2018 Book MetabolicProfiling

Methods in
Molecular Biology 1738
Georgios A. Theodoridis
Helen G. Gika
Ian D. Wilson Editors
Metabolic
Profiling
Methods and Protocols
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
Metabolic Profiling
Methods and Protocols
Edited by
Georgios A. Theodoridis
Department of Chemistry, Aristotle University of Thessaloniki, Thessaloniki, Greece
Helen G. Gika
School of Medicine, Aristotle University of Thessaloniki, Thessaloniki, Greece
Ian D. Wilson
Department of Surgery and Cancer, Imperial College London, London, UK
Editors
Georgios A. Theodoridis Helen G. Gika
Department of Chemistry School of Medicine
Aristotle University of Thessaloniki Aristotle University of Thessaloniki
Thessaloniki, Greece Thessaloniki, Greece
Ian D. Wilson
Department of Surgery and Cancer
Imperial College London
London, UK
ISSN 1064-3745 ISSN 1940-6029 (electronic)

Methods in Molecular Biology
ISBN 978-1-4939-7642-3 ISBN 978-1-4939-7643-0 (eBook)
https://doi.org/10.1007/978-1-4939-7643-0
Library of Congress Control Number: 2018931170
© Springer Science+Business Media, LLC, part of Springer Nature 2018

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.
The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Humana Press imprint is published by the registered company Springer Science+Business Media, LLC part of
Springer Nature.
The registered company address is: 233 Spring Street, New York, NY 10013, U.S.A.
Preface
This book provides a number of protocols for “global metabolic profiling,” also known as
metabonomics and/or metabolomics. Metabolomics deals with the holistic analysis of small
molecules aiming to characterize the metabolic content of the studied samples/systems and
reveal changes that result from alterations to them as a result of, e.g., different physiological
states or the onset and progression of disease, etc. Over the last few decades, there have
been significant developments in both analytical technologies and multivariate statistical
methods that have greatly facilitated the growth of these holistic analytical approaches.
In putting together this volume, the editors have placed emphasis on obtaining chap-
ters that illustrate the different approaches taken by researchers to develop tools to address
the important challenges of the field. The first part of the book contains chapters on the
challenges and perspective of the topic (Gika et al.), the use of quality control measures
(QC) and validation issues (Begou et al.), data mining (Riccadonna and Francheschi), and
bio- and chemoinformatic tools for metabolomics (Witting). These chapters highlight basic
concepts such as experimental design, data treatment, metabolite identification, the need
for harmonization, and the linking of data obtained by different analytical modes (also
combining metabolomics results with data from other omics fields).
The second section, which is concerned with methodology, describes protocols for sam-
ple preparation centered on techniques for tissues, feces, and blood samples (Michopoulos,
Deda et al., also addressed by Vorkas et al.) and chemical derivatization for GC-MS (Hušek
et al.). The methods used for metabolite analysis and profiling are covered with chapters on
GC-MS metabolic profiling (Klapa et al.), LC-MS profiling using both targeted methods
(Virgiliou et al.) and IPC-LC-MS (Michopoulos), and untargeted (Want) profiling
approaches. The profiling of polar charged metabolites still remains a challenge, and this
section includes a chapter on the use of CE-MS for this purpose (Ramautar). NMR spectro-
scopic methods for profiling biological fluids (Benaki and Mikros) are also considered.
The volume concludes with two application sections covering the use of metabolomics
in life sciences with examples of methodologies that can be found in food science or bio-
marker discovery for disease diagnosis and human well-being. In the case of food and natu-
ral products, the protocols describe the analytical methods used and their application in
food quality control, where the use of NMR spectroscopy is described (Schripsema and
Dagnino) and the evaluation of product authenticity and geographical origin (Spyros and
coworkers). Both these issues represent major challenges for the food industry and are still
a great concern for the health of the consumer. The use of proton-transfer-reaction time-
of-flight mass spectrometry (PTR-TOF-MS) for the analysis of volatile organic chemicals
(VOCs) is also described (Farneti). Arapitsas and Mattivi describe a protocol on the analysis
of wine by LC-MS with application to the classification of wine according to the grape
variety. In the case of applications in life sciences, the use of metabolic profiling for bio-
marker discovery in cardiovascular disease (Vorkas et al.) and the targeted analysis of ste-
roids (Rudaz and coworkers) are described. Finally, Siopi and Mougios discuss experimental
design and considerations on sample collection for studies involving human subjects.
v
vi Preface
While still an area of rapid technical development, the place of “omic” metabolic phe-
notyping where the objective is to gain unbiased, global knowledge of the content of the
studied system, is firmly fixed as a means of gaining insights into the conditions under
study, thereby enhancing our knowledge and detailed understanding of the phenomena
under investigation.
Thessaloniki, Greece Georgios A. Theodoridis

