
CFP-Venus tandem protein sample collect Idonor, Iacceptor, and
IFRET.
6. Image the recommended FRET negative control. Imaging the
cytosolic CFP and Venus sample, collect Idonor, Iacceptor, and
IFRET.
3.3 Image All image processing steps can be performed with Fiji (see Note 8).
Processing and The menu navigation in Fiji for the required functions is in Table 3.
Analysis Images are referred to by the sample and the imaging condition, for
example CFPIdonor is the CFP sample acquired with donor imaging
conditions.
1. Calculate background from the medium only Idonor, Iacceptor,
and IFRET images. Background is a 20 frame average of the
medium only sample:
FRET Microscopy 337
Table 3
ImageJ/Fiji operations
Operation Menu location

Adjust LUT Image ! Adjust ! Brightness/contrast
Apply LUT Image ! Look up tables
Image math Process ! Image calculator
Math Process ! Math
Measure Image ! Measure
Open images Plugins ! LOCI ! Bioformats importer
Save ROIs Analyze ! Tools ! ROI manager
Stack average Image ! Stacks ! Z project. . .; projection type ¼ average intensity
BGdonor ¼ AverageðmediumI donor Þ

BGacceptor ¼ Average mediumI acceptor
BGFRET ¼ AverageðmediumI FRET Þ
2. Subtract BGdonor from all Idonor images, BGacceptor from all

Iacceptor images, and BGFRET from all IFRET images. All
subsequent steps assume background subtracted images
(see Note 9).
3. Calculate the contribution factor α of directly excited acceptor
fluorescence in the FRET channel from the Venus only images.
The result should be a 32-bit float image.
VenusI FRET
α¼
VenusI acceptor
Select regions of the image where there was Venus fluores-

cence and measure the average α. Measure each field of view
and average all α to obtain the final bleed-through correction
value α (see Note 10).
4. Calculate the contribution factor β of donor fluorescence
bleed-through in the FRET image from the CFP only images.
The result should be a 32-bit float image.
CFPI FRET
β ¼
CFPI donor
Select regions of the image where there was CFP fluores-
cence and measure the average β. Measure for each field of view
and average all β to obtain the final bleed-through correction
value β.
5. Calculate corrected FRET (NFRET) for experimental and

control data. The result should be a 32-bit float image.
I FRET αI acceptor βI donor
N FRET ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
I acceptor I donor
4 Notes
1. This protocol can be adapted for fixed cell samples that do not
require labeling with antibodies. Transfect samples with donor
and acceptor fusion proteins as described. Following adequate
expression time, fix with 4 % paraformaldehyde and mount the
cells on slides for microscopy.
2. The donor and acceptor only controls are needed to measure the
bleed-through of direct excitation and emission into IFRET. Cyto-
solic expression of CFP and Venus will simplify accurate measure-
ments of these bleed-through contributions; however, protein-
A-CFP and protein-B-Venus fusion proteins can be utilized.
3. The placement of the fluorescent tags in the fusion protein is
critical for FRET. A negative “no FRET” result does not rule
out protein interaction. If the distance between the donor and
acceptor is too great (opposite sides of a complex) or if the
orientation is not optimal, there may be no FRET even when
protein-A and protein-B are in fact interacting. The labeling
strategy should be carefully considered and different tag loca-
tions explored [13].
4. Sensitized emission FRET measurements can be conducted on
a widefield or epi-fluorescence microscope. The widefield
microscope should be equipped with a fluorescence light source
(Hg lamp or LED), a scientific CCD or CMOS camera, and
three filter cubes corresponding to the three imaging channels
(Table 2). For CFP-Venus FRET the three filter cubes should
be: Donor Cube: 430/20 ex 480/30 em; Acceptor Cube:
500/20 ex 535/30 em; and FRET Cube: 430/20 ex 535/
30 em. The protocol can be followed substituting the appro-
priate filter cube for Idonor, Iacceptor, and IFRET.
5. To reduce background, the cell imaging medium should be
phenol-red free. When observed with widefield fluorescence,
the sample is photobleaching. Therefore it is important to
minimize the time the fluorescence lamp shutter is open and
to use neutral density (ND) filters to attenuate the lamp illumi-
nation intensity.
6. The Nyquist sampling theorem is commonly utilized to deter-
mine the ideal pixel size for optimal high-resolution imaging.
7. To focus in the medium it is easiest to focus on some dust or
particles on the coverslip and then up into the medium. Make
sure the areas imaged do not contain any non-uniformities.
FRET Microscopy 339
8. This analysis can be performed in many different imaging soft-

ware. Macros can be developed in Fiji and there are plugins
available for ImageJ/Fiji that streamline the analysis (i.e., pix-
FRET; free download: http://www.unil.ch/cig/page16989.
html).
9. If the illumination intensity varies across the field of view, a
shade correction, obtained by imaging a uniform fluorescence
sample, should be applied.
10. The bleed-through of acceptor and donor (α and β) are often
considered constants. This is generally true, however, intensity
correlations have been reported. If the intensities of the sample
varies significantly or if the intensity of the controls is signifi-
cantly different than that of the sample this should be tested.
If an intensity correlation exists, it can be corrected by fitting
the bleed-through values.
Acknowledgements
The authors thank Dr. Claire E. Atkinson for critical reading of

the manuscript. This work was supported by NIH grant
1RO1CA142858 awarded to A.I.M. and NIH grant
1R21AR066920 awarded to A.L.M.
References
1. Lakowicz JR (1988) Principles of frequency- 8. Piston DW, Rizzo MA (2008) FRET by fluo-
domain fluorescence spectroscopy and applica- rescence polarization microscopy. Methods
tions to cell membranes. Subcell Biochem Cell Biol 85:415–430
13:89–126 9. Hoppe A, Christensen K, Swanson JA (2002)
2. Stryer L (1978) Fluorescence energy transfer as Fluorescence resonance energy transfer-based
a spectroscopic ruler. Annu Rev Biochem stoichiometry in living cells. Biophys J
47:819–846 83:3652–3664
3. Lam AJ, St-Pierre F, Gong Y et al (2012) 10. Gordon GW, Berry G, Liang XH et al (1998)
Improving FRET dynamic range with bright Quantitative fluorescence resonance energy
green and red fluorescent proteins. Nat Meth- transfer measurements using fluorescence
ods 9:1005–1012 microscopy. Biophys J 74:2702–2713
4. Gautier I, Tramier M, Durieux C et al (2001) 11. Goedhart J, Hink MA, Jalink K (2014) An
Homo-FRET microscopy in living cells to introduction to fluorescence imaging techni-
measure monomer-dimer transition of GFP- ques geared towards biosensor applications.
tagged proteins. Biophys J 80:3000–3008 Methods Mol Biol 1071:17–28
5. Mattheyses AL, Hoppe AD, Axelrod D (2004) 12. North AJ (2006) Seeing is believing? A begin-
Polarized fluorescence resonance energy trans- ners’ guide to practical pitfalls in image acqui-
fer microscopy. Biophys J 87:2787–2797 sition. J Cell Biol 172:9–18
6. Padilla-Parra S, Tramier M (2012) FRET micros- 13. Miyawaki A, Tsien RY (2000) Monitoring
copy in the living cell: different approaches, protein conformations and interactions by
strengths and weaknesses. Bioessays 34:369–376 fluorescence resonance energy transfer
7. Piston DW, Kremers GJ (2007) Fluorescent between mutants of green fluorescent protein.
protein FRET: the good, the bad and the Methods Enzymol 327:472–500
ugly. Trends Biochem Sci 32:407–414
Chapter 21
Utilizing ELISA to Monitor Protein-Protein Interaction

Zusen Weng and Qinjian Zhao
Abstract
Enzyme-linked immunosorbent assay (ELISA) is a commonly used method in analyzing biomolecular
interactions. As a rapid, specific, and easy-to-operate method, ELISA has been used as a research tool as well
as a widely adopted diagnostic method in clinical settings and for microbial testing in various industries.
Inhibition ELISA is a one-site binding analysis method, which can monitor protein-protein interactions in
solution as opposed to more commonly used sandwich ELISA in which the analyte capture step is required
on a solid surface either through specific capture or through passive adsorption. Here, we introduce
inhibition ELISA procedures, using a recombinant viral protein as an example, with emphasis on how
inhibition ELISA could be used to probe subtle protein conformational changes in solution impacting
protein–protein binding affinity. Inhibition ELISA is used to probe one binding site at a time for binding
partners in solution with unrestricted conformation. The assay can be performed in a quantitative manner
with a serially diluted analyte in solution for solution antigenicity or binding activity assessment.
Key words Inhibition ELISA, Monoclonal antibody, Antigen–antibody interaction, The half-
maximal inhibitory concentration (IC50), Conformational change, Disulfide bond, Binding affinity,
Binding analysis in solution
1 Introduction
Enzyme-linked immunosorbent assay (ELISA) is a popular immu-

nochemical assay [1] that uses a solid-phase enzyme immunoassay
(EIA, or an earlier and more sensitive method, radiolabeled immu-
noassay commonly referred to as RIA) [2, 3] to detect the presence
of a substance in a liquid sample. The specific binding was normally
achieved through highly specific antigen–antibody interactions.
ELISA has replaced RIA as the most popular method for analyzing
a specific interaction due to the lack of need to use radiolabeled
compounds. The ELISA was first reported in the 1960s [4] and
now has been used as a diagnostic tool in medicine and plant
pathology, as well as a quality-control assays in various industries
and in quarantine laboratories at border controls and during pan-
demics [5, 6]. There are many ways one can perform an ELISA to
probe a specific interaction in a qualitative or quantitative way.
341
342 Zusen Weng and Qinjian Zhao
In one example of ELISA, an antigen is affixed to a solid surface or

onto synthetic beads through passive adsorption, and then a specific
antibody or sample is applied over the surface so that it can bind to
the antigen via a highly specific interaction. This detection antibody
(or a secondary antibody) is covalently linked to an enzyme (the
most common examples are horseradish peroxidase, HRP, and
alkaline phosphatase, AP), and in the final step a substance is
added that the enzyme can convert a substrate to a different form
with detectable absorption or fluorescence signals [7–9].
There are several different types of ELISA, such as direct bind-
ing ELISA on Ag-coated plate, sandwich ELISA, competitive
ELISA and so on. These methods are usually used to monitor
antigen–antibody interaction through the detectable signals in the
assay with an enzymatic reaction for amplification in the final step.
Similarly, these methods can also be used to monitor two interact-
ing proteins such as a protein and its receptor or an enzyme and its
effector, and are not limited to an antigen–antibody pair. As a
practical example, a competitive ELISA (or inhibition ELISA) was
performed by Cuervo et al. to monitor the interaction between the
Hepatitis B virus surface antigen (HBsAg) lots and polyclonal Ab
for quality control on product consistency of a commercial vaccine
[10]. The inhibition ELISA is a kind of competitive ELISA where a
pair of binding partners is premixed prior to transferring to the
assay plate to detect the residual labeled partner [11, 12]. The
samples in solution to be detected in the inhibition ELISA compete
for protein binding sites with the same protein coated on a solid
surface, namely, the higher the binding activity in solution, the less
the labeled binding partner left available for the surface immobi-
lized partner to bind [13].
There are two unique characteristics for the inhibition ELISA we
introduce here: it monitors a single epitope at a time as opposed two
different epitopes in more commonly used sandwich ELISA [14];
and the test protein is interacting in solution with no restriction on its
conformation as opposed to binding events on a surface-adsorbed
protein as in the case for most ELISAs [15]. Therefore, it reflects
more faithfully the biological interactions owing to fewer artifacts for
this solution interaction based method and no wash cycles disrupting
the interaction being probed [16]. Since there is no need for the
analyte to be bound onto solid surface and to survive the wash cycles
of most ELISAs [17], the inhibition ELISA, unlike dot blot, Western
blot, direct binding ELISA and Biacore assay, can also be used on
particulate proteins, such as a protein with the binding sites (i.e.,
epitopes) of interests on a cell surface or an antigen adsorbed onto
particulate adjuvant in a vaccine formulation [18].
In this chapter, as an example of inhibition ELISA, we moni-
tored the solution interaction of a pair of interacting proteins
(recombinant HBsAg and a monoclonal antibody IgG1 5F11) in
a quantitative manner. The binding strength can be quantitatively
Protein-Protein Interaction by ELISA 343
described by IC50 (the half-maximal inhibitory concentration)

values of the analyte in solution [19]. IC50 is a measure of the
effectiveness of a test sample (or a reference) in solution in binding
to a biomolecule as a tracer in the assay. The larger an IC50 value
(normally expressed in ng/mL) is, the weaker the interaction
between an HBsAg sample and 5F11 in solution.
Since recombinant HBsAg is a cysteine-rich protein and the
epitopes on HBsAg are highly dependent on the correct disulfide
bind pairings [20], we tested the binding activity of 5F11 to
HBsAg after the reduction of the disulfide bonds during pretreat-
ment to probe the effect on the binding affinity. To reduce the
disulfide bonds, the HBsAg are treated with 0.5 mM dithiothreitol
(DTT). Subsequently, DTT was removed by dialysis to allow a
certain degree of maturation (HBsAg would spontaneously
undergo oxidative maturation after DTT removal) [20–22]. The
epitope was found to mature back to some degree after the removal
of the added reductant. We demonstrated that 0.5 mM DTT
treated HBsAg and “Post DTT removal HBsAg with oxidative
maturation” showed lower affinity (68- and 8-fold weaker, respec-
tively) to 5F11 through their IC50 change relative to the untreated
HBsAg control. The usefulness of this convenient solution-based
inhibition ELISA method to monitor protein-protein interaction
was demonstrated in a quantitative way.
2 Materials
2.1 Solutions Prepare all solutions using double-distilled or deionized water

(dd- or di-H2O) and analytical grade reagents. Prepare and store
all reagents at 4 C (unless indicated otherwise). The reagents
should be prepared as listed below.
1. Carbonate buffer (CB, 20 ): 0.3 M Na2CO3, 0.7 M
NaHCO3. Weigh 31.8 g of Na2CO3 and transfer to a 1-L
graduated cylinder or a glass beaker. Add 500 mL deionized
water and mix with a glass bar until the Na2CO3 is completely
dissolved. Then add 58.8 g of NaHCO3 to the above solution
and bring up to 1 L with dd-H2O. Mix until no solid can be
seen in the solution.
2. Saturated ammonium sulfate: Transfer 90 g of ammonium
sulfate to a 100 mL graduated cylinder. Bring up to 100 mL
with dd-H2O. Incubate the solution at 80 C in a water bath
and dissolve the ammonium sulfate by stirring until the solid is
completely dissolved. Incubate at 4 C until the solution
reaches 4 C. Adjust the pH to 7.4 with 1.0 M NaOH (see
Note 1).
3. Phosphate buffered saline (PBS) (10 ): 20 mM KH2PO4,
0.1 M Na2HPO4, 1.37 M NaCl, 27 mM KCl. Weigh 2.7 g of
KH2PO4, 14.2 g of Na2HPO4, 80.0 g of NaCl, and 2.0 g of

KCl and transfer to a 1-L graduated cylinder. Add dd-H2O to a
volume of 800 mL. Mix and adjust the pH to 7.4 with HCl.
Bring up to 1.00 L with dd-H2O.
4. Blocking buffer: PBS (2 ), 0.5 % casein, 2 % gelatin, 0.1 %
ProClin. Add 100 mL of PBS (10 ) to a 1-L graduated cylin-
der. Weigh 5.0 g of casein and transfer to the cylinder. Incubate
gelatin at 37–40 C for 10 min to make the gelatin less sticky.
Then add 20 mL of gelatin to the PBS (see Note 2). Add 1 mL
of ProClin to the above solution and bring up to 1 L with dd-
H2O. Mix the solution with a magnetic stirring apparatus until
all reagents are dissolved (see Note 3). Store the blocking
buffer at 4 C in a well-sealed bottle.
5. Assay diluent: 20 mM Tris–HCl (pH 8.0), newborn calf serum,
1 % casein, 10 % sucrose. Dissolve 2.42 g of Tris base to 100 mL
dd-H2O and adjust the pH to 8.0 with HCl. Then add 40 mL
of newborn calf serum, 10 g of casein, and 100 g of sucrose to
the solution. Bring up to 1 L with dd-H2O (see Note 4). Mix
the solution using a magnetic stirring apparatus until all
reagents are dissolved.
6. PBS-T (wash buffer): PBS containing 0.05 % (w/v) Tween 20.
Add 1 mL of Tween 20 to 2 L 1 PBS.
2.2 Antigens and 1. Purified HBsAg: The HBsAg (NIDVD, Xiamen, China) was
Conjugates expressed in CHO (Chinese hamster ovary) cells and purified
to >95 % purity. The concentration of HBsAg was 1.0 mg/mL.
2. Purified HBsAg conformational monoclonal antibody IgG1
5F11 (see Note 5).
3. Horseradish peroxidase (HRP), DTT, NaIO4, glycol, and
NaBH4 were all bought from Sigma.
4. TMB (Tetramethylbenzidine) Reagent A and TMB Reagent B
(INNOVAX, Beijing, China).
5. TMB Stop Solution: 2 M H2SO4 (Beijing Wantai Biopharma-
ceuticals, Beijing, China).
3 Methods
3.1 HRP (Horseradish 1. Dialyze 4 mL of 5F11 solution (1 mg/mL) in 1 L of 50 mM

Peroxidase) CB (1 ). Stir with a stirrer at 4 C for 2 h. Change the CB to
Conjugation to newly prepared CB twice. Each time stir for 2 h with a magnetic
Antibody 5F11 stirring apparatus.
2. Dissolve 0.2 g of HRP in 10 mL of dd-H2O and dissolve 0.2 g
of NaIO4 in 10 mL dd-H2O (the final concentration of both
HRP and NaIO4 is 20 mg/mL).
3. Mix 200 μL of 20 mg/mL HRP and 200 μL of 20 mg/mL

NaIO4 gently (see Note 6). Incubate the mixture at 4 C for
30 min.
4. Add 4 μL of glycol to above mixture slowly with gently mixing
(see Note 7). Incubate the solution in a dark place at room
temperature for 30 min.
5. Add the solution in step 4 to the dialyzed 5F11 solution (see
Note 8) and dialyze the resulting mixture in 1 L of 50 mM CB
three times, each time for 2 h at 4 C.
6. Dissolve 0.2 g of NaBH4 in 10 mL of dd-H2O (see Note 9).
Add 40 μL of NaBH4 to the dialyzed mixture in step 5.
Incubate the mixture in a dark place at 4 C for 2 h. Mix
every 30 min.
7. Slowly add an equal volume of saturated ammonium sulfate
solution to the mixture in step 6 while stirring (see Note 10).
After sufficient mixing, incubate the mixture at 4 C for 4 h.
8. Transfer the mixture to a 4 mL Eppendorf tube and centrifuge
at 14,000 g for 10 min. Remove the liquid carefully.
9. Add 200 μL of newborn calf serum to 1.8 mL of dd-H2O and
mix with 2 mL of glycerol. Dissolve the precipitate in step
8 with the mixed solution. The final mixture should contain
1.0 mg/mL of 5F11-HRP (see Note 11). Store at 20 C.
3.2 Determination of 1. Dilute the HBsAg stock to 1.0 μg/mL with PBS for plate
the Proper Working coating.
Concentration of 5F11- 2. Add 100 μL to each well on 96-well plates. Seal the plates and
HRP incubate at 37 C for 6 h.
3. Wash the 96-well plates with 300 μL of PBS-T per well. Tap the
plates on bibulous paper when the wells point down until the
wells are dry.
4. Add 200 μL of assay diluent to each well on 96-well plates (see
Note 12). Seal the plates with plate sealers and incubate at
37 C for 2 h.
5. Wash the 96-well plates with PBS-T and tap again. Seal the
plates and store the plates at 4 C.
6. Dilute 1 μL of 1.0 mg/mL 5F11-HRP solution in 1 mL of
PBS. Then dilute the diluent at serial twofold dilutions (see
Note 13).
7. Transfer 100 μL of the diluents in step 6 to each well on the
plates in step 5. Seal plates and incubate at 37 C for 1 h.
8. Wash the 96-well plates with 300 μL of PBS-T per well five
times. Then tap the plates until the wells are dry as mentioned
before.
Fig. 1 The ELISA profile of different concentration of 5F11-HRP in a direct

binding assay to determine the working concentration of the enzyme
conjugate. The 5F11-HRP was diluted using a set of twofold serial dilutions in
order to determine the appropriate working concentration/dilution in the assay.
The initial concentration of 5F11-HRP is 2.0 ug/mL
9. Mix 6 mL of TMB Reagent A with 6 mL TMB Reagent B for

each plate prior to use. Transfer 100 μL to each well. Incubate
at 37 C (5–20 min) for color development. To stop the color
development, add 50 μL of TMB Stop Solution (see Note 14).
10. Read the optical density (OD) for each well with a microplate
reader set to 450 nm. Analyze the results and determine the
appropriate concentration of 5F11-HRP used (see Note 15)
(Fig. 1).
3.3 Inhibition ELISA The flowchart of solution inhibition ELISA is shown in Fig. 2. The
protocol for inhibition ELISA below should correspond to the
flowchart shown (Fig. 2).
1. Dilute the 5F11-HRP with assay diluent to 10 ng/mL.
2. Dilute the HBsAg samples used to detect twofold serial
dilutions—starting with 4 105 ng/mL (see Note 16). Each
sample should be more than 60 μL.
3. Mix 60 μL of diluted HBsAg sample with 60 μL of diluted
5F11-HRP in each well on plates (see Note 17). Seal the plates
and incubate at 37 C for 30 min. Then centrifuge the plates
for 5 min at 1,500 g.
4. Transfer 100 μL of the supernatant in step 3 to each well of
plates from step 5 of Subheading 3.2 (see Note 18). Seal the
plates and incubate at 37 C for 1 h.
5. Wash the 96-well plates with 300 μL of PBS-T per well five
times. Then tap the plates until no residual liquid remains, as
mentioned before.
Fig. 2 Flowchart of a solution inhibition ELISA. The reagent HBsAg passively adsorbed on the surface in the
microwells is shown black. The serially diluted analyte in solution, the HBsAg reference or test sample, are
shown in gray. The IC50 range of analyte to reference indicates binding affinity changes to antibody 5F11 (see
Note 20)
6. Mix 6 mL of TMB Reagent A with 6 mL of TMB Reagent B for

each plate immediately prior to use. Transfer 100 μL to each
well. Incubate at 37 C for color development as mentioned
before. Add 50 μL of TMB Stop Solution to stop the color
reaction.
7. Read the optical density (OD) for each well with a microplate
reader set to 450 nm. Analyze the results and monitor the
interaction between HBsAg and 5F11 through IC50 shift (see
Note 19) (Fig. 3).
4 Notes
1. The temperature of the water bath can range from 60 C to

90 C since the purpose of this step is to dissolve the ammo-
nium sulfate at a high temperature. During incubation at 4 C,
Fig. 3 Inhibition ELISA curves for different HBsAg samples (i.e., HBsAg refer-
ence, 0.5 mM DTT treated HBsAg, Post DTT removal HBsAg with oxidative
maturation). Fitted curves and the IC50 values were obtained using “Prism 5.”
The IC50 values for HBsAg reference, 0.5 mM DTT treated HBsAg, Post DTT
removal HBsAg with oxidative maturation are 97, 6,572, and 810 ng/mL,
respectively (or relative IC50 of 1, 0.015, and 0.12, respectively). Nearly tenfold
increase in solution antigenicity was observed as a results of HBsAg spontane-
ous maturation
the ammonium sulfate precipitates out. Ammonium sulfate

solid can be seen at the bottom of the solution. 1.0 M NaOH
can be used to adjust the solution pH. When making the
NaOH solution, be aware of heat release during the dissolution
of solid NaOH.
2. Gelatin is stored at 4 C when it is solid. It can melt to liquid at
a temperature of 37–40 C. The measuring cylinder used to
measure gelatin always contains residual gelatin on the walls.
The residual gelatin should be washed with dd-H2O and trans-
ferred to the solution to ensure the volume of gelatin is
accurate.
3. The stirring time should be more than 1 h since it is difficult to
dissolve casein powder. It is better that the rotor is set at high
rotate speed.
4. If desired, a pH indicator such as phenol red can be added to
the solution to indicate pH during storage or experiments.
5. The ELISA with conformationally sensitive monoclonal anti-
body can be used to monitor the integrity of protein binding
sites [21], which is critical in quality control of pharmaceuticals
or vaccines. When the binding affinity to a conformationally
sensitive antibody weakens, the epitope integrity of the protein
analyte may have been compromised.
6. This step is to oxidize the glycan components on HRP using
NaIO4. Mixing should be gentle. The weight of HRP and
NaIO4 in the mixture should be equal.
7. The step is to terminate the oxidation of HRP, volume of glycol

(Vglycol, μL)–weight of HRP (MHRP, mg) ¼ 1:1.
8. The weight of antibody (M5F11, mg)–the weight of HRP
(MHRP, mg) ¼ 1:1. The experiment in this chapter is to moni-
tor protein interactions through inhibition ELISA in solution
with one binding partner labeled with HRP.
9. The NaBH4 solution should be prepared immediately prior to
use. NaBH4 is a strong reductant, making the protein–protein
conjugate into a more stable form.
10. Saturated ammonium sulfate should be added slowly with
stirring at 4 C. It is used to salt out the 5F11-HRP.
11. The concentration of 5F11-HRP is reported as the 5F11 con-
centration as it is hard to quantify the HRP conjugate. Loss of
5F11 is considered minimal in the process of HRP conjugation
procedures.
12. This step is to block the nonspecific binding sites in the micro-
wells. The blocking buffer should be added at a larger volume
than the coating HBsAg solution so that blocking buffer can
cover the walls that HBsAg-coated areas.
13. Serial dilution of 5F11-HRP should cover a wide enough range
so that an appropriate concentration can be determined. A
negative control should be used to monitor the ELISA
background.
14. The color reagent solution turns blue as a result of surface
bound HRP catalysis. The solution turns yellow after adding
acidic TMB Stop Solution. The same color development time
should be used throughout all the experiments.
15. In this experiment, the initial concentration of 5F11-HRP is
2,000 ng/mL; it can be larger when the proper initial concen-
tration is unknown. The concentration of 5F11-HRP we deter-
mined is 5.0 ng/mL because the OD450 that correlates to this
concentration is about 1.5. An OD450 that is too large leads to
inaccurate measurement due to too little light transmitted. An
antibody concentration that is too high could lead to incom-
plete inhibition, while too low of an antibody concentration
would result in low signals.
16. The test HBsAg samples could be different production lots or
different samples that underwent different pretreatments, such
as 0.5 mM DTT-treated. Reference (HBsAg sample without
any pretreatment) should be used as a control analyte for
quantitatively analyzing the changes in IC50 values. Duplicate
wells are used to improve assay reproducibility.
17. The plates used here are non-adsorptive plates. Other plates
mentioned in this chapter are adsorptive plates. If the test
analyte were allowed to bind to the binding partner first in
solution (prior to being transferred to the assay plate), the

inhibition ELISA can also be used to determine the true solu-
tion affinity (KD) constant of biologically relevant binding
partners such as antibody–antigen or receptor–ligand
pairs when reaching equilibrium for all the mixtures is
needed [23]. There is no need to reach equilibrium to deter-
minate the IC50.
18. The labeled binding partner and a serially diluted protein ana-
lyte can also be added directly to the plates without premixing.
This is more commonly referred to as “competitive ELISA” as
the labeled molecules bind to either the partner in solution or
to the “competing” partner on the solid surface. It is an alter-
native method to monitor protein-protein interactions; how-
ever, different IC50 values will be obtained when compared to
the method described in this chapter.
19. The binding curve (as indicated by OD450) is shown in Fig. 3.
The IC50 (the half-maximal inhibitory concentration) can be
calculated by software such as “Prism 5”. IC50 is a measure of
the effectiveness of a sample in inhibiting biological or bio-
chemical function. The larger an IC50 is, the lower affinity a
HBsAg sample has to the same antibody. The HBsAg reference
is an untreated sample. The HBsAg samples were also treated
with 0.5 mM DTT and the DTT was later removed by dialysis.
The results showed that the 0.5 mM DTT treated HBsAg has
the largest IC50, whereas the IC50 of HBsAg reference is the
lowest (differing by 68-fold). This means that 0.5 mM DTT
treatment made the interaction between HBsAg and 5F11
much weaker. After DTT removal, only a fraction of the 5F11
binding activity of HBsAg was observed in “Post DTT removal
HBsAg with oxidative maturation.” The relative IC50 (Refer-
ence IC50/Sample IC50) can also be used to show the IC50
range between sample and reference. The interaction between
two interacting proteins, with one binding partner labeled with
an enzyme or a fluorophore/chromophore [7], can be moni-
tored in a similar way, particularly when the binding site could
be changed subtly but with an impact on the binding affinity.
20. Since the immune complex of analyte and labeled binding
partner forms in solution and there is no need for such a
complex to bind onto a surface, depletion of the labeled bind-
ing partner is the measure of the analyte binding activity [18].
Therefore, inhibition ELISA can be employed in the study of
active binding site in whole cells or particulate vaccine antigens
on adjuvants. Serial dilution of the analyte enables the quanti-
tative analysis of the binding activity of different lots or pre-
parations of a given protein. With a sensitive probe, impact on
the binding affinity due to some intentional manipulations
could be monitored using the assay described here.
Acknowledgement
The work on chapter writing was enabled with the funding of

Chinese Ministry of Science and Technology 863 Project
(2012AA02A408) and National Science Foundation of China
(81471934) to Q.Z.
References
1. Leng SX, McElhaney JE, Walston JD et al to the standardization process. Biologicals
(2008) ELISA and multiplex technologies for 32:171–176
cytokine measurement in inflammation and 11. Li H, Shen DT, Jessup DA et al (1996) Preva-
aging research. J Gerontol A Biol Sci Med Sci lence of antibody to malignant catarrhal fever
63:879–884 virus in wild and domestic ruminants by
2. Catt K, Tregear G (1967) Solid-phase radioim- competitive-inhibition ELISA. J Wildl Dis
munoassay in antibody-coated tubes. Science 32:437–443
158:1570 12. Tai HC, Campanile N, Ezzelarab M et al
3. Aviñó A, Gómara M, Malakoutikhah M et al (2006) Measurement of anti-CD154 mono-
(2012) Oligonucleotide-peptide conjugates: clonal antibody in primate sera by competitive
solid-phase synthesis under acidic conditions inhibition ELISA. Xenotransplantation
and use in ELISA assays. Molecules 13:566–570
17:13825–13843 13. Raj GD, Rajanathan TM, Kumar CS et al
4. Yalow RS, Berson SA (1996) Immunoassay of (2008) Detection of peste des petits ruminants
endogenous plasma insulin in man. Obes Res virus antigen using immunofiltration and
4:583–600 antigen-competition ELISA methods. Vet
5. Smitsaart EN, Fernandez E, Maradei E et al Microbiol 129:246–251
(1994) Validation of an inhibition ELISA 14. Dessy FJ, Giannini SL, Bougelet CA et al
using a monoclonal antibody for foot-and- (2008) Correlation between direct ELISA, sin-
mouth disease (FMD) primary diagnosis. Zen- gle epitope-based inhibition ELISA and
tralbl Veterinarmed B 41:313–319 pseudovirion-based neutralization assay for
6. Aleshukina A, Iagovkin É, Bondarenko V measuring anti-HPV-16 and anti-HPV-18
(2011) Change of proinflammatory cytokine antibody response after vaccination with the
profile in human intestine in dysbacteriosis AS04-adjuvanted HPV-16/18 cervical cancer
caused by the antibiotics therapy. Zh Mikrobiol vaccine. Hum Vaccin 4:425–434
Epidemiol Immunobiol: 81–85 15. Giffroy D, Mazy C, Duchene M (2006) Vali-
7. Meng Y, High K, Antonello J et al (2005) dation of a new ELISA method for in vitro
Enhanced sensitivity and precision in an potency assay of hepatitis B-containing vac-
enzyme-linked immunosorbent assay with cines. Pharmeuropa Bio 2006:7
fluorogenic substrates compared with com- 16. Chovel Cuervo ML, Sterling AL, Nicot AI et al
monly used chromogenic substrates. Anal Bio- (2008) Validation of a new alternative for
chem 345:227–236 determining in vitro potency in vaccines con-
8. D€ahnrich C, Komorowski L, Probst C et al taining Hepatitis B from two different manu-
(2013) Development of a standardized ELISA facturers. Biologicals 36:375–382
for the determination of autoantibodies against 17. Friguet B, Chaffotte AF, Djavadi-Ohaniance L
human M-type phospholipase A2 receptor in et al (1985) Measurements of the true affinity
primary membranous nephropathy. Clin Chim constant in solution of antigen-antibody com-
Acta 421:213–218 plexes by enzyme-linked immunosorbent assay.
9. Crowther JR (2000) The ELISA guidebook, J Immunol Methods 77:305–319
vol 149. Humana press, Totowa, New Jersey 18. Zhao Q, Modis Y, High K et al (2012) Disas-
10. Cuervo MLC, de Castro Yanes AF (2004) sembly and reassembly of human papillomavi-
Comparison between in vitro potency tests rus virus-like particles produces more virion-
for Cuban Hepatitis B vaccine: contribution like antibody reactivity. Virol J 9:52
19. Harrison RO, Goodrow MH, Hammock BD treatment: Multifaceted biochemical and
(1991) Competitive inhibition ELISA for the immunochemical characterization. Vaccine
s-triazine herbicides: assay optimization and 29:7936–7941
antibody characterization. J Agric Food Chem 22. Mulder AM, Carragher B, Towne V et al
39:122–128 (2012) Toolbox for non-intrusive structural
20. Zhao Q, Wang Y, Abraham D et al (2011) Real and functional analysis of recombinant VLP
time monitoring of antigenicity development based vaccines: a case study with hepatitis B
of HBsAg virus-like particles (VLPs) during vaccine. PLoS One 7:e33235
heat- and redox-treatment. Biochem Biophys 23. High K, Meng Y, Washabaugh MW et al
Res Commun 408:447–453 (2005) Determination of picomolar equilib-
21. Zhao Q, Towne V, Brown M et al (2011) rium dissociation constants in solution by
In-depth process understanding of enzyme-linked immunosorbent assay with
RECOMBIVAX HB® maturation and fluorescence detection. Anal Biochem
potential epitope improvements with redox 347:159–161
Chapter 22
Glutathione-S-Transferase (GST)-Fusion Based Assays

for Studying Protein-Protein Interactions
Haris G. Vikis and Kun-Liang Guan
Abstract
Glutathione-S-transferase (GST)-fusion proteins have become an effective reagent to use in the study of
protein-protein interactions. GST-fusion proteins can be produced in bacterial and mammalian cells in large
quantities and purified rapidly. GST can be coupled to a glutathione matrix, which permits its use as an
effective affinity column to study interactions in vitro or to purify protein complexes in cells expressing the
GST-fusion protein. Here, we provide a technical description of the utilization of GST-fusion proteins as
both a tool to study protein-protein interactions and also as a means to purify interacting proteins.
Key words GST, Affinity chromatography, Protein-protein interactions, Ras, Raf
1 Introduction
A convenient method for the analysis of protein-protein interactions

is through the use of chimeric proteins comprising glutathione-S-
transferase (GST) linked in frame to a protein of interest. Commonly
referred to as GST-fusion proteins, these chimeras can typically be
expressed in Escherichia coli (E. coli) and rapidly affinity-purified
under non-denaturing conditions. Fusion of GST to a particular
protein commonly enhances the protein’s solubility and because
GST has virtually no toxicity in cells, these proteins can generally be
expressed in large amounts. Numerous convenient expression vectors
now exist for production of GST-fusions in both bacterial and mam-
malian cells.
GST-fusion proteins are commonly used to study protein-
protein interactions mainly because GST has a strong affinity for
glutathione and thus can be coupled at high concentration to an
immobilized glutathione matrix. Furthermore, the interaction of
GST with glutathione is particularly robust and resistant to strin-
gent buffer conditions. The GST moiety of the fusion protein
generally does not interfere with the accessibility of the fused
protein because a long flexible linker region exists which permits
353
354 Haris G. Vikis and Kun-Liang Guan
Protein mixture
GSH
GSH
Agarose
Bound protein
GST X
GSH
GSH
GST-X agarose affinity column
Unbound Protein
Fig. 1 A GST-fusion protein affinity column. Cell lysates comprising numerous

proteins are incubated with GST-X fusion protein bound to a glutathione
(GSH)–agarose matrix. After unbound proteins are washed away, only proteins
that interact with the GST-X fusion protein can be retrieved from the mixture
each half of the fusion to behave as separate domains. All of these

attributes make GST-fusion proteins bound to a glutathione matrix
simple and effective affinity columns which many investigators have
used as a tool to isolate novel interacting proteins (Fig. 1). In this
chapter we use GST-Ras expressed in E. coli and GST-B-Raf
expressed in mammalian cells to illustrate how to express and purify
GST-fusion proteins and use them in the analysis of protein-protein
interactions.
2 Materials
1. GEX-KG expression vector (or a similar GST-fusion expression

vector).
2. (DH5α, BL21(DE3)-RIL) E. coli strains.
GST-Fusion Proteins 355
3. LB Media (agar, liquid).

4. Antibiotic.
5. 37 C, 30 C, and room temperature incubators.
6. IPTG.
7. Spectrophotometer.
8. Lysis buffer A: 50 mM Tris–HCl, pH 7.5, 100 mM NaCl,
1 mM DTT, 0.2 mM PMSF, EDTA-free protease inhibitor
cocktail.
9. French press or sonicator.
10. Triton X-100 (TX100) detergent.
11. Refrigerated centrifuge.
12. Elution buffer: 50 mM Tris–HCl, pH 8, 100 mM NaCl,
10 mM reduced glutathione, 1 mM DTT.
13. SDS-PAGE equipment.
14. Glutathione (GSH)–agarose.
15. Reduced glutathione.
16. Liquid N2, 80 C storage.
17. Rocking/rotating platform.
18. Transfection reagent (e.g., Lipofectamine, Invitrogen).
19. Lysis buffer B: 20 mM Tris–HCl, pH 7.5, 100 mM NaCl, 1 %
NP-40, 5 mM MgCl2, 1 mM DTT, 0.2 mM PMSF, Complete
EDTA-free protease inhibitor tablet.
20. Antibodies: α-C-Raf, α-HA, α-GST.
3 Methods
Here we outline the methods involved in the general cloning,

expression, and purification of GST-fusion proteins and then
describe the specific use of bacterially expressed GST-Ras mutants
to analyze their interaction with C-Raf. We also describe the expres-
sion of a mammalian GST-fusion protein, GST-B-Raf, and analyze
its interaction with AKT.
3.1 Expression Vectors that direct expression of GST fusion proteins in E. coli are
Plasmids widely available commercially. Furthermore, a large collection of
empty vectors and vectors with cDNA inserts are also available
3.1.1 pGEX-KG E. coli through resources such as Addgene (addgene.org) and PlasmID
Expression Vector (plasmid.med.harvard.edu/PLASMID). We commonly use the
pGEX-KG expression vector for production of GST-fusion proteins
in E. coli [1] (Fig. 2). This vector is based on a set of vectors initially
constructed by Pharmacia Biotech. It is typical of many prokaryotic
expression vectors in that it contains an origin of replication
(pBR322 ori), an ampicillin resistance gene (Ampr), and a strong
MCS:
5’ CTG GTT CCG CGT GGA TCC CCG GGA ATT TCC GGT GGT GGT GGT
BamHI SmaI
GGAATT CTA GAC TCC ATG GGT CGA CTC GAG CTC AAG CTT AAT TCA 3’
EcoRI XbaI NcoI SalI XhoI SacI HindIII
L MCS
ST Stop codons
G
Am
pr
P tac
pGEX-KG
lac
lq
pBR322 ori
Fig. 2 pGEX-KG E. coli expression vector. pGEX-KG vector shown with essential components: GST gene,
L (linker region), MCS (multiple cloning site), Ampr (ampicillin resistance gene), pBR322 ori (origin of
replication), lacIq (lac repressor), and Ptac (promoter). The MCS is shown with restriction sites
promoter (Ptac) which is kept in the “off” state by the lac repressor
(lacIq). Addition of the non-hydrolyzable lactose analog, isopropyl
β-D-1-thiogalactopyranoside (IPTG) inactivates the lac repressor
and permits activation of the promoter. The GST gene lies down-
stream of the promoter and is followed by a linker region which is a
stretch of amino acids predicted to be very flexible and with no
apparent structure. This is then followed by the multiple cloning
site (MCS) which contains a number of commonly used restriction
sites for insertion of the cDNA of interest.
3.1.2 Cloning Strategy Insertion of the cDNA of interest into the pGEX-KG expression
vector MCS is performed by conventional molecular biology clon-
ing manipulations. It is worth noting that the cDNA need not
contain an initiator methionine codon; however, it must be inserted
in the same frame as the GST gene. DNA sequencing may be a
necessary diagnostic method to confirm the construct.
3.2 Production of Following construction and verification of the expression con-

GST-Fusion Proteins struct, the construct must then be transformed into the appropriate
in E. coli bacterial host strain. We typically use two E. coli strains for protein
expression: (1) DH5α and (2) BL21(DE3)-RIL (see Note 1).
3.2.1 E. coli 1. Transform plasmid expression vector into E. coli strain using
Transformation and standard heat shock methods and plate on an LB/Amp agar
Induction of Protein plate overnight at 37 C.
Expression 2. Pick an individual colony the next day (after 12–18 h of
growth), inoculate 10 mL of LB/Amp (50–100 μg/mL) liquid
media, and grow overnight in a 37 C shaker (see Note 2).
3. The next morning pour the 10 mL culture of E. coli into 1 L of
LB/Amp and grow for 2–5 h until an OD600 of 0.5–0.8 has
been reached (log phase growth).
4. Add IPTG to 0.1 mM and shake cells for 3–4 h at 37 C or
overnight at room temperature (see Note 3).
3.2.2 Protein Extraction 1. Centrifuge cells at 5,000 g for 10 min in polypropylene

bottles.
2. Decant supernatant. If desired, cells can be stored as a pellet at
20 C.
3. Resuspend cells in 15–30 mL of cold lysis buffer (see Note 4).
The use of an aluminum beaker to perform the following lysis
step will ensure the cells stay cold.
4. Cell lysis: there are two methods commonly used; (a) pass
through a French press twice or (b) sonicate (4 20–30 s
bursts). The French press is more cumbersome than the soni-
cator, but it is usually preferred when purifying a protein whose
activity can be destroyed by heat generated by the metal tip of
the sonicator. Therefore, we recommend keeping the sample as
cold as possible while sonicating (e.g., keep on ice with con-
stant mixing) and performing four 20–30 s sonication bursts
allowing 1–2 min intervals between bursts.
5. Add Triton X-100 (TX100) detergent to 1 %. This is added
after the lysis step to help solubilize proteins and to avoid any
frothing that may occur during sonication.
6. Centrifuge lysate at 30,000 g for 20 min at 4 C.
7. At this point NaCl can be added up to 1 M to prevent

co-purifying nonspecific proteins. Addition of NaCl will not
affect the affinity of GST for the glutathione matrix.
8. Add 1 mL of GSH–agarose or sepharose (50 % slurry, pre-
washed in lysis buffer) to the E. coli supernatant and mix for
at least 1 h at 4 C. The binding capacity of glutathione–agar-
ose is approximately 10 mg of GST protein per mL of 50 %
slurry.
9. After incubation, the glutathione–agarose beads must be
washed (see Note 5). Wash steps: (a) 2 lysis buffer + 1 M
NaCl; (b) 2 lysis buffer; (c) 3 lysis buffer (- TX100).
At this point, the protein can be stored bound to the gluta-
thione beads (see Subheading 3.3) or be eluted off the beads.
3.2.3 Elution and Dialysis 1. Elute the protein twice with 0.5 mL of elution buffer (10 min
of GST-Fusion Protein each).
2. To remove the glutathione from the buffer, the eluted protein
can be dialyzed against a buffer of choice at 4 C. We typically
dialyze 1 mL of eluted protein against 1 L of elution buffer
(without glutathione) and replace 1 mM DTT with 0.1 %
β-Me. However, we recommend choosing whichever dialysis
buffer system you deem most appropriate for your protein.
3. Quantify the amount of protein by running an SDS-PAGE gel
against known protein quantities (e.g., BSA). Electrophoresis/
Coomassie staining is preferable to using the Bradford assay
because of the ability to assess protein purity as well (Fig. 3).
Performing a Western blot with α-GST antibody will also allow
you to assess purity and the presence of degradation products.
GST-Ras
Coomassie
1 2 3
Fig. 3 IPTG induction and purity of GST-Ras protein. Samples were run on a
12.5 % SDS-PAGE gel and stained with Coomassie blue. Lane 1, E. coli lysate
prior to IPTG induction; lane 2, E. coli lysate after 3 h of IPTG induction; lane 3,
purified GST-Ras protein. The arrow indicates the location of the GST-Ras protein
in the induced lysate and after purification by glutathione–agarose
3.2.4 Storage of Various methods of storage are used and each should be tested to
GST-Fusion Proteins see whether it affects protein stability, activity, etc. GST-fusions can
be stored on beads or in solution. Common storage methods
include:
1. Addition of glycerol to 50 % and storage at 20 C (prevents
sample from freezing).
2. Addition of glycerol to 5–10 % and storage at 80 C.
3. Addition of glycerol to 5–10 % and snap-freeze in liquid N2,
prior to storage at 80 C.
After thawing and prior to use in an experiment, the beads
should be washed to remove glycerol.
3.3 Using GST-Ras to To illustrate the use of a GST-fusion protein in a binding reaction,
Analyze the Interaction we analyzed the interaction of the H-Ras GTPase with the Raf
with Raf kinase. We constructed pGEX-KG-H-RasV12 and -H-RasN17
expression vectors which express two mutants of the human
H-Ras protein. The mutation G12V locks Ras in the GTP nucleo-
tide bound conformation, while the T17N mutation locks the
protein in the GDP nucleotide bound conformation [2, 3]. GTP
bound Ras is in the “on” conformation and is able to interact with
numerous effector molecules such as the protein kinase, Raf,
through the Raf N-terminal domain. GDP bound Ras is in the
“off” conformation and unable to interact with Raf.
We purified GST-RasV12 and GST-RasN17 using the protocol
described in Subheading 3.2. It is important to note that these
small GTPases are required to be purified in the presence of 5 mM
MgCl2 to prevent the loss of bound nucleotide. The following
procedure describes the analysis of the interaction between E. coli-
expressed GST-Ras mutants and mammalian expressed N-terminal
domain (amino acids 1–269) of Raf. Raf 1–269 expression was
directed from the plasmid pCDNA3-cRaf-1-269, which was trans-
fected into mammalian HEK293 cells according to standard lipo-
fection protocols.
1. Transfect a 10 cm dish of mammalian HEK293 cells (using
Lipofectamine (Invitrogen)) with 10 μg of the pCDNA3-cRaf
(1–269) plasmid and grow for a further 48 h.
2. Wash cells once with 10 mL of phosphate buffered saline (PBS)
and lyse with 1 mL of lysis buffer B.
3. Place cells on ice for 10 min, scrape and collect into an eppen-
dorf tube.
4. Centrifuge the lysate at 15,000 g for 15 min at 4 C and
collect the supernatant.
5. Add 10 μg GST, GST-RasV12, and GST-RasN17 into three
separate tubes and to these tubes add the 293 cell lysate col-
lected in step 4. Mix by incubating at 4 C for 1–2 h on a
rocking/rotating platform.
293 cell lysate

(+ Raf 1-269)
GST-RasN17
GST-RasV12
GST
Pulldown:
WB: α -cRaf
GST-RasV12
GST-RasN17
coomassie
GST
Fig. 4 Interaction of Raf and Ras. Top panel, Raf (1–269) binds GST-RasV12 as
detected by α-Raf Western blot. Bottom panel, Coomassie blue stained gel of the
GST-fusions used in the assay
6. Add 10–15 μL GSH–agarose to each tube and rock for a

further 0.5–1 h (see Note 6). Wash the GSH–agarose beads
four times with 1 mL of lysis buffer B.
7. Elute bound GST-Ras with the elution buffer as explained in
Subheading 3.2.3, or with 1 SDS sample buffer (see Note 7).
8. Run samples on an SDS-PAGE gel and subject to a Western
blot using α-c-Raf antibody (Fig. 4).
3.4 Production of One benefit of expression in mammalian cells is that many eukary-
GST-Fusion Proteins otic proteins undergo modifications that do not occur in E. coli. For
in Mammalian Cell example, the Ras protein is C-terminal prenylated when expressed
Culture in mammalian cells, but does not undergo this modification in
E. coli because the appropriate modification enzymes are not pres-
ent in bacteria. Furthermore, mammalian proteins that typically
cannot be expressed in E. coli or are easily degraded in E. coli can
often be expressed in mammalian cells where conditions (tRNA,
folding machinery, etc.) are more suitable. However, the fact that
expression in mammalian cells results in less protein compared to E.
coli makes it less economical.
Using glutathione–agarose to purify GST-fusion proteins from
mammalian cells bypasses the need for using antibodies and thus
immunoprecipitation methods. This is particularly useful because
immunopurified proteins cannot easily be eluted from the antibody
and must be eluted by boiling which releases the antibody into the
mixture. Since GST-fusions can easily be eluted with glutathione,

no antibody is present. This makes identification of co-purified
proteins by protein sequencing or mass spectrometry much easier
because of the lack of large amounts of contaminating antibody
present.
Note that the purification protocol for mammalian GST-fusion
proteins is the same as in Subheading 3.2.2 except during the lysis
step detergents such as 1 % TX-100 or 1 % NP-40 are used to lyse
the cells instead of a French press or sonicator.
3.5 Detecting the There are a number of commercially available cytomegalovirus

Interaction Between (CMV) promoter driven GST-fusion expression vectors for use in
B-Raf and AKT Using mammalian cells. The vector pEBG-3X drives the expression of
Mammalian GST-fusion proteins from a very strong E1Fα promoter. We have
GST-Fusion Proteins constructed a modified version of pEBG-3X vector where more
restriction sites have been added to the MCS and have named it
pEBG-3X-HV (Fig. 5). Transfection of a 10 cm plate of HEK293
cells with pEBG-3X-HV can produce up to 5–10 μg of GST protein
which is roughly an order of magnitude more protein than what
most CMV promoter based vectors can direct.
1. Transfect HEK 293 cells (in a 6-well plate) with the following
combination of plasmids:
(a) pEBG-3X-HV (0.5 μg) + pCDNA3-HA-B-Raf (0.5 μg)
(b) pEBG-3X-HV-AKT (0.5 μg) + pCDNA3-HA-B-Raf
(0.5 μg)
(c) pEBG-3X-HV (0.5 μg) + pDNA3-HA-AKT (0.5 μg)
(d) pEBG-3X-HV-B-Raf (0.5 μg) + pDNA3-HA-AKT
(0.5 μg)
2. Forty-eight hours post-transfection, lyse the cells in 250 μL of
lysis buffer B (without MgCl2).
3. Centrifuge the lysate at 15,000 g for 10 min at 4 C.
4. Collect the supernatant and add 15 μL of GSH–agarose (50 %
slurry).
pEBG-3X-HV multiple cloning site:
BamHI NdeI EcoRV NheI

_______ _______ _______ _______
5’ ATC GAA GGT CGT GGG ATC GGA TCC CAT ATG GAT ATC GCT AGC
XmaI SalI SpeI ClaI NotI
_______ _______ _______ _______ __________
CCC GGG GTC GAC ACT AGT ATC GAT GCG GCC GCT GAA TAG 3’
Fig. 5 pGEX-3X-HV vector multiple cloning site

HA- HA-
BRaf AKT
GST-B-Raf
GST-AKT
GST
GST
glutathione-agarose:
HA-B-Raf
pulldown,
WB: α-HA
HA-AKT
GST-B-Raf
* GST-AKT
pulldown,
WB: α-GST
GST
lysate, HA-B-Raf
WB: α-HA *
HA-AKT
1 2 3 4
Fig. 6 Interaction of B-Raf with AKT. HEK293 cells were transfected with the
plasmid combinations indicated and grown for 48 h. Cells were lysed and GST-
fusion proteins were purified by addition of glutathione–agarose. To detect the
presence of co-purified proteins, SDS-PAGE was performed followed by an α-HA
Western blot. Asterisk nonspecific band
5. Incubate for 1–2 h on a rocker/rotating platform at 4 C.

6. Wash GSH–agarose beads four times with lysis buffer.
7. Elute using glutathione elution buffer (Subheading 3.2.3) or
boil in 1 SDS sample buffer.
8. Perform SDS-PAGE followed by Western blot analysis with
α-HA (Fig. 6).
3.6 Passing A common method to identify proteins that interact with a

Radiolabelled Lysates GST-fusion protein is to metabolically label cells with 35S-Met/
over GST-Fusion Cys and pass the cell lysates over a GST-fusion affinity column. This
Columns provides a very sensitive method of identifying novel interacting
proteins. However, there are several procedural changes to be aware
of to make this an effective method.
1. When lysing radiolabelled cells in mammalian culture, layer the

lysis buffer onto the radiolabelled cells so as to not detach
the cells from the cell culture plate. Furthermore, do not scrape
the cells after the lysis. Simply tilt the dish and remove as much
of the soluble lysate as possible. This will help reduce any
insoluble fragments that might not be completely removed in
the centrifugation step.
2. Often nonspecific proteins bind the glutathione–agarose
matrix during the binding step. One way to reduce this is to
block the beads in lysis buffer with the addition of 1 % BSA.
In addition, preclearing the lysate with glutathione–agarose
prior to incubating with glutathione–agarose/GST-fusion pro-
tein will help remove matrix-interacting proteins. This step
typically requires about 100 μL of glutathione agarose per
300 μL of lysate.
3.7 Different Lysis Reagents in the binding buffer may influence protein-protein inter-
Buffers to Use actions. It is often useful to try different buffer systems when
analyzing binding. Three commonly used buffer systems are
provided in order of increasing stringency:
1. NP-40 buffer: 20 mM Tris–HCl, pH 7.5, 100 mM NaCl, 1 %
Nonidet P-40
2. Triton buffer: 20 mM Tris–HCl, pH 7.5, 100 mM NaCl, 1 %
TX-100
3. RIPA buffer: 0.1 % SDS, 1 % Triton X-100, 0.5 % deoxycholate,
50 mM Tris–HCl, pH 7.5, 150 mM NaCl.
4 Notes
1. BL21(DE3)-RIL cells contain some human tRNAs that are

underrepresented in E. coli and hence enhance expression of
certain human proteins in E. coli. Production of soluble mam-
malian proteins can be aided by cloning shorter cDNAs and/or
encoding for less contiguous hydrophobic amino acids [4].
2. We recommend starting from a fresh colony for maximum
expression. This is particularly important in the case of the
RIL cells because often we see no protein expression if using
colonies more than 1 day old.
3. We find 0.1 mM IPTG is adequate and higher concentrations
do not seem to increase expression. To optimize expression,
one might try different conditions such as length of induction
and temperature of induction on small scale cultures first. For
example, if solubility is a problem, this can often be improved if
cells are induced overnight at 30 C.
4. We have found that the extent of protein degradation in E. coli

depends more on the expression conditions and less on the
concentration of protease inhibitors during the purification.
Hence it is important to vary the conditions as explained in
Note 3 to minimize protein degradation.
5. For washing of 1 mL of glutathione–agarose matrix, we typi-
cally spin down beads for 20 s at 5,000 g and wash in
Eppendorf tubes using multiple 1 mL washes. The washes
need not contain protease inhibitors.
Since proteins expressed in E. coli are produced at a rapid
rate, they sometimes do not fold efficiently. When this happens
the E. coli chaperone Hsp70 binds these misfolded proteins.
Therefore, it is common to find Hsp70 (at 70 kDa) associated
with purified GST-fusion proteins. One method to remove
Hsp70 from the GST-fusion is to wash the beads twice (prior
to elution) with 1 mL of 500 mM triethanolamine–HCl
(pH 7.5), 20 mM MgCl2, 50 mM KCl, 5 mM ATP, 2 mM
DTT for 10 min at room temperature.
6. Another method is to use GST-Ras which is still bound to
GSH–agarose. Either way is acceptable as we have not observed
any differences between the two methods.
7. Eluting the protein is more specific, but will release less GST-
fusion protein than boiling for 3 min in SDS sample buffer.
Acknowledgments
The authors would like to thank Huira Chong and Jennifer Aur-
andt for critical review of the manuscript. This work was supported
by grants from National Institutes of Health and Walther Cancer
Institute (K.L.G.). K.L.G. is a MacArthur Fellow. H.V. is sup-
ported by a Rackham Predoctoral Fellowship.
References
1. Guan KL, Dixon JE (1991) Eukaryotic proteins 3. Bourne HR, Sanders DA, McCormick F (1991)
expressed in Escherichia coli: an improved The GTPase superfamily: conserved structure and
thrombin cleavage and purification procedure molecular mechanism. Nature 349:117–127
of fusion proteins with glutathione S-transferase. 4. Dyson MR, Shadbolt SP, Vincent KJ, Perera RL,
Anal Biochem 192:262–267 McCafferty J (2004) Production of soluble
2. Katz ME, McCormick F (1997) Signal transduc- mammalian proteins in Escherichia coli: identifi-
tion from multiple Ras effectors. Curr Opin cation of protein features that correlate with
Genet Dev 7:75–79 successful expression. BMC Biotechnol 4:32
Chapter 23
Hexahistidine (6xHis) Fusion-Based Assays

for Protein-Protein Interactions
Mary C. Puckett
Abstract
Fusion-protein tags provide a useful method to study protein-protein interactions. One widely used fusion
tag is hexahistidine (6xHis). This tag has unique advantages over others due to its small size and the
relatively low abundance of naturally occurring consecutive histidine repeats. 6xHis tags can interact with
immobilized metal cations to provide for the capture of proteins and protein complexes of interest. In this
chapter, a description of the benefits and uses of 6xHis-fusion proteins as well as a detailed method for
performing a 6xHis-pulldown assay are described.
Key words Hexahistidine tag, 6xHis pull down, Affinity chromatography, Protein-protein
interactions
1 Introduction
Protein-protein interactions play vital roles in the regulation of

cellular functions, both physiological and pathophysiological.
These interactions have been traditionally studied by immunopre-
cipitation assays where one protein is isolated through the use of a
specific antibody and bound protein partners are analyzed. Immu-
noprecipitation, however, is extremely dependent on the character-
istics of the antibody used, leading to significant disadvantages if an
antibody is of poor quality or shows a high level of nonspecific
binding [1]. Furthermore, the light and heavy chains of the anti-
body itself can frequently be detected by Western blot, which has
the potential to obscure results obtained by this method.
To eliminate some of the disadvantages of immunoprecipita-
tions in protein-protein interaction studies, many fusion-protein
tags have been developed to specifically isolate proteins of interest.
A variety of protein tags have been developed, including the com-
monly used glutathione S-transferase (GST), FLAG, HA, and hex-
ahistidine (6xHis) tags [2]. While all of these tags have advantages
365
366 Mary C. Puckett
Fig. 1 Schematic of immobilized metal ions with histidine residues. Imidazole

rings form coordination bonds with transition metal cations
and disadvantages, 6xHis tags are particularly useful for a variety of

reasons and are discussed further in this chapter.
Polyhistidine tags were first used to aid in protein purification
[3], and have since come to have many useful applications in
protein-protein interaction studies. 6xHis tags are particularly use-
ful due to their strong interactions with immobilized metal cations.
The imidazole rings found in histidine residues can form coordina-
tion bonds with transition metal ions through the imidazole’s
Nitrogen atoms (Fig. 1), which serve as electron donors [4].
While Ni2+ is commonly used, other transition metals such as
Co2+, Cu2+, and Zn2+ can also interact with polyhistidine tags
[2]. The low abundance of naturally occurring histidine chains
and the minimal interaction of other amino acid residues with
these metals allows for highly specific capture of the tagged protein.
Additionally, a low concentration of imidazole in the lysis buffer is
sufficient to compete with native histidine residues on untagged
proteins and the metal–histidine interactions can easily be com-
peted by additional free imidazole or altered pH to recover the
protein or protein complexes of interest.
In addition to their unique ability to bind immobilized metal
cations, 6xHis tags have advantages due to their small size. Their
small size minimizes tag interference in native protein folding and
protein-protein interactions, which is sometimes noted with larger
fusion tags. Similarly, 6xHis tags can be added to either the N- or
6xHis Binding Assays 367
Fig. 2 Schematic of a 6xHis Pulldown assay. Lysates containing 6xHis

tagged proteins and other proteins of interest are prepared. The lysates are
then incubated with Ni2+ beads, which bind the 6xHis tag. Ni2+ beads are then
washed with washing and binding buffers to decrease nonspecific interactions,
leaving 6xHis-tagged protein complexes bound. The resulting complexes can
then be recovered and analyzed
C-terminus of a protein, allowing for greater flexibility. Vectors also

exist for the production of a cleavable His tag, allowing for non-
tagged recombinant proteins to be recovered.
Due to these advantages, many applications have been devel-
oped utilizing 6xHis tags. Purification of proteins from bacteria is
often accomplished by fusing the protein to a 6xHis tag, which can
then be used to isolate the protein quickly and efficiently and can be
cleaved, leaving an untagged protein available if desired [5]. Fur-
thermore, with the development of fluorophore-conjugated anti-
bodies, 6xHis-tagged fusion proteins can be used to analyze
protein-protein interactions via fluorescence resonance energy
transfer (FRET) as an alternative to traditional methods. The
most common use of 6xHis fusion proteins, however, is in the
His Pulldown assay. In this assay, a protein of interest is fused
with a 6xHis tag and overexpressed in mammalian cells. The
6xHis fusion protein can then be isolated, and complexed proteins
analyzed (Fig. 2).
2 Materials
1. 6xHis mammalian expression vector (e.g., pDEST26,

Invitrogen).
2. 6xHis-tag affinity resin.
3. Chromatography column.
368 Mary C. Puckett
4. Lysis buffer: 1 % NP-40, 137 mM NaCl, 40 mM Tris–HCl

pH 8.0, 60 mM imidazole, 5 mM Na4P2O7, 5 mM NaF, 2 mM
Na3VO4, 1 mM PMSF, 10 mg/L aprotinin, 10 mg/L
leupeptin.
5. 1 PBS.
6. Washing buffer: 1 % NP-40, 500 mM NaCl, 20 mM Tris–HCl
pH 8.0, 60 mM imidazole.
7. Binding buffer: 500 mM NaCl, 20 mM Tris–HCl pH 8.0,
5 mM imidazole.
8. Charging buffer: 50 mM NiSO4.
9. 6 SDS sample buffer: 7 mL 4 Tris–HCl/SDS (6.05 g Tris
base, 0.4 g SDS, 40 mL H2O, pH 6.8), 3.0 mL glycerol, 1.0 g
SDS, 0.93 g DTT, 1.2 mg bromophenol blue, H2O to 10 mL
total volume (aliquot and store at 20 C).
10. Rotator.
11. Antibodies for Western blotting: His-probe (Santa cruz,
sc-803), His-tag (Cell signaling, #2365).
3 Methods
3.1 Expression Many commercial plasmids are available for the production of
Plasmids 6xHis fusion proteins. Two commonly used vectors are pDEST17
for bacterial expression and pDEST26 for mammalian expression
(Invitrogen). These plasmids can be used within the gateway clon-
ing system to easily shuttle genes of interest into various vectors.
However, other 6xHis vectors are commercially available.
3.2 Expression in The following procedure is based on mammalian expression of

Mammalian Cell 6xHis-tagged proteins grown in six well plastic tissue culture plates.
Culture and Lysate Figure 3 shows interaction of 6xHis-14-3-3γ and HA-ASK1 as an
Preparation example.
1. Transfect mammalian cells, such as HEK293T or COS7, with
6xHis-fusion protein expression plasmids using an appropriate
transfection reagent (e.g., Fugene HD, Xtremegene HP, or
Lipofectamine) according to the manufacturer’s protocol.
2. Forty-eight hours after transfection, aspirate the medium off
cells and wash each well with 1 mL of 1 PBS. After washing,
add 200 μL of lysis buffer (containing imidazole) to each well
3. Scrap cells into lysis buffer and transfer to microcentrifuge
tubes. Incubate samples on ice for 10–30 min.
4. Spin samples down in a microcentrifuge for 10 min at maxi-
mum speed at 4 C to clarify lysates.
6xHis Binding Assays 369
H2O2 (1mM) - +
HA-ASK1
6xHis
Pull-Down 6xHis-14-3-3
HA-ASK1
Cell Lysate
6xHis-14-3-3
Fig. 3 Example of a 6xHis pulldown assay. Cells were transfected with HA-ASK1
and 6xHis-14-3-3γ. 48 h after transfection, cells were treated with H2O2 and
harvested. Lysate samples were analyzed by performing a 6xHis pulldown
assay, SDS-PAGE, and Western blot
5. Reserve 20 μL of clarified lysate supernatant to serve as the

input or whole cell lysate control. Add 6 μL of 6 SDS sample
buffer and boil for 5 min. Store samples at 20 C or 80 C
until ready for SDS-PAGE/Western blot analysis.
3.3 Performing a His 1. Add 160 μL of the lysate to 25 μL of charged his resin beads (see
Pulldown Subheading 3.4) in a microcentrifuge tube.
2. Rotate samples slowly at 4 C for 2 h (see Note 3).
3. Quickly centrifuge beads and discard the supernatant.
4. Wash the beads twice by adding 200 μL of washing buffer and
once with 200 μL of binding buffer. For each wash, gently
agitate the beads by turning the tube up and down four to
five times, quickly centrifuge the beads, and discard the super-
natant (see Note 4).
5. Recover bound proteins from the beads by boiling in 20 μL of
2 SDS sample loading buffer for 5 min.
6. Analyze proteins by SDS-PAGE and Western blotting. Anti-
6xHis antibodies are available from commercial sources (see
Note 5). A couple commonly used antibodies are listed in
Subheading 2.
3.4 Charging 6xHis 1. Add 3 mL his resin to a chromatography column, such as the
Resin Beads Bio-Rad Poly-Prep column. (Volumes can be scaled up or
down depending on need.)
2. Add 10 mL of nanopure water to the column, and allow to flow
through by gravity. Repeat a total of three times. Allow the
water to flow through completely before each repetition.
3. Add 15 mL of charging buffer to the column.
4. Add 20 mL of binding buffer to the column.
370 Mary C. Puckett
5. Add 3 mL of binding buffer to the column to make a 50 %

slurry of charged beads.
6. Store at 4 C for in capped columns or microcentrifuge tubes
for up to 6 months.
4 Notes
1. Detergent and NaCl concentrations can be adjusted in the lysis

buffer to alter the stringency of the assay. Increased detergent
and salt concentrations will increase the stringency of the assay
and decrease nonspecific binding.
2. The length of time between cell transfection and lysis can be
altered to optimize protein expression.
3. The time of incubation at 4 C can be lengthened to increase
capture of 6xHis fusion proteins on the surface of the immo-
bilized metal beads.
4. The number of washes following a 6xHis pulldown can be
increased to decrease nonspecific binding and weak
interactions.
5. Before performing Western blot analysis, confirm that the
6xHis antibody chosen can recognize the fusion tag utilized
in the assay. 6xHis tags may be fused to either the N- or C-
terminus of a protein, and the antibody used should be able to
recognize the desired orientation. Similarly, anti-His antibodies
may be raised against epitopes with varying numbers of histi-
dine residues, and the ability of the antibody to recognize the
6xHis tag used should be confirmed by checking the manufac-
turer’s datasheet.
References
1. Bjerrum OJ (1977) Immunochemical investiga- purification of recombinant proteins with novel

tion of membrane proteins. A methodological metal chelate adsorbent. Nat Biotechnol
survey with emphasis placed on immunoprecipita- 6:1321–1325
tion in gels. Biochim Biophys Acta 472:135–195 4. Kuo WH, Chase HA (2011) Exploiting the
2. Terpe K (2003) Overview of tag protein fusions: interactions between poly-histidine fusion tags
from molecular and biochemical fundamentals and immobilized metal ions. Biotechnol Lett
to commercial systems. Appl Microbiol Biotech- 33:1075–1084
nol 60:523–533 5. Bornhorst JA, Falke JJ (2000) Purification of
3. Hochuli E, Bannwarth W, Dobeli H, Gentz R, proteins using polyhistidine affinity tags. Meth-
Stuber D (1988) Genetic approach to facilitate ods Enzymol 326:245–254
Chapter 24
Studying Protein-Protein Interactions via Blot

Overlay/Far Western Blot
Randy A. Hall
Abstract
Blot overlay is a useful method for studying protein-protein interactions. This technique involves
fractionating proteins on SDS-PAGE, blotting to nitrocellulose or PVDF membrane, and then incubating
with a probe of interest. The probe is typically a protein that is radiolabeled, biotinylated, or simply
visualized with a specific antibody. When the probe is visualized via antibody detection, this technique is
often referred to as “Far Western blot.” Many different kinds of protein-protein interactions can be studied
via blot overlay, and the method is applicable to screens for unknown protein-protein interactions as well as
to the detailed characterization of known interactions.
Key words Protein-protein interactions, Blot overlay, Far Western blot, Protein, Receptor,
Association, Nitrocellulose, SDS-PAGE, Binding
1 Introduction
During preparation for SDS-PAGE, proteins are typically reduced

and denatured via treatment with Laemmli sample buffer [1]. Since
many protein-protein interactions rely upon aspects of secondary
and tertiary protein structure that are disrupted under reducing and
denaturing conditions, it might seem likely that few if any protein-
protein interactions could survive treatment with SDS-PAGE sam-
ple buffer. Nonetheless, it is well-known that many types of
protein-protein interaction do in fact still occur even after one of
the partners has been reduced, denatured, run on SDS-PAGE, and
Western blotted. Blot overlays are a standard and very useful
method for studying interactions between proteins.
In principle, a blot overlay is similar to a Western blot. For both
procedures, samples are run on SDS-PAGE gels, transferred to
nitrocellulose or PVDF, and then overlaid with a soluble protein
that may bind to one or more immobilized proteins on the blot.
In the case of a Western blot, the overlaid protein is antibody. In the
case of a blot overlay, the overlaid protein is a probe of interest,
371
372 Randy A. Hall
often a fusion protein that is easy to detect. The overlaid probe can
be detected either via incubation with an antibody (this method is
often referred to as a “Far Western blot”), via incubation with
streptavidin (if the probe is biotinylated), or via autoradiography
if the overlaid probe is radiolabeled with 32P. The specific method
that will be described here is a Far Western blot overlay that was
used to detect the binding of blotted hexahistidine-tagged PDZ
domain fusion proteins to soluble GST fusion proteins
corresponding to adrenergic receptor carboxyl-termini [2].
However, this method may be adapted to a wide variety of
applications.
2 Materials
1. SDS-PAGE mini-gel apparatus.