Thessaloniki, Greece Helen G. Gika
London, UK Ian D. Wilson
Contents
Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Contributors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Part I Fundamentals
1 Metabolic Profiling: Status, Challenges, and Perspective�� 3

Helen G. Gika, Georgios A. Theodoridis, and Ian D. Wilson
2 Quality Control and Validation Issues in LC-MS Metabolomics�� 15
Olga Begou, Helen G. Gika, Georgios A. Theodoridis, and Ian D. Wilson
3 Data Treatment for LC-MS Untargeted Analysis�� 27
Samantha Riccadonna and Pietro Franceschi
4 Bio- and Chemoinformatics Approaches for Metabolomics Data Analysis�� 41
Michael Witting
Part II Methods
5 HILIC-MS/MS Multi-Targeted Method for Metabolomics Applications�� 65

Christina Virgiliou, Helen G. Gika, and Georgios A. Theodoridis
6 Ion Pair Chromatography for Endogenous Metabolites LC-MS
Analysis in Tissue Samples Following Targeted Acquisition �� 83
Filippos Michopoulos
7 LC-MS Untargeted Analysis �� 99
Elizabeth J. Want
8 NMR-Based Metabolic Profiling Procedures for Biofluids
and Cell and Tissue Extracts �� 117
Dimitra Benaki and Emmanuel Mikros
9 Untargeted GC-MS Metabolomics �� 133
Matthaios-Emmanouil P. Papadimitropoulos, Catherine G. Vasilopoulou,
Christoniki Maga-Nteve, and Maria I. Klapa
10 Rat Fecal Metabolomics-Based Analysis�� 149
Olga Deda, Helen G. Gika, and Georgios A. Theodoridis
11 GC-MS Metabolomic Profiling of Protic Metabolites
Following Heptafluorobutyl Chloroformate Mediated Dispersive
Liquid Microextraction Sample Preparation Protocol�� 159
Petr Hušek, Zdeněk Švagera, Dagmar Hanzlíková, Iva Karlínová,
Lucie Řimnáčová, Helena Zahradníčková, and Petr Šimek
12 Sheathless Capillary Electrophoresis-Mass Spectrometry
for the Profiling of Charged Metabolites in Biological Samples�� 183
Rawi Ramautar
vii
viii Contents
Part III Plant/Food Applications
13 Two-Phase Extraction for Comprehensive Analysis of the Plant

Metabolome by NMR�� 195
Jan Schripsema and Denise Dagnino
14 NMR Spectroscopy Protocols for Food Metabolomics Applications�� 203
Evangelia Ralli, Maria Amargianitaki, Efi Manolopoulou,
Maria Misiak, Georgios Markakis, Sofia Tachtalidou,
Alexandra Kolesnikova, Photis Dais, and Apostolos Spyros
15 Direct Injection Analysis of Fruit VOCs by PTR-ToF-MS:
The Apple Case Study�� 213
Brian Farneti
16 LC-MS Untargeted Protocol for the Analysis of Wine �� 225
Panagiotis Arapitsas and Fulvio Mattivi
Part IV Life Science Applications
17 Tissue Multiplatform-Based Metabolomics/Metabonomics

for Enhanced Metabolome Coverage�� 239
Panagiotis A. Vorkas, M. R. Abellona U, and Jia V. Li
18 UHPLC-HRMS Analysis for Steroid Profiling in Serum (Steroidomics)�� 261
Federico Ponzetto, Julien Boccard, Raul Nicoli, Tiia Kuuranne,
Martial Saugy, and Serge Rudaz
19 Metabolomics in Human Acute-Exercise Trials: Study Design
and Preparation�� 279
Aikaterina Siopi and Vassilis Mougios
Index �� 289