2. SDS-PAGE 4–20 % mini gels.
3. Western blot transfer apparatus.
4. Power supply.
5. Nitrocellulose.
6. SDS-PAGE pre-stained molecular weight markers.
7. SDS-PAGE sample buffer: 20 mM Tris–HCl, pH 7.4, 2 %
SDS, 2 % β-mercaptoethanol, 5 % glycerol, 1 mg/ml bromo-
phenol blue.
8. SDS-PAGE running buffer: 25 mM Tris–HCl, pH 7.4,
200 mM glycine, 0.1 % SDS.
9. SDS-PAGE transfer buffer: 10 mM Tris–HCl, pH 7.4,
100 mM glycine, 20 % methanol.
10. Purified hexahistidine-tagged fusion proteins.
11. Purified GST-tagged fusion proteins.
12. Anti-GST monoclonal antibody.
13. Goat anti-mouse HRP-coupled secondary antibody.
14. Phosphate-buffered saline: 137 mM NaCl, 2.7 mM KCl,
10 mM Na2HPO4, 2 mM KH2PO4, pH 7.4.
15. Blocking buffer: 2 % nonfat powdered milk, 0.1 % Tween-20 in
PBS.
16. Enhanced chemiluminescence (ECL) kit.
17. Blot trays.
18. Autoradiography cassette.
19. Clear plastic sheet protector.
20. Film.
Blot Overlay 373
3 Methods
3.1 SDS-PAGE and The purpose of this step is to immobilize the samples of interest on
Blotting nitrocellulose or an equivalent matrix, such as PVDF. It is very
important to keep the blot clean during the handling steps involved
in the transfer procedure, since contaminants can contribute to
increased background problems later on during detection of the
overlaid probe.
1. Place gel in SDS-PAGE apparatus and fill chamber with
running buffer.
2. Mix purified hexahistidine-tagged fusion proteins with SDS-
PAGE sample buffer to a final concentration of approximately
0.1 μg/μl of fusion protein (see Note 1).
3. Load 20 μl of fusion protein (2 μg total) in each lane of the gel.
If there are more lanes than samples, load 20 μl of sample buffer
in the extra lanes (see Note 2).
4. In at least one lane of the gel, load 20 μl of SDS-PAGE
molecular weight markers.
5. Run gel for approximately 1 h at 150 V using the power supply.
6. Stop gel, turn off the power supply, remove the gel from its
protective casing, and place in transfer buffer.
7. Place pre-cut nitrocellulose in transfer buffer to wet it.
8. Put nitrocellulose and gel together in transfer apparatus, and
transfer proteins from gel to nitrocellulose for 90 min at 200 V
using a power supply.
3.2 Overlay During the overlay step, the probe is incubated with the blot and
unbound probe is then washed away. The potential success of the
overlay depends heavily on the purity of the overlaid probe. GST
and hexahistidine-tagged fusion proteins should be purified as
extensively as possible. If the probe has many contaminants, this
may contribute to increasing the background during the detection
step, making visualization of the specifically bound probe more
difficult.
1. Block blot in blocking buffer for at least 30 min (see Note 3).
2. Add GST fusion proteins to a concentration of 25 nM in 10 ml
of blocking buffer.
3. Incubate GST fusion proteins with blot for 1 h at room tem-
perature while rocking slowly.
4. Discard GST fusion protein solution and wash blot three times
for 5 min each with 10 ml of blocking buffer while rocking
slowly.
374 Randy A. Hall
5. Add anti-GST antibody at 1:1,000 dilution (approximately

200 ng/ml final) to 10 ml of blocking buffer.
6. Incubate anti-GST antibody with blot for 1 h while rocking
slowly.
7. Discard anti-GST antibody solution and wash blot three times
slowly.
8. Add goat anti-mouse HRP-coupled secondary antibody at
1:2,000 dilution to 10 ml of blocking buffer.
9. Incubate secondary antibody with blot for 1 h while rocking
slowly.
10. Discard secondary antibody solution and wash blot three times
slowly (see Note 4).
11. Wash blot one time for 5 min with phosphate-buffered saline,
pH 7.4.
3.3 Detection of The final step of the overlay is to detect the probe that is bound
Overlaid Proteins specifically to proteins immobilized on the blot. In viewing differ-
ent exposures of the visualized probe, an effort should be made to
obtain the best possible signal-to-noise ratio. Nonspecific back-
ground binding will increase linearly with time of exposure. Thus,
shorter exposures may have more favorable signal-to-noise ratios.
1. Incubate blot with enhanced chemiluminescence solution for
60 s (see Note 5).
2. Remove excess ECL solution from blot and place blot in a clear
plastic sheet protector.
3. Tape sheet protector into an autoradiography cassette.
4. Move to the darkroom and place one sheet of film into the
autoradiography cassette with the blot.
5. Expose film for 5–2,000 s, depending on the intensity of the
signal.
6. Develop the film in standard film developer.
4 Notes
1. The protocol described here is intended for the in-depth study

of a protein-protein interaction that is already known. How-
ever, blot overlays can also be utilized in preliminary screening
studies to detect novel protein-protein interactions. For this
application, tissue lysates would typically be loaded onto the
SDS-PAGE gel instead of purified fusion protein samples.
The blotted tissue lysates would then be overlaid with the
Blot Overlay 375
probe of interest. The advantages of this technique are (1)

many tissue samples can be screened in a single blot and (2)
the molecular weight and tissue distribution of probe-
interacting proteins can be immediately determined. The dis-
advantages of this method are (1) due to the multiple washing
steps involved in the procedure, a fairly high-affinity interaction
is required for the interaction to be detected, (2) detection of
probe-interacting proteins is dependent upon their level of
expression in native tissues, and (3) interactions requiring
native conformations of both proteins will not be detected.
Tissue lysate overlays have been utilized as screening tools to
detect not only the interaction of the β1-adrenergic receptor
with MAGI-2 described here [2] (Fig. 1) but also the interac-
tion of the β2-adrenergic receptor with NHERF-1 [3] and the
interactions of a number of different proteins with actin [4–6],
calmodulin [7, 8], and the cyclic AMP-dependent protein
kinase RII regulatory subunit [9–12]. Tissue lysate overlay
approaches have also been effectively utilized to detect
phosphorylation-specific interactions between SH2 domains
and various phosphoproteins [13–16]. Associations involving
modular protein domains such as SH2 or PDZ domains are
often detected extremely well via blot overlay approaches, since
such modular domains typically interact with short motifs on
Fig. 1 Overlay of GST-tagged adrenergic receptor carboxyl-termini onto

hexahistidine-tagged PDZ domains. Equal amounts (2 μg) of purified His-
tagged fusion proteins corresponding to PDZ domains from PSD-95, nNOS,
MAGI-1, MAGI-2, and NHERF-1 were immobilized on nitrocellulose. Overlays
with the carboxyl-terminus of the β1-adrenergic receptor expressed as a GST
fusion protein (β1AR-CT-GST) (25 nM) revealed strong binding to PSD-95 PDZ3
and MAGI-2 PDZ1, moderate binding to MAGI-1 PDZ1, and no detectable binding
to the first two PDZ domains of PSD-95 or to the PDZ domains of nNOS or
NHERF-1. In contrast, overlays with the β2-adrenergic receptor expressed as a
GST fusion protein (β2AR-CT-GST) (25 nM) revealed strong binding to NHERF-1
PDZ1 but no detectable binding to any of the other PDZ domains examined.
These data demonstrate that selective and specific binding can be obtained in
overlay assays
376 Randy A. Hall
their binding partners in a manner that is not disrupted by

denaturation of the binding partners on SDS-PAGE gels.
Screens for novel protein-protein interactions via blot overlay
have also been performed by probing panels of purified fusion
proteins instead of tissue lysates; this method has been success-
fully utilized to identify a number of novel PDZ domain-
mediated interactions [17–26].
2. Since some probes can exhibit extensive nonspecific binding to
blotted proteins, it is important in overlay assays to have nega-
tive controls for probe binding. When the blotted proteins are
GST fusion proteins, GST by itself is a good negative control
(as illustrated in Fig. 2). When the blotted proteins are His-
tagged fusion proteins, as illustrated in Fig. 1, it is helpful to
have one or more His-tagged fusion proteins on the same blot
that will not bind to the probe. In this way, it is possible to
demonstrate the specificity of binding and to rule out the
possibility that the observed interaction is due to the tag.
3. The blocking of the blot is a very important step in every
overlay assay. The idea is to block potential nonspecific sites
of protein attachment to the blot, so that nonspecific binding
of the probe will be minimized. When a high amount of
Fig. 2 Overlay of hexahistidine-tagged MAGI-2 PDZ1 onto GST-tagged adrenergic receptor carboxyl-termini.
(a) In the reverse of the overlay experiments illustrated in Fig. 1, equal amounts (2 μg) of purified GST fusion
proteins corresponding to the carboxyl-termini of various adrenergic receptor subtypes were immobilized on
nitrocellulose. Overlay with His/S-tagged MAGI-2 PDZ1 (20 nM) revealed strong binding to β1AR-CT-GST but
no detectable binding to control GST, β2AR-CT-GST or α1AR-CT-GST. These data demonstrate that the
interaction between the β1AR-CT and MAGI-2 PDZ1 can be visualized via overlay in either direction. (b)
Estimate of the affinity of the interaction between β1AR-CT and MAGI-2 PDZ1. Nitrocellulose strips containing
2 μg β1AR-CT-GST (equivalent to lane 2 in the preceding panel) were incubated with His/S-tagged MAGI-
2 PDZ1 at six concentrations between 1 and 300 nM. Specific binding of MAGI-2 PDZ1 did not increase
between 100 and 300 nM, and thus the binding observed at 300 nM was defined as “maximal” binding. The
binding observed at the other concentrations was expressed as a percentage of maximal binding within each
experiment. The bars and error bars shown on this graph indicate mean SEM (n ¼ 3). The KD for MAGI-
2 PDZ1 binding to β1AR-CT was estimated at 10 nM (see Note 6)
Blot Overlay 377
nonspecific background binding is observed, it is often helpful

to block for a longer time or with a higher concentration of
milk. Some investigators favor bovine serum albumin or other
proteins in place of milk for blocking blots prior to overlay.
4. The washing of the blot is of critical importance. If the washes
are not rigorous enough, the nonspecific background binding
of the probe will be undesirably high. Conversely, if the washes
are too rigorous, specific binding of the probe may be lost and
the protein-protein interaction of interest may be difficult to
detect. Thus, if a large amount of nonspecific background
binding is observed, one should consider increasing the rigor
of the washes, while conversely if the background is low but
little or no specific binding is observed, one should consider
decreasing the rigor of the washes. The rigor of the washes is
dependent upon (1) time, (2) volume, (3) speed, and (4)
detergent concentration. To make washes more rigorous, one
should wash for a longer time, wash in a larger volume, increase
the rate at which the gels are rocked during the washes, and/or
increase the detergent concentration in the buffer used for
washing.
5. There are a number of ways to visualize bound probe in an
overlay assay. The method described here depends upon detec-
tion of the probe with an antibody, which is often referred to as
a “Far Western blot.” One alternative approach is to biotinylate
the probe and then detect it with a streptavidin/enzyme con-
jugate [5–7]. The appeal of this approach is that it can be quite
sensitive, since the streptavidin-biotin interaction is one of the
highest affinity interactions known. The main drawback of this
approach is that biotinylation of the probe may alter its proper-
ties, such that it may lose the ability to interact with partners it
normally binds to. An additional approach to probe detection is
phosphorylation of the probe using 32P-ATP to make the
probe radiolabeled [10–12]. A primary advantage of this
method is that once the probe is overlaid onto the blot, no
further detection steps are necessary (i.e., no incubations with
antibody or streptavidin are required). This cuts down on the
number of washing steps and may aid in the detection of
protein-protein interactions that are of somewhat lower affin-
ity. The main disadvantages of the phosphorylation approach
are (1) radioactive samples require special handling, and (2) as
with biotinylation, phosphorylation of the probe may alter its
properties, such that certain protein-protein interactions may
be disrupted.
6. As is illustrated in Figs. 1 and 2, detection of the interactions
between adrenergic receptor carboxyl-termini and their PDZ
domain-containing binding partners is completely reversible.
Either partner can be immobilized on the blot and overlaid
378 Randy A. Hall
with the other. Many other protein-protein interactions can

similarly be detected in a reversible manner, but some interac-
tions can only be detected in one direction due to a require-
ment for the native conformation of one of the partners. As is
also illustrated in Fig. 2, the affinity of a given protein-protein
interaction may be estimated via blot overlay saturation bind-
ing curves. This method involves increasing the concentration
of overlaid probe until a maximal amount of specific binding
is obtained. An estimate for the affinity constant (KD) of the
interaction can then be determined from the slope of
the binding curve. Estimates such as these must be evaluated
with the caveat that they are derived under artificial conditions
involving many hours of incubation time, washing, and detec-
tion. Nonetheless, affinity constant estimates derived via this
method are useful in comparing affinities between proteins
examined under the same conditions and overlaid with the
same probe.
Acknowledgments
R.A.H. is supported by grants from the National Institutes of

Health.
References
1. Laemmli UK (1970) Cleavage of structural actin binding site. J Biol Chem

proteins during the assembly of the head of 275:32331–32337
bacteriophage T4. Nature 227:680–685 7. Pennypacker KR, Kyritsis A, Chader GJ et al
2. Xu J, Paquet M, Lau AG et al (2001) beta 1- (1988) Calmodulin-binding proteins in human
adrenergic receptor association with the synap- Y-79 retinoblastoma and HTB-14 glioma cell
tic scaffolding protein membrane-associated lines. J Neurochem 50:1648–1654
guanylate kinase inverted-2 (MAGI-2). Differ- 8. Murray G, Marshall MJ, Trumble W et al
ential regulation of receptor internalization by (2001) Calmodulin-binding protein detection
MAGI-2 and PSD-95. J Biol Chem using a non-radiolabeled calmodulin fusion
276:41310–41317 protein. Biotechniques 30:1036–1042
3. Hall RA, Premont RT, Chow CW et al (1998) 9. Lohmann SM, DeCamilli P, Einig I et al
The beta2-adrenergic receptor interacts with (1984) High-affinity binding of the regulatory
the Na+/H+exchanger regulatory factor to subunit (RII) of cAMP-dependent protein
control Na+/H+ exchange. Nature kinase to microtubule-associated and other cel-
392:626–630 lular proteins. Proc Natl Acad Sci U S A
4. Luna EJ (1998) F-actin blot overlays. Methods 81:6723–6727
Enzymol 298:32–42 10. Bregman DB, Bhattacharyya N, Rubin CS
5. Li Y, Hua F, Carraway KL et al (1999) The (1989) High affinity binding protein for the
p185(neu)-containing glycoprotein complex regulatory subunit of cAMP-dependent pro-
of a microfilament-associated signal transduc- tein kinase II-B. Cloning, characterization,
tion particle. Purification, reconstitution, and and expression of cDNAs for rat brain P150. J
molecular associations with p58(gag) and Biol Chem 264:4648–4656
actin. J Biol Chem 274:25651–25658 11. Carr DW, Hausken ZE, Fraser ID et al (1992)
6. Holliday LS, Lu M, Lee BS et al (2000) The Association of the type II cAMP-dependent
amino-terminal domain of the B subunit of protein kinase with a human thyroid RII-
vacuolar H+ATPase contains a filamentous anchoring protein. Cloning and
Blot Overlay 379
characterization of the RII-binding domain. J regulates receptor activity. J Biol Chem

Biol Chem 267:13376–13382 281:29949–29961
12. Hausken ZE, Coghlan VM, Scott JD (1998) 20. Balasubramanian S, Fam SR, Hall RA (2007)
Overlay, ligand blotting, and band-shift tech- GABAB receptor association with the PDZ
niques to study kinase anchoring. Methods scaffold Mupp1 alters receptor stability and
Mol Biol 88:47–64 function. J Biol Chem 282:4162–4171
13. Machida K, Thompson CM, Dierck K, Jablo- 21. Lee SF, Kelly M, McAlister A, Luck SN, Garcia
nowski K, Karkkainen S, Liu B, Zhang H, Nash EL, Hall RA, Robins-Browne RM, Frankel G,
PD, Newman DK, Nollau P, Pawson T, Hartland EL (2008) A C-terminal class I PDZ
Renkema GH, Saksela K, Schiller MR, Shin binding motif of EspI/NleA modulates the
DG, Mayer BJ (2007) High-throughput phos- virulence of attaching and effacing Escherichia
photyrosine profiling using SH2 domains. Mol coli and Citrobacter rodentium. Cell Microbiol
Cell 26:899–915 10:499–513
14. Machida K, Mayer BJ (2009) Detection of 22. Kunkel MT, Garcia EL, Kajimoto T, Hall RA,
protein-protein interactions by far-western Newton AC (2009) The protein scaffold
blotting. Methods Mol Biol 536:313–329 NHERF-1 interacts with PKD and controls
15. Evans JV, Ammer AG, Jett JE, Bolcato CA, the amplitude and duration of localized PKD
Breaux JC, Martin KH, Culp MV, Gannett activity. J Biol Chem 284:24653–24661
PM, Weed SA (2012) Src binds cortactin 23. Stalker TJ, Wu J, Morgans A, Traxler EA,
through an SH2 domain cystine-mediated link- Wang L, Chatterjee MS, Lee D, Quertermous
age. J Cell Sci 125:6185–6197 T, Hall RA, Hammer DA, Diamond SL,
16. Gao Z, Poon HY, Li L, Li X, Palmesino E, Brass LF (2009) Endothelial cell specific adhe-
Glubrecht DD, Colwill K, Dutta I, Kania A, sion molecule (ESAM) localizes to platelet-
Pawson T, Godbout R (2012) Splice-mediated platelet contacts and regulates thrombus
motif switching regulates disabled-1 phosphor- formation in vivo. J Thromb Haemost
ylation and SH2 domain interactions. Mol Cell 7:1886–1896
Biol 32:2794–2808 24. Ritter SL, Asay MJ, Paquet M, Paavola KP,
17. Fam SR, Paquet M, Castleberry AM, Oller H, Reiff RE, Yun CC, Hall RA (2011) GLAST
Lee CJ, Traynelis SF, Smith Y, Yun CC, Hall stability and activity are regulated by interac-
RA (2005) P2Y1 purinergic receptor signaling tion with the PDZ scaffold NHERF-2. Neu-
is controlled by interaction with the PDZ scaf- rosci Lett 487:3–7
fold NHERF-2. Proc Natl Acad Sci U S A 25. O’Neill A, Gallegos L, Justilien V, Garcia EL,
102:8042–8047 Leitges M, Fields A, Hall RA, Newton AC
18. He J, Bellini M, Inuzuka H, Xu J, Xiong Y, (2011) PKCα promotes cell migration through
Wang X, Castleberry AM, Hall RA (2006) Pro- a PDZ-dependent interaction with its novel
teomic analysis of β1-adrenergic receptor inter- substrate discs large homolog (DLG) 1. J Biol
actions with PDZ scaffold proteins. J Biol Chem 286:43559–44368
Chem 281:2820–2827 26. Matsumoto M, Fujikawa A, Suzuki R, Shimizu
19. Paquet M, Asay MJ, Fam SR, Inuzuka H, Cas- H, Kuboyama K, Hiyama TY, Hall RA, Noda
tleberry AM, Oller H, Smith Y, Yun CC, Tray- M (2012) SAP97 promotes the stability of Nax
nelis SF, Hall RA (2006) The PDZ scaffold channels at the plasma membrane. FEBS Lett
NHERF-2 interacts with mGluR5 and 586:3805–3812
Chapter 25
Co-immunoprecipitation from Transfected Cells

Yoshinori Takahashi
Abstract
Co-immunoprecipitation (Co-IP) is one of the most widely used methods to identify novel proteins that
associate with a protein of interest or to determine complex formation between known proteins. For this
technique, a protein of interest is captured using a specific antibody. The antibody-bound protein, as well as
any proteins bound to the protein of interest, is then precipitated using a resin (immunoprecipitation, IP).
Proteins that are not bound to the protein of interest are then removed from the sample with a series of
washes. The resulting immunocomplexes are then analyzed by immunoblot. As the requirements for
protein-protein interactions vary, optimal experimental conditions for examining the interacting partners
of different proteins of interest must be determined empirically. Once appropriate experimental conditions
have been established, the IP/Co-IP procedure is simple and straightforward. In this chapter, a standard
protocol for IP/co-IP, with several key factors for the success of IP/co-IP analyses, is discussed.
Key words Immunoprecipitation, Co-immunoprecipitation, Protein-protein interaction, Antibody,

Transfection
1 Introduction
Protein-protein interactions regulate many biological processes

both within and between cells [1, 2]. Co-immunoprecipitation
(Co-IP) is a simple, yet effective method to determine complex
formation between proteins of interest [3–5]. In this assay, proteins
that directly or indirectly interact with a protein of interest can be
co-purified using specific antibodies (Fig. 1) (see Note 1). The
results of Co-IP are highly reproducible, and the assay is relatively
inexpensive.
The first step of the co-IP procedure is to prepare lysate from
cells or tissue samples that express (either endogenously or
exogenously) the protein of interest. In order to ensure that
protein-protein interactions will remain intact, cells need to be
lysed using a proper lysis buffer. In general, non-ionic detergents
(e.g., NP-40, Triton X-100) disrupt lipid–protein but not protein-
protein interactions, while ionic detergents (e.g., Chaps) affect
both. Therefore, lysis buffers containing non-ionic detergents
381
382 Yoshinori Takahashi
1. Transfect expression vectors

and/or treat cells
2. Lyse cells total cell lysate
3. Add antibody-immobilized sepharose/agarose resin
4. Precipitate immune-complexes & wash the precipitates

immune-complex
X Y
Z
5. Analyze by immunoblotting
Fig. 1 Schematic of the co-immunoprecipitation procedure. Cells may be

transfected with expression vectors that encode proteins of interest and/or
treated to induce protein complex formation (step 1). The cells are then lysed
in an optimal lysis buffer (step 2). Total cell lysate (TCL) is then incubated with
resin-immobilized antibodies, which can specifically recognize the protein of
interest, to form immune-complexes (step 3). The resultant immune-complexes
are then precipitated by centrifugation, washed to remove unbound proteins, and
analyzed by immunoblotting (steps 4 and 5)
tend to preserve protein-protein interactions more than those

containing ionic detergents (see Note 2). Proteins of interest are
then captured by incubating total cell/tissue lysate with specific
antibodies. The resultant immunocomplexes (composed of anti-
body, protein of interest (antigen), and antigen-associated proteins)
can be precipitated using a resin (e.g., agarose, sepharose, or mag-
netic beads) that is conjugated with IgG-binding Protein A/G
(see Note 3). Following a series of washes to remove irrelevant,
non-binding proteins, antigens and any proteins that are bound
are eluted by boiling the precipitated resins in denaturing Laemmli
buffer or by incubating with large amounts of peptides containing
Co-immunoprecipitation 383
the epitope of the immunoprecipitation antibody. The eluted

proteins are then analyzed by SDS-PAGE/immunoblotting and/
or mass spectrometry.
As mentioned above, proper experimental conditions must be
determined for each protein-protein interaction. Selection of an
optimal lysis buffer and immunoprecipitation antibody are the
two most important aspects for the success of a co-IP experiment.
However, the requirement of antibodies that specifically recognize
the antigen without interfering with its interaction with binding
partners limits the application of co-IP for detection of complexes
between many endogenously expressed proteins. To overcome this
hurdle, the protein of interest (and in some cases associating pro-
teins of interest) is frequently fused with a small peptide sequence
(~15aa), known as an epitope tag (e.g., flag, myc, HA, his, V5), or a
fluorescence protein (e.g., GFP, DsRed) at the amino- or carboxy-
terminus, and ectopically expressed in cells (see Note 4). Antibodies
for these “tags” that are compatible with IP/co-IP have been well
developed and are commercially available from multiple manufac-
turers. In this chapter, a protocol for transient transfection of 293 T
cells using the calcium-phosphate method, the most frequently
used and least expensive strategy to express proteins of interest, in
addition to standard procedures for co-IP, is described.
2 Materials
1. Cell line: HEK293, 293 T, or a variant such as 293 T/F17

(all available from ATCC).
2. Growth medium: Dulbecco’s Modified Eagle’s Medium
(DMEM) supplemented with 10 % fetal bovine serum and
1 % penicillin/streptomycin.
3. Expression plasmids.
4. Transfection reagents: 2.5 M CaCl2; 2 HEPES-buffered
saline (HeBS): 140 mM NaCl, 1.5 mM Na2HPO4, 50 mM
HEPES, pH 7.05 (see Note 5).
5. Cell scraper/lifter (for adherent cells).
6. Phosphate buffered saline (PBS): 137 mM NaCl, 2.7 mM KCl,
10 mM Na2HPO4, 2 mM KH2PO4, pH 7.4.
7. Lysis buffer: prepare stock solution (e.g., 137 mM NaCl,
20 mM Tris–HCl, pH 8.0, 10 % glycerol, 1 % NP-40) and
add protease and phosphatase (optional) inhibitor cocktails
prior to use (see Note 6).
8. Commercially available kit for measuring protein concentration
(e.g., BCA kit, Thermo Scientific).
9. Antibody that is specific for the protein of interest, and is

indicated for use in immunoprecipitation experiments.
10. Protein A or G-immobilized resin (e.g., sepharose, agarose)
(see Note 3).
11. 2 Laemmli sample buffer: prepare stock solution (125 mM
Tris–HCl, pH 6.8, 4 % SDS, 0.01 % bromophenol blue, 20 %
glycerol) and add ~5–10 % of β-mercaptoethanol prior to use.
12. Equipment: Basic lab equipment including a tube shaker/
rotator and cold room/chromatography refrigerator.
13. Reagents and equipment for immunoblotting.
3 Methods
3.1 Transfection of Day 1

Plasmids Expressing
1. Trypsinize 293 T/F17 cells grown under standard culture
the Gene/Proteins of
conditions and pellet 4.5 106 cells per sample.
Interest (See Note 7)
2. Resuspend the cells in 2 ml of medium/sample. Seed the cells
in 10 cm culture dishes containing 8 ml of growth medium.
Place the dishes in a 37 C incubator supplemented with 5 %
CO2 and allow the cells to attach and grow overnight.
Day 2 (The cells should be approximately 50–60 % confluent prior
to transfection)
3. Mix the following reagents in a 5 ml tube for each sample:
Expression plasmids (5 μg each) + sterile deionized water 450 μl

2.5 M CaCl2 50 μl
4. Gently add 500 μl of 2 HeBS in a drop-wise manner. Mix by

vortexing (2 s 5) and incubate at room temperature for
15 min.
5. Gently add the transfection mixture in a drop-wise manner to
the cells.
6. Change the medium 6–20 h after transfection.
7. Treat the cells if necessary (see Note 8).
3.2 Preparation of 1. Twenty-four to 48 h after transfection, harvest the cells using a

Total Cell Lysate cell scraper/lifter and pellet the cells by centrifugation at
1,500 g for 5 min at 4 C.
2. Wash the cells by resuspending the pellet in 1 ml of ice-cold
PBS, transfer the samples to a new 1.7 ml tube, centrifuge at
1,500 g for 5 min at 4 C, and discard as much of the
supernatant as possible (see Note 9).
3. Add ice-cold lysis buffer (~3 –5 the pellet volume) and

lyse the pellet by gently pipetting until all clumps disappear
(~5–10 times).
4. Incubate the samples on ice for 30 min.
5. Centrifuge at 15,000 g for 10 min at 4 C and transfer the
supernatant to a new 1.7 ml tube (see Note 9).
6. Measure protein concentration.
7. Check the expression of proteins of interest by SDS-PAGE/
immunoblot analyses (this step is optional, but highly recom-
mended) (see Note 10).
3.3 Preclearing of 1. Using a large-orifice tip, transfer the appropriate IgG-cross-

Cell Lysate (Optional) linked resins (15–20 μl (sample number + 1)) (see Notes 12
(See Note 11) and 13) to a microcentrifuge tube containing 1 ml of PBS.
Pellet the resin by centrifugation at 6,000 g for 30 s and
aspirate the supernatant.
2. Resuspend the resin in 1 ml lysis buffer (without protease/
phosphatase inhibitors), pellet the resin by centrifugation at
6,000 g for 30 s, and aspirate the supernatant.
3. Repeat step 2 three times for a total of four washes.
4. Resuspend the resin in ice-cold lysis buffer (~30–35 μl (sample
number + 1)).
5. Transfer 750 μg of total cell lysate to a new tube and adjust the
total volume to 200 μl with ice-cold lysis buffer.
6. Add 50 μl of the resin slurry prepared in Subheading 3.3 step 4
using a large-orifice tip. Incubate the tubes on a tube shaker/
rotator at a slow speed for 1 h at 4 C.
7. Pellet the resin by centrifugation at 6,000 g for 30 s at 4 C
and transfer 200 μl of the supernatant to a new tube. Save the
pelleted resins to use as negative controls for the experiment
(see Note 14).
3.4 Immunoprecipi- 1. In a new tube, dilute the appropriate amount of immunopre-

tation cipitation antibodies (1–10 μg of affinity purified antibodies,
1–5 μl of immune serum, or 10–100 μl of hybridoma superna-
tant,) with lysis buffer (containing the appropriate inhibitors)
to 100 μl/sample (see Note 15).
2. Add 100 μl of the antibody diluent to a precleared lysate-
containing tube prepared in step 7 in Subheading 3.3 and
incubate the tubes on a tube shaker/rotator for 1 h (to over-
night) at 4 C to allow the formation of immune complexes.
3. During the incubation, prepare Protein A or G-immobilized
resin slurry (as described in steps 1–4 in Subheading 3.3).
4. Add 50 μl of the resin slurry to each reaction tube in step 2 and

incubate the tubes on a tube shaker/rotator for 1 h at 4 C to
allow the antibody to bind to the protein complexes.
5. Pellet the resin to precipitate the immune complexes by centri-
fugation at 6,000 g for 30 s at 4 C. Transfer 100 μl of
supernatant to a new tube and aspirate the remaining superna-
tant using an 18G needle (see Note 16).
6. Resuspend the pelleted resin in 500 μl of ice-cold lysis buffer,
centrifuge at 6,000 g for 30 s at 4 C, and aspirate the
remaining supernatant to remove non-binding proteins.
7. Repeat the wash step two to three more times.
8. Resuspend the resin-bound immune complexes in 20 μl of
2 Laemmli buffer, boil for 5 min, and analyze by SDS-
PAGE/immunoblot analysis. If detecting proteins of interest
by the indirect method, avoid using antibodies developed in
the same species as those used for immunoprecipitation
(see Notes 17, 18, and 19).
4 Notes
1. A protein detected in immune-complexes does not necessarily

mean that it can directly bind to the antigen. For example,
while Beclin 1 (Protein Z in Fig. 1), a component of the class
III PI3-kinase that directly interacts with UVRAG (Protein Y),
can be detected in an immune-complex isolated using anti-Bif-
1 antibodies (immunoprecipitation antibodies), knockdown of
UVRAG or inhibition of UVRAG-Beclin 1 interaction results
in failure to detect Beclin 1in the anti-Bif-1 immune-complex
[6]. This indicates that Bif-1 (Protein X) indirectly associates
with Beclin 1 through UVRAG. In addition, physiologically
irrelevant interactions may be artificially induced during the
preparation of whole cell lysates (e.g., association of a nuclear
protein with a cytoplasmic protein). Therefore, it is important
to validate protein-protein interactions found during co-IP
experiments by other methods such as GST-pull down with
purified proteins (for the detection of direct protein-protein
interactions in vitro), FRET and/or BiFC (for spatial and tem-
poral analyses of protein-protein interactions) (see Chapters 20,
22, 31–33; 35–37). Colocalization analysis by confocal micros-
copy or deconvolution microscopy can also demonstrate that
two (or more) proteins localize in the same compartment of a
cell (see Chapter 34).
2. It is important to keep in mind that this is not always the case
and optimal detergents vary depending on proteins of interest.
For example, non-ionic detergents have been shown to induce
the conformational change of the inactive form of proapoptotic

Bax that artificially leads to the heterodimerization of Bax with
antiapoptotic Bcl-xL [7].
3. Protein A and G are immunoglobulin-binding proteins derived
from bacteria. The affinity of each protein binding to immuno-
globulins differs among species and the subclass of IgG. Typically,
Protein A and Protein G are used for immunoprecipitation of
antibodies developed in rabbit and mouse systems, respectively.
For further information regarding which immunoglobulin
should be used for a specific antibody, see the manufacture’s
product data sheet.
4. Fusion with an epitope tag/fluorescent protein may affect the
intracellular localization or function of a protein of interest. It is
therefore important to confirm that a tagged-protein is func-
tional in cells (e.g., by expressing it in cells that are deficient for
the protein of interest and comparing the results to wild-type
cells).
5. pH is a critical factor for the formation of DNA precipitates.
It is important to precisely adjust the pH to 7.05. If the
efficiency of transfection is too low, recalibrate the pH meter
and make solutions with several different pHs (e.g., 0.3) to
optimize an ideal pH condition for your particular experiment.
6. Selection of an appropriate lysis buffer is one of the most
important steps in a co-immunoprecipitation assay. A suitable
lysis buffer varies depending on the proteins of interest
(see Subheading 1). Addition of glycerol (5–10 %, final) in the
lysis buffer may increase the stability or formation of protein
complexes.
7. If a specific IP antibody for the protein of interest is available,
choose an optimal cell line/tissue lysate in which your proteins
of interest are known to be functionally expressed, treat the
cells if necessary (see Note 8) and start from Subheading 3.2
step 2.
8. Many protein-protein interactions are regulated by post-
transcriptional modifications (e.g., phosphorylation) that are
induced upon exposure to certain circumstances. However, in
some cases, overexpressing proteins of interest can induce the
formation of protein complexes that would normally require a
certain stimulation to be detected at the endogenous expres-
sion level (e.g., protein complex formation that is regulated by
post-transcriptional modifications of proteins of interest).
9. If you wish to stop and continue the following steps later, the
samples can be stored at 80 C. However, multiple free-
ze–thaw cycles should be avoided to prevent/minimize protein
denaturation and degradation.
10. If the proteins of interest are routinely expressed in your lab,

the expression levels can simply be quantified by dot blotting.
The dot blotting procedure is as follow: (1) Spot 5 μl of
sequentially diluted lysates (e.g., 0, 2, 4, and 8 μg/μl) onto a
strip of nitrocellulose membrane and mark each spot with a
pencil. (2) Dry the membrane strip. (3) Soak the membrane
strip in Tris-buffered saline with Tween 20 (TBS-T: 0.05 %
Tween 20, 150 mM NaCl, 20 mM Tris–HCl, pH 7.5) and
follow the procedure for immunoblotting starting at the block-
ing step.
11. Preclearing lysates will remove proteins that nonspecifically
bind to resins and/or immunoglobulins and thus reduce back-
ground signals.
12. Resins are generally stored as slurry in a buffer containing
ethanol. Calculate the amount of resin in the slurry and use a
volume of 15–20 μl of resin per reaction (e.g., if the resin you
are using is stored as a 50 % slurry, use 30–40 μl of slurry/
sample).
13. To preclear lysates, use IgG derived from the same species as
the antibodies being used for immunoprecipitation. Alterna-
tively, Protein A or G-immobilized resin without immunopre-
cipitation antibodies can be used.
14. Avoid taking the pelleted resin. After transferring the superna-
tant, aspirate the remaining supernatant using a 27 G needle
(avoid aspirating resin), wash the resin 3–4 times with lysis
buffer, resuspend in 20 μl of 2 Laemmli buffer, and boil for
5 min along with the co-IP samples. The resins used to preclear
the lysate can be used as a negative control for the experiment.
15. The appropriate amount of immunoprecipitation antibody will
vary by their affinity to the antigen. Contact the antibody
source to obtain detailed information on appropriate dilutions.
16. The post-IP supernatant can be used to determine whether the
amount of antibodies (and Protein A or G-immobilized resin)
used for the experiment was sufficient for immunoprecipitating
all, or the majority of, the protein of interest, and how much of
the protein of interest can be associated with the antibody. If a
substantial amount of the protein of interest remains in the
supernatant, increase the amount of immunoprecipitation anti-
body (or decrease the amount of total cell lysate).
17. In the indirect detection system, proteins of interest are detected
using HRP-conjugated secondary antibodies that recognize the
IgG of the primary antibody. Boiling of the precipitated resin in
Laemmli buffer results in the elution of not only the antigens and
antigen-associated proteins but also the immunoprecipitation
antibodies that are composed of heavy (~50 kDa) and light
(~25 kDa) chains. Therefore, immunoblotting by the indirect
method with antibodies developed in the same species as those

used for IP results in the detection of very intense signals from
heavy and light chain proteins that frequently mask the signals
from proteins of interest. If selection of antibodies developed
from different species is not optional, it is strongly recommended
to crosslink the immunoprecipitation antibodies to resins or to
use the direct detection method for immunoblotting. Alterna-
tively, using detection antibodies that preferentially recognize
the native disulfide forms of the IP antibodies (e.g., TrueBlot
reagents, Rockland Immunochemicals Inc.), the appearance of
heavy and light chains can be minimized from blots.
18. Once a protein (prey) has successfully been co-precipitated
with a protein of interest (bait), it is important to switch the
“bait” and “prey” and confirm that the two proteins can still be
co-precipitated in order to exclude false positives from nonspe-
cific binding.
19. Factors that may affect co-IP results: (1) lysis buffer; (2) immu-
noprecipitation antibodies; (3) the location where an epitope tag
is fused; (4) expression levels of proteins of interest; (5) culture
conditions (including certain treatments); and (6) the amount of
lysate, antibodies, and/or resin used for IP.
References
1. Fu H (2004) Protein-Protein Interactions. endogenous proteins. Anal Bioanal Chem
Methods and Applications. Methods in Molecu- 389:461–473
lar Biology 261. 5. Berggard T, Linse S, James P (2007) Methods
2. Braun P, Gingras AC (2012) History of protein- for the detection and analysis of protein-protein
protein interactions: from egg-white to complex interactions. Proteomics 7:2833–2842
networks. Proteomics 12:1478–1498 6. Takahashi Y, Coppola D, Matsushita N et al
3. Dwane S, Kiely PA (2011) Tools used to study (2007) Bif-1 interacts with Beclin 1 through
how protein complexes are assembled in signal- UVRAG and regulates autophagy and tumori-
ing cascades. Bioeng Bugs 2:247–259 genesis. Nat Cell Biol 9:1142–1151
4. Markham K, Bai Y, Schmitt-Ulms G (2007) Co- 7. Hsu YT, Youle RJ (1997) Nonionic detergents
immunoprecipitations revisited: an update on induce dimerization among members of the Bcl-
experimental concepts and their implementation 2 family. J Biol Chem 272:13829–13834
for sensitive interactome investigations of
Chapter 26
In Vivo Protein Cross-Linking

Fabrice Agou and Michel Véron
Abstract
In the cell, homo- and hetero-associations of polypeptide chains evolve and take place within subcellular
compartments that are crowded with many other cellular macromolecules. In vivo chemical cross-linking of
proteins is a powerful method to examine changes in protein oligomerization and protein-protein interac-
tions upon cellular events such as signal transduction. This chapter is intended to provide a guide for the
selection of cell membrane permeable cross-linkers, the optimization of in vivo cross-linking conditions,
and the identification of specific cross-links in a cellular context where the frequency of random collisions is
high. By combining the chemoselectivity of the homo-bifunctional cross-linker and the length of its spacer
arm with knowledge on the protein structure, we show that selective cross-links can be introduced
specifically on either the dimer or the hexamer form of the same polypeptide in vitro as well as in vivo,
using the human type B nucleoside diphosphate kinase as a protein model.
Key words In vivo cross-linking, Oligomerization, NDPK-B, Nucleoside nucleotide metabolism,

Multifunctional enzyme, Cysteine-mediated cross-linking, Histidine kinase, DNA binding protein,
Metastasis
1 Introduction
Chemical cross-linking is a powerful technique that has long been

used to characterize protein-protein interactions [1, 2]. Over the
past decade, in vitro cross-linking combined with mass spectrome-
try has further augmented the technique’s utility. By harnessing
these two techniques, investigators have studied protein subunit
composition and architectural organization of large protein com-
plexes. They have also detailed structural aspects of protein-protein
interfaces and identified unknown protein partners within
proteome-wide protein interaction networks [3, 4]. In vitro cross-
linking has also been successfully employed as a complementary
method for determining the crystal structure of large multi-subunit
protein assemblies when classical structural biology approaches
failed [5]. However, all these studies take place outside of a cellular
context, and only a few examples of cross-linking performed in
living cells currently exist in the literature. This is mainly due to
391
392 Fabrice Agou and Michel Véron
the difficulty of targeting cross-links in specific proteins in a cellular

environment like those of E. coli and human cells, in which the total
concentrations of protein are in the range of 200 and 50–100 g/l,
respectively [6]. Such a situation increases the frequency of random
collisions, which may lead to nonspecific intermolecular cross-
linking. For instance, tetrameric hemoglobin was formerly shown
to form octamers when intact erythrocyte cells were treated with a
membrane permeable cross-linker [7]. Moreover, cross-linker
application to cells usually produces less in vivo cross-linked sites
than those observed in vitro [8]. Therefore, it requires more
sophisticated cross-linking reagents and advanced technology and
informatics, especially when cross-linking experiments are coupled
with mass spectrometry analysis.
Using structural information provided by X-ray crystallography
or NMR, it is possible to overcome some of the problems related to
cross-linking in living cells. Indeed, more selective cross-links can
be introduced in a protein taking into account the spatial arrange-
ment of its nucleophilic residues.
This chapter presents a brief general survey of in vivo cross-
linking approaches based on structural design to probe the quater-
nary structure of the human type B nucleoside diphosphate kinase
(NDPK-B, also known as NME2/NM23-H2), an enzyme which is
involved in the maintenance of the cellular nucleoside triphosphate
(NTPs) pool [9, 10]. This enzyme has also been identified as a
DNA binding protein and is involved in transcriptional activation of
the c-myc gene [11, 12]. Only the hexameric enzyme is active in
NTPs synthesis, whereas the dimeric protein is not [13, 14]. Con-
versely, the dimer binds to DNA with a higher affinity than the
hexamer [15], suggesting the possibility that the dimer may act as a
more efficient transcriptional activator than the hexamer. As this
dual activity depends on hexameric and dimeric structures, differ-
ent oligomeric states of the protein were investigated in cells. In the
first attempt, S100 extracts from different tumor cells were ana-
lyzed by gel filtration method. However the resolution of the gel
was too low to separate the different oligomeric forms due to the
interconversion of different species during the protein elution.
Thus, we developed an in vivo cross-linking method to monitor
the quaternary structure of NDPK-B.
We first showed by in vitro cross-linking that specific cross-links
on cysteine residues could be selectively introduced on the NDPK
dimer or hexamer, depending on the length of the spacer arm of the
homo-bifunctional cross-linker. We used this difference in the pro-
tein reactivity with long and short cross-linkers to establish an
in vitro direct correlation between the oligomeric state of protein
and the amount of cross-linked dimeric subunits. Differential cross-
linking was then performed with long and short membrane perme-
able cross-linkers in HeLa cells after overexpressing either the
dimer or the hexamer. A positive correlation could be established
Protein Cross-Linking in Living Cells 393
between the amount of overexpressed dimeric mutant protein and

the amount of cross-links quantified by immunoblotting. This
shows that our in vivo protein cross-linking method based on
structural design is highly selective and can be used to probe the
quaternary structure of proteins in intact cells.
2 Materials
2.1 Pure The recombinant wild type hexamer was purified as described in
Recombinant Proteins ref. [16]. The recombinant dimeric mutant P96S-A146 stop was
constructed by the overlap extension method [17]. The point
mutation P96S and the removal of the 7 C-terminal amino acids
in the crystallographic hexamer interface has been shown to greatly
affect hexamerization of NDPK [14]. The double mutant was
purified according to a procedure similar to the one used for the
wild type protein except that all purification buffers contained
0.05 mM of dodecyl-β-D-maltoside (DDM) detergent. Proteins
were stored at 20 C in 20 mM potassium phosphate pH 7.0
buffer containing 1 mM DTE and 50 % (v/v) glycerol.
2.2 SDS-PAGE 1. Laemmli SDS-PAGE equipment and buffers.

2. Coomassie R250 dye and Coomassie staining buffers.
2.3 Western Blotting 1. Western blotting equipment (nitrocellulose membrane, trans-

fer chamber, filter paper, X-ray film).
2. Western blotting buffers: Tris–HCl/glycine transfer buffer
(pH 8.3), blocking and washing buffers.
3. Specific antibodies against the protein of interest and anti-
Ig-peroxidase (secondary antibody). In our experiments we
routinely used polyclonal anti-NDPK-B antibodies at a final
concentration of 10 μg/ml.
4. Enhanced chemiluminescence Plus reagents (ECL Plus) with a
laser scanner system such as Storm (GE Healthcare life
sciences) for detection and quantification.
2.4 Protein Cross- 1. Membrane permeable sulfydryl-reactive homo-bifunctional

Linking via Cysteine cross-linker: Bis-maleimidoethane (BMOE); bis-maleimidohex-
Residues ane (BMH) freshly dissolved in DMSO.
2. Membrane impermeable and water-soluble sulfhydryl-reactive
homo-bifunctional cross-linker: 1,8-bis-maleimidotriethylene-
glycol (BM[PEO3]) freshly dissolved in water according to the
manufacturer’s instructions.
3. Quenching and reaction buffers for in vitro cross-linking:
2 quenching buffer, 2 Laemmli loading buffer containing
100 mM DTE; 2 reaction buffer, 100 mM potassium phos-

phate pH 7.2, and 300 mM potassium chloride.
4. 6 quenching buffer for in vivo cross-linking: 180 mM DTE
in water.
2.5 Cell Culture 1. Cell culture equipment (plastic culture vessels, laminar flow
hood, CO2 incubator, water bath).
2. Complete culture medium (DMEM, fetal calf serum (FCS),
antibiotics).
3. HeLa cells (ATCC CCL-2) in serum supplemented DMEM.
4. Ca2+ and Mg2+-free phosphate-buffered saline (CMF-PBS).
5. Trypsin–EDTA solution in CMF-PBS.
2.6 DNA 1. Mammalian expression vectors bearing the cytomegalovirus

Transfections promoter (e.g., pcDNA3, Invitrogen).
2. Cationic liposomes (e.g., FuGENE 6, Roche).
2.7 Whole Cell Urea lysis buffer (SBLU): 62.5 mM Tris–HCl, pH 7.8, 2 % SDS,
Extracts 6 M urea, 100 mM DTE, and 0.05 % bromophenol blue.
3 Methods
3.1 General Cross-links can be more efficiently introduced in a protein when its
Considerations: In Vivo three-dimensional structure is known. The criteria to be considered
Selectivity in Cross- for optimal selectivity in in vivo protein cross-linking (hetero- or
Linking Based on homo-bifunctional reagent; type of reactive group; length and
Structural Design rigidity of the linker structure) are almost the same as those previ-
ously described in vitro [1, 18] except for two:
1. The cross-linker must diffuse across the membrane, and there-
fore its linker structure should be mostly hydrophobic.
2. Individual amino acids likely to react with cross-linker should be
limited in the 3D space to minimize nonspecific intermolecular
bridges in vivo (see Note 1). The use of a low cells–cross-linker
ratio is also advised to limit the extent of protein modifications,
thereby enhancing the relative differences in chemical reactivity
between the amino acids.
We used a non-cleavable, membrane-permeable, homo-
bifunctional reagent containing maleimide groups at both ends
for introducing selective cysteine cross-links in vivo. In addition
to the advantages of cysteine-based cross-linking, the extent of
protein modification through cysteine residues is usually reduced
due to the low content of cysteine in protein, minimizing nonspe-
cific intermolecular cross-linking in vivo.
Fig. 1 Protein cross-linking based on structural design to probe quaternary

structure. (a) Side view of the hexameric structure of NDPK-B. Out of the six
identical subunits which compose the NDPK-B hexamer, only four subunits are
clearly visible in this Connelly surface representation and are referred as A, D, C,
and F subunits. (b) Connelly surface representation of the dimer rotated by 180
around the threefold axis (hexameric interface view). Each subunit contains two
cysteines C109 and C145 (colored in black) which are differently exposed to
solvent in dimeric and hexameric states. In the hexamer, only the C145
cysteines are solvent-accessible and close in the 3D space (solid line 4.5 Å,
hexameric contact between A chain and F chain), whereas in the dimer, C145 (A)
of one monomer and C109 (D) of the second monomer are solvent-accessible
and are distant from each other by 14.6 Å (solid line). All other cysteine pairs in
the dimer are separated by more than 31 Å and are indicated in dotted lines.
Given the different spatial arrangement of cysteine pairs in the dimeric and
hexameric states, selective cross-linking can be introduced on the dimer and the
hexamer using a homo-bifunctional reagent varying only in length and flexibility
of the spacer arm which connects both specific cysteine reactive groups
We chose Human NDPK-B as a model protein because (1) its

crystal structure has already been determined [19] and (2) the
protein only contains two cysteine residues (C145 and C109)
which are differently exposed to solvent in dimeric and hexameric
states (see Fig. 1 for details).
Whatever the protein studied, those who wish to use com-
mercially available reagents may consult the “Crosslinking Tech-
nical Handbook,” which can be freely downloaded from Thermo
Scientific Pierce (http://www.piercenet.com). The website also
provides an online interactive cross-linker selection guide, which
is a practical starting point for those who do not know which
reagents and protocols to use to cross-link their protein specifi-
cally in vivo, taking also into account all the criteria we have listed
above.
3.2 In Vitro Protein BM[PEO]3 and BMOE are homo-bifunctional uncleavable

Cross-Linking of bis-maleimide cross-linkers that selectively react with thiols to
NDPK-B via Cysteine form stable covalent thioether linkages at pH 6.5–7.5. BMOE
Residues Using and BM[PEO]3 differ only in the length of the spacer arm and in
Variable Spacer Arms their solubility in water. The two maleimide groups are distant from
14.7 Å and 8 Å in BM[PEO]3 and BMOE, respectively. In the
following section, BM[PEO]3 will therefore be referred to as the
long cross-linker and BMOE as the short cross-linker.
3.2.1 In Vitro Cross- 1. Dialyze extensively or desalt the protein with a small G25 col-
Linking of the NDPK-B umn into 50 mM sodium phosphate pH 7.2 reaction buffer
Hexamer (See Fig. 2) containing 150 mM sodium chloride to remove any trace of
reducing agents.
Fig. 2 In vitro cross-linking of the NDPK-B hexamer with short and long cysteine specific cross-linkers. NDPK-
B, 0.02 mg/ml (1.2 μM, subunit concentration), was reacted with either the short (BMOE) (a) or the long cross-
linker (BM[PEO]3) (b) using a concentration of 0.1 mM following the experimental procedure described in
Subheading 3.2.1. At the time indicated, the reaction was quenched by addition of SDS-PAGE loading buffer
and the samples were subjected to 15 % SDS-PAGE followed by Coomassie staining. M, D*, H*, and NS* refer
to the positions of monomers, cross-linked dimers, cross-linked hexamers, and nonspecific higher order
cross-linked species of NDPK-B subunits, respectively. (c): Scans of the SDS-PAGE gel shown in (a) in order to
quantify cross-linked species. The hexameric cross-linked species was very low as compared to the cross-
linked dimer even though the protein at 0.02 mg/ml was a hexamer as judged by the equilibrium sedimenta-
tion [16]. This reflects the very low efficiency of these short and long cross-linkers to cross-bridge all six
subunits together. Patterns were similar using a higher concentration of cross-linker or using different protein
concentrations (0.002 mg/ml or 0.2 mg/ml) confirming the low cross-linking efficiency in the formation of the
hexameric cross-linked species
2. Dissolve the long cross-linker in water according to the manu-

facturer’s instructions and the short cross-linker in DMSO,
both at 1 mM.
3. Prepare several eppendorf tubes each containing 25 μl of
2 reaction buffer, 1 or 10 μg of the protein of interest and
complete with water to a 50 μl final volume. Add also 10 μl of
2 quenching buffer in one extra tube to control the quench-
ing efficiency.
4. Start the reaction by adding 5 μl of the cross-linker solution
(1/10 volume) and mix well by vortexing.
5. Stop the reaction at different times by adding 10 μl of
2 quenching buffer and stir vigorously by vortexing.
6. Heat samples for 3–5 min at 100 C.
7. Load samples on a Laemmli gel with a desired percentage of
acrylamide. Generally, use 5 % gels for SDS-denatured proteins
of 60–200 kDa, 10 % gels for 16–70 kDa, and 15 % gels for
12–45 kDa. Visualize different cross-linked species with Coomas-
sie blue staining according to the standard protocol (see Note 2).
3.2.2 In Vitro Cross- Gel filtration analyses showed that the purified P96S-A146stop
Linking of the NDPK-B mutant formed a stable dimer in the protein concentration range
Dimeric Mutant (P96S- of 0.5–50 μM (subunit concentration). Experimental procedures to
A146stop) (See Fig. 3) cross-link the NDPK-B dimeric mutant with the long and short
cross-linkers were similar to those used for the wild type hexamer
except that the reaction buffer contained 0.05 mM of dodecyl-β-D-
maltoside (DDM) detergent, to stabilize the dimer.
Obviously the in vitro characterization of protein cross-linking
with pure components (as illustrated in Fig. 4) is not a prerequisite
step to carry out in vivo cross-linking. This was shown here for a
direct comparison of the selectivity of protein cross-linking in vitro
and in a cellular context.
3.3 Cross-Linking of Since the long cross-linker BM[PEO]3 is membrane-impermeable,

NDPK-B in Living Cells an analogous reagent such as BMH which crosses the cell
(See Note 3) membrane was used. Like BM[PEO]3, BMH is a thiol-specific,
non-cleavable, homo-bifunctional cross-linker that connects both
maleimide groups via a spacer arm whose length is close to that of
BM[PEO]3 (16.1 Å). In the following protocol of in vivo protein
cross-linking, the long and short cross-linkers are then referred to
as BMH and BMOE, respectively. To check whether overexpressing
the stable dimeric mutant P96S-C145stop in HeLa cells induced a
shift towards a hexameric state, GFP fusion proteins with the
hexamer wild type or the dimeric mutant were made. After tran-
sient transfections of HeLa cells, analysis by fluorescence imaging
showed different protein localization patterns. Whereas the wild
type hexamer-GFP fusion protein was mainly located in the cytosol,
Fig. 3 In vitro cross-linking of the dimeric mutant of NDPK-B with short and long cysteine specific cross-
linkers. Pure dimeric mutant, 0.02 mg/ml 1.2 μM (subunit concentration) was treated with either the short
(BMOE) (a), or the long cross-linker (BM[PEO]3) (b), according to the instructions described in Subheading 3.2.2
using 0.1 mM of the cross-linker. The reaction was quenched by addition of SDS-PAGE loading buffer at the
time indicated. Cross-linked species were then analyzed by SDS-PAGE and Coomassie blue staining. M and D*
denote the monomer and the cross-linked dimer. (c) Scans of the SDS-PAGE gel shown in (a). The graph
represents the percentage of each species with time and each curve was fitted with a monoexponential (solid
line) (see also Note 3). Scans of the SDS-PAGE gel shown in (b) are not represented under these experimental
conditions because the reaction was too fast
the dimeric mutant-GFP fusion protein was present in both com-

partments (nucleus and cytosol, data not shown). These results
suggested that the dimeric mutant was not totally converted into
a hexamer when highly expressed in HeLa cells.
3.3.1 Transient 1. The day before the transfections, trypsinize and count the
Transfections of HeLa Cells HeLa cells, plate them at the appropriate plating density in a
to Overexpress the Dimeric complete medium so that they are 50–60 % confluent on the
Mutant and the Wild Type day of transfection. For cationic lipid-mediated transfections a
Hexamer 6-well plate was used, with each well containing 0.2 106
HeLa cells in a total complete media volume of 2 ml
(DMEM, 10 % FCS).
2. On the day of transfection, add the requisite volume of serum-
free medium as diluent to a total volume of 100 μl in a sterile
polystyrene or polypropylene tube, and then the cationic
lipid transfection reagent such as FuGENE 6 (Roche). Tap
gently to mix.
Fig. 4 Diagram of in vitro cross-linking efficiencies of NDPK-B with short and long cysteine-specific cross-
linkers as a function of the oligomeric states. Bimolecular rate constants relative to short (BMOE) and long (BM
[PEO]3) cross-linkers with the dimer and the hexamer were determined using three independent cross-linker
concentrations as described in Subheading 3.2. (see Note 3). The arrow surfaces relative to cross-linking
reactions with short (in black) and long cross-linkers (in gray) are proportional to the magnitude of each rate
constant. The reaction D ! D* is the formation of the dimeric mutant into the cross-linked dimer; H ! D* is
the formation of the hexamer into two cross-linked subunits; D* ! H* is the formation of the cross-linked
dimer into the cross-linked hexamer and H ! H* is the direct conversion of the wild type into the cross-linked
hexamer. The rate constants k1long D ! D* and k1short D ! D* were 900 50 M1.s1 and 9.6
0.5 M1.s1, respectively, meaning that the cross-linking efficiency on the dimer is enhanced about
100-fold with the long cross-linker compared to the short cross-linker. The cross-linking efficiency
corresponding to the formation of two cross-linked subunits within the hexamer is lower with both reagents.
Bimolecular rate constants, k1long H ! D* and k1short H ! D*, were 0.9 M1.s1 and 3 M1.s1, respec-
tively, indicating that the short cross-linker is slightly more effective as compared to the long cross-linker in
hexamer cross-linking. Non-measurable rates of the cross-linked hexamer or the intermediate cross-linked
species (trimer, tetramer, or pentamer) are shown as dotted arrows
3. Add the DNA solution at a final concentration of 0.1 μg/μl to

the prediluted transfection reagent. In our experiments a 3:1
(μl/μg) ratio of FuGENE 6 reagent to DNA was used for
optimal transfection efficiency into HeLa cells. Mix gently
and let the DNA–lipid complex form for 45 min at room
temperature.
4. While DNA–lipid complexes are forming, replace medium on
cells with the appropriate volume of fresh complete medium.
5. Add DNA-lipid complexes directly to each well with the
desired range of DNA. In our experiments 0, 2, 5, 10, and
20 μl of transfection medium containing 0.1 μg/ml of DNA
were added directly on the cells. Distribute the DNA-lipid
mixture around each well and swirl the 6-well plate to ensure
even dispersal. Incubate for 24 h at 37 C in 5 % CO2.
3.3.2 In Vivo Protein On the day of the treatment with cross-linkers, HeLa cells should
Cross-Linking (See Fig. 5) be 80 % confluent.
1. Dissolve the long and short cross-linkers in DMSO to a final
concentration of 20 mM.
2. Trypsinize each well with 100 μl of Trypsin–EDTA solution
and incubate for 5 min at 37 C. Stop the reaction by adding
400 μl of complete medium and count the cells.
3. Immediately after, transfer the cells still in suspension into a
sterile polystyrene or polypropylene tube. Wash the cells twice
with 500 μl of complete medium. Incubate cells for 30 min on
ice.
4. Distribute HeLa cells in sterile polystyrene or polypropylene
tubes with a cell density of 0.8 106 cells in 20 μl of com-
plete medium.
5. Add the cross-linker directly to the cell suspension at a final
concentration of 1 mM (1 μl of 20 mM stock in DMSO in a
total medium volume of 20 μl). Immediately mix gently. Take
one half of the cells without cross-linker treatment and add the
same volume of DMSO as a control. To check the quenching
efficiency, add the cross-linker in a tube already containing
30 mM of DTE. Incubate samples for 1 h at 37 C in 5 %
CO2. Swirl the suspension occasionally during the 37 C incu-
bation (see Notes 4 and 5).
6. Stop the cross-linking reaction by adding 4 μl of 6 quench-
ing buffer (for a final concentration of 30 mM). Let the cells
incubate for 10 min at 37 C to allow the quencher to fully
diffuse into the cells.
7. Visualize and compare cells which were treated and mock
treated by light microscopy, in order to check for cellular death.
8. Pellet cells by centrifugation at 4 C and wash twice with cold
CMF-PBS buffer at 4 C.
3.3.3 Preparation 1. Prepare protein extracts by directly adding 40 μl of SBLU 1

of Whole Cell Extracts buffer, vortex well, and heat cells at 100 C until lysis is
complete.
2. Let cell lysates return to room temperature before centrifuging
at 16,000 g for 10 min at 4 C. Transfer the soluble protein
fraction to a new microcentrifuge tube without touching the
pellet (see Note 6).
3. Samples are then analyzed by SDS-PAGE after appropriate
dilution with SBLU buffer (in general 1/10) and cross-linked
proteins are identified by Western blotting according to stan-
dard protocols.
Fig. 5 In vivo cross-linking of NDPK-B in HeLa cells. HeLa cells were treated for 1 h with long (BMH, top gels)
or short (BMOE, bottom gels) membrane permeable cross-linkers after transient transfections with a variable
amount of DNA encoding either the dimeric mutant or the wild type hexamer as indicated. After quenching the
reaction with DTE, samples were taken up in SDS-urea gel-solubilizing extraction buffer (SBLU) and were
analyzed by SDS-PAGE and Western blotting using a polyclonal antibody against NDPK-B as described in
Subheading 3.3.3. M, D*, H*, and NS* refer to the electrophoretic migrations of monomers, cross-linked
dimers, cross-linked hexamers, and higher-order nonspecific multimers of NDPK-B subunits, respectively.
Dimeric cross-links were quantified using a Storm system and the percentages of cross-linked dimers related
to either dimer (left) or hexamer (right) overexpression were represented versus the amount of DNA
transfected into HeLa cells. Note the double position of monomers especially when HeLa cells overexpressing
the dimer were treated with the short cross-linker. This was due to unmodified and one-point modified
monomers (see Note 2). Note also the considerable increase of dimeric cross-linked species with dimer
overexpression in HeLa cells treated by the long cross-linker. This increase was not as significant when the
same transfected cells were treated with the short cross-linker. In cells that overexpressed the hexamer there
were only small increases of dimer cross-links and the cross-linking was slightly more efficient with the short
cross-linker. Taken together, these results were very similar to those obtained in vitro (Fig. 3), indicating the
high selectivity of our protein cross-linking in a cellular context
3.3.4 Quantification of The technique we currently use for immunodetection of

Cross-Linked Species on cross-linked species is based on a chemiluminescent substrate sys-
Western Blots tem for horseradish peroxidase (HRP) which can be visualized both
on film and blot imaging system (ECL Plus, Amersham Bios-
ciences). Immunodetection is carried out according to the suppli-
er’s instructions and the quantification is performed using Storm
(Amersham Biosciences), which is better than film detection.
3.4 Concluding In vivo protein cross-linking based on structural design can greatly
Comments facilitate the introduction of selective cross-links on proteins to
study protein oligomerization or protein-protein interactions.
The combination of the chemoselectivity and regioselectivity of
the reagent with the protein structure can even allow examination
of changes in protein oligomerization upon cellular events and
specific protein-protein interactions upon ligand binding. By con-
sidering the specificity of the maleimide group for cysteine residues
and the spatial arrangement of these residues, selective cross-linking
can be introduced specifically on either the dimer or the hexamer
in vitro as well as in intracellular environments that are crowded
with many other cellular macromolecules. The method described
here allows investigation into the oligomerization of the NDPK-B
protein in metastatic, tumor, and normal cells (unpublished
results). This approach can be extended to signaling proteins,
such as the NEMO protein, an essential modulator involved in
the NF-κB pathway, which are usually less abundant in cells. We
have successfully used these methods to show that NEMO forms a
protein complex in cells, termed IκB kinase (IKK) complex, with
two IKKα/β kinases [20]. Strikingly, the molecular mass of the
cross-linked IKK complex perfectly matched the subunit composi-
tion of the IKK complex, which was accurately determined in vitro
more than one decade after the in vivo cross-linking experiments
[21]. Given the high sensitivity of this method to detect unstable
homo- and hetero-associations of proteins, the experimenter will
also have to address the question of whether the cross-links corre-
spond to a mature protein association or to an assembly intermedi-
ate. This is particularly true when in vivo cross-linking results are
compared in different cells at variable stages of proliferation.
4 Notes
1. Precautions should be taken when interpreting in vivo chemical

cross-linking results. If cross-links are observed in vivo, it may be
specific (reflecting a relevant structure) or it may be nonspecific.
Cross-linking on living cells increases the probability of detect-
ing unstable oligomers, but it also increases the probability of
nonspecifically cross-linking proteins together. When cross-links
are observed, the assay should be repeated with reduced

concentrations of cross-linker, at a reduced temperature or for
shorter periods of time to detect preferential cross-links. If iden-
tical results are obtained over a broad concentration range of
cross-linkers or with several reagents, the in vivo cross-links are
likely to be specific.
2. Often a monomer containing an internal cross-link forms a
complex with SDS that has a smaller Stokes radius than does
the uncross-linked monomer-SDS complex. This cross-linked
complex will then be observed on SDS-PAGE as a band migrat-
ing with lower apparent molecular weight than does the
uncross-linked monomer. This slight change of the electropho-
retic migration can also be observed for the cross-linked species
of dimer, trimer, etc., on Laemmli SDS-PAGE.
3. General equation for in vivo protein cross-linking
The formation of a specific cross-linking with a homo- or
hetero- cross-linker can be written as follows:
k1 ¼k0 ½CL k2
½P ----------! ½P ----! ½P
where [P] is the unmodified protein, [CL] the cross-linker,
[P*] the protein cross-linked to one end of the bifunctional
reagent, and [P**] the protein cross-linked to both ends of the
reagent. Usually for protein cross-linking [CL]Total>>[P]Total
we can then approximate [CL] by [CL]T and we can consider
the first reaction as pseudo-first-order. The general equation
for the variation of [P**] with time is:

k1 expðk2 t Þ k2 expðk1 t Þ
Equation 1: ½P = 1 þ ½P 0
k2 k1
where t is the time, k1 the pseudo-first-order rate constant, and
P0 the protein concentration at time 0. The first reaction is
usually the rate-determining step of the reaction with a bimo-
lecular reactant, meaning that whenever a P* protein is formed,
it is rapidly converted into P** (k2>>k1). Equation 1 then
becomes a more simple expression:
½P ¼ f1 expðk1 t Þg½P0

4. Because many proteins contained in the serum can interfere
with in vivo cross-linking, it is recommended to carry out the
reaction in PBS buffer, if cells can endure such a treatment up
to 2 h. There is no general rule to calculate the effective
concentration of cross-linker in serum-containing medium. In
our experiments with NDPK-B the effective concentration of
both cross-linkers in PBS is tenfold less compared to that in
medium containing 10 % serum.
5. It is possible to incubate the cross-linker for longer periods. We

performed a range of experiments with different times of incu-
bation, 10 , 60 and 120 min. The cross-linking efficiency was
optimal at 60 and 120 min, whereas the reaction was not
complete at 10 min. This obviously depends on the chemos-
electivity of the cross-linker as well as the type of functional
groups that are accessible on the protein. It also depends
mostly on the subcellular localization of the protein. The
in vivo cross-linking of the NEMO protein, a cytosolic signal-
ing protein that is located near the membrane, was optimal
after an incubation time of only 10 min [20].
6. It is important to note that the use of a high cross-linker
concentration can reduce the total amount of extracted pro-
teins. This is due to the in vivo formation of nonspecific inter-
molecular cross-links with a high molecular mass that are
removed by centrifugation. It is therefore crucial for any
in vivo protein cross-linking to determine the minimal effective
concentration of the cross-linker in order to reduce the proba-
bility of random collisions and to get a satisfactory yield of
protein after extraction. To examine the extent of global pro-
tein cross-linking, it is also advised to analyze the crude extracts
from the mock-treated and the reagent-treated cells by SDS-
PAGE followed by Coomassie or silver staining. If only a few
proteins were entirely cross-linked into species of high molecu-
lar mass, the cross-linking is likely specific.
Acknowledgments
The authors are very grateful to Samuel Levy, MD for his critical
reading of the manuscript. This work was supported, in whole or in
part, by the Fondation ARC pour la recherche sur le cancer and La
Ligue contre le cancer.
References
1. Ji TH (1983) Bifunctional reagents. Methods 5. Pornillos O, Ganser-Pornillos BK, Kelly BN
Enzymol 91:580–609 et al (2009) X-ray structures of the hexameric
2. Kluger R, Alagic A (2004) Chemical cross- building block of the HIV capsid. Cell
linking and protein-protein interactions-a 137:1282–1292
review with illustrative protocols. Bioorg 6. Mika JT, Poolman B (2011) Macromolecule
Chem 32:451–472 diffusion and confinement in prokaryotic
3. Petrotchenko EV, Borchers CH (2010) Cross- cells. Curr Opin Biotechnol 22:117–126
linking combined with mass spectrometry for 7. Wang K, Richards FM (1975) Reaction of
structural proteomics. Mass Spectrom Rev dimethyl-3,3’-dithiobispropionimidate with
29:862–876 intact human erythrocytes. Cross-linking of
4. Stengel F, Aebersold R, Robinson CV (2012) membrane proteins and hemoglobin. J Biol
Joining forces: integrating proteomics and Chem 250:6622–6626
cross-linking with the mass spectrometry of 8. Bruce JE (2012) In vivo protein complex
intact complexes. Mol Cell Proteomics 11 topologies: sights through a cross-linking
(R111):014027 lens. Proteomics 12:1565–1575
9. Boissan M, Lacombe ML (2011) Learning 15. Mesnildrey S, Agou F, Veron M (1997) The
about the functions of NME/NM23: lessons in vitro DNA-binding properties of NDP
from knockout mice to silencing strategies. kinase are related to its oligomeric state. FEBS
Naunyn Schmiedebergs Arch Pharmacol Lett 418:53–57
384:421–431 16. Agou F, Raveh S, Mesnildrey S et al (1999)
10. Lascu I, Gonin P (2000) The catalytic mecha- Single strand DNA specificity analysis of
nism of nucleoside diphosphate kinases. J Bioe- human nucleoside diphosphate kinase B.
nerg Biomembr 32:237–246 J Biol Chem 274:19630–19638
11. Dexheimer TS, Carey SS, Zuohe S et al (2009) 17. Pogulis RJ, Vallejo AN, Pease LR (1996) In
NM23-H2 may play an indirect role in tran- vitro recombination and mutagenesis by over-
scriptional activation of c-myc gene expression lap extension PCR. Methods Mol Biol
but does not cleave the nuclease hypersensitive 57:167–176
element III(1). Mol Cancer Ther 18. Kluger R (1997) Chemical cross-linking and
8:1363–1377 protein function. In: Creighton TE (ed) Pro-
12. Postel EH, Berberich SJ, Flint SJ et al (1993) tein function a practical approach. IRL Press,
Human c-myc transcription factor PuF identi- Oxford, p 185
fied as nm23-H2 nucleoside diphosphate 19. Morera S, Lacombe M-L, Xu Y et al (1995)
kinase, a candidate suppressor of tumor metas- X-Ray structure of nm23 Human Nucleoside
tasis. Science 261:478–480 Diphophate Kinase B complexed with GDP at
13. Mesnildrey S, Agou F, Karlsson A et al (1998) 2A resolution. Structure 3:1307–1314
Coupling between catalysis and oligomeric 20. Agou F, Ye F, Goffinont S et al (2002)
structure in NDP kinase. J Biol Chem NEMO trimerizes through its coiled-coil C-
273:4436–4442 terminal domain. J Biol Chem 277:
14. Karlsson A, Mesnildrey S, Xu Y et al (1996) 17464–17475
Nucleoside diphosphate kinase. Investigation 21. Napetschnig J, Wu H (2013) Molecular Basis
of the intersubunit contacts by site-directed of NF-kappaB Signaling. Annu Rev Biophys
mutagenesis and crystallography. J Biol Chem 42:443–468
271:19928–19934
Part IV
Cell-Based Bimolecular Interaction Reporter Assays

Chapter 27
Identification of Protein-Protein Interactions by Standard

Gal4p-Based Yeast Two-Hybrid Screening
Jeroen Wagemans and Rob Lavigne
Abstract
Yeast two-hybrid (Y2H) screening permits identification of completely new protein interaction partners for
a protein of interest, in addition to confirming binary protein-protein interactions. After discussing the
general advantages and drawbacks of Y2H and existing alternatives, this chapter provides a detailed protocol
for traditional Gal4p-based Y2H library screens in Saccharomyces cerevisiae AH109. This includes bait
transformation, bait auto-activation testing, prey library transformation, Y2H evaluation, and subsequent
identification of the prey plasmids. Moreover, a one-on-one mating protocol to confirm interactions
between suspected partners is given. Finally, a quantitative α-galactosidase assay protocol to compare
interaction strengths is provided.
Key words Protein-protein interactions, Gal4p, Yeast two-hybrid, α-galactosidase assay
1 Introduction
The development of the yeast two-hybrid (Y2H) system by Stanley

Fields in 1989 [1] represented a major milestone in the study of
protein-protein interactions, which play a crucial role in almost all
biological processes. Although originally used to detect binary
protein-protein interactions between known interaction partners,
it is now clear that Y2H facilitates the identification of completely
new protein interaction partners for a protein of interest.
1.1 Gal4p-Based The standard Y2H system exploits the modular nature of eukary-
Yeast Two-Hybrid otic transcription factors like Gal4p of Saccharomyces cerevisiae. This
transcriptional activator has two functional domains: an amino
terminal DNA binding domain and a carboxy terminal activation
domain. While the former binds to the upstream activating
sequence (UAS) of the galactose metabolism genes in S. cerevisiae,
the latter activates their transcription by recruiting the RNA poly-
merase to the promoter region. Moreover, Keegan et al. [2]
demonstrated that Gal4p remains functional even if the two
409
410 Jeroen Wagemans and Rob Lavigne
Fig. 1 The Y2H system. (a) Transcriptional activator Gal4p of S. cerevisiae is composed of two functional
domains: the DNA binding domain (DBD) that binds to the upstream activating sequence (UAS) and the
activation domain (AD) that recruits the RNA polymerase and activates transcription of a reporter gene. Both
components are required to produce a selectable phenotype. Y2H uses the modularity of Gal4p. (b) The bait
protein X is fused to the Gal4p DBD, which cannot activate transcription on its own. (c) The prey protein Y is
fused to the Gal4p AD. Since this fusion protein is not recruited to the promoter, it is also unable to activate
transcription on its own. (d) Only when X and Y interact, the DBD and AD are both present at the promoter site,
restoring Gal4p function resulting in a detectable phenotype
domains are not covalently attached as long as they are both present
at the promoter site. Fields [1] picked up this idea to fuse interact-
ing proteins to the two Gal4p domains, which physically brings the
two domains together and restores Gal4p function.
In Gal4p-based Y2H, the protein fused to the C-terminus of
the Gal4p DNA binding domain is known as the bait, while the
other interaction partner, fused to the C-terminus of the Gal4p
activation domain, is called the prey. These constructs are both
transformed into the same yeast cell. Finally, a third component
of the Y2H system (Fig. 1) is the reporter construct in this bait- and
prey-expressing yeast cell. This reporter gene is located down-
stream from a promoter with a UAS recognized by the Gal4p
DNA binding domain. If bait and prey interact, the Gal4p activa-
tion domain is tethered to the promoter region and transcription of
the reporter gene is activated. This results in a specific, selectable
phenotype.
While Fields originally designed the system to investigate
known interactions, Chien et al. [3] applied the technique 2 years
later to unravel completely new interactions for a particular protein
of interest. The known protein, fused to the Gal4p DNA binding
domain, was used as a bait to screen for interaction partners in a
whole library of potential interaction partners, fused to the activa-
tion domain. After Y2H interaction analysis, the prey plasmid was
harvested from those yeast colonies able to activate all reporter
genes. Finally, by sequencing the prey plasmid inserts, the interac-
tion partner of the protein of interest was identified.
Gal4p-Based Yeast Two-Hybrid 411
1.2 Drawbacks of As for any other protein-protein interaction analysis technique, one
the Yeast Two-Hybrid should realize it is almost impossible to detect all interacting pro-
System and Possible teins using Y2H. In contrast, Y2H can also identify interactions
Solutions that actually do not occur in vivo. These are called false negative and
false positive results, respectively. These often arise due to the use of
artificial fusion proteins in combination with an artificial reporter
construct, the removal of the proteins from their natural biological
context, their translocation to the yeast nucleus, or the production
of the hybrid protein at a different level than its natural concentra-
tion in the cell [4].
One potential problem is the bait protein, which could inde-
pendently lead to reporter gene activation (e.g., if the protein has
transcription activation activity or interacts nonspecifically with the
Gal4p activation domain). This is called auto-activation or self-
activation and can lead to false positive results. While this activity
can be related to the biological function of the protein, it can also
be a consequence of the fusion with the DNA binding domain or
the removal of the protein from its natural context [4]. Because of
this phenomenon, auto-activation of the bait protein should always
be checked before performing the YH2 screen. If this auto-
activation test is positive, the bait as such is not suited for the
screen. To resolve this problem, the transcription activation domain
could be removed and the screen performed with the other
domain, or another Y2H system could be used.
While only the lacZ reporter gene, encoding β-galactosidase,
was originally used in a classical blue/white screening assay [1],
current Y2H reporter yeast strains have several secondary auxotro-
phic reporter genes, which permit selection for interaction by
monitoring yeast cell growth on selective media. For instance,
Y2H strain S. cerevisiae AH109 [5] has the HIS3 and ADE2
reporter genes, encoding imidazole glycerol phosphate dehydratase
and phosphoribosylaminoimidazole carboxylase, respectively.
These enzymes are crucial in the biosynthesis of histidine and
adenine. By plating on medium lacking histidine and/or adenine,
interacting baits and preys are selected. In addition, S. cerevisiae
AH109 contains the MEL1 reporter gene, encoding α-galactosi-
dase, a secreted enzyme enabling direct blue/white screening on
X-α-gal indicator plates. The use of more than one reporter gene
reduces the number of false positive results [5].
Moreover, when using HIS3 as a reporter gene, as is the case
with strain S. cerevisiae AH109, 3-amino-1,2,4-triazole (3-AT)
should be added to the medium. HIS3 is the most leaky construct
of the three reporter genes in AH109, so there is always a low
background transcription activation of this gene. 3-AT is a compet-
itive inhibitor of the HIS3 product and will reduce false positive
results due to the background activation [6]. At the same time, the
3-AT concentration should not be too high to avoid losing identi-
fication of weaker interactions. The optimal concentration that
should be used in the Y2H screen is also determined in the auto-

activation test.
Although Y2H tests for interactions in vivo, the use of yeast as a
host can be a potential disadvantage. It is possible that the fusion
proteins will not be stable in the yeast cell or will require specific
posttranslational modifications. Moreover, some proteins might be
toxic to yeast. Bacterial two-hybrid [7] and mammalian two-hybrid
[8] assays are possible alternatives to overcome these problems.
Targeting of non-nuclear proteins to the yeast nucleus can also be
inefficient or could cause nonspecific interactions, which is likely for
proteins with hydrophobic domains. For instance, membrane pro-
teins forced into the nucleus tend to bind nonspecifically to other
proteins [4]. To address this issue, the membrane yeast two-hybrid
system, based on the split-ubiquitin assay [9], was developed.
Furthermore, interactions depending on more than two part-
ners are not detected using Y2H. The yeast three-hybrid assay [10]
provides a possible solution.
A last but common class of false positive results occurs because
of spontaneous mutations in the yeast strain, bait or prey constructs
causing reporter gene activation [11]. Hence, it is important to
confirm the potential interacting proteins in an independent Y2H
experiment using fresh yeast cells.
In addition to false positive results, the use of artificial fusion
proteins can also lead to false negative results (e.g., if the expression
levels are too low or when the fusion alters the actual conformation
of the bait or prey protein). This could result in a reduced activity or
the exposure of artificial surfaces that nonspecifically interact with
other proteins. Moreover, the domain fused to the protein could
occlude the natural interaction domain or cause steric hindrance
[4]. To reduce the number of false negative results and to maximize
the number of identified interactions, it is advisable to perform
screens in more than just one Y2H vector system. Although the
original system uses the modularity of Gal4p, other DNA binding
domains like LexA [12], the λ repressor protein cI [13], or the
human estrogen receptor protein [14] could be used. Similarly,
different transcription activation domains like the herpes simplex
virus protein VP16, or its alternative B42, have been applied [12].
Application of different Gal4p-based vector systems is also a
possibility. As the expression level of the bait and prey fusion con-
structs influences the outcome of the Y2H analysis, vectors with
varying promoters and copy numbers due to a different origin of
replication (ori) could be exploited [15, 16]. For Y2H, our lab uses
a combination of the pGBKT7g-pGADT7g (Gateway variants of
the Clontech Matchmaker 3 vectors) and the pDEST32-pDEST22
(Life Technologies) plasmids that lead to different levels of expres-
sion (see Table 1). Hence, their results are complementary.
Although one can easily use classical restriction enzyme cloning to
construct bait plasmids and prey libraries, it is preferable to use the
Table 1
Summary of applied yeast two-hybrid vectors
Characteristics Selection
Vector Promoter Ori E. coli Yeast Reference

pGBKT7g (bait) Truncated pADH1 2μ Kanamycin TRP1 [17]
pGADT7g (prey) Full-length pADH1 2μ Ampicillin LEU2 [17]
pDEST32 (bait) Full-length pADH1 CEN Gentamicin LEU2 Life Technologies
pDEST22 (prey) Full-length pADH1 CEN Ampicillin TRP1 Life Technologies
Gateway system from Life Technologies to facilitate subcloning

into different Y2H systems. By first constructing a prey library in
an entry vector (e.g., pENTR1A, Life Technologies), one can easily
shuttle the library to different prey vectors and use these different
libraries in subsequent screens.
1.3 Advantages of Despite its drawbacks, the Y2H system has some clear advantages
the Yeast Two-Hybrid over other methods. Firstly, it is clear that an in vivo assay more
Approach closely resembles cellular conditions. Furthermore, Y2H is able to
detect weak and transient interactions since a continuous activation
of the reporter genes is not necessary to see a signal. The fact that
Y2H only detects binary interactions can also be an advantage.
In contrast to methods such as affinity purification, which detect
whole complexes, Y2H directly identifies the two interaction part-
ners. Moreover, there is no need for recombinant proteins, which is
a huge advantage. Finally, although it is labor intensive, Y2H has a
relatively low investment cost. Only minimal laboratory equipment
and low-cost microbial growth media are needed [4].
This chapter provides a detailed protocol for Gal4p-based Y2H
screens using one protein as bait against a prey library. Assuming a
bait and prey library has been constructed, this protocol will first
describe small-scale yeast transformation of the bait plasmid.
In addition, this chapter includes a bait auto-activation test, large-
scale yeast transformation of the prey library, evaluation of the Y2H
results, yeast plasmid isolation, and identification of the inserts by
DNA sequencing. Since Y2H is prone to false positive results, it is
important to confirm the potential interactions. A one-on-one
protocol using mating between the two different S. cerevisiae strains
AH109 (mating type a) [5] and Y187 (mating type α) [18] will also
be discussed. This protocol can also be applied to directly confirm
interaction between two suspected interaction partners. Finally, a
quantitative α-galactosidase assay to compare different interaction
strengths will be provided.
2 Materials
2.1 Bait Plasmid 1. Yeast strain: S. cerevisiae AH109 with genotype MATa,
Transformation trp1-901, leu2-3,112, ura3-52, his3-200, gal4Δ, gal80Δ, LYS2::
GAL1UAS-GAL1TATA-HIS3,GAL2UAS-GAL2TATA-ADE2,
URA3::MEL1UAS-MEL1TATA-lacZ (Clontech) [5].
2. Bait plasmids: Gene of interest cloned in bait vector (e.g.,
pGBKT7g [17] or pDEST32 (Life Technologies)).
3. YPDA liquid medium: 10 g yeast extract, 20 g peptone. Bring
up to 880 ml with deionized water and autoclave. After sterili-
zation, let the medium cool to room temperature. Add 100 ml
of 20 % glucose stock solution and 20 ml of 2 mg/ml adenine
stock solution for a final volume of 1 L. Store at room
temperature.
4. YPDA solid medium: 10 g yeast extract, 20 g peptone, 20 g
agar. Bring up to 880 ml with deionized water and autoclave.
After sterilization, let the medium cool to 60 C. Add 100 ml of
20 % glucose stock solution and 20 ml of 2 mg/ml adenine
stock solution for a final volume of 1 L. Pour plates in a sterile
hood and let the medium solidify. Store plates at 4 C for 2–3
weeks.
5. 2 mg/ml Adenine stock solution: 200 mg adenine hemisulfate
salt. Bring up to 100 ml with deionized water and sterilize
through a 0.22 μm filter.
6. 20 % Glucose stock solution: Put 800 ml of ultrapure water in a
beaker and stir. Weigh 200 g α-D (+)-glucose (anhydrous) and
add it to the ultrapure water in small amounts to ensure that it is
completely dissolved. Autoclave.
7. Sterile water: autoclave ultrapure water.
8. 1 M LiOAc: 5.1 g lithium acetate dihydrate. Bring up to 50 ml
with ultrapure water and autoclave. Store at room temperature.
9. 100 mM LiOAc: 0.51 g lithium acetate dihydrate. Bring up to
50 ml with ultrapure water and autoclave.
10. 50 % (w/v) PEG3350: Add 50 g polyethylene glycol (average
molecular weight 3,350) to 50 ml of ultrapure water. Stir until
completely dissolved. Filter sterilize and store at room temper-
ature, securely capped to prevent evaporation. Water loss will
increase the PEG concentration and severely reduce the yield of
transformants. To avoid this, make fresh PEG3350 solution
every few months.
11. ssDNA: 10 mg/ml single-stranded fish sperm DNA, MB
grade. Store at 20 C in aliquots.
12. Selective liquid yeast medium: 6.9 g yeast nitrogen base without
amino acids. Bring up to 700 ml with ultrapure water and
autoclave. Let the medium cool to room temperature and add

100 ml of 10 amino acid stock solution, 100 ml of 20 %
glucose stock solution, and 100 ml of 10 amino acid dropout
solution for a final volume of 1 L.
13. Selective liquid yeast medium (“synthetic defined” or “SD
medium”): 6.9 g yeast nitrogen base without amino acids,
20 g bacto agar. Bring up to 700 ml with ultrapure water and
autoclave. Let the medium cool to 60 C and add 100 ml of
10 amino acid stock solution, 100 ml of 20 % glucose stock
solution, and 100 ml of 10 amino acid dropout solution for a
final volume of 1 L. If needed, add 3-AT and/or X-α-gal and
pour the plates in a sterile culture hood. Store at 4 C for up to
2–3 weeks.
14. 10 amino acid stock solution: 500 mg L-arginine, 800 mg
L-aspartic acid, 500 mg L-isoleucine, 500 mg L-lysine,
200 mg L-methionine, 500 mg L-phenylalanine, 1,000 mg
L-threonine, 500 mg L-tyrosine, 1,400 mg L-valine, and
200 mg uracil. Bring up to 1 L with ultrapure water and
sterilize with a 0.22 μm filter. Store at 4 C for up to 1 month.
15. 10 amino acid dropout (DO) stock solutions: Depending on
the selective medium, add the remaining components not
provided by the 10 amino acid stock solution.
SD-Trp medium (medium without tryptophan): add 10
DO-Trp: 40 mg adenine hemisulfate salt, 20 mg L-histi-
dine, 100 mg L-leucine dissolved in 100 ml of ultrapure
water (filter sterilize).
SD-Leu medium (medium without leucine): add 10 DO-Leu:
40 mg adenine hemisulfate salt, 20 mg L-histidine,
50 mg L-tryptophan in 100 ml of ultrapure water (filter
sterilize).
SD-Trp-Leu medium (medium lacking tryptophan and leu-
cine): add 10 DO-Trp-Leu: 40 mg adenine hemisulfate
salt, 20 mg L-histidine in 100 ml of ultrapure water (filter
sterilize).
SD-Trp-Leu-His medium (medium lacking tryptophan, leu-
cine, and histidine): add 10 DO-Trp-Leu-His: 40 mg ade-
nine hemisulfate salt in 100 ml of ultrapure water (filter
sterilize).
SD-Trp-Leu-His-Ade medium: add 10 DO-Trp-Leu-His-
Ade, which is the same as sterile ultrapure water.
16. 1 M 3-AT stock solution: 0.84 g 3-amino-1,2,4-triazole in 10 ml
of ultrapure water. Filter sterilize. Store at 4 C.
17. X-α-gal stock solution: 200 mg 5-bromo-4-chloro-3-indolyl-α-
D-galactopyranoside dissolved in 10 ml of N,N-dimethylforma-
mide. Store at 20 C in the dark (e.g., a conical tube wrapped
in aluminum foil). Add 2 ml of stock solution to 1 L of SD

medium. As X-α-gal is a heat-sensitive compound, let the
medium first cool to 60 C before adding.
2.2 Bait Auto- 1. Yeast strain: S. cerevisiae AH109 containing the bait plasmid.
activation Test 2. Prey plasmids: empty prey vectors or prey vectors with an
independent protein inserted (e.g., if the bait is constructed
in pGBKT7g, use the empty pGADT7g vector as a control; if
the bait is constructed in pDEST32, use the empty pDEST22
as a control).
3. Solid media needed (25 ml medium/90 mm petri dish):
2 SD-Trp-Leu plates
1 SD-Trp-Leu-His plate
1 SD-Trp-Leu-His +1 mM 3-AT plate (25 μl 1 M 3-AT/25 ml
medium)
1 SD-Trp-Leu-His +3 mM 3-AT plate (75 μl 1 M 3-AT/25 ml
medium)
1 SD-Trp-Leu-His +5 mM 3-AT plate (125 μl 1 M 3-AT/
25 ml medium)
1 SD-Trp-Leu-His +10 mM 3-AT plate (250 μl 1 M 3-AT/
25 ml medium)
1 SD-Trp-Leu-His-Ade plate
2.3 Prey Library 1. Yeast strain: S. cerevisiae AH109 containing the bait plasmid.
Transformation 2. Prey library: library cloned in prey vector (e.g., pGADT7g [17]
or pDEST22 (Life Technologies)).
3. 2 YPDA liquid medium: 20 g yeast extract, 40 g peptone.
Bring up to 760 ml with deionized water and autoclave. After
sterilization, let the medium cool down. Add 200 ml of 20 %
glucose stock solution and 40 ml of 2 mg/ml adenine stock
solution for a final volume of 1 L.
4. Solid media needed (25 ml medium/90 mm petri dish; 80 ml
medium/150 mm petri dish):
12 SD-Trp-Leu 90 mm plates
30 SD-Trp-Leu-His 150 mm plates + x mM 3-AT (concentra-
tion determined from auto-activation test)
2.4 Evaluation of 1. Solid media needed (40 ml medium/Nunc OmniTray dish or

Library Transformation similar microtiter plate):
1 OmniTray SD-Trp-Leu
1 OmniTray SD-Trp-Leu-His + x mM 3-AT (concentration
determined from auto-activation test)
1 OmniTray SD-Trp-Leu-His-Ade + 40 mg/L X-α-gal
2.5 Yeast Plasmid 1. 0.67 mM K2HPO4 pH 7.5: 0.0306 g K2HPO4. Dissolve in

Isolation 150 ml of deionized water, adjust pH to 7.5, bring up to
200 ml with deionized water and autoclave.
2. Zymolyase solution: 12.5 mg zymolyase 20T (250 U) in 50 μl of
0.67 mM K2HPO4, pH 7.5, for each yeast colony. Prepare
fresh before each use.
2.6 Identification 1. Competent Escherichia coli cells: such as One Shot TOP10
of Prey Inserts chemically competent E. coli (Life Technologies).
2. LB medium: 10 g tryptone, 5 g yeast extract, 10 g NaCl (15 g
agar). Bring up to 1 L with deionized water and autoclave. Let
cool before adding antibiotics.
3. Resuspension buffer: 6.06 g tris base, 3.72 g ethylenediamine-
tetraacetic acid disodium salt dihydrate (Na2EDTA 2H2O).
Add approximately 800 ml of deionized water and adjust pH
to 8. Bring up to 1 L with deionized water and autoclave. After
autoclaving, add DNase- and proteinase-free RNase A to a final
concentration of 100 μg/ml and store at 4 C.
4. Lysis buffer: 8 g NaOH, 10 g sodium dodecyl sulfate (SDS).
Bring up to 1 L with demineralized water. Filter sterilize and
store in a plastic bottle at room temperature. When stored too
cold, a white precipitate will appear. The precipitated SDS can
be dissolved again by microwaving.
5. Neutralization buffer: 294.5 g potassium acetate. Dissolve in
500 ml of deionized water and adjust pH to 5.5 with glacial
acetic acid (more than 100 ml will be needed). Bring up to 1 L
and autoclave. Store at room temperature.
6. 70 % ice cold ethanol: 70 ml ethanol, 30 ml deionized water.
Store at 20 C.
7. Sequencing primers: specific for the prey plasmids. E.g:
pDEST22_F: 50 -TATAACGCGTTTGGAATCACT-30
pDEST22_R: 50 -AGCCGACAACCTTGATTGGAGAC-30
pGADT7g_F: 50 -CTATTCGATGATGAAGATACCCCAC-
CAAACCC-30
pGADT7g_R: 50 -GTGAACTTGCGGGGTTTTTCAGTATC-
TACGATT-30
2.7 One-on-One Y2H 1. Yeast strains:

Confirmation Test S. cerevisiae AH109 containing the bait plasmid (generated in
Subheading 3.1).
S. cerevisiae AH109 containing an empty bait vector or a bait
vector with an independent gene inserted (this strain can
be generated with the same protocol as in Subheading 3.1).
S. cerevisiae Y187 with genotype MATα, ura3-52, his3-200,

ade2-101, trp1-901, leu2-3,112, gal4Δ, met, gal80Δ,
URA3::GAL1UAS-GAL1TATA-lacZ (Clontech) [18].
2. Solid media needed (40 ml medium/Nunc OmniTray dish)
1 OmniTray YPDA
1 OmniTray SD-Trp-Leu
1 OmniTray SD-Trp-Leu-His + x mM 3-AT
1 OmniTray SD-Trp-Leu-His-Ade + 40 mg/L X-α-gal
2.8 Quantitative 1. Assay buffer: 2 volumes of 1 NaOAc buffer combined with 1

α-Galactosidase Assay volume of 100 mM PNP-α-Gal solution. Mix well. Prepare
fresh before each use.
2. 100 mM PNP-α-Gal solution: 30.13 mg p-nitrophenyl-α-D-
galactopyranoside in 1 ml of deionized water. Filter sterilize.
Prepare fresh before each use.
3. 1X NaOAc: 2.05 g sodium acetate in 50 ml of deionized water
(0.5 M NaOAc). Adjust pH to 4.5.
4. 10X Stop solution: 10.6 g Na2CO3 in 100 ml of deionized water
(1 M Na2CO3).
3 Methods
3.1 Bait Plasmid This protocol is adapted from the lithium acetate/single-stranded
Transformation carrier DNA/polyethylene glycol method by Gietz and Woods
[19], which permits high-efficiency transformation of yeast. This
method generally yields 105 transformants per μg of DNA.
1. Streak S. cerevisiae strain AH109 on a fresh YPDA agar plate
starting from a 25 % glycerol cell stock (stored at 80 C).
Incubate for 2–3 nights at 30 C (see Note 1).
2. Inoculate two 5 ml YPDA cultures (in 50 ml flasks) with one
big (3 mm) or two to four small (1–2 mm) colonies. Inoculate
different numbers of colonies for the two cultures to ensure
different cell densities the next day. Incubate overnight in a
shaking incubator at 30 C. The next morning, measure the
optical density at 600 nm (OD600). A preculture with a value
below 1 is preferred as these cells are still not stationary and will
have a shorter lag phase. Use this overnight culture to inoculate
a fresh yeast culture.
3. Inoculate 50 ml of YPDA (in a 500 ml flask) until an OD600
value of approximately 0.120–0.140 is achieved (see Note 2).
Incubate at 30 C until the OD600 reaches 0.480–0.560 (this
usually takes 4–5 h). At least two cell divisions are needed for a
good transformation efficiency. This efficiency remains con-
stant during the next three to four cell divisions.
4. From this point forward, all steps are performed at room

temperature. Centrifuge the cells for 5 min at 3,500 g.
5. Discard the medium and resuspend the cells in a half volume of
sterile water (25 ml). Centrifuge again for 5 min at 3,500 g
to wash the cells.
6. Discard the water and resuspend the cell pellet in 1 ml of
100 mM LiOAc (see Note 3). Transfer to a 1.5 ml microcen-
trifuge tube.
7. Pellet the cells (14,000 g, 15 s) and remove the LiOAc with a
micropipette. Resuspend the cells in 400 μl of 100 mM LiOAc.
The final volume is now approximately 500 μl. This is sufficient
for ten transformations (see Note 4). Divide the cells into 50 μl
aliquots.
8. Centrifuge (14,000 g, 15 s) and remove the LiOAc. The
cells are now ready for transformation. The transformation
mixture is added in the following order (see Note 5):
240 μl PEG3350 (50 % w/v)
36 μl 1 M LiOAc
10 μl ssDNA (10 mg/ml)
x μl (1 μg) plasmid DNA
74 x μl sterile water
Total 360 μl
9. Vortex for 1 min and incubate the transformation mixture in a
water bath set at 42 C for 30 min. Invert tubes every 5–10 min
(see Note 6).
10. Centrifuge the mixture for 15 s at 3,500 g and remove the
supernatant with a micropipette. Resuspend the cells in 1 ml of
sterile water.
11. After resuspension, plate the cells on selective medium (10 μl,
100 μl, remaining volume) (e.g., SD-Trp for pGBKT7g and
SD-Leu for pDEST32) and incubate at 30 C for 3 nights (see
Note 7).
3.2 Bait Auto- This protocol also uses the LiOAc/ssDNA/PEG method. The
activation Test important differences are highlighted in bold.
1. Streak S. cerevisiae AH109 containing the bait plasmid
(obtained in Subheading 3.1) on a fresh selective agar plate.
2. Inoculate two 5 ml SD medium cultures (in 50 ml flasks) with
one big (3 mm) or two to four small (1–2 mm) colonies.
Incubate overnight in a shaking incubator at 30 C. The next
morning, determine the OD600 and use the overnight culture
to inoculate a fresh yeast culture.
3. Steps 3–10 are exactly the same (see Note 9). In step 8, use
1 μg of empty prey vector or an independent prey construct to
transform the S. cerevisiae AH109 containing the bait plasmid.
4. After resuspension in 1 ml of sterile water, 10 μl and 100 μl are
plated on SD-Trp-Leu to check the transformation efficiency.
The remaining volume (890 μl) is divided between SD-Trp-
Leu-His +0, 1, 3, 5 and 10 mM 3-AT (100 μl on every plate)
and SD-Trp-Leu-His-Ade (the remaining volume) plates.
All plates are incubated at 30 C for 5 days.
5. Evaluation: a background of very small colonies, due to leaky
expression of the HIS3 reporter, can usually be observed. Only
large colonies on the SD-Trp-Leu-His plates (or maybe even
on the SD-Trp-Leu-His-Ade plate) are an indication of auto-
activation. Out of the different concentrations of 3-AT, choose
one concentration that will be used in the Y2H screen. This
concentration should not be too low to reduce the number of
false positive “interactions” but also not too high, to reduce the
chance of missing weaker interactions (see Note 10).
3.3 Prey Library 1. Streak S. cerevisiae AH109 containing the bait plasmid
Transformation (obtained in Subheading 3.1) on a fresh selective agar plate.
2. Day 1: Inoculate the freshly streaked strain in 5 ml of selective
SD medium in a 50 ml flask (use several small colonies). Incu-
bate overnight at 30 C.
3. Day 2: Inoculate two 100 ml cultures of selective SD medium
in 1 L baffled flasks with the overnight culture (use two differ-
ent volumes: 1 and 2 ml) (see Note 12). The next day, one of
the 100 ml cultures should have an OD600 below 1 (still
exponentially growing, but preferably also close to 1 to have
enough inoculum for a 600 ml culture). Prewarm 600 ml
2 YPDA in a 2 L baffled flask at 30 C and autoclave two
additional empty 2 L baffled flasks and a 250 ml graduated
cylinder.
4. Day 3: Measure the OD600. Centrifuge the culture with the
value the closest to, but still below, 1 (divide between two
50 mL tubes) (3,500 g, 5 min). Discard the supernatant
and resuspend the cell pellets in 1 ml of 2 YPDA. Use this
concentrated cell solution to inoculate the 600 ml 2 YPDA
culture to an OD600 of approximately 0.120–0.140. Divide the
600 ml culture between the three 2 L baffled flasks using the
sterile graduated cylinder. Incubate the three cultures at 30 C
in a shaking incubator until the OD600 reaches 0.480–0.560.
This will take 4–5 h.
5. Divide the culture in twelve 50 ml sterile tubes. Centrifuge
(3,500 g, 5 min).
6. Discard the supernatant and wash the cells in a half volume

(300 ml) of sterile water (50 ml/2 cell pellets) (see Note 13).
Centrifuge again.
7. Discard the supernatant and wash the cells a second time with
300 ml of sterile water. Collect the cells by centrifugation.
Resuspend all cell pellets in 4 ml of sterile water.
8. Prepare the 120 transformation mixture (see Note 14):
28.8 ml PEG3350 (50 % w/v)
4.32 ml 1 M LiOAc
1.2 ml ssDNA (10 mg/ml)
x ml (120 μg) prey library DNA
4.88 x ml sterile water
9. Vortex the transformation mixture vigorously and add it to the
competent cells which were resuspended in 4 ml of water.
Vortex the entire mixture for 10 min (see Note 15).
10. Divide the solution into six 15 ml conical tubes and incubate
for 30 min at 30 C followed by 45 min at 42 C in a water bath
(see Note 16). Invert the tubes every 5–10 min.
11. Collect all cells in one 50 ml tube and centrifuge for 5 min at
3,500 g. Discard the transformation solution and resuspend
the cells in 12 ml of sterile water.
12. Plate three dilution series on SD-Trp-Leu to calculate the
transformation efficiency. For each series, 10 μl of cells are
diluted in 990 μl of sterile water (102). 100 μl from this
dilution is further diluted in 900 μl water (103). Repeat for a
104 and 105 dilution. Mix every dilution again right before
plating 100 μl on SD-Trp-Leu.
13. Divide the remaining volume (400 μl per plate) over 30 SD-
Trp-Leu-His + x mM 3-AT 150 mm plates (use the appropri-
ate concentration of 3-AT determined from the auto-activation
test) (see Note 17).
14. After 3 nights at 30 C, the dilution series can be counted.
Calculate the average. The transformant yield is the average
10 (100 μl plated) dilution factor 12 (total volume).
The minimum number of transformants needed depends
on the library used (see Note 18). Incubate the 150 mm
SD-Trp-Leu-His plates at 30 C for 5–7 nights.
3.4 Evaluation of 1. Check all 30 plates and select a maximum of 96 yeast colonies
Library Transformation to analyze (see Note 19).
2. Pick the selected colonies with a toothpick and transfer them to
a microtiter plate containing 100 μl of sterile water.
3. Shake for 15 min on a shaker to resuspend all cells.
4. Spot in parallel 2 μl on an SD-Trp-Leu, SD-Trp-Leu-His + x

mM 3-AT, and SD-Trp-Leu-His-Ade + 40 mg/L X-α-gal
OmniTray plate (see Note 20).
5. Incubate at 30 C for 5–7 nights.
6. Add 100 μl of sterile water +30 % glycerol to the remaining
volume of resuspended cells. This backup cell stock can be
stored at 80 C.
7. Evaluation: everything should grow on the SD-Trp-Leu and
the SD-Trp-Leu-His + x mM 3-AT plates. Interactions should
also be strong enough to activate not only the HIS3 reporter
but also the ADE2 and MEL1 reporter genes, so colonies of
interest are those that can grow on the SD-Trp-Leu-His-Ade +
X-α-gal plate and have a blue halo around them caused by the
α-galactosidase activity. A maximum of 24 blue colonies should
be selected for further analysis (see Note 21).
3.5 Yeast Plasmid 1. Inoculate the colonies in 5 ml of selective SD medium and

Isolation incubate for 2 nights at 30 C (see Note 22). It is only necessary
to select for the prey plasmid.
2. Centrifuge for 5 min at 3,500 g and discard the supernatant.
Resuspend the cell pellet in 250 μl of 0.67 mM K2HPO4,
pH 7.5. Transfer to a 1.5 ml tube.
3. Add 50 μl of zymolyase solution and incubate in a water bath at
35 C. Invert regularly.
4. Collect cells by centrifugation for 5 min at 3,500 g. Discard
the supernatant.
5. The zymolyase will degrade the yeast cell wall and create spher-
oplasts; from now on a standard bacterial miniprep kit can be
used to isolate the yeast plasmid.
3.6 Identification Because of its low yield and impurity, the isolated yeast prey plasmid
of Prey Inserts cannot be directly used as a template in DNA sequencing analysis.
To identify the prey inserts, the plasmid should first be transformed
back to E. coli. A direct PCR is not advisable, as there is still some
genomic yeast DNA present that may cause nonspecific products
and there can be more than one type of prey plasmid present.
1. For each blue colony, transform 5 μl of yeast plasmid DNA to
competent E. coli cells. Plate cells on LB medium selecting for
the prey plasmid.
2. For each transformation, four colonies (derived from one yeast
colony) are selected for further analysis (see Note 23). Grow
the yeast overnight in a 4 ml selective LB culture and isolate the
prey plasmid using a standard miniprep kit. In the case of a
larger number of yeast colonies, it is better to use a microtiter
plate scaled plasmid isolation method. Pick four colonies per
transformation in a microtiter plate with 100 μl selective LB

and incubate for 2 h at 37 C without shaking.
3. First inoculate 15 μl of the cells in a microtiter plate containing
150 μl of selective LB. Close the lid and seal the plate with
parafilm around the borders. Grow overnight at 37 C on a
microtiter plate (MTP) shaker. To the remaining culture, add
100 μl of LB with 40 % glycerol. This backup glycerol cell stock
can be stored at 20 C.
4. The next day, collect the cells by centrifugation at 4,000 g
for 10 min. Discard the supernatant (see Note 24).
5. Add 35 μl of resuspension buffer, shake for approximately
5 min on an MTP shaker until everything is resuspended.
6. Add 35 μl of lysis buffer, shake for about 2 min until the
solution is clear again. Do not allow lysis to proceed for longer
than 5 min.
7. Add 49 μl of neutralization buffer, shake for another 5 min.
A white precipitate of cell debris and genomic DNA will form.
8. Centrifuge for 20 min at 4,000 g. Transfer the supernatant
containing the plasmid DNA to a new round-bottomed micro-
titer plate (approximately 110 μl).
9. Add 60 μl of isopropanol and mix by pipetting up and down.
Incubate for 15 min without shaking at room temperature.
10. Centrifuge 30 min at 4,000 g to precipitate the plasmid
DNA. Remove the supernatant by quickly inverting the plate
above a sink.
11. Add 70 μl of 70 % ice cold ethanol to wash the pellet. Centri-
fuge 10 min at 4,000 g at 4 C. Discard supernatant by
quickly inverting the plate above the sink and tapping the
plate on paper tissues (see Note 25). Let the DNA dry for
about 15 min at room temperature.
12. Resuspend the DNA pellets in 20 μl of ultrapure water by
pipetting up and down.
13. Digest the plasmid DNA with a restriction enzyme cutting out
the insert (for miniprep on a column: use 2 μl of plasmid DNA
as template; for miniprep in MTP format: use 6 μl of plasmid
DNA as template) and visualize using DNA agarose gel
electrophoresis.
14. Evaluation: this digest checks whether all prey plasmids origi-
nated from one blue yeast cell contain the same insert (based on
length alone). As yeast can contain more than one prey plasmid,
in contrast to E. coli that can only harbor one, it is possible to
see different insert lengths. If this is the case, both types of
inserts have to be analyzed. One construct per insert length
should be chosen for DNA sequencing analysis using prey

plasmid-specific primers to identify the potential interaction
partners for the protein of interest.
3.7 One-on-One Y2H All the observed interactions could be false positive results and
Confirmation Test should be confirmed using an independent Y2H experiment.
Choose one prey construct (the plasmid isolated from E. coli in
Subheading 3.6) for each different potential interaction partner.
These prey constructs will be transformed to the α-mating type
yeast strain Y187 [18] and will be mated with an AH109 strain
containing the bait construct and another strain containing an
empty bait vector as a negative control. An independent protein
inserted in the bait vector could also be used. If there are not too
many different prey plasmids, the same protocol provided in Sub-
heading 3.1 can easily be followed to transform the plasmids into
S. cerevisiae Y187. After transformation, one can proceed with
step 15 from this protocol. In the case of a higher number of
different preys or different Y2H screens confirmation tests com-
bined in one experiment, the protocol given in this section, which is
adapted for up to 96 yeast transformations in parallel, can be
followed.
1. Streak S. cerevisiae Y187 on a fresh YPDA agar plate starting
from a 25 % glycerol cell stock (stored at 80 C). Incubate for
2–3 nights at 30 C.
2. Inoculate two tubes of 4 ml of YPDA with several small colo-
nies. Shake overnight at 30 C.
3. Dilute the overnight culture with an OD600 value below 1 in
100 ml of YPDA (in a 1 L baffled flask) until an OD600 of
0.120–0.140. Incubate at 30 C until the OD600 reaches
0.480–0.560.
4. Divide the culture between two conical tubes and centrifuge
(3,500 g, 5 min).
5. Discard the medium and resuspend the cells in 2 25 ml
sterile water. Centrifuge again.
6. Discard the supernatant and resuspend the cells in 2 1 ml of
100 mM LiOAc. Transfer to 1.5 ml microcentrifuge tubes.
7. Pellet the cells (14,000 g, 15 s) and remove the LiOAc with a
micropipette. Resuspend each cell pellet in 400 μl of 100 mM
LiOAc. The final volume is now around 2 500 μl. This is
enough for 96 small-scale transformations. Centrifuge
(14,000 g, 15 s) and remove the LiOAc.
8. Resuspend the two pellets in 1,480 μl of sterile water and
transfer to a 15 ml conical tube. The transformation mixture
is now added in the following order to the cells:
4,800 μl PEG3350 (50 % w/v)
720 μl 1 M LiOAc
200 μl ssDNA (10 mg/ml)
9. Vortex for 10 min. Prepare a PCR plate by adding to each well
100 ng of plasmid DNA (a different plasmid per well).
10. Divide the transformation mixture in the PCR plate by adding
25 μl to each well. Pipette up and down a few times to mix.
Place a lid on the plate during incubation (see Note 26).
11. Incubate for 10 min at 30 C followed by 30 min at 42 C to
heat shock the cells. This step can be done on a PCR block.
12. Centrifuge the plate at 3,500 g for 1 min and discard the
supernatant by pipetting. Resuspend the cells in 100 μl of
sterile water.
13. Plate the entire cell mixture on selective SD medium and
incubate at 30 C for 3 nights (see Note 27).
14. On the same day as step 13, freshly streak the different AH109
strains on selective agar plates. Incubate for 3 nights at 30 C.
15. Fill a microtiter plate with 50 μl of sterile water and resuspend
the different S. cerevisiae Y187 and AH109 strains in different
wells (pick several colonies).
16. Mating: spot 2 μl a-yeast on a YPDA agar plate. Let dry. Spot
2 μl α-yeast on the same position as the a-yeast. Do this for all
different combinations (e.g., each prey in S. cerevisiae Y187 is
mated with S. cerevisiae AH109 containing the bait construct
and S. cerevisiae AH109 containing the empty bait vector).
There will be two positions for each prey which has to be
tested. Incubate overnight at 30 C.
17. The next day, mating has occurred. To select for diploid cells,
pick some cell material from every position in 50 μl of sterile
water and spot the different wells (2 μl) on SD-Trp-Leu. Incu-
bate for two nights at 30 C.
18. Resuspend the diploid cells again in 50 μl of sterile water and
spot 2 μl on SD-Trp-Leu-His + x mM 3-AT. Incubate for 5–7
nights at 30 C.
19. Pick the different cells in 50 μl of sterile water and spot 2 μl on
SD-Trp-Leu-His-Ade +40 mg/L X-α-gal. Incubate for 5–7
nights at 30 C.
20. Evaluation: those prey plasmids that only generate colonies
with a halo for the bait-prey combination but not when com-
bined with the empty bait vector are confirmed interactions.
3.8 Quantitative To compare different interaction strengths, a quantitative assay to

α-Galactosidase Assay measure the α-galactosidase activity (encoded by the MEL1
reporter gene) should be performed. The stronger the interaction
between bait and prey, the more transcription of MEL1, and conse-
quently the more α-galactosidase is produced. This enzyme is
secreted in the medium, so there is no need to lyse the yeast cells in

this protocol (adapted from ref. 23).
1. Inoculate 3 ml of liquid SD-Trp-Leu-His medium with a bait-
and prey-expressing yeast cell (see Note 28). Do this in tripli-
cate for every type of yeast colony that needs to be analyzed.
Incubate overnight at 30 C in a shaking incubator.
2. Vortex each culture for 30 s to disperse cell clumps, then
transfer 1 ml to a cuvette to measure the OD600 (should be
between 0.5 and 1; if higher, dilute the cell suspension).
3. Transfer 1 ml of the vortexed cell suspension to a 1.5 ml
microcentrifuge tube. Centrifuge (14,000 g, 2 min). Imme-
diately proceed to step 4 to reduce loss of enzyme activity.
4. Prepare enough assay buffer for all samples and controls, and
let it equilibrate to room temperature. For each assay 48 μl of
buffer is needed (see Note 29).
5. Transfer 16 μl of the supernatant (of step 3) into a well of a
clear 96-well flat-bottom microtiter plate. Also include one
well containing sterile SD-Trp-Leu-His. This blank will be
used to zero the spectrophotometer in step 8.
6. Add 48 μl of assay buffer to each well and close the plate.
Incubate at 30 C for 1 h.
7. Stop the reaction by adding 136 μl of 10 stop solution to each
well.
8. Measure the OD at 410 nm (OD410) relative to the blank
sample using a spectrophotometer that can read microtiter
plates.
9. Calculate the α-galactosidase units. One unit is defined as the
amount of enzyme that hydrolyzes 1 μmol of p-nitrophenyl-α-
D-galactoside to p-nitrophenol and D-galactose in 1 min at
30 C in acetate buffer, pH 4.5 [20].
In the following formula, the OD600 of the overnight culture is
used to normalize the OD410 to the amount of cells. Remember
to take into account the dilution factor from step 2, if necessary.
α‐galactosidase in milliunits per ml per cell
¼ ðOD410 V f 1, 000Þ=ðε b t V i OD600 Þ
t ¼ elapsed time of incubation (in min) (60 min)
Vf ¼ final volume of the assay (200 μl)
Vi ¼ volume of supernatant added (16 μl)
ε ¼ p-nitrophenol molar absorptivity at 410 nm
b ¼ the light path (in cm)
ε b ¼ 10.5 ml/μmol for the microtiter plate format
4 Notes
1. Always use freshly streaked cells for your transformations. This

will give the best transformation efficiency. The plate used to
inoculate the overnight culture can be up to 2 weeks old for
routine transformations, but for the prey library transforma-
tion, it is definitely advisable to use fresh colonies.
2. The fresh culture in the morning should have 5 106 cells/ml
because at the start of transformation 2 107 cells/ml are
needed. For strain AH109, this start density corresponds to
an OD600 of 0.130. To reach this, first vortex the overnight
culture to disperse all cell clumps, add the cells in smaller steps,
and measure the OD600 in between until an OD600 of 0.130 is
reached. It is critical that all cell clumps are dispersed to obtain a
healthy liquid culture. If the culture does not reach the final
OD600 in the expected 4–5 h, it is best to start over with an
overnight culture and do the transformation the following day,
especially when the efficiency should be high (e.g., for the
library transformation).
3. It is not required to prepare a separate 100 mM stock of LiOAc,
a 1 M stock suffices. In this case, first resuspend the cells in
sterile water and then add the required LiOAc. For instance, if
the cells need to be resuspended in 400 μl of 100 mM LiOAc,
first resuspend them in 360 μl of sterile water and then add
40 μl of 1 M LiOAc.
4. When less than ten transformations are needed, all washing
volumes can be adapted. Five milliliters of yeast culture is
needed for one transformation. For every 5 ml of exponentially
growing cells, the washing volumes are subsequently 2.5 ml of
water, 100 μl of 100 mM LiOAc, and 40 μl of 100 mM LiOAc.
5. Adding the PEG3350 before the LiOAc is important. LiOAc is
harmful for the cells if the concentration is too high. The PEG
will first make a shield around the cells. When transforming a
higher number of different plasmids, it can be useful to resus-
pend the cells in batches in the transformation mixture without
the plasmids and then divide the cells over different tubes each
containing a different plasmid. This does not affect the trans-
formation efficiency. For instance, for ten transformations, stop
in step 7 when the amount of cells is 500 μl. Centrifuge and
remove the LiOAc. Then resuspend the cells in 740 y μl
of sterile water (where y is the total amount of plasmid DNA)
and transfer to a small conical tube. Add 2.4 ml of PEG3350
and vortex. Add 360 μl of 1 M LiOAc and 100 μl of ssDNA and
vortex again for 2 min. Add 360 x μl of transformation
mixture to each tube with x μl of plasmid DNA.
6. The vortexing step is crucial for a good transformation

efficiency. Cells have to be fully resuspended and surrounded
by all the compounds to take up the plasmid DNA. Cells that
are not vortexed sufficiently will settle faster to the bottom of
the tube and the transformant yield will be lower.
7. Usually the 100 μl plate will have a sufficient number of single
colonies. Once you have become familiar with the protocol,
only plate 100 μl and keep the rest of the transformation
mixture at 4 C. Check the plate after 2 days and when more
colonies are needed, the remaining cells can be plated.
8. It is possible to directly use the colonies obtained in Subhead-
ing 3.1 if you directly proceed. Then there is no need to streak
them again before performing the auto-activation test.
9. Also use YPDA for this culture. Growth in SD medium select-
ing for the bait would take too long.
10. Most of the time, 1 or 3 mM 3-AT is sufficient.
11. For the library transformation, one should always use fresh cells
to obtain a high transformant yield.
12. One overnight culture would not give enough cells to inoculate
600 ml of culture, which is needed for the library transforma-
tion. Hence, two steps are needed. Inoculation of the second
overnight culture (100 ml culture) early in the afternoon
(2–3 PM) is usually successful to get an OD close to 1 the
next morning.
13. First add 2 ml of water per pellet to resuspend them by
pipetting up and down. Then adjust to the final volume by
pouring.
14. The transformation mixture can be made and vortexed during
the washing steps. As the PEG3350 is viscous, pipette it slowly
to obtain the correct volume.
15. Vortexing for 10 min can easily be done by taping the tubes on
the vortex machine.
16. After dividing the mixture in the small conical tubes, make sure
the level is below the water level. If this is not the case, divide
over more tubes.
17. It is recommended to prepare the selective plates only 2–3 days
before your planned experiment, especially for the library trans-
formations. In that case, they can be kept at room temperature,
so the plates will be dried, allowing the transformation mix-
tures to dry much faster after plating.
18. The number of transformants needed depends on the library.
For random genomic libraries, this can be calculated using the
Clarke and Carbon formula [21] that indicates that the actual
number of clones required equals ln(1 P)/ln(1 (f/G))
where P is the probability, f is the mean fragment size of the

plasmid inserts, and G is the genome size (e.g., for E. coli K-12
[22] with a genome size of 4,639,221 bp and a mean fragment
size of 600 bp, 35,605 transformants are needed to cover the
whole genome once with a probability of 0.99). Knowing that
in Y2H fusion proteins the fragment orientation and reading
frame has to be correct, the actual total number of yeast trans-
formants needed is 6 35605 or 2.14 105.
19. The total number of colonies growing on the SD-Trp-Leu-His
plates depends on the bait, the 3-AT concentration, and the
efficiency. The higher the transformation efficiency, the more
colonies will be growing. In this case the library is covered
multiple times, so there will be colonies that contain the same
prey plasmid. Because it is too labor intensive to analyze all of
the colonies in this case, it is important to make a selection.
Limit the selection to 96 colonies, which is still an easy format
to work with. Careful inspection of all plates and numbering
and sorting the colonies in groups of different sizes is one way
to remove duplicates (e.g., 1, 2, 3, >3 mm). Although size
might reflect interaction strength, there is no reason to assume
that weak interactions have less biological significance, so
always select different sizes of colonies from all different plates.
When picking the colonies to array them in a microtiter plate,
check the consistency of the colony. Based on our experience,
granular colonies that fall apart after picking generally contain
false positive interactions. So it is preferable to only pick
smooth and creamy colonies.
20. Other single well microtiter plates or a small petri dish in the
case of a smaller number of colonies can be used.
21. Limit this selection to 24 colonies again by carefully inspecting
and sorting the colonies in groups of different colors (e.g.,
white, pale blue, dark blue, dark blue with a blue halo).
22. It is easier to use glass culture tubes than flasks. Yet, when using
glass tubes, the cells will settle more easily to the bottom of the
tube. To grow a culture sufficiently dense for plasmid yield,
vortex after inoculation, shake the tubes vigorously (250 rpm),
and vortex them again after 1 day of incubation.
23. Different E. coli colonies derived from the same yeast colony
are selected because yeast can contain more than one prey
plasmid. As E. coli can only contain one, this step will show
the different inserts.
24. Supernatant removal can be performed using a vacuum pump.
The microtiter plate method to recover plasmids requires some
practice but is an inexpensive and efficient solution when
screening large numbers of colonies.
25. To remove the supernatant, invert the plate very quickly to

avoid well-to-well contamination.
26. Use a microtiter plate lid on top of the PCR plate. When using
strips to close the plate, it is too difficult to remove the strips
after incubation which will cause well-to-well contamination.
27. To reduce the amount of medium and to make the plating
more convenient, QTray bioassay trays from Genetix can be
used for plating. These contain a 48-position grid, so only two
plates are needed to plate 96 transformation mixtures.
28. Inoculate the cells starting from an SD-Trp-Leu or SD-Trp-
Leu-His plate. Use fresh cells (maximum 2 weeks old).
29. When the number of samples is smaller or a spectrophotometer
to read microtiter plates is not available, a 1 ml assay can also be
performed. Refer to [23]) for a detailed protocol.
Acknowledgements
Jeroen Wagemans holds a predoctoral fellowship of the

“Agentschap voor Innovatie door Wetenschap en Technologie in
Vlaanderen” (IWT, Belgium).
References
1. Fields S, Song OK (1989) A novel genetic 8. Buchert M, Schneider S, Adams MT et al
system to detect protein protein interactions. (1997) Useful vectors for the two-hybrid sys-
Nature 340:245–246 tem in mammalian cells. Biotechniques
2. Keegan L, Gill G, Ptashne M (1986) Separa- 23:396–398, 400, 402
tion of DNA binding from the transcription- 9. Johnsson N, Varshavsky A (1994) Split ubiqui-
activating function of a eukaryotic regulatory tin as a sensor of protein interactions in vivo.
protein. Science 231:699–704 Proc Natl Acad Sci U S A 91:10340–10344
3. Chien CT, Bartel PL, Sternglanz R et al (1991) 10. Brachmann RK, Boeke JD (1997) Tag games
The two-hybrid system: a method to identify in yeast: the two-hybrid system and beyond.
and clone genes for proteins that interact with a Curr Opin Biotechnol 8:561–568
protein of interest. Proc Natl Acad Sci U S A 11. Uetz P (2002) Two-hybrid arrays. Curr Opin
88:9578–9582 Chem Biol 6:57–62
4. Van Criekinge W, Beyaert R (1999) Yeast two- 12. Golemis EA, Khazak V (1997) Alternative
hybrid: state of the art. Biol Proced Online yeast two-hybrid systems. The interaction trap
2:1–38 and interaction mating. Methods Mol Biol
5. James P, Halladay J, Craig EA (1996) Genomic 63:197–218
libraries and a host strain designed for highly 13. Serebriiskii I, Khazak V, Golemis EA (1999) A
efficient two-hybrid selection in yeast. Genetics two-hybrid dual bait system to discriminate
144:1425–1436 specificity of protein interactions. J Biol Chem
6. Durfee T, Becherer K, Chen PL et al (1993) 274:17080–17087
The retinoblastoma protein associates with the 14. Le Douarin B, Pierrat B, vom Baur E et al
protein phosphatase type 1 catalytic subunit. (1995) A new version of the two-hybrid assay
Genes Dev 7:555–569 for detection of protein-protein interactions.
7. Hu JC, Kornacker MG, Hochschild A (2000) Nucleic Acids Res 23:876–878
Escherichia coli one- and two-hybrid systems 15. Caufield JH, Sakhawalkar N, Uetz P (2012) A
for the analysis and identification of protein- comparison and optimization of yeast two-
protein interactions. Methods 20:80–94 hybrid systems. Methods 58:317–324
16. Chen YC, Rajagopala SV, Stellberger T et al 20. Lazo PS, Ochoa AG, Gascon S (1978) alpha-
(2010) Exhaustive benchmarking of the yeast Galactosidase (melibiase) from Saccharomyces
two-hybrid system. Nat Methods 7:667–668, carlsbergensis: structural and kinetic properties.
author reply 668 Arch Biochem Biophys 191:316–324
17. Uetz P, Dong YA, Zeretzke C et al (2006) 21. Clarke L, Carbon J (1976) A colony bank con-
Herpesviral protein networks and their interac- taining synthetic Col El hybrid plasmids repre-
tion with the human proteome. Science sentative of the entire E. coli genome. Cell
311:239–242 9:91–99
18. Harper JW, Adami GR, Wei N et al (1993) The 22. Blattner FR, Plunkett G 3rd, Bloch CA et al
p21 Cdk-interacting protein Cip1 is a potent (1997) The complete genome sequence of
inhibitor of G1 cyclin-dependent kinases. Cell Escherichia coli K-12. Science 277:1453–1462
75:805–816 23. Clontech Yeast Protocols Handbook. Clontech
19. Gietz RD, Woods RA (2002) Transformation Laboratories, Inc. www.clontech.com. Proto-
of yeast by lithium acetate/single-stranded car- col No. PT3024-1. 2. Version No. PR973283
rier DNA/polyethylene glycol method. Meth-
ods Enzymol 350:87–96
Chapter 28
Reverse Two-Hybrid Techniques in the Yeast

Saccharomyces cerevisiae
Matthew A. Bennett, Jack F. Shern, and Richard A. Kahn
Abstract
Use of the yeast two-hybrid system has provided definition to many previously uncharacterized pathways
through the identification and characterization of novel protein-protein interactions. The two-hybrid
system uses the bifunctional nature of transcription factors, such as the yeast enhancer Gal4, to allow
protein-protein interactions to be monitored through changes in transcription of reporter genes. Once a
positive interaction has been identified, either of the interacting proteins can be mutated by site-specific or
randomly introduced changes, to produce proteins with a decreased ability to interact. Mutants generated
using this strategy are very powerful reagents in tests of the biological significance of the interaction and in
defining the residues involved in the interaction. Such techniques are termed reverse two-hybrid methods.
We describe a reverse two-hybrid method that generates loss-of-interaction mutations of the catalytic
subunit of the Escherichia coli heat-labile toxin (LTA1) with decreased binding to the active
(GTP-bound) form of human ARF3, its protein cofactor. While newer methods are emerging for
performing interaction screens in mammalian cells, instead of yeast, the use of reverse two-hybrid in yeast
remains a robust and powerful means of identifying loss-of-interaction point mutants and compensating
changes that remain among the most powerful tools of testing the biological significance of a protein-
protein interaction.
Key words Reverse two-hybrid, Two-hybrid, Protein interaction, Loss-of-interaction mutation
1 Introduction
Protein interactions define key biochemical and regulatory

networks in the cell. The identification and characterization of
protein-protein interactions through the use of two-hybrid assays
have provided definition to many previously uncharacterized
pathways. Yeast two-hybrid systems use the bifunctional nature of
transcription factors, e.g., Gal4 or LexA, to assay protein interac-
tions in live cells [1, 2]. The DNA binding and transcriptional
activation domains of the transcription factor are each expressed
as fusion proteins and promote transcription only when joined by
the binding of the other portion of the fusion proteins (see Fig. 1
and Note 1). Screens of two-hybrid libraries have led to the
433
434 Matthew A. Bennett et al.
Fig. 1 Expression of reporter genes (e.g., LacZ) in two-hybrid systems is dependent upon the interaction of two
fusion proteins that can be lost as a result of a point mutation. The Gal4 protein, two interacting proteins of
interest, and resulting fusion proteins are shown pictorially on the left. An “x” in a protein represents a point
mutation. Interaction between the two fusion proteins generates a transactivator capable of promoting
transcription of reporter genes (LacZ) downstream of the Gal4 upstream activating sequence (UAS). Point
mutations in either fusion protein that decrease binding result in decreases in or elimination of reporter gene
expression. The level of expression is shown on the right as the amount of β-galactosidase (β-gal) activity
identification of a large number of protein binding partners

(e.g., refs. 3, 4) and proteome based screens (e.g., refs. 5–8) have
increased the number of putative interactions markedly. Both the
methods and the level of complexity or detail in the interactomes
continue to improve [7, 9–17].
After a two-hybrid interaction is identified, either of the inter-
acting proteins can be mutated to decrease the binding affinity. The
generation and use of such mutants is collectively referred to as
“reverse two-hybrid” methods. Such loss-of-interaction mutants
can become essential reagents to test the functional importance of
specific protein interactions in live cells, to obtain a low-resolution
map of the binding interface, or to generate pairs of mutants that
interact independently of endogenous proteins (see also Chappell
and Gray [18]). A number of variants and uses for reverse two-
Reverse Two-Hybrid 435
hybrid have emerged over the years that offer greater versatility,
depending on needs [19–21]. These approaches are particularly
attractive when one of the proteins under study binds multiple
protein partners as mutagenesis and two-hybrid screens can identify
specific loss of binding mutants that retain binding to other part-
ners. In this chapter, we describe a reverse two-hybrid method for
generating mutations of the catalytic subunit of the Escherichia coli
heat-labile toxin (LTA1) with reduced binding to the activated
(GTP bound) form of human ARF3, [Q71L]ARF3 [22, 23].
As seen here and in other studies (e.g., Das et al. [24]), the use of
reverse two-hybrid is an attractive means of testing the relevance of
protein interactions in pathogen biology.
The main features of this approach include PCR-based random
mutagenesis over a defined region of any coding region, and iden-
tification of full-length mutants, expressed to the same levels as
wild-type protein by immunoblot analysis using an epitope tag on
the fusion proteins. The conditions used here have yielded approxi-
mately one change per 250 bp amplified. PCR allows the extent of
mutagenesis to be controlled by modifying the concentration of
nucleotides or manganese in the PCR reaction [25]. Although the
use of selectable markers fused to the C-terminus of the mutated
protein allows for genetic selection of full-length proteins [26–28],
which is important to high-throughput screens and can speed up
this rate limiting step in reverse two-hybrid screens, we prefer the
use of immunoblots to allow estimates of protein expression as well
as confirmation that the protein is full-length. Using the methods
described below we have been able to generate hundreds of colo-
nies with reduced β-galactosidase activity in less than a week and
typically 25–50 % of those were found to result from point muta-
tions in full-length proteins, suitable for further analysis. Thus, tens
of mutants can be obtained, their interactions with multiple effec-
tors characterized, and sequences determined in less than 1 month.
The loss of an interaction is limited in interpretations due to the
negative nature of the data. However, loss of one binding partner
with retention of others provides a powerful probe of activities in
live cells. In addition, after the generation of loss-of-interaction
mutants in one protein it is even easier to then mutate the other
binding partner and screen for mutants that regain the interaction.
Such pairs of mutants can provide a formal proof of specific func-
tionalities in live cells. Such mutants may have the added benefit of
not binding the wild-type protein and can allow cell studies to be
carried out independently of interference or complications pre-
sented by the presence of endogenous proteins. As with any tech-
nique, the data and conclusions from the use of reverse two-hybrid
techniques are strengthened by confirmation using independent
methods.
2 Materials
1. pACT2 vector [29] (Genbank accession #U29899), a 2 μ

plasmid carrying the selectable LEU2 gene and the ADH1
promoter driving expression of the activation domain (AD) of
Gal4p (Gal4 Region II [1], residues 768–881), followed by the
HA epitope, sites for insertion of the open reading frame of
your protein, and the ADH1 terminator.
2. pBG4D vector [30], a 2 μ plasmid carrying the selectable TRP3
marker and the Gal4 DNA Binding Domain (BD) followed by
the HA epitope, designed to add the Gal4-BD to the 30 end of
any inserted open reading frame, with expression of the result-
ing fusion protein driven by the ADH1 promoter. pBG4D was
made by Robert M. Brasas by subcloning the SacI/HindIII
fragment from the pAS1-CYH plasmid into the plasmid D133.
3. Yeast strain Y190 (MAT a gal4 gal80 his3 trp1-901 ade2-101
ura3-52 leu2-3,112 + URA3::GAL!lacZ, LYS2::GAL(UAS)-
HIS3 cyhR) [29] kindly provided by Steve Elledge.
4. Whatman filter paper no. 1.
5. Nitrocellulose filters (Schleicher & Schuell Protran, 82 mm).
6. Monoclonal HA antibody 12CA5 (Boehringer), diluted
1:1,000 (final ¼ 1 μg/mL) for immunoblots.
7. X-gal (5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside),
100 mg/ml stock solution in N,N-dimethylformamide
(DMF).
8. Low dATP nucleotide mix (2.5 mM dCTP, 2.5 mM dGTP,
2.5 mM dTTP, and 0.625 mM dATP in water).
9. Taq Polymerase.
10. Synthetic (SD) minimal, complete, and dropout media. Mini-
mal SD medium: 1.5 g yeast nitrogen base lacking amino acids,
20 g glucose, 5 g (NH4)2SO4 per liter. Complete SD medium:
20 mg adenine, 20 mg histidine, 20 mg methionine, 20 mg
tryptophan, 20 mg uracil, 30 mg leucine, 30 mg lysine per liter,
each is omitted as needed to generate the dropout media, e.g.,
SD leu trp.
11. SD Plates, made with the same components as SD media plus
20 g agar per liter.
12. SD leu/cycloheximide, SD leu containing 2.5 μg/mL
cycloheximide.
13. Agarose gel and DNA sequencing equipment.
14. SDS-PAGE (sodium dodecylsulfate polyacrylamide gel electro-
phoresis) and protein electrotransfer equipment for
immunoblotting.
15. 1 SDS sample buffer (50 mM Tris–HCl, pH 6.8, 10 %

glycerol, 1 % SDS, 1 % (v:v) β-mercaptoethanol, bromophenol
blue).
16. Z-Buffer (60 mM Na2HPO4, 40 mM NaH2PO4, 10 mM KCl,
2 mM MgSO4, pH 7.0).
17. Oligonucleotide primers for PCR mutagenesis:
#827 (sense primer): GCGTATAACGCGTTTGGAATC,
binds 70 bp 50 of the pACT2 MCS.
#828 (anti-sense primer): GAGATGGTGCACGATGCACAG,
binds 70 bp 30 of the pACT2 MCS.
18. Sequencing Oligonucleotides:
#1042 (sense primer): GCTTACCCATACGATGTTCCA—
binds within the pACT2 MCS at the 50 end.
#1043 (anti-sense primer): TGAACTTGCGGGGTTTTT-
CAG—binds within the pACT2 MCS at the 30 end.
19. DNA purification kits (e.g., Qiagen mini prep, QiaQuick gel
extraction, Wizard PCR purification kit).
20. pAB157, a plasmid encoding [Q71L]ARF3 with the Gal4-BD
at the C-terminus by insertion into the BamHI site of pBG4D,
to direct expression of [Q71L]ARF3-BD.
21. pXJ12, a plasmid carrying the open reading frame of the cata-
lytic (A1) subunit of the E. coli heat labile toxin (LTA1) cloned
into the NcoI and BamHI sites of pACT2, putting it in frame
with the Gal4-AD, to direct expression of AD-LTA1 (see Fig. 2).
22. YAB457, the yeast strain, derived from Y190, carrying the
pAB157 plasmid and thus expressing [Q71L]ARF3-BD.
23. SOS: 3.33 g/L yeast extract, 6.66 g/L dextrose, 6.66 g/L
peptone, 6.5 mM CaCl2.
24. TE: 10 mM Tris–HCl, pH 8.0 and 1 mM EDTA, pH 8.0.
25. 0.1 M lithium acetate in TE buffer.
26. LiOAc/PEG solution: 40 % (w:v) PEG 3350 suspended in
0.1 M lithium acetate in TE buffer.
27. Sonicated herring sperm DNA, 10 mg/mL.
28. STES: 500 mM NaCl, 200 mM Tris–HCl, pH 7.6, 10 mM
EDTA, 1 % SDS.
29. Phenol–chloroform: 50 % phenol, 50 % chloroform.
30. Chloroform
31. Polymerase chain reaction (PCR) buffer (1 ): 10 mM
Tris–HCl, pH 8.3, 50 mM KCl, 0.0001 % gelatin, which is
made from a 10 solution.
Fig. 2 Reverse two-hybrid screens use random PCR mutagenesis and gap repair
in yeast to generate a large library of mutants, ready for screening. Amplification
of the coding region of LTA1 in pXJ12 is performed with primers that anneal
50 bp outside of the region targeted for mutagenesis to yield a product with
ends homologous to those in the gapped plasmid, shown below. The error rate of
the polymerase is increased by carrying out the PCR under conditions of reduced
stringency. Note that any region of DNA can be targeted for mutagenesis. Co-
transformation of the mutated PCR product with gapped pACT2 plasmid,
prepared by restriction digestion, into yeast allows for repair of the plasmid by
homologous recombination. When transformed into a yeast strain expressing the
[Q71L]ARF3-BD fusion proteins, the resulting transformants can be assayed
directly for loss of interaction using the colony β-galactosidase assay
3 Methods
The methods listed in this section describe: (1) PCR-introduced,

random mutagenesis of LTA1, (2) co-transfection of mutated
LTA1 and gapped pACT2 plasmids into YAB457 to yield yeast
colonies expressing mutant LTA-AD and [Q71L]ARF3 fusion
proteins, (3) assaying for loss of interaction via a colony β-galacto-
sidase assay, (4) rescuing the plasmids expressing mutant LTA1, and
(5) characterizing the loss-of-interaction LTA1 mutants. These
methods begin after a two-hybrid interaction has been identified;
e.g., between the protein products of the plasmids pXJ12 (AD-
LTA1) and pAB157 ([Q71L]ARF3-BD).
3.1 Generation This method exploits the lack of proofreading by Taq polymerase
of Mutant Substrates and ability to replicate DNA with low fidelity and generate random
for Plasmids changes into any region of DNA, in this case the LTA1-encoding
portion of the AD-LTA1 plasmid (see Fig. 2 and Note 2).
3.1.1 Mutagenesis Mutagenesis of LTA1 is accomplished through PCR amplification

of LTA1 of pXJ12 with primers #827 and #828 under conditions that
promote polymerase-induced errors (see Fig. 2). These error-
prone conditions include the addition of MnCl2 and lowering the
concentration of dATP, relative to other nucleotides, as described
previously [31, 32]. The 50 μL PCR reaction consists of the
following: 1.5 mM MgCl2, 0.05 mM MnCl2, 0.25 mM each of
dCTP, dGTP, and TTP, 0.0625 mM dATP, 2 μM Primer #827,
2 μM Primer #828, 1 μL Taq Polymerase (0.05 U/μL), 25 ng
pXJ12 in 1 PCR Buffer.
PCR amplification/mutagenesis is performed in 25 cycles
using an annealing temperature of 55 C (15 s) and an extension
temperature of 72 C (1 min for products <1 kb), with a denatur-
ation step (94 C, 15 s) before each cycle. Purify the product using
a commercially available PCR cleanup kit according to manufac-
turer’s directions. The resulting purified PCR product will contain
randomly introduced mutations in the LTA1 coding region and
approximately 80 bp of flanking pACT2-derived DNA upstream
and downstream of the open reading frame. The yield from PCR
under these conditions will be lower than normal but should
produce plenty of product for later steps.
3.1.2 Gapped Plasmid Create the gapped plasmid by digesting pXJ12 with NcoI and
(pACT2) Production XhoI (see Note 3). Purify the resulting linearized, 8 kb vector back-
bone from a 1 % agarose gel using a commercially available gel
extraction kit.
3.1.3 Estimating Vector Estimate the concentration of the purified, gapped (pACT2-
and PCR Product derived) plasmid and PCR products by comparing the ethidium
Concentration bromide fluorescence intensity from each sample in a 1 % agarose
gel to that of a known amount of a λ HindIII DNA marker of
similar size.
3.2 Co- The insertion of the mutagenized open reading frame into the
transformation of gapped plasmid can be accomplished by taking advantage of the
Gapped Plasmid and high level of homologous recombination in yeast cells.
PCR Product Co-transformation of yeast with gapped plasmid and PCR products
that contain regions of identity of 50 bp at each end is approxi-
mately 50 % as efficient as transformation with the circular plasmid
alone (see Fig. 2 and Note 4).
1. Inoculate 50 mL of SD trp with YAB457 and grow at 30 C
to OD600 ~ 0.6.
2. Harvest the cells by centrifugation at 1,000 g for 5 min.

3. Resuspend the cells in TE buffer and collect cells again by
centrifugation.
4. Wash the cells in 0.1 M lithium acetate in TE and collect by
centrifugation.
5. Resuspend the cells 0.5 mL of 0.1 M lithium acetate/TE.
6. Aliquot 100 μL of cell suspension into each of five sterile
microfuge tubes.
7. Add 5 μL of freshly boiled sonicated herring sperm DNA and
0.7 mL 40 % PEG/LiAc/TE to each tube (single stranded
carrier DNA improves transformation frequency).
8. Add the transforming DNA (gapped plasmid and PCR pro-
ducts) to each tube. Use equal molar amounts of gapped plas-
mid and PCR products. The amount of DNA will vary with the
experimentally determined transformation frequency, designed
to yield 400–600 transformants per 100 mm plate, but will
probably be in the range of 0.1–1 μg per transformation
9. Incubate each transformation reaction at 30 C for 30 min.
10. Transfer tubes to a 42 C water bath and heat shock for 15 min.
11. Add 600 μL of SOS to each transformation tube.
12. Plate an appropriate volume (100 μL) of the transformed cell
suspension on SD trp leu plates and incubate at 28 C until
small, distinct colonies became visible (approximately 3 days).
3.3 Assaying The X-gal filter assay is used to measure levels of β-galactosidase
Transformants for activity in yeast colonies as an indicator of protein interactions [33].
Loss-of-Interaction The two fusion proteins (often referred to as bait and prey) interact,
thereby activating transcription of the lacZ gene and production of
β-galactosidase, which catalyzes the hydrolysis of X-gal to yield a
blue product. Both the time of incubation and the darkness of the
blue product formed on the filter should be monitored and com-
pared to the parent plasmid to allow initial determination of loss-of-
interaction mutations.
1. Place a nitrocellulose filter on each plate for 30 s (or until wet)
to replica the yeast colonies onto the filter.
2. Drop the filter into liquid nitrogen and carefully remove after
20 s. Avoid breaking the filters that will be brittle while frozen.
3. Transfer the frozen nitrocellulose filter, with the yeast side
facing up, to the lid of a 100 mm dish on which has been placed
Whatman filter paper wetted with X-gal in Z buffer (1.5 mL of
Z-buffer with 15 μL of 100 mg/mL X-gal solution to yield a
final X-gal concentration of 1 mg/mL).
4. Cover with the empty dish to prevent dehydration and

incubate the filter at 30 C. Compare the blueness of colonies
that develop on the co-transformation plates with that of the
control plates at different times. The time will vary with the
bait–prey combination. Severe changes (white colonies) are
often later found to result from the introduction of premature
stop codons but can also produce a mutant with the most
dramatic difference in binding. It is a good idea to initially
select groups of colonies with small, intermediate, and large
differences in activity.
5. Streak colonies from the experimental plates, which turned less
blue or remained white in comparison to the (parental) positive
control plate, onto SD trp leu plates to select for retention
of both plasmids. After colonies appear, they may be
re-screened using histidine auxotrophy for selection. We have
found this to be a more sensitive assay than the X-gal filter assay.
3.4 Recovering Loss- Recover plasmids from yeast using a modification of the standard
of-Interaction Mutant glass bead method [34], also known as a “smash and grab” prepa-
Plasmids and ration. The modification simply adds the use of commercial plasmid
Retesting purification columns to further remove contaminants that inhibit
transformation. The use of plasmid purification columns is not
required or used in all laboratories but we have found it increases
success rates, sometimes markedly.
1. Streak colonies with diminished β-galactosidase activity onto SD
leu cycloheximide plates to select for loss of the [Q71L]ARF3-
BD expressing plasmid. This step is not required but often facil-
itates the isolation of the desired (AD-LTA1 mutant) plasmid
from strains carrying two different plasmids (see Note 6).
2. Incubate the plates for 3 days at 28 C.
3. Inoculate 20 mL of SD leu media with colonies from the SD
leu cycloheximide plates and grow each culture to an OD600
nm ~ 1.
4. Collect cells by centrifugation at 1,000 g for 10 min.
5. Wash each pellet in 1 mL of STES buffer and collect again.
6. Resuspend in 100 μL of STES and add an equal volume of glass
beads (see Note 7).
7. Vortex each tube for 2 min.
8. Add 100 μL of STES and 200 μL phenol–chloroform and
vortex for 5–30 min. Times will vary with the vortex used.
9. Pellet cells and debris and separate phases in a microfuge at
maximal speed (approximately 14,000 g) for 10 min, and
transfer the aqueous phase to a new tube.
10. Extract with 200 μL of phenol–chloroform, and then with

chloroform alone.
11. Further purify the plasmid in the aqueous phase using a com-
mercial mini-prep plasmid prep, according to manufacturer’s
directions. Then use the product to transform competent bac-
terial (DH5α) cells.
12. Purify plasmids from transformed DH5α using a commercial
plasmid mini-prep kit, according to manufacturer’s instruc-
tions. Check by restriction digestion analysis to confirm the
correct plasmid has been obtained.
13. Transform purified plasmids into YAB457 and assay transfor-
mants for interaction with [Q71L]ARF3 using the X-gal filter
assay. The color developed at different times (15, 30, 60,
180 min) should be compared to that of the YAB457-derived
strain expressing wild-type LTA1. Score activities visually on a
three plus scale, with three plus equal to that of wild-type LTA1
(see Note 8).
3.5 Characterizing Multiple factors can be responsible for a loss of interaction, some of
the Noninteracting which are far less informative than a point mutation in a critical
Proteins amino acid residue. Two types of uninteresting changes, premature
stop codons and decreased levels of protein expression, can be
screened for by immunoblotting cell lysates with an HA antibody
to determine the presence and relative abundance of the epitope tag
in the LTA1-AD fusion proteins. The levels of protein expressed in
each strain expressing mutant LTA1 should be compared to wild-
type LTA1 (see Note 9). Protein preparations from total yeast cell
lysates can be obtained using a modification of Horvath et al. [35].
1. Grow five OD of each yeast strain in selective liquid medium.
2. Collect cells by centrifugation at 1,000 g for 5 min.
3. Wash cells with water and collect again at 1,000 g for 5 min.
4. Resuspend each pellet in 20 μL of 1 SDS sample buffer and
boil at 95 C for 5 min.
5. Add one cell volume of acid-washed glass beads and vortex
3 min at maximum speed.
6. Add 120 μL sample buffer and boil for 5 min.
7. Pellet cells and debris by centrifugation in a microfuge for
15 min at maximal setting. Transfer supernatant to another
tube and boil for 5 min.
8. Load 7.5 μL of the sample on a 12 % SDS gel and resolve
proteins using standard methods.
9. Transfer the resolved proteins to nitrocellulose at 60 V for 2 h
and immunoblot with the 12CA5 (HA) antibody using stan-
dard methods.
If the mutant protein library was transfected into an

ARF3-expressing strain, then you should detect two bands in
immunoblots, at sizes corresponding to the AD-LTA1 and ARF3-
BD fusion proteins. Y190-derived strains also express a protein of
approximately 50 kDa that is bound by the 12CA5 antibody Y190
should be used as a control, at least initially. A shift to a smaller size
or the absence of (or clearly reduced) immunoreactivity is sugges-
tive of premature termination or alteration in the level of expres-
sion, respectively. Plasmids encoding full-length LTA1 mutants
that were expressed at approximately the same level as wild type
LTA1 should be sequenced with pACT2 sequencing primers
#1259 and #1260 to determine any changes [23].
4 Notes
1. The two-hybrid system described here was that developed in

the Elledge lab and uses a Gal4-based reporter system [29].
Y190 is a strain of S. cerevisiae engineered to express the HIS3
and lacZ gene products as reporters, in response to the pres-
ence of a functional Gal4 transactivator. Y190 is able to detect
protein-protein interactions by making two Gal4 hybrid pro-
teins involving two different domains of Gal4: one protein is
fused with the Gal4 Activation Domain to generate a Gal4-AD
fusion protein, while the other protein is fused with the Gal4
DNA Binding Domain to yield a Gal4-BD fusion protein.
If, when expressed in Y190, the two fusion proteins fail to
interact, the yeast fail to activate transcription at the GAL4
promoter and hence of the two reporters. However, if the
two proteins do interact, a functional Gal4 transactivator is
created. Thus, interaction can be observed by selecting and
assaying for histidine prototrophy and β-galactosidase activity.
Other two-hybrid systems have been described: that use
other Gal4-based plasmids [36], that use more than two repor-
ters [37], and that use other transcriptional activators (e.g.,
LexA; for reviews see refs. 38, 39). With only slight alterations,
the steps outlined in this chapter can be applied to other such
systems to perform a reverse two-hybrid screen.
2. The lack of proofreading by Taq polymerase facilitates the
introduction of mutations, most often A-T and A-G mutations
[25, 40]. Alternate DNA polymerases (e.g., Stratagene’s Muta-
zyme) designed to introduce a broader variety of mutations can
also be used.
3. Gapped plasmid was generated from pXJ12 (as opposed to
empty pACT2) to ensure that any uncut plasmid present in
the gapped plasmid isolation would encode a protein that
would retain interaction with [Q71L]ARF3 and remain blue
in the X-gal filter lift assay.
4. Transformation efficiency was determined by transforming

YAB457 with undigested pXJ12. We estimated the efficiency
for gapped plasmid repair in the co-transformation reaction to
be half that of intact plasmid when transforming with an equi-
molar amount of gapped plasmid and PCR product. The vol-
ume of cell suspension resulting from the co-transformation
reaction plated on each SD trp leu plate was adjusted to
yield between 400 and 600 transformants per 100 mm plate.
5. Up to four control transformation reactions can be performed;
including transformation of (1) uncut pXJ12 plasmid to deter-
mine general transformation efficiency, (2) no DNA, (3) the
gapped plasmid alone or (4) the PCR product alone provide
estimates of the background levels of transformations that will
appear on the experimental co-transformation plates. Although
not required, these controls are strongly encouraged the first
time reverse two-hybrid techniques are used in a lab.
6. pAS1-CYH and pBG4D both contain the CyhS gene that
causes sensitivity to cycloheximide. By growing cells containing
pACT2 and pBG4D-based plasmids on cycloheximide-
containing medium, cells not expressing pBG4D-based plas-
mids are selected and allow for easier isolation of pACT2-LTA1
plasmids.
7. This and the next step are critical and are the ones most likely to
cause low recovery of plasmids. Too few beads and the cells
simply rotate in the tube and too many beads and a slurry forms
that also decreases shearing forces that produce cell lysis.
8. As an alternative to the filter lift assay, a liquid culture β-galac-
tosidase activity assay can be used to generate a more quantifi-
able measure of activity [41]. Briefly, duplicate strains of yeast
can be grown in 5 ml of selective medium overnight at 30 C.
The next day, 5 ml of fresh selective medium is added to
20–50 μL of the overnight culture and grown to mid log
phase (OD600 ¼ 0.3–0.7). Cells are collected by centrifugation
and resuspended in 1 ml Z-Buffer and put on ice. Cells
(50–100 μL) are then mixed with Z buffer to a volume of
1 ml, and a drop of 0.1 % SDS and two drops of chloroform
are added to the sample with a Pasteur pipette. The mixtures
are vortexed for 15 s and incubated for 15 min at 30 C.
O-nitrophenol α-D-galactopyranoside (ONPG, 0.8 mg) is
added, the solution vortexed for 5 s, and incubated at 30 C
until a medium yellow color develops in the positive controls.
The reaction is stopped by adding 0.5 ml of 1 M Na2CO3, and
the time is recorded. OD420 and OD550 are determined after
the cell debris is removed by centrifugation. Units of activity
are determined using the equation:
U ¼ 1, 000ðOD420 OD550 Þ=ðT V OD600 Þ;

where V is the volume of the culture used in the assay (μL), T is

the time of the reaction (min), OD600 is the cell density at the
start of the assay, OD420 represents the combination of absor-
bance by ONPG and the light scattering by cell debris, and
OD550 represents the light scattering by cell debris.
9. In addition to providing the Gal4 activation domain, pACT2
also adds an HA-epitope after the Gal4-AD and before the
protein of interest. Other plasmids encoding the Gal4 DNA
binding domain, such as pBG4D or pAS1-CYH, also include
the HA tag to their fusion protein products and therefore will
also appear when blotting with an HA antibody. When com-
paring levels of protein expression by immunoblotting, it is
necessary to determine the protein concentration of each sam-
ple and load equivalent protein onto the gel as differences in
cell lysis and protein recovery can occur. Note that not all
proteins are expressed in yeast to a level that can be detected
by immunoblotting, but these proteins can still be used in two-
hybrid and reverse two-hybrid methods.
References
1. Fields S, Song O (1989) A novel genetic system 9. Maier RH, Maier CJ, Hintner H et al (2012)
to detect protein-protein interactions. Nature Quantitative real-time PCR as a sensitive
340:245–246 protein-protein interaction quantification
2. Chien CT, Bartel PL, Sternglanz R et al (1991) method and a partial solution for non-
The two-hybrid system: a method to identify accessible autoactivator and false-negative mol-
and clone genes for proteins that interact with a ecule analysis in the yeast two-hybrid system.
protein of interest. Proc Natl Acad Sci U S A Methods 58:376–384
88:9578–9582 10. Pellet J, Meyniel L, Vidalain PO et al (2009)
3. Boman AL, Zhang C, Zhu X et al (2000) A pISTil: a pipeline for yeast two-hybrid Interac-
family of ADP-ribosylation factor effectors that tion Sequence Tags identification and analysis.
can alter membrane transport through the BMC Res Notes 2:220
trans-Golgi. Mol Biol Cell 11:1241–1255 11. Rajagopala SV, Uetz P (2009) Analysis of
4. Van Valkenburgh H, Shern JF, Sharer JD et al protein-protein interactions using array-based
(2001) ADP-ribosylation factors (ARFs) and yeast two-hybrid screens. Methods Mol Biol
ARF-like 1 (ARL1) have both specific and 548:223–245
shared effectors: characterizing ARL1-binding 12. Ratushny V, Golemis E (2008) Resolving the
proteins. J Biol Chem 276:22826–22837 network of cell signaling pathways using the
5. Ito T, Chiba T, Ozawa R et al (2001) A com- evolving yeast two-hybrid system. Biotechni-
prehensive two-hybrid analysis to explore the ques 44:655–662
yeast protein interactome. Proc Natl Acad Sci 13. Cagney G, Uetz P (2001) High-throughput
U S A 98:4569–4574 screening for protein-protein interactions
6. Ito T, Tashiro K, Muta S et al (2000) Toward a using yeast two-hybrid arrays. Curr Protoc
protein-protein interaction map of the bud- Protein Sci Chapter 19:Unit 19 16.
ding yeast: a comprehensive system to examine 14. Parrish JR, Gulyas KD, Finley RL Jr (2006)
two-hybrid interactions in all possible combi- Yeast two-hybrid contributions to interactome
nations between the yeast proteins. Proc Natl mapping. Curr Opin Biotechnol 17:387–393
Acad Sci U S A 97:1143–1147 15. Vidalain PO, Boxem M, Ge H et al (2004)
7. Uetz P (2002) Two-hybrid arrays. Curr Opin Increasing specificity in high-throughput yeast
Chem Biol 6:57–62 two-hybrid experiments. Methods
8. Uetz P, Giot L, Cagney G et al (2000) A com- 32:363–370
prehensive analysis of protein-protein interac- 16. Legrain P, Selig L (2000) Genome-wide pro-
tions in Saccharomyces cerevisiae [see tein interaction maps using two-hybrid sys-
comments]. Nature 403:623–627 tems. FEBS Lett 480:32–36
17. Walhout AJ, Boulton SJ, Vidal M (2000) Yeast 29. Durfee T, Becherer K, Chen PL et al (1993)
two-hybrid systems and protein interaction The retinoblastoma protein associates with the
mapping projects for yeast and worm. Yeast protein phosphatase type 1 catalytic subunit.
17:88–94 Genes Dev 7:555–569
18. Chappell TG, Gray PN (2008) Protein interac- 30. Boman AL, Kuai J, Zhu X et al (1999) Arf
tions: analysis using allele libraries. Adv Bio- proteins bind to mitotic kinesin-like protein 1
chem Eng Biotechnol 110:47–66 (MKLP1) in a GTP-dependent fashion. Cell
19. Endoh H, Vincent S, Jacob Y et al (2002) Motil Cytoskeleton 44:119–132
Integrated version of reverse two-hybrid sys- 31. Kuai J, Kahn RA (2000) Residues forming a
tem for the postproteomic era. Methods Enzy- hydrophobic pocket in ARF3 are determinants
mol 350:525–545 of GDP dissociation and effector interactions.
20. Endoh H, Walhout AJ, Vidal M (2000) A FEBS Lett 487:252–256
green fluorescent protein-based reverse two- 32. Muhlrad D, Hunter R, Parker R (1992) A
hybrid system: application to the characteriza- rapid method for localized mutagenesis of
tion of large numbers of potential protein- yeast genes. Yeast 8:79–82
protein interactions. Methods Enzymol 33. Bai C, Elledge SJ (1996) Gene identification
328:74–88 using the yeast two-hybrid system. Methods
21. Vidal M, Endoh H (1999) Prospects for drug Enzymol 273:331–347
screening using the reverse two-hybrid system. 34. Rose MD, Winston F, Hieter P (1990) In:
Trends Biotechnol 17:374–381 Rose MD, Winston F, Hieter P (eds) Labora-
22. Zhu X, Kahn RA (2001) The Escherichia coli tory course manual for methods in yeast genet-
heat labile toxin binds to Golgi membranes and ics. Cold Spring Harbor Laboratory, Cold
alters Golgi and cell morphologies using ADP- Spring Harbor, NY
ribosylation factor-dependent processes. J Biol 35. Horvath A, Riezman H (1994) Rapid protein
Chem 276:25014–25021 extraction from Saccharomyces cerevisiae. Yeast
23. Zhu X, Kim E, Boman AL et al (2001) ARF 10:1305–1310
binds the C-terminal region of the Escherichia 36. Bartel P, Chien CT, Sternglanz R et al (1993)
coli heat-labile toxin (LTA1) and competes for Elimination of false positives that arise in using
the binding of LTA2. Biochemistry the two-hybrid system. Biotechniques
40:4560–4568 14:920–924
24. Das S, Kalpana GV (2009) Reverse two-hybrid 37. James P, Halladay J, Craig EA (1996) Genomic
screening to analyze protein-protein interac- libraries and a host strain designed for highly
tion of HIV-1 viral and cellular proteins. Meth- efficient two-hybrid selection in yeast. Genetics
ods Mol Biol 485:271–293 144:1425–1436
25. Cadwell RC, Joyce GF (1992) Randomization 38. Brent R, Finley RL Jr (1997) Understanding
of genes by PCR mutagenesis. PCR Methods gene and allele function with two-hybrid meth-
Appl 2:28–33 ods. Annu Rev Genet 31:663–704
26. Leanna CA, Hannink M (1996) The reverse 39. Vidal M, Legrain P (1999) Yeast forward and
two-hybrid system: a genetic scheme for selec- reverse ‘n’-hybrid systems. Nucleic Acids Res
tion against specific protein/protein interac- 27:919–929
tions. Nucleic Acids Res 24:3341–3347 40. Shafikhani S, Siegel RA, Ferrari E et al (1997)
27. Puthalakath H, Strasser A, Huang DC (2001) Generation of large libraries of random
Rapid selection against truncation mutants in mutants in Bacillus subtilis by PCR-based plas-
yeast reverse two-hybrid screens. Biotechni- mid multimerization. Biotechniques
ques 30:984–988 23:304–310
28. Vidal M, Brachmann RK, Fattaey A et al 41. Guarente L (1983) Yeast promoters and lacZ
(1996) Reverse two-hybrid and one-hybrid fusions designed to study expression of cloned
systems to detect dissociation of protein- genes in yeast. Methods Enzymol
protein and DNA-protein interactions. Proc 101:181–191
Natl Acad Sci U S A 93:10315–10320
Chapter 29
MAPPIT, a Mammalian Two-Hybrid Method for In-Cell

Detection of Protein-Protein Interactions
Irma Lemmens, Sam Lievens, and Jan Tavernier
Abstract
MAPPIT (MAmmalian Protein-Protein Interaction Trap) is a two-hybrid technology that facilitates the
detection and analysis of interactions between proteins in living mammalian cells. The system is based on
type 1 cytokine receptor signaling. The bait protein of interest is fused to a chimeric signaling-deficient
cytokine receptor, the signaling competence of which is restored upon recruitment of a prey protein that is
coupled to a functional cytokine receptor domain. MAPPIT exhibits an excellent signal-to-noise ratio,
detects a wide variety of protein-protein interactions (PPIs) including transient and indirect interactions,
and has been shown to be highly complementary to other two-hybrid methods with respect to the
interactions it can detect. Variants of the method were developed to allow large-scale PPI screening,
mapping of protein interaction interfaces, PPI inhibitor screening and drug profiling. This chapter
describes a basic 4-day MAPPIT protocol for the analysis of interaction between two designated proteins.
Key words MAPPIT, Two-hybrid, Protein-protein interaction, Cytokine receptor, Interactomics,

Mammalian cells
1 Introduction
MAPPIT (MAmmalian Protein-Protein Interaction Trap) is a

mammalian two-hybrid method that is based on type 1 cytokine
receptor signaling. Type 1 cytokine receptors, such as the erythro-
poietin receptor (EpoR) and leptin receptor (LepR), lack intrinsic
kinase activity but make use of constitutively associated Janus
Kinases (JAK) to relay signals intracellularly. When the cytokine
receptor is stimulated by its ligand, the associated JAKs are acti-
vated by cross-phosphorylation and subsequently phosphorylate
tyrosine residues in the cytoplasmic tail of the receptor. These
phosphorylated tyrosine motifs recruit STAT (Signal Transducer
and Activator of Transcription) proteins that are in their turn
phosphorylated by the activated JAKs, upon which the STAT
molecules migrate to the nucleus and activate transcription of a
selected set of target genes.
447
448 Irma Lemmens et al.
In the MAPPIT assay, a signaling-deficient chimeric type 1

cytokine receptor is used. Its extracellular domain is typically
derived from the homodimeric EpoR and is fused to the transmem-
brane domain and cytoplasmic tail of the LepR. This chimeric
receptor is made signaling-deficient by mutating the 3 tyrosines
present in the cytoplasmic tail to phenylalanine. Upon ligand
administration the associated JAKs of this chimeric receptor are
activated, but since there are no tyrosines left in the receptor tail
that can be phosphorylated, no STAT recruitment sites will be
created and no signal is transmitted to the nucleus. The protein of
interest that will serve as bait is C-terminally fused to this signaling-
deficient receptor. The protein which one wants to test for interac-
tion with the bait protein, referred to as the prey, is coupled to a
portion of another cytokine receptor, the glycoprotein 130 recep-
tor (gp130). The gp130 domain that is used contains tyrosine
motifs that after phosphorylation by the JAKs serve as STAT3
recruitment sites. If the two proteins of interest (bait and prey)
interact, the JAKs will be able to phosphorylate these tyrosines
upon cytokine ligand activation. This generates functional STAT3
binding sites, and recruited STATs, upon activation by the JAKs
will migrate to the nucleus to activate transcription of a luciferase
reporter gene (Fig. 1) [1].
EpoR
Cellular membrane
JAK JAK
LepR F F
F F
F F
P
P B
B B
B P STAT3
Nuclear membrane
induction of STAT3
responsive reporter gene
Fig. 1 Schematic representation of the MAPPIT principle. Interaction between

bait and prey protein results in the recruitment of a gp130 fragment containing
STAT3 recruitment sites, thereby complementing the signaling-deficient
chimeric receptor bait. Activation of STAT3 can be monitored using the
STAT3-responsive rat Pap reporter gene fused to luciferase (for details see
text). EpoR, erythropoietin receptor; LepR, leptin receptor; JAK, Janus kinase;
F, a tyrosine mutated to a phenylalanine; gp130, glycoprotein 130; STAT3, signal
transducer and activator of transcription 3; B, bait; P, prey. Phosphorylation
events are indicated by an arrow and a light-colored P
MAPPIT, In-Cell PPI Detection 449
One of the main assets of MAPPIT is that the interaction

occurs in intact mammalian cells, representing a near-physiological
environment to study human proteins. The interaction (cytoplasm)
and detector (nucleus) site are physically separated which reduces
background problems (e.g., those caused by transcriptional auto-
activators in the yeast two-hybrid system). Moreover, the fact that
the readout is dependent on ligand administration provides an
additional background control. MAPPIT has been successfully
used for in-depth analyses of the protein interaction profiles of
different type 1 cytokine receptors, such as the EpoR, LepR and
growth hormone receptor, as well as for toll-like receptor signaling
and for the detection of interactions with viral proteins [2–10].
These analyses have shown that the approach can detect transient as
well as indirect interactions. Extensive benchmarking against other
PPI methods indicated that MAPPIT is complementary to other
two-hybrid methods with respect to the interactions it can detect,
and is a valuable tool for large-scale protein-protein interactomics
studies [11]. Downscaling the assay from a 96-well to a 384-well
format using liquid handling robotics increases the throughput,
allowing the parallel analysis of several thousands of interactions
in order to validate the quality of various protein-protein interac-
tomes, generated by a high-throughput yeast two-hybrid method
[12–16]. This robotized platform is also applied for detailed char-
acterization of protein-protein interaction interfaces in an unbiased
manner, utilizing a random mutagenesis based strategy [17–19].
To use MAPPIT as a primary screening tool for the detection of
novel interactions, ArrayMAPPIT was developed, which allows a
bait protein to be screened against a large collection of preys
(currently over 12.500) in only 4 days [20, 21]. Alternatively, a
FACS-based cDNA library screening variant can also be used [22].
Owing to its robustness, the MAPPIT assay has also proven useful
for high-throughput screening for small molecule disruptors of
PPIs, which represent an emerging drug target class [23]. Finally,
compound-protein interaction profiling can be performed using
the MASPIT variant where a methotrexate fusion of a compound
of interest is linked to a dihydrofolate reductase (DHFR) bait
receptor [24]. For a more detailed overview of the different appli-
cations of the MAPPIT technology platform we refer to Lievens
et al. [25].
Here we describe a basic MAPPIT protocol that enables detec-
tion of an interaction between two proteins of interest. In brief, this
procedure involves the proteins of interest to be cloned in the
appropriate MAPPIT bait and prey plasmids, co-transfecting these
together with a luciferase reporter in mammalian (typically
Hek293T) cells, stimulating the cells with the cytokine ligand,
and lastly measuring luciferase reporter activity.
2 Materials
2.1 Plasmids 1. The pSEL(+2L) plasmid. The backbone of this plasmid is

derived from pSV-SPORT, which carries an early SV40
promoter for low-level expression in mammalian cells and
encodes the signaling-deficient chimeric receptor to which the
bait should be C-terminally fused after a glycine–glycine–serine
linker. Alternative plasmids can be used for cloning the bait (see
Note 1, [1, 26]).
2. The pMG1 construct. The prey should be cloned C-terminally,
after a glycine–glycine–serine hinge, in this plasmid that
encodes an N-terminally FLAG-tagged part of the gp130
receptor (amino acids 760–918) using the strong SRα pro-
moter of the pMET7 vector [1].
3. The reporter construct pXP2d2-rPAP1-luci. This reporter
contains a STAT3-dependent promoter fragment derived
from the rat Pancreatitis Associated Protein 1 (Pap) gene
fused to luciferase [1].
2.2 Cell Culture 1. Human embryonic kidney (Hek) 293 T cell line.
2. Growth medium: Dulbecco’s modified Eagle’s medium
(DMEM) supplemented with 10 % Fetal Calf Serum and anti-
biotics like gentamycin or penicillin/streptomycin.
3. Human Epo.
2.3 Transient 1. 2.5 M CaCl2. Prepare in distilled water. Filter-sterilize by

Ca3(PO4)2 Transfection passage through a 0.45 μM nitrocellulose membrane and
of Hek293T Cells store at 20 C.
2. 2 Hepes-Buffered Saline (HeBS): 280 mM NaCl, 1.5 mM
Na2HPO4, 50 mM HEPES. Adjust pH to 7.05 with NaOH.
Filter-sterilize by passage through a 0.45 μM nitrocellulose
membrane and store at 20 C.
2.4 Luciferase Assay 1. Luciferase lysis buffer: 25 mM Tris-phosphate, pH 7.8, 2 mM

DTT, 2 mM 2,2 diaminocyclohexane-N,N,N0 ,N0 -tetra-acetate
(DCTA), 10 % glycerol, 1 % Triton X-100. Store at 20 C.
2. Luciferase substrate buffer: 40 mM tricine, 2.14 mM
(MgCO3)4 Mg(OH)2 5H2O, 5.34 mM MgSO4 , 66.6 mM
DTT, 0.2 mM EDTA, Coenzyme A, 734 μM Adenosine 50
triphosphate, 940 μM D-luciferin. Store at 20 C. The
reagent is light sensitive.
3. Luminescence counter.
3 Methods
3.1 Cloning Bait and 1. Design primers flanking your gene of interest that
Prey of Interest incorporate relevant restriction sites to clone the desired bait
in the pSEL(+2L) plasmid and the prey in the pMG1 construct.
Suitable restriction sites to clone the bait in the pSEL(+2L)
vector are SalI or SacI in combination with NotI or XbaI. The
prey can be cloned in the pMG1 vector using EcoRI combined
with NotI or XbaI (see Notes 1 and 2). Both plasmids also exist
in a gateway recombinatorial cloning compatible version,
allowing the transfer of bait and prey using an LR clonase
reaction (Invitrogen).
2. Perform a Polymerase Chain Reaction using a polymerase with
proofreading activity (e.g., Pfu DNA polymerase).
3. Use traditional cloning methods or gateway recombination to
insert the genes of interest into the appropriate vectors (Fig. 2).
4. Transform the resulting plasmids into a suitable bacterial line.
5. Prepare plasmid DNA of transfection-suitable quality, and vali-
date the plasmids by restriction digest or sequencing.
Fig. 2 Overview of the MAPPIT bait and prey plasmids. Plasmid pSEL(+2L) is
used for cloning the bait; plasmid pMG1 to fuse the prey C-terminally to the
gp130 moiety. A hinge (amino acids Gly–Gly–Ser) is placed between the fusions
for extra flexibility. EpoR, erythropoietin receptor; +2L, two extra leucines were
added in the transmembrane domain to optimize the fusion between the
extracellular domain of the EpoR and the cytoplasmic tail of the LepR. LepRF3,
leptin receptor variant with 3 tyrosines mutated to phenylalanines that lacks a
functional STAT3 recruitment site; gp130, glycoprotein 130; SV40, simian virus
40 early promoter; SRα, promoter comprising SV40 and the R-U5 segment of
human T-cell leukemia virus type 1 long terminal repeat; GGS, glycine–glyci-
ne–serine; pA, SV40 polyadenylation signal
3.2 Seeding Cells 1. For each bait–prey combination to be tested seed 1 104
subconfluent Hek293T cells in six wells of a black 96-well
plate in 100 μl growth medium per well. Cell lines other than
Hek293T cells can be used (see Notes 3 and 4).
Besides testing the bait–prey interaction the following con-
trols should be included: a bait–irrelevant prey and irrelevant
bait–prey combination to control for a possible background
caused by the bait or prey, respectively. Additionally, a prey (like
SH2beta) that can interact with the bait receptor independent
of the bait can serve as an additional control to check the
signaling capacity of the bait receptor.
2. Grow overnight in a humidified atmosphere at 37 C and 5–8 %
CO2.
3.3 Transient 1. For each bait–prey combination, make a DNA and CaCl2
Ca3(PO4)2 Transfection mixture containing 250 ng of the bait plasmid, 500 ng of the
prey plasmid, and 50 ng of the reporter (pXP2d2-rPAP1-luci)
with 5 μl of sterile 2.5 M CaCl2 in a total volume of 50 μl of
distilled water in a 96-well plate (see Notes 5 and 6).
2. Add 50 μl of 2 HeBS to each well containing the DNA and
CaCl2 mixtures.
3. Shake the plate for 1 min at 800 rpm.
4. Resuspend the DNA, CaCl2 and 2 HeBS mixture.
5. Add 10 μl to the cells in the 96-well plate in sextuple for every
bait–prey combination.
6. Incubate the cells overnight in a humidified atmosphere at
37 C and 5–8 % CO2.
The precipitate can be checked under a microscope: the parti-
cles should look like small speckles to obtain optimal transfection
efficiencies (see Note 7). Alternative commercially available trans-
fection reagents like lipofectamine can be used as well.
3.4 Stimulation of 1. For each bait–prey combination, stimulate three of the six wells
the Cells with 50 μl Epo at a final concentration of 5 ng/μl. To the
remaining three wells 50 μl of growth medium (without cyto-
kine) should be added.
2. Incubate the cells overnight in a humidified atmosphere at
37 C and 5–8 % CO2.
3.5 Measurement of 1. Remove growth medium from the black 96-well plates.
the Luciferase Activity 2. Add 50 μl of luciferase lysis buffer to each well and incubate for
10 min at room temperature.
3. Add 35 μl of luciferase substrate buffer to each well and imme-
diately measure the luminescent signals using a chemilumines-
cence reader suitable for 96-well format.
3.6 Data Analysis 1. Calculate the fold induction value for each bait–prey
combination by dividing the average of the values of the sti-
mulated wells by the average of the values of the non-
stimulated wells.
2. The bait–prey interaction is scored positive if the fold induction
value obtained by the bait–prey combination is at least three
times higher than both the fold induction value of the
bait–irrelevant prey and irrelevant bait–prey interactions.
4 Notes
1. Alternative plasmids for cloning the bait can be used, such as

the pCLL vector wherein the extracellular domain of the
homodimeric EpoR receptor is replaced by the oligomeric
LepR. Additionally, the cytoplasmic tail of the LepR after the
JAK binding site can be replaced by glycine–glycine–serine
repeats. For a more detailed description of these different
plasmids, we refer to Lemmens et al. [26]. An alternative prey
plasmid, named pMG2, can be used as well. In this plasmid, a
part of the gp130 moiety (amino acids 905–918) is duplicated
thereby adding two additional tyrosines [27]. Although this
can increase the obtained signal, the background is often
elevated as well.
2. When deciding which one of the protein pair of interest to be
tested should be cloned as bait or prey, consider if it involves a
protein that is targeted to a particular cellular organelle (like
the nucleus or mitochondria). It is better to clone these genes
as bait since detection of the interaction in MAPPIT takes place
in the submembranal space of the cytoplasm. For a functional
prey fusion, the targeting signals (such as a nuclear localization
signal) should be omitted. Likewise, full-length integral trans-
membrane proteins will be problematic to study using
MAPPIT.
3. Cells other than Hek293T cells (e.g., the hematopoietic TF-1
or neuronal N38 cell lines [2, 28]) can be used as long as
they are transfectable and contain a sufficient amount of endo-
genous STAT3 molecules to transmit the signal to the nucleus.
An alternative MAPPIT method wherein the gp130 moiety is
replaced with a part of the beta common receptor chain, which
carries STAT5 instead of STAT3 recruitment sites, can be more
suitable for certain cell lines like the hematopoietic Ba/F3
cells which express little endogenous STAT3 but sufficient
STAT5 [29].
4. It is important to use cells that are in a logarithmic growth
phase (subconfluent), overgrown cultures won’t transfect
efficiently.
5. Different amounts of bait and prey DNA may yield better

results in some settings.
6. If background caused by the prey is observed when testing the
irrelevant bait–prey interaction, lowering the amount of prey
DNA can reduce or diminish the background.
7. If no precipitation can be observed, check the pH of the
2 HeBS buffer, as this is critical for the formation of the
precipitate.
Acknowledgements
pXP2d2 was a gift from Dr. S. Nordeen, Colorado Health Sciences

Center, Department of Pathology, Denver, CO, 80262, USA. This
work was supported by a grant from the Fund for Scientific
Research-Flanders (FWO-V grant to I.L.).
References
1. Eyckerman S, Verhee A, Van der Heyden J et al 9. Pattyn E, Lavens D, Van der Heyden J et al
(2001) Design and application of a cytokine- (2008) MAPPIT (MAmmalian Protein-
receptor-based interaction trap. Nat Cell Biol Protein Interaction Trap) as a tool to study
3:1114–1119 HIV reverse transcriptase dimerization in intact
2. Montoye T, Lemmens I, Catteeuw D et al human cells. J Virol Methods 153:7–15
(2005) A systematic scan of interactions with 10. Ulrichts P, Peelman F, Beyaert R et al (2007)
tyrosine motifs in the erythropoietin receptor MAPPIT analysis of TLR adaptor complexes.
using a mammalian 2-hybrid approach. Blood FEBS Lett 581:629–636
105:4264–4271 11. Braun P, Tasan M, Dreze M et al (2009) An
3. Uyttendaele I, Lemmens I, Verhee A et al experimentally derived confidence score
(2007) Mammalian protein-protein interaction for binary protein-protein interactions. Nat
trap (MAPPIT) analysis of STAT5, CIS, Methods 6:91–97
and SOCS2 interactions with the growth 12. Boxem M, Maliga Z, Klitgord N et al (2008) A
hormone receptor. Mol Endocrinol 21: protein domain-based interactome network for
2821–2831 C. elegans early embryogenesis. Cell
4. Lavens D, Montoye T, Piessevaux J et al (2006) 134:534–545
A complex interaction pattern of CIS and 13. Simonis N, Rual JF, Carvunis AR et al (2009)
SOCS2 with the leptin receptor. J Cell Sci Empirically controlled mapping of the Caenor-
119:2214–2224 habditis elegans protein-protein interactome
5. Lavens D, Ulrichts P, Catteeuw D et al (2007) network. Nat Methods 6:47–54
The C-terminus of CIS defines its interaction 14. Simonis N, Rual JF, Lemmens I et al (2012)
pattern. Biochem J 401:257–267 Host-pathogen interactome mapping for
6. Piessevaux J, Lavens D, Montoye T et al (2006) HTLV-1 and -2 retroviruses. Retrovirology
Functional cross-modulation between SOCS 9:26
proteins can stimulate cytokine signaling. 15. Venkatesan K, Rual JF, Vazquez A et al (2009)
J Biol Chem 281:32953–32966 An empirical framework for binary interactome
7. Piessevaux J, De Ceuninck L, Catteeuw D et al mapping. Nat Methods 6:83–90
(2008) Elongin B/C recruitment regulates 16. Yu H, Braun P, Yildirim MA et al (2008) High-
substrate binding by CIS. J Biol Chem quality binary protein interaction map of the
283:21334–21346 yeast interactome network. Science
8. Ulrichts P, Tavernier J (2008) MAPPIT analy- 322:104–110
sis of early Toll-like receptor signalling events. 17. Bovijn C, Ulrichts P, De Smet AS et al (2012)
Immunol Lett 116:141–148 Identification of interaction sites for
dimerization and adapter recruitment in Toll/ high-throughput drug screening. Methods

interleukin-1 receptor (TIR) domain of Toll- 58:335–342
like receptor 4. J Biol Chem 287:4088–4098 24. Caligiuri M, Molz L, Liu Q et al (2006)
18. Uyttendaele I, Lavens D, Catteeuw D et al MASPIT: three-hybrid trap for quantitative
(2012) Random mutagenesis MAPPIT analysis proteome fingerprinting of small molecule-
identifies binding sites for Vif and Gag in both protein interactions in mammalian cells.
cytidine deaminase domains of Apobec3G. Chem Biol 13:711–722
PLoS One 7:e44143 25. Lievens S, Peelman F, De Bosscher K et al
19. Bovijn C, De Smet AS, Uyttendaele I et al (2011) MAPPIT: a protein interaction
(2013) Identification of Binding Sites for Mye- toolbox built on insights in cytokine receptor
loid Differentiation Primary Response Gene 88 signaling. Cytokine Growth Factor Rev
(MyD88) and Toll-like Receptor 4 in MyD88 22:321–329
Adapter-like (Mal). J Biol Chem 26. Lemmens I, Lievens S, Tavernier J (2008)
288:12054–12066 MAPPIT: a versatile tool to study cytokine
20. Lievens S, Vanderroost N, Defever D et al receptor signalling. Biochem Soc Trans
(2012) ArrayMAPPIT: a screening platform 36:1448–1451
for human protein interactome analysis. 27. Lemmens I, Eyckerman S, Zabeau L et al
Methods Mol Biol 812:283–294 (2003) Heteromeric MAPPIT: a novel strategy
21. Lievens S, Vanderroost N, Van der Heyden J to study modification-dependent protein-
et al (2009) Array MAPPIT: high-throughput protein interactions in mammalian cells.
interactome analysis in mammalian cells. J Pro- Nucleic Acids Res 31:e75
teome Res 8:877–886 28. Wauman J, De Smet AS, Catteeuw D et al
22. Lievens S, Van der Heyden J, Vertenten E et al (2008) Insulin receptor substrate 4 couples
(2004) Design of a fluorescence-activated cell the leptin receptor to multiple signaling path-
sorting-based Mammalian protein-protein ways. Mol Endocrinol 22:965–977
interaction trap. Methods Mol Biol 29. Montoye T, Piessevaux J, Lavens D et al (2006)
263:293–310 Analysis of leptin signalling in hematopoietic
23. Lievens S, Caligiuri M, Kley N et al (2012) The cells using an adapted MAPPIT strategy.
use of mammalian two-hybrid technologies for FEBS Lett 580:3301–3307
Chapter 30
Bioluminescence Resonance Energy Transfer to Detect

Protein-Protein Interactions in Live Cells
Nicole E. Brown, Joe B. Blumer, and John R. Hepler
Abstract
Bioluminescence resonance energy transfer (BRET) is a valuable tool to detect protein-protein interactions.
BRET utilizes bioluminescent and fluorescent protein tags with compatible emission and excitation proper-
ties, making it possible to examine resonance energy transfer when the tags are in close proximity (<10 nm)
as a typical result of protein-protein interactions. Here we describe a protocol for detecting BRET from two
known protein binding partners (Gαi1 and RGS14) in HEK 293 cells using Renilla luciferase and yellow
fluorescent protein tags. We discuss the calculation of the acceptor/donor ratio as well as net BRET and
demonstrate that BRET can be used as a platform to investigate the regulation of protein-protein interac-
tions in live cells in real time.
Key words Bioluminescence resonance energy transfer (BRET), Renilla luciferase (RLuc), Yellow
fluorescent protein (YFP), RGS14, G protein regulatory (GPR) motif, G protein
1 Introduction
Bioluminescence Resonance Energy Transfer (BRET) is a method

of studying protein-protein interactions in live cells [1]. BRET
utilizes non-radiative energy transfer between energy donor and
energy acceptor protein tags. The energy transfer occurs when the
protein tags are in close proximity, as described by the Förster
distance [2]. As shown in Fig. 1, BRET serves as a molecular
ruler, detecting protein-protein interactions under 10 nm
(see Note 1). For a comprehensive review, see refs. [3, 4].
BRET makes use of a bioluminescent energy donor while the
energy acceptor is a fluorophore. The choice of BRET pair is based
on the overlap of the bioluminescent protein (donor) emission
spectrum with the excitation spectrum of the fluorescent protein
(acceptor). For BRET experiments, the most commonly chosen
bioluminescent donor is luciferase from the sea pansy Renilla reni-
formis. Renilla luciferase (RLuc) catalyzes the oxidation of its
substrate, coelenterazine, to produce blue light at 482 nm.
457
458 Nicole E. Brown et al.
Fig. 1 BRET is dependent on the distance between the donor luciferase and the
acceptor fluorophore. Addition of the cell permeant Renilla luciferase substrate
coelenterazine (ctz) results in oxidation of the substrate to coelenteramide,
which produces blue light at 482 nm. When protein-protein interactions between
Protein X and Protein Y bring the donor luciferase (RLuc) and acceptor fluor-
ophore (YFP) in close proximity (<10 nm), the energy from the donor can be
transferred to the acceptor and light is produced at 527 nm. When the BRET tags
are not in close enough proximity, light is only emitted at 482 nm
Fig. 2 The energy transfer between BRET pairs depends on the overlap of the
donor emission spectrum with the excitation spectrum of the acceptor. For
Renilla luciferase, oxidation of coelenterazine results in an emission peak at
482 nm. This emission overlaps well with the excitation spectrum of yellow
fluorescent protein (excitation peak: 514 nm). The resulting energy transfer
yields yellow light with an emission peak of 527 nm
The emission spectrum of RLuc overlaps well with the excitation

spectra of the yellow fluorescent protein (YFP) family of proteins
including the mutant YFP variants enhanced YFP (EYFP) and
Venus [5] which emit light at ~527 nm (Fig. 2). For more informa-
tion on BRET pairs, see Note 2.
BRET 459
BRET has distinct advantages over other techniques to detect

protein-protein interactions. First, BRET is amenable to detecting
interactions in live cells, thus proteins retain posttranslational mod-
ifications and cellular trafficking regulations that may be important
for protein-protein interactions. BRET is readily adaptable to
almost any cell type that allows expression of the donor and accep-
tor proteins. In live cells, protein-protein interactions can be moni-
tored in real time over a time-course or for a fixed time interval in
response to cellular treatments such as exposure to GPCR agonists,
growth factors, or other drugs as an approach to define the regula-
tion of protein complexes [6–12]. Additionally, fusion of BRET
pairs to the same recombinant protein can be used to develop small
molecule biosensors [13]. Methods have also been developed using
BRET as a reporter for movement and subcellular location of target
proteins [14]. Moreover, unlike the similar technique fluorescence
resonance energy transfer (FRET), BRET does not require external
excitation but instead relies on the addition of the cell permeant
substrate coelenterazine to initiate the assay, thereby endowing the
experimenter with temporal control over the assay and preventing
unintentional activation of the acceptor fluorophore. Given these
many advantages, BRET can be readily adapted for high through-
put screening for small molecule modulators of protein-protein
interactions. For review, see refs. [9, 15].
Below we describe a BRET experiment to explore the
interactions between Regulator of G protein Signaling 14
(RGS14) and its binding partner Gαi1. RGS14 has previously
been shown to interact with Gαi1 through its G protein regulatory
(GPR) motif by traditional biochemical methods [16, 17]. We
detail transfection of a C-terminal luciferase tagged RGS14
(RGS14-Luc) donor and internal YFP tagged Gαi1 (Gαi1-YFP)
acceptor. We demonstrate a robust BRET signal between wild
type RGS14 and Gαi1 that is disrupted with a mutant RGS14
(Q515A/R516A) that can no longer bind Gαi1. In our example,
we show how to vary the acceptor protein expression level to
achieve optimal net BRET signal. We describe how to calculate
net BRET, acceptor/donor ratio, and fit the data using graphing
software.
2 Materials
2.1 Cell Lines 1. Maintain HEK 293 cells in 1 Dulbecco’s Modified Eagle
Medium (DMEM) without phenol red indicator, supplemented
with 2 mM L-glutamine, 100 U/mL penicillin, 100 mg/mL
streptomycin, and 10 % fetal bovine serum (5 % for transfection).
Grow cells in a humidified incubator with 5 % CO2 at 37 C.
2.2 Buffer 1. BRET buffer (Tyrode’s solution): 140 mM NaCl, 5 mM KCl,

Compositions/Stock 1 mM MgCl2, 1 mM CaCl2, 0.37 mM NaH2PO4, 24 mM
Solutions NaHCO3, 10 mM HEPES, pH 7.4, 0.1 % glucose.
2. Polyethylenimine (PEI) transfection reagent stock solution:
Dissolve PEI (1 mg/mL) in dH2O at 80 C while stirring,
cool and adjust to pH 7.2 using 0.1 N HCl and filter-sterilize.
Aliquot and store at 80 C. Use each aliquot only once.
3. Luciferase substrate stock solution: 2 mM benzyl coelentera-
zine H in 100 % ethanol containing 60 mM HCl, aliquot and
store at 80 C.
2.3 Instrumentation 1. Microplate reader with 485 nm emission and excitation filters,
530 nm emission filters, and compatible white-bottomed
96-well plates (see Note 3).
2. Compatible microplate reader, spreadsheet, and graphing
software (see Note 3).
2.4 Plasmids 1. Donor plasmids can be constructed by inserting your gene of

interest into a vector containing the humanized RLuc gene.
For construction of RGS14-Luc constructs presented below,
rat RGS14 cDNA was inserted into the phRLuc-N2 vector,
which places the RLuc tag at the C-terminus of RGS14. Deter-
mining the optimal location of the Rluc tag relative to the
protein of interest is an important parameter and must be
determined empirically (see Note 1).
2. For many BRET experiments, acceptor plasmids can be con-
structed by inserting your gene of interest into a commercially
available vector encoding YFP, EYFP or Venus. For construc-
tion of Gαi1-YFP used below, insertion of the YFP tag at either
terminus compromised the function of Gαi1. Thus, the Gαi1-
YFP construct was engineered by inserting the YFP coding
sequence between the b and c helices of the helical domain of
Gαi1 which was then expressed in a pcDNA3.1 vector [18].
3 Methods
3.1 Experimental 1. Below we describe an experiment where the donor expression

Setup level is set and the acceptor expression level is varied. This
experimental setup will allow the acceptor to saturate the
donor and provide a maximal BRET signal. The expression
level should be empirically determined; however, we typically
use donor plasmid amounts that yield relative luminescence
units (RLU) of 100,000–350,000 in our microplate reader
(TriStar LB 941). In our example, 5 ng of phRLuc-N2::
RGS14 yields an RLU of ~100,000–300,000 but this will
vary for other donor constructs depending on transfection
BRET 461
efficiency, the ability of the transfected cells to express the

donor protein and the type of instrument used for detection.
Additionally, on our microplate platform, we typically use
acceptor plasmid amounts that yield relative fluorescence
units (RFUs) of 30,000–200,000. In our example experiment,
pcDNA3.1::Gαi1-YFP typically yields ~30,000–60,000 RFUs.
2. For the present experiment, 5 ng of the donor plasmid
(RGS14-WT-Luc or RGS14-Q515A/R516A-Luc) was trans-
fected with increasing amounts of acceptor (0, 10, 50, 100,
250, and 500 ng pcDNA3.1::Gαi1-YFP) (see Note 4).
3.2 Transient 1. Seed 8 105 cells per well in six-well plates in 2 mL medium
Transfection with per well, grow in a humidified incubator at 37 C overnight
Polyethylenimine (PEI) with 5 % CO2.
(See Note 5) 2. Prior to transfection, change medium to 1 DMEM contain-
ing 2 mM L-glutamine, 100 U/mL penicillin, 100 mg/mL
streptomycin, and 5 % fetal bovine serum, 2 mL per well.
3. Generate solution A by adding 8 μL of 1 mg/mL PEI from
stock to 92 μL of serum-free medium for each well, allow this
solution to incubate for 3 min.
4. Generate solution B by adding up to 1.5 μg of DNA to 100 μL
of serum-free medium for each condition in 1.5 mL microcen-
trifuge tubes. DNA amount is adjusted to a final concentration
of 1.5 μg by adding empty pcDNA3.1 plasmid.
5. Add 100 μL of solution A to microcentrifuge tubes containing
solution B to create solution C.
6. Cap the 1.5 mL microcentrifuge tube and immediately vortex
for 3 s.
7. Incubate solution C at room temperature for 15 min.
8. Add solution C (~200 μL) dropwise to the appropriate well of
cells in the six-well plates.
9. Allow cells to grow for 1–2 days (the medium does not need to
be changed for PEI transfection).
3.3 BRET 1. Immediately prior to beginning the BRET experiment, prepare

coelenterazine by diluting the stock solution to 50 μM in room
temperature BRET buffer (see Note 6).
2. Aspirate the transfection medium from the six-well plates.
3. To each well, add 750 μL of room temperature BRET buffer,
using a pipette to gently remove the cells from the plate.
4. Plate 90 μL of the cells in triplicate into white-bottomed
96-well plates.
5. Load plate into plate reader and detect fluorescence levels

(excitation: 485 nm, emission: 530 nm) using microplate
reader software (see Note 7).
6. Add 10 μL of coelenterazine solution to each well (5 μM final
concentration of coelenterazine per well).
7. Incubate the cells with coelenterazine for 2 min at room
temperature.
8. Take BRET readings by measuring luminescence at 485 20
nm and fluorescence at 530 20 nm (see Note 8).
3.4 Analysis 1. Export fluorescence and BRET data into spreadsheet software.
2. The BRET ratio can be determined by dividing fluorescence by
luminescence (BRET readings at 530/485 nm).
3. Calculate net BRET by subtracting out background lumines-
cence (BRET readings in cells expressing the donor without
any acceptor).
4. Calculate the acceptor/donor ratio by dividing the initial fluo-
rescence measurements (530 nm) by the luminescence mea-
surements (485 nm).
5. Using graphing software, plot the acceptor/donor ratio against
the net BRET as in Fig. 3. The data can then be fit using a
nonlinear regression, (typically a one-site binding [hyperbola]
is the most appropriate) to observe BRET saturation as a key
indicator of signal specificity.
4 Notes
1. BRET efficiency (donor energy transfer to acceptor) is sensitive

to the donor-acceptor proximity and is inversely proportional
to the sixth power of distance between them. Thus, BRET
signals generally reflect direct protein association; however,
non-robust BRET signals can be detected due to close proxim-
ity without the occurrence of direct binding. For example,
when a third intermediate protein brings the donor and accep-
tor into close proximity [10–12]. In addition to proximity,
BRET also depends on the orientation of the protein tag
dipoles. Inefficient dipole coupling can prevent energy transfer,
despite protein-protein interactions. Thus, it is advantageous
to use linkers (typically four or more Gly residues) inserted
between the proteins of interest and the BRET tags to allow
sufficient movement of the tags. Moreover, placement of tags
must be considered when engineering recombinant proteins
with BRET tags. Placement of the BRET tag at the
N-terminus, internally, or at the C-terminus of the protein
can have a profound impact on the observed BRET signal and
BRET 463
Fig. 3 HEK 293 cells were transfected with increasing amounts of Gαi1-YFP (0,
10, 50, 100, 250, and 500 ng) and either 5 ng RGS14-WT-Luc or RGS14-Q515A/
R516A-Luc. Wild type RGS14 shows a robust BRET signal with Gαi1. Conversely,
the RGS14 mutant (Q515A/R516A) that can no longer bind Gαi1 shows a
drastically reduced maximal BRET signal indicating a disruption in the protein-
protein interaction. The above data is representative of three independent
experiments. Curves were generated with GraphPad Prism 5 using the one-
site binding curve fitting function. Additionally, Gαi1-YFP expression levels were
verified by immunoblot analysis
whether protein function is compromised. For example, place-

ment of the luciferase tag at the N-terminus of RGS14 rather
than the C-terminus results in a dramatic reduction in BRET
signal with Gαi1-YFP. As another example, the acceptor YFP
tag in Gαi1 cannot be placed at either termini without affecting
protein function, and was placed internally between the b and c
helices of the alpha helical domain with minimal consequences
to Gαi1 function [18]. Due to these considerations, two inter-
acting proteins may not always be detected by BRET due to the
distance between the tags or abnormal protein function or
localization through improper tag placement.
2. Additional BRET pairs and BRET substrates have been devel-
oped. Many of these BRET pairs can be used with the method
described above. For a comprehensive review of other BRET

pairs and substrates, see ref. [19].
3. We use the TriStar LB 941 Multimode Microplate Reader
(Berthold Technologies) for our BRET experiments; however,
other plate readers can be used with similar results. The plate
reader must detect light signals at two distinct wavelengths
either simultaneously or sequentially. In our assay, we use Bert-
hold Technologies filters at 485 nm to measure luminescence
and at 530 nm to measure fluorescence though similar filters
can be purchased from other vendors. To collect BRET data,
we use the MikroWin 2000 software (Mikrotek). MikroWin is
specialized for microplate experiments and optimized to run
with a variety of instruments from various manufacturers. Addi-
tionally, data collected in MikroWin can be exported and fur-
ther analyzed in spreadsheet software such as Microsoft Excel.
Data can then be graphed in graphing software such as Graph-
Pad Prism.
4. In order to calculate net BRET, it is necessary to include a
donor-only control (donor transfected without any acceptor).
The donor-only control is used to assess any background
BRET observed in the absence of acceptor. BRET from
donor-only controls are subtracted from observed BRET
values to calculate the net BRET.
5. Other transfection reagents can be used with similar results. We
choose to use PEI as it yields high transfection efficiency and
reproducibility at very affordable cost.
6. Coelenterazine is light sensitive and should be kept away from
light exposure until ready to use. Dilute coelenterazine stock
immediately before performing BRET to prevent breakdown
of the substrate.
7. Initial fluorescence levels are taken to determine the acceptor/
donor ratio as well as an internal control to verify expression of
the fluorescently tagged protein. For experiments where the
amount of donor is held constant and the amount of acceptor is
increased, corresponding increases of the acceptor should be
observed in the fluorescence measurement.
8. As stated above, for detection on the TriStar platform, ideal
luminescence should be about 100,000–350,000 relative lumi-
nescence units (RLUs). Ideal fluorescence should range
between 30,000 and 200,000 relative fluorescence units
(RFUs). As a corollary, typical acceptor/donor ratios range
between 0 and 15; however, this number will vary depending
on transfection efficiency and the expression level of the accep-
tor. In addition, lowering the level of donor expression will also
inflate the acceptor/donor ratio.
BRET 465
References
1. Xu Y, Piston DW, Johnson CH (1999) A inhibitors of cholinesterase-8A (Ric-8A) both
bioluminescence resonance energy transfer regulate the regulator of g protein signaling 14
(BRET) system: application to interacting cir- RGS14.Galphai1 complex in live cells. J Biol
cadian clock proteins. Proc Natl Acad Sci U S A Chem 286:38659–38669
96:151–156 11. Oner SS, Maher EM, Breton B et al (2010)
2. Wu P, Brand L (1994) Resonance energy trans- Receptor-regulated interaction of activator of
fer: methods and applications. Anal Biochem G-protein signaling-4 and Galphai. J Biol
218:1–13 Chem 285:20588–20594
3. Xu Y, Kanauchi A, von Arnim AG et al (2003) 12. Oner SS, An N, Vural A et al (2010) Regula-
Bioluminescence resonance energy transfer: tion of the AGS3.G{alpha}i signaling complex
monitoring protein-protein interactions in liv- by a seven-transmembrane span receptor. J Biol
ing cells. Methods Enzymol 360:289–301 Chem 285:33949–33958
4. Lohse MJ, Nuber S, Hoffmann C (2012) Fluo- 13. Jiang LI, Collins J, Davis R et al (2007) Use of
rescence/bioluminescence resonance energy a cAMP BRET sensor to characterize a novel
transfer techniques to study G-protein-coupled regulation of cAMP by the sphingosine 1-
receptor activation and signaling. Pharmacol phosphate/G13 pathway. J Biol Chem
Rev 64:299–336 282:10576–10584
5. Nagai T, Ibata K, Park ES et al (2002) A variant 14. Lan TH, Liu Q, Li C et al (2012) Sensitive and
of yellow fluorescent protein with fast and effi- high resolution localization and tracking of
cient maturation for cell-biological applica- membrane proteins in live cells with BRET.
tions. Nat Biotechnol 20:87–90 Traffic 13:1450–1456
6. Romero-Fernandez W, Borroto-Escuela DO, 15. Couturier C, Deprez B (2012) Setting Up a
Tarakanov AO et al (2011) Agonist-induced Bioluminescence Resonance Energy Transfer
formation of FGFR1 homodimers and signal- High throughput Screening Assay to Search
ing differ among members of the FGF family. for Protein/Protein Interaction Inhibitors in
Biochem Biophys Res Commun 409:764–768 Mammalian Cells. Front Endocrinol 3:100
7. Gales C, Rebois RV, Hogue M et al (2005) 16. Hollinger S, Taylor JB, Goldman EH et al
Real-time monitoring of receptor and G- (2001) RGS14 is a bifunctional regulator of
protein interactions in living cells. Nat Meth- Galphai/o activity that exists in multiple
ods 2:177–184 populations in brain. J Neurochem
8. Angers S, Salahpour A, Joly E et al (2000) 79:941–949
Detection of beta 2-adrenergic receptor dimer- 17. Kimple RJ, De Vries L, Tronchere H et al
ization in living cells using bioluminescence (2001) RGS12 and RGS14 GoLoco motifs
resonance energy transfer (BRET). Proc Natl are G alpha(i) interaction sites with guanine
Acad Sci U S A 97:3684–3689 nucleotide dissociation inhibitor Activity. J
9. Hamdan FF, Audet M, Garneau P et al (2005) Biol Chem 276:29275–29281
High-throughput screening of G protein- 18. Gibson SK, Gilman AG (2006) Gialpha and
coupled receptor antagonists using a biolumi- Gbeta subunits both define selectivity of G
nescence resonance energy transfer 1-based protein activation by alpha2-adrenergic recep-
beta-arrestin2 recruitment assay. J Biomol tors. Proc Natl Acad Sci U S A 103:212–217
Screen 10:463–475 19. Bacart J, Corbel C, Jockers R et al (2008) The
10. Vellano CP, Maher EM, Hepler JR et al (2011) BRET technology and its application to screen-
G protein-coupled receptors and resistance to ing assays. Biotechnol J 3:311–324
Chapter 31
Mapping Biochemical Networks with Protein Fragment

Complementation Assays
Ingrid Remy and Stephen W. Michnick
Abstract
Cellular biochemical machineries, what we call pathways, consist of dynamically assembling and
disassembling macromolecular complexes. Although our models for the organization of biochemical
machines are derived largely from in vitro experiments, do they reflect their organization in intact, living
cells? We have developed a general experimental strategy that addresses this question by allowing the
quantitative probing of molecular interactions in intact, living cells. The experimental strategy is based on
Protein fragment Complementation Assays (PCA), a method whereby protein interactions are coupled to
refolding of enzymes from cognate fragments where reconstitution of enzyme activity acts as the detector of
a protein interaction. A biochemical machine or pathway is defined by grouping interacting proteins into
those that are perturbed in the same way by common factors (hormones, metabolites, enzyme inhibitors,
etc.). In this chapter we review some of the essential principles of PCA and provide details and protocols for
applications of PCA, particularly in mammalian cells, based on three PCA reporters, dihydrofolate reduc-
tase, green fluorescent protein, and β-lactamase.
Key words Protein fragment Complementation Assays, Dihydrofolate reductase, Green fluorescent
protein, TEM β-lactamase, Two-hybrid, Protein-protein interactions, Methotrexate, CCF2/AM,
Nitrocefin, Fluorescein, Flow cytometry, CHO, COS, HEK 293 cells
1 Introduction
A first step in defining the function of a novel gene is to determine

its interactions with other gene products in an appropriate context;
that is, since proteins make specific interactions with other proteins
as part of functional assemblies, an appropriate way to examine the
function of the product of a novel gene is to determine its physical
relationships with the products of other genes. This is the basis
of the highly successful Yeast Two-Hybrid system [1–6]. The
central problem with two-hybrid screening is that detection of
protein-protein interactions occurs in a fixed context, the nucleus
of S. cerevisiae, and the results of a screening must be validated as
biologically relevant using other assays in appropriate cell, tissue, or
organism models. Although this would be true for any screening
467
468 Ingrid Remy and Stephen W. Michnick
strategy, it would be advantageous if one could combine library

screening with tests for biological relevance into a single strategy,
thus tentatively validating a detected protein as biologically relevant
and eliminating false-positive interactions immediately. It was with
these challenges in mind that our laboratory developed the Protein-
fragment Complementation Assays (PCA). In this strategy, the
gene for an enzyme is rationally dissected into two pieces. Fusion
proteins are constructed with two proteins that are thought to bind
to each other, fused to either of the two probe fragments. Folding
of the probe protein from its fragments is catalyzed by the binding
of the test proteins to each other, and is detected as reconstitution of
enzyme activity. We have already demonstrated that the PCA strat-
egy has the following capabilities: (1) allows for the detection of
protein-protein interactions in vivo and in vitro in any cell type; (2)
allows for the detection of protein-protein interactions in appropri-
ate subcellular compartments or organelles; (3) allows for the
detection of induced versus constitutive protein-protein interac-
tions that occur in developmental, nutritional, environmental, or
hormone-induced signals; (4) allows for the detection of the kinetic
and equilibrium aspects of protein assembly in these cells; (5) allows
for screening of cDNA libraries for protein-protein interactions in
any cell type.
In addition to the specific capabilities of PCA described above,
there are special features of this approach that make it appropriate
for screening of molecular interactions, including the following: (1)
PCAs are not a single assay but a series of assays; an assay can be
chosen because it works in a specific cell type appropriate for
studying interactions of some class of proteins; (2) PCAs are inex-
pensive, requiring no specialized reagents beyond that necessary for
a particular assay and off the shelf materials and technology; (3)
PCAs can be automated and high-throughput screening could be
done; (4) PCAs are designed at the level of the atomic structure of
the enzymes used; because of this, there is additional flexibility in
designing the probe fragments to control the sensitivity and strin-
gencies of the assays; (5) PCAs can be based on enzymes for which
the detection of protein-protein interactions can be determined
differently including by dominant selection or production of a
fluorescent or colored product.
The selection of enzymes and design of PCAs have been dis-
cussed in detail [7–19] and here we review only the most basic
ideas. Polypeptides have evolved to code for all of the chemical
information necessary to spontaneously fold into a stable, unique
three-dimensional structure [20–22]. It logically follows that the
folding reaction can be driven by the interaction of two peptides
that together contain the entire sequence, and in the correct order
of a single peptide that will fold. This was demonstrated in the
classic experiments of Richards [23] and Taniuchi and Anfinsen
[24]. In practice this does not easily work since the major driving
Mapping Biochemical Networks with PCA 469
a Interaction-directed folding b Weakly associating

from protein fragments (PCA) subunits
Fig. 1 Two alternative strategies to achieve complementation. (a) The PCA

strategy requires that unnatural peptide fragments be chosen that are unfolded
prior to association of fused interacting proteins. This prevents spontaneous
association of the fragments (pathway X) that can lead to a false signal. (b)
Naturally occurring subunits that are already capable of folding can be mutated
to interact with lower affinity. However, to some extent, this will always occur,
requiring the selection of cells that express protein partner fusions at low enough
levels that background is not detected
force for protein folding is the hydrophobic effect, so also is

nonspecific aggregation of unfolded peptides. However, if one
adds soluble interacting proteins to the fragments that by interact-
ing increase the effective concentration of the fragments, correct
folding could be favored over any other non-productive process
[25–27]. If the protein that folds from its constitutive fragments is
an enzyme, whose activity could be detected in vivo, then the
reconstitution of its activity can be used as a measure of interaction
of the interacting proteins (Fig. 1a). Furthermore, this binary, all or
none folding event, provides for a very specific measure of protein
interactions dependent on not mere proximity, but the absolute
requirement that the peptides must be organized precisely in space
to allow for folding of the enzyme from the cognate fragments. We
select proteins to dissect into fragments that are not capable of
spontaneously folding from their complementary fragment into a
functional and complete protein. These facts distinguish the PCA
strategy from complementation of naturally occurring and weakly
associating subunits of enzymes [28], in which some spontaneous
assembly occurs, as illustrated in Fig. 1b.
A number of PCAs have been developed based on dominant-
selection, colorimetric, fluorescent, or luminescent outputs. In this
chapter, we discuss three different PCAs, based on the enzymes
murine dihydrofolate reductase (mDHFR) and TEM ß-lactamase,
and also provide protocols for a PCA based on green fluorescent
protein (GFP).
The DHFR PCA can be used in a variety of applications to

perform both simple survival-selection as readout and simulta-
neously, a fluorescent assay allowing quantitative detection and
the cellular localization of protein interactions can be performed
[29–33]. The ß-lactamase assay can be used as a very sensitive
in vivo or in vitro quantitative detector of protein interactions as,
unlike DHFR, one measures the continuous conversion of sub-
strate to colored or fluorescent product [34, 35]. However it
should be noted that generation of a product by an enzyme does
not guarantee that signal to background would be superior to that
of fixed fluorophore reporters like GFP and fluorescein-conjugated
methotrexate (fMTX) bound to DHFR. Observable signal to back-
ground depends, for example, on the quantum yield of the fluor-
ophore, retention of fluorophore by a cell, the optical properties of
the cells used, and the extent to which fluorophores are retained in
individual cellular compartments. For instance, in spite of no enzy-
matic amplification, the DHFR fluorescence assay requires between
only 1,000 and 3,000 molecules of reconstituted DHFR to clearly
distinguish a positive response from background.
Reconstitution of DHFR activity can be monitored in vivo by
cell survival in DHFR-negative cells (CHO-DUKX-B11, for exam-
ple) grown in the absence of nucleotides. The principle of the
DHFR PCA survival assay is that cells simultaneously expressing
complementary fragments of DHFR fused to interacting proteins
or peptides will survive in media depleted of nucleotides. This is an
extraordinarily sensitive assay. In mammalian cells, survival is
dependent only on the number of molecules of DHFR reas-
sembled, and we have determined that this number is approxi-
mately 25 molecules of DHFR per cell [29]. The second
approach is a fluorescence assay based on the detection of fMTX
binding to reconstituted DHFR. The basis of the DHFR PCA
fluorescence assay is that complementary fragments of DHFR,
when expressed and reassembled in cells, will bind with high affinity
(Kd ¼ 540 pM) to fMTX in a 1:1 complex. fMTX is retained in
cells by this complex, while the unbound fMTX is actively and
rapidly transported out of the cells [29, 36, 37]. In addition,
binding of fMTX to DHFR results in a 4.5-fold increase in quan-
tum yield. Bound fMTX, and by inference reconstituted DHFR,
can then be monitored by fluorescence microscopy, fluorescence-
activated cell sorting (FACS), or spectroscopy [29–31]. It is impor-
tant to note that, although fMTX binds to DHFR with high
affinity, it does not induce DHFR folding from the fragments in
the PCA. This is because the folding of DHFR from its fragments is
obligatory; if binding of the oligomerization domains does not
induce folding, no binding sites for fMTX are created. Therefore,
the number of complexes observed as measured by number of
fMTX molecules retained in the cell is a direct measure of the
equilibrium number of oligomerization domain complexes formed,
a b c
d e f
g h
Relative intensity
NO
Number of cells
1.0
EPO
EMP1
0.5
0.0
0
10 101 10 2
10 3 10-13 10-11 10-9 10-7
FITC units Concentration EPO
Fig. 2 Applications of the DHFR PCA to detecting the localization of protein

complexes and quantitating protein interactions. (a–c) different protein pairs
showing (a), plasma membrane, (b), cytosol, and (c), whole cell localization in
transiently transfected COS cells. (d–f), localization of a protein complex in plant
protoplasts. Potato protoplasts expressing two proteins implicated in response
to salicylic acid (SA) fused to DHFR fragments. These are NPR1/NIM1-DHFR
(F[1,2]) and TGA2-DHFR (F[3]) examined by fluorescent microscopy in the
presence of fMTX and DAPI (4,6-diamidino-2-phenylindole). (d), A protoplast
that has not been treated with SA or (e), treated with SA shows that that complex
is induced to relocalize from cytosol to nucleus by SA. (f) Nuclear counter-
staining with DAPI in the same protoplast. (g), FACS results of DHFR PCA. CHO
cells expressing the erythropoietin receptor fused to complementary DHFR
fragments. Receptor activation (conformation change) induced by erythropoietin
(EPO) or a peptide agonist (EMP1) lead to an increase in fluorescence.
(h) Dose–response curve for Epo-induced fluorescence as detected by FACS
results in g
independent of binding of fMTX [29]. The other obvious applica-

tion of the DHFR PCA fluorescence assay is in determining the
location in the cell of interactions as illustrated in a number of cell
types (Fig. 2). The GFP assay can be used for this purpose as well,
but has the distinct advantage that no additional fluorophore is
necessary to do this assay. However, readers should be cautioned
that this assay is only appropriate for high-affinity, very stable com-
plexes and applications to studying transient assembly and disas-
sembly of protein complexes is very limited, due to the slow folding
of the protein and maturation of the fluorophore.
ß-Lactamase is strictly a bacterial enzyme and has been geneti-
cally deleted from many standard E. coli strains. It is not present at
all in eukaryotes. Thus, the ß-lactamase PCA can be used
a
48.18
Hydrolysis of CCF2/AM ate 460nm
b
50.00
(fluorescence units/min)
40.00
30.00
c
20.00
11.59
9.44 8.48
10.00
0.00
FRB+ZIP pMT3 CCF2AM ZIP+ZIP
Fig. 3 ß-Lactamase PCA using the fluorescent substrate CCF2/AM. (a) ZIP (GCN4 leucine zipper-forming
sequences) are tested in HEK 293 cells as described in the text. FRB (rapamycin-FKBP binding domain of
FRAP) is used as a negative control. pMT3 is the expression vector alone and ZIP + ZIP is the positive control.
Data recorded in white microtiter plates on a Perkin Elmer HTS 7000 plate reader. (b, c) Fluorescent
micrographs of cells expressing ß-lactamase PCA showing negative (b, FRB + ZIP) or positive (c, ZIP + ZIP)
response
universally in eukaryotic cells and many prokaryotes, without any

intrinsic background. Also, assays are based on catalytic turnover of
substrates with rapid accumulation of product. This enzymatic
amplification should allow for relatively weak molecular interac-
tions to be observed. The assay can be performed simultaneously
or serially in a number of modes, such as the in vitro colorimetric
assay or the in vivo fluorescence assay (Fig. 3) or the survival assay
in bacteria. Assays can be performed independent of the measure-
ment platform and can easily be adapted to high-throughput for-
mats requiring only one pipetting step.
2 Materials
2.1 DHFR PCA 1. 12-well plates, tissue culture treated; six-well plates, tissue
Survival Assay culture treated.
2. Minimum essential medium: alpha medium without ribonu-
cleosides and deoxyribonucleosides (α-MEM).
3. Dialyzed fetal bovine serum (Hyclone, Cat. no: SH30079-03).
4. Adenosine; deoxyadenosine; thymidine.
5. Transfection reagent (e.g., Lipofectamine Plus reagent, Life
Technologies).
6. Trypsin–EDTA.
7. Cloning cylinders.
2.2 DHFR PCA 1. 12-well plates, tissue culture treated.

Fluorescence Assay 2. Dulbecco’s modified Eagle medium (DMEM); minimum
essential medium: alpha medium without ribonucleosides and
deoxyribonucleosides (α-MEM).
3. Cosmic calf serum; dialyzed fetal bovine serum.
Technologies).
5. Fluorescein-conjugated methotrexate (fMTX) (Molecular
Probes).
6. Dulbecco’s Phosphate-Buffered Saline (PBS).
7. Aqueous mounting medium.
8. Trypsin–EDTA.
9. Micro cover glasses, 18 mm circles, No. 2.
10. Microscope slides, glass, 25 75 1.0 mm.
11. 96-well white microtiter plates (Dynex no 7905).
12. Protein quantification reagents (e.g., Bio-Rad protein assay).
2.3 GFP PCA 1. 12-well plates, tissue culture treated.

Fluorescence Assay 2. Dulbecco’s modified Eagle medium (DMEM).
3. Cosmic calf serum.
Technologies).
6. Aqueous mounting medium.
7. Trypsin–EDTA.
8. Micro cover glasses, 18 mm circles, No 2.
9. Microscope slides, glass, 25 75 1.0 mm.
10. 96-well black microtiter plates (Dynex no 7805).
11. Protein quantification reagents (e.g., Bio-Rad protein assay).
2.4 ß-Lactamase 1. 12-well plates, tissue culture treated.

PCA Colorimetric 2. Dulbecco’s modified Eagle medium (DMEM).
Assay
4. Transfection reagent (e.g., FuGENE 6).
5. Trypsin–EDTA.
6. 100 mM Phosphate Buffer.
7. Nitrocefin.
8. 96-well plates.
2.5 ß-Lactamase 1. 12-well plates, tissue culture treated.

PCA Fluorometric 2. Dulbecco’s modified Eagle medium (DMEM).
Assay
4. Transfection reagent (e.g., FuGENE).
5. Trypsin–EDTA.
7. CCF2-AM (kindly provided by Roger Tsien).
8. 96-well white microtiter plates (Dynex no 7905).
9. Normal saline: 140 mM, NaCl, 5 mM KCl, 2 mM CaCl2,
10 mM Hepes, 6 mM sucrose, 10 mM glucose, pH 7.35.
10. Physiological saline solution: 10 mM HEPES, 6 mM sucrose,
10 mM glucose, 140 mM NaCl, 5 mM KCl, 2 mM MgCl2,
2 mM CaCl2, pH 7.35.
11. 15 mm glass coverslip.
3 Methods
3.1 DHFR PCA 1. Twenty-four hours before transfection, plate 1 105 CHO
Survival Assay DUKX-B11 cells (DHFR-negative; could also be done in
other cells lines, see Note 1) in 12-well plates in α-MEM
medium enriched with 10 % dialyzed fetal bovine serum and
supplemented with 10 μg/ml of adenosine, deoxyadenosine,
and thymidine.
2. Co-transfect cells with the PCA fusion partners (see Note 2)
using a transfection reagent (e.g., Lipofectamine Plus reagent)
according to the manufacturer’s instructions.
3. Forty-eight hours after the beginning of the transfection, split
cells at approximately 5 104 in 6-well plates in selective
medium consisting of α-MEM enriched with dialyzed FBS
but without addition of nucleotides (see Notes 3 and 4).
4. Change medium every 3 days. The appearance of distinct colo-
nies usually occurs after 4–10 days of incubation in selective
medium. Colonies are observed only for clones that simulta-
neously express both interacting proteins fused to one or the
other complementary DHFR fragments. Only interacting pro-
teins will be able to achieve normal cell division and colony
formation.
For further analysis of the interacting protein pair:
5. Isolate three to five colonies per interacting partners by trypsi-
nization (trypsin–EDTA) using cloning cylinders and grow
them separately.
6. Select the best expressing clone by immunoblot (Western blot)

or using the DHFR PCA fluorescence assay (see Subhead-
ing 3.2). Amplification of the expressed gene using methotrex-
ate resistance can be done afterwards if desired, to obtain clones
with increased expression [38].
7. Carry out functional analysis of the clone stably expressing the
interacting proteins pair fused to the complementary DHFR
fragments by using the DHFR PCA fluorescence assay.
3.2 DHFR PCA 1. Twenty-four hours before transfection, plate 1 105 COS
Fluorescence Assay cells (this assay can be performed using any other cell line, see
Note 5) on 18 mm circular glass coverslips in 12-well plates in
3.2.1 Fluorescence
DMEM medium enriched with 10 % Cosmic calf serum.
Microscopy
2. Transiently co-transfect cells with the PCA fusion partners (see
Note 2) using a transfection reagent (e.g., Lipofectamine Plus
reagent) according to the manufacturer’s instructions.
3. The next day, change medium and add fluorescein-conjugated
methotrexate (fMTX) to the cells at a final concentration of
10 μM (see Note 6).
For stable cell lines:
For CHO DUKX-B11 cells (or other cell line) stably expres-
sing PCA fusion partners, seed cells to approximately 2 105
on 18 mm glass coverslips in 12-well plates in α-MEM medium
enriched with 10 % dialyzed FBS. The next day, fMTX is added
to the cells at a final concentration of 10 μM.
4. After incubation with fMTX for 22 h at 37 C, remove the
medium and wash the cells with PBS and re-incubate for
15–20 min at 37 C in the culture medium to allow for efflux
of unbound fMTX (see Note 7). Remove medium and wash the
cells four times with cold PBS on ice and finally mount the
coverslips on microscope glass slides with an aqueous mount-
ing medium.
5. Perform fluorescence microscopy on live cells (see Note 8).
These experiments must be performed within 30 min of the
wash procedure. If the negative control (untransfected cells
treated with fMTX) is too fluorescent, the wash procedure
must be modified (see Note 9).
3.2.2 Flow Cytometry Preparation of cells for fluorescence-activated cell sorting (FACS)
Analysis analysis is the same as described for fluorescence microscopy, except
that following the PBS wash (twice in this case), cells are gently
trypsinized (trypsin–EDTA), suspended in 500 μl of cold PBS and
kept on ice prior to flow cytometric analysis within 30 min. Data are
collected on a FACS analyzer with stimulation with an argon laser
tuned to 488 nm with emission recorded through a 525 nm band-
width filter.
3.2.3 Fluorometric Preparation of cells for fluorometric analysis is the same as described
Analysis for fluorescence microscopy, except that following the PBS wash
(twice in this case), cells are gently trypsinized (trypsin–EDTA).
Plates are put on ice and 100 μl of cold PBS is added to the cells.
The total cell suspensions are transferred to 96-well white microti-
ter plates (Dynex) and keep on ice prior to fluorometric analysis.
The assay can be performed on any microtiter plate reader; we use a
Perkin-Elmer HTS 7000 Series Bio Assay Reader in the fluores-
cence mode. The excitation and emission wavelengths for the
fMTX are 497 nm and 516 nm, respectively. Afterward, the data
are normalized to total protein concentration in cell lysates.
3.3 GFP PCA All procedures describe for the DHFR PCA fluorescence assays are
Fluorescence Assay the same for GFP PCA, except that there are no use of fMTX and
no washing steps. The wash procedure is obviously irrelevant in the
case of the GFP PCA where the folded/reassembled protein is a
fluorophore itself.
3.4 In Vitro 1. Twenty-four hours before transfection, plate 1 105 COS or

ß-Lactamase PCA HEK 293 T cells (this assay can be performed using any other
Colorimetric Assay cell line) in 12-well plates in DMEM medium enriched with
10 % Cosmic calf serum.
2. Transiently co-transfect cells with the PCA fusion partners (see
Note 10) using a transfection reagent (e.g., FuGENE 6)
3. Forty-eight hours after transfection, wash cells three times with
cold PBS and resuspended in 300 μl of cold PBS and keep on
ice. Centrifuge cells at 4 C for 30 s, discard the supernatant,
and resuspend cells in 100 μl of cold 100 mM, phosphate
buffer pH 7.4 (ß-lactamase reaction buffer).
4. Freeze in dry ice/ethanol for 10 min and thaw in a water bath
at 37 C for 10 min, then lyse cells with 3 cycles of freeze and
thaw. Remove cell membrane and debris are by centrifugation
at 4 C for 5 min (10,000 g). Collect the supernatant whole
cell lysate and store at 20 C until assays are performed.
5. Perform assays in 96-well microtiter plates. For testing ß-lacta-
mase activity, 100 μl of 100 mM phosphate buffer is allocated
into each well. Add 78 μl of H2O and 2 μl of 10 mM Nitrocefin
(final concentration of 100 μM). Finally, add 20 μl of unfrozen
cell lysate (final buffer concentration of 60 μM).
6. The assays can be performed on any microtiter plate reader; we
use a Perkin-Elmer HTS 7000 Series Bio Assay Reader in the
absorption mode with a 492 nm measurement filter.
3.5 In Vivo 1. Twenty-four hours before transfection, plate 1 105 COS or

Enzymatic Assay HEK293 T cells in 12-well plates in DMEM medium enriched
and Fluorescent with 10 % Cosmic calf serum.
Microscopy with 2. Transiently co-transfect cells with the PCA fusion partners (see
CCF2/AM Note 10) using a transfection reagent (e.g., FuGENE 6)
3. Twenty-four hours after transfection, split cells again to ensure
50 % confluency the following day (1.5 105) (see Note 11).
Split cells either onto 12-well plates for suspension enzymatic
assay or onto 15 mm glass coverslips for fluorescent
microscopy.
4. Forty-eight hours after transfection, wash cells three times with
PBS to remove all traces of serum (see Note 12).
5. Load cells with the following: 1 μM of CCF2/AM diluted into
a physiologic saline solution for 1 h.
For in vivo enzymatic assay:
6. Wash cells washed twice with the physiologic saline and resus-
pend into the same solution. Aliquot 1 106 cells into a 96-
well fluorescence white plate and measure blue fluorescence
with a Perkin Elmer HTS 7000 Series Bio Assay Reader in the
fluorescence Top reading mode with a 409 nm excitation filter
and a 465 emission filter.
For fluorescence microscopy:
7. Wash cells twice with physiologic saline as in step 5, prior to
examination under the microscope (see Note 13).
We have used two substrates to study the ß-lactamase PCA. The
first one is the cephalosporin called Nitrocefin. This substrate is used
in the in vitro colorimetric assay. ß-lactamase has a kcat/km of
1.7 104 mM1 s1. Substrate conversion can be easily
observed by eye; the substrate is yellow in solution while the prod-
uct is a distinct ruby red color. The rate of hydrolysis can be
monitored quantitatively with any spectrophotometer by measuring
the appearance of red at 492 nm. Signal to background, depending
on the mode of measurement can be greater than 30 to 1.
We have also developed an in vivo fluorometric assay using the
substrate CCF2/AM [39, 40]. Although not as good a substrate as
nitrocefin (kcat/km of 1,260 mM1 s1), CCF2/AM has
unique features that make it a useful reagent for in vivo PCA.
First, CCF2/AM contains butyryl, acetyl and acetoxymethyl esters,
allowing diffusion across the plasma membrane where cytoplasmic
esterases catalyze the hydrolysis of its ester functionality releasing
the polyanionic (4 anions) ß-lactamase substrate CCF2. Because of
the negative charge of CCF2, the substrate becomes trapped in the
cell. In the intact substrate, fluorescence resonance energy transfer
(FRET) can occur between a coumarin donor and fluorescein
acceptor pair covalently linked to the cephalosporin core. The

coumarin donor can be excited at 409 nm with emission at
447 nm, which is within the excitation envelope of the fluorescence
acceptor (maximum around 485 nm), leading to remission of green
fluorescence at 535 nm. When ß-lactamase catalyzes hydrolysis of
the substrate the fluorescein moiety is eliminated as a free thiol.
Excitation of the coumarin donor at 409 nm then emits blue
fluorescence at 447 nm whereas the acceptor (fluorescein) is
quenched by the free thiol.
4 Notes
1. Alternatively, recessive selection can be achieved in eukaryotic

cells by using DHFR fragments containing one or more of
several mutations (for example F31S mutation, see below)
that reduce the affinity of refolded DHFR to the anti-folate
drug methotrexate and growing cells in the absence of nucleo-
tides with selection for methotrexate resistance. This would
obviously be necessary in working with mouse ES cells as,
with all eukaryotes, DHFR activity is present.
2. The best orientations of the fusions for the DHFR PCA are:
protein A-DHFR[1,2] + protein B-DHFR[3] or DHFR[1,2]-
protein A + protein B-DHFR[3], where proteins A and B are
the proteins to test for interaction. We typically insert a 10-
amino-acid flexible polypeptide linker consisting of (Gly.Gly.
Gly.Gly.Ser)2 between the protein of interest and the DHFR
fragment (for both fusions). DHFR[1,2] corresponds to amino
acids 1–105, and DHFR[3] corresponds to amino acids
106–186 of murine DHFR. The DHFR[1,2] fragment that
we use also contains a phenylalanine to serine mutation at
position 31 (F31S), rendering the reconstituted DHFR resis-
tant to methotrexate (MTX) treatment.
3. It is crucial that cell density is kept to a minimum and cells are
well separated when split to avoid cells “harvesting” nutrients
from adjacent cells on dense plates, or colonies might appear to
be forming from clumps of cells that were not sufficiently
separated during the splitting procedure.
4. The choice of dialyzed FBS manufacturer is crucial. Cells need
very little nucleotide in the medium to propagate and this will
result in false positives. The Hyclone dialyzed FBS has proven a
particularly reliable source.
5. The fluorescence DHFR PCA assay is universal and in theory
can be used in any cell type or organism. This assay has already
been shown to work in several mammalian cell lines as well as in
plant cells and insect cells.
6. A stock solution of 1 mM fMTX should be prepared as follows:

Dissolve 1 mg of fMTX in 1 ml of dimethyl formamide (DMF).
To facilitate the dissolution, incubate 15 min at 37 C and mix
by vortexing every 5 min. Protect the tube from light. Store at
20 C.
7. Complementary fragments of DHFR fused to interacting pro-
tein partners, when expressed and reassembled in cells, will
bind with high affinity (Kd ¼ 540 pM) to fMTX in a 1:1
complex. fMTX is retained in cells by this complex, while the
unbound fMTX is actively and rapidly transported out of the
cells.
8. All of the work reported to date has been performed in live
cells. While cells can be fixed, there is a significant reduction in
observable fluorescence.
9. Particular attention must be given to optimizing the fMTX
load and “wash” procedures. Important variables include the
time of loading, temperatures at which each wash step is per-
formed, the number and length of wash steps and the time
between washing and visualization. Too little washing will
mean that background cannot be distinguished from a positive
result. One should scrutinize the relevant parameters in the
same sense as one would for say, a Western blot. Results may
also vary with the way the cells are plated and the types of cells
used. Generally, as in other fluorescent microscopy procedures,
the shape of cells and the localization of the fluorophore will
result in better or worse results. For stable cell lines, the inten-
sity of fluorescence will also depend on the levels of expression
of the fusion proteins. The loading times and concentrations of
fMTX (22 h, 10 μM) used may result in a nonspecific and
punctate fluorescence that is observed with any filter set. We
do not know the source of this background, but it should not
be mistaken for the real fluorescence signal produced by the
PCA, which should be observed strictly with a filter that is
optimal for observation of fluorescein. We have observed that
loading fMTX for between 2 and 5 h at lower (5 μM) concen-
trations prevents this nonspecific signal, although fewer cells
are labeled. Loading times and concentrations must be opti-
mized for specific cell types.
10. The best orientations of the fusions for the ß-lactamase PCA are:
protein A-BLF[1] + protein B-BLF[2] or BLF[1]-protein A +
protein B-BLF[2], where proteins A and B are the proteins to
test for interaction. We typically insert a 15-amino-acid flexible
polypeptide linker consisting of (Gly.Gly.Gly.Gly.Ser)3 between
the protein of interest and the ß-lactamase fragment (for both
fusions). BLF[1] corresponds to amino acids 26–196 (Ambler
numbering), and BLF[2] corresponds to amino acids 198–290
of TEM-1 ß-lactamase.
11. The maximum loading efficiency of CCF2-AM is observed at

50 % confluence.
12. Serum may contain esterases that can destroy the substrate.
13. We perform fluorescence microscopy on live HEK 293 or COS
cells with an inverse Nikon Eclipse TE-200 (objective plan
fluor 40 dry, numerically open at 0.75). Images were taken
with a digital CCD cooled (50 C) camera, model Orca-II
(Hamamatsu Photonics (exposure for 1 s, binning of 2 2
and digitalization 14 bits at 1.25 MHz). Source of light is a
Xenon lamp Model DG4 (Sutter Instruments). Emission filters
are changed by an emission filter switcher (model Quanto-
scope) (Stanford Photonics). Images are visualized with ISee
software (Inovision Corporation) on an O2 Silicon Graphics
computer. The following selected filters are used: Filter set
#31016 (Chroma Technologies); Excitation filter: 405 nm
(passing band of 20 nm); Dichroic Mirror: 425 nm DCLP;
Emission filter #1: 460 nm (passing band of 50 nm); Emission
filter #2: 515 nm (passing band of 20 nm).
References
1. Drees BL (1999) Progress and variations in 9. Hu CD, Kerppola TK (2003) Simultaneous
two-hybrid and three-hybrid technologies. visualization of multiple protein interactions
Curr Opin Chem Biol 3:64–70 in living cells using multicolor fluorescence
2. Evangelista C, Lockshon D, Fields S (1996) complementation analysis. Nat Biotechnol
The yeast two-hybrid system: prospects for 21:539–545
protein linkage maps. Trends Cell Biol 10. Paulmurugan R, Gambhir SS (2003) Monitor-
6:196–199 ing protein-protein interactions using split syn-
3. Fields S, Song O (1989) A novel genetic system thetic renilla luciferase protein-fragment-
to detect protein-protein interactions. Nature assisted complementation. Anal Chem
340:245–246 75:1584–1589
4. Vidal M, Legrain P (1999) Yeast forward and 11. Luker KE, Smith MC, Luker GD et al (2004)
reverse ‘n’-hybrid systems. Nucleic Acids Res Kinetics of regulated protein-protein interac-
27:919–929 tions revealed with firefly luciferase comple-
5. Walhout AJ, Sordella R, Lu X et al (2000) mentation imaging in cells and living animals.
Protein interaction mapping in C. elegans Proc Natl Acad Sci U S A 101:12288–12293
using proteins involved in vulval development. 12. Magliery TJ, Wilson CG, Pan W et al (2005)
Science 287:116–122 Detecting protein-protein interactions with a
6. Uetz P, Giot L, Cagney G et al (2000) A com- green fluorescent protein fragment reassembly
prehensive analysis of protein-protein interac- trap: scope and mechanism. J Am Chem Soc
tions in Saccharomyces cerevisiae. Nature 127:146–157
403:623–627 13. Remy I, Michnick SW (2006) A highly sensi-
7. Michnick SW, Remy I, Campbell-Valois FX tive protein-protein interaction assay based on
et al (2000) Detection of protein-protein inter- Gaussia luciferase. Nat Methods 3:977–979
actions by protein fragment complementation 14. Wehr MC, Laage R, Bolz U et al (2006) Mon-
strategies. Methods Enzymol 328:208–230 itoring regulated protein-protein interactions
8. Paulmurugan R, Umezawa Y, Gambhir SS using split TEV. Nat Methods 3:985–993
(2002) Noninvasive imaging of protein- 15. Michnick SW, Ear PH, Manderson EN et al
protein interactions in living subjects by using (2007) Universal strategies in research and
reporter protein complementation and recon- drug discovery based on protein-fragment
stitution strategies. Proc Natl Acad Sci U S A complementation assays. Nat Rev Drug Discov
99:15608–15613 6:569–582
16. Remy I, Michnick SW (2007) Application of 29. Remy I, Michnick SW (1999) Clonal selection
protein-fragment complementation assays in and in vivo quantitation of protein interactions
cell biology. Biotechniques 42:137, 139, 141 with protein-fragment complementation
passim assays. Proc Natl Acad Sci U S A
17. Stefan E, Aquin S, Berger N et al (2007) Quan- 96:5394–5399
tification of dynamic protein complexes using 30. Remy I, Wilson IA, Michnick SW (1999)
Renilla luciferase fragment complementation Erythropoietin receptor activation by a
applied to protein kinase A activities in vivo. ligand-induced conformation change. Science
Proc Natl Acad Sci U S A 104:16916–16921 283:990–993
18. Michnick SW, Ear PH, Landry C et al (2010) A 31. Remy I, Michnick SW (2001) Visualization of
toolkit of protein-fragment complementation biochemical networks in living cells. Proc Natl
assays for studying and dissecting large-scale Acad Sci U S A 98:7678–7683
and dynamic protein-protein interactions in liv- 32. Remy I, Campbell-Valois FX, Michnick SW
ing cells. Methods Enzymol 470:335–368 (2007) Detection of protein-protein interac-
19. Michnick SW, Ear PH, Landry C et al (2011) tions using a simple survival protein-fragment
Protein-fragment complementation assays for complementation assay based on the enzyme
large-scale analysis, functional dissection and dihydrofolate reductase. Nat Protoc
dynamic studies of protein-protein interactions 2:2120–2125
in living cells. Methods Mol Biol 756:395–425 33. Tarassov K, Messier V, Landry CR et al (2008)
20. Anfinsen CB, Haber E, Sela M et al (1961) The An in vivo map of the yeast protein interac-
kinetics of formation of native ribonuclease tome. Science 320:1465–1470
during oxidation of the reduced polypeptide 34. Galarneau A, Primeau M, Trudeau LE et al
chain. Proc Natl Acad Sci U S A 47:1309–1314 (2002) Beta-lactamase protein fragment com-
21. Anfinsen CB (1973) Principles that govern the plementation assays as in vivo and in vitro sen-
folding of protein chains. Science sors of protein protein interactions. Nat
181:223–230 Biotechnol 20:619–622
22. Gutte B, Merrifield RB (1971) The synthesis of 35. Remy I, Ghaddar G, Michnick SW (2007)
ribonuclease A. J Biol Chem 246:1922–1941 Using the beta-lactamase protein-fragment
23. Richards FM (1958) On the Enzymic Activity complementation assay to probe dynamic
of Subtilisin-Modified Ribonuclease. Proc Natl protein-protein interactions. Nat Protoc
Acad Sci U S A 44:162–166 2:2302–2306
24. Taniuchi H, Anfinsen CB (1971) Simultaneous 36. Israel DI, Kaufman RJ (1993) Dexamethasone
formation of two alternative enzymology active negatively regulates the activity of a chimeric
structures by complementation of two overlap- dihydrofolate reductase/glucocorticoid recep-
ping fragments of staphylococcal nuclease. J tor protein. Proc Natl Acad Sci U S A
Biol Chem 246:2291–2301 90:4290–4294
25. Pelletier JN, Campbell-Valois FX, Michnick 37. Kaufman RJ, Bertino JR, Schimke RT (1978)
SW (1998) Oligomerization domain-directed Quantitation of dihydrofolate reductase in
reassembly of active dihydrofolate reductase individual parental and methotrexate-
from rationally designed fragments. Proc Natl resistant murine cells. Use of a fluorescence
Acad Sci U S A 95:12141–12146 activated cell sorter. J Biol Chem
26. Pelletier JN, Michnick SW (1997) A protein 253:5852–5860
complementation assay for detection of 38. Kaufman RJ (1990) Selection and coamplifica-
protein-protein interactions in vivo. Protein tion of heterologous genes in mammalian cells.
Eng 10:89 Methods Enzymol 185:537–566
27. Johnsson N, Varshavsky A (1994) Split ubiqui- 39. Zlokarnik G, Negulescu PA, Knapp TE et al
tin as a sensor of protein interactions in vivo. (1998) Quantitation of transcription and
Proc Natl Acad Sci U S A 91:10340–10344 clonal selection of single living cells with beta-
28. Rossi F, Charlton CA, Blau HM (1997) Moni- lactamase as reporter. Science 279:84–88
toring protein-protein interactions in intact 40. Zlokarnik G (2000) Fusions to beta-
eukaryotic cells by beta-galactosidase comple- lactamase as a reporter for gene expression
mentation. Proc Natl Acad Sci U S A in live mammalian cells. Methods Enzymol
94:8405–8410 326:221–244
Chapter 32
Detection of Protein-Protein Interaction Using Bimolecular

Fluorescence Complementation Assay
Cau D. Pham
Abstract
The bimolecular fluorescence complementation (BiFC) assay is a versatile technique for investigating
protein-protein interaction (PPI) in living systems. The BiFC assay exploits the color-emitting moiety
and the modular structure of fluorescent proteins to provide both temporal and spatial information of the
PPI. The modular property of fluorescent proteins enables researchers to strategically partition a fluorescent
protein into two nonfluorescent units, which can be independently fused to other proteins. When the
fusion proteins interact with each other, the nonfluorescent fragments reconstitute to generate a fluores-
cence signal. PPI can then be detected by capturing the fluorescence signal with a fluorescence microscope.
In this chapter, the Venus fluorescent protein is employed to demonstrate the application of the BiFC assay.
Key words Fluorescent proteins, BiFC technology, Protein-protein interaction, Protocol
1 Introduction
In 1962, the green fluorescent protein (GFP) of the Aequorea

victoria jellyfish was the first protein identified to possess an intrin-
sic bioluminescence property [1]. Five decades later, more than 40
different species of fluorescent proteins are available for various
applications with color spanning beyond the visible spectrum
[2, 3]. Despite growing in number, fluorescent proteins share
similar molecular weight and backbone features such as a barrel-
like structure comprising of 11 β-sheets and an α-helix filament that
traverses through the entire barrel at the center. The β-sheet barrel
and the α-helix structure are essential for fluorescent protein matu-
ration or chromophore formation, a necessary step for color-
emission. Chromophore formation also requires oxygen, which
facilitates the modification of at least four highly conserved amino
acids on the fluorescent proteins: the tyrosine at position 66, the
glycine at position 67, the arginine at position 96, and the gluta-
mate at position 222 [3, 4].
483
484 Cau D. Pham
Although GFP was discovered in 1962, it took three more

decades for scientists to figure out how to harness the unique
properties of GFP. In addition to being able to generate fluores-
cence signal, fluorescent proteins are nontoxic and their maturation
is not dependent on any particular host [2, 3]. Fluorescent proteins
can also be modified and/or linked to another protein without
sacrificing their fluorescence signal. These traits enable scientists
to utilize them as reporter proteins in different biological systems.
In 1992, full length GFP was first cloned and used as a reporter
protein to study gene expression in bacteria and nematodes [2].
From that time on, full length fluorescent proteins have been
widely used to visualize protein expression, protein localization,
and protein complexes in live cells [3, 4]. Furthermore, the bimo-
lecular fluorescence complementary (BiFC) assay, a newly devel-
oped fluorescent protein-based method, has enabled scientists to
visualize protein-protein interaction (PPI) in living systems [5–7].
The BiFC assay involves splitting a fluorescent protein into two
fragments that are incapable of generating fluorescence signal on
their own [5]. However, these fragments can reconstitute and
produce fluorescence signal if they are brought together such as
when they are fused to two proteins that interact with one another
(Fig. 1). BiFC technology was successfully demonstrated by Hu
et al. in 2002 by tethering Fos to the C-terminal fragment and Jun
to the N-terminal fragment of the enhanced yellow fluorescent
protein (EYFP) [5]. In addition to YFP, many other fluorescent
proteins and multiple splitting positions have also been successfully
developed for BiFC (Table 1).
The ability to examine PPI in living systems is essential to our
understanding of biological processes [9, 10]. Toward this end, the
BiFC assay holds several key advantages over other PPI detection
methods such as yeast two-hybrid and co-immunoprecipitation
[11, 12]. Firstly, the BiFC assay allows real-time visualization of
PPI with minimal perturbation to their natural environment, such
Fig. 1 The principle of Venus BiFC technology. The Venus fluorescent protein is strategically split into two
nonfluorescent fragments. The reconstitution of Venus protein and signal from the nonfluorescent constituents
can be mediated by two interacting proteins. Venus protein structure was drawn using the I-TASSER program [8]
Table 1
Fluorescent proteins commonly utilized for BiFC technology
Brightness Oligomerization
Color Percentage half-lifeb (in of fully mature
Protein Organism Ex/Em class of brightnessa seconds) protein BiFC split position Reference
CFP Aequorea 433/475 Cyan 19 64 Monomer 154–155;172–173 [2, 6]
victoria
Cerulean Aequorea 433/475 Cyan 39 36 Weak dimer 154–155;172–173 [2, 6]
victoria
GFP Aequorea 485/500 Green 49 174 Weak dimer 157–158 [2, 6]
victoria
YFP Aequorea 514/527 Yellow 74 60 Weak dimer 154–155;172–173 [2, 6]
victoria
Venus Aequorea 515/528 Yellow 76 15 Weak dimer 154–155;172–173 [2, 6]
victoria
Citrine Aequorea 516/529 Yellow 85 49 Monomer 154–155;172–173 [2, 6]
victoria
mRFP1-Q66T Discosoma sp. 549/570 Red ND ND ND 154–155;168–169 [2, 6]
BiFC Assay for PPI Detection

mCherry Discosoma sp. 587/610 Red 23 96 Monomer 159–160 [2, 6]
A summary of the BiFC systems that have been developed in the last 10 years
a
Percentage of brightness of intact fluorescent protein relative to fluorescein (100 %)
b
The Brightness half-life was calculated by measuring the time it took to bleach the emission rate of 1,000 photon/s down to 500 photon/s [2]
485
486 Cau D. Pham
as in live cells or animals [11, 12]. Secondly, the BiFC assay can
provide valuable information regarding the sub-cellular location of
the PPI [12]. Such information has good predictive value about the
functional roles of the interaction. These and other aforementioned
attributes are the reason why BiFC is a valuable tool to have in one’s
molecular cabinet. In this chapter, one will learn how to detect PPI
in mammalian cells using Venus BiFC.
2 Materials
1. Thermocycler.
2. DNA polymerase and buffer (e.g., Phusion, New England
BioLabs).
3. P38 and MKK3 cDNAs (Open Biosystems).
4. Gateway® entry vector (Life Technologies).
5. TE (10 mM Tris–HCl, 1 mM EDTA) pH 8.0.
6. BP Clonase™ II enzyme mix (Life Technologies).
7. Shaker-incubator and water bath.
8. Competent bacteria (e.g., NEB10β).
9. Bacterial growth media including S.O.C and LB. LB medium is
supplemented with the appropriate antibiotic (100 μg/mL
Ampicillin, 50 μg/mL Kanamycin).
10. Mini-prep/midi-prep kits.
11. BsrGI restriction endonuclease and buffer.
12. Agarose gel electrophoresis reagents (molecular grade agarose,
Tris-Borate-EDTA buffer, ethidium bromide).
13. Gene-specific oligonucleotide primers for cloning and
sequencing.
14. pSCM167-NV and pDEST26-CV mammalian expression vec-
tors (developed in the lab of Dr. Haian Fu, Emory University).
15. LR Clonase™ II enzyme mix (Life Technologies).
16. A nucleic acid spectrophotometer (e.g., NanoDrop).
17. Cell growth medium for the specific cell line (e.g., DMEM
supplemented with 5 % FBS and 1 % Penicillin–Streptomycin).
18. Cell line that can be transfected (e.g., the LN229 cell line).
19. Cell culture incubator with CO2 supply.
20. 0.05 % Trypsin/0.53 mM EDTA.
21. Opti-MEM.
22. Transfection reagent (e.g., FuGENE® HD transfection
reagent, Promega).
23. Fluorescence microscope and filters.
BiFC Assay for PPI Detection 487
3 Methods
Described below are (1) the steps to create BiFC vectors and (2)
how to examine PPI in mammalian cells (see Note 1). A previously
known interacting protein pair MKK3 and P38 [13] will be used to
demonstrate the use of a BiFC assay to monitor PPI. The BiFC
reporter used in this example is the Venus fluorescent protein, a
variant of EYFP (see Note 2). Venus fluoresces brighter, matures
rapidly at 37 C, and is less sensitive to environmental conditions
[3, 14, 15]. Both Venus fragments are in the Gateway® vector
system (Life Technologies) and will generate N-terminal fusion
proteins (the Venus fragments fuse to the N-terminus of the pro-
tein of interest) (Fig. 2). The N-terminal fragment of Venus (NV),
Fig. 2 Gateway® cloning technology and a schematic of the Venus BiFC constructs. (a) Gateway® vectors
allow the shuttling of DNA fragments from one vector to another compatible vector via site-specific homolo-
gous recombination catalyzed by recombinase enzymes. (b) N-terminal fusion constructs regulated by the
CMV promoter. For these vectors, the Venus fragments are attached to the protein of interest through a
(2xGGGS) linker sequence
488 Cau D. Pham
Fig. 3 The interaction between P38 and MKK3 was used to demonstrate the Venus BiFC assay. Cells that
expressed both P38 and MKK3 proteins fused to compatible Venus fragments exhibit strong Venus signals.
The Venus signal in cells that expressed only one fusion protein along with its complementary Venus fragment
was more than two times lower
residues 1–173, along with a linker (see Note 3), was inserted into
the pSCM167 vector (Haian Fu lab) to generate the pSCM167-NV
expression vector (see Note 4). The C-terminal fragment of Venus
(CV), residues 156–239, along with a linker, was inserted into the
pDEST26 vector (Life Technologies) to make the pDEST26-CV
expression vector (see Note 4).
3.1 Creating BiFC Gateway® cloning technology can be used to clone DNA fragments
Constructs for PPI into the Venus BiFC expression vectors (as an example, P38 and
Detection MKK3 were cloned into pSCM167-NV and pDEST26-CV, respec-
tively) (Figs. 2 and 3). Gateway® technology employs a two-vector
system, an entry and destination vector, to shuttle DNA fragments
between vectors. To utilize this technology, a DNA fragment is first
inserted into an entry vector. From the entry vector, the DNA
fragment can be transferred to Gateway® destination/expression
vectors via site-specific homologous recombination.
3.1.1 Cloning of DNA 1. Use a polymerase chain reaction (PCR) to add attB sites to the
Fragments into a 50 and 30 ends of the DNA fragment as described by the
Gateway® Entry Vector manufacturer (Life Technologies) (see Note 5). For example,
attB sequences {50 -GGG GAC AAG TTT GTA CAA AAA
AGC AGG CTT C-30 and 50 -GGG GAC CAC TTT GTA
CAA GAA AGC TGG GTC TTA CTA-30 } were added to the
50 and 30 , respectively, of both P38 and MKK3. The attB
sequences facilitate the insertion of DNA fragments into the
Gateway® entry vectors via site-specific homologous
recombination.
2. Perform the cloning of the DNA fragment, flanked by the attB
sites, into the Gateway® entry vector (in this case pDONR201)
using BP clonase enzyme and bacterial transformation as sug-
gested by the manufacturer’s instruction (with minor
modifications as described below). First, combine purified

attB-PCR product (15–150 ng) and pDONR201 vector
(150 ng) in a 1.5 mL microcentrifuge tube. Next, if necessary,
adjust the reaction volume to 8 μL with TE buffer. Then, add
2 μL of BP Clonase™ II enzyme mix to the reaction, vortex to
mix, brief centrifugation, and incubate the reaction at room
temperature (25 C) for 1 h. Finally, stop the reaction by
adding 1 μL of proteinase K, vortex to mix, and incubate at
37 C for 10 min. The BP reaction is now ready for bacterial
transformation.
3. Bacterial transformation is achieved by adding 2 μL of the BP
reaction to 20 μL of freshly thawed competent bacteria. Gently
tap the tube to mix the reaction and follow with a 30 min
incubation on ice. Next, perform a 30 s heat-shock by incubating
the reaction in a 42 C water bath. Then, place the reaction back
on ice and add 250 μL of S.O.C medium. Transfer the reaction to
a 37 C shaker-incubator and incubate (with shaking) for 1 h.
Finally, spread about 100–200 μL of the transformation mixture
onto a LB plate that contains 50 μg/mL of Kanamycin and
incubate the plate overnight at 37 C (see Note 6).
4. To determine the correct clone, pick a few colonies from the
transformation plate and inoculate each of them in 4 mL of LB
broth containing 50 μg/mL of Kanamycin. Allow the bacteria
to populate overnight at 37 C in a shaker-incubator. Next, use
a miniprep kit (follow the manufacturer’s protocol) to extract
the plasmid DNA from the bacteria. Then, perform BsrGI
restriction digestion (one BsrGI restriction site is present at
each att site) of the plasmid followed by agarose gel electro-
phoresis analysis to verify the insertion of the DNA fragment
into the entry vector. Finally, perform sequence analysis to
check the fidelity and the orientation of the inserted DNA
fragment.
3.1.2 Cloning of the DNA Both the pSCM167-NV and the pDEST26-CV are Gateway®-
Fragment into the based expression vectors. This enables the transfer of DNA frag-
pSCM167-NV and ments from entry vectors (such as pDONR201) directly into the
pDEST26-CV Vectors pSCM167-NV and pDEST26-CV vectors via site-specific homolo-
gous recombination. The transferring of a DNA fragment from
Gateway® entry vectors into Gateway® destination vectors is cata-
lyzed by the LR Clonase™ enzyme and is performed as instructed
by the manufacturer (with minor modifications).
1. First, combine the purified entry vector (15–150 ng) and
150 ng of the destination vector (pSCM167-NV or
pDEST26-CV) in a 1.5 mL microcentrifuge tube. Next, if
necessary, adjust the reaction volume to 8 μL with TE buffer.
Then, add 2 μL of LR Clonase™ II enzyme mix to the reaction,
490 Cau D. Pham
vortex to mix, briefly centrifuge, and incubate the reaction at

room temperature (25 C) for 1 h. Finally, stop the reaction by
adding 1 μL of proteinase K, vortex to mix, and incubate at
37 C for 10 min. The LR reaction is now ready for bacterial
transformation.
2. Transform competent bacteria, as described above, with the
product of the LR reaction. Positive clones can be selected by
plating the bacteria on LB plates containing 100 μg/mL of
Ampicillin. Finally, perform BsrGI restriction digestion and
DNA sequence analysis to verify the insertion of the DNA
fragment in the destination vector (see Note 7).
3. Vectors with the correct fusion of the DNA fragment to the
respective Venus fragments can then be used in subsequent
BiFC assays.
3.2 Performing In order to perform the BiFC assay, cells must uptake the expres-
the BiFC Assay sion vectors. Prior knowledge about the model system is essential
(see Note 1). Prior to perform BiFC assay, one needs to confirm the
expression level of the fusion proteins. This can be determined by
Western blot (see Note 8). One also needs to consider checking the
folding of the proteins of interest. Incorrect folding of the proteins
of interest caused by the fused fluorescent protein fragments could
affect the stability and localization of the protein [11] (see Note 9).
Localization of fusion proteins and untagged wild-type proteins
can be assessed by indirect immunofluorescence assay [11].
3.2.1 Transfecting the As an example, pSCM167-NV-P38 and pDEST26-CV-MKK3

Expression Vectors into a vectors were transiently co-transfected into LN229 cells using
Cell Line FuGENE® HD transfection reagent. Likewise, pSCM167-NV-
P38 and pDEST26-CV as well as pDEST26-CV-MKK3 and
pSCM167-NV vector combinations were also co-transfected into
the LN229 cell line to serve as negative controls (see Note 10).
In addition to negative controls, one should also consider a positive
control and a transfection control to monitor PPI and transfection
efficiency, respectively (see Note 10). The transfection protocol
below was adapted from the FuGENE® HD manufacturer’s
instructions and was optimized for the LN229 cell line. All plasmid
DNAs were prepared using Qiagen midi-prep kit as described in the
manufacturer’s protocol. DNAs were quantified using a NanoDrop
(see Note 11).
3.2.2 Cell Growth In this example, low passage LN229 cells were cultured in DMEM
Conditions, Cell Seeding, supplemented with 5 % FBS and 1 % Penicillin/Streptomycin
and Cell Transfection (please note that the cell culture and transfection conditions will
need to be adjusted for the cell lines being used). Cells were kept in
a water-jacketed incubator at 37 C and supplemented with 5 %
CO2 humidity. For transfection, the following steps were taken.
1. Pre-culture 25 mL of LN229 cells in a 75 mL flask until the

growing surface of the flask is about 90 % covered with cells.
2. Harvest the cells by trypsinization.
3. Completely remove all growth medium.
4. Add 1.5 mL of 0.05 % Trypsin/0.53 mM EDTA.
5. Incubate at room temperature for 3–5 min or until most cells
have detached from the surface of the plate.
6. Add 10 mL of cell growth medium to the flask to deactivate the
trypsin (see Note 12).
7. Pipet the cells eight to ten times with a 10 mL transfer pipet to
dissociate the cells, and determine the cell concentration using
a hemocytometer.
8. Prepare a six-well plate of cells for transfection by adding 2 mL
(300,000 cell/mL) of trypsinized cells, in growth medium, to
each well. Then, gently swirl the plate to evenly disperse the
cells around the wells.
9. Allow the cells to cover about 60 % of the well’s surface by
keeping the plate at 37 C and 5 % CO2 humidity in a water-
jacketed incubator (usually takes about 18 h). When the cells
are about 60 % confluent, they are ready for transfection.
10. Prepare the transfection reaction by performing the following
steps. First, transfer 200 μL of Opti-MEM to a 1.5 mL sterile
microcentrifuge tube. Next, add 320 ng of the pSCM167-NV
and 320 ng of the pDEST26-CV expression vectors (see Note
13). Gently tap the tube to mix the DNAs. Finally, add 1.92 μL
of transfection reagent directly to the solution (see Note 14)
and gently tap the tube to mix the reaction.
11. Incubate the reaction at room temperature (25 C) for 15 min.
12. Use a micropipettor and slowly dispense 100 μL of the trans-
fection reaction drop-by-drop over the whole well.
13. Gently swirl the plate to facilitate diffusion of the transfection
reagent.
14. Place the plate back into the cell culture incubator.
3.2.3 Detecting PPI with Proteins that show strong interaction can be detected 24 h post-
a Fluorescence Microscope transfection (see Note 15). Once the cells have been transfected
with the appropriate plasmids, PPI can be observed using a fluores-
cence microscope with the correct filter. The YFP filter set can be
used to detect Venus signal. As demonstrated in Fig. 3, strong
Venus signal was observed in cells that were co-transfected with
the pSCM167-NV-P38 and the pDEST26-CV-MKK3 vectors.
Noticeably, a low Venus signal was also observed in cells that were
co-transfected with control vectors; the pSCM167-NV-P38 and
pDEST26-CV pair as well as pDEST26-CV-MKK3 and
492 Cau D. Pham
pSCM167-NV pair (see Notes 10 and 16). However, the signals

from these cells were much lower than the cells that harbored both
the P38 and MKK3 fusion proteins (Fig. 3). For a novel protein
interaction pair, the interaction can be further validated using a
co-immunoprecipitation assay or tag pull-down assay with GST or
FLAG tags.
4 Notes
1. The model system must able to uptake foreign DNA such as

plasmids. Also keep in mind that fluorescent protein matura-
tion requires oxygen. Therefore, this assay will not work in
organisms that are sensitive to oxygen or in an environment
that lacks oxygen. Temperature can also affect the maturation
process of fluorescent proteins. For example, YFP-BiFC opti-
mal maturation temperature is 30 C while other fluorescent
proteins typically mature at 37 C [11].
2. Keep in mind that the signal from the reconstituted fluorescent
proteins is not as bright as intact fluorescent proteins. For weak
interacting proteins, it is best to use a strong fluorescence signal
BiFC system such as Venus.
3. A linker is a short peptide that connects the fluorescent protein
fragment to the protein of interest. The linker provides limited
independent movement between the fluorescent protein frag-
ment and its fusion protein. The length and the amino acid
sequence of the linkers have been known to affect PPI detec-
tion. Linker sequences such as RSIAT, RPACK-
IPNDLKQKVMNH, AAANSSIDLISVPVDSR, and (GGGS)
n have been successfully employed by many groups [11].
The GGGSGGGS linker sequence was used for the example
experiment.
4. When choosing expression vectors, make sure that the vector
system is compatible with model system (i.e., replication, pro-
moter, etc.). Both pSCM167-NV and pDEST26-CV are
expression vectors for mammalian cells and protein expression
is regulated by the CMV promoter. Using these vectors, the
proteins of interest will be fused to the Venus fragments at their
N-terminus. Both Venus fragments have a start codon and no
stop codon. Both vectors have an additional tag. The
pSCM167-NV vector has a FLAG tag while the pDEST26-
CV has a 6XHis tag. These tags don’t interfere with the BiFC
assay, but do provide another platform for detecting PPI
(i.e., FLAG or HIS pull-down).
5. The attB sequences are added to the PCR-primers during
primer synthesis. Make sure that the attB sequences as well as
their orientation are correct when designing primers for

Gateway® cloning. If the DNA fragment is intended for expres-
sion, it is essential to check the reading frame and the presence
of a start as well as a stop codon.
6. According to the manufacturer, Gateway® cloning is highly
efficient. Gateway® vectors (i.e., PDONR201, pSCM167-NV,
and pDEST26-CV) contain the ccdB gene that encodes for a
bacterial toxin (use ccdB resistant bacteria to propagate such
plasmids). The ccdB gene is replaced by the target DNA frag-
ment during the BP and LR clonase reaction. Therefore, one
only needs to check a few colonies to find a clone with the
correct insert. If needed, increasing the incubation time (i.e.,
overnight) of BP and LR reactions can increase the number of
colonies with the desired insert.
7. The sequencing primers should be located within the Venus
fragments and about 100 base-pairs upstream from the protein
of interest. Orientation and reading frame alignment can both
be determined using these primers.
8. The assay will not work if either fusion protein fails to express.
On the other hand, high protein expression of the fusion
proteins could lead to nonspecific interaction [15]. Depending
on the location of the epitope used to generate the antibody, a
GFP antibody can be used to detect both the N- and C-Venus
fragments. Protein specific antibodies can also be used to check
for fusion protein expression.
9. If protein folding or PPI is affected by the fused reporter
fragment(s), the problem can be mitigated by changing the
fusion site and/or reporter fragment. There are 8 possible
fusion combinations for each pair of proteins (e.g.,
NV-proteinA and CV-proteinB, CV-proteinA and
NV-proteinB, NV-proteinA and proteinB-CV). For a novel
pair of proteins, it is imperative to try all eight pairing combina-
tions before deciding that the proteins do not interact.
10. Controls are essential to assess the success of the BiFC assay.
Negative controls are necessary to rule out false-positive,
strong fluorescence signals from a pair of non-interacting pro-
teins. Positive controls allow the researcher to determine
whether the assay worked or did not work. If possible, choose
a protein pair that can serve as positive PPI (wild type proteins)
as well as negative PPI controls (non-interacting mutants). One
can also use the BiFC fragments (such as pSCM167-NV-P38
and pDEST26-CV, pDEST26-CV-MKK3 and pSCM167-NV,
and pDEST26-CV and pSCM167-NV) as negative controls to
monitor self-assembly of the BiFC fragments. To monitor
transfection efficiency, one also needs to consider using an
internal control reporter-plasmid (i.e., full length fluorescent
protein of a different color, luciferase reporter proteins, etc.).
494 Cau D. Pham
11. Good quality DNA (260/280 ratio close to 2.0) is essential for
a successful assay. Poor quality DNA can reduce transfection
efficiency or can be detrimental to the cell.
12. Make sure that FBS is present in the growth medium. FBS
inhibits trypsin activity [16].
13. Too much DNA can be toxic to some cell lines. Moreover, too
much DNA could lead to high background noise [15].
14. Various transfection reagents may be toxic to some cell lines.
If using FuGENE® HD reagent, minimize contact of the
reagent with plastic surface (i.e., microcentrifuge tube’s wall)
because this could lead to a decrease in transfection efficiency.
15. Strong interacting pairs of proteins can be detected as early as
24 h post-transfection. Long incubation times can lead to high
background signal.
16. Self-assembly of the Venus BiFC fragments as well as of other
BiFC fragments is often a drawback for this assay. Self-assembly
of the BiFC fragments is a common contributor of false-
positive results. Self-assembly signals are typically lower than
protein-interaction assisted reconstitution signals for many
PPIs. However, self-assembly signals may pose a problem for
detecting weak PPIs. Replacing certain amino acids in BiFC
systems (i.e., Venus) can reduce self-assembly [14].
Acknowledgements
I would like to thank Drs. Haian Fu and Jonathan Havel for their
generous gifts of the pSCM167-NV and pDEST26-CV Gateway®
destination vectors as well as constructive input to make this assay
work.
References
1. Shimomura O, Johnson FH, Saiga Y (1962) 5. Hu CD, Chinenov Y, Kerppola TK (2002)
Extraction, purification, and properties of Visualization of interactions among bZIP and
aequorin, a bioluminescent protein from the Rel family proteins in living cells using bimo-
luminous hydromedusan, Aequorea. J Cell lecular fluorescence complementation. Mol
Comp Physiol 59:223–239 Cell 9:789–798
2. Shaner NC, Steinbach PA, Tsien RY (2005) A 6. Vidi PA, Watts VJ (2009) Fluorescent and bio-
guide to choosing fluorescent proteins. Nat luminescent protein-fragment complementa-
Methods 2:905–909 tion assays in the study of G protein-coupled
3. Kremers G-J, Gilbert SG, Cranfill PJ, Davidson receptor oligomerization and signaling. Mol
MW, Piston DW (2011) Fluorescent proteins Pharmacol 75:733–739
at a glance. J Cell Sci 124:157–160 7. Ozawa T (2006) Designing split reporter pro-
4. Remington SJ (2006) Fluorescent proteins: teins for analytical tools. Anal Chim Acta
maturation, photochemistry, and photophy- 556:58–68
sics. Curr Opin Struct Biol 16:714–721
8. Zhang Y (2008) I-TASSER server for protein 3D network mapping. Curr Opin Biotechnol
structure prediction. BMC Bioinformatics 9:40 14:610–6177
9. Remy I, Michnick SW (2007) Application of 13. Raingeau J, Whitmarsh AJ, Barrett T, Derijard
protein-fragment complementation assays in B, Davis RJ (1996) MKK3- and MKK6-
cell biology. Biotechniques 42:137–145 regulated gene expression is mediated by the
10. Gandhi TK, Zhong J, Mathivanan S, Karthick p38 mitogen-activated protein kinase signal
L, Chandrika KN, Mohan SS, Sharma S, Pin- transduction pathway. Mol Cell Biol
kert S, Nagaraju S, Periaswamy B, Mishra G, 16:1247–1255
Nandakumar K, Shen B, Deshpande N, Nayak 14. Kodama Y, Hu CD (2010) An improved bimo-
R, Sarker M, Boeke JD, Parmigiani G, Schultz lecular fluorescence complementation assay
J, Bader JS, Pandey A (2006) Analysis of the with a high signal-to-noise ratio. Biotechni-
human protein interactome and comparison ques 49:793–805
with yeast, worm and fly interaction datasets. 15. Shyu YJ, Liu H, Deng X, Hu CD (2006) Iden-
Nat Genet 38:285–293 tification of new fluorescent protein fragments
11. Kerppola TK (2006) Design and implementation for bimolecular fluorescence complementation
of bimolecular fluorescence complementation analysis under physiological condition. Bio-
(BiFC) assays for the visualization of protein techniques 40:61–66
interactions in living cells. Nat Protoc 16. Rogers M, Dani JA (1995) Comparison of
1:1278–1286 quantitative calcium flux through NMDA,
12. Michnick SW (2003) Protein fragment ATP, and ACh receptor channels. Biophys J
complementation strategies for biochemical 68:501–506
Chapter 33
Split-Luciferase Complementation Assay to Detect

Channel–Protein Interactions in Live Cells
Alexander S. Shavkunov, Syed R. Ali, Neli I. Panova-Elektronova,
and Fernanda Laezza
Abstract
The understanding of ion channel function continues to be a significant driver in molecular pharmacology.
In this field of study, protein-protein interactions are emerging as fundamental molecular determinants of
ion channel function and as such are becoming an attractive source of highly specific targets for drug
development. The investigation of ion channel macromolecular complexes, however, still relies on conven-
tional methods that are usually technically challenging and time-consuming, significantly hampering our
ability to identify, characterize and modify ion channel function through targeted molecular approaches. As
a response to the urgent need of developing rapid and albeit accurate technologies to survey ion channel
molecular complexes, we describe a new application of the split-luciferase complementation assay to study
the interaction of the voltage-gated Na + channel with the intracellular fibroblast growth factor 14 and its
dynamic regulation in live cells. We envision that the flexibility and accessibility of this assay will have a broad
impact in the ion channel field complementing structural and functional studies, enabling the interrogation
of protein–channel dynamic interactions in complex cellular contexts and laying the basis for new frame-
works in drug discovery campaigns.
Key words Protein-protein interactions, Protein fragment complementation, Split luciferase comple-
mentation, Firefly luciferase, Bioluminescence, Two-way protein fragment complementation assay,
Application of LCA, Sub-cloning of ion channel
1 Introduction
The split-luciferase complementation assay is a flexible and conve-

nient method that allows the interrogation of a variety of
protein-protein interactions in live cells [1–11]. Construction of a
split-luciferase complementation assay requires cloning of the inter-
acting proteins of interest into suitable mammalian expression vec-
tors. The vectors are designed to express the target proteins in
frame with complementary luciferase fragments spaced by a flexible
linker region. The resulting DNA constructs are then co-
transfected into mammalian cells, which are allowed to grow and
497
498 Alexander S. Shavkunov et al.
express the fusion proteins. The cells are then re-plated onto
multi-well plates, substrate is added, and luminescence resulting
from complementation of luciferase fragments is measured as a
function of the protein-protein interaction of interest. At this step,
additional experimental manipulations (e.g., pharmacological) can
be introduced to test the effect on the protein-protein interaction
of interest in live cells. Additional control experiments are utilized to
rule out possible artifacts that may arise from changes in the reporter
construct abundance (expression and/or degradation levels), viabil-
ity and metabolic state of transfected cells, or direct impact on the
reporter enzymatic activity non-related to the efficiency of the
complex assembly. In this chapter, we describe a new application
of LCA for the study of transmembrane ion channels.
Recent advancements in biology and pharmacology highlight
the crucial role of protein-protein interactions for ion channel
proper functioning and regulation [12–14]. The protein interfaces
through which these interactions are mediated present a novel class
of promising drug targets with high potential for improved speci-
ficity of therapeutic effects [15–19]. However, characterization of
the interactions of ion channels with their regulatory proteins, as
well as high-throughput screening for potential modulators of
these interactions, proves to be quite challenging with conventional
techniques [20, 21]. Direct analysis of protein-protein interactions
by traditional biochemical methods, like enzyme-linked immuno-
sorbent assays (ELISA), surface plasmon resonance (SPR), and
fluorescence polarization (FP), is complicated by the nature of ion
channels, which are difficult to express, purify, and reconstitute
in vitro. The functional effect of protein binding to ion channels
has been also studied using manual and/or automated patch-clamp
electrophysiology [22], fluorescence-based methods [23], or ion
flux assays [24]. Yet, despite certain advantages of these methods
over biochemical assays (higher cost efficiency, no need for protein
purification, measurements done in the cellular context), they pos-
sess inherent shortages and limitations of their own, such as lower
processivity, complicated manual operation, etc.
These considerations stimulated us to seek alternative strategies
to characterize the interaction between the voltage-gated sodium
channel Nav1.6 and the intracellular fibroblast growth factor
14 (FGF14), a biologically relevant regulatory protein which con-
trols gating, stability, and targeting of Nav channel α subunits
(Nav1.1–Nav1.9) through a high affinity interaction with the intra-
cellular C-terminal tail [25–30]. Our approach relied on the
bioluminescence-based split luciferase complementation assay
(LCA) introduced by Luker et al. [31]. This method is based on
functional complementation between two separated fragments of
the Photinus pyralis firefly luciferase which are fused to the pair of
interacting proteins of interest. The interaction of the respective
binding partners drives the complementation of the luciferase
LCA to Detect Channel–Protein Interactions 499
Fig. 1 Application of split luciferase complementation assay (LCA) to studying the

FGF14-Nav interaction. The principle of the split luciferase complementation
assay (LCA) for detection of the FGF14-Nav1.6Ctail channel complex formation
fragments leading to recovery of enzymatic activity and generation

of luminescent signal in the presence of the substrate. The LCA
provides a quantitative and reversible real-time readout of protein-
protein interactions in vitro and in vivo and is the emerging
alternative to fluorescence-based assays (for example, FRET [23])
that are limited by cellular autofluorescence, low signal-to-noise
ratio, and narrow dynamic range [1, 32, 33]. It combines the
processivity and scalability of conventional biochemical assays
with the dynamic readout, more adequate biological context and
higher cost efficiency of electrophysiological and other in cellulo
approaches. The format of the assay and its reliability allow its
application for high-throughput screening experiments.
We have successfully adapted the LCA to detect the assembly of
the C-terminal tail of the Nav1.6 and FGF14 in live cells (Fig. 1)
and demonstrated its applicability for identification of intracellular
signaling pathways which modulate this interaction [10, 11].
2 Materials
2.1 Cell Lines and 1. HEK-293 cells (or another cell line that is easily transfected).
Bacterial Strains 2. Chemically competent E. coli (such as One Shot TOP10
Chemically Competent E. coli, Invitrogen).
2.2 Tissue Culture 1. Dulbecco’s Modified Eagle’s Medium (DMEM).

2. DMEM without L-glutamate and Phenol Red.
3. F12 nutrient mixture.
4. Fetal bovine serum.
5. Penicillin–Streptomycin, liquid.
6. 10 Stock solution phosphate-buffered saline (PBS): 0.2 M
phosphate (0.038 M NaH2PO4, 0.162 M Na2HPO4), 1.5 M
NaCl, pH 7.4, titrated with HCl.
7. 0.25 % Trypsin–EDTA.
8. 25 cm2 or 75 cm2 cell culture flask.
9. 24-well tissue culture plates.
10. 96-well tissue culture white/μClear® plates (Greiner Bio-One).
11. Transfection reagent (such as Lipofectamine 2000, Invitrogen).
2.3 Cloning and DNA 1. FRB-NLuc (Gift from Dr. Piwnica-Worms, Washington Uni-
Preparation versity, St. Louis, MO [1]).
2. CLuc-FKBP (Gift from Dr. Piwnica-Worms, Washington Uni-
versity, St. Louis, MO [1]).
3. CD4-Nav1.6Ctail (Gift of Dr. Benedict Dargent, INSERM,
France [34]).
4. FGF14-1b-GFP [28].
5. pGL3 firefly luciferase plasmid (Promega, Madison, WI).
6. Phusion High-Fidelity DNA Polymerase.
7. DNA restriction enzymes (see Note 1).
8. Primers.
9. T4 DNA ligase.
10. Gel Extraction Kit.
11. LB broth: 10 g of Bacto-Tryptone, 5 g of yeast extract, 10 g of
NaCl dissolved in 1 l of distilled or deionized H2O and ster-
ilized by autoclaving.
12. LB agar: 10 g of Bacto-Tryptone, 5 g of yeast extract, 10 g of
NaCl, 15 g of agar dissolved in 1 l of distilled or deionized H2O
with heating and sterilized by autoclaving.
13. Ampicillin.
14. Selection media: LB agar plates supplemented with 100 μg/ml
ampicillin.
15. Miniprep Kit.
16. EndoFree Plasmid Maxi Kit.
2.4 Luciferase Assay 1. D-luciferin.

2. SP600125 (1,9-pyrazoloanthrone).
3. BAY 11-7082 ((E)3-[(4-methylphenyl) sulfonyl]-2-
propenenitrile).
4. Dimethyl sulfoxide (DMSO).
2.5 Cell Viability 1. Cell Proliferation Assay Kit (such as CyQUANT, Invitrogen).
Assay
2.6 Western Blot 1. Cell scraper.

2. 7.5 % SDS-PAGE gel.
3. PVDF membranes.
4. Lysis buffer: 20 mM Tris base, 150 mM NaCl, 1 % NP-40.
5. Protease inhibitor cocktail for mammalian cells.
6. 10 Tris/Glycine/SDS running buffer: 250 mM Tris–HCl,
1.92 M glycine, 1 % SDS, pH 8.3.
7. Transfer buffer: 20 mM Tris–HCl, 150 mM glycine, pH 8.0.
8. 4 SDS-PAGE sample buffer: 250 mM Tris–HCl, 40 % glyc-
erol, 4 % SDS, 0.002 % bromophenol blue, pH 6.8.
9. Bond-Breaker TCEP solution: 50 mM tris(2-carboxyethyl)
phosphine.
10. TBS-T: 20 mM Tris–HCl, 137 mM NaCl, pH 7.6, 0.1 %
Tween 20.
11. Blocking buffer: 2.5 % nonfat dry milk in TBS-T buffer.
12. Anti-luciferase goat polyclonal antibody.
13. Anti-calnexin rabbit polyclonal antibody.
14. Peroxidase horse anti-goat IgG antibody.
15. Peroxidase goat anti-rabbit IgG antibody.
16. ECL Advance Western Blotting Detection kit.
2.7 Instrumentation 1. Thermal cycler.

2. UV/Vis spectrophotometer.
3. Sonicator.
4. Synergy H4 Multi-Mode Microplate Reader.
5. Power source.
6. Electrophoresis unit.
2.8 Analysis 1. Microsoft Excel.

Software 2. Origin 8.6 (Origin Lab Corporation).
3. FluorChem HD2 System.

4. AlphaView 3.1 software (ProteinSimple).
3 Methods
3.1 DNA One of the major advantages of LCA is the possibility to monitor
Manipulation protein-protein interactions in live cells. This usually requires trans-
fection of mammalian cells with DNA constructs expressing hybrid
proteins bearing complementary fragments of the luciferase
reporter. As an example, we describe the sub-cloning of the DNA
coding sequences of our proteins of interest into the expression
vectors encoding complementary Photinus pyralis luciferase frag-
ments which were designed and constructed by Luker et al. [31].
The resulting plasmids were used to express hybrid protein con-
structs CLuc-FGF14, containing the C-terminus fragment (CLuc,
aa. 398–550) of the luciferase and the full-length human FGF14
isoform 1b, and CD4-Nav1.6-NLuc, containing a chimera of the
CD4ΔCtail (aa. 1–395) and the Nav1.6 C-tail (aa. 1763–1976)
fused to the N-terminus fragment (NLuc, aa. 2–416) of the lucif-
erase (Figs. 1 and 2). Using the CD4-Nav1.6 chimera has a number
advantages, as it is expressed more efficiently than the full-length
Nav1.6 in a heterologous system, ensures proper membrane target-
ing [34] and exposes the Nav channel C-tail on the plasma mem-
brane in the correct orientation. This also allows for the isolation of
the C-tail from the rest of the Nav channel, limiting any potential
indirect modulatory effects on the protein–channel complex
induced by other intracellular domains of the Nav channel. The
design of fusions between the reporter fragments and the proteins
of interest may vary depending on the particular pair of interaction
partners. The N-Luc or the C-Luc fragment of the reporter may be
placed at either terminus of the polypeptide chain of each of the
proteins of interest. Important factors to consider are the spatial
orientation of the interaction partners, in case structural
Fig. 2 Map of the expression vectors used for of the LCA. The detailed
description of the cloning procedure can be found in Subheading 3.1
information is available, as well as their cellular localization and

membrane topology (in the case of peripheral or integral mem-
brane proteins).
1. Amplify the coding sequence of FGF14-1b by PCR using the
following primers containing 50 -BsiWI and 30 -NotI restriction
sites:
Sense: 50 -CTCGTACGCGTCCCGGGGCGTAAAACCGGT
GCCCCTCTTC-30 .
Antisense: 50 -GTTTAGCGGCCGCCTATGTTGTCTTACTC
TTGTTGACTGG-30 .
Use the following conditions:
Segment # of cycles Temperature ( C) Duration

1 1 95 2 min
2 30 95 30 s
60 30 s
72 1 min
3 1 72 10 min
4 Hold 4 1
2. Amplify the coding sequence of CD4-Nav1.6 by PCR using the

following primers containing 50 -BamHI and 30 -BsiWI restric-
tion sites:
Sense: 50 -CGGGGTACCCAAGCCCAGAGCCCTGCCATTT-
CTGTGGGCTCAGGT-3.
Antisense: 50 -CGCGTACGAGATCTGGCACTTGGACTCC-
CTGACCT CTTTTTGCCT-30 .
Use the following conditions:
Segment # of cycles Temperature ( C) Duration

1 1 95 2 min
2 30 95 30 s
65 30 s
72 3.5 min
3 1 72 10 min
4 Hold 4 1
3. Resolve the PCR products by electrophoresis in 1 % agarose gel

and purify them using a gel extraction kit according to the
manufacturer’s instructions.
4. Measure the concentrations of the purified PCR products by
UV spectrophotometry (optical density at 260 nm).
5. Digest 50 ng of the FGF14-1b PCR product with 5 units of
NotI for 8 h at 37 C; add 5 units of BsiWI and continue
digestion overnight at 55 C (see Note 1).
6. Digest 1 μg of CLuc-FKBP with 5 units of NotI for 8 h at

37 C; add 5 units of BsiWI and continue digestion overnight at
55 C.
7. Digest 50 ng of the CD4-Nav1.6 PCR product with 5 units of
BamHI for 8 h at 37 C; add 5 units of BsiWI and continue
digestion overnight at 55 C.
8. Digest 1 μg of FRB-NLuc with 5 units of BamHI for 8 h at
37 C; add 5 units of BsiWI and continue digestion overnight at
55 C.
9. Resolve the digested fragments by electrophoresis in 1 % aga-
rose gel and purify the required fragments using a gel extraction
kit according to the manufacturer’s instructions.
10. Measure the concentrations of the purified DNA fragments by
UV spectrophotometry (optical density at 260 nm).
11. Set up ligation reactions for FGF14 coding sequence with
CLuc and CD4-Nav1.6 with NLuc in 10 μl final volume
using 400 units of T4 DNA ligase per reaction according to
the manufacturer’s instructions. A typical reaction contains
60 ng (CD4-Nav1.6) or 33 ng (FGF14-1b) of the insert and
100 ng of the corresponding plasmid vector, with the vector
and the insert at 1:3 molar ratio (see Note 2). Incubate over-
night at 16 C.
12. Use 0.5–5 μl of the ligation reaction mixture to transform 20 μl
of competent E. coli cells, add 200 μl of LB broth, and incubate
at 37 C for 1 h prior to plating on selective LB agar plates
supplemented with 100 μg/ml ampicillin.
13. Select individual colonies and grow 5-ml overnight cultures in
LB broth containing 100 μg/ml ampicillin.
14. Purify plasmid DNA using a Miniprep Kit and confirm the
identity of recombinant plasmids by restriction digestion and
DNA sequencing.
15. Select the colonies which yielded correct recombinant plasmids
and grow 500-ml cultures in LB broth containing 100 μg/ml
ampicillin.
16. Purify plasmid DNA using EndoFree Plasmid Maxi Kit.
3.2 Transient 1. Grow and maintain HEK293 cells in 25 cm2- or 75 cm2-tissue

Transfection culture flasks at 37 C with 5 % CO2 in medium composed of
of HEK293 Cells equal volumes of DMEM and F12 supplemented with 10 %
fetal bovine serum, 0.05 % glucose, 0.5 mM pyruvate, 100 U/
ml penicillin, and 100 μg/ml streptomycin (supplemented
DMEM/F12).
2. Plate 4.5 105 HEK293 cells per each well of a 24-well tissue
culture plate and incubate overnight to give monolayers at
90–100 % confluency (see Note 3).
3. The next day, pre-warm sterile PBS and DMEM/F12 without
serum and antibiotics at 37 C.
4. Prepare two Eppendorf tubes with equal volumes of DMEM/
F12 without serum or antibiotics (50 μl of medium per each
well of cells in the 24-well plate).
5. Add 1 μg of CD4-Nav1.6-NLuc and 1 μg of CLuc-FGF14
plasmid DNA per each well to be transfected into tube 1.
6. Add 2 μl of transfection reagent (e.g., Lipofectamine 2000) per
each well to be transfected (1 μl of the transfection reagent per
1 μg of plasmid DNA) into tube 2. Incubate for 5 min at room
temperature (see Note 4).
7. Transfer the medium containing DNA from tube 1 into tube
2 with the medium containing the transfection reagent. Mix by
gentle tapping (do not vortex). Incubate the transfection mix-
ture for 20 min at room temperature.
8. During incubation, aspirate the medium from the wells of the
24-well plate with HEK293 cells, wash briefly with PBS and
replace with 100 μl per well of DMEM/F12 without serum or
antibiotics.
9. Carefully dispense 100 μl of the transfection mixture to the
respective well of the 24-well plate containing cells (see Note
5).
10. Incubate the plate with the cells for 6 h at 37 C with 5 % CO2.
11. Add 800 μl of supplemented DMEM/F12 per well and let the
cells grow for 24 h (see Note 6).
3.3 Preparation 1. 24 h post transfection aspirate the culture medium from the 24-
of Transiently well plate with transfected HEK293 cells (see Note 7).
Transfected 2. Wash each well briefly with 400 μl of PBS and dispense 100 μl
HEK293 Cells of 0.04 % trypsin solution (0.25 % trypsin diluted with PBS) per
for Luminescence well (see Note 8).
Measurement 3. Place the plate in the CO2 incubator and monitor the cells
periodically under a microscope (see Note 9).
4. As soon as the cells detach from the plastic (typically 3–5 min)
add 800 μl of supplemented DMEM/F12 per each well to stop
trypsinization.
5. Wash the cells off the plastic by careful pipetting and transfer
the cell suspension from individual wells into a 15-ml tube with
a conical bottom.
6. Centrifuge for 5 min at 800 g.
7. Aspirate the medium and resuspend the cells in fresh

supplemented DMEM/F12 (800 μl of medium per each well
of transfected cells from the 24-well plate).
8. Dispense 200 μl of the cell suspension into each well of a 96-
well plate with white walls and clear flat bottom (see Note 10).
9. Place the plate on a rotary shaker at low speed for 2 min. Check
the wells under microscope to make sure that the cells are
evenly distributed across the bottom.
10. Let the cells grow for 48 h.
3.4 Luminescence 1. Set up the experimental protocol for the Synergy H4 Multi-
Reader Method Setup Mode Microplate Reader with the following parameters
and Data Collection l Maintain temperature at 37 C.
l Dispense 100 μl of substrate solution per well prior to
measurement.
l Shake the plate for 3 s.
l Perform luminescence readings for 30 min at 2 min intervals
(open hole; integration time 0.5 s, sensitivity 245).
2. Replace the culture medium in the wells of the 96-well plate
with HEK293 cells (prepared as described in Subheadings 3.2
and 3.3) with 100 μl of DMEM/F12 without Phenol Red (see
Note 11).
3. Prepare working solution of 1.5 mg/ml of D-luciferin in PBS
(see Note 12).
4. Prime the substrate dispenser of the Synergy H4 Multi-Mode
Microplate Reader with 1,200 μl of D-luciferin working
solution.
5. Load the 96-well plate with cells into the Synergy H4 and
initiate the experimental protocol sequence.
3.5 Identification The LCA provides opportunities to test the effects of various
of Compounds pharmacological agents on the stability of a protein complex [1].
Modulating We have successfully applied this methodology to identify protein
the Protein-Protein kinase pathways which modulate the interaction between FGF14
Interaction Using and the Nav channel (Fig. 3a, b) [10, 11]. It is important to exclude
the LCA false positive hits due to direct influence of the compound on the
luciferase enzymatic activity; this type of artifact has been reported
previously with conventional luciferase reporter systems [5, 35].
We address this issue by parallel testing of candidate inhibitors on
transfected cells expressing full length Photinus pyralis luciferase
(from which the NLuc and CLuc fragments of the reporter have
been engineered) treated under the same conditions as in the LCA
experiment (Fig. 3c).
Fig. 3 The FGF14-Nav1.6Ctail complex formation is regulated by specific kinase inhibitors. (a) HEK293 cells
were transiently transfected with CLuc-FGF14 and CD4-Nav1.6-NLuc and treated with the c-JNK inhibitor
SP600125 (50 μM; closed triangle), the IKK inhibitor, BAY 11-7082 (10 μM; closed diamond) or dimethylsulf-
oxide (DMSO; 0.5 % control; open circle); control without treatment is also shown (open square). Assembly of
the LCA pair is detected as luminescence upon the addition of the D-luciferin substrate at time zero and
normalized to % maximal luminescence signal in the untreated control; data are mean SEM, representing
quadruplicates from one representative experiment. (b) Bar graph represents mean SEM expressed as %
maximal luminescence of control (0.5 % DMSO) from at least three independent experiments. The graphs
illustrates the effect of 50 μM SP600125 and 10 μM BAY 11-7082 on the FGF14-Nav1.6 channel C-tail
assembly. BAY 11-7082 causes a twofold reduction of the LCA reporter assembly and luminescence output,
while SP600125 has no significant impact. (c) HEK293 cells were transiently transfected pGL3 expressing
1. Transfect HEK293 cells with the vectors expressing CLuc-

FGF14 and CD4-Nav1.6-NLuc and prepare them for lucifer-
ase activity measurement as described above in Subheadings 3.2
and 3.3.
2. Transfect HEK293 cells with the pGL3 plasmid expressing full
length firefly luciferase and prepare them following the same
procedure; use 1 μg of plasmid DNA and 1 μl of transfection
reagent per each well of a 24-well plate.
3. Wash the cells briefly with pre-warmed sterile PBS and change
the medium to DMEM/F12 without Phenol Red, serum and
antibiotics, 100 μl per well of the 96-well plate.
4. Prepare dilutions of protein kinase inhibitors in DMSO. The
inhibitor concentration in the intermediate stocks should be
200 the intended final concentration in the culture medium
(see Note 13).
5. Dispense 0.5 μl of protein kinase inhibitor solution into each
experimental well and 0.5 μl of DMSO into each control well of
the 96-well plate with transfected cells (see Note 14).
6. Gently pipette the culture medium in each well one to two
times to ensure uniform concentration of the inhibitor
throughout the volume and to prevent it from precipitating.
7. Incubate the plate for 1 h at 37 C with 5 % CO2.
8. Proceed to luminescence measurement (as described in
Subheading 3.4).
3.6 Data Analysis 1. Export the relative luminescence values measured by Synergy
H4 and organized by well position and time point into Micro-
soft Excel.
2. Plot the luminescence values of the corresponding experimen-
tal wells as a function of time.
3. To compare the luminescence levels across experimental con-
ditions, calculate maximal signal intensity for each well as the
ä
Fig. 3 (Continued) full-length Photinus pyralis firefly luciferase holoenzyme and treated for 1 h prior to the
assay with the indicated compounds. Bar graph expressed as % maximal luminescence illustrates the effect
of the indicated compounds on the intrinsic enzymatic activity of luciferase; no statistically significant change
in full-length lucidferase activity was observed. (d) HEK293 cells were transiently transfected with CLuc-
FGF14 and CD4-Nav1.6-NLuc and treated with the indicated compounds. The effect of compounds on cell
viability was determined using a CyQUANT Cell Proliferation Assay Kit (Promega). Bar graph expressed as %
control fluorescence illustrates the effect of the indicated compounds on cell viability. Data are mean SEM;
***p < 0.001, Student’s t-test. (e) Representative example of Western blots of lysates (equal amount of
protein per lane) from cells transfected with CLuc-FGF14 + CD4-Nav1.6-NLuc and treated for 1 h with 0.5 %
DMSO (2, 4), 50 μM SP600125 (3), and 10 μM BAY 11-7082 (5). Untreated control is also shown (1). Western
blots were probed with a polyclonal anti-luciferase antibody; immunodetection of calnexin was used as
loading control
average of maximum luminescence values at three consecutive

time points (see Note 15).
4. Plot the results using appropriate graphical software (e.g., Ori-
gin 8.6).
3.7 Cell Viability Cell toxicity of test compounds is one of the factors which may
Testing affect the LCA output and produce false-positive results. To
account for this type of artifact, positive identification of com-
pounds by the LCA should be followed by tests on cell viability
under conditions identical to the LCA (Fig. 3d).
1. Transfect the cells with the LCA reporter constructs as
described in Subheading 3.2 and re-plate them onto a 96-
well plate as in Subheading 3.3.
2. Treat the cells with the test compounds as described in Sub-
heading 3.4, under the same conditions as in the LCA
experiment.
3. After incubation with the test compounds carefully aspirate the
culture medium and wash the cells with PBS.
4. Aspirate the PBS, freeze the cells, and store the plate at 80 C
until the samples are to be assayed (see Note 16).
5. Prepare the working solution for the cell proliferation assay
(the procedure below is for the CyQUANT assay).
l Dilute Component B (Cell lysis buffer) 1:20 in nuclease-free
distilled H2O (1 ml +19 ml).
l Thaw Component A (CyQUANT GR Dye) at room tem-
perature; add 250 μl (1:80) to the solution of Component B
prepared as above (see Note 17).
6. Thaw the cells at room temperature and add 200 μl of the
working solution to each well, including empty wells as blanks.
7. Incubate for 2–5 min at room temperature protected from light.
8. Measure fluorescence using a Synergy™ H4 Multi-Mode
Microplate Reader (excitation λ ¼ 485 nm, emission λ ¼
528 nm) or any suitable microplate reader with filters appro-
priate for ~480 nm excitation and ~520 nm emission maxima.
9. Subtract the mean fluorescence value of the blank from fluores-
cence values of the wells containing cells; calculate relative cell
viability as percent mean fluorescent signal intensity in the mock-
treated control samples from the same experimental plate.
4 SDS-Polyacrylamide Gel Electrophoresis and Western Blotting
One of the specific concerns when applying LCA for compound

screening is a possibility that the compounds may affect reporter
activity by changing the expression levels or degradation rates of the
reporter proteins. Thus, it is important to verify the reporter

expression level in HEK293 cells after treatment with inhibitors
positively identified by LCA under identical conditions (Fig. 3e).
Polyclonal anti-luciferase antibody allows for the detection of both
reporter constructs through their corresponding luciferase frag-
ments; thus, no specific antibodies are required for each new pair
of constructs, and the relative expression level of different con-
structs can be compared more easily (see Note 18).
1. Wash transfected HEK293 cells in the wells of the 24-well plate
with warm PBS (see Note 19).
2. Add 50 μl of lysis buffer freshly supplemented with protease
inhibitor cocktail for mammalian cells.
3. Incubate the plate with cells on ice for 5 min with periodic
stirring.
4. Scrape the cells off the plastic with a cell scraper, triturate by
careful pipetting and transfer lysate from each well into a micro-
centrifuge tube.
5. Sonicate the cell suspensions on ice for 20 s with 1 s pulses.
6. Clear the lysates by centrifugation (15,000 g, 15 min at
4 C) and transfer the supernatant into new tubes.
7. Add 33 μl of 4 SDS-PAGE sample buffer and 11 μl of Bond-
Breaker TCEP solution to each sample.
8. Incubate the samples at 65 C for 10 min; let them cool down
to room temperature and spin briefly.
9. Load 50 μl of each sample and 5 μl of a protein molecular mass
standard onto a 7.5 % SDS-PAGE gel and run electrophoresis at
120 V constant.
10. Open the cassette with the gel, assemble it with the PVDF
membrane and load into an electrophoretic transfer cell with
prechilled transfer buffer. Place the transfer cell in an ice
bucket.
11. Run electrophoretic transfer for 2 h at 75 V constant.
12. Incubate the membrane with the transferred protein in block-
ing buffer for 30 min at room temperature.
13. Incubate the membrane overnight with anti-luciferase goat
polyclonal antibody or anti-calnexin rabbit polyclonal antibody
(1:1,000 final dilution) in blocking buffer at 4 C.
14. Wash the membrane with TBS-T three times for 10 min.
15. Incubate the membrane with HRP-labeled goat anti-rabbit or
horse anti-goat secondary antibody (1:5,000).
16. Wash the membrane with TBS-T three times for 10 min.
17. Incubate the membrane with luminescent substrate solution

for 5 min at room temperature.
18. Visualize luminescent signal and record the blot image using
FluorChem HD2 System and AlphaView 3.1 software.
19. Measure the pixel intensity of protein bands using AlphaView
3.1. Normalize the raw values of the bands of interest in each
sample to the corresponding values of calnexin (loading
control).
5 Future Directions: Dual-Color Split Luciferase Complementation Assay (DC-LCA)
A complex problem in protein-protein interaction studies is to

characterize two or more interactions occurring simultaneously.
Advancement in this direction is the introduction of a new type of
LCA, dual-color split luciferase complementation assay (DC-LCA),
which allows the interaction of a particular protein with two bind-
ing partners to be simultaneously monitored [36]. This approach
utilizes two N-terminal fragments derived from structurally related
luciferases of different click beetle species, one with green light and
the other one with red light emission, and a common C-terminal
fragment originating from the green-emitting luciferase. The C-
terminal fragment fused to the protein of interest produces lumi-
nescence in the presence of D-luciferin upon interaction with either
of the two N-terminal fragments fused to the protein binding
partners, resulting in red or green luminescence production
depending on the type of protein pair driving the reporter assembly.
The DC-LCA method could resolve the important question of
whether FGF14 binds to the Nav channel as a monomer, as sug-
gested by structural studies [26], or in a dimeric form, which would
greatly advance our knowledge of the mechanisms underlying ion
channel regulation.
6 Notes
1. We recommend High Fidelity BamHI and NotI enzymes to

avoid star activity.
2. Calculations for the optimal vector-insert ratio can be found at
http://www.insilico.uni-duesseldorf.de/Lig_Input.html.
3. Typically, three 25-cm2 tissue culture flasks or one 75-cm2
tissue culture flask yields enough cells for two 24-well tissue
culture plates.
4. You must proceed to the next step within 25 min.
5. HEK293 cells easily come off the plastic, and therefore it is
important to dispense the media very carefully. Cell loss
can be minimized if the media is slowly poured along the wall

of the wells.
6. It is normal to observe some cell loss after transfection. How-
ever, it is recommended to check the plate with transfected cells
under a microscope and discard wells with abnormally high cell
loss.
7. Do not leave the cells without media. This becomes critical
when cells are split (passaged) into the 96-well plate. We rec-
ommend aspirating one row and then dispensing the media
before proceeding to the second row of the 96-well plates.
8. To minimize cell loss from aspiration, we recommend using a
10 pipette tip fitted on a glass Pasteur pipette to reduce the
suction of liquid flow.
9. Avoid long exposure of cells to trypsin.
10. It is recommended to use 96-well plates with non-transparent
(black or white) walls to avoid the bleed-through of the lumi-
nescent signal between the adjacent wells. A clear (transparent)
bottom well enables routine microscopic examination of the
cell monolayer quality.
11. Using medium without Phenol Red helps to maximize the
luminescence output.
12. Prepare stock solution of D-luciferin in PBS (30 mg/ml) and
store aliquots of appropriate size at 20 C. Keep solutions
containing luciferin protected from light.
13. Protein kinase inhibitors are generally better soluble and more
stable in DMSO than in aqueous solutions. It is recommended
to prepare a stock solution of the inhibitor in DMSO and store
it at 80 C in aliquots of appropriate size (e.g., 10 μl) to avoid
multiple freezing and thawing. Using a 20 mM stock of a kinase
inhibitor is convenient for most applications.
Addition of DMSO to the culture medium helps to prevent
kinase inhibitors from precipitating and facilitates intracellular
penetration. However, high concentrations of DMSO may be
detrimental, largely due to compromised integrity of the cell
membrane. In our experiments, concentrations of DMSO in
the culture medium up to 1 % had no significant effect on the
split luciferase reporter activity after 1 h of incubation (unpub-
lished data); however, in cell-based assays it is generally not
recommended to exceed 0.5 % of DMSO in the culture
medium [37]. Using 200 intermediate dilutions of the inhi-
bitors provides a 0.5 % final concentration of DMSO in the
medium with different inhibitor final concentrations. It is
recommended to keep the DMSO and protein kinase inhibitor
solutions protected from direct light.
14. It is recommended to have several identically treated wells of a

96-well plate per each experimental condition. We typically use
4 wells per condition (quadruplicate).
15. Normalization of signal intensities across each experimental
plate to the average luminescence value of the untreated con-
trol wells in a corresponding plate helps to improve reproduc-
ibility of results between individual experiments and minimize
variability arising from non-related factors (e.g., overall trans-
fection efficiency).
16. Samples for the CyQUANT Cell Proliferation Assay can be
stored at 80 C for up to 4 weeks.
17. The working solution can be prepared in advance and stored
for up to 2–3 h at room temperature protected from light.
18. Because HEK293 cells easily detach from plastic during con-
secutive steps of washing and changing solutions (see Note 5)
in-cell Western assay and similar techniques are not optimal for
this purpose. Besides, this type of assay does not allow discrim-
ination between the two reporter constructs when immunode-
tection is done using anti-luciferase polyclonal antibody.
19. It is preferable to pre-warm PBS to 37 C to prevent cells
coming off the plastic.
References
1. Luker KE, Smith MC, Luker GD et al (2004) 6. Herbst KJ, Allen MD, Zhang J (2011) Lumi-
Kinetics of regulated protein–protein interac- nescent kinase activity biosensors based on a
tions revealed with firefly luciferase comple- versatile bimolecular switch. J Am Chem Soc
mentation imaging in cells and living animals. 133:5676–5679
Proc Natl Acad Sci U S A 101:12288–12293 7. Stefan E, Aquin S, Berger N et al (2007) Quan-
2. Misawa N, Kafi AK, Hattori M et al (2010) tification of dynamic protein complexes using
Rapid and high-sensitivity cell-based assays of Renilla luciferase fragment complementation
protein–protein interactions using split click applied to protein kinase A activities in vivo.
beetle luciferase complementation: an Proc Natl Acad Sci U S A 104:16916–16921
approach to the study of G-protein-coupled 8. Thorne N, Inglese J, Auld DS (2010) Illumi-
receptors. Anal Chem 82:2552–2560 nating insights into firefly luciferase and other
3. Paulmurugan R, Gambhir SS (2003) Monitor- bioluminescent reporters used in chemical
ing protein–protein interactions using split biology. Chem Biol 17:646–657
synthetic renilla luciferase protein-fragment- 9. Auld DS, Thorne N, Nguyen DT et al (2008)
assisted complementation. Anal Chem A specific mechanism for nonspecific activation
75:1584–1589 in reporter-gene assays. ACS Chem Biol
4. Paulmurugan R, Umezawa Y, Gambhir SS 3:463–470
(2002) Noninvasive imaging of protein–pro- 10. Shavkunov A, Panova N, Prasai A et al (2012)
tein interactions in living subjects by using Bioluminescence methodology for the detec-
reporter protein complementation and recon- tion of protein–protein interactions within the
stitution strategies. Proc Natl Acad Sci voltage-gated sodium channel macromolecular
99:15608–15613 complex. Assay Drug Dev Technol
5. Herbst KJ, Allen MD, Zhang J (2009) The 10:148–160
cAMP-dependent protein kinase inhibitor H- 11. Shavkunov AS, Wildburger NC, Nenov MN
89 attenuates the bioluminescence signal pro- et al (2013) The fibroblast growth factor 14
duced by Renilla luciferase. PLoS One 4: (FGF14)/voltage-gated sodium channel com-
e5642 plex is a new target of glycogen synthase kinase
3 (GSK3). J Biol Chem 288 26. Goetz R, Dover K, Laezza F et al (2009) Crys-
(27):19370–19385 tal structure of a fibroblast growth factor
12. Thayer DA, Jan LY (2010) Mechanisms of dis- homologous factor (FHF) defines a conserved
tribution and targeting of neuronal ion chan- surface on FHFs for binding and modulation of
nels. Curr Opin Drug Discov Devel voltage-gated sodium channels. J Biol Chem
13:559–567 284:17883–17896
13. Leterrier C, Brachet A, Fache MP et al (2010) 27. Laezza F, Gerber BR, Lou JY et al (2007) The
Voltage-gated sodium channel organization in FGF14(F145S) mutation disrupts the interac-
neurons:protein interactions and trafficking tion of FGF14 with voltage-gated Na + chan-
pathways. Neurosci Lett 486:92–100 nels and impairs neuronal excitability. J
14. Catterall WA (2010) Signaling complexes of Neurosci 27:12033–12044
voltage-gated sodium and calcium channels. 28. Lou JY, Laezza F, Gerber BR et al (2005)
Neurosci Lett 486:107–116 Fibroblast growth factor 14 is an intracellular
15. Jubb H, Higueruelo AP, Winter A et al (2012) modulator of voltage-gated sodium channels. J
Structural biology and drug discovery for pro- Physiol 569:179–193
tein–protein interactions. Trends Pharmacol 29. Goldfarb M, Schoorlemmer J, Williams A et al
Sci 33:241–248 (2007) Fibroblast growth factor homologous
16. Smith MC, Gestwicki JE (2012) Features of factors control neuronal excitability through
protein–protein interactions that translate into modulation of voltage-gated sodium channels.
potent inhibitors: topology, surface area and Neuron 55:449–463
affinity. Expert Rev Mol Med 14:e16 30. Dover K, Solinas S, D’Angelo E et al (2010)
17. Wells JA, McClendon CL (2007) Reaching for Long-term inactivation particle for voltage-
high-hanging fruit in drug discovery at pro- gated sodium channels. J Physiol 589(Pt
tein–protein interfaces. Nature 6):1505
450:1001–1009 31. Luker KE, Piwnica-Worms D (2004) Optimiz-
18. Buchwald P (2010) Small-molecule protein–- ing luciferase protein fragment complementa-
protein interaction inhibitors: therapeutic tion for bioluminescent imaging of
potential in light of molecular size, chemical protein–protein interactions in live cells and
space, and ligand binding efficiency considera- animals. Methods Enzymol 385:349–360
tions. IUBMB Life 62:724–731 32. Yang KS, Ilagan MX, Piwnica-Worms D et al
19. Zinzalla G, Thurston DE (2009) Targeting (2009) Luciferase fragment complementation
protein–protein interactions for therapeutic imaging of conformational changes in the epi-
intervention: a challenge for the future. Future dermal growth factor receptor. J Biol Chem
Med Chem 1:65–93 284:7474–7482
20. Clare JJ (2010) Targeting ion channels for 33. Ilagan MX, Lim S, Fulbright M et al (2011)
drug discovery. Discov Med 9:253–260 Real-time imaging of notch activation with a
luciferase complementation-based reporter. Sci
21. Wickenden A, Priest B, Erdemli G (2012) Ion Signal 4:rs7
channel drug discovery: challenges and future
directions. Future Med Chem 4:661–679 34. Garrido JJ, Fernandes F, Giraud P et al (2001)
Identification of an axonal determinant in the
22. Kiss L, Bennett PB, Uebele VN et al (2003) C-terminus of the sodium channel Na(v)1.2.
High throughput ion-channel pharmacology: EMBO J 20:5950–5961
planar-array-based voltage clamp. Assay Drug
Dev Technol 1:127–135 35. Thorne N, Auld DS, Inglese J (2010) Apparent
activity in high-throughput screening: origins
23. Masi A, Cicchi R, Carloni A et al (2010) Opti- of compound-dependent assay interference.
cal methods in the study of protein–protein Curr Opin Chem Biol 14:315–324
interactions. Adv Exp Med Biol 674:33–42
36. Villalobos V, Naik S, Bruinsma M et al (2010)
24. Terstappen GC (2004) Nonradioactive rubid- Dual-color click beetle luciferase heteroprotein
ium ion efflux assay and its applications in drug fragment complementation assays. Chem Biol
discovery and development. Assay Drug Dev 17:1018–1029
Technol 2:553–559
37. Lazo JS, Brady LS, Dingledine R (2007)
25. Laezza F, Lampert A, Kozel MA et al (2009) Building a pharmacological lexicon: small
FGF14 N-terminal splice variants differentially molecule discovery in academia. Mol Pharma-
modulate Nav1.2 and Nav1.6-encoded sodium col 72:1–7
channels. Mol Cell Neurosci 42:90–101
Chapter 34
Confocal Microscopy for Intracellular Co-localization

of Proteins
Toshiyuki Miyashita
Abstract
Confocal laser scanning microscopy is the best method to visualize intracellular co-localization of proteins
in intact cells. Because of the point scan/pinhole detection system, light contribution from the neighbor-
hood of the scanning spot in the specimen can be eliminated, allowing high Z-axis resolution. Fluorescence
detection by sensitive photomultiplier tubes allows the usage of filters with a narrow bandpath, resulting in
minimal cross-talk (overlap) between two spectra. This is particularly important in demonstrating co-
localization of proteins with multicolor labeling. Here, the methods outlining the detection of transiently
expressed tagged proteins and the detection of endogenous proteins are described. Ideally, the intracellular
co-localization of two endogenous proteins should be demonstrated. However, when antibodies raised
against the protein of interest are unavailable for immunofluorescence or the available cell lines do not
express the protein of interest sufficiently enough for immunofluorescence, an alternative method is to
transfect cells with expression plasmids that encode tagged proteins and stain the cells with anti-tag
antibodies. However, it should be noted that the tagging of proteins of interest or their overexpression
could potentially alter the intracellular localization or the function of the target protein.
Key words Confocal microscopy, Immunofluorescence, Multicolor labeling, Laser, Fluorophore
1 Introduction
Confocal laser scanning microscopy is the best method to visualize

intracellular co-localization of proteins in intact cells. Confocal laser
scanning microscopy offers significant advantages for viewing sub-
cellular localization of proteins compared with conventional fluo-
rescence microscopy.
First of all, the point scan/pinhole detection system eliminates
light contribution from the neighborhood of the scanning spot in
the specimen (Fig. 1). Therefore, high Z-axis resolution as well as
an X-Y image can be obtained. Fluorescence detection by sensitive
photomultiplier tubes (PMT) allows the usage of filters with a
narrow bandpath, resulting in minimal cross-talk (overlap) between
two spectra. This is particularly important in demonstrating the
515
516 Toshiyuki Miyashita
Photomultiplier
Pinhole
Pinhole
Pinhole
Laser Beam Splitter
(Dichroic Mirror)
Scanner
Objective Lens
Object in Focal Plane

Object not in Focal Plane
Fig. 1 Schematic diagram of the light path
co-localization of proteins with multicolor labeling. In addition,

digital images obtained with PMT are easy to record, modify, and
transfer. Transmitted light images can be obtained by use of a
transmitted light detector, but the images will not be confocal
images. Such transmitted light image can be used in combination
with a fluorescence image to demonstrate fluorescent localization.
Notably, co-localization of two proteins does not necessarily mean
that there is a physical association between the proteins. Other
methodologies, such as co-immunoprecipitation, should be used
to confirm the interaction.
The procedure described in this chapter includes transient
transfection of constructs which encode GFP- and HA-tagged
proteins, followed by immunofluorescence using anti-HA anti-
body. Finally, the interaction of the two proteins of interest is
observed under a confocal laser scanning microscope by demon-
strating the intracellular co-localization after a merged image is
generated. The procedure to detect the intracellular co-localization
of two endogenous proteins by multicolor labeling is also
described.
2 Materials
1. Confocal laser scanning microscope (CLSM) (e.g., Olympus

FLUOVIEW FV300 or equivalent).
2. Tissue culture equipment.
3. HeLa cells.
Confocal Microscopy 517
4. DMEM medium with 10 % fetal bovine serum.

5. Constructs encoding GFP-tagged or HA-tagged proteins of
interest.
6. Chamber slide with cover.
7. Transfection reagent (Effectene transfection reagent (QIA-
GEN) is used as an example).
8. PBS (phosphate-buffered saline).
9. 4 % Paraformaldehyde in PBS.
10. Permeabilization solution: 0.1 % TritonX-100 in PBS.
11. Preblock solution: 10 mM Tris–HCl, pH 8.0, 150 mM NaCl,
0.1 % Tween 20, 5 % skim milk, 2 % bovine serum albumin
(should be filtered through filter paper just before use).
12. Anti-HA mouse monoclonal antibody (HA.11, MMS-101R,
CRP Inc.) (see Note 1).
13. Secondary antibodies: TRITC (tetramethylrhodamine-isothio-
cyanate)-labeled rabbit anti-mouse immunoglobulin (R0270,
DAKO), Alexa Fluor 488-labeled goat anti-mouse immuno-
globulin (A-11029, Molecular Probes), Alexa Fluor 546-
labeled goat anti-rabbit immunoglobulin (A-11035, Molecular
Probes).
14. Vectashield® mounting medium (Vector laboratories, Inc.).
15. Micro cover glass.
16. Rubber cement or nail polish.
3 Methods
The methods described below outline (1) the detection of tran-

siently expressed tagged proteins and (2) the detection of endoge-
nous proteins. To obtain multicolor images with optimum contrast
and minimal cross-talk, appropriate combinations of fluorophores
should be selected. The spectral properties of frequently used
fluorophores and recommended lasers are listed in Table 1.
3.1 Detection of When antibodies raised against a protein of interest are unavailable
Transiently Expressed for immunofluorescence or the available cell lines do not express
GFP- and HA-Tagged the protein of interest sufficiently enough for immunofluorescence,
Proteins an alternative method is to transfect cells with expression plasmids
that encode tagged proteins and stain the cells with anti-tag
antibodies. In addition, the protein of interest can be fused with a
fluorophore for direct detection. Currently, a wide variety of
fluorescent proteins have been developed for generating fusion
proteins. By selecting a proper combination (e.g., EGFP and
DsRed2-Monomer), it is possible to investigate the subcellular
Table 1
Approximate spectral properties of representative fluorophores used for multicolor labeling
Absorption Emission
Fluorophore (nm) (nm) Laser Notes
Group 1 (green)
FITC 490 512 Ar.488 Most widely used green fluorescent dye,
easy to photobleach
EGFP 488 507 Ar.488 Brighter than wild-type GFP, for
generating fusion proteins
Alexa Fluor 488 495 519 Ar.488 Brighter and more photostable than FITC
Group 2 (red)
TRITC 541 572 HeNe543 or Commonly used red fluorescent dye in
Ar.Kr.568 combination with FITC
Texas red 596 620 HeNe543 or Good spectral separation from FITC,
Ar.Kr.568 slightly higher background staining [4]
DsRed2-monomer 556 586 HeNe543 or For generating fusion proteins
Ar.Kr.568
Cy3 552 565 HeNe543 or Brighter than TRITC
Ar.Kr.568
Alexa Fluor 546 556 573 HeNe543 or Brighter than TRITC or Cy3
Ar.Kr.568
Propidium iodide 530 615 HeNe543 or For DNA/RNA staining
Ar.Kr.568
Group 3 (blue)
DAPI 345 425 UV.Ar.364 For nuclear staining
Hoechst33342 355 465 UV.Ar.364 For nuclear staining
For multicolor labeling, choose one fluorophore from each group depending on the laser sources equipped in the CLSM,
thus allowing up to three-color labeling
localization of two proteins of interest in living cells [1]. Further-

more, when the CLSM is equipped with a proper stage incubator in
which temperature and CO2 are precisely controlled, long-term
time lapse imaging can be obtained [2].
When overexpressing a tagged protein of interest, it is manda-
tory to verify the plasmid by DNA sequencing and confirm the
proper molecular weight of the expressed protein by Western blot-
ting [3]. It should also be noted that the tagging of proteins of
interest or their overexpression could potentially alter the intracel-
lular localization or the function of the target protein. To demon-
strate that the protein of interest is localized in a particular
organelle, a number of organelle-specific probes or vectors as
well as organelle-specific monoclonal antibodies are available.
To visualize mitochondria, for example, MitoTracker probes

(Molecular Probes) and mitochondria localization vectors, such as
pECFP-Mito (Clontech; Palo Alto, CA), can be used.
3.1.1 Transfection 1. The day before transfection, seed HeLa cells at a density of
7 104 cells/well (when two well chamber slides are used).
2. Incubate the cells at 37 C and 5 % CO2 in DMEM medium.
3. Prepare 0.5 μg of each plasmid that encodes GFP- or HA-
tagged protein. The minimum DNA concentration should be
0.1 μg/μl.
4. Dilute the DNA in 60 μl of Buffer EC (Effectene transfection
reagent kit). Add 6 μl of Enhancer and mix by vortexing for 1 s
(see Note 2).
5. Incubate at room temperature for 2–5 min and spin down the
mixture for a few seconds.
6. Add 10 μl of Effectene Transfection Reagent to the mixture.
Mix by vortexing for 10 s.
7. Incubate the samples for 5–10 min at room temperature to
allow complex formation.
8. During step 7, aspirate the medium from the slides and wash
the cells once with PBS. Add 1 ml of fresh growth medium to
each well.
9. Add the transfection complexes drop-wise onto the cells.
Gently swirl the plates to ensure uniform distribution of the
complexes.
10. Incubate the cells with the complexes at 37 C and 5 % CO2 for
24 h to allow for gene expression.
3.1.2 Immunofluore- 1. Carefully aspirate the medium and add 0.5 ml/well of 4 %
scence paraformaldehyde in PBS drop-wise onto the cells (see Note 3).
2. To fix the cells, incubate the samples for 1 h at 4 C.
3. Carefully replace the solution with 0.5 ml/well of permeabili-
zation solution.
4. Incubate the samples for 5 min at room temperature.
5. Carefully replace the solution with 0.5 ml/well of preblock
solution and incubate the samples for 1 h at room temperature.
6. Carefully aspirate the preblock solution and add 0.5 ml/well of
anti-HA mouse monoclonal antibody diluted with preblock
solution 1:200.
7. Incubate the samples for 1 h at room temperature (see Note 4).
8. Carefully aspirate the antibody solution and add 1 ml/well of
PBS.
9. Incubate the samples for 10 min at room temperature. Gentle

swirling facilitates washing efficiency.
10. Replace the solution with another 1 ml/well of fresh PBS and
repeat step 9.
11. Carefully aspirate PBS and add 0.5 ml/well of TRITC-labeled
rabbit anti-mouse immunoglobulin diluted with preblock solu-
tion. A 1:20 dilution is recommended.
12. Incubate the samples for 1 h at room temperature.
13. Carefully aspirate the antibody solution and wash the samples
three times as described in steps 8 and 9 (see Note 5).
14. Blot off excess PBS with a tissue, taking care not to allow the
cells to dry out.
15. Place one drop of Vectashield® over each well of cells and then
place a micro cover glass over the entire well, taking care to
minimize air bubbles.
16. Secure the cover glass with rubber cement or nail polish.
3.1.3 Confocal Laser Since the operation procedure of a CLSM depends on manufac-
Scanning Microscopy turer and systems, it is not possible to generalize this procedure.
This chapter describes techniques for the Olympus CLSM. The
manufacturer’s manual should be consulted for details.
1. Choose the correct combination of laser, barrier filters, and
excitation dichroic mirrors for the dyes in use according to
the manufacturer’s recommendation (for GFP/TRITC, the
following combination is appropriate. Laser combination;
Ar488nm + HeNe543nm, Barrier filters; BA510IF +
BA530RIF for channel 1, BA565IF + BA590 for channel 2,
dichroic mirror; DM488/543).
2. Set the power switch of each unit to ON. To stabilize the laser
beam output, allow the system to warm up for at least 10 min
after turning the laser power ON.
3. Start the FLUOVIEW software by double-clicking the FLUO-
VIEW icon on the desktop.
4. Place the specimen in an inverted position (in combination
with an inverted microscope) on the microscope stage and
turn the light path selector to the “Binocular” section.
5. Engage the optimum cube for specimen dye by operating the
cube turret and focus on the specimen initially using transmit-
ted light and then quickly observe the cells with fluorescence.
6. When the area to be observed with confocal microscopy is
determined, turn the light path selector to the “Side port”
position and rotate the cube turret so that no cube is engaged.
7. Choose the highest scan speed and set the zoom ratio to “X1.”
Set the channel to be acquired (brightness should be adjusted
one channel at a time).
8. Click the “Focus” button to acquire repeated images at a high
speed.
9. Focus on the cells of interest and adjust the image brightness.
Parameters to be considered at this step are “PMT voltage,”
“Offset,” and “Gain,” each of which can be adjusted indepen-
dently. The ND filters can also be selected using the LASER
INTENSITY turret. Select the optimum ND filters according
to the brightness of the specimen. Please note that a strong
laser intensity will make the fluorescence fade quickly.
10. When it is necessary to observe the detail of a specific area, the
image of a limited area can be acquired by using the “Zoom”
scale and the “Pan” buttons.
11. Once the appropriate parameters have been determined, stop
repeated scanning, set a lower scan speed and acquire the image
by selecting the “Once” button (see Notes 6 and 7).
12. After the desired image has been acquired, save it to a disk. It is
possible to save images acquired with more than one channel at
a time. The user can select the file types used for saving an
image in a file. Fluoview Multi Tiff format is used for image
analysis and processing on FLUOVIEW. Single TIF is a format
for maximum portability between Macintosh and IBM PCs and
programs such as Adobe Photoshop. To demonstrate the co-
localization of two proteins, the merged image of the two
proteins in a single view should be shown as well as displaying
images of the two channels side by side. For example, when the
expression of a GFP-fused protein is displayed in green and that
of an HA-tagged protein is shown in red, then co-localization
of the two proteins is visualized in yellow in the merged image
(Fig. 2a–c).
3.2 Detection of Demonstrating the intracellular co-localization of two endogenous

Endogenous Proteins proteins is more convincing than showing subcellular localizations
Using Labeled of epitope- or GFP-tagged proteins. However, results should be
Secondary Antibodies obtained with highly specific primary antibodies. For, example
antibodies that detect major cross-reactive bands in Western blot-
ting are not suitable for immunofluorescence. Antibodies should
also be affinity-purified prior to use to reduce background. In
addition, the cell lines used for this analysis should express the
proteins of interest sufficiently enough for immunofluorescence.
The two primary antibodies used in these experiments should be
raised in two different species. Secondary antibodies used for simul-
taneous detection of more than one antigen should also be carefully
selected according to the following criteria. The secondary
Fluorophore α
Fluorophore β
Primary antibody: anti-X raised in animal A
Primary antibody: anti-Y raised in animal B
Secondary antibody: Fluorophore α-labeled

anti-A Ig raised in animal C
Secondary antibody: Fluorophore β -labeled

anti-B Ig raised in animal C
(1)
(2)
X Y X Y
Fig. 2 Schematic diagram of multicolor detection of endogenous proteins
antibodies should (1) be derived from the same host species so that
they do not recognize one another (Fig. 2 (1)), (2) not cross-react
with other primary antibodies used in the assay system (Fig. 2 (2)),
(3) not cross-react with endogenous proteins present in the cell
lines under investigation. Secondary antibodies absorbed against
the sera of a number of species to minimize cross-reactivity are
commercially available and are usually marked as “highly cross-
absorbed” or “for multiple labeling.” When cells growing in sus-
pension are to be stained, they should be cytocentrifuged prior to
the fixation using a Cytospin or equivalent that allows low-speed
centrifugal force to separate the cells on slides while maintaining
cellular integrity.
1. Carefully aspirate the medium and add 0.5 ml/well of 4 %
paraformaldehyde in PBS drop-wise onto the cells (see Note 3).
2. To fix the cells, incubate the samples for 1 h at 4 C.
3. Carefully replace the solution with 0.5 ml/well of permeabili-
zation solution.
4. Incubate the samples for 5 min at room temperature.
anti-protein X mouse monoclonal antibody diluted with
preblock solution. Because staining protocols vary with appli-

cation, the appropriate dilution of antibody should be deter-
mined empirically.
8. Carefully aspirate the antibody solution and add 1 ml/well of
PBS.
9. Incubate the samples for 10 min at room temperature. Gentle
swirling facilitates washing efficiency.
10. Replace the solution with another 1 ml/well of fresh PBS and
repeat step 9.
11. Carefully aspirate PBS and add 0.5 ml/well of Alexa Fluor 488-
labeled goat anti-mouse immunoglobulin diluted with pre-
block solution (1:200 or higher dilution).
12. Repeat steps 7–10.
This step can be omitted depending on the antibody.
anti-protein Y rabbit polyclonal antibody diluted appropriately
with preblock solution.
15. Repeat steps 7–10.
16. Carefully aspirate PBS and add 0.5 ml/well of Alexa Fluor 546-
labeled goat anti-rabbit immunoglobulin diluted with preblock
solution (1:200 or higher dilution).
18. Carefully aspirate the antibody solution and wash the samples
three times as described at steps 8 and 9 (see Note 5).
19. Blot off excess PBS with a tissue, taking care not to allow the
cells to dry out.
20. Place one drop of Vectashield® over each well of cells and then
place a micro cover glass over the entire well, taking care to
minimize air bubbles.
21. Secure the cover glass with rubber cement or nail polish.
Confocal laser scanning microscopy can be performed using
the same combinations of lasers, barrier filters, and dichroic mirror
as described in Subheading 3.1.3 (channel 1 for Alexa Fluor 488,
channel 2 for Alexa Fluor 546). Examples of multicolor labeling of
HeLa cells are demonstrated in Fig. 3.
Fig. 3 Multicolor labeling of HeLa cells. (a–c) HeLa cells were transfected with pSDHIp-EGFP and pHA-Bak
that encode the EGFP-tagged iron sulfur subunit of human succinate dehydrogenase [5] and HA-tagged
proapoptotic Bcl-2 family member, Bak [6, 7], respectively. At 24 h after transfection, cells were stained with
anti-HA antibody followed by TRITC-labeled secondary antibody. Panels (a) and (b) show the subcellular
localizations of SDHIp and Bak, respectively. A merged picture is shown in panel (c). SDHIp is a mitochondrial
protein and Bak is shown to co-localize with SDHIp. Arrows indicate cells transfected with only pSDHIp-EGFP.
(d–f) HeLa cells were incubated with mouse anti-GM130 monoclonal antibody (BD transduction Laboratories)
and rabbit anti-DRPLA antibody [8]. The primary antibodies were detected by secondary antibodies conjugated
to Alexa Fluor 488 or 546. DRPLA protein, a product of the gene responsible for a genetic neurodegenerative
disease, dentatorubral-pallidoluysian atrophy (DRPLA) [9], is localized in the nucleus [10]. However, a closer
observation reveals that it is also localized in the juxtanuclear region (arrowhead) where it co-localizes with
GM130 (f), which is a cis-Golgi network marker [11]. (g–i) HeLa cells were treated similarly as in (d–f) except
that anti-PML mouse monoclonal antibody (Santa Cruz) was used as one of the primary antibodies. The
promyelocytic leukemia (PML) protein is a major component of nuclear dot-like structures known as PML
nuclear bodies (NBs) or PML oncogenic domains (POD) [12, 13] (g). These panels show that the nuclear DRPLA
protein is not recruited to NBs or POD
4 Notes
1. Other representative anti-tag antibodies are described below.

Recommended starting dilutions are indicated in parentheses.
Anti-FLAG (M2) (mouse monoclonal) (1:100) (F3165,
SIGMA; St. Louis, MO).
Anti-His (His-probe, H-15) (rabbit polyclonal) (1:200)
(sc-803, Santa Cruz; Santa Cruz, CA).
Anti-c-Myc (9E10) (mouse monoclonal) (1:100) (sc-40, Santa
Cruz).
2. To achieve optimal transfection efficiency for every new cell
line/plasmid DNA combination, it is recommended to opti-
mize the amounts of transfection reagent, DNA, and the cell
number prior to transfection according to the manufacturer’s
protocol.
3. The alternative method to fix and permeabilize cells is to use
methanol or acetone. The method should be selected empiri-
cally depending on the cell lines and antibodies.
4. Fluorophore-conjugated anti-tag antibodies are also available.
In this case, go to step 13.
5. If the confocal laser scanning microscope is equipped with a
UV laser, it is recommended to add 1.25 μM of Hoechst33342
(Table 1) to the second wash for nuclear counterstaining. This
allows observation of the location of the nucleus.
6. To improve the image quality by reduction of noise, Kalman
accumulation is recommended. It allows acquiring of images
for a specified number of times while averaging the images.
7. For a double-labeled specimen, there are two alternatives to
obtain the image, sequential acquisition and simultaneous
acquisition. With the former image capturing method, an
image slice from each laser excitation is obtained using one
PMT at a time. This scan is advantageous in acquiring an
image with a lower cross-talk between two wavelengths. In
contrast, with a simultaneous scan, images are obtained with-
out a time lag between the two wavelengths.
Acknowledgments
I thank Yuko Ohtsuka and Mami U for their technical assistance. I

am also grateful to Drs. Yoshiaki Shikama and Yuko Okamura-Oho
for their valuable discussions.
References
1. Abe Y, Oka A, Mizuguchi M, Igarashi T, Modulation of apoptosis by the widely
Ishikawa S, Aburatani H, Yokoyama S, Asahara distributed Bcl-2 homologue Bak. Nature
H, Nagao K, Yamada M, Miyashita T (2009) 374:736–739
EYA4, deleted in a case with middle interhemi- 8. Miyashita T, Okamura-Oho Y, Mito Y, Naga-
spheric variant of holoprosencephaly, interacts fuchi S, Yamada M (1997) Dentatorubral pal-
with SIX3 both physically and functionally. lidoluysian atrophy (DRPLA) protein is cleaved
Hum Mutat 30:E946–E955 by caspase-3 during apoptosis. J Biol Chem
2. Yamagata K, Suetsugu R, Wakayama T (2009) 272:29238–29242
Long-term, six-dimensional live-cell imaging 9. Nagafuchi S, Yanagisawa H, Ohsaki E, Shir-
for the mouse preimplantation embryo ayama T, Tadokoro K, Inoue T, Yamada M
that does not affect full-term development. (1994) Structure and expression of the gene
J Reprod Dev 55:343–350 responsible for the triplet repeat disorder, den-
3. Sambrook J, Russell DW (2001) Molecular tatorubral and pallidoluysian atrophy
cloning, a laboratory manual, 3rd edn. Cold (DRPLA). Nat Genet 8:177–182
Spring Harbor Laboratory Press, Cold Spring 10. Miyashita T, Nagao K, Ohmi K, Yanagisawa H,
Harbor, NY Okamura-Oho Y, Yamada M (1998) Intracel-
4. Wessendorf MW, Brelje TC (1992) Which lular aggregate formation of dentatorubral-
fluorophore is brightest? A comparison of the pallidoluysian atrophy (DRPLA) protein with
staining obtained using fluorescein, tetra- the extended polyglutamine. Biochem Biophys
methylrhodamine, lissamine rhodamine, Texas Res Commun 249:96–102
red, and cyanine 3.18. Histochemistry 11. Nakamura N, Rabouille C, Watson R, Nilsson
98:81–85 T, Hui N, Slusarewicz P, Kreis TE, Warren G
5. Kita K, Oya H, Gennis RB, Ackrell BC, Kasa- (1995) Characterization of a cis-Golgi matrix
hara M (1990) Human complex II (succinate- protein, GM130. J Cell Biol 131:1715–1726
ubiquinone oxidoreductase): cDNA cloning 12. Dyck JA, Maul GG, Miller WHJ, Chen JD,
of iron sulfur (Ip) subunit of liver mitochon- Kakizuka A, Evans RM (1994) A novel macro-
dria. Biochem Biophys Res Commun molecular structure is a target of the
166:101–108 promyelocyte-retinoic acid receptor oncopro-
6. Chittenden T, Harrington EA, O’Connor R, tein. Cell 76:333–343
Flemington C, Lutz RJ, Evan GI, Guild BC 13. Weis K, Rambaud S, Lavau C, Jansen J, Carvalho
(1995) Induction of apoptosis by the Bcl-2- T, Carmo-Fonseca M, Lamond A, Dejean A
homologue Bak. Nature 374:733–736 (1994) Retinoic acid regulates aberrant nuclear
7. Kiefer MC, Brauer MJ, Powers VC, Wu JJ, localization of PML-RAR alpha in acute pro-
Umansky SR, Tomei LD, Barr PJ (1995) myelocytic leukemia cells. Cell 76:345–356
Part V
High Throughput Screening Assays for Protein-Protein

Interactions: Case Studies
Chapter 35
Fluorescence Polarization Assay to Quantify

Protein-Protein Interactions in an HTS Format
Yuhong Du
Abstract
Fluorescence polarization (FP) technology is based on the measurement of molecule rotation, and has been
widely used to study molecular interactions in solution. This method can be used to measure binding and
dissociation between two molecules if one of the binding molecules is relatively small and fluorescent. The
fluorescently labeled small molecule (such as a small peptide) rotates rapidly in the solution. Upon
excitation by polarized light, the emitted light remains depolarized and gives rise to a low FP signal.
When the fluorescent small molecules in solution are bound to bigger molecules (such as a protein), the
movement of the complex becomes slower. When such a complex is excited with polarized light, much of
the emitted light is polarized because of the slow movement of the complex. Thus, the binding of a
fluorescently labeled small molecule to a bigger molecule can be monitored by the change in polarization
and measured by the generation of an increased FP signal. This chapter aims to provide a step-by-step
practical procedure for developing an FP assay in a multi-well plate format to monitor protein-protein
interaction (PPI) in a homogenous format.
Key words Fluorescence polarization (FP), Protein-protein interaction (PPI), Small molecule, Probe,
Multi-well plate, Plate reader, Dissociation constant (Kd), Competition assay
1 Introduction
Fluorescence polarization (FP) is a very powerful and sensitive

nonradioactive technique for the study of molecular interactions
in solution. First described by Perrin in 1926 [1], FP is based on the
observation of the molecular movement of fluorescent molecules in
solution. When a fluorescently labeled molecule is excited by polar-
ized light, it emits light with a degree of polarization that is
inversely proportional to the rate of molecular rotation. This prop-
erty can be used to measure binding and dissociation between two
molecules if one of the binding molecules is relatively small and
fluorescent. Molecular rotation is largely dependent on molecular
mass, with larger masses showing slower rotation. The basic princi-
ple of FP is depicted in Fig. 1. When a fluorescently labeled small
529
530 Yuhong Du
Polarized
Polarized bigger molecules
excitation light
excitation light (protein, > 10, 000 Da)
Probe
(fluorescent
small molecules,
< 1500 Da)
Rapid rotation Slow rotation
Emitted light is Emitted light

depolarized is polarized
Low FP FP signal
Fig. 1 Schematic of the theory behind FP assays. A rapidly rotating small

molecule fluorophore gives a low FP signal (low mP). The association of a
relatively large molecule, such as a protein, with the small molecule fluorophore
slows down the rotation of the fluorophore, leading to an increased FP signal
molecule (typically <1,500 Da), such as a peptide or ligand, is free

in solution, it rotates rapidly. Upon excitation by polarized light,
the emitted light will be largely depolarized light. When this fluo-
rescent small molecule is bound to a bigger molecule (typically
>10 kDa), such as a protein, the rotational movement of the
fluorophore becomes slower due to the significantly reduced rota-
tional speed of the complex. Thus, the emitted light will remain
polarized. Therefore, the binding of a fluorescently labeled small
molecule to a protein can be monitored by the change in polariza-
tion from low to high.
Protein-protein interactions (PPI) can be monitored by FP if
one of the components of the PPI is small and fluorescent. There-
fore, the first step to develop an FP binding assay for PPI is to
identify a small binding epitope or peptide for one of the partners,
which can then be labeled with a fluorophore dye. Fluorescently
labeled peptides, which can be synthesized by commercial services,
are typically referred to as a “tracer” or “probe” in an FP assay.
Commonly used fluorophores for FP tracers include fluorescein,
rhodamine, BODIPY, and Cy5 dyes, or their derivatives. The
molecular weight of fluorescently labeled small molecules for FP
assay is typically less than 1,500 Da, although up to 5,000 Da can
be acceptable if the binding partner is very large.
Once the probe has been designed and synthesized, the assay
procedure for FP with modern instrumentation that is capable of
measuring an FP signal in a multiwell plate (96/384/1,536) is
quite simple and easy. An FP binding experiment is performed by
titrating a protein of interest with a fixed concentration of tracer in a
multiwell plate. The FP signal, which is expressed as millipolariza-
tion units (mP), is then measured with a multimode plate reader
equipped with a proper set of filters for the specific fluorophore.
Fluorescence Polarization for Bimolecular Interactions 531
The tracer alone without protein should give rise to only minimal
FP signal for the free tracer. Increasing FP signal with increasing
protein concentrations in the presence of tracer should be observed
if the protein binds to the tracer. At higher protein concentrations,
the FP signal should rise to a plateau that corresponds to complete
binding of all tracers. A binding curve can be generated by plotting
FP signal against protein concentration, and used for dissociation
constant (Kd) calculation.
The following protocol describes a general step-by-step proce-
dure of developing an FP binding assay to monitor the interaction
of two biomolecules. The results of FP assay development for
monitoring the interaction of 14-3-3 protein with a fluorescently
labeled phosphopeptide binding partner are presented as a case
study [2]. The 14-3-3 proteins mediate phosphorylation-
dependent protein-protein interactions. Through binding to
numerous client proteins, 14-3-3 controls a wide range of physio-
logical processes and has been implicated in a variety of diseases,
including cancer and neurodegenerative disorders [3]. In order to
better understand the structure and function of 14-3-3 proteins
and to develop small molecule modulators of 14-3-3 proteins for
physiological studies and potential therapeutic interventions, a
highly sensitive FP-based 14-3-3 assay was designed and optimized.
Using the interaction of 14-3-3 with a fluorescently labeled phos-
phopeptide derived from Raf-1 as a model system (see Fig. 3a), the
detailed procedure to develop an FP assay for monitoring the
interaction of 14-3-3 protein with its binding partners is described.
This 14-3-3 FP binding assay is a simple one-step “mix-and-read”
method for analyzing 14-3-3 proteins. This solution based, versa-
tile method can be used to monitor the binding of 14-3-3 with a
variety of client proteins. While this chapter uses 14-3-3 protein
interactions as an example, the procedure described in this chapter
can be adapted and applied to any FP-suited system for monitoring
the interaction of two molecules.
2 Materials
1. Protein (14-3-3γ): recombinant GST-14-3-3 protein was

expressed in Escherichia coli strain BL21 (DE3) as a GST-tagged
product and purified as described [2].
2. LB medium for auto-induction.
3. Assay buffer: HEPES buffer containing 10 mM HEPES,
150 mM NaCl, 0.05 % Tween-20, and 0.5 mM DTT, pH 7.4
(see Note 1).
4. Resuspension buffer for protein purification: phosphate-
buffered saline (PBS) containing 1.0 mM phenylmethylsulfonyl
fluoride (PMSF), 1 μg/ml leupeptin, and 1 μg/ml aprotinin.
532 Yuhong Du
5. Glutathione-Sepharose beads.
6. Elution buffer: PBS, pH 9, containing 20 mM reduced gluta-
thione, 0.25 mM PMSF, 5 mM dithiothreitol (DTT), and
0.1 % Triton X-100.
7. De-salt gel filtration column (such as a PD-10 gel filtration
column from GE Healthcare).
8. Fluorescent Probe (TMR-pS259-Raf): a phosphopeptide
derived from a well-studied 14-3-3 binding protein, Raf-1,
was synthesized and labeled with 5/6 carboxytetramethylrho-
damine (TMR). The TMR-pS259-Raf contains 15 residues:
5/6-TMR-LSQRQRST[pS]TPNVHM, and was dissolved in
HEPES buffer as 200 μM stock and stored at 20 C
before use.
9. Unlabeled antagonist or control peptides: R18
(PHCVPKNLSWLNLEAN MCLP) [4] and mutated R18,
R18Lys, where D12 and E14 were changed to K [5], were
dissolved in HEPES buffer as 200 μM stock and stored at
20 C before use.
10. Plate: 384-well solid bottom black plate (see Note 2).
11. Plate reader for FP signal measurements (see Note 3): For 14-
3-3 FP assay development, FP measurements were performed
on an Analyst HT plate reader (Molecular Devices) using an FP
protocol. For the tetramethylrhodamine (TMR)-labeled probe
(Ex: 545 nm; Em: 610 nm), a dichroic mirror of 565 nm was
used.
3 Methods
3.1 Protein E. coli BL21 (DE3) strains were grown in LB medium with ampi-
Purification cillin (100 μg/ml) for protein auto-induction. After overnight
incubation with shaking at 37 C, the cells were harvested and the
pellet was resuspended in ice-cold resuspension buffer. The GST-
14-3-3 γ protein was then purified by using Glutathione-Sepharose
beads, and the bound protein was eluted in elution buffer. Salts
were removed from the pooled elution fractions containing the
GST-14-3-3γ fusion protein using a PD-10 gel filtration column
equilibrated in assay buffer without DTT. The protein concentra-
tion was measured and GST-14-3-3γ protein was stored at 20 C
before use.
3.2 Determination of To determine the concentration of TMR-pS259-Raf peptide for

the Concentration of the 14-3-3 binding FP assay, the peptide was serially diluted in assay
Fluorescent Probe buffer in a 384-well plate. An example plate format for the serial
(See Note 4) dilution of peptide is shown in Fig. 2a. The final volume for each
well is 30 μl.
a Peptide titration plate format:

Peptide (nM) (30 µl/well)
0 0.1 0.2 0.4 0.8 1.6 3.1 6.3 12.5 25 50 100
1 2 3 4 5 6 7 8 9 10 11 12
Replicate 1 A
Replicate 2 B
Replicate 3 C
b Protein titration plate format:

Protein (μM) (15 µl/well)
0 0.005 0.010 0.02 0.08 0.16 0.31 0.63 1.25 2.5 5
1 2 3 4 5 6 7 8 9 10 11 12
Replicate 1 D
Replicate 2 E
Replicate 3 F
Fig. 2 An example plate format for probe (a) and protein (b) titrations for FP assay development in a 384-well
plate format. The FP binding assay is then preformed by adding 15 μl of probe at a selected concentration to
the serially diluted protein. The final volume for each well of the 384-well plate is 30 μl
1. Add 30 μl of assay buffer to columns 1–11 in rows A, B, and C,

as triplicates.
2. In a 1.5 ml microcentrifuge tube, make 200 μl of the highest
concentration of peptide tested (e.g., 100 nM of TMR-pS259-
Raf peptide) in assay buffer.
3. Pipette 60 μl of the highest concentration of peptide (100 nM
of TMR-pS259-Raf) to column 12 of rows A, B, and C.
4. Load 3 tips using a multichannel pipette for 384-well plates,
transfer 30 μl of the highest concentration of peptide in tripli-
cate (column 12, rows A, B, and C) to the next column (col-
umn 11, rows A, B, and C) containing 30 μl of assay buffer. The
same three tips may be used for the entire serial dilution
process.
5. Mix the solutions in column 11 (rows A, B, and C) by pipetting
30 μl of the solution up and down at least five times.
6. Transfer 30 μl of the mixture in column 11 (rows A, B, and C)
to the next column (column 10, rows A, B, and C) containing
30 μl of assay buffer.
7. Repeat the above process until reaching column 2. After mixing
the solution in column 2, aspirate and discard 30 μl of solution,
so the final volume of the wells in column 2 is 30 μl. At this
point, all wells contain an equal volume and the concentrations
of peptide range from 0 (column 1) to 100 nM (column 12)
with triplicates for each concentration.
534 Yuhong Du
8. Centrifuge the plate at 1,000 rpm for 2 min to get rid of any
bubbles in the wells. This step is important especially for
handling low volumes, as the bubbles will affect the reading.
9. Load the plate on the plate reader.
10. Set up the proper settings for the instrument software for the
FP measurement.
For 14-3-3 FP assay development, an Analyst HT plate
reader was used. An integration time of 100 ms was used, and
the Z height was set at 2.15 mm (middle). The excitation
polarization was set at “static,” and emission polarization was
set at “dynamic.” For the TMR-labeled probe, an excitation
filter at 545 nm and an emission filter at 610–675 nm were used
with a dichroic mirror of 565 nm. Three panels of the data were
obtained from the FP measurement (see Note 3).
11. Data analysis for peptide titration:
Caluculate the average fluorescence intensity (FI) in parallel
channel for each sample. The concentrations of peptide that
exhibit about ten times or more FI signal compared to buffer-
only background controls should be selected for the following
FP binding assay development (see Note 4).
For the 14-3-3 FP binding assay, 1 nM TMR-pS259-Raf
peptide was chosen based on the observation that this concen-
tration of the TMR-labeled peptide exhibited about ten times
more FI signal compared to buffer-only background controls.
3.3 Development of After selecting the probe concentration, the binding of 14-3-3
the FP Binding Assay protein to TMR-pS259-Raf peptide was performed by titrating
(See Note 5) 14-3-3 protein using a fixed concentration of TMR-pS259-Raf
peptide (1 nM) (see Note 6). Serial dilution of protein was carried
out using a similar procedure as the peptide serial dilution except
the initial volume for each well was different. An example plate
format for protein dilution is shown in Fig. 2b, the final volume for
each well was 15 μl. The FP binding reaction was then performed
by adding 15 μl of peptide at the selected final concentration to all
wells containing increasing concentrations of protein.
1. Add 15 μl of assay buffer to columns 1–11, rows D, E, and F, as
triplicates, in the same plate as the peptide titration.
2. In a 1.5 ml microcentrifuge tube, make 100 μl of 2 of the
highest concentration of protein tested (e.g., 5 μM of GST-14-
3-3γ protein) in assay buffer.
3. Pipette 30 μl of 5 μM GST-14-3-3γ protein to column 12, rows
D, E, and F as triplicates.
4. Load three tips using a multichannel pipette for 384-well
plates, transfer 15 μl of the highest concentration of protein
in triplicate (column 12, rows D,E, and F) to the next column

(column 11, rows D,E, and F) containing 15 μl of assay buffer.
5. Mix the solutions in column 11 (rows D, E, and F) by pipetting
15 μl of the solution up and down at least five times.
6. Transfer 15 μl of the mixture in column 11 (rows D, E, and F)
to the next column (column 10, rows D, E, and F) containing
15 μl of assay buffer.
7. Repeat the above process until reaching column 2. After mixing
the solution in column 2, aspirate and discard 15 μl of solution,
so the final volume of the wells in column 2 is 15 μl. At this
point, all wells contain an equal volume (15 μl) and the con-
centrations of protein range from 0 (column 1) to 5 μM (col-
umn 12) with triplicates (rows D, E, and F) for each
concentration.
8. In a 1.5 ml microcentrifuge tube, make 600 μl of 2 peptide
(2 nM of TMR-pS259-Raf) in assay buffer and transfer to a
reagent reservoir.
9. Using a 384-multichannel pipette, add 15 μl of 2 nM TMR-
pS259-Raf peptide to the serially diluted protein in columns
1–12, rows D, E, and F. At this point, all wells contain an
equal volume (30 μl) with the same concentration of peptide
(final: 1 nM). The final concentrations of protein range from
0 (column 1) to 2.5 μM (column 12) with triplicates (rows D,
E, and F) for each concentration.
10. Centrifuge the plate at 1,000 rpm for 2 min.
12. After incubating the plate at room temperature (RT) for vari-
ous time periods (see Note 7), read the FP signal using the
proper settings.
3.4 Data Analysis for The dynamic range of the FP binding assay (i.e., the FP assay
the FP Binding Assay window) is defined by the difference of the measured FP signal
between bound fluorescent probe in the presence of protein and
unbound (free) probe in the absence of protein.
1. Calculate the FP assay window for each protein concentration
using the following equation:
FP assay window ¼ mPb mPf
mPb is the recorded mP value for the wells containing

TMR-pS259-Raf peptide and 14-3-3 protein, which defines
the binding FP signal from bound peptide. mPf is the recorded
mP value for the samples with TMR-pS259-Raf peptide only in
the absence of 14-3-3 protein, which defines the minimal, or
background, FP signal from unbound peptide.
536 Yuhong Du
Fig. 3 Development of an FP binding assay for monitoring 14-3-3 protein interactions. (a) 14-3-3 binding FP
assay design. The 14-3-3 binding peptide, pS259-Raf, is labeled with 5/6 carboxytetramethylrhodamine
(TMR). The unbound peptide (fluorescent small molecule) rotates fast in solution and gives a low FP signal
when excited with polarized light. The association of 14-3-3 protein (large molecule) slows down the rotation
of the peptide, leading to an increased FP signal. (b) Interaction of 14-3-3 TMR-pS259-Raf peptide induces the
generation of an FP signal. TMR-pS259-Raf peptide (1 nM, final) was incubated with increasing concentra-
tions of GST-14-3-3γ or GST proteins in a 384-well black microplate. The FP signals were recorded after 1 h
of incubation using an Analyst HT. (c) The assay window was calculated by subtracting free peptide
polarization values from values of bound peptide recorded in the presence of specified protein concentrations
of GST-14-3-3. The data were fit to the model FP = ð½14-3-3 BmaxÞ=ð½14-3-3 þ KdÞ using nonlinear
regression analysis, where Bmax is the maximal binding (Prism 4.0; Graphpad). Data shown are the average
values from triplicate samples with standard deviation (SD)
2. Generate a binding isotherm for the calculation of association

parameters such as Kd and maximal binding for the interaction
of 14-3-3 with TMR-pS259-Raf peptide. The obtained FP
assay window from the protein titration should be plotted
against the increasing concentrations of 14-3-3 protein. The
Kd for the binding can be calculated using Graphpad Prism 4.0
software nonlinear regression analysis or another curve fitting
software.
As an example, results from the 14-3-3/TMR-pS259-Raf
FP assay are shown in Fig. 3b. Interaction of 14-3-3 with
TMR-pS259-Raf gave rise to a significant FP signal with mini-
mal background polarization with the peptide probe alone.
With increasing amounts of GST-14-3-3γ protein, polarization
values progressively increased to reach saturation, suggesting
that a greater fraction of fluorescent peptide was bound to the

14-3-3 protein. This FP signal is likely due to the specific
interaction of TMR-pS259-Raf with 14-3-3γ because incuba-
tion of TMR-pS259-Raf with GST alone was incapable of
generating an increase in FP signal. The maximum FP assay
window reached approximately 150 mP with an estimated
dissociation constant, Kd, of 0.41 μM for the Raf peptide
(Fig. 3c). This procedure provides a simple, quantitative
“mix-and-measure” FP assay for studying 14-3-3 protein
interactions.
3.5 Development of Once the FP binding assay has been developed with a reasonable
an FP Competition maximum FP signal window (above ~60 mP), the specificity of the
Assay to Validate the FP signal should be evaluated in a competition FP binding assay
Specificity of the FP format. Such an assay works by measurement of the decrease in FP
Binding Signal signal caused by a competitor, e.g., an un-labeled or antagonist
ligand that displaces the binding of the labeled probe to the pro-
tein. To establish the competition assay, titration of a competitor to
a reaction mixture containing fixed concentrations of protein and
probe is performed as follows.
1. Select the protein concentration for the competition assay.
A concentration of the protein should be chosen that gen-
erates a signal between the EC50 to EC80 of the maximum FP
signal window based on the established FP binding curve.
Higher concentrations of the protein will lie in the plateau
region of the binding curve, which may lead to the competition
assay being insensitive to competitors. Lower concentrations of
the protein will not provide a good assay window for measure-
ment of the decreased signal by displacement of the probe.
Based on the 14-3-3 FP binding curve (Fig. 3b), 0.5 μM
GST-14-3-3γ was selected for the competition assay. This con-
centration of the protein produced ~70 mP FP assay window
and was close to its Kd or EC50 concentration (0.4 μM).
2. Prepare serially diluted competitors in a separate plate (see
Note 8).
In a 384-well round bottom plate, make 5 serially diluted
antagonist or control peptide in assay buffer using the
procedures described in Subheadings 3.2 and 3.3. An example
plate format is given in Fig. 4a.
3. Prepare two mixtures containing 1.25 the selected concen-
tration of peptide and protein in two 1.5 ml microcentrifuge
tubes. Mixture 1 should contain 1.25 nM TMR-pS259-Raf
peptide only and mixture 2 should contain both the peptide
and protein at 1.25 the selected concentrations (1.25 nM of
TMR-pS259-Raf and 0.63 μM of GST-14-3-3γ).
538 Yuhong Du
1. Make competitor plate:

make 5X serial dilution of competitor in a 384-well round-bottom plate
Competitor (μM) (30 μl/well)
0 0 0.1 0.2 0.4 0.8 1.6 3.1 6.3 12.5 25 50
1 2 3 4 5 6 7 8 9 10 11 12
Competitor 1 A
Competitor 2 B
2. Make assay plate:

add 20 μl of probe or protein/probe at selected concentrations to a 384-well black plate
Probe
only Protein + Probe
1 2 3 4 5 6 7 8 9 10 11 12
Replicate 1 G
Replicate 2 H
Replicate 3 I
Replicate 4 J
Replicate 5 K
Replicate 6 L
3. Perform competition assay

Transfer 5 μl from the competitor plate to the assay plate using a multi-channel pipette
Competitor (μM) (25 μl/well)
0 0 0.02 0.04 0.08 0.16 0.31 0.63 1.25 2.5 5 10
1 2 3 4 5 6 7 8 9 10 11 12
Replicate 1 G
Competitor
Replicate 2 H
1
Replicate 3 I
Replicate 1 J
Competitor
Replicate 2 K
2
Replicate 3 L
Fig. 4 An example plate format and procedure for developing a competition FP assay. (1) Make serial dilutions
of competitors in a 384-well round-bottom plate. (2) Prepare an assay plate in a 384-well black plate
containing selected concentrations of probe and protein. Probe only without protein is included as a
background control which gives rise to minimal FP signals. (3) Perform the competition assay by transferring
the competitor to assay plate with triplicate samples
4. As shown in the competition assay plate format (Fig. 4), pipette

20 μl of mixture 1 (peptide only) to column 1, rows G, H, and I
as triplicates. Add 20 μl of mixture 2 (peptide and protein) to
the rest of the wells (columns 2-12, rows G, H, and I).
5. Load a multichannel pipette with 12 pipette tips. Transfer 5 μl/
well of serially diluted competitor in row A (column 1–12) in
the competitor plate to the corresponding well of the assay

plate (row G). Repeat the procedure to transfer the diluted
competitor to rows H and I of the assay plate to generate
triplicate samples for each point (see Note 9).
6. Incubate the plate at RT for a period of time (see Note 10).
8. Read the FP signal using the proper instrument settings.
9. Analyze the data for the competition assay
(a) Calculate percentage of control (% of control) for each dose
of competitor:
The effect of the competitors on the disruption of the
interaction is typically expressed as % of control and calcu-
lated as the following equation:
% of control ¼ ðrecorded FP signalcompetitor
recorded FP signalblank Þ=
ðrecorded FP signalcontrol
recorded FP signalblank Þ 100
Where recorded FP signalblank is the average measured FP

signal from peptide only wells (triplicates), which defines
the minimal FP signal. The recorded FP signalcontrol is the
average FP signal from wells containing protein and pep-
tide, which defines the maximal FP signal, and the recorded
FP signalcompetitor is the FP signal from wells containing
protein and peptide in the presence of competitors.
(b) Calculate the IC50 of the competitor:
Plot the % of control against the competitor concentra-
tion using Graphpad software, or another graphing soft-
ware with curve fitting, to obtain the IC50, the competitor
concentration that leads to displacement of 50 % of the
maximal binding.
10. To obtain the optimal conditions, the competition assay should
be repeated with several different protein concentrations or
more dose points for the competitor tested (see Note 11).
Figure 5a shows an example competition curve from the
14-3-3 FP competition assay. A well-characterized 14-3-3 peptide
antagonist, R18, was used as a competitor in the assay. The addition
of R18 dose-dependently decreased the FP signal for the binding of
14-3-3 to the TMR-pS259-Raf peptide. The negative control pep-
tide, R18Lys, which is a mutated form of R18 peptide, did not have
any effect on the FP signal. These results validate the specificity of
the 14-3-3 FP binding assay for studying 14-3-3 protein-protein
interactions.
540 Yuhong Du
Fig. 5 Competition FP assay. (a) Competitive inhibition of 14-3-3 antagonist peptide, R18, on the 14-3-3/Raf
peptide interaction FP assay. A mutant derivative of R18, R18Lys, which cannot bind 14-3-3, was used as a
negative control. Increasing concentrations of R18 or R18Lys were added to the reaction containing 1 nM
TMR-pS259-Raf peptide and 0.5 μM GST-14-3-3γ. After incubation for 1 h at room temperature, the FP
signals were recorded. Percent of control was calculated as described in Subheading 3.5 and plotted against
peptide concentration. (b) Competition of R18 peptide for the binding of TMR-pS259-Raf peptide to increasing
concentrations of GST-14-3-3γ. Increasing concentrations of R18 peptide were added to four panels of
reactions containing 0.25, 0.5, 1, and 2.5 μM of GST-14-3-3γ and 1 nM of TMR-pS259-Raf peptide. The FP
signals were recorded after 1 h incubation at RT. Percent of control were calculated and plotted against R18
peptide concentrations. Data shown are the average signal from triplicates with SD
4 Notes
1. The buffer used for an FP assay must have low fluorescence

background. Also, the chosen buffer must be one in which the
protein and peptide are stable. Frequently used buffers have a
neutral pH such as PBS, HEPES, or Tris.
2. Other types of multiple-well plates can also be used depending
on the working volume. For example, a minimum of 50 μl
solution per well is required for 96-well plates, 15–50 μl per
well is required for a 384-well plate, and 2–5 μl per well is
sufficient for a 1,536-well plate. To avoid fluorescence cross-
talk, the use of solid-bottom black plates is recommended.
Typically, one 384-well plate should be sufficient for develop-
ing an FP binding assay using the example plate format/layout
as presented in Figs. 2 and 3.
3. Many commercially available instruments are capable of mea-
suring an FP signal from solution in 96/384/1,536-well
microtiter plate format. The fluorescence is measured using
polarized excitation and emission filters. Two measurements
are performed on every well and fluorescence polarization is
defined and calculated as:
Polarization ¼ P ¼ ðI vertical I horizontal Þ=ðI vertical þ I horizontal Þ

Where Ivertical is the intensity of the emission light parallel to the

excitation light plane and Ihorizontal is the intensity of the emis-
sion light perpendicular to the excitation light plane [6]. All
polarization values are expressed as millipolarization units
(mP).
All commercial microplate readers have built-in software for
mP calculation. Depending on the instrument used, three sets
of data are generally reported: (1) calculated mP values; (2) raw
fluorescence intensity counts of vertical (or parallel/S-
channel); and (3) horizontal (or perpendicular/P-channel)
measurements for each well.
mP calculation for different instruments requires the proper
use of measured fluorescence intensity of parallel/S-channel
and perpendicular/P-channel. As optical parts of fluorometers
possess unequal transmission or varying sensitivities for verti-
cally or horizontally polarized light, such instrument artifacts
should be corrected for accurate calculation of the absolute
polarization state of the molecule using fluorescent readers.
This correction factor is known as the “G Factor” which is
instrument-dependent. The G factor corrects for any bias
towards the horizontal (or perpendicular/P-channel) measure-
ment. Most commercially available instruments have an option
for correcting a single-point polarization measurement with a
G factor. For example, the mP values for an FP measurement
with an Envision Multilabel plate reader are calculated as:
mP ¼ 1000 ðS G P Þ=ðS þ G PÞ
In practice for HTS applications, however, it is unnecessary to

measure absolute polarization states; the assay window is the
important factor. The assay window is not significantly changed
by G factor variation.
4. In order to select the proper concentration of fluorescent probe
for an FP binding assay, increasing concentrations of fluores-
cent probe are prepared in assay buffer without the binding
protein using serial dilution. The background control is assay
buffer alone without peptide. In general, the concentration of
peptide that generates a fluorescence intensity (FI) signal of
more than about ten times the “buffer-only” control sample is
selected for developing the FP binding assay. Notice that the FP
signal is expressed as a ratio of fluorescence intensities. Thus,
the FP signal is not influenced by changes in intensity brought
about by changes in the probe concentration. Typically, low
nM fluorescent probe concentrations should be sufficient for
FP binding assay development.
5. For a protein titration experiment, it is recommended to start
with the highest protein concentration possible to make sure
the plateau region of the binding curve is well defined.
542 Yuhong Du
Depending on the solubility of the protein, the starting

concentration of a protein can be in the μM range (e.g.,
40–100 μM). More protein dilution points that cover low nM
to high μM are necessary to obtain the initial portion of the
binding curve, which is especially important if the affinity of the
binding is very high.
6. To determine the specificity of the FP binding signal for a PPI,
it is recommended to include a non-binding protein as a nega-
tive control that should generate only minimal FP signal. In the
case of the 14-3-3 binding assay, GST protein was used as a
negative control which did not give rise to any FP signal in the
presence of the TMR-pS259-Raf peptide (Fig. 3b).
7. After mixing the peptide with the protein, it is recommended to
read the FP binding signal over a period of time, for example,
5 min, 30 min, 1 h, 2 h, 4 h, 18 h. Depending on the interac-
tion, the incubation time to reach the equilibration may vary.
8. The final volume for the diluted competitors in the plate will
be determined based on the number of assay wells for each
sample. Depending on the type of plate, 5–10 μl of dead
volume per well should be prepared for transferring the
diluted competitors to an assay plate containing the reaction
mixture. Multiple competitors or controls for the competition
assay may be tested at the same time by adding more rows of
dilution for additional competitors/controls. Repeating the
assay using a different concentration range of competitor
may be necessary.
9. If testing multiple competitors, the assay wells should be
increased accordingly. Figure 4 shows an example assay format
for testing a second competitor (competitor 2) or control.
10. The FP signal should be measured at various time points after
addition of antagonists, e.g., 30 min, 1 h, 2 h, and overnight.
11. For competition assay development for high throughput
screening (HTS) of PPI inhibitors, it is recommended to test
the effect of competitors in an FP assay using several different
concentrations of protein with reasonable FP signal windows
(above 60 mP). As an example, Fig. 5b shows the effect of the
14-3-3 antagonist peptide on the 14-3-3 competition assay
when the concentration of the protein ranges from EC20 to
EC100 (maximal) of the binding curve. As expected, the sensi-
tivity to the antagonist peptide, R18, decreased as the concen-
tration of protein increased as reflected by the decreased IC50
values with the use of increasing concentrations of GST-14-3-
3γ protein in the FP assay. The IC50 values of 1.4, 2.4, 3.5, and
greater than 10 μM were observed when 0.25, 0.5, 1.0, and
2.5 μM of GST-14-3-3γ were used, respectively. Although the
assay window was increased with increasing concentrations of
14-3-3 protein (Fig. 3c), the sensitivity to the antagonist was

decreased. This aspect is especially important for selecting FP
assay conditions for HTS. The protein concentration selected
in an FP competition assay should strike a balance between
large dynamic range and sensitivity to inhibitors.
5 Summary
FP-based technology has a number of key advantages for moni-

toring bimolecular interactions. It is nonradioactive in a homoge-
nous “mix-and-read” format without wash steps, multiple
incubations, or separation required. FP measurement is directly
carried out in solution; no perturbation of the sample is required,
making the measurement faster and more quantitative than plate-
based methods for bimolecular interactions. It is readily adaptable
to low volume (20–30 μl per well for a 384-well plate or 2–5 μl
per well for a 1,536-well plate). Due to its technical simplicity
and low cost, the FP assay is well-suited for HTS applications to
discover PPI modulators. In addition to monitoring PPI, the FP
assay has been successfully used to study a wide variety of targets
including kinases, phosphatases, proteases, G-protein-coupled
receptors (GPCRs), and nuclear receptors [7, 8]. The procedure
described here is generally applicable to develop an FP assay for
various applications.
Acknowledgments
I thank Dr. Haian Fu for his insightful comments and members of

the Fu laboratory for constructive suggestions. This work was
supported in part by grants from the National Institutes of
Health/National Institute of General Medical Sciences (R01
GM53165 to H.F.) and a Lung Cancer program seed grant from
the Winship Cancer Institute. Yuhong Du is a recipient of the
Emory Drug Development and Pharmacogenomics Academy
Research Fellowship.
References
1. Perrin F (1926) Polarization of light of fluores- regulation. Annu Rev Pharmacol Toxicol
cence, average life of molecules. J Phys Radium 40:617–647
7:390–401 4. Wang B, Yang H, Liu YC et al (1999) Isolation
2. Du Y, Masters SC, Khuri FR et al (2006) Moni- of high-affinity peptide antagonists of 14-3-3
toring 14-3-3 protein interactions with a homo- proteins by phage display. Biochemistry
geneous fluorescence polarization assay. J 38:12499–12504
Biomol Screen 11:269–276 5. Masters SC, Fu H (2001) 14-3-3 proteins medi-
3. Fu H, Subramanian RR, Masters SC (2000) ate an essential anti-apoptotic signal. J Biol
14-3-3 proteins: structure, function, and Chem 276:45193–45200
544 Yuhong Du
6. Jameson DM, Croney JC (2003) Fluo- perspectives and primer. J Biomol Screen
rescence polarization: past, present and future. 5:297–306
Comb Chem High Throughput Screen 8. Burke TJ, Loniello KR, Beebe JA et al (2003)
6:167–173 Development and application of fluorescence
7. Owicki JC (2000) Fluorescence polarization and polarization assays in drug discovery. Comb
anisotropy in high throughput screening: Chem High Throughput Screen 6:183–194
Chapter 36
Estrogen Receptor Alpha/Co-activator Interaction

Assay: TR-FRET
Terry W. Moore, Jillian R. Gunther, and John A. Katzenellenbogen
Abstract
Time-resolved fluorescence resonance energy transfer, TR-FRET, is a time-gated fluorescence intensity
measurement which defines the relative proximity of two biomolecules (e.g., proteins, peptides, or DNA)
based on the extent of non-radiative energy transfer between two fluorophores with overlapping emission/
excitation spectra. In these assays, an excited lanthanide ion acts as a “donor” that transfers energy to an
“acceptor” fluorophore through dipole–dipole interactions. A FRET signal is reported as the ratio of
acceptor to donor emission following donor excitation. When a donor-conjugated protein interacts with
an acceptor-conjugated protein, the donor and acceptor fluorophores are brought in close proximity
allowing energy transfer from the donor to the acceptor resulting in a FRET signal. Because the lanthanide
donors have a long emission half-life, the energy transfer measurement can be time-gated, which dramati-
cally reduces assay interference (due to background autofluorescence and direct acceptor excitation) and
thereby increases data quality. Here, we describe a TR-FRET assay that monitors the interaction of the
estrogen receptor (ER) α ligand binding domain (labeled with a terbium chelate via a streptavidin–biotin
interaction) with a sequence of coactivator protein SRC3 (labeled directly with fluorescein) and the
disruption of this interaction with a peptide and a small molecule inhibitor.
Key words TR-FRET, Protein-protein interaction, Lanthanide, Long-lifetime donor, Fluorophore,

Protein labeling
1 Introduction
The transcription-regulating function of the estrogen receptors

(ERs), ERα and ERβ, relies on their interaction with coactivator
proteins. The best studied coactivators are members of the p160
class of steroid receptor coactivators (SRCs) that functionally link
ER with modification of chromatin structure and activation of the
basal transcriptional machinery [1]. The interaction of the SRCs
with ER is regulated by ligand-induced conformation of the ER
ligand-binding domain (LBD) where coactivator proteins are
bound to ER in the presence of agonist ligands, such as estradiol.
Thus, it is intuitive that a reliable and robust assay able to probe the
interaction state of the estrogen receptor with its coactivator
545
546 Terry W. Moore et al.
protein would allow further elucidation of the dynamics of this

interaction, as well as provide a tool which could be used in high-
throughput screening for the identification and development of
inhibitors of these two biologically relevant proteins.
A number of assays have been developed to study
receptor-coactivator interactions. For instance, glutathione
S-transferase (GST)-pull down and related assays have been used
to study receptor-coactivator interactions, but these assays are
rather labor intensive [2, 3]. An easily employed assay that we [4]
and others [5, 6] have described is based on the principle of
fluorescence polarization (FP) and monitors the interaction of a
fluorophore-labeled SRC LXXLL peptide with the ER LBD.
Unfortunately, this assay has low dynamic range and also requires
high ER concentrations (200 nM), which make accurate determi-
nation of Ki values difficult and costly. Some groups have reported
FRET-based assays (see chapter 20 for more details regarding FRET
assays) to examine nuclear receptor-coactivator interactions, but we
have found certain features of these assays make them less than
ideal: blocking and washing steps, expensive lanthanide-conjugated
antibodies [7–9], or expensive biologic fluorophores [8].
As we found the state-of-the-art assays that were available to
study ER/coactivator interactions less than optimal, we developed
a TR-FRET assay that is amenable to a high-throughput screening
format [10, 11]. The assay we developed uses TR-FRET to moni-
tor the interaction between the ER LBD labeled (via a streptavidin-
biotin interaction) with a terbium chelate and a fluorescein-labeled
sequence of the SRC3 coactivator protein (see Fig. 1). Terbium
functions as a long-lifetime (ca. millisecond) luminescent donor,
and the fluorescein serves as the TR-FRET acceptor. This assay is
superior to organic dye FRET because the emission half-life of
fluorescein is short (nsec) relative to that of the terbium complex
(msec half-life) [12]. Background emission stemming from direct
excitation of fluorescein or endogenous cellular fluorophores can
thus be eliminated by pulsing the terbium complex at the excitation
wavelength and gating the emission with a 50-μs delay. When
properly optimized, the TR-FRET method gives a good signal-
to-noise ratio and can be run in a straightforward, mix-and-mea-
sure format with very low concentrations of terbium-labeled strep-
tavidin and biotin-labeled ER-LBD.
We note that, in a previous publication [11], we detailed the
use of a Cy5-europium pair that was developed in collaboration
with our colleagues at Emory University. Using this pair is advan-
tageous because it allows the monitoring of acceptor emissions at
longer wavelengths than fluorescein. Autofluorescent compounds
found in libraries typically emit at wavelengths shorter than
550 nm; thus, when used in a high-throughput format, the Eu-
Cy5 system is a better choice for minimizing false positives arising
from interfering emission patterns. The reason we have detailed the
Estrogen Receptor/Coactivator TR-FRET Assay 547
340 nm
(D λex) 520 nm
FRET(i.e.A λex) (A λem)
Tb
Fl
ER
SRC
SA B E2
495 nm
340 nm
(D λem)
(D λex)
Tb
Fl
ER
SA CBI SRC
B
E2
Fig. 1 Schematic of the time-resolved fluorescence resonance energy transfer

(FRET) assay. (Top) In general, FRET occurs when an emission wavelength of a
donor molecule (D λem; e.g., 495 nm) overlaps with the excitation wavelength of
a nearby acceptor (A λex), resulting in an emission signal from the acceptor (A
λem; e.g., 520 nm). FRET occurs between the streptavidin-terbium (SA-Tb) donor
and the fluorescein-steroid receptor coactivator (SRC-Fl) acceptor when SRC-Fl
is recruited to the biotin-labeled estrogen receptor (B-ER) bound with the agonist
ligand 17 β-estradiol (E2). (Bottom) In the presence of coactivator binding
inhibitor (CBI), this assembly is disrupted, and the FRET signal decreases
terbium-fluorescein pair here is because we found it to give better

signal-to-noise than the Eu-Cy5 pair when using our particular
plate reader (VICTOR multi-label plate reader) for routine
dose–response assays. In general, if an assay is needed for a high-
throughput screen, we would recommend the Eu-Cy5 pair.
We have developed this assay and explained in detail below the
steps necessary to replicate it using ER alpha. We [13–16] and
others [17–21] have since generalized the assay to other nuclear
hormone receptors and coactivator protein segments, and we
encourage other users to do the same. This does sometimes require
fine-tuning of assay component concentrations, but, generally, the
results are very accurate and reliable.
2 Materials
Prepare all solutions using autoclaved, deionized water and analyti-

cal grade reagents. Prepare and store all reagents at room tempera-
ture (unless indicated otherwise).
1. SRC1-Box II peptide. Store wrapped in foil at 20 C (see

Note 1).
2. Pyrimidine coactivator binding inhibitor (CBI) 1. Store at
20 C (see Note 2).
3. 20 mM solutions of test compounds in DMSO (or DMF).
Store at 20 C.
4. ERα-417-biotin and SRC-3-NRD-fluorescein. Store at 20 C
(see Note 3).
5. 200 nM fluorescein-SRC-3-NRD. Store at 4 C (see Note 4).
6. 1 mM solution of 17β-estradiol in DMF. Store at 20 C.
7. LanthaScreen™ Streptavidin-Terbium (Invitrogen). Store at
20 C.
8. TR-FRET buffer: 20 mM Tris–HCl, pH 7.5, 0.01 % NP40,
50 mM NaCl; adjust pH with conc. HCl to pH 7.5 (see Notes
5–8).
9. 96-Well black HE high efficiency microplate (Molecular
Devices) (see Note 9).
10. 96-Well polypropylene plate (see Note 10).
11. Eight-channel autopipettor with tips (0–20 μL capacity) (see
Note 11).
12. Single-channel manual pipettors with tips (0–1,000 μL capac-
ity) (see Note 12).
13. VICTOR™ Multilabel plate reader (Perkin-Elmer) (see Note
13).
3 Methods
1. Prepare a 4 stock solution (Solution A) containing the fol-

lowing components at the following concentrations: ERα-417
(8 nM), 17β-estradiol (4 μM), and LanthaScreen™
Streptavidin-Terbium (2 nM) in TR-FRET buffer. Keep this
solution at 0–5 C (see Notes 14 and 15).
2. In a 96-well colorless polypropylene plate, serially dilute (see
Note 16) each 20 mM solution of test compound, including
positive controls pyrimidine CBI 1 and SRC1-Box II peptide,
using DMF (see Note 17).
3. In a second 96-well colorless polypropylene plate, dilute (see
Note 18) the previously prepared DMF-diluted test com-
pounds in a 1:10 fashion into TR-FRET buffer (see Note 19).
4. Add 10 μL of each of the buffer-diluted compound solutions to
a black 96-well plate, starting with the highest concentration at
the top left hand corner of the plate, with decreasing
Fig. 2 Representative data from a fluorescence resonance energy transfer (FRET) assay. By plotting the ratio of
the emission intensities of the acceptor to the donor (A/D 1,000) against the log of ligand molar
concentration, dose–response curves for displacement of SRC-3-NRD-fluorescein by the steroid receptor
coactivator peptide (left ) and pyrimidine coactivator binding inhibitor 1 (right) can be generated. Varying the
concentration of the agonist 17β-estradiol (E2; 500 nM (filled square) and 50 μM (filled triangle)) has no
substantial effect on the IC50 values of the compounds, implying that these positive control compounds do not
compete with E2 for the ligand-binding pocket but act by direct displacement of the SRC-3-NRD-fluorescein
concentrations down the plate. For every compound tested,

the eight concentration points are tested in duplicate (two full
columns for each test compound). Add 5 μL of Solution A to
each well of the black 96-well plate.
5. Incubate the plate at 0–5 C (on a foil-covered ice bucket or in
the refrigerator) for 20 min.
6. Add 5 μL of 200 nM fluorescein-SRC-3-NRD to each well of
the black plate (see Notes 20 and 21).
7. Incubate the plate at room temperature for 1 h in the dark (see
Notes 22 and 23).
8. Measure the TR-FRET signal using an excitation filter at
340/10 nm (see Note 24), and emission filters for terbium
and fluorescein at 495/20 and 520/25 nm, respectively, with
an acquisition gated with a 50-μs delay (see Notes 25–28) (see
Fig. 2 for an example of data obtained).
4 Notes
1. The sequence of the SRC1-Box II peptide is NH2-CLTERH-

KILHRLLQE-CO2H. It was synthesized at a private protein
sciences facility. Fluorescein labeling was through the
N-terminal cysteine residue.
2. The pyrimidine CBI 1 positive control was synthesized by our
laboratory as outlined in references [4, 16] below. When design-
ing a TR-FRET assay to monitor a desired protein-protein
interaction, it is best to include a known positive control.
HN
N NH
pyrimidine CBI 1
3. The mutant ER protein labeled with biotin and the

SRC-3-NRD protein fragment labeled with fluorescein were
made in our laboratory, as outlined in the references below
[16, 22], and stored as a 1:1 glycerol–buffer mixture at
20 C. When working with these proteins, try to minimize
the time at room temperature or even on the cooling block. It is
also advisable to aliquot these proteins so that some can be
stored untouched in a freezer without daily disturbance.
4. Diluted fluorescein-SRC-3-NRD can be made by diluting
stock protein into TR-FRET buffer. It should be made prior
to use, and any extra should be discarded.
5. Stock solutions of 1 M Tris at various pH values are used. For
example, to make this buffer, 20 mL of 1 M Tris was added to a
1 L bottle with NP40 and NaCl, and filled to 1 L.
6. NP40 is very viscous. Take time to ensure correct measure-
ment. Thoroughly rinse the graduated cylinder used for mea-
suring and add rinse to the buffer bottle.
7. This concentration was achieved by diluting from a stock solu-
tion of 0.5 M NaCl.
8. TR-FRET buffer is very stable and can be stored at room
temperature.
9. Other 96-well, 20-μL plates may work, but we have found
these to give optimal results.
10. These plates are used for dilution. Any plates that can handle
solvents such as DMF or DMSO without breakdown and can
accommodate the necessary volumes could be used.
11. Autopipettors with low μL capability reduce the error asso-
ciated with this assay and are highly recommended.
12. Use caution when adding low μL volumes to plates or even to
stock solutions. It is very difficult to see if small volumes have
been added to a black plate. This can be checked using a manual
pipettor to aspirate one of the wells and determine the μL
volume in the well, but it is generally not recommended as
some solution will be lost on the pipet tip.
13. This assay has been performed on other plate readers with TR-
FRET capabilities. Concentrations of reagents may need to be
adjusted to generate optimal signal-to-noise results.
14. This stock solution can be divided into eight wells of a dilution
plate so that it can be dispensed into the final black plate using
an autopipettor.
15. This solution should be kept in a cooling block on ice. Any
extra should be discarded at the end of the day.
16. We prefer eight concentrations for dose–response assays using
1:10 and 1:3 serial dilutions. For example, to make approxi-
mately 100 μL of each concentration point, put 100 μL of
20 mM solution in the first well. Make the second well by
adding 30 μL of 20 mM stock to 70 μL of DMF. Then serially
dilute each of these two wells 1:10 into DMF. In this case, the
DMF concentrations would be 20, 6, 2, 0.6, 0.2, 0.06, 0.02,
and 0.006 mM.
17. We often use DMF to prepare stock solutions because DMF
does not freeze and is not as hygroscopic as DMSO. DMF is,
however, less inert than DMSO, and, in other systems, it may
denature proteins. We have found that, in this assay, either
DMSO or DMF can be used with no noticeable effect on
assay performance.
18. For example, add 10 μL of DMF solution to 90 μL TR-FRET
buffer. At this point, the compound concentrations are 2,000,
600, 200, 60, 20, 6, 2, and 0.6 μM, with 10 % DMF.
19. It is not recommended to keep the diluted compounds either in
solvent or in buffer. Even with the most careful covering with
acetate covers and foil wrap, some evaporation does occur,
changing the concentration of the compounds. If the dilution
plates are to be used for several hours, keep covered using a
solvent-resistant plate cover and wrap the plate in foil. Keep the
plate on a chilled block.
20. The 200 nM fluorescein-SRC-3-NRD stock solution can be
divided into eight wells of a dilution plate so that it can be
dispensed into the final black plate using an autopipettor.
21. At this final point, the compound concentrations are 1,000,
300, 100, 30, 10, 3, 1, and 0.3 μM, with 5 % DMF. The
concentrations of each assay component are, for ERα-417,
2 nM; for 17β-estradiol, 1 μM; and for LanthaScreen™
Streptavidin-Terbium, 0.5 nM.
22. Although we recommend waiting for an hour for incubation of
the assay components before reading, the final dose–response
curve and, therefore, Ki value are nearly identical after incuba-
tion for only 5 min. Taking a measurement at an earlier time-
point can be helpful if protein viability needs to be checked or
preliminary results are needed extremely quickly.
23. Fluorescein is a light-sensitive fluorophore. Covering the plate

with aluminum foil is typically sufficient to remove ambient
light.
24. The nomenclature “340/10 nm” refers to the allowance of the
filter used; thus, “340/10 nm” implies that light of wavelengths
from 335 to 345 nm is allowed to pass through the filter.
25. The plate reader should output data giving emission intensities
for the donor (D) and acceptor (A). We calculate the final ratio
used with the following formula: A/D 1,000. This number
can then be plotted against concentration to give a dose–re-
sponse curve.
26. If a compound or compound series does show significant inter-
ference with the fluorescence filters used in this assay, consider
switching to a different acceptor, or donor/acceptor pair (easily
researched on many commercial websites).
27. If unexpected results occur (e.g., a compound gives increasing
interaction with increasing dose) it is likely due to fluorescent
interference of the compound instead of a true increased asso-
ciation of the proteins or stabilizing effect. Check the fluores-
cence spectrum of the compound for any interference.
28. If this assay does not produce reliable, robust results, it is most
commonly due to degradation of protein over time. Realisti-
cally, the proteins should last months when stored and handled
appropriately. Replace the proteins when the signal-to-noise
ratio begins to decrease or the Ki values of the standard peptide
start to drift.
References
1. Tamrazi A, Carlson KE, Rodriguez AL et al estrogen receptor with coactivator peptides.
(2005) Coactivator proteins as determinants Angew Chem Int Ed 46:4471–4473
of estrogen receptor structure and function: 6. Ozers MS, Ervin KM, Steffen CL et al (2005)
spectroscopic evidence for a novel coactivator- Analysis of ligand-dependent recruitment of
stabilized receptor conformation. Mol Endo- coactivator peptides to estrogen receptor
crinol 19:1516–1528 using fluorescence polarization. Mol Endocri-
2. Bramlett KS, Wu YF, Burris TP (2001) Ligands nol 19:25–34
specify coactivator nuclear receptor (NR) box 7. Gowda K, Marks BD, Zielinski TK et al (2006)
affinity for estrogen receptor subtypes. Mol Development of a coactivator displacement
Endocrinol 15:909–922 assay for the orphan receptor estrogen-related
3. Mueller SO (2004) Xenoestrogens: mechan- receptor-gamma using time-resolved fluores-
isms of action and detection methods. Anal cence resonance energy transfer. Anal Biochem
Bioanal Chem 378:582–587 357:105–115
4. Rodriguez AL, Tamrazi A, Collins ML et al 8. Liu JW, Knappenberger KS, Kack H et al
(2004) Design, synthesis, and in vitro (2003) A homogeneous in vitro functional
biological evaluation of small molecule inhibi- assay for estrogen receptors: coactivator
tors of estrogen receptor a coactivator binding. recruitment. Mol Endocrinol 17:346–355
J Med Chem 47:600–611 9. Zhou GC, Cummings R, Li Y et al (1998)
5. Becerril J, Hamilton AD (2007) Helix Nuclear receptors have distinct affinities for
mimetics as inhibitors of the interaction of the coactivators: characterization by fluorescence
resonance energy transfer. Mol Endocrinol 16. Parent AA, Gunther JR, Katzenellenbogen JA
12:1594–1604 (2008) Blocking estrogen signaling after the
10. Gunther JR, Moore TW, Collins ML et al hormone: pyrimidine-core inhibitors of estro-
(2008) Amphipathic benzenes are designed gen receptor-coactivator binding. J Med Chem
inhibitors of the estrogen receptor α/steroid 51:6512–6530
receptor coactivator interaction. ACS Chem 17. Shukla SJ, Nguyen DT, MacArthur R et al
Biol 3:282–286 (2009) Identification of pregnane X receptor
11. Gunther JR, Du Y, Rhoden E et al (2009) A set ligands using time-resolved fluorescence reso-
of time-resolved fluorescence resonance energy nance energy transfer and quantitative high-
transfer assays for the discovery of inhibitors of throughput screening. Assay Drug Dev Tech-
estrogen receptor-coactivator binding. J Bio- nol 7:143–169
mol Screen 14:181–193 18. Wang Y, Yang DZ, Chang A et al (2012) Syn-
12. Mathis G (1995) Probing molecular- thesis of a ligand-quencher conjugate for the
interactions with homogeneous techniques ligand binding study of the aryl hydrocarbon
based on rare-earth cryptates and fluorescence receptor using a FRET assay. Med Chem Res
energy-transfer. Clin Chem 41:1391–1397 21:711–721
13. Kim SH, Gunther JR, Katzenellenbogen JA 19. Schaufele F (2011) FRET analysis of androgen
(2010) Monitoring a coordinated exchange pro- receptor structure and biochemistry in living
cess in a four-component biological interaction cells. Methods Mol Biol 776:147–166
system: development of a time-resolved terbium- 20. Vogel KW, Marks BD, Kupcho KR et al (2010)
based one-donor/three-acceptor multicolor Improved nuclear receptor binding assays for
FRET system. J Am Chem Soc 132:4685–4692 HTS by conversion from FP to TR-FRET read-
14. Jeyakunnar M, Carlson KE, Gunther JR et al out. Endocr Rev 31
(2011) Exploration of dimensions of estrogen 21. Hilal T, Puetter V, Otto C et al (2010) A dual
potency parsing ligand binding and coactivator estrogen receptor TR-FRET assay for simulta-
binding affinities. J Biol Chem 286: neous measurement of steroid site binding and
12971–12982 coactivator recruitment. J Biomol Screen
15. Gunther JR, Parent AA, Katzenellenbogen JA 15:268–278
(2009) Alternative inhibition of androgen 22. Tamrazi A, Carlson KE, Daniels JR et al (2002)
receptor signaling: peptidomimetic pyrimi- Estrogen receptor dimerization: ligand binding
dines as direct androgen receptor/coactivator regulates dimer affinity and dimer dissociation
disruptors. ACS Chem Biol 4:435–440 rate. Mol Endocrinol 16:2706–2719
Chapter 37
High Content Screening Biosensor Assay to Identify

Disruptors of p53–hDM2 Protein-Protein Interactions
Yun Hua, Christopher J. Strock, and Paul A. Johnston
Abstract
This chapter describes the implementation of the p53–hDM2 protein-protein interaction (PPI) biosensor
(PPIB) HCS assay to identify disruptors of p53–hDM2 PPIs. Recombinant adenovirus expression
constructs were generated bearing the individual p53–GFP and hDM2–RFP PPI partners. The N-terminal
p53 transactivating domain that contains the binding site for hDM2 is expressed as a GFP fusion protein
that is targeted and anchored in the nucleolus of infected cells by a nuclear localization (NLS) sequence.
The p53–GFP biosensor is localized to the nucleolus to enhance and facilitate the image acquisition and
analysis of the PPIs. The N-terminus of hDM2 encodes the domain for binding to the transactivating
domain of p53, and is expressed as a RFP fusion protein that includes both an NLS and a nuclear export
sequence (NES). In U-2 OS cells co-infected with both adenovirus constructs, the binding interactions
between hDM2 and p53 result in both biosensors becoming co-localized within the nucleolus. Upon
disruption of the p53–hDM2 PPIs, the p53–GFP biosensor remains in the nucleolus while the shuttling
hDM2–RFP biosensor redistributes into the cytoplasm. p53–hDM2 PPIs are measured by acquiring
fluorescent images of cells co-infected with both adenovirus biosensors on an automated HCS
imaging platform and using an image analysis algorithm to quantify the relative distribution of the
hDM2–RFP shuttling component of the biosensor between the cytoplasm and nuclear regions of
compound treated cells.
Key words Protein-protein interaction biosensors, High content screening, Imaging, Image analysis
1 Introduction
The combination of high content screening (HCS) with fluorescent

biosensors allows for high spatial and temporal resolution of
protein-protein interaction (PPI) partners that have been fused to
spectrally distinct fluorescent reporter proteins [1–9]. The distri-
bution of macromolecules within compartments of the cell is a
tightly regulated process, exemplified by the passage of proteins
through the nuclear pore complexes (NPCs) in the nuclear enve-
lope that separates the cytoplasm and nucleus [10]. While small
molecules pass through NPCs by passive diffusion, the passage of
cargos larger than ~40 kDa require a facilitated active import and
555
556 Yun Hua et al.
export process mediated by specific receptor proteins [10]. Passage

through the NPC involves the assembly of a trimeric import-
complex in the cytoplasm between an importin-α adaptor receptor
that recognizes the nuclear localization sequence (NLS) of the
cargo and the importin-β transport receptor that facilitates docking
interactions with the nucleoporins [10, 11]. Since the nucleus and
cytoplasm represent subcellular compartments that can readily be
distinguished by HCS methods, investigators have taken advantage
of the regulated nucleocytoplasmic transport of proteins to design
biosensors to measure PPIs [1, 2, 4, 5]. To construct translocation
or positional biosensors, the “bait” biosensor is typically anchored
to a specific location within the cell while the “prey” biosensor is
designed to shuttle between different compartments or regions of
the cell, and productive PPIs are required for the co-localization of
both biosensors [1, 2, 4, 5].
In an example involving p53 and hDM2, Stauber and collea-
gues generated a “bait” biosensor encoding the interacting domain
of p53 fused to HIV-1 Rev and blue fluorescent protein (BFP) that
when transfected into cells was expressed in the nucleolus region of
the nucleus [5]. They constructed a shuttling “prey” biosensor
encoding the interacting domain of mdm2 fused with a SV40
NLS sequence, an HIV-1 Rev NES sequence, GST, and green
fluorescent protein (GFP) to generate an NLS-GFP-GST-mdm2-
NES expression plasmid that when transfected into cells was
predominantly localized in the cytoplasm [5]. Co-transfection of
cells with both the p53 “bait” and mdm2 “prey” biosensors
resulted in translocation of the mdm2 biosensor into the nucleus
and co-localization with the p53 biosensor in the nucleolus [5].
Nutlin-3 treatment of cells co-transfected with both biosensors
caused a redistribution of the mdm2 prey biosensor into the cyto-
plasm [5]. We have recently described the development and imple-
mentation of a similarly designed p53–hDM2 protein-protein
interaction biosensor (PPIB) HCS assay in which we screened
220,017 of the NIH’s compound library and identified three
novel structurally related methylbenzo-naphthyridin-5-amine
(MBNA) hits [1, 2]. Hit follow-up studies revealed that the
MBNAs triggered the expected biological responses, including
enhanced p53 protein and p53 target gene levels, p53-dependent
cell cycle arrest in G1, apoptosis, and growth inhibition with
IC50s ~4 μM [2].
This chapter describes the employment of the p53–hDM2
PPIB HCS assay to identify p53–hDM2 PPI disruptors [1, 2].
Recombinant adenovirus expression constructs were created for
the individual p53–GFP and hDM2–RFP PPI partners (Fig. 1).
The N-terminal residues 1–131 of p53 include the p53 transacti-
vating domain that contains the binding site for hDM2. This GFP
fusion protein fragment is targeted to and anchored in the nucleo-
lus by the inclusion of nuclear localization (NLS) and nucleolus
p53–hDM2 Protein-Protein Interaction Biosensor HCS Assay 557
Fig. 1 p53–hDM2 protein-protein interaction biosensor design and distribution phenotypes. Recombinant
adenovirus expression constructs were generated bearing the individual p53–GFP and hDM2–RFP PPI
partners. The N-terminal residues 1–131 of p53 include the transactivating domain that contains the binding
site for hDM2, and this protein fragment is expressed as a GFP fusion protein that is targeted and anchored in
the nucleolus of infected cells by the inclusion of nuclear localization (NLS) and nucleolus localization (NLOS)
sequences. The p53–GFP component is directed to the nucleolus to enhance and facilitate the image
acquisition and analysis of the PPIs. Route 1: In cells infected with the p53–GFP “bait” biosensor adenovirus
alone, the p53–GFP remains localized to bright fluorescent puncta in the nucleolus and its distribution does not
alter upon exposure to Nutlin-3 or other p53–hDM2 disruptors. The N-terminal residues 1–118 of hDM2
encode the domain for binding to the p53 transactivating domain, and this fragment is expressed as a RFP
fusion protein that includes both an NLS and a nuclear export sequence (NES). Route 2: In cells infected with
the hDM2–RFP adenovirus “prey” biosensor alone, hDM2–RFP expression is localized only in the cytoplasm of
cells and does not change upon exposure to Nutlin-3 or other p53–hDM2 disruptors. Route 3: In U-2 OS cells
that are co-infected with both adenovirus constructs, however, the binding interactions between the hDM2
and p53 components of the biosensors result in both proteins becoming localized to the nucleolus. Route 4:
Upon disruption of the p53–hDM2 protein-protein interaction with a compound like Nutlin-3, the p53–GFP
interaction partner remains nucleolar, while the shuttling hDM2–RFP interaction partner redistributes into the
cytoplasm. Route 5: Upon removal of a disrupting agent like Nutlin 3, the shuttling hDM2–RFP interaction
partner redistributes back into the p53–GFP containing nucleolus
localization (NLOS) sequences (Fig. 1) [1, 2], which enhances and

facilitates image acquisition and analysis of the PPIs. In cells
infected with the p53–GFP “bait” biosensor adenovirus alone,
the p53–GFP remains localized to bright fluorescent puncta in
558 Yun Hua et al.
Fig. 2 Color composite images of U-2 OS cells infected with (a) the p53–GFP biosensor alone, (b) the
hDM2–RFP biosensor alone, or (c) co-infected with both biosensors and then treated 10 μM Nutlin-3 for
90 min. U-2 OS cells were infected with (a) the p53–GFP biosensor alone, (b) the hDM2 biosensor alone, or (c)
co-infected with both biosensors and 2,500 cells were seeded into the wells of 384-well assay plates,
cultured overnight at 37 C, 5 % CO2, and 95 % humidity, and then treated 10 μM Nutlin-3 for 90 min. Cells
were then fixed and stained with Hoechst 33342 and 20 0.4 NA images of three fluorescent channels
(Ch1—Hoechst blue, Ch2—p53–GFP green, and Ch3—hDM2–RFP red) were acquired on the ArrayScan VTI
automated imaging platform as described above. Representative color composite images of control and
Nutlin-3 treated U-2 OS cells individually infected with or co-infected with both the p53–GFP and hDM2–RFP
biosensors are presented. In cells infected with each of the biosensors alone, the p53–GFP biosensor remains
localized to bright fluorescent puncta within the nucleoli of Hoechst stained nuclei and its distribution does not
alter upon exposure to Nutlin-3, while the hDM2–RFP biosensor remains localized in the cytoplasm of cells
and does not change upon exposure to Nutlin-3. In U-2 OS cells co-infected with both biosensors, however,
the two biosensors become co-localized to the nucleolus, and upon exposure to Nutlin-3 the hDM2–RFP
interaction partner redistributes into the cytoplasm while the p53–GFP interaction partner remains nucleolar
the nucleolus and its distribution does not alter upon exposure to
Nutlin-3 or when co-infected with the hDM2–RFP biosensor
(Figs. 1, route 1 and 2a). The N-terminal residues 1–118 of
hDM2 encode the domain for binding to the N-terminal transacti-
vating domain of p53, and this fragment is expressed as a RFP
fusion protein that includes both an NLS and a nuclear export
sequence (NES) (Fig. 1) [1, 2]. In cells infected with the
hDM2–RFP adenovirus “prey” biosensor alone, hDM2–RFP
expression is localized only in the cytoplasm of cells and does not

change upon exposure to Nutlin-3 (Figs. 1, route 2 and 2b). In cells
that are co-infected with both adenovirus constructs, however, the
interaction of the hDM2 and p53 components of the biosensors
results in both proteins becoming localized to the nucleolus
(Figs. 1, route 3 and 2c) [1, 2]. Upon disruption of the
p53–hDM2 protein-protein interaction with a compound like
Nutlin-3, the p53–GFP interaction partner remains nucleolar,
while the shuttling hDM2–RFP interaction partner redistributes
into the cytoplasm (Figs. 1, route 4 and 2c) [1, 2].
The disruption of the p53–HDM2 interaction biosensor is
measured by acquiring images on an automated HCS imaging
platform, and an image analysis algorithm is used to quantify the
relative distribution of the hDM2–RFP shuttling component of the
biosensor between the cytoplasm and nuclear regions of compound
treated cells that were co-infected with both adenovirus biosensors
(Figs. 2 and 3) [1, 2]. The images of the three fluorescent channels
(Hoechst, GFP, and RFP) of the p52–hDM2 PPIB were analyzed
using the molecular translocation image analysis algorithm
(Fig. 3a). The nucleic acid dye Hoechst 33342 was used to stain
and identify the nucleus, and this fluorescent signal was used to
focus the instrument and define a nuclear mask in channel 1 (Ch1).
Objects in Ch1 that exhibited the appropriate fluorescent intensi-
ties above background and size (width, length, and area) character-
istics were identified and classified by the image segmentation as
nuclei (dark blue ring, Ch1). The nuclear mask was eroded 1 pixel
to reduce cytoplasmic contamination within the nuclear area, and
the reduced mask was used to quantify the amount of target chan-
nel, p53–GFP (Ch2) or hDM2–RFP (Ch3), fluorescence within
the nucleus (yellow Circ masks in Ch2 and Ch3). The nuclear mask
was then dilated to cover as much of the cytoplasmic region as
possible without going outside the cell boundary. Removal of the
original nuclear region from this dilated mask creates a Ring mask
that covers the cytoplasmic region outside the nuclear envelope
(magenta Ring mask in Ch2 and cyan Ring mask in Ch3). The
image analysis algorithm outputs quantitative data such as: the total
and average fluorescent intensities of the Hoechst stained objects
(Ch1) the selected object or cell count from Ch1, the total and
average fluorescent intensities of the p53–GFP (Ch2) and the
hDM2–RFP (Ch3) signals in the nucleus (Circ) or cytoplasm
(Ring) regions as an overall well average value, or on an individual
cell basis. To quantify the translocation of the hDM2–RFP from the
nucleus to the cytoplasm induced by disruptors of the p53–hDM2
protein-protein interaction the image analysis algorithm calculates
a mean average intensity difference by subtracting the average
hDM2–RFP intensity in the Ring (Cytoplasm) region from the
average hDM2–RFP intensity in the Circ (Nuclear) region of
Ch3; Mean Circ-Ring Average Intensity Difference Channel 3
560 Yun Hua et al.
Fig. 3 Molecular translocation image analysis algorithm. U-2 OS cells co-infected with the p53–hDM2 PPIB
adenoviruses were seeded at 2,500 cells per well in 384-well Greiner collagen-coated assay plates, cultured
overnight at 37 C, 5 % CO2, and 95 % humidity, and were then exposed to the indicated concentrations of
Nutlin-3 in 0.5 % DMSO for 90 min prior to fixation with 3.7 % formaldehyde containing 2 μg/mL Hoechst
33342. Images in three fluorescent channels were sequentially acquired on the ArrayScan VTI platform using a
20 0.4 NA objective with the XF93 excitation and emission filter set and were analyzed with the molecular
translocation (MT) image analysis algorithm. (a) Image segmentation. In Ch1, Hoechst stained objects that
exhibited the appropriate morphological characteristics (width, length, and area) with fluorescent intensities
sufficiently above a background threshold were identified and classified by the image segmentation as nuclei.
The nuclear mask derived from Ch1 (dark blue ring) was then used to segment the images from Ch2 and Ch3
(MCRAID-Ch3) (Fig. 3b). On average, Nutlin-3 exhibited an IC50

of 0.608 0.382 μM (n ¼ 5) in the p53–hDM2 PPIB assay
(Fig. 3b). To implement the p53–hDM2 PPIB assay in screening,
the mean MCRAID-Ch3 value of the DMSO maximum plate
control wells (n ¼ 32) and the mean MCRAID-Ch3 value of the
10 μM Nutlin-3 minimum plate control wells (n ¼ 24) were
utilized to normalize the MCRAID-Ch3 compound data and to
represent 0 % and 100 % disruption/inhibition of the p53–hDM2
interactions, respectively [1, 2].
2 Materials
1. Nutlin-3.
2. Formaldehyde.
3. Dimethyl sulfoxide (DMSO) (99.9 % high performance liquid
chromatography-grade, under argon).
4. Hoechst 33342.
5. U-2 OS osteosarcoma cell line.
6. Culture medium: McCoy’s 5A medium, 2 mM L-glutamine,
10 % fetal bovine serum, and 100 U/mL penicillin and
streptomycin.
7. Humidified incubator at 37 C, 5 % CO2, and 95 % humidity.
8. Dulbecco’s Mg2+ and Ca2+ free phosphate buffered saline
(PBS).
Fig. 3 (continued) into nuclear (Circ) and cytoplasmic (Ring) regions. To reduce cytoplasm contamination
within the nuclear area the nuclear mask was eroded, and the reduced mask (yellow ring) was used to
quantify the amount of target channel, p53–GFP in Ch2 and hDM2–RFP in Ch3, fluorescence within the
nuclear region. To measure fluorescence in a region of the cytoplasm in Ch2 and Ch3, the nuclear mask
derived from Ch1 was dilated to cover as much of the cytoplasm region as possible without going outside the
cell boundary. Removal of the original nuclear region from this dilated mask creates a ring mask that covers
the cytoplasm region outside the nuclear envelope. The number of pixels away from the nuclear mask and the
number of pixels (width) between the inner and outer ring masks were selectable within the MT bio-
application software. The ring masks were then used to quantify the amount of target channel, p53–GFP
(Ch2, mauve rings) or hDM2–RFP (Ch3, light blue rings), fluorescence within the cytoplasm region. (b)
Chemical structure of Nutlin-3 and concentration dependent disruption of p53–hDM2 PPIs. The molecular
translocation image analysis algorithm produces a mean average intensity difference in Ch3 (MCRAID-Ch3)
parameter calculated by subtracting the average hDM2–RFP intensity in the Ring (Cytoplasm) region from the
average hDM2–RFP intensity in the Circ (Nuclear) region of Ch3. High MCRAID-Ch3 values indicate that the
hDM2–RFP biosensor is predominantly localized within the nuclear region, while low MCRAID-Ch3 values
indicate a more prominent localization within the cytoplasm. Nutlin-3 treatment induced a concentration
dependent decrease in the hDM2–RFP MCRAID-Ch3 signal that was consistent with the redistribution of the
hDM2–RFP from the nucleus to the cytoplasm. Nutlin-3 consistently exhibited an IC50 of 0.608 0.382 μM
(n ¼ 5) for the disruption of the p53–hDM2 PPIs
562 Yun Hua et al.
9. Trypsin 0.25 %, 1 g/L EDTA solution.

10. Recombinant adenovirus p53–GFP and hDM2–RFP Biosen-
sors: Recombinant adenovirus expression constructs bearing
the individual p53–GFP (TagGFP, Evrogen, Inc.) and
hDM2–RFP (Tag RFP, Evrogen, Inc.) protein-protein interac-
tion partners were obtained from Cyprotex [1, 2].
11. 384-well collagen-coated barcoded assay microplate.
3 Methods
1. Aspirate spent tissue culture medium from U-2 OS cells in

tissue culture flasks that are <70 % confluent (see Note 1),
wash cell monolayers 1 with PBS, and expose cells to
trypsin-EDTA until they detach from the surface of the tissue
culture flasks. Add serum containing tissue culture medium to
neutralize the trypsin. Transfer the cell suspension to a 50 mL
capped sterile centrifuge tube and centrifuge at 500 g for
5 min to pellet the cells. Resuspend cells in serum containing
tissue culture medium and count the number of trypan blue
excluding viable cells using a hemocytometer.
2. Co-infect 1 107 U-2 OS cells with p53–GFP and
hDM2–RFP adenovirus by incubating cells with the manufac-
turer’s recommended volume of virus (see Note 2), typically
5 μL/107 cells, in 1.5 mL of culture medium for 1 h at 37 C,
5 % CO2 in a humidified incubator with periodic inversion
(every 10 min) to maintain cells in suspension.
3. Dilute co-infected cells to 5.6 104 cells/mL and dispense
45 μL (2,500 cells) in each well of a 384-well collagen-coated
barcoded assay microplate using a liquid handler (see Note 3).
4. Incubate assay plates overnight at 37 C, 5 % CO2 in a humi-
dified incubator (see Note 4).
5. Use an automated liquid handling device to add diluted com-
pounds or plate controls (Nutlin3, 10 μM final, or 5 μL of
DMSO, 0.25–0.5 % final) to appropriate wells for a final screen-
ing concentration of 25 μM (see Note 5).
6. Incubate plates at 37 C, 5 % CO2 in a humidified incubator for
90 min (see Note 6).
7. Fix samples by the adding 50 μL of pre-warmed (37 C) 7.4 %
formaldehyde and 2 μg/mL Hoechst 33342 in PBS using a
liquid handler and incubate at room temperature for 30 min
(see Note 7).
8. Aspirate the liquid and wash the plates twice with 85 μL of PBS
using a liquid handler (see Note 8). Seal with adhesive alumi-
num plate seals (Abgene) with the last 85 μL wash of PBS in
place.
9. Acquire fluorescent images in three channels on an automated

imaging platform (see Note 9) (e.g. ArrayScan VTI, Thermo-
fisher Scientific) (Fig. 2). Images of three fluorescent channels
(Hoechst, GFP, and RFP) are sequentially acquired on an
ArrayScan VTI platform using either a 10 0.3 NA or a 20
0.4 NA objective with the XF93 excitation and emission filter
set (Hoechst, FITC, and TRITC) (Fig. 2). In our system,
excitation is provided by an X-CITE™ 120 W high pressure
metal halide arc lamp with Intelli-Lamp™ technology (Photo-
nic Solutions Inc. Mississauga, Canada), and images are
captured on a high sensitivity cooled Orca CCD Camera
(Photometrics Quantix). Typically with the 10 0.3 NA objec-
tive, the ArrayScan VTI platform is set up to acquire 250
selected objects (nuclei) or two fields of view, whichever
comes first. With the 20 0.4 NA objective, the ArrayScan
VTI platform is set up to acquire four fields of view.
10. Analyze the images of the three fluorescent channels (Hoechst,
GFP, and RFP) of the p53–hDM2 PPIB using the molecular
translocation image analysis algorithm as described above
(Fig. 3) (see Note 10). The MCRAID-Ch3 output is utilized
to quantify the disruption of p53–hDM2 PPIs (Fig. 3b).
4 Notes
1. Typically better responses are obtained when the p53–hDM2

adenovirus biosensors are used to co-infect U-2 OS cells
harvested from tissue culture flasks that are <70 % confluent.
2. The optimal volume of each batch of recombinant adenovirus
biosensor per 106 U-2 OS cells is determined empirically in
virus titration experiments. Increasing amounts of virus are
incubated with the same number of cells and then the levels
of biosensor expression and % of cells infected are determined
on the HCS platform after 24 h in culture. Performing infec-
tions in cells suspended in a low volume of media combined
with periodic inversion (every 10 min) to maintain cells in
suspension enhances both the rate of infection and expression
levels.
3. Determining the optimal cell seeding density is a critical assay
development parameter for any cell-based assay, and especially
for HCS assays. The objective is to minimize the cell culture
burden while ensuring that sufficient cells are captured per
image to give statistical significance to the image analysis para-
meters of interest. Typically variability increases as the number
of cells captured and analyzed decreases.
564 Yun Hua et al.
4. The optimal length of time in culture post viral infection should

be determined empirically for each adenovirus biosensor.
In general, we have found 24 h post infection to be optimal
for most recombinant adenovirus biosensor constructs.
5. Determining the DMSO tolerance is a critical assay develop-
ment parameter for any cell-based assay, and especially for HCS
assays. DMSO has two major effects on HCS assays [1–3, 8, 9,
12, 13]. At DMSO concentrations 5 % there is a significant
cell loss due to cytotoxicity and/or reduced cell adherence.
At DMSO concentrations >1 % but <5 %, cells change from a
well-spread and well-attached morphology to a more rounded
loosely attached morphology that interferes with the ability of
the image analysis algorithm to segment images into
distinct cytoplasm and nuclear regions. The maximum DMSO
tolerance of the HCS assay and the stock concentration of the
compound library are the major determining factors of the
compound concentration selected for primary screening,
confirmation, and follow-up studies.
6. Determining the appropriate compound exposure period is a
critical assay development parameter for any cell-based assay,
including HCS assays, and should be determined empirically
for each new assay. Longer compound exposure periods can
result in increased levels of cytotoxicity, which may significantly
hinder the ability to make reliable measurements.
7. Determining appropriate cell fixation and nuclear staining con-
ditions is a critical assay development parameter for all end point
HCS assays. Combining cell fixation with Hoechst nuclear
staining in a single procedure saves time and reduces the num-
ber of steps in the protocol. Cell fixation is also important
because the scanning and acquisition of a full 96-well or 384-
well assay plate on an automated imaging platform depends
upon the number of fluorescent channels and images captured
per well and the focusing options selected. Scanning times may
range anywhere from 10–15 min to 2–3 h per plate depending
upon the complexity of the image acquisition procedure.
8. To control and reduce environmental exposure levels to form-
aldehyde and for long term storage of fixed assay plates, we
recommend the aspiration and washing of plates on an auto-
mated plate washing platform.
9. Although we have described the image acquisition process on the
ArrayScan VTI platform, most automated imaging platforms
designed for HCS would be capable of capturing these images.
10. Although we have described the image analysis of the molecular
translocation algorithm provided by the ArrayScan VTI plat-
form, most automated imaging platforms designed for HCS
provide similar image analysis programs that would be capable
of analyzing these images.
References
1. Dudgeon D, Shinde SN, Shun TY, Lazo JS, HIV-1 rev mutants in living cells. Virology
Strock CJ, Giuliano KA, Taylor DL, Johnston 251:38–48
PA, Johnston PA (2010) Characterization and 8. Trask O, Nickischer D, Burton A, Williams
optimization of a novel protein-protein inter- RG, Kandasamy RA, Johnston PA, Johnston
action biosensor HCS assay to identify disrup- PA (2009) High-throughput automated con-
tors of the interactions between p53 and focal microscopy imaging screen of a kinase-
hDM2. Assay Drug Dev Technol 8:437–458 focused library to identify p38 mitogen-
2. Dudgeon D, Shinde SN, Hua Y, Shun TY, activated protein kinase inhibitors using the
Lazo JS, Strock CJ, Giuliano KA, Taylor DL, GE InCell 3000 analyzer. Methods Mol Biol
Johnston PA, Johnston PA (2010) Implemen- 565:159–186
tation of a 220,000 compound HCS campaign 9. Trask OJ Jr, Baker A, Williams RG et al (2006)
to identify disruptors of the interaction Assay development and case history of a 32 K-
between p53 and hDM2, and characterization biased library high-content MK2-EGFP trans-
of the confirmed hits. J Biomol Screen location screen to identify p38 mitogen-
15:152–174 activated protein kinase inhibitors on the
3. Johnston PA, Shinde SN, Hua Y, Shun TY, ArrayScan 3.1 imaging platform. Methods
Lazo JS, Day BW (2012) Development and Enzymol 414:419–439
validation of a high-content screening assay to 10. Kumar S, Saradhi M, Chaturvedi NK, Tyagi
identify inhibitors of cytoplasmic dynein- RK (2006) Intracellular localization and
mediated transport of glucocorticoid receptor nucleocytoplasmic trafficking of steroid recep-
to the nucleus. Assay Drug Dev Technol tors: an overview. Mol Cell Endocrinol
10:432–456 246:147–156
4. Knauer S, Moodt S, Berg T, Liebel U, Pepper- 11. Stauber R (2002) Analysis of nucleocytoplas-
kok R, Stauber RH (2005) Translocation bio- mic transport using green fluorescent protein.
sensors to study signal-specific nucleo- Methods Mol Biol 183:181–198
cytoplasmic transport, protease activity and 12. Nickischer D, Laethem C, Trask OJ Jr et al
protein-protein interactions. Traffic (2006) Development and implementation of
6:594–606 three mitogen-activated protein kinase
5. Knauer S, Stauber RH (2005) Development of (MAPK) signaling pathway imaging assays to
an autofluorescent translocation biosensor sys- provide MAPK module selectivity profiling for
tem to investigate protein-protein interactions kinase inhibitors: MK2-EGFP translocation, c-
in living cells. Anal Chem 77:4815–4820 Jun, and ERK activation. Methods Enzymol
6. Lundholt B, Heydorn A, Bjorn SP, Praeste- 414:389–418
gaard M (2006) A simple cell-based HTS 13. Williams RG, Kandasamy R, Nickischer D et al
assay system to screen for inhibitors of p53- (2006) Generation and characterization of a
Hdm2 protein-protein interactions. Assay stable MK2-EGFP cell line and subsequent
Drug Dev Technol 4:679–688 development of a high-content imaging assay
7. Stauber R, Afonina E, Gulnik S, Erickson J, on the Cellomics ArrayScan platform to screen
Pavlakis GN (1998) Analysis of intracellular for p38 mitogen-activated protein kinase inhi-
trafficking and interactions of cytoplasmic bitors. Methods Enzymol 414:364–389
Chapter 38
Case Study: Discovery of Inhibitors of the MDM2–p53

Protein-Protein Interaction
Liu Liu, Denzil Bernard, and Shaomeng Wang
Abstract
The p53 protein, a tumor suppressor, is inactivated in many human cancers through mutations or by its
interaction with an oncoprotein, MDM2. Blocking the MDM2–p53 protein-protein interaction has the
effect of activating wild-type p53 and has been pursued as a novel anticancer strategy. Small-molecule
inhibitors of the MDM2–p53 interaction have been discovered through various approaches, and a number
of them have progressed into clinical trials for cancer treatment. Here, we describe the methods and
techniques used in the discovery of small-molecule inhibitors of the MDM2–p53 interaction.
Key words p53, MDM2, Protein-protein interaction, Cancer therapy, High-throughput screening,
Virtual screening
1 Introduction
Protein-protein interactions (PPI) regulate many biological pro-

cesses, and targeting those interactions that have been implicated in
human diseases has become an important strategy in drug discovery
[1, 2]. One such PPI target is the interaction between murine
double minute 2 (MDM2) and the tumor suppressor protein
p53. p53 is involved in processes such as regulation of the cell
cycle, apoptosis, DNA repair, and senescence [3–6]. The critical
role of p53 as a tumor suppressor protein is evident from the fact
that it is among the most frequently mutated proteins in human
tumors and alterations of its precursor p53 gene are seen in about
50 % of human cancers and lead to either inactivation or loss of p53
protein [7, 4]. In many cancers, wild-type p53 is effectively inhib-
ited by the MDM2 (HDM2 in humans) protein, which was origi-
nally discovered as an overexpressed protein in a mouse tumor cell
line [4, 7–10]. The interaction of MDM2 with p53 [11–13] was
shown to inhibit transactivation mediated by p53 [8] enhancing the
tumorigenic potential of cells overexpressing MDM2 [14–20].
Thus, overexpression of MDM2 in multiple types of human cancers
567
568 Liu Liu et al.
is related to their poor prognosis and response to therapy [14–19].

It was also demonstrated by the rescue of MDM2-null mice with
simultaneous deletion of the p53 gene [21, 22] that the
MDM2–p53 interaction is the crucial determinant in death asso-
ciated with the absence of MDM2. In addition, a study of over
3,000 tumors [18] revealed a negative correlation between MDM2
amplification and p53 mutations. Thus in cancers retaining wild-
type p53, the MDM2–p53 protein-protein interaction is an impor-
tant therapeutic target [23–25].
1.1 Regulation of p53 The interaction between MDM2 and p53 regulates the cellular
and MDM2 basal levels and activity of p53 through an autoregulatory feedback
loop (Fig. 1). The expression of MDM2 protein is induced by p53
which binds to the P2 promoter of the MDM2 gene leading to its
transcriptional activation. MDM2 binds to p53 protein and inhibits
its activity by various mechanisms such as inhibition of the transac-
tivation function of p53, export of p53 out of the nucleus, or direct
ubiquitination of p53 leading to its proteasomal degradation
[26–28]. Thus inhibition of the MDM2–p53 interaction can lead
to activation of p53 and control of its tumor suppressing capacity.
1.2 Structural Basis In 1996, the crystal structure of HDM2 in a complex with an
of the MDM2–p53 N-terminal fragment of the p53 protein was published [29], reveal-
Interaction ing the structural basis for the MDM2–p53 protein-protein
interaction. On the MDM2 protein surface, there was observed a
well-defined binding pocket involving mostly hydrophobic resi-
dues, including Met50, Leu54, Leu57, Gly58, Ile61, Met62, Tyr67,
His73, Val75, Phe91, Val93, His96, Ile99, and Tyr100, that comple-
mented the hydrophobic surface of the α-helical p53 peptide.
Interestingly, three residues, Phe19, Trp23, and Leu26, from the
p53 peptide were found to occupy this hydrophobic pocket, with
the formation of a hydrogen bond between the backbone carbonyl
of Leu54 in MDM2 and the indole nitrogen of Trp23 in p53
(Fig. 2). The compact binding pocket on the MDM2 protein
surface is suitable for the development of small-molecule com-
pounds blocking the MDM2 interaction with p53, and a variety
of such inhibitors have been reported in the literature and in
patents [30–32].
Fig. 1 MDM2–p53 autoregulatory feedback loop

Case Study: Discovery of Inhibitors of the MDM2–p53 Protein-Protein Interaction 569
Fig. 2 Crystal structure of MDM2 protein in complex with p53 peptide (PDB ID:
1YCR). The MDM2 protein surface is shown with the p53 peptide in a green
cartoon representation. The p53 residues Phe19, Trp23, and Leu26 are shown in
stick form. The hydrogen bond between Leu54 of MDM2 and Trp23 of p53 is
indicated by a dashed line
2 Discovery of Small-Molecule Inhibitors Targeting the MDM2–p53 Interaction
Since publication of the crystallographic structure indicating the

small, well-defined p53 binding pocket on MDM2, several classes
of small-molecule inhibitors have been reported. The most studied
representatives are cis-imidazolines [33], spiro-oxindoles [34, 35],
benzodiazepinediones [36–38], chromenotriazolopyrimidines
[39], m-chloropiperidinones [40], indolyl hydantoins [41], imida-
zothiazoles [42], terphenyl compounds [43, 44], quinolinols [45],
and sulfonamides [46]. In most cases the initial lead compounds
were discovered by application of protein-protein interaction
exploration methods which are discussed below in detail. Generally,
the methods applied may be classified as experimental (both high
and regular throughput) and computational (virtual) screenings.
2.1 Experimental Screening of large diverse chemical libraries using high-throughput

Screening of Both screening (HTS) methods to identify hit compounds is an approach
Large Diverse and commonly employed in protein-protein interaction studies and
Small Focused drug discovery. HTS methods are able to test millions of com-
Chemical Libraries pounds with largely diverse structural scaffolds, from which hit
compounds may be identified as starting points for further design
and optimization of chemical compounds. HTS is particularly use-
ful for those cases in which no detailed structural information of the
interacting interfaces is available. Screening small libraries with
certain preferred properties is also quite useful and may be more
570 Liu Liu et al.
valuable than HTS of large libraries. Structural insights regarding

the target or the nature of ligands binding to it can often permit
screening of a discreet segment of the chemical space, thereby
focusing the search on likely hits. In addition to methods utilized
for HTS, methods which cannot be adapted to an HTS format may
be developed and used here. These methods, unsuited to HTS, can
also be utilized as secondary assays to confirm hit compounds
identified by HTS. Both approaches have been utilized successfully
for the screening of nonpeptidic small molecules capable of dis-
rupting the MDM2–p53 interaction and have been responsible for
the discoveries of various lead compounds, including those men-
tioned above.
2.1.1 Surface Plasmon Surface plasmon resonance (SPR) currently is one of the most
Resonance (SPR) widely used label-free methods to evaluate protein–protein and
protein–small molecule interactions. In addition to affinities, it
can determine kinetic parameters, including kon and koff rates,
which provide additional useful information for protein-protein
interaction studies and drug discovery.
1. SPR as a primary assay
At Hoffmann-La Roche, lead compounds of the first class of potent
and selective small-molecule inhibitors of the MDM2–p53 interac-
tion, named Nutlins, were discovered by HTS of a diverse library of
synthetic compounds using SPR [47]. Although the throughput of
SPR is typically not high enough to be compatible with HTS,
researchers of Roche successfully integrated a competitive interac-
tion strategy into the SPR assay to increase its throughput. Instead
of directly evaluating the interaction between compounds and the
MDM2 protein, the interaction between MDM2 and p53 in the
presence of test compounds was evaluated to determine the inhibi-
tory abilities of those compounds. No details of this work were
provided however.
Materials The popular Biacore series of SPR instruments was initially intro-
duced by Sweden’s Biacore AB Corporation, which was purchased
by GE Healthcare in 2006. Currently, SPR instruments are avail-
able from a variety of manufacturers, such as Biosensing Instru-
ment, Sensia, Thermo, and Bio-Rad. Most of these are much less
expensive than those from GE/Biacore [48]. In the Roche study,
competitive assays were performed on a Biacore S51 SPR instru-
ment. CM5 sensor chips were used for protein immobilization.
Reagents included an EDC/NHS activation kit from GE Health-
care, PentaHis antibody from Qiagen, His-tagged p53 protein, and
MDM2 protein. The running buffer was 10 mM HEPES with
0.15 % NaCl and 2 % DMSO. The Evaluation software provided
with the instrument was used to fit the kinetic sensorgrams and
Microsoft Excel was used to calculate the IC50 values of the com-
pounds tested.
Methods In this study, the capture approach rather than the commonly used
direct immobilization approach was used to load p53 protein onto
the sensor chip. PentaHis antibody (Qiagen) was pre-immobilized
on the Biacore CM5 sensor chip by amine coupling chemistry,
which subsequently was able to capture the His-tagged p53 protein.
Since the location of the HisTag can be precisely controlled on
either the N- or C-terminus of the p53 protein, the protein orienta-
tion in the immobilization/capture process can be controlled and
the influence on the p53–MDM2 interacting model is minimized.
To assure data consistency, the level of p53 captured was controlled
at around 200 response units (RU; one RU corresponds to 1 pg of
protein per mm2). In all the assays, 300 nM of MDM2 protein was
included and 10 mM DMSO stock solutions of test compounds
were further serially diluted. The binding of immobilized p53 pro-
tein and 300 nM of MDM2 protein in the presence of test com-
pounds was determined as a percentage of binding in the absence of
inhibitor. IC50 values were calculated using Microsoft Excel.
Notes This class of Nutlin compounds, developed from hits identified

through competitive SPR, has a cis-imidazoline core structure.
Structure-based chemical modifications of the lead compound,
monitored by NMR [49], led to compounds with high affinity.
Among two possible enantiomers that were formed, only one was
active, Nutlin-3a (1, Fig. 3), which was isolated from racemic
Nutlin-3 and found to disrupt the MDM2–p53 interaction with
an IC50 value of 90 nM. Subsequent crystallization of MDM2 in a
complex with a Nutlin compound (PDB ID: 1RV1) showed that it
binds to the p53 binding site in MDM2. The interaction of the p53
peptide with MDM2 is mimicked by Nutlin-3a placing one of its
bromophenyl moieties in the Trp pocket, the other bromophenyl
group in the Leu pocket, and the ethyl ether side chain in the Phe
pocket. Nutlin-3 has been extensively evaluated for its therapeutic
potential and mechanism of action in human cancer. A derived
compound, RG7112 (2, Fig. 3), developed by Roche scientists,
binds to MDM2 with an IC50 value of 18 nM from a Homoge-
neous Time Resolved Fluorescence (HTFR) assay and was the first
MDM2–p53 inhibitor to enter clinical trials [50].
However, this competitive SPR method cannot determine the
actual Ki values of compounds tested and the IC50 values deter-
mined rely significantly on the experimental conditions. Thus, this
method is mainly suitable for parallel comparison, with assay con-
ditions carefully controlled by experienced operators; extreme cau-
tion needs to be taken when comparing results from different labs
or scientists.
2. SPR as a secondary assay
SPR is however an excellent method when used as a secondary assay
to confirm hit compounds identified from HTS. Researchers from
Amgen developed SPR direct binding methods which were able to
572 Liu Liu et al.
Fig. 3 Small-molecule inhibitors of the MDM2–p53 interaction discovered by experimental screening
accurately determine the Kd values of compounds. This led to the

development of two classes of hit compounds, chromenotriazolo-
pyrimidines [39] and m-chloropiperidinones [40], identified from
Homogeneous Time-Resolved Fluorescence (HTRF), that
potently target the p53–MDM2 interaction. Details of this HTRF
method and these two classes of compounds are discussed below.

For the SPR direct binding method, the p53 protein used in the
competitive SPR method discussed above was no longer required.
The binding target, MDM2 protein, was loaded on the sensor chip
and the interaction between the tested compounds and MDM2
protein was monitored directly.
Materials Assays were performed on the Biacore S51 SPR instrument and
CM5 sensor chips used just as in the Roche competitive assay. As
mentioned above, the major difference of materials used here was
that p53 protein was no longer needed. Both GST and AviTAG
labeled MDM2 proteins, goat anti-GST antibody, and Streptavidin
(SA) were used to prepare MDM2 surface on the sensor chips
through the capture approach. Other common reagents for SPR
included an EDC/NHS amine activation kit, acetate buffer for
Anti-GST antibody or SA immobilization, MDM2 protein capture,
and assay running buffer (25 mM Tris–HCl, pH 7.5, 150 mM
NaCl, 0.2 mM TCEP, and 0.005 % Tween 20). Instead of the
Evaluation software, Scrubber2 software (BioLogic Software Pty
Ltd.) was used to fit the kinetic data.
Methods CM5 chips were activated with the EDC/NHS activation kit fol-
lowing protocols provided by Biacore. Immediately after activation,
Anti-GST antibody or SA was immobilized on the CM5 chips
which would capture GST or AviTAG labeled MDM2 proteins to
prepare a stable MDM2 surface where interactions with small
molecules would take place. Serial dilutions of p53 peptide, as the
positive control, and compound stock solutions in DMSO were
further diluted to the desired concentration range in the assay
buffer. Compound working solutions were then injected over cap-
tured MDM2 surfaces. Two minutes of association and 3–4 min of
dissociation were monitored and raw data collected were fit, using
Scrubber2, to a typical 1:1 binding model from which Kd, kon, and
koff were obtained.
Notes Recently, kinetic parameters, particularly the dissociation rate con-

stants (koff) have captured the attention of quite a few researchers.
koff is the key component with the most potential to enhance
compound affinity against the target protein. Furthermore, besides
potent affinity, relatively slow off-rate of the compound is desired
for drug action as well. Significant progress has been made to
design and develop screening assays based on koff values.
Experimentally, direct binding assay is more straightforward
and relatively easier to be developed and optimized than the com-
petitive assay discussed above. However, sensorgram fitting may be
confusing and even misleading since quite a few matters, such as
compound aggregation, DMSO drift, especially non-specific inter-
actions while compound concentrations are relatively high, can
produce artificial SPR responses. Raw SPR sensorgrams need to
574 Liu Liu et al.
be carefully checked before performing kinetic fitting in order to

obtain accurate kon, and koff values.
2.1.2 Homogeneous Homogeneous time-resolved fluorescence, also known as Time-

Time-Resolved Resolved Fluorescence Resonance Energy Transfer (TR-FRET), is
Fluorescence (HTRF) one of the most widely used homogenous methods for primary
HTS and also for secondary confirmation. It is a quite simple and
versatile method, and due to the bounty of various robust fluores-
cent donor–acceptor combinations and protein labeling protocols,
it is suitable for most protein-protein interaction studies.
A class of chromenotriazolopyrimidine compounds was discov-
ered at Amgen using HTRF-based HTS of about 1.4 million
compounds. This screen detected the quenching by the com-
pounds tested of the FRET signal from fluorescent donor-labeled
MDM2 to acceptor labeled p53 protein. The initial hit compound,
chromenotriazolopyrimidine 3 (Fig. 3), a racemic mixture of the
syn- and anti-diastereoisomers was identified, with an IC50 of
3.88 μM [39]. The binding of this compound to MDM2 was
further confirmed by the SPR direct binding method as described
above. Further modifications and separation of isomers led to
compound 4 (Fig. 3), which has an IC50 of 3.0 μM in a luciferase
reporter assay.
Materials The essential materials for an HTRF assay are interacting partners
labeled with fluorescent donor and acceptor, respectively. In this
case, GST-MDM2 protein which would be recognized by anti-GST
antibody labeled with europium cryptate and Avi-p53 protein teth-
ered to SA-Xlent were designated as the donor–acceptor pair.
Besides these carefully designed fluorescently labeled proteins,
other materials included phosphate assay buffer, an Envision plate
reader, Perkin Elmer white 384 Opti plates, and some typical data
processing software. All materials are commonly needed for typical
in vitro biochemical assays and excellent evidence of the simplicity
of HTRF method.
Methods Generally, HTRF assays are performed through a competitive

approach in which compounds are tested for their ability to inhibit
the interaction between protein partners of interest. In the Amgen
screen, serial dilutions of compounds in DMSO were added to the
reaction plate followed by assay buffer, GST-MDM2 and Avi-p53
proteins. Reaction mixtures were incubated at room temperature
for 60 min during which equilibrium amongst the three compo-
nents, compound-MDM2–p53, was established. A detection mix-
ture composed of both SA-Xlent and Eu-anti-GST was added to
the reaction mixture. After incubation at room temperature for
18 h, the reaction mixture was transferred to assay plates, which
were read on an Envision plate reader with an excitation wavelength
of 320 nm that is specific for Eu. Emissions were measured at

665 nm and 615 nm, corresponding to Xlent and Eu, respectively,
and the ratio of Em665/Em615 represents the binding between
MDM2 and p53. IC50 values were calculated from the inhibition
curves obtained.
Notes Further co-crystallization studies on the anti-diastereoisomer of

compound 3 and the MDM2 protein illustrated that the three
hydrophobic pockets of MDM2, where key p53 residues bind,
were successfully occupied by the hydrophobic groups of com-
pound 3, indicating that this compound inhibits p53 competitively.
The discovery and validation of this new class of inhibitors was an
important step towards testing the hypothesis of the utility of
p53–MDM2 inhibition in the treatment of cancer. However,
some fundamental disadvantages of this class of compounds include
their chemical instability, poor physical properties, and moderate
affinities, which compromised their potential for development as
clinical lead compounds. Based on this class of compound, another
class of m-chloropiperidinone compounds was developed recently
by Amgen. The most potent AM-8553 (5, Fig. 3) had an IC50
value of 1.1 nM in the same HTRF assay and excellent pharmaco-
kinetic properties and in vivo efficacy [40].
Recently, researchers from Hoffmann-La Roche, using an
HTRF (TR-FRET) method, discovered a new class of small-
molecule inhibitors targeting the p53–MDM2 interaction [41].
Initial HTS was performed for the p53–MDMX interaction. An
indolyl hydantoin class was identified as a hit, and compounds in
this class were further evaluated in a p53–MDM2 HTRF assay. A
dual inhibitor, RO-5963 (6, Fig. 3), was discovered which targets
both MDM2 and MDMX with IC50 values to MDM2 and MDMX
of 17.3 nM and 24.7 nM, respectively. These cases demonstrated
that HTRF is versatile and efficient, quite straightforward to per-
form, and suitable for both HTS and secondarily for hit confirma-
tion. However, as in the competitive SPR method discussed above,
the most important questions concerning the HTRF method
include its inability to determine Ki values of compounds accurately
and the significant influence of assay conditions on the IC50 values
obtained. Again, data comparison between different laboratories or
different methods should be carried out with extra caution.
2.1.3 ThermoFluor ThermoFluor microcalorimetry is another widely used method for

Microcalorimetry protein–protein and protein–small molecule interaction studies.
Different from most other methods, which typically monitor the
change of a certain physical or chemical property of the protein or
complex before and after the introduction of compounds, Thermo-
Fluor monitors protein unfolding as a function of temperature.
Protein unfolding proceeds differently in the presence of small
576 Liu Liu et al.
molecules that interact with the protein. A fluorescent dye, which

exhibits different fluorescence intensities depending on the folding
status of the protein binding to it, is used in the assay. Compounds
binding to the target protein are detected by a consequent increase
in thermal stability, which is determined as the change in midpoint
transition temperature (ΔTm) upon the addition of the compound
at a specific concentration. The larger the shift in Tm the higher the
affinity of the binder. Due to its simplicity, this method works
perfectly as an HTS method enabling very high throughput.
A series of benzodiazepinedione-based MDM2 inhibitors were
discovered by researchers at Johnson and Johnson using
ThermoFluor-based HTS. This involved the use of ThermoFluor
microcalorimetry technology to screen a library of 22,000 benzo-
diazepinediones, designed using Directed Diversity software [37],
and also a library of 338,000 compounds developed by combinato-
rial chemistry [36].
Materials Based on the principles of this assay, an instrument which is able to

precisely control reaction temperatures and simultaneously moni-
tor fluorescence is required. Although initially a specially designed
ThermoFluor instrument was required, which was used in these
studies, it was subsequently demonstrated that a typical real-time
quantitative PCR (qPCR) worked well for the ThermoFluor
method [51, 52]. This discovery made ThermoFluor accessible in
most biochemical and biological laboratories and has significantly
enhanced the application of ThermoFluor.
Since the method directly monitors the interaction between the
target protein and a small molecule, MDM2 is the only protein
needed (p53 protein is not needed). Besides the required com-
pounds, the only reagent needed is an appropriate fluorescent
dye, dapoxyl sulfonic acid, which was used in studies from the
same group. Two other commonly used dyes are 1-anilinonaphtha-
lene-8-sulfonic acid (ANS) and SYPRO Orange, the latter being
used typically for assays performed with qPCR.
Methods The procedure of ThermoFluor is quite straightforward. In the

work done by Johnson and Johnson, purified MDM2 protein was
added to each well of a 384-well polypropylene plate followed by
the fluorescent probe, dapoxyl sulfonic acid, and the compound to
be assayed. After mixing, the plates were heated in the Thermo-
Fluor instrument to a maximum temperature of 85 C and the
fluorescence signal monitored at 500–530 nm with 8 and 12 bit
CCD cameras. In order to minimize evaporation, particularly at
higher temperatures, 5–10 μL of mineral oil may be added to each
well to overlay the reaction mixture. The midpoint transition tem-
perature (ΔTm) was obtained by fitting the transition curve of
fluorescence intensities versus reaction temperatures.
Notes Although ThermoFluor is a quick and easy method and feasible for
most laboratories, one must understand that the correlation
between melting temperature change obtained from ThermoFluor
and affinity is qualitative, and an increase in melting temperature
does not necessarily correspond to a specific interaction of interest.
A secondary specific and quantitative assay is required to obtain
accurate affinity information. In these studies, MDM2 binders were
tested with an FP-based peptide displacement binding assay (see
below) to identify bona fide inhibitors of the MDM2–p53 interac-
tion. This led to the identification of some benzodiazepinedione-
based MDM2 inhibitors: compounds 7 (IC50 ¼ 420 nM) and
8 (IC50 ¼ 490 nM) which disrupt the MDM2–p53 interaction,
and the optimized compound 9 whose Kd value is 80 nM (Fig. 3).
Crystal structures of 9 in a complex with MDM2 revealed that 9 fits
into the same pockets in the MDM2 binding cleft used by the p53
peptide side chains Phe19, Trp23, and Leu26. As with the Nutlins,
the interactions of these inhibitors with the MDM2 protein are
primarily through van der Waals contacts. Another series of benzo-
diazepinedione compounds identified by the same ThermoFluor
method followed by optimization led to TDP521252 (10) and
TDP665759 (11) (Fig. 3) which bind to MDM2 with IC50 values
of 708 nM and 704 nM, respectively [38].
2.1.4 Fluorescence The FP assay, which can involve a direct binding or a competitive
Polarization (FP) approach, is probably the most widely used method for screening
small molecules targeting protein-protein interactions. Through the
direct binding approach, Kd values can be easily determined accu-
rately by monitoring the change in FP of the fluorescently labeled
small molecule or peptide while binding to different concentrations
of protein. This approach is not suitable for HTS since it is not
practical to fluorescently label each compound tested. However, it
can be easily adapted to HTS by employing a competitive approach
in which compounds compete with a fluorescently labeled small-
molecule compound or peptide, typically termed a tracer, for bind-
ing to the target protein. In most cases, the Ki values of compounds
based on the experimental IC50 values obtained can be accurately
determined, as can the Kd value of the tracer to the target protein, the
latter being determined with the direct binding approach.
There are many successful cases reported of discovery and
development of MDM2–p53 interaction inhibitors using the FP
method. Among them, the most successful one probably is the
spiro-oxindole class of compounds developed at the University of
Michigan [34, 35, 53, 54]. Starting from the crystal structure of
the MDM2 protein complexed with the p53 peptide, a structure-
based de novo rational strategy was employed to design new classes
of small-molecule inhibitors targeting the MDM2–p53 interaction.
With the assistance of computational screening, a spiro-oxindole
578 Liu Liu et al.
core structure was discovered that mimics the interaction between

the p53 peptide and the MDM2 protein, which subsequently was
used as the starting point for rational design. A sensitive and
quantitative FP-based binding assay was designed and developed
to determine the accurate binding affinities of newly synthesized
compounds. Accurate SAR data guided the de novo design pre-
cisely. Starting from compound MI-5 (12, Fig. 3) with a Ki value of
8.5 μM, a new class of highly potent, small-molecule inhibitors,
such as MI-219 (13, Fig. 3) and MI-147 (14, Fig. 3) with Ki
values of 13 nM and 0.6 nM, respectively, were obtained. A fully
optimized compound, SAR405838, in this class entered clinical
trials in 2012.
Materials Since the FP assay was performed as a competitive approach, in

addition to the target MDM2 protein, a fluorescently labeled mod-
ified p53 peptide serving as the tracer was the most important
reagent in the competitive assay [55]. Other materials included
phosphate assay buffer, typical Microtiter assay plates, a plate reader
capable of reading FP, and typical data processing software, such as
Prism, all of which are common in most biochemical laboratories.
Compared to the similar HTRF method discussed above, the FP
assay is even easier to perform since only one fluorescently labeled
component is required and this molecule is typically a small peptide
or organic molecule which is much more easily labeled than pro-
teins used in the HTRF method.
Methods One of the major advantages of the FP method is that Ki values of

compounds tested can generally be determined accurately. In order
to calculate Ki, the Kd value of the tracer to the target protein must
first be determined. This can be achieved by monitoring the total
FP of mixtures containing the tracer at a fixed concentration and
the target protein at increasing concentrations up to full saturation.
Thus, serial dilutions of the MDM2 protein in the assay buffer
(100 mM potassium phosphate, pH 7.5, 100 μg/ml bovine γ-
globulin, 0.02 % sodium azide, with 0.01 % Triton X-100 and 4 %
DMSO) were prepared in a 96-well black assay plate followed by
addition of the tracer solution with a fixed concentration in each
well. The wells were mixed and incubated at room temperature for
60 min with gentle shaking to ensure equilibrium. The FP values in
millipolarization units (mP) were measured at an excitation wave-
length of 485 nm and an emission wavelength of 530 nm, with
Fluorescein amidite (FAM) as the fluorophore. Equilibrium disso-
ciation constants (Kd) were calculated with Graphpad Prism 5.0
software (Graphpad Software, San Diego, CA) by measuring the
sigmoidal dose-dependent FP increases as a function of protein
concentrations.
The Ki values of compounds tested were determined in a dose-

dependent competitive binding experiment. Mixtures of serial dilu-
tions of compounds dissolved in DMSO with preincubated
MDM2/tracer complex at fixed concentrations in the assay buffer
were added into assay plates and incubated at room temperature
with gentle shaking. Negative controls containing MDM2/tracer
complex only (equivalent to zero inhibition), and positive controls
containing only free tracer (equivalent to 100 % inhibition),
were included in each assay plate. Using Prism, IC50 values
were determined by nonlinear regressive fitting of the sigmoidal
dose-dependent FP decreases as a function of total compound
concentrations. The Ki values of tested compounds were calculated
using the measured IC50 values, the Kd value of the tracer, and
the concentrations of the MDM2 protein and the tracer in the
competitive assay [56].
Notes As previously mentioned, Ki values can be calculated in most cases.

However, when the compound is more potent than the tracer used,
the Ki value cannot be determined accurately since under this
circumstance, the experimental IC50 value is not correlated linearly
with Ki [57]. Mathematical solutions have been proposed to deal
with this issue [57–60] but none of them performs perfectly with
extremely potent compounds. Among all these interpretations, that
proposed by Zhang [60] utilized a unique approach in which Ki
values were obtained by directly fitting the inhibition curves with-
out involving IC50 values. This approach was beneficial in some
cases in which calculations based on the IC50 produced meaningless
negative values.
FP was also used by other laboratories in which different classes
of small-molecule inhibitors targeting MDM2–p53 interaction
were discovered. The p53 peptide binds to MDM2 in an amphi-
phatic manner via an alpha helix that locates to the hydrophobic
residues in the binding pocket. Using this knowledge, a small
library of terphenyl compounds which can mimic the alpha helical
peptide [61] was developed and screened using the FP method. By
substituting alkyl or aryl groups at the o-positions of the terphenyl
scaffold, a small library of 21 terphenyl derivatives was synthesized
[43]. The most potent of these, compound 15 (Fig. 3) had a Ki
value with MDM2 of 182 nM. In addition to the displacement of
p53 peptide in the FP assay, 15N HSQC NMR spectroscopy
showed that the terphenyl 15 interacts with the p53 binding region
of MDM2.
In a patent, Novartis disclosed compounds from an imidazole-
indole class as inhibitors of MDM2 and MDMX. The most potent
compound has an IC50 of 15 nM for MDM2 in an FP binding assay
and interestingly also binds to MDMX with IC50 ¼ 1.32 μM in a
TR-FRET binding assay [62].
580 Liu Liu et al.
2.1.5 ELISA, Isothermal Other methods to study protein–protein and protein–small mole-
Titration Calorimetry (ITC), cule interactions were used successfully on the MDM2–p53 inter-
and NMR action. These included the ELISA, ITC, and NMR methods.
Compounds 16 and 17 (Fig. 3), which inhibit the MDM2–p53
interaction with IC50 values of 10 μM and 15 μM, respectively, are
terphenyl compounds which were identified using an in vitro,
quantitative, ELISA-based assay as a screen [44]. The screening of
a small library of 16 chalcones, compounds known to have antican-
cer properties [63], with a similar ELISA-based assay monitoring
disruption of the MDM2–p53 interaction resulted in the identifi-
cation of two compounds 18 and 19 (Fig. 3) [64] which have IC50
values of 49 μM and 117 μM, respectively. NMR titration experi-
ments were used to confirm the binding, and the compounds were
also found to inhibit the MDM2–p53 interaction in an in vitro
DNA binding electrophoretic mobility super-shift assay (EMSA).
Recently, researchers from Daiichi Sankyo developed an imida-
zothiazole class of potent MDM2 inhibitors starting from the
core structure of the Nutlins. Based on extensive SAR information
obtained from ELISA, the most potent compound 20 (Fig. 3) had
an IC50 value of 1.2 nM in their ELISA assay [65].
ITC, another quite popular and unique method for protein–
protein and protein–small molecule interaction studies, is the only
100 % real homogeneous and label-free method available. In addi-
tion to the binding affinities, it can also produce very useful infor-
mation on the stoichiometry and ΔH of the interaction of interest,
data which are obtained only with great difficulty by any other
method. However, it also has huge disadvantages over other meth-
ods. The most important of these are the significantly large amount
of protein that is consumed and the very low throughput that is
obtained. Thus the most suitable application for ITC is its use as a
confirmatory assay for fully optimized compounds that are ready to
enter the next research stage. Researchers from Amgen designed
and developed a robust ITC assay and confirmed their new class of
MDM2 inhibitors that were identified by HTRF, as discussed
above. In this way, novel structural features were discovered of
the MDM2–small molecule interaction model involving the
MDM2 N-terminal residues that had previously been regarded as
“structureless” [66].
2.2 Virtual Database In contrast to direct experimental screening of chemical libraries,

Screening when the structural basis of the interaction of the target proteins is
known, computational database screening of millions of com-
pounds can be performed to produce a select set of compounds
for further experimental study.
2.2.1 Pharmacophore Pharmacophores represent the groups in a molecule and their

Screen spatial juxtaposition that are important for its interaction with a
target protein. Using different sets of available data, namely (a) p53
mutagenesis data affecting MDM2 binding [67], (b) truncation

studies of peptidic inhibitors [68], and (c) peptidic inhibitors with
unnatural amino acids [55], multiple pharmacophore models can
be generated and used to screen for hits [46]. In a typical study of
this type, the compounds selected were then tested for inhibition of
MDM2–p53 interaction with an ELISA-based assay. A sulfon-
amide, NSC 279287 (21, Fig. 4), was identified and found to
inhibit the interaction between full-length MDM2 and p53 pro-
teins with an IC50 value of 31.8 μM. However, in a p53 reporter
gene assay only a 20 % increase in p53 transcriptional function was
seen after treatment with 100 μM of NSC 279287, indicating its
low potency [46].
A set of MDM2–p53 inhibitors based on a pharmacophore
model developed from the crystal structure of MDM2 was
obtained by molecular dynamics simulation, which considered pro-
tein flexibility [69]. Hits obtained from pharmacophore screening
of ~50,000 synthetic compounds were tested in a fluorescence-
polarization MDM2 binding assay. This led to the discovery of
five nonpeptidic, small-molecule inhibitors, the most potent of
which, 22 (Fig. 4), had a Ki of 110 nM.
2.2.2 Combined A group of 2,599 compounds was obtained by application of a

Pharmacophore and pharmacophore screen to a database of over 100,000 compounds,
Structure-Based Screening prefiltered for drug-like properties and including both synthetic
and natural products [45]. In this case, the pharmacophore was
derived from the MDM2–p53 crystal complex and known small-
molecule inhibitors such as the Nutlins and benzodiazepinediones.
The hits from the first screen were then subjected to a second screen
using structure-based docking with the GOLD program [70, 71]
to rank them by their predicted binding affinity to the p53 binding
site. After rescoring with Chemscore [72] and X-Score [73], the
top 200 compounds were then combined to obtain a set of 354
non-redundant compounds, each of which was complementary to
the MDM2 target site. From a final set of 67 of these compounds,
tested with a competitive FP MDM2 binding assay, ten hits were
identified as compounds with a Ki value <10 μM. The most potent
of these, NSC 66811 (23, Fig. 4), had a Ki value of 120 nM and
dose-dependently activated p53 in LNCaP prostate cancer cells.
Fig. 4 Small-molecule inhibitors of the MDM2–p53 interaction discovered by virtual database screening
582 Liu Liu et al.
3 Concluding Remarks
Several small-molecule inhibitors targeting the MDM2–p53

protein-protein interaction have been successfully developed in
the last decade with the assistance of various biochemical and
biophysical methods including both experimental and virtual data-
base screenings. Among them, RG7112 and RG7388, two Nutlin
derivatives from Roche, AMG 232 from Amgen, and MI-77301/
SAR405838, a spiro-oxindole discovered from our laboratory at
the University of Michigan, are examples that have advanced to
clinical trials. Targeting the MDM2–p53 protein-protein interac-
tion using small molecules which reactivate the p53 is a potentially
attractive cancer therapeutic strategy for the treatment of human
cancers retaining wild-type p53. Clinical testing of these new agents
will provide the ultimate test of the usefulness of this therapeutic
strategy for the treatment of cancer and the critical role of PPI in
key human disease-related biological processes.
These successful cases are excellent examples of the applications
of some most widely used methods, which are valuable for those
interested in other protein–protein and protein–small molecule
interactions. To develop methods successfully, typical criteria that
must be considered include assay throughput, the information that
each method can provide, the suitability of each technology to the
protein–protein and protein–small molecule interactions of inter-
est, and cost. Some other essential issues that need also to be
considered include accuracy, reproducibility, and the relative ease
of performance.
Acknowledgements
Funding from the National Cancer Institute of the National Insti-

tutes of Health is greatly appreciated.
Disclosure statement
S.W. and the University of Michigan own equity in Ascenta Thera-

peutics, Inc., which has licensed the technologies related to the
MDM2 inhibitors of the spiro-oxindole class. S.W. serves as a
consultant for Ascenta.
References
1. Fry DC, Vassilev LT (2005) Targeting protein- 2. Murray JK, Gellman SH (2007) Targeting
protein interactions for cancer therapy. J Mol protein-protein interactions: lessons from
Med 83:955–963 p53/MDM2. Biopolymers 88:657–686
3. Fridman JS, Lowe SW (2003) Control of apo- implications for chemotherapy. Curr Cancer
ptosis by p53. Oncogene 22:9030–9040 Drug Targets 5:27–41
4. Hainaut P, Hollstein M (2000) p53 and human 18. Momand J, Jung D, Wilczynski S et al (1998)
cancer: the first ten thousand mutations. Adv The MDM2 gene amplification database.
Cancer Res 77:81–137 Nucleic Acids Res 26:3453–3459
5. Vogelstein B, Lane D, Levine AJ (2000) Surf- 19. Gunther T, Schneider-Stock R, Hackel C et al
ing the p53 network. Nature 408:307–310 (2000) Mdm2 gene amplification in gastric
6. Vousden KH, Lu X (2002) Live or let die: the cancer correlation with expression of Mdm2
cell’s response to p53. Nat Rev Cancer protein and p53 alterations. Mod Pathol
2:594–604 13:621–626
7. Feki A, Irminger-Finger I (2004) Mutational 20. Bond GL, Hu W, Levine AJ (2005) MDM2 is a
spectrum of p53 mutations in primary breast central node in the p53 pathway: 12 years and
and ovarian tumors. Crit Rev Oncol Hematol counting. Curr Cancer Drug Targets 5:3–8
52:103–116 21. Jones SN, Roe AE, Donehower LA et al (1995)
8. Momand J, Zambetti GP, Olson DC et al Rescue of embryonic lethality in Mdm2-
(1992) The mdm-2 oncogene product forms a deficient mice by absence of p53. Nature
complex with the p53 protein and inhibits p53- 378:206–208
mediated transactivation. Cell 69:1237–1245 22. de Oca M, Luna R, Wagner DS, Lozano G
9. Fakharzadeh SS, Trusko SP, George DL (1995) Rescue of early embryonic lethality in
(1991) Tumorigenic potential associated with mdm2-deficient mice by deletion of p53.
enhanced expression of a gene that is amplified Nature 378:203–206
in a mouse tumor cell line. EMBO J 23. Vassilev LT (2007) MDM2 inhibitors for can-
10:1565–1569 cer therapy. Trends Mol Med 13:23–31
10. Fakharzadeh SS, Rosenblum-Vos L, Murphy M 24. Chene P (2003) Inhibiting the p53-MDM2
et al (1993) Structure and organization of interaction: an important target for cancer
amplified DNA on double minutes containing therapy. Nat Rev Cancer 3:102–109
the mdm2 oncogene. Genomics 15:283–290 25. Shangary S, Wang S (2008) Targeting the
11. Lane DP, Crawford LV (1979) T antigen is MDM2-p53 interaction for cancer therapy.
bound to a host protein in SV40-transformed Clin Cancer Res 14:5318–5324
cells. Nature 278:261–263 26. Freedman DA, Wu L, Levine AJ (1999) Func-
12. Linzer DI, Levine AJ (1979) Characterization of tions of the MDM2 oncoprotein. Cell Mol Life
a 54 K dalton cellular SV40 tumor antigen pres- Sci 55:96–107
ent in SV40-transformed cells and uninfected 27. Juven-Gershon T, Oren M (1999) Mdm2: the
embryonal carcinoma cells. Cell 17:43–52 ups and downs. Mol Med 5:71–83
13. DeLeo AB, Jay G, Appella E et al (1979) 28. Wu X, Bayle JH, Olson D et al (1993) The
Detection of a transformation-related antigen p53-mdm-2 autoregulatory feedback loop.
in chemically induced sarcomas and other Genes Dev 7:1126–1132
transformed cells of the mouse. Proc Natl 29. Kussie PH, Gorina S, Marechal V et al (1996)
Acad Sci U S A 76:2420–2424 Structure of the MDM2 oncoprotein bound to
14. Bond GL, Hu W, Bond EE et al (2004) A the p53 tumor suppressor transactivation
single nucleotide polymorphism in the domain. Science 274:948–953
MDM2 promoter attenuates the p53 tumor 30. Millard M, Pathania D, Grande F et al (2011)
suppressor pathway and accelerates tumor for- Small-molecule inhibitors of p53-MDM2
mation in humans. Cell 119:591–602 interaction: the 2006-2010 update. Curr
15. Oliner JD, Kinzler KW, Meltzer PS et al (1992) Pharm Des 17:536–559
Amplification of a gene encoding a p53- 31. Zhan C, Lu W (2011) Peptide activators of the
associated protein in human sarcomas. Nature p53 tumor suppressor. Curr Pharm Des
358:80–83 17:603–609
16. Zhou M, Gu L, Abshire TC et al (2000) Inci- 32. Zak K, Pecak A, Rys B et al (2013) Mdm2 and
dence and prognostic significance of MDM2 MdmX inhibitors for the treatment of cancer: a
oncoprotein overexpression in relapsed child- patent review (2011–present). Expert Opin
hood acute lymphoblastic leukemia. Leukemia Ther Pat 23:425–448
14:61–67
33. Vassilev LT (2004) Small-molecule antagonists
17. Rayburn E, Zhang R, He J et al (2005) MDM2 of p53-MDM2 binding: research tools and
and human malignancies: expression, clinical potential therapeutics. Cell Cycle 3:419–421
pathology, prognostic markers, and
584 Liu Liu et al.
34. Ding K, Lu Y, Nikolovska-Coleska Z et al database screening strategy. J Med Chem

(2006) Structure-based design of spiro- 49:3759–3762
oxindoles as potent, specific small-molecule 46. Galatin PS, Abraham DJ (2004) A nonpeptidic
inhibitors of the MDM2-p53 interaction. J sulfonamide inhibits the p53-mdm2 interac-
Med Chem 49:3432–3435 tion and activates p53-dependent transcription
35. Ding K, Lu Y, Nikolovska-Coleska Z et al in mdm2-overexpressing cells. J Med Chem
(2005) Structure-based design of potent non- 47:4163–4165
peptide MDM2 inhibitors. J Am Chem Soc 47. Vassilev LT, Vu BT, Graves B et al (2004) In
127:10130–10131 vivo activation of the p53 pathway by small-
36. Grasberger BL, Lu T, Schubert C et al (2005) molecule antagonists of MDM2. Science
Discovery and cocrystal structure of benzodia- 303:844–848
zepinedione HDM2 antagonists that activate 48. Schasfoort RBM, Tudos AJ (2008) Handbook
p53 in cells. J Med Chem 48:909–912 of surface plasmon resonance. Royal Society of
37. Parks DJ, Lafrance LV, Calvo RR et al (2005) Chemistry Publishing, Cambridge, UK
1,4-Benzodiazepine-2,5-diones as small mole- 49. Fry DC, Graves B, Vassilev LT (2005) Devel-
cule antagonists of the HDM2-p53 interac- opment of E3-substrate (MDM2-p53)-bind-
tion: discovery and SAR. Bioorg Med Chem ing inhibitors: structural aspects. Methods
Lett 15:765–770 Enzymol 399:622–633
38. Koblish HK, Zhao S, Franks CF et al (2006) 50. Vu B, Wovkulich P, Pizzolato G et al (2013)
Benzodiazepinedione inhibitors of the Hdm2: Discovery of RG7112: A small-molecule
p53 complex suppress human tumor cell pro- MDM2 inhibitor in clinical development.
liferation in vitro and sensitize tumors to doxo- ACS Med Chem Lett 4:466–469
rubicin in vivo. Mol Cancer Ther 5:160–169 51. Ericsson UB, Hallberg BM, Detitta GT et al
39. Allen JG, Bourbeau MP, Wohlhieter GE et al (2006) Thermofluor-based high-throughput
(2009) Discovery and optimization of chrome- stability optimization of proteins for structural
notriazolopyrimidines as potent inhibitors of studies. Anal Biochem 357:289–298
the mouse double minute 2-tumor protein 53 52. Lo MC, Aulabaugh A, Jin G et al (2004) Eval-
protein–protein interaction. J Med Chem uation of fluorescence-based thermal shift
52:7044–7053 assays for hit identification in drug discovery.
40. Rew Y, Sun D, Gonzalez-Lopez De Turiso F Anal Biochem 332:153–159
et al (2012) Structure-based design of novel 53. Shangary S, Qin D, McEachern D et al (2008)
inhibitors of the MDM2-p53 interaction. J Temporal activation of p53 by a specific
Med Chem 55:4936–4954 MDM2 inhibitor is selectively toxic to tumors
41. Graves B, Thompson T, Xia M et al (2012) and leads to complete tumor growth inhibi-
Activation of the p53 pathway by small-mole- tion. Proc Natl Acad Sci U S A 105:3933–3938
cule-induced MDM2 and MDMX dimeriza- 54. Yu S, Qin D, Shangary S et al (2009) Potent
tion. Proc Natl Acad Sci U S A and orally active small-molecule inhibitors of
109:11788–11793 the MDM2-p53 interaction. J Med Chem
42. Miyazaki M, Kawato H, Naito H et al (2012) 52:7970–7973
Discovery of novel dihydroimidazothiazole 55. Garcia-Echeverria C, Chene P, Blommers MJ
derivatives as p53-MDM2 protein-protein et al (2000) Discovery of potent antagonists of
interaction inhibitors: synthesis, biological the interaction between human double minute
evaluation and structure-activity relationships. 2 and tumor suppressor p53. J Med Chem
Bioorg Med Chem Lett 22:6338–6342 43:3205–3208
43. Yin H, Lee GI, Park HS et al (2005) 56. Nikolovska-Coleska Z, Wang R, Fang X et al
Terphenyl-based helical mimetics that disrupt (2004) Development and optimization of a
the p53/HDM2 interaction. Angew Chem Int binding assay for the XIAP BIR3 domain
Ed Engl 44:2704–2707 using fluorescence polarization. Anal Biochem
44. Chen L, Yin H, Farooqi B et al (2005) p53 332:261–273
alpha-helix mimetics antagonize p53/MDM2 57. Huang X (2003) Fluorescence polarization
interaction and activate p53. Mol Cancer Ther competition assay: the range of resolvable inhib-
4:1019–1025 itor potency is limited by the affinity of the
45. Lu Y, Nikolovska-Coleska Z, Fang X et al fluorescent ligand. J Biomol Screen 8:34–38
(2006) Discovery of a nanomolar inhibitor of 58. Kenakin TP (1993) Pharmacologic analysis of
the human murine double minute 2 (MDM2)- drug-receptor interaction, 2nd edn. Raven,
p53 interaction through an integrated, virtual New York
59. Munson PJ, Rodbard D (1988) An exact cor- 67. Lin J, Chen J, Elenbaas B et al (1994) Several
rection to the “Cheng-Prusoff” correction. J hydrophobic amino acids in the p53 amino-
Recept Res 8:533–546 terminal domain are required for transcrip-
60. Zhang R, Mayhood T, Lipari P et al (2004) tional activation, binding to mdm-2 and the
Fluorescence polarization assay and inhibitor adenovirus 5 E1B 55-kD protein. Genes Dev
design for MDM2/p53 interaction. Anal Bio- 8:1235–1246
chem 331:138–146 68. Bottger V, Bottger A, Howard SF et al (1996)
61. Kutzki O, Park HS, Ernst JT et al (2002) Identification of novel mdm2 binding peptides
Development of a potent Bcl-x(L) antagonist by phage display. Oncogene 13:2141–2147
based on alpha-helix mimicry. J Am Chem Soc 69. Bowman AL, Nikolovska-Coleska Z, Zhong H
124:11838–11839 et al (2007) Small molecule inhibitors of the
62. Boettcher A, Buschmann N, Furet P, et al. MDM2-p53 interaction discovered by
(2008) 3-imidazolyl-indoles for the treatment ensemble-based receptor models. J Am Chem
of proliferative diseases. US Patent WO Soc 129:12809–12814
2008119741 70. Jones G, Willett P, Glen RC et al (1997) Devel-
63. Go ML, Wu X, Liu XL (2005) Chalcones: an opment and validation of a genetic algorithm
update on cytotoxic and chemoprotective for flexible docking. J Mol Biol 267:727–748
properties. Curr Med Chem 12:481–499 71. Verdonk ML, Cole JC, Hartshorn MJ et al
64. Stoll R, Renner C, Hansen S et al (2001) Chal- (2003) Improved protein-ligand docking
cone derivatives antagonize interactions using GOLD. Proteins 52:609–623
between the human oncoprotein MDM2 and 72. Eldridge MD, Murray CW, Auton TR et al
p53. Biochemistry 40:336–344 (1997) Empirical scoring functions: I. The
65. Uoto K, Kawato H, Sugimoto Y, et al. (2009) development of a fast empirical scoring func-
Imidazothiazole derivative having 4,7-diazas- tion to estimate the binding affinity of ligands
piro[2.5]octane ring structure. US Patent in receptor complexes. J Comput Aided Mol
WO/2009/151069 Des 11:425–445
66. Michelsen K, Jordan JB, Lewis J et al (2012) 73. Wang R, Lai L, Wang W (2002) Further devel-
Ordering of the N-terminus of human MDM2 opment and validation of empirical scoring
by small molecule inhibitors. J Am Chem Soc functions for structure-based binding affinity
134:17059–17067 prediction. J Comput Aided Mol Des 16:11–26
Chapter 39
Biophysical Methods for Identifying Fragment-Based

Inhibitors of Protein-Protein Interactions
Samuel J. Pfaff, Michael S. Chimenti, Mark J.S. Kelly,
and Michelle R. Arkin
Abstract
Fragment-based lead discovery complements high-throughput screening and computer-aided drug design
for the discovery of small-molecule inhibitors of protein-protein interactions. Fragments are molecules with
molecular masses ca 280 Da or smaller, and are generally screened using structural or biophysical
approaches. Several methods of fragment-based screening are feasible for any soluble protein that can be
expressed and purified; specific techniques also have size limitations and/or require multiple milligrams of
protein. This chapter describes some of the most common fragment-discovery methods, including surface
plasmon resonance, nuclear magnetic resonance, differential scanning fluorimetry, and X-ray
crystallography.
Key words Fragment, Ligand, Discovery, Biophysics, SPR, NMR, DSF, Crystallography, Protein-
protein, Interaction
1 Introduction
Fragment-based lead discovery (FBLD) is an increasingly popular

screening modality for discovering small-molecule inhibitors of
protein-protein interactions (PPI) [1–6]. “Fragments” are organic
compounds roughly ½ the size of a drug-like molecule, with
molecular masses roughly 120–280 Da; fragment-screening
libraries range in size from ca 500 to 15,000 members. Fragments
are typically screened using a structural or physical method that
identifies binders to the protein of interest. These binders are then
optimized using structure-guided design; in favorable cases, frag-
ments that bind to adjacent sites are linked together to create a
tight-binding lead compound.
There are several potential advantages to FBLD for PPI. First,
small-molecule inhibitors of PPI are likely rare in a standard high-
throughput screening (HTS) library because chemical space is
vastly larger than an HTS library (1060 vs. 106 members) and
587
588 Samuel J. Pfaff et al.
selection criteria could be biased towards previously “drugged”

targets. Since chemical space is smaller (ca 106) for fragment-
sized compounds, fragment libraries can sample chemical space
more broadly. Second, the binding surfaces of PPI tend to comprise
3–5 subsites, rather than one large, deep cavity, as is observed for
many enzymes [7]. Because the binding surfaces are complex, it is
difficult to satisfy all the necessary binding interactions through
random screening. By contrast, a fragment just needs to fill one
subsite. Third, PPI tend to contain regions of structural flexibility,
and fragments have been particularly successful at selecting such
regions. This chapter provides procedures for some of the most
common technologies for fragment screening. For more informa-
tion on the design of fragment libraries and chemical optimization
of fragments, the reader is referred to published reviews [8–10].
FBLD screens usually employ binding-based methods such as
surface plasmon resonance (SPR; biosensor) and nuclear magnetic
resonance (NMR), protein-stabilization methods such as differen-
tial scanning fluorimetry (DSF), or structure-based methods for
which X-ray crystallography is the gold standard. These biophysical
approaches circumvent many of the artifacts common in high-
throughput screening; in fact, the procedures outlined here are
highly recommended for secondary assays following an inhibition-
or activity-based screen. In contrast to activity-based screens, bind-
ing methods provide direct evidence of target engagement, and
usually also provide important details about binding stoichiometry
(generally, one fragment/protein is desired), binding kinetics, and
binding affinity. In the case of protein-detected NMR, Tethering
[11, 12], and X-ray crystallography, the binding site on the protein
is also known; X-ray provides the most detailed characterization of
the protein/fragment interaction.
The ideal methodology depends on the size of the library, the
size and availability of protein, and the existence of NMR and/or
X-ray structures. At least one of the methods here should be
suitable for a given protein-of-interest, provided that the protein
can be expressed and purified on the milligram scale.
1.1 Surface Plasmon Surface plasmon resonance (SPR) is a sensitive biophysical

Resonance technique that provides binding affinities, kinetics, and stoichiom-
etry for PPI and protein/small-molecule interactions. SPR and
biolayer interferometry belong to a broader class of technologies
known as biosensors. These techniques aim to measure the kinetics
and stoichiometry of a bimolecular interaction through physical
changes at a measurement surface. An SPR assay commonly con-
sists of a protein tethered to a gold surface by a flexible chemical
layer, such as dextran. The tethered protein remains under a steady
flow of buffer, and is interrogated by injection of small-molecule or
protein “analyte.” As analyte binds to the protein surface, the
refractive index (RI) of the buffer above the gold layer changes,
Fragment-Based PPI Inhibitor Discovery 589
and this change in RI is measured. After an analyte injection is

complete, buffer is flowed over the surface, allowing monitoring
of the analyte’s dissociation.
Binding analysis by SPR makes use of kinetic and equilibrium
measurements. Most SPR instruments measure the signal at a rate of
1–10 Hz, allowing real-time measurement of complex formation
(association) and dissociation. Importantly, the change in RI is
directly proportional to the mass of analyte that accumulates on
the surface; thus, by knowing the amount of material immobilized
on the surface and the molecular weights of the protein and analyte,
one can calculate the stoichiometry of the binding interaction. The
binding affinity of the complex is then determined by measuring the
binding stoichiometry at multiple doses or from fitting binding/
dissociation kinetics. Fragments tend to bind to proteins with KD
values in the 0.1–10 mM range. For these low-affinity interactions,
there is limited kinetic information in the data, and equilibrium is
reached within ~1 s. Fragments are therefore typically measured at
multiple doses, and KD is determined by binding isotherm (Fig. 1).
The power of SPR for fragment discovery lies in its very high
sensitivity and the ability to derive quantitative binding informa-
tion. In addition, SPR requires only a few milligrams of protein,
and modern instruments have sufficient throughput to monitor
~103–104 binding interactions in 1 week. GE Healthcare’s Biacore
4000 instrument, for instance, can screen four independent targets
against four compounds in a single 3–5 min injection cycle, and
holds ten 384-well plates plus control solutions. Other instru-
ments, such as Biacore’s T200, ForteBio’s OctetRed, Pioneer’s
SensiQ, and Bio-Rad’s Proteon offer somewhat different capabil-
ities in terms of throughput, automation, sensitivity, cost, and
innovative features.
SPR has also found widespread use as a secondary assay for
prioritizing compounds from HTS campaigns. Aggregation, cova-
lent binding, and nonspecific binding are easy to identify by SPR,
since such mechanisms show high stoichiometry, low selectivity,
and/or slow kinetics. Thus, SPR is very sensitive to misbehavior
by small molecule analytes [13]. These non-desirable binding
modes commonly generate false positives in high-throughput
biochemical or cell-based screens, and can lead to a loss of time
and resources if incorrectly prioritized [14].
1.2 Differential Differential scanning fluorimetry (DSF; alternatively called thermal

Scanning Fluorimetry shift or ThermoFluor) is an inexpensive and flexible method that
has recently become commonplace in biophysical drug screening.
DSF uses a thermocycler and fluorescent dyes to provide a measure
of the thermal stability of a target protein. In the context of drug
screening, DSF is used to identify compounds that increase the
thermal stability of a protein (Fig. 2). This thermal shift between
apo and bound protein forms is proportional to the change in
b c 80
80
60
Response (RU)
60
40 40
RU
20
20
0
0
0 50 100 0 250 500 750 1000 1250
Time (seconds) [Fragment (µM)]
Fig. 1 Fragment screening by SPR. (a) Primary screening data (courtesy of Stacie Bulfer). Green circles are
positive controls and black squares are fragments screened at 250 μM. The red line represents zero response
and the blue dashed line represents the hit cutoff of three standard deviations. (b) Fragment hit follow-up:
duplicate dose–response data for a fragment hit. This compound shows canonical fragment binding kinetics
with rapid approaches to equilibrium at the beginning and end of each injection. The injection start is indicated
by a closed arrow and the injection end (dissociation phase) is indicated by an open arrow. The data used to
generate equilibrium binding values are the averages of the values between the dashed lines. (c) Equilibrium
binding analysis of the compound used in (b). Data were fit to a simple 1:1 hyperbolic binding isotherm
Gibbs free energy upon binding of the compound [15]. The

reagents that enable DSF are a class of environmentally sensitive
dyes, including ANS, DAPOXYL derivatives, and SYPRO orange.
Most critically, these dyes are strongly quenched in aqueous envir-
onments and bind preferentially and nonspecifically to hydrophobic
surfaces. When bound to a hydrophobic surface, quenching is
relieved and a fluorescent signal can be measured. As a protein
denatures, hydrophobic core residues become exposed to solvent
and are available for dye binding, allowing enhanced fluorescence.
These characteristics allow protein unfolding to be monitored and
quantified in a straightforward experiment.
DSF can be employed at many stages in a fragment screening
campaign. Primary library screening by DSF has significant advan-
tages over other biophysical techniques. Unlike SPR, X-ray, or
NMR, DSF requires very little optimization of the initial conditions
∆Tm
{
Normalized Fluorescence
1.0
0.5
0.0
40 50 60 70 80 90 100
Temperature (°C)
Fig. 2 Protein stabilization measured by DSF. Thermal denaturation of a cation-

sensitive protein was measured by DSF in the presence of either 0.5 M K+ (blue)
or 0.5 M Na+ (orange) (Data courtesy of Samuel Pfaff). Melting temperatures
were calculated by fitting to the Boltzmann equation and are represented by
dashed lines. The rightward shift in Tm seen for the Na+ condition is analogous to
that of a potent small molecule hit
[16]; the initial concentrations of protein and dye are the only
parameters that typically require optimization. This allows diver-
gent targets to be screened under very similar conditions. Buffer
and salt conditions can have significant effects on protein melting
curves, and can be optimized as well. However, it is usually desir-
able to employ a simple buffer system that best reflects the physio-
logical environment of the target protein. Assay throughput is high,
since 384 wells can be read simultaneously in 45–90 min, depend-
ing on assay parameters. Thus, an average-sized fragment library
can be screened in a day with plate-handling automation. Protein
consumption during screening is also quite low (typically 5 mg), as
very small volumes (<5 μL/well) can be used during the assay. For
targets that are challenging to isolate, this can be critical. Finally,
the reagents and instrumentation are relatively inexpensive. RT-
PCR instruments have become commonplace at research centers
in the last decade, and they are far less complex to maintain than
SPR, NMR, or in-house X-ray equipment. The dye necessary for a
screen is very inexpensive, and 384 well PCR plates are comparable
in price to most assay plates. On the other hand, For fragments,
thermal shifts can be within 1 C, and binding stoichiometry is not
directly measured. Taking advantage of its speed, DSF screening is
often used as a primary screen to identify binders, followed by
higher resolution methods.
DSF is also valuable as a first step in structural biology experi-
ments. Proteins can show wide ranges of thermal stabilities as a
function of buffer pH, salt identity and concentration, reducing
agents, and other additives. Using DSF, one can rapidly screen a
matrix of conditions that can help to identify optimal storage

buffers and stabilizing classes of reagents that suggest possible
crystallization conditions. This type of prescreen is widely
employed in structural genomics and drug discovery environments,
and can decrease the time required to solve structures.
1.3 Nuclear Nuclear Magnetic Resonance (NMR) spectroscopy is a powerful

Magnetic Resonance and flexible biophysical technique that has assumed a key role
in examining the structural biology of ligand–protein and protein-
protein interactions. A sample containing a solution of protein is
inserted into a strong electromagnet (9–21 T) and sensitive
electronics record the resonances of NMR-active nuclei in response
to carefully designed perturbations. A number of different proper-
ties of the NMR-active nuclei are recordable in this manner, yield-
ing information about the chemical environment and relaxation of
the protein or ligand nuclei under observation. Because the binding
of a ligand to a protein will alter the environment of both partners,
NMR-based techniques can report on the nature and extent of
binding events. Methods have been developed to screen for either
the disruption or stabilization of protein-protein interactions
directly; however, that is outside the scope of this chapter [17].
Of the diverse palette of NMR experiments that aid small-molecule
discovery efforts [18], we focus on ligand- and protein-observed
experiments that detect the typically weak binding (μM–mM) of
initial fragment hits (Fig. 3). Ligand-observed methods make
use of the alterations in chemical shift and/or relaxation of the
compound upon binding to the larger protein. Protein-observed
methods generally focus on perturbations in chemical shift in the
protein resonances.
NMR-based screening offers several potential benefits. First,
the chemical shift is exquisitely sensitive to the chemical environ-
ment of each nucleus, making it possible to describe precisely which
moieties of a ligand are binding to the protein, and vice versa,
providing a detailed structural understanding of the interaction.
Binding affinities can be measured and competition between frag-
ments can also be determined. Second, both ligand and protein can
be detected; hence, NMR allows visualization of the whole system,
and is particularly sensitive to common artifacts such as precipita-
tion or aggregation of the ligands or the target. Third, the NMR
experiment is measured in solution, so interactions between pro-
teins and ligands can be measured in vitro under near physiological
conditions (pH, salt, and protein concentration). Furthermore,
because proteins are not restricted by a support matrix (as in SPR
or crystallography), they are free to adopt different conformations,
and potential fragment-binding sites are not occluded. Recent
developments in NMR, such as In-Cell-NMR and experiments
using cell extracts, allow monitoring of the interactions between
individual proteins and/or their ligands in complex milieu [19].
b 8.8 8.6 8.4 8.2 8.0 7.8 7.6

125 125
L72
126 R83 126
127 [Fragment], mM 127

R155 0
ω1 - 15N (ppm)
A31 0.2
128 1 128
2
5
C184
129 129
I182
130 130
V176
131 V181 131
132 132
8.8 8.6 8.4 8.2 8.0 7.8 7.6
1H
2- (ppm)
Fig. 3 NMR experiments for fragment screening. (a) Saturation transfer difference. 1H NMR spectrum of a
fragment (blue) is reduced when it binds to protein (red). Measuring the difference spectrum then gives a
negative signal with the same chemical shifts as the fragment (not shown). (b) 1H-15N HSQC. Each backbone
NH has a characteristic chemical shift. When fragments bind, they induce changes to these shifts in dose-
dependent manner (see legend) (Data courtesy of Mark Kelly, Michael Chimenti, and Rich Tjhen)
Several technological developments have given NMR the

throughput and sensitivity to serve as a tool for primary fragment
screening. High-field magnets and cryogenically cooled electronics
provide increased sensitivity and resolution; robotic sample prepa-
ration, automated data acquisition, and supporting software have
increased throughput. Practically speaking, the throughput of
NMR screens varies dramatically, depending on the instrumentation
and assay format. For protein-detected and 19F ligand-detected
formats, primary screens of several thousand fragments have been
described using mixtures of ~100 compounds [20–23]. Screening
in mixtures biases towards tighter binders due to competition and
often lower concentrations of each compound. Other common
ligand-detected formats, such as saturation transfer difference
(STD), are best performed with smaller pools of ~10 compounds,

reducing the overall throughput. Although a number of groups
currently perform NMR primary screens with libraries on the
order of 2,000–10,000 fragments, others prefer to apply NMR to
smaller libraries and as a secondary assay to obtain confirmatory or
higher resolution binding information.
Protein targets are suitable for NMR fragment screening if they
have good solubility, are mono-disperse and are stable in solution
for the duration of the screening run. Stability and consistency can
be determined by including known ligands as positive controls
throughout automated screening runs. For protein-observed
experiments, one can also compare 2D inverse-correlation spectra
(e.g., 2D [15N, 1H]-HSQC) for signs of protein denaturation and
degradation. The size of the protein is not critical for ligand-
detected experiments; for protein-observed NMR, the size dictates
what methods (pulse sequences) are most suitable. NMR typically
requires relatively large amounts of pure protein, so the expression
protocol must allow access to 10s of mg of protein. Protein con-
centrations average 10–20 μM protein for ligand-observed meth-
ods and ~50–200 μM protein for protein-observed techniques
(depending on target size). Below, we describe experiments per-
formed with the most common formats, using 5 mm outside
diameter standard or Shigemi NMR tubes (500 μl and 250 μl
sample volumes, respectively). NMR accessories exist for smaller
tubes (3 mm), capillaries, and LC systems that significantly reduce
protein requirements; however, with the capillary systems, precipi-
tation of fragments and/or protein can lead to blockage.
NMR has the potential to define the binding sites of fragments
on the protein target. In the case of protein-observed methods,
site-specific assignments of resonances are usually required. In some
cases, these have been performed previously and are available in
the literature. If not previously determined, assignments for key
residues, i.e., “probe-residues,” can be obtained by mapping
perturbations from known ligands, residue type-specific labeling
or mutagenesis (e.g., to identify specific methyl-groups). In
the case of ligand-observed experiments, competition with
natural ligands or substrates can be used to aid identification of
binding sites.
Many considerations that apply generally to the design of frag-
ment libraries are also important for NMR; however, a few special
criteria are important to mention here. First, high solubility in
aqueous buffer at acceptable d6-DMSO concentrations (1–5 %) is
essential, given NMR’s unique ability to detect very weak binders in
the 10 mM range. It is also critical to have adequate buffering
capacity in samples and to monitor the pH of samples following
addition of ligands. Compounds are commonly screened as mix-
tures of 5–10 fragments to reduce instrument time and protein
requirements. Pool size is selected based on total compound
solubility and concerns about possible interactions between

fragments. For ligand-observed methods, the use of mixtures is
very attractive because individual hits can be identified in NMR
spectra from resonance changes associated with binding. For such
experiments, fragments should be combined in mixtures such that
as few resonances as possible overlap in the 1D 1H NMR spectrum.
1D proton reference spectra NMR of fragments are acquired
prior to screening for quality control (QC) to eliminate compounds
that have low solubility and to enable fragments with minimal
spectral overlap to be combined in mixtures. The solution concen-
tration is estimated by comparing the fragment’s signal intensities
to either the residual 1H signal from the 99.9 % d6-DMSO or to the
DSS chemical shift reference. For ligand observed experiments, at
least one ligand signal must not overlap with the water resonance.
In principle, libraries of several thousand fragments can be
screened in a reasonable time using the methods outlined here.
However, when resources and/or protein are limiting, fragments
can be preselected based on predicted solubility (logP), diversity
(Tanimoto coefficient), and/or virtual screening (docking scores).
1.4 X-Ray The ideal situation for most fragment screening efforts is to obtain
Crystallography atomic-resolution structures of fragment hits bound to the target
(Fig. 4). High-resolution structures allow medicinal chemists to
improve compound potency by rationally altering chemical struc-
ture to better complement the target site. For small fragments,
whose binding modes are difficult to model precisely or predict a
Fig. 4 High-resolution X-ray structure of a PPI inhibitor fragment hit. RadA bound
to fragment hit indazole at 1.3 Å resolution (PDB: 4B2I) [3]. 2Fo-Fc electron
density map is shown for the compound contoured at 1.5σ. The electrostatic
surface and figure were generated using PyMOL [42]
priori, experimental structures by X-ray or NMR are almost

essential. While X-ray crystallography is commonly a follow-up
method attempted with a small set of preferred fragment hits,
many groups use it very successfully for initial library screening
[24, 25]. Placing structural efforts first is very attractive in that
every screening hit provides fodder for chemical optimization.
In contrast to the other techniques discussed, knowledge of a
hit compound’s potency is completely devalued in this context.
Compounds that produce interpretable electron density have a
known binding mode and are known to be soluble enough to at
least partially saturate the binding sites in a crystal. This is a wealth
of information, even in the absence of initial potencies. It is not
possible to rank order the binding affinity of hits by X-ray, but
affinity can be obtained using a functional assay or one of the
other binding methods described in this chapter.
The key prerequisites for screening by crystallography are large
quantities of pure target protein and reproducible conditions for
growing high-resolution crystals. Since most fragment discovery
campaigns move to structure determination at some point, the
availability of crystal structures is generally important for FBLD.
When used as a screening methodology, however, crystallization
methods must be solved up-front. Furthermore, diffraction-quality
crystals must be available in a high-throughput and reproducible
manner.
There are two primary means of achieving fragment occupancy
in a crystal: crystal soaking and co-crystallization. Crystal soaking is
far more common in fragment screening [26]. In this experiment,
pre-grown crystals are transferred from their growth wells into
wells containing the same buffer plus high concentrations of one
or several fragments. Each test well requires one large, high-quality
crystal, making screening of an average-sized library of singletons
daunting. Five to ten fragments are commonly included in each
well, both reducing the number of crystals needed and increasing
the likelihood of finding a hit in any well.
1.5 Summary The techniques overviewed above are the most commonly used
techniques in FBLD and each technique has a proven record of
success in identifying promising chemical matter against many
targets. Given below are detailed experimental protocols for devel-
oping these assays and employing them in an FBLD campaign
aimed at a protein or PPI target.
2 Materials
2.1 Surface Plasmon 1. 0.25 mg/mL neutravidin in 10 mM Na-Acetate, pH 4.5.

Resonance 2. 0.2 μm bottle-top filter.
3. 1:1 mixture of 0.1 M N-hydroxysuccinimide (NHS) and 0.4 M

1-ethyl-3-[3-dimethylaminopropyl]carbodiimide (EDC).
4. 1 M ethanolamine, pH 8.5.
5. Biotinylated target protein.
6. Immobilization buffer: 10 mM HEPES, pH 7.4, 150 mM
NaCl, 0.25 mM TCEP, 0.05 % (v/v) Tween 20.
7. 0.22 μm, 0.5 mL spin filter.
8. 0.2 mg/mL Amine-PEG-Biotin.
9. Running buffer: 10 mM Tris–HCl, pH 8.0, 150 mM NaCl,
0.25 mM TCEP, 0.05 % (v/v) Tween 20, 5 % (v/v) DMSO.
10. Positive control molecule, peptide, or protein.
11. Dimethyl sulfoxide (DMSO).
12. Assay plates.
13. Fragment library.
2.2 Differential 1. PCR plates suitable for the instrument being used.
Scanning Fluorimetry 2. Fluorescent dye.
3. Protein of interest.
4. Fragment library.
5. Assay Buffer: 20 mM HEPES, pH 7.4, 150 mM NaCl,
0.25 mM TCEP.
6. Optically clear sealing tape.
2.3 Nuclear 1. Ligand(s) of interest or fragment library.

Magnetic Resonance 2. Protein of interest.
3. d6-DMSO.
4. D2O.
5. NMR buffer: 100 % D2O PBS, pH 7.5, 1 mM NaN3, and
1 mM DTT, 0.5 % DMSO with 10 μM DDS.
6. 10 % (w/v) solution of NaN3 for dilution into assay buffer.
3 Methods
3.1 Surface Plasmon These methods assume that the user is familiar with the operation
Resonance of the SPR instrumentation in his/her laboratory.
3.1.1 Select SPR Chip The first step in performing any SPR experiment is selecting a strategy
Surface for immobilizing the target protein to the sensor surface. The two
most common immobilization strategies are (a) covalent coupling
through lysine side chains and N-termini using EDC/NHS-activated
carboxyl groups on the sensor surface (random-amine coupling) and
(b) capture of biotinylated proteins by avidin surfaces. Other

methods for immobilization include thiol coupling, NiNTA coupling
of hexa-histidine-tagged proteins, and capture by specific antibodies
(see Chapter 7 for more details).
For fragment screening, it is essential to couple a high density
of target protein on the sensor surface. The maximum SPR binding
signal for an analyte is given by:
ðImmobilization DensityÞ ðMW analyteÞ=ðMW protein targetÞ
The mass ratio above becomes very small for fragments against
average-sized protein targets (0.01 or less), necessitating a high
immobilization density to compensate. This is one negative aspect
of screening by SPR. High-density surfaces have often been asso-
ciated with artifactual protein behavior, due primarily to the high
concentration of protein at the surface. Additionally, some immo-
bilized protein might be inactive towards small-molecule binding
due to denaturation or orientation/immobilization to the SPR
surface. For these reasons, it is critical to optimize the protein-
immobilization procedure using a well-behaved control com-
pound, peptide, or protein binding partner as a control. This
positive control will also be used to monitor the stability of the
protein surface during screening.
Our preferred method of immobilization is biotin–avidin cap-
ture using site-specifically biotinylated proteins [27]. The AviTag
sequence is a short peptide that is a substrate for E. coli biotin ligase
BirA. If protein expression is performed in E. coli, the AviTagged
target protein can be coexpressed with BirA to yield protein that is
singly biotinylated at a specific lysine in the AviTag sequence.
Vectors for appending an AviTag to the N or C termini of targets
are commercially available. We prefer this method to random
biotinylation (coupling to lysine or cysteine residues) or covalent
coupling, as it is a consistent, predictable method that requires no
extra processing steps, and it does not compromise potentially
relevant protein amines.
A generic protocol for immobilization using biotinylated
protein follows. Streptavidin and neutravidin-coated SPR surfaces
(chips) can be purchased from many instrument vendors or can be
prepared from carboxymethyl chips, such as Biacore CM5 or CM7
(these two chips differ in density of carboxyl groups, see Chapter 7
for more details). We prefer neutravidin, a mutant form of strepta-
vidin, as it contains fewer potential small-molecule binding sites,
and is more amenable to covalent coupling due to its increased pI.
Steps 2–5 describe the protocol for immobilizing neutravidin; this
protocol is similar for covalent coupling of any target to a carbox-
ymethyl chip surface by EDC/NHS chemistry.
3.1.2 Prepare SPR Chip 1. Install SPR chip into instrument per vendor recommendation.
Surface
Covalently modify SPR chip with neutravidin:
2. Dilute neutravidin to 0.25 mg/mL into 10 mM Na-Acetate,
pH 4.5. Buffer should first be filtered through 0.2 μm bottle-
top filter (see Note 1).
3. Inject a 1:1 mixture of 0.1 M NHS and 0.4 M EDC for 5 min
to activate the chip surfaces.
4. Inject 0.25 mg/mL neutravidin for 4–7 min across all surfaces.
Using a Biacore CM5 chip this should yield 8,000–12,000 RU
of neutravidin density on all surfaces. Leave one surface as a
neutravidin-only reference.
5. Inject 1 M ethanolamine, pH 8.5, for 5 min across all surfaces
to block all remaining reactive sites.
Immobilize biotinylated protein
6. Dilute the biotinylated target protein to low concentration
(5–50 μg/mL) in immobilization buffer.
7. Filter the target protein solution using a 0.22 μm, 0.5 mL spin
filter.
8. Inject the target protein on an avidin-labeled flow cell for
1–5 min, or until the desired amount has been captured.
A surface of ~10,000 RUs for a protein of average size
(20–50 kDa) is a reasonable goal (see Note 2).
9. Inject 0.2 mg/mL Amine-PEG-Biotin (or a similar, soluble
biotin analog) for 4 min across all surfaces to block any
remaining biotin binding sites.
3.1.3 Establish the It is essential to determine the quality of the immobilized protein
Activity of the SPR Surfaces before proceeding with a fragment screening effort.
1. Select a running buffer. A typical buffer could contain 10 mM
Tris–HCl, pH 8.0, 150 mM NaCl, 0.25 mM TCEP, 0.05 %
(v/v) Tween 20, 5 % (v/v) DMSO (see Note 3).
2. Select a small molecule, peptide, or protein that binds to the
site of interest to be used as a positive control.
3. Serially dilute the control in DMSO and transfer to a running
buffer without DMSO, so that the % DMSO is identical in the
sample well and running buffer. For example, for a final DMSO
concentration of 5 % in 100 μL sample volume, dilute 5 μL of
compound into 95 μL of running buffer without DMSO.
Select a wide concentration range to span the expected binding
constant. Inject samples over the chip surface and measure the
SPR signal.
4. Calculate the KD using kinetic or equilibrium measurements,

and compare to the expected value. Determine the fraction of
active protein on the surface using the following equation:
surface activity ¼ ðRU analyteÞ=ðRU immobilized proteinÞ
ðMW proteinÞ=ðMW analyteÞ
1=expected binding stoichiometry
Surface activities of >0.8 can often be achieved, but lower

(ca 0.5) are also common.
3.1.4 Run Fragment 1. Set up assay plates using a standardized format based on the
Screen capabilities of your instrument. As with the control com-
pounds, the DMSO concentration and buffer should exactly
match the running buffer. Typical concentrations for fragments
are 100–500 μM in 2–5 % DMSO.
2. Run the screen using injection times between 10 and 30 s,
followed by dissociation monitoring for 30 s. Because frag-
ments should bind weakly and therefore reach equilibrium
quickly, these short injections provide a significant amount of
data with which to classify compounds.
3. Inject buffer blanks (containing DMSO) and positive control
molecules regularly throughout the screen (e.g., at the end of
a column or each plate), to evaluate assay stability. Blank
injections should also be used during data analysis to “dou-
ble-reference,” which is especially critical in fragment screening
where the signal is expected to be low [28]. The controls
should also be used to calculate z0 or other statistical factors
to assess the quality of the screen [29].
3.1.5 Identify Screening SPR data allows for hit selection based on both the % fractional
Hits occupancy the compound achieves, and also the kinetic profile it
produces.
1. Reference SPR data (“sensorgrams”) to the unmodified flow cell
and double-reference to blank injections using the instrument’s
software or third-party software such as Scrubber (Biologic
Software, http://www.biologic.com.au/). Additional data
smoothing and scaling can reduce complexity arising from
variation in positive control binding and experiments run on
separate surfaces [28].
2. Prepare a scatterplot (Fig. 1a) of the control and test
compounds. Select a time-range near the end of the sample
injection, where the positive control has achieved equilibrium.
Use this time range to determine the RU for each control and
test compound, and plot on the y-axis. Scatterplots can also be
scaled to binding stoichiometry:
stoichiometry ¼ ðRU analyteÞ=ðRU immobilized proteinÞ

ðMW proteinÞ=ðMW analyteÞ
ðsurface activityÞ
3. Select compounds for follow-up based on approximate binding

stoichiometry and kinetics. The standard threshold for hit
selection is three times the standard deviation of the non-
interacting compounds (Fig. 1a) (see Note 4).
4. Cherry-pick or prepare fresh DMSO stocks of selected frag-
ments and retest at multiple doses (Fig. 1c). As with positive
control compounds, hits should be serially diluted in DMSO
and then transferred to running buffer that does not contain
DMSO. If possible, dose responses should start severalfold
above the concentration for primary screening; eight 3-fold
dilutions or ten 2-fold dilutions are common. The Kd is esti-
mated by converting RU to stoichiometry at each concentra-
tion point and fitting the concentration vs. stoichiometry
binding isotherm (or log[analyte] vs. stoichiometry). This
experiment yields dissociation constants and stoichiometry for
well-behaved compounds and serves to identify bad-acting
compounds that looked passable at a single concentration.
3.1.6 Initiate Secondary At this stage, repurchase or synthesize the most potent compounds
Assays with low stoichiometry and move forward to secondary screening
(such as functional activity), chemical optimization, and structure
determination by NMR or X-ray crystallography (see below).
3.2 Differential 1. Select reagents. The choice of plates will depend on the model
Scanning Fluorimetry of RT-PCR instrument to be used; in general, use 384-well
PCR plates with optically clear sealing tape. There are also
3.2.1 Develop DSF Assay
many potential dyes. No published study has systematically
compared the appropriate dyes for overall performance, and
target-specific effects of particular dyes may be possible.
SYPRO Orange is cost-effective and widely used. It is provided
as a 5,000 stock in DMSO, and our experience has shown it
to work well at 1 final concentration across a range of targets
and buffers.
2. Prepare a matrix containing several concentrations of protein
and several concentrations of dye, (e.g., SYPRO Orange); each
combination should be tested in triplicate. Typical starting
parameters are 2 μM protein with 5,000 diluted SYPRO
Orange [30]. Assay volumes can be as low as 5 μL with appro-
priate 384-well plates.
3. Measure the melting temperature of each dye–protein combi-
nation using a qRT-PCR instrument. Record fluorescence data
at the appropriate dye wavelengths, using a temperature range
of 25–95 C in 1 steps with 30 s per step. Load the plate into

the thermocycler and initiate the experiment. Analyze the data
by plotting the fluorescence intensity vs. temperature; the
melting temperature (Tm) is the temperature at the inflection
point (see Subheading 3.2.2 below for details).
4. Select dye/protein combinations. Thinking ahead to the
screen, it is useful to minimize the amount of protein and dye
required per well. Thus, select the minimal reagent concentra-
tions and well volumes that give reproducible Tm values.
5. Establish stability of the protein in DMSO/buffer solutions
(see Note 5).
3.2.2 Run Fragment The following procedure is a general method for fragment screen-
Screen ing by DSF in 384-well plates with a final well volume of 20 μL and
a final compound concentration of 1 mM.
1. Prepare reagents for screening as follows: dilute protein from a
concentrated stock to 1.33 the predetermined optimal assay
concentration (usually between 0.5 and 5 μM) into assay buffer
without DMSO. Dilute dye to 5 the predetermined optimal
concentration into assay buffer without DMSO. Prepare or
store compounds at 20 final concentration in 100 % DMSO
(20 mM for a final assay volume of 1 mM).
2. Add 15 μL diluted protein to each well of a 384-well PCR
plate.
3. Add 1 μL DMSO to one column of the plate. These wells serve
as the negative controls.
4. If a positive control compound is available, add 1 μL of 20
positive control in DMSO to one column of the plate.
5. Add 1 μL of test compound to each non-control well using a
multichannel pipette or an automated liquid handler.
6. Add 4 μL of diluted dye to each well using a multichannel
pipette or an automated liquid handler. Seal the plate with
optically clear sealing tape, cover with foil, and mix by gentle
shaking at room temperature for 10–20 min.
7. Remove foil and centrifuge plate for 2 min at 1,000–2,000 g
to remove bubbles. The plate is now ready to be assayed.
8. Set up the RT-PCR instrument to record fluorescence data at
the appropriate dye wavelengths, for a temperature range of
25–95 C in 1 steps with 30 s per step. Load the plate into the
thermocycler and run the experiment. With these parameters,
each plate should take between 35 and 45 min.
9. Determine the Tm for fragments. Plot the fluorescence vs.
temperature data and fit to the Boltzmann equation:
F ¼ min þ ðmax minÞ=ð1 þ expððT m x Þ=slopeÞÞ
Where F is the measured fluorescence, min is the minimum

observed fluorescence, max is the maximum observed fluores-
cence, x is the temperature, Tm is the melting temperature,
and slope is the slope of the curve at the Tm. Tm and slope
are to be fit, while the other parameters are given by the data
(see Note 6).
3.2.3 Identify Screening To select screening hits, prepare a scatter plot of Tm value vs.
Hits compound and control wells. The control wells should produce a
precise Tm with low standard deviation (<1 ). Fragments that
produce a Tm increase greater than two times the standard devia-
tion of control wells are typically considered hits. This threshold
can be modified if the number of hits above 2 standard deviations is
thought to be too high or too low. Setting too low of a hit
threshold is, however, likely to generate many false positive com-
pounds that cannot be confirmed. In this case it might be advanta-
geous to rescreen at a higher compound concentration.
3.2.4 Initiate Secondary If desired, affinities can be directly determined by DSF using serial
Assays dilutions of fragments. Thermal shift dose–response data is not
fitted to a standard hyperbolic curve, but rather to a more compli-
cated expression that represents the free energy change of folding as
well as that of compound binding. Matulis et al. have derived a
model to fit DSF titration curves [31]. Accurate binding affinities
may be more readily obtained from other biophysical methods,
such as SPR, NMR, or ITC, as there is no reliance upon an inde-
pendently determined or estimated parameter, as is the case with
DSF. Since every assay format has advantages and limitations, it is
generally advisable to follow up hits from any fragment screen using
an orthogonal method.
3.3 Nuclear The detailed operation of NMR instruments is beyond the scope of
Magnetic Resonance this chapter. Omitted are basic preparative setup steps for acquiring
NMR spectra, such as shimming and tuning, as it is assumed the user
is familiar with these or will have support from an NMR spectrosco-
pist. For further reading, see Till Maurer’s excellent introduction to
NMR methods in fragment-based ligand screening and Bertini for
more detailed protocols and troubleshooting [32, 33].
3.3.1 Develop a The Saturation Transfer Difference (STD) experiment relies on the
Saturation Transfer nuclear Overhauser effect (nOe) between nuclei in the protein and
Difference (STD) Assay in a small-molecule ligand that binds transiently. The compounds
that interact with the protein will show reduced intensity in their
NMR signals relative to non-binders, whose signals will be unaf-
fected by the presence of protein. STD is performed as a subtractive
measurement, where the 1-D ligand spectra are collected in the
presence and absence of protein irradiation (“on-resonance” and
“off-resonance,” respectively). These spectra are subtracted, so that

bound ligands yield an NMR signal, while non-binders are not
observed.
Ligand–protein ratios, NMR irradiation time, and temperature
are best optimized empirically with a known ligand; however, stan-
dard conditions may be applied initially. If no control is available,
screening may proceed initially until a “hit” is found.
1. To avoid spectral overlap with fragments, buffers should con-
tain deuterated organic salts and other additives (see Note 7).
2. For STD, it is critical to maintain a large excess of free ligand,
ideally at a ratio of 10–50:1 ligand to protein. Typically, we
screen with 1–5 % d6-DMSO, 500–1,000 μM ligand, and
10–50 μM protein.
3. Generally, low temperatures (10 C) are used for smaller
proteins (6–20 kDa), since the strength of the nOe is depen-
dent on the protein’s correlation time, which increases with
lower temperatures. Additionally, modifications of the standard
pulse sequences are available which include filters to suppress
protein resonances that may overlap with ligand signals.
4. Select a portion of the protein spectrum away from ligand
resonances (typically near the methyl region, 0.5 to
1.5 ppm). Selectively irradiate protein using a shaped 270
pulse train (e.g., an 80 ms Gaussian pulse on a 500 MHz
instrument) for several seconds. The NMR irradiation time is
an optimizable parameter. It is good practice to record an on-
resonance STD spectrum of the fragment solution (in the
absence of protein) to serve as a reference.
5. In a second reference experiment, move the saturation pulse
“off-resonance,” so as not to excite any protein nuclei, usually
by 20–50 ppm (we irradiate at 20 ppm). Care needs to be taken
that the selective “saturation” pulse length, shape, and power
are calibrated such that ligand resonances outside of the methyl
region are not affected by the saturation pulse (to check this,
the reference spectrum from step 4 can be compared to a
simple 1D 1H spectrum of the ligand).
6. Collect the STD experiment as a “pseudo-2D” where both the
on- and off-resonance free induction decays (FIDs) are
collected in an interleaved manner and stored separately on
disk. We recommend this approach because it allows both
experiments to be handled separately. However, many com-
monly used pulses sequences subtract the on-resonance FID
from the off-resonance FID (reference) during acquisition.
For both experiment types, the data are apodized (1–3 Hz
line-broadening), zero-filled once, Fourier transformed, and
phased.
In the case of overlaid pseudo 2D data, peaks that show an STD

effect will be reduced in intensity in the on-resonance experiment.
By scaling the off-resonance spectrum to the on-resonance
spectrum, the “percent STD” reduction for each peak can be
measured. The two STD spectra can then be subtracted to generate
a difference spectrum if desired.
3.3.2 Run an STD The following is a general outline for a protocol when performing
Fragment Screen STD experiments in standard 5 mm diameter NMR tubes.
Typically ~100 samples (5–10 fragments each) or more can be
screened in a 24 h period with automated hardware.
1. Prepare the sample(s) without protein in NMR buffer (final
volume ~450 μL). Samples will include fragment(s), ~10 μM
DSS (for referencing), a deuterated buffer and D2O (99.96 %)
as solvent (see Note 8).
2. Acquire a simple 1D spectrum at the temperature to be used for
screening for each fragment to be used as a reference spectrum
during data analysis.
3. Add protein to each sample. Place samples in a rack. Make a
“master mix” containing protein (plus fresh DTT if indicated),
and NaN3 (to control bacterial growth) along with 10–20 μL
of NMR buffer per sample. Dispense the correct volume of
master mix (usually ~50 μL) into each tube with a pipette
(by placing NMR tubes into a pipette tip box, an 8-channel
pipette can be used for this). Carefully swing the rack up and
down to force the solution to the bottom of each tube (cover
the top of the rack with a piece of cardboard and grip it firmly).
Allow the tubes to sit for 1–2 h or overnight at 4 C to mix.
4. After tuning, shimming, etc., load parameters for the STD
NMR experiment. Be sure to optimize the 1H transmitter
frequency to obtain good suppression of any residual water.
The number of scans is typically set to 16 or 32, saturation time
to 1.3 s, recycle delay to 3.7 s with a spectral width of 16 ppm
and collecting 4,096 complex data points. Experiments can be
acquired in 10–15 min per sample.
5. The data are then apodized (1–3 Hz line-broadening), zero-
filled once, Fourier transformed, and phased. 1D on- and
off-resonance STD are extracted from the pseudo 2D spectrum
and overlaid for comparison. The “off-resonance” experiment
is scaled down to the “on-resonance” experiment and the “%
STD” reduction for each peak measured.
6. When working with mixtures of ligands, repeating the experi-
ment with the individual fragments from a pool that showed a
“hit” can confirm the identity of the binding fragment(s).
3.3.3 Identifying In the STD and ligand-detected experiments, hit identification is

Fragment Hits determined by examining the on-resonance and off-resonance
(reference) experiments, either by overlaying the two for compari-
son or by computing the difference spectrum (Fig. 3a). Ligands
that have no interaction with the protein target should not exhibit
changes between the on- and off-resonance experiments, and give
no signals in the difference spectrum. Ligands with strong interac-
tions will give large changes in intensity, and unambiguous negative
signals in the difference spectra. Ligands that bind weakly (like most
fragments) will give small, but interpretable changes in the
spectrum. Unless a ligand with known affinity is available as a
reference, a cutoff value is usually chosen to determine which
fragments are candidates for “hit” follow-up. Typically fragments
with STD difference values of >5 % are classified as a “hit”.
Secondary assays for STD include measuring dose–response
behavior of the signal to rank-order fragments. Binding selectivity
can be tested by running the STD experiment against a control
protein or homolog of the target. STD gives limited information
about the binding site or stoichiometry, and thus this method is
complemented by the orthogonal techniques (ITC, SPR, X-ray,
protein-detected NMR).
3.3.4 WaterLOGSY Similar to STD NMR, WaterLOGSY NMR is a ligand-detected

method. Details are described in [34]. WaterLOGSY relies on
transfer of magnetization from bulk water to a ligand through
waters bound in or adjacent to the small-molecule binding site,
and from magnetization from bulk water that is transferred
through protons in the target by spin-diffusion. For some targets,
detection of small-molecule binding can be stronger by Water-
LOGSY, and new PO-WaterLOGSY experiments offer higher
sensitivity [35]. Implementation of the PO-WaterLOGSY can be
more involved, however. Unlike STD-NMR, WaterLOGSY cannot
be performed in 100 % D2O; 10 % H2O is usually sufficient to
obtain a strong signal, and also allows an STD experiment to be run
on the same sample.
3.3.5 HSQC and HMQC One of the first fragment screening methods, called “SAR by
NMR NMR,” was based on a protein-observed HSQC experiment
[36]. A key advantage of protein-observed experiments is that the
structure and state of the target protein are monitored in each
experiment.
HSQC and related chemical-shift perturbation methods rely on
the availability of labeled protein, either 15N-labeling or 13C-labeling
for larger systems. The collection of a 13C/15N 2D inverse-
correlation or HSQC/HMQC type experiment yields a 2D spec-
trum, where cross-peaks corresponding to the frequencies of two
directly bound nuclei (e.g., Hn–15N or 1H–13C in a methyl group)
are dispersed on a 2D plane based on their chemical shifts (Fig. 3b).

Protein-detected techniques are limited by the need to obtain a
well-resolved 2D correlation spectrum, placing an upper limit on
the domain size of around 50 kDa when using 15N-labeling. Large
proteins or complexes can be screened by resorting to selective-
labeling strategies where the methyl groups of Isoleucine, Leucine,
Valine, Alanine, and Methionine are labeled with 13C combined with
deuteration of other aliphatic groups.
The details of running HSQC experiments are dependent on
the NMR instrument and the target. Some general considerations
for fragment screening by HSQC/HMQC follow:
1. The spectrum is collected with and without ligand or a pool of
ligands. If the ligand binds to the target, frequency changes
(shifts in peak positions) or broadening of peaks will be
observed for groups close to the binding site.
2. Due to the sensitivity of chemical shifts of groups (in particular
amide protons) and protein structure to changes in solvent pH,
it is essential to monitor the pH of samples following the
addition of fragments.
3. Deuterated DMSO (1–5 %) is usually used as a co-solvent as in
ligand-observed experiments, but it is important to record a
spectrum of the target in the presence of DMSO as a reference
as it may induce shifts and also to test for changes in activity.
4. Where mixtures (5–10) of fragments are screened, it is neces-
sary to deconvolute pools that contain a “hit” by screening
individual fragments. The size of the pools of fragments is
usually kept fairly small (5–10) to allow weak binding frag-
ments to be screened at concentrations of 1–2 mM while
reducing the chance of solubility limitations.
5. If chemical shift assignments are available or can be obtained,
the location of the binding site can be identified at atomic
resolution. If binding of a ligand causes allosteric changes in
the target, peaks corresponding to residues remote from the
binding site may also be affected. It is reasonable to predict
however, that the strongest shifts will cluster to a region that
forms the binding pocket.
6. The location of the binding pocket can be confirmed by iden-
tifying short-range interactions (<6 Å) between the ligand and
target, such as by recording NOESY experiments.
Once a binding fragment has been identified, a second frag-
ment that binds in close proximity can be sought by repeating the
screen in the presence of a high concentration of the first ligand.
This approach was originally described by Shuker et al. [36]. When
the HSQC spectrum has been assigned for the target protein,
HSQC-based screens offer a unique combination of throughput,
simplicity of design, and high-resolution binding data.
3.4 X-Ray 1. Select the appropriate protein construct (see Note 9).
Crystallography 2. Select a storage buffer that confers good protein stability.
3.4.1 Validate Crystal Prescreening for protein thermal stability in a range of possible
Forms for Fragment storage buffers can be easily accomplished using differential
Screening scanning fluorimetry (DSF; see above).
3. Screen crystallography conditions. Obtaining protein crystals
that diffract to better than 2.5 Å resolution is almost always the
result of brute force screening. In a typical case, 3–10 protein
expression constructs are screened against >1,000 initial
conditions from sparse-matrix, commercial crystal screens
(see Note 10).
4. Assess the stability of crystals in cryogenic conditions [26].
Nearly all protein crystallography is carried out under cryo-
genic conditions, requiring the optimization of freezing con-
ditions that prevent crystal damage. Typically, this involves
temporary soaking of each crystal in a condition containing a
high concentration (20–30 % (v/v)) of glycerol or high molec-
ular weight PEG. This should be performed with several test
solutions in a trial and error fashion.
5. Assess the stability of crystals in DMSO (~10–15 % (v/v)) to
simulate the conditions that will be used for soaking in test
compounds. Several crystals should be transferred to drops
containing varying concentrations of DMSO and observed
intermittently over the next 24 h. Some crystals cannot tolerate
DMSO and will become visibly damaged. X-ray data sets
should then be collected from crystals that remain intact in
DMSO to test for more subtle damage caused by the addition
of organic solvent.
6. Plate compounds for an X-ray based fragment screen.
To increase throughput and reduce the number of crystals
needed, compounds are typically pooled in sets of 5–10.
Premixed and plated compounds can be stored in DMSO or
as dried powder mixes. In either case, after the compounds are
solubilized in DMSO, crystallization buffer is added to each
well to a final volume of 2–5 μL. Crystals are then transferred
into each test well and incubated for a variable length of time.
The time necessary to soak compounds into a crystal is highly
protein dependent.
3.4.2 Collect and Process 1. Prepare crystals for data collection. Data collection and crystal
X-Ray Diffraction Data handling procedures are dictated by the X-ray generation site,
be it a home source or an external light source. For screening,
some measure of automation in crystal handling is critical, as it
can greatly increase efficiency and decrease the risk of manual
error and sample loss. Many beamlines are now outfitted with
frozen crystal handling robots that mostly automate the
process of crystal mounting/dis-mounting and X-ray beam

centering. The process of freezing crystals must still be carried
out manually, but no further human contact with the crystals is
required.
2. Determine a data-collection strategy and collect data. Data
collection throughput is largely driven by the number of data
frames that must be collected per crystal. This number primar-
ily depends on the symmetry parameters of the crystal and can
only be optimized to a certain degree (see Note 11).
3. Process data and generate a model. These steps have become
much more automated since the advent of fragment screening.
We recommend XDS [37], ELVES [38], or HKL2000 for
indexing, scaling, and merging of diffraction data. ELVES is
the most automated package for robust data processing, and is
well suited for processing dozens of data sets quickly. For
phasing and model refinement, we recommend Phenix [39],
which is amenable to scripting through the command line
interface, and has an advanced graphical interface as well. For
model visualization and manual building, Coot [40] is the
current standard.
3.4.3 Identify Hits from 1. Search electron density maps for bound fragments (Fig. 4).
X-Ray Screens Fragment occupancy can be evaluated manually or by compu-
tational methods. Both Coot and Phenix contain modules for
locating large unfilled regions of electron density.
2. When fragment density is observed, attempt to identify its
structure from among the compounds in the screening pool
of 5–10 fragments. This analysis is easiest when the fragment
pools are designed to contain very dissimilar compounds, the
fractional occupancy of the site is high, and the resolution of
the crystal is good (~2 Å). When the identity is not obvious,
one can build separate models for each possible compound and
refine them; the refinement process should eliminate some
compounds from consideration. More automated approaches
to considering fragment electron density have been reported by
industrial labs, but are not publicly available [25]. In many
cases, it is best to recrystallize the fragment/protein complex
from a pure solution of the compound.
3.4.4 Characterize Hits Developing secondary assays for an X-ray fragment screen depends
from X-Ray Screens on many factors, including target type, chemistry resources, and
availability of secondary biophysical assays to measure binding
affinities. For a PPI target, it is ideal to assess whether or not the
fragments inhibit the interaction, using some of the techniques
described elsewhere in this book. Biophysical competition assays
by SPR, analytical ultracentrifugation, analytical gel filtration, or
static light scattering can also give a measure of functional potency.
4 Conclusion
FBLD holds promise for the discovery of small-molecule inhibitors

of PPI. Particular advantages include the ability to search chemical
space efficiently and the use of binding-based assays to identify
ligands. Each of the primary screening methods described in this
chapter is widely utilized in FBLD, but each has its limitations.
Judicious combination of orthogonal methods will ensure that
fragments are selected based on lead-like criteria, including high
ligand efficiency (binding affinity) and binding-site specificity.
5 Notes
1. For direct amine coupling of a test protein, prepare an immo-

bilization buffer that is logical for your protein target, within
the following guidelines: (a) select buffer pH below the pI for
the protein. This allows the positively charged protein to “pre-
concentrate” at the carboxyl surface, thus increasing coupling
efficiency, (b) avoid amine-based buffers, including Tris, (c)
avoid dithiothreitol (DTT) because it reduces the reactivity of
activated NHS-esters, (d) include a low amount of non-ionic
detergent (Tween 20 at 0.05 % or less). All buffers should be
kept at room temperature and filtered through 0.2 μm bottle-
top filters prior to use.
2. Depending on the architecture of the chip and instrument,
different configurations are possible. Using a Biacore CM5
chip and T200 instrument, there can be three active flow cells
and one reference flow cell (where no test protein is added).
Having a reference flow cell is mandatory, but the configuration
of the three active flow cells is completely variable. It is com-
mon to add three different immobilization densities of the
same protein. One can also screen three targets in parallel, or
immobilize variants (such as point mutants) of the primary
target [41].
3. This step is highly target-dependent, but the following guide-
lines should be considered: (a) the buffer should maximize the
stability and activity of the immobilized protein, (b) addition of
detergents, such as 0.01 % Tween or 0.1 % Triton is highly
recommended to reduce nonspecific binding of test com-
pounds to the protein and microfluidics, (c) carrier protein,
such as Prionex or beta-gamma globulin, can further reduce
nonspecific effects, (d) using the same buffer conditions in SPR
as will be used for secondary assays can increase reproducibility
across assays [28]. Finally, it is very important to include
DMSO in the running buffer such that the %DMSO is identical
to that in the sample wells. Fragment screening is typically

performed in a background of 2–5 % DMSO to aid in
compound solubility.
4. Compounds that bind in excess of their maximum theoretical
binding level are likely to be nonspecific; at the primary screen-
ing stage, we generally accept compounds that bind with appar-
ent stoichiometries <3. All sensorgrams that pass these filters
should be inspected by eye to take advantage of the high data
content of an SPR-based screen. In almost all cases, fragments
should bind and dissociate with rapid kinetics (Fig. 1b), which
produces a sensorgram with a square pulse-like shape. Discern-
ible kinetics have been reported for some low affinity fragments,
but this is relatively uncommon.
5. As fragment concentrations in the screen will be high, a final
concentration of 5 % (v/v) DMSO is typical. Therefore, moni-
tor protein melting at several DMSO concentrations (e.g., 0, 1,
3, 5 %). If a significant reduction in Tm is seen at any DMSO
concentration, it is advisable to stay below that concentration in
the screen.
6. The Tm can also be estimated from the minimum of the nega-
tive first derivative of the raw data. This process has limited
resolution, because the data is only collected in 1 steps, and the
Tm is estimated by visual inspection. For this reason, we prefer
to more robustly fit the Tm using the raw data. Several simple
and free tools exist for fitting DSF data using this method. We
use FIT, developed at the University of Washington (http://
skuld.bmsc.washington.edu/FIT/). FIT can be used as a web
server or the script can be downloaded and run on a local
machine. Each well in a data set is individually fit to both single
and double transition models, with figures and statistics auto-
matically generated in useful formats.
7. Deuterated Good buffers are available for a range of pH values;
deuterated acetate or phosphate buffer may also be used.
Reductants such as DTT and TCEP are also available in deut-
erated form. Sufficient salt should be included to maintain
protein activity and stability, however higher concentrations
(>50 mM) reduce sensitivity, particularly for cryogenic probes.
For a more detailed discussion of sample preparation, see Bertini
Ch. 15 [33].
8. Tubes can be the inexpensive type, such as Norell XR series
bulk NMR tubes. If using automation, it is essential to allow
sufficient time (>3 min) for a sample to thermally equilibrate
after it is inserted into the magnet. As noted, smaller targets
(6–25 kDa) will yield stronger STD signals at lower tempera-
tures, but may require longer equilibration times.
9. For targets that have previously been crystallized, it is often

useful to create an expression construct that begins at the first
modeled residue at the N-terminus and ends at the last mod-
eled residue at the C-terminus of the published structure. By
removing disordered termini, proteins can crystallize more
readily and with higher resolution.
10. Miniature-scale crystallization robots such as the Mosquito and
Honeybee systems have made the process of screening for
crystal conditions much faster and less resource intensive.
They allow for drop sizes lower than 100 nL, whereas tradi-
tional screening by manual pipetting is performed with at least
500 or 1,000 nL of concentrated protein stock per test
condition.
11. X-ray test frames are analyzed computationally to give a
preliminary guess at the space group and unit cell parameters.
This information is then used to devise an efficient strategy
for collecting a fully complete data set. As most crystals of
the same target in the same condition will be isomorphous,
the same collection strategy can be used throughout the
screen. The type of X-ray detector also influences collection
throughput [25].
References
1. Rees DC, Congreve M, Murray CW et al protein-protein interface. Drug Discov Today
(2004) Fragment-based lead discovery. Nat 14:155–161
Rev Drug Discov 3:660–672 8. Lau WF, Withka JM, Hepworth D et al (2011)
2. Hajduk PJ, Greer J (2007) A decade of Design of a multi-purpose fragment screening
fragment-based drug design: Strategic library using molecular complexity and orthog-
advances and lessons learned. Nat Rev Drug onal diversity metrics. J Comput Aided Mol
Discov 6:211–219 Des 25:621–636
3. Scott DE, Ehebauer MT, Pukala T et al (2013) 9. Na J, Hu Q (2011) Design of screening collec-
Using a fragment-based approach to target tions for successful fragment-based lead discov-
protein-protein interactions. Chembiochem ery. Methods Mol Biol 685:219–240
14:332–342 10. Chen IJ, Hubbard RE (2009) Lessons for frag-
4. Braisted AC, Oslob JD, Delano WL et al ment library design: analysis of output from
(2003) Discovery of a potent small molecule multiple screening campaigns. J Comput
IL-2 inhibitor through fragment assembly. J Aided Mol Des 23:603–620
Am Chem Soc 125:3714–3715 11. Erlanson DA, Wells JA, Braisted AC (2004)
5. Arkin MR, Randal M, Delano WL et al (2003) Tethering: fragment-based drug discovery.
Binding of small molecules to an adaptive Annu Rev Biophys Biomol Struct 33:199–223
protein-protein interface. Proc Natl Acad Sci 12. Wilson CG, Arkin MR (2013) Probing struc-
U S A 100:1603–1608 tural adaptivity at PPI interfaces with small
6. Petros AM, Huth JR, Oost T et al (2010) molecules Drug Discovery Today: Technolo-
Discovery of a potent and selective bcl-2- gies 10 (4):e501–e508
inhibitor using SAR by NMR. Bioorg Med 13. Giannetti AM, Koch BD, Browner MF (2008)
Chem Lett 20:6587–6591 Surface plasmon resonance based assay for the
7. Fuller JC, Burgoyne NJ, Jackson RM (2009) detection and characterization of promiscuous
Predicting druggable binding sites at the inhibitors. J Med Chem 51:574–580
14. Babaoglu K, Simeonov A, Irwin JJ et al (2008) through of fragment lead identification using
Comprehensive mechanistic analysis of hits surface plasmon resonance. Methods Enzymol
from high-throughput and docking screens 493:169–218
against beta-lactamase. J Med Chem 29. Zhang JH, Chung TD, Oldenburg KR (1999)
51:2502–2511 A simple statistical parameter for use in evalua-
15. Cimmperman P, Baranauskiene L, Jachimovi- tion and validation of high throughput screen-
ciute S et al (2008) A quantitative model of ing assays. J Biomol Screen 4:67–73
thermal stabilization and destabilization of 30. Niesen FH, Berglund H, Vedadi M (2007) The
proteins by ligands. Biophys J 95:3222–3231 use of differential scanning fluorimetry to
16. Kranz JK, Schalk-Hihi C (2011) Protein ther- detect ligand interactions that promote protein
mal shifts to identify low molecular weight stability. Nat Protoc 2:2212–2221
fragments. Methods Enzymol 493:277–298 31. Matulis D, Kranz JK, Salemme FR et al (2005)
17. Rizo J, Rosen MK, Gardner KH (2012) Thermodynamic stability of carbonic anhy-
Enlightening molecular mechanisms through drase: measurements of binding affinity and
study of protein interactions. J Mol Cell Biol stoichiometry using thermofluor. Biochemistry
4:270–283 44:5258–5266
18. Pellecchia M, Bertini I, Cowburn D et al 32. Maurer T (2011) Advancing fragment binders
(2008) Perspectives on NMR in drug discov- to lead-like compounds using ligand and
ery: a technique comes of age. Nat Rev Drug protein-based NMR spectroscopy. Methods
Discov 7:738–745 Enzymol 493:469–485
19. Ito Y, Selenko P (2010) Cellular structural 33. Bertini I, Molinari H, Niccolai N (1991) NMR
biology. Curr Opin Struct Biol 20:640–648 and biomolecular structure, vol xvii. VCH,
20. Dalvit C, Fagerness PE, Hadden DT et al Weinheim, 209 p
(2003) Fluorine-NMR experiments for high- 34. Dalvit C, Pevarello P, Tato M et al (2000)
throughput screening: theoretical aspects, Identification of compounds with binding
practical considerations, and range of applica- affinity to proteins via magnetization transfer
bility. J Am Chem Soc 125:7696–7703 from bulk water. J Biomol NMR 18:65–68
21. Dalvit C, Flocco M, Veronesi M et al (2002) 35. Gossert AD, Henry C, Blommers MJ et al
Fluorine-NMR competition binding experi- (2009) Time efficient detection of protein-
ments for high-throughput screening of large ligand interactions with the polarization opti-
compound mixtures. Comb Chem High mized PO-WaterLOGSY NMR experiment. J
Throughput Screen 5:605–611 Biomol NMR 43:211–217
22. Hajduk PJ, Meadows RP, Fesik SW (1999) 36. Shuker SB, Hajduk PJ, Meadows RP et al
NMR-based screening in drug discovery. Q (1996) Discovering high-affinity ligands for
Rev Biophys 32:211–240 proteins: SAR by NMR. Science
23. Hajduk PJ, Gerfin T, Boehlen JM et al (1999) 274:1531–1534
High-throughput nuclear magnetic resonance- 37. Kabsch W (2010) Xds. Acta Crystallogr D Biol
based screening. J Med Chem 42:2315–2317 Crystallogr 66:125–132
24. Murray CW, Blundell TL (2010) Structural 38. Holton J, Alber T (2004) Automated protein
biology in fragment-based drug design. Curr crystal structure determination using ELVES.
Opin Struct Biol 20:497–507 Proc Natl Acad Sci U S A 101:1537–1542
25. Spurlino JC (2011) Fragment screening purely 39. Adams PD, Afonine PV, Bunkoczi G et al (2010)
with protein crystallography. Methods Enzy- Phenix: a comprehensive python-based system
mol 493:321–356 for macromolecular structure solution. Acta
26. Bottcher J, Jestel A, Kiefersauer R et al (2011) Crystallogr D Biol Crystallogr 66:213–221
Key factors for successful generation of 40. Emsley P, Cowtan K (2004) Coot: model-
protein-fragment structures requirement on building tools for molecular graphics. Acta
protein, crystals, and technology. Methods Crystallogr D Biol Crystallogr 60:2126–2132
Enzymol 493:61–89 41. Hamalainen MD, Zhukov A, Ivarsson M et al
27. Prakash O, Eisenberg MA (1979) Biotinyl 5’- (2008) Label-free primary screening and affin-
adenylate: corepressor role in the regulation of ity ranking of fragment libraries using parallel
the biotin genes of Escherichia coli k-12. Proc analysis of protein panels. J Biomol Screen
Natl Acad Sci U S A 76:5592–5595 13:202–209
28. Giannetti AM (2011) From experimental 42. Schrodinger, Llc (2010) The PyMOL molecu-
design to validated hits a comprehensive walk- lar graphics system, version 1.3r1
INDEX
A Bioinformatics .....................................133, 286, 295, 296

Biolayer interferometry
Affinity constant ................................ 126, 165, 167, 170, binding kinetics ........... 165, 167, 168, 174, 176, 180
177, 181, 378 biosensors ...................................................... 140, 141,
Affinity tags.......... 114. See also Chromatography, Affinity
165–175, 178, 179, 181, 459, 555–559,
for co-immunoprecipitation ........................ 292, 381, 562, 563, 588
383, 386, 388, 389 octet ............................................... 171, 173–178, 180
Glutathione-S-transferase (GST) ................. 144–149,
Pall ForteBio .................................................. 171, 179
353–363, 365, 372–376, 386, 492, 531, 532, Bioluminescence resonance energy transfer
534, 536, 537, 540, 542, 556, 573, 574 (BRET) ............................................... 457–464
6X His.....................................................365–370, 492
Biosensor .................................................... 117, 131, 133,
α-galactosidase assay. See Yeast two-hybrid system 139–143, 150, 165–175, 178–181, 459,
Analyte ...............................110–115, 120–122, 125–132, 555–564, 588
135–137, 141, 165–172, 174, 175, 177,
β-lactamase assays
179–181, 342, 343, 347–350, 588, 589, 598, fluorescent microscopy ......................... 471, 477, 479
600, 601 in vitro colorimetric assay ....................................... 477
Analytical ultracentrifugation. See Sedimentation in vivo enzymatic assay .................................. 477–478
equilibrium
Blot overlay........................................................... 371–378
Association constant .................................. 130, 167, 184,
185, 187–190, 198, 212, 216, 218, 242, 251, C
253, 255, 257, 260
Autoradiography ......................................... 309, 372, 374 Chromatography. See also Gel filtration chromatography
affinity ............................................................. 270, 291
B gel filtration .................................................... 223–232
immobilized metal-affinity...................................... 291
β-galactosidase ............................................ 411, 434, 435, ion exchange................................................... 291, 294
438, 440, 441, 443, 444 size exclusion .................................206, 217, 223, 233
Biacore/GE Healthcare. See Surface plasmon resonance
Chromophores ............................................ 240, 242, 243
Bimolecular Fluorescence Complementation Assay. Circular dichroism (CD)
See Protein fragment complementation assay extrinsic.................................................. 240, 251, 256
Binding affinity.......................................36, 94, 112, 129,
intrinsic ..........................................239, 240, 251, 256
143, 144, 148, 151, 154, 184, 185, 252, 260, methods for analysis
326, 343, 347, 348, 350, 434, 581, 588, 589, constrained multilinear regression
596, 610
(LINCOMB) ............ 244, 246–249, 258, 259
Binding constant ............................... 26, 29, 32, 33, 129, convex constraint algorithm (CCA) ............... 249,
136, 239–242, 251–253, 255, 260, 599 258, 259
Binding equations
neural network programs
Hill plot ..................................................................... 28 (CDNN and K2D)............249, 250, 258, 259
Klotz plot.............................................................25, 26 nonconstrained multilinear regression
Scatchard plot............................................... 25, 27, 28 (MLR)................................ 246–250, 258, 259
Binding equilibrium........................................................ 33
ridge regression (CONTIN) .................. 246, 249,
Binding site.................................... 17, 18, 25–27, 35, 36, 250, 258, 259
72, 79, 80, 85, 97–99, 103, 105, 112–114, singular value decomposition (SVD, VARSLC,
117, 130, 143, 146, 150, 167, 179, 196,
SELCON)..........................249, 250, 258, 259
197, 201, 252, 269, 342, 348, 350, 448, Computational prediction
453, 470, 556, 557, 571, 581, 588, 592, accession number ................................................61, 62
594, 596, 598, 599, 606, 607, 610, 160, 165
vol. 1278, DOI 10.1007/978-1-4939-2425-7, © Springer Science+Business Media New York 2015
615
616 Index
Computational prediction (cont.) E
BLAST ............................................. 65, 71, 78, 86, 89
docking Electrophoresis ....................... 29, 66, 73, 282, 286, 358,
AutoDock .........................................81, 82, 86, 87 423, 436, 486, 489, 501, 503, 504, 509, 510.
DOCK ................................................................. 81 See also SDS-PAGE
flexX ..................................................................... 81 Enhanced chemiluminescence (ECL).........................311,
glide ..................................................................... 81 315–317, 372, 374, 393, 402, 501
GOLD .........................................................81, 581 Enthalpy........................ 8, 184, 185, 187–192, 197–199,
motif prediction .................................................. 60–73 201, 256, 257, 261–263
Pfam .......................................................................3, 15 Entropy .................................. 8, 184, 185, 191, 198, 257
scansite Enzyme-linked immunosorbent assay (ELISA)
database search ..............................62, 66–68, 286, competitive/inhibition ................. 342, 343, 346–350
295–298, 300 direct binding .......................................................... 342
input motif, using .........................................62, 64 sandwich .................................................................. 342
protein sequence .............................57, 61, 62, 65, Equilibrium constants.........................133, 206, 207, 221
66, 69, 78–80, 86, 282, 310, 314
F
public database, from.......................................... 58
quick motifs method.....................................66, 72 Far Western blot. See Blot overlay
scansite motif, using............................... 69, 71, 72 Fast protein liquid chromatography
sequence input ..............................................62, 66 (FPLC) .................................. 226, 230. See also
virtual screening Gel filtration chromatography
bindingDB ........................................................... 83 Flow cytometry ............................................................. 475
ChEMBL ............................................................. 83 Fluorescein ................................323, 324, 456, 470, 473,
drugBank ............................................................. 83 475, 477–479, 530, 578
OEChem ............................................................. 82 Fluorescence
PubChem............................................................. 83 emission ...............................329, 331, 332, 334, 338,
ZINC ................................................................... 82 457, 462, 476, 477, 509, 518, 540, 547, 549
Computational prediction. ........................ 39, 46–48, 50, excitation ............................. 330–332, 334, 457, 458,
53, 57–73. See also Internet resources 462, 476–448, 509, 534, 546, 547, 549
Confocal microscopy Fluorescence-activated cell sorting .............................. 475
detection of endogenous proteins ....... 517, 521, 522 Fluorescence anisotropy. See Fluorescence polarization
detection of transiently expressed Fluorescence Polarization (FP)
proteins ............................................... 116, 517 adaptation for HTS .............................. 472, 476, 477,
Coomassie blue ................................. 286, 292, 301, 358, 529–543, 574, 576, 577
360, 397, 398 fluorophores .................................323, 329, 330, 470,
Crystallography ............................. 4, 5, 58, 80, 101, 102, 517, 518, 530, 546
246, 267, 308, 392, 588, 592, 595, Fluorescence resonance energy transfer (FRET)
596, 601, 608. acceptor fluorophore.....................329, 330, 458, 459
See also X-ray crystallography donor fluorophore .................................................. 329
Fluorescence lifetime imaging microscopy
D (FILM)......................................................... 331
Differential scanning fluorimetry Förster radius/distance.................................. 331, 457
for fragment-based screening ........................ 587–612 live cell imaging.............................................. 334, 335
Dihydrofolate reductase (DHFR) fluorescence assay Time-resolved FRET (TR-FRET) ................ 545–552
flow cytometry analysis ........................................... 475 lanthanide .......................................................... 546
fluorometric analysis ............................................... 476 long-lifetime donor........................................... 546
microscopic analysis ................................................ 512 Fluorescent proteins................................... 331, 483–485,
survival assay .......................................... 470, 472, 474 492, 517. See also DsRed2-Monomer;
Dissociation constant ................................. 24, 27, 30, 32, Fluorescein; Green fluorescent protein (GFP);
33, 126, 130, 136, 138, 162, 167, 184, Red fluorescent protein (RFP); Venus; Yellow
187–190, 199, 201, 240, 251–257, 323, 531, fluorescent protein (YFP)
537, 578, 601 Fluorophore ..............................323, 324, 329, 330, 350,
Druggability ...........................80, 84–86, 94, 95, 99, 104 367, 457, 459, 470, 471, 476, 479, 517, 518,
DsRed2-Monomer............................................... 517, 518 525, 530, 546, 552, 578
Index 617
Förster resonance energy transfer. See Fluorescence Immunoprecipitation
resonance energy transfer co-immunoprecipitation .......................308, 381–389,
Fragment-based drug discovery .............................93–105 484, 492, 516
Fragment-based library............................... 102, 591, 597 Interactomics ................................................................. 449
Internet resources
G bookshelf at NCBI ........................46, 61, 65, 71, 286
Gal4...................409–430, 433, 434, 436, 437, 443, 445 databases
Gateway vectors................................. 368, 412, 413, 451, BIND ...................................................... 45, 46, 51
BioGRID ............... 39, 41, 43, 45, 47, 50, 51, 80
486–489, 493, 494
Gel filtration chromatography CATH .....................................................................5
AKTApurifier FPLC system.................................... 226 DIP ...................................................43, 45–47, 51
MINT ...............................................44–47, 51, 80
size exclusion chromatography .................... 206, 217,
223, 233 Pathguide............................................................. 40
stokes radii ............................................................... 223 Protein Data Bank (PDB) ........................ 3, 8, 14,
15, 58, 78, 82–89, 250, 273–275, 569,
Glutathione matrix...................................... 353, 354, 358
Glutathione-S-transferase (GST) ........................ 353–364 571, 595
Green fluorescent protein (GFP) ....................... 323–326, Reactome .............................39, 42, 44–47, 50, 51
STRING .................................... 39, 41, 47, 50, 51
383, 397, 398, 469–471, 473, 476, 483–485,
493, 500, 516–521, 524, 556–559, 561, 563 Domain searches and interaction predictions
GST-fusion techniques GeneMANIA ....................................39, 47, 49–51
expression and purification interdom .............................................................. 44
InterPro Scan ................................................69, 71
in E. coli ....................................................353–364
in mammalian cells ..................353, 360, 361, 368 Protein-Protein Interaction Prediction
expression plasmids .......................355, 368, 383, 517 (PIPs) ..........................................47, 50, 57–73
ScanSite (see Computational prediction)
lysis buffers ..................................................... 363, 381
radiolabeled technique for identification of novel ExPASy Molecular Biology Server ................ 133, 221
interacting proteins ............................ 354, 362 Locating reagents and protocols
BioSupplyNet ...................................459, 462, 464
H protocol Online................................................. 395
PubMed ...............................................................46, 50
Hexahistidine (6X His)......................114, 116, 365–370, Isopropyl-β-D-1-thiogalactopyranoside
372, 373, 375, 376 (IPTG) ........................................355–358, 363
High content screening Isothermal Titration Calorimetry (ITC) ........... 184–187,
biosensor assay for detection of PPI ............. 555–564 189, 191, 192, 194, 195, 199, 580, 603, 606
High performance liquid chromatography
(HPLC)...................230, 234, 235, 237, 285, K
289, 301
Kinetics ............................... 32, 110, 113, 117, 128, 132,
High throughput screening................................ 139, 141,
143, 149, 151, 449, 459, 468, 498, 499, 542, 133, 137, 141, 165, 167, 168,
546, 569 171–176, 178, 180, 267, 588–590,
601, 611
Homogeneous time resolved fluorescence (HTRF).
See Time-resolved FRET (TR-FRET)
L
Horseradish peroxidase..................... 311, 315, 316, 342,
344, 402 Label-free technology ................................................... 165
LacZ reporter ................................................................ 411
I LexA............................................................. 412, 433, 443
Ligand...............................................7, 13–15, 23–36, 43,
IC50, half maximal inhibitory concentration................ 30,
343, 347–350, 539, 542, 549, 556, 561, 570, 78, 79, 81–85, 89, 98, 100, 110–120,
571, 574, 575, 577, 579–581 122–128, 130–137, 151, 155, 160–162,
165–170, 172–177, 179–181, 183–186, 189,
Immunoblot/immunodetection........................ 282, 292,
382–386, 388, 389, 393, 402, 435, 436, 200, 240, 249, 252, 267–269, 273, 274,
442, 443, 445, 463, 475, 508, 513. See also 277, 324, 350, 402, 447–449, 530,
Western blot 537, 545, 547, 549, 570, 592–595,
597, 603–609
Immunofluorescence .................490, 516, 517, 519, 521
618 Index
Light scattering (LC) Polymerase chain reaction (PCR) ...................... 422, 425,
interferometric refractometer ........................ 234, 236 430, 435, 437–440, 444, 451, 488, 489, 492,
stoichiometry.................................................. 233–238 503, 504, 576, 591, 597, 601, 602
Luciferase Protein
assay ................................................................ 450, 501 co-localization ...................... 3, 46, 51, 329, 515–525
firefly luciferase ...................................... 498, 500, 508 conformation ........................ 239, 243, 244, 249, 332
renilla luciferase .............................................. 457, 458 cross-linking
split-luciferase complementation assay (See Protein cross-linkers ........... 392, 396, 397, 399–401, 403
fragment complementation assay) quantification............................................ 393, 402
in vitro ..................................... 391–393, 396–398
M in vivo ........................................................391–404
digestion ................................................ 283, 285, 300
Mass spectrometry
iTRAQ/TMT....................... 283, 285, 293, 298, 299 domain
LC-MS/MS................. 282, 283, 285–295, 298, 300 interaction......................... 43, 44, 50, 60, 78, 375
search .............................................................44, 78
MALDI-TOF ................................................. 283, 284
shotgun proteomics ....................................... 281, 282 folding............................................288, 366, 469, 493
spectral counting ........................................... 283, 287, identification........................281, 285, 286, 290, 295,
299, 300, 303
291, 298
stable isotope labeling with amino acids in cell motif ..................................14, 60–69, 142, 169, 375,
culture (SILAC) ................................ 283, 285, 447, 448, 459
286, 292, 293, 298, 299, 301, 302 oligomerization ..................................... 205, 402, 485
phosphorylation ............................... 17, 18, 142, 143,
Microscopy ..................................4, 5, 19, 329, 331, 334,
338, 386, 400, 470, 471, 475–477, 479, 480, 290, 377, 387, 447, 448, 531
515, 520, 523 post-translational modification..........................44, 51,
58, 72, 285, 290–291, 299, 412, 459
N quantification................................285–287, 291, 296,
298, 300, 473
Nuclear magnetic resonance (NMR) secondary structure (see Protein Secondary structure)
chemical-shift perturbation .................................... 606 thermodynamics ............................240, 241, 253, 257
for fragment-based screening ...............588, 591–595, unfolding ........................................................ 575, 590
603, 605 Protein A/Protein G ................................. 115, 251–254,
Heteronuclear single quantum correlation 332, 334–336, 338, 382, 384, 385, 387, 388,
(HSQC) ................... 270–272, 274, 276, 277, 478, 479, 576
579, 593, 594, 606, 607 Protein fragment complementation assay (PCA)
molecular docking......................................78, 86, 268 applications ..................................................... 470, 471
paramagnetic relaxation enhancement.................. 268, general principle ............................................. 467–480
269, 273 PCA reporters
saturation transfer difference ......................... 593, 603 β-lactamase assays (see β-lactamase assays)
WaterLOGSY........................................................... 606 Dihydrofolate reductase (DHFR) fluorescence
assay (see Dihydrofolate reductase (DHFR)
O
fluorescence assay)
Overlay assay. See Blot overlay Green fluorescent protein (GFP) assay ............ 471
Split-luciferase complementation
P assay......................................................497–513
Venus bimolecular fluorescence complementation
Peptide arrays
assay.................................... 484–488, 490–494
alanine screening ....................................308, 315–317
Protein Secondary structure
fmoc-chemistry........................................................ 307
identifying pockets
ResPep SL SPOT ........................................... 308, 312
CavBase................................................................ 79
Peptide library ................................................................. 99
fpocket .............................................. 79–81, 84–86
Phage display ................................................................... 99
FTSite ...............................................79, 80, 84–86
Pharmacophore screen......................................... 580–581
SiteMap............................................. 79–81, 84–86
Phosphopeptide ......................................... 286, 291, 302,
Voronoi tessellation............................................. 79
531, 532
Proteomics.............................. 18, 48, 67, 109, 221, 224,
Photobleaching ........................................... 331, 336, 338
281–283, 285, 293, 295
Polyacrylamide gel. See Electrophoresis
Index 619
Q Binding kinetics...................113, 117, 128, 133, 137,
165, 167, 168, 174, 176, 180, 588, 590
QCM-D chip selection .........................................111–126, 128,
binding affinity ........................................................ 154 130, 133, 136, 597, 598
EDC/NHS crosslinking ......................................... 158 for fragment-based screening ........................ 102, 603
sauerbrey equation ........................154, 160, 161, 164 sensor surface immobilization ...................... 112, 114,
sensor chip ...................................................... 161, 163 597, 598
Quantitative analysis ...................... 23–36, 294, 302, 350
T
R
ThermoFluor microcalorimetry .......................... 276, 575
Red fluorescent protein (RFP) ........... 556–559, 561–563 Time-resolved FRET (TR-FRET) ...................... 545–552
Resonant waveguide grating Transfection...................... 292, 334, 355, 361, 368–370,
biosensor.................................................139–143, 150 383, 384, 387, 394, 397–399, 401, 438,
corning Epic ................................................... 139–151 450–452, 459–461, 464, 472–477, 486, 490,
evanescent wave.............................................. 140, 141 491, 493, 494, 500, 502, 504, 505, 508, 512,
513, 516, 517, 519, 524, 525, 556
S
Transformation............................ 27, 195, 357, 413–425,
Sacchromyces cerevisia strains 427–430, 438–441, 444, 488–490
AH109 ........................ 411, 413, 414, 416–420, 424, Two-hybrid systems
425, 427 mammalian two-hybrid assay
Y187...............................................413, 418, 424, 425 MAmmalian Protein-Protein Interaction Trap
Y190....................................................... 436, 437, 443 (MAPPIT) ...........................................447–454
Scansite. See Computational prediction Plasmids ............................................449–451, 453
Screening library ........................................................... 101 reverse two-hybrid ......................................... 433–445
SDS-PAGE ................................................. 355, 358, 360, Characterization of isolated mutants .......433–445
362, 369, 371–374, 376, 383, 385, 393, 396, loss-of-interaction mutation ................... 434, 435,
398, 400, 401, 403, 425, 430, 436, 501, 510. 438, 440
See also Western blot mutagenesis ......................................435, 437–439
Sedimentation equilibrium plasmids ................................... 438, 439, 441–445
equilibrium constants............................ 206, 207, 221 strains (see Sacchromyces cerevisia strains)
heterologous system ............................................... 219 Yeast two-hybrid system ................................ 409–429
monomeric species ......................................... 214–216 3-AT (3-amino-1,2,4-triazole)........411, 418, 425
Sepharose beads ............................................................ 532 Bait auto-activation test .................................... 413
Size exclusion chromatography........ 206, 217, 223, 233. confirmation test ...................................... 417, 424
See also Gel filtration chromatography mating ...............................................413, 424, 425
Split-luciferase complementation assay ............... 497–513 medium................................... 411, 414–422, 424,
Dual-color split luciferase complementation 425, 428, 430
assay.............................................................. 511 plasmid isolation......................413, 417, 422, 443
Stoichiometry ................................ 26, 27, 29, 30, 32, 33, plasmids ..................................412, 414, 416, 417,
42, 83, 103, 113, 177, 196, 201, 212, 423–425, 427, 429
233–238, 255, 290, 332, 580, 588, 589, quantitative α-galactosidase assay........... 413, 418,
591, 600, 601, 606 425–426
Structural basis screening libraries..................................... 102, 587
crystallography ................................. 4, 5, 58, 80, 101, strains (see Sacchromyces cerevisia strains)
102, 246, 267, 308, 392, 588, 592, 595, 596,
601, 608 U
Domain-domain interaction.................................7, 23 Ubiquitin .............................................290, 300, 412, 568
Domain-peptide interaction ............... 6, 15, 151, 540
"Hot spot," ................................... 7–9, 11–13, 81, 94 V
signaling complex........................................... 6, 12, 13
Surface plasmon resonance (SPR) Venus ..................................................331, 332, 334–338,
Biacore/GE Healthcare...............110, 121, 122, 211, 458, 460, 484–488, 490–494
393, 532, 570, 589 Virtual Screening. See Computational prediction
620 Index
W X
Websites ............ 16, 286, 552. See also Internet resources X-α-Gal (5-bromo-4-chloro-3-indolyl-α-D-galactop-
CD analysis ..................................................... 239–263 yranoside)............................................ 415, 436
Cross-linkers ......................................... 392, 396, 397, X-ray crystallography
399–401, 403 for fragment-based screening ....................... 588, 595,
pGEX vectors ................................ 355–357, 359, 361 596, 601, 608
Protein Data Bank (PDB) ........................3, 8, 14, 15,
58, 78, 83–89, 273–275, 569, 571, 595 Y
Protein domains ............................44, 50, 60, 78, 375 Yeast two-hybrid. See Two-hybrid systems
Scansite ................................................... 61–63, 65–73 Yeast two-hybrid system
Sedimentation equilibrium ............................ 205–222
Medium
Two-hybrid systems .............................. 433, 434, 445 LB .....................................................417, 422, 423
Western blot ............................................... 292, 300, 342, SD .........415, 416, 419, 420, 422, 425, 426, 428
358, 360, 362, 365, 368–377, 393, 400–402, YPDA ........................................................ 414, 416
475, 479, 490, 501, 508, 509, 518, 521.
Yellow fluorescent protein (YFP) .............. 336, 458–461,
See also Immunoblot/immunodetection 463, 484, 485, 491, 492
Far Western blot (see Blot overlay)

Protein Protein Interactions Methods and

Uploaded by

Copyright:

Available Formats

Protein Protein Interactions Methods and

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Protein Protein Interactions Methods and

Uploaded by

Copyright:

Available Formats

Methods in

Molecular Biology 1278

For further volumes:

Methods and Applications

ISSN 1064-3745 ISSN 1940-6029 (electronic)

Springer New York Heidelberg Dordrecht London

Printed on acid-free paper

Humana Press is a brand of Springer

From regulation of DNA replication, RNA transcription, and protein translation to

Atlanta, GA, USA Cheryl L. Meyerkord

1 Structural Basis of Protein-Protein Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

PART II LABEL–FREE BIOSENSORS AND TECHNIQUES

7 Studying Protein-Protein Interactions Using Surface Plasmon Resonance . . . . 109

16 Protein-Protein Interaction Analysis by Nuclear Magnetic

PART III TAG/AFFINITY–BASED METHODS

PART IV CELL–BASED BIMOLECULAR INTERACTION REPORTER ASSAYS

27 Identification of Protein-Protein Interactions by Standard

30 Bioluminescence Resonance Energy Transfer to Detect

PART V HIGH THROUGHPUT SCREENING ASSAYS FOR PROTEIN-PROTEIN

35 Fluorescence Polarization Assay to Quantify Protein-Protein

FABRICE AGOU Département de Biologie Cellulaire et Infection, Institut Pasteur, Unité de

KUN-LIANG GUAN Department of Pharmacology and Moores Cancer Center, University of

INGRID REMY Département de Biochimie, Université de Montréal, Montréal, QC, Canada

ZHIPING WU Department of Structural Biology, Department of Neurodevelopmental

Structural Basis of Protein-Protein Interactions

Key words Structure, Complex, Energetics, Crystallography, Allostery, Regulation, Force,

Proteins live, work, and die in a highly crowded environment and

Fig. 1 Atomic structure from single-particle electron microscopy. Structure of the

small subset of the complete human binary protein-protein interac-

well to automation, neither has the crystallization of large

2 Classification of Binary Protein Complexes

Interacting domain-pairs may be classified as reversible and irrevers-

An independent classification distinguishes “domain–domain”

3 The Architecture and Energetics of Domain–Domain Interfaces

A number of general principles have arisen from the study of known

4 Residue Usage at Interfaces

As the number and type of interfaces found in the PDB have

interfaces, as enumerated below. Note that while this discussion

of each residue, grouped by type in increasing order of their

4.2 Interface A popular measure of interface composition is the “interface pro-

and polar interactions, all of which are required to bury comple-

5 Complex Formation and Conformational Changes

The integrin family of αβ heterodimeric plasma membrane

Fig. 5 Conformational changes in the α2 I domain upon binding collagen. Upper

The β-subunit of integrins also contains an I-domain. Although

conformational changes within the C-terminal helix linked to

6 Protein-Peptide Interactions: A Key Mediator of Complexity in Higher Eukaryotes

As noted above, recent years have witnessed an explosion in the

a larger intrinsically disordered region (IDR) that is encoded by a

alternative exon usage, which is enriched in SLiMs and IDRs,

7 Intramolecular Protein-Protein Interactions and Signal Transduction

Signal transduction involves much more than the regulated,

molecules are likely to be bound, at least transiently.

8 Force and the Single Molecule

Many intracellular signaling/scaffolding molecules switch from a

The complexity of conformational switches in signal transduction

Quantitative Analysis of Protein-Protein Interactions

In the post-genomic era, the importance of protein-protein inter-

2. All receptor molecules are equivalent and independent.

½Rf ½L f ð½Rt ½RL Þð½L t ½RL Þ

0 ¼ ½RL 2 ð½Rt þ ½L t þ K d Þ þ ½Rt ½L t ; ð6Þ

2a 3 þ 9ab 27c

constant of 1013 M. Here, binding may take as long as hours and

[Rt] Kd If the concentration of Rt is much greater than Kd then Eq. 8 can