Contributors
M.R. Abellona U • Section of Biomolecular Medicine, Division of Computational

and Systems Medicine, Department of Surgery and Cancer, Faculty of Medicine,
Imperial College London, London, UK
Maria Amargianitaki • NMR Laboratory, Chemistry Department,
University of Crete, Heraklion, Crete, Greece
Panagiotis Arapitsas • Department of Food Quality and Nutrition, Research and
Innovation Centre, Fondazione Edmund Mach (FEM), San Michele all’Adige, Italy
Olga Begou • Department of Chemistry, Aristotle University of Thessaloniki,
Thessaloniki, Greece
Dimitra Benaki • Department of Pharmaceutical Chemistry, National and Kapodistrian
University of Athens, Athens, Greece
Julien Boccard • School of Pharmaceutical Sciences, University of Geneva,
University of Lausanne, Geneva 4, Switzerland; Swiss Center of Applied Human
Toxicology (SCAHT), University of Basel, Basel, Switzerland
Denise Dagnino • Grupo Metabolômica, Universidade Estadual do Norte Fluminense,
Campos dos Goytacazes, Rio de Janeiro, Brazil
Photis Dais • NMR Laboratory, Chemistry Department, University of Crete,
Heraklion, Crete, Greece
Olga Deda • Department of Chemistry, Aristotle University of Thessaloniki,
Thessaloniki, Greece
Brian Farneti • Genomics and Biology of Fruit Crop Department, Research and
Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy
Pietro Franceschi • Computational Biology Unit, Research and Innovation Centre,
Fondazione E. Mach, Trento, Italy
Helen G. Gika • School of Medicine, Aristotle University of Thessaloniki, Thessaloniki,
Greece
Dagmar Hanzlíková • Institute of Laboratory Diagnostics, Department of Biochemistry,
University Hospital Ostrava, Ostrava, Czech Republic
Petr Hušek • Institute of Laboratory Diagnostics, Department of Biochemistry, University
Hospital Ostrava, Ostrava, Czech Republic; Biology Centre, Institute of Entomology,
Analytical Biochemistry & Metabolomics, Czech Academy of Sciences, České Budějovice,
Czech Republic
Iva Karlínová • Biology Centre, Institute of Entomology, Analytical Biochemistry &
Metabolomics, Czech Academy of Sciences, České Budějovice, Czech Republic
Maria I. Klapa • Metabolic Engineering and Systems Biology Laboratory, Institute of
Chemical Engineering Sciences, Foundation for Research & Technology - Hellas
(FORTH/ICE-HT), Patras, Greece; Department of Chemical and Biomolecular
Engineering, University of Maryland, College Park, MD, USA; Department of
Bioengineering, University of Maryland, College Park, MD, USA
ix
x Contributors
Alexandra Kolesnikova • NMR Laboratory, Chemistry Department, University of Crete,

Tiia Kuuranne • Swiss Laboratory for Doping Analyses, University Center of Legal
Medicine Geneva and Lausanne, Centre Hospitalier Universitaire Vaudois and
University of Lausanne, Lausanne, Switzerland
Jia V. Li • Section of Biomolecular Medicine, Division of Computational and Systems
Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College
London, London, UK; Centre for Digestive and Gut Health, Institute of Global Health
Innovation, Imperial College London, London, UK
Christoniki Maga-Nteve • Metabolic Engineering and Systems Biology Laboratory,
Institute of Chemical Engineering Sciences, Foundation for Research & Technology -
Hellas (FORTH/ICE-HT), Patras, Greece; School of Medicine, University of Patras,
Patras, Greece
Efi Manolopoulou • NMR Laboratory, Chemistry Department, University of Crete,
Georgios Markakis • NMR Laboratory, Chemistry Department, University of Crete,
Fulvio Mattivi • Department of Food Quality and Nutrition, Research and Innovation
Centre, Fondazione Edmund Mach (FEM), San Michele all’Adige, Italy; Center
Agriculture Food Environment, University of Trento, San Michele all’Adige, Italy
Filippos Michopoulos • IMED Oncology, AstraZeneca, Macclesfield, Cheshire, UK
Emmanuel Mikros • Department of Pharmaceutical Chemistry, National and
Kapodistrian University of Athens, Athens, Greece
Maria Misiak • NMR Laboratory, Chemistry Department, University of Crete,
Vassilis Mougios • School of Physical Education and Sport Science at Thessaloniki,
Aristotle University of Thessaloniki, Thessaloniki, Greece
Raul Nicoli • Swiss Laboratory for Doping Analyses, University Center of Legal Medicine
Geneva and Lausanne, Centre Hospitalier Universitaire Vaudois and University of
Lausanne, Lausanne, Switzerland
Matthaios-Emmanouil P. Papadimitropoulos • Metabolic Engineering and Systems
Biology Laboratory, Institute of Chemical Engineering Sciences, Foundation for Research
& Technology - Hellas (FORTH/ICE-HT), Patras, Greece; Division of Genetics, Cell &
Developmental Biology, Department of Biology, University of Patras, Patras, Greece
Federico Ponzetto • Swiss Laboratory for Doping Analyses, University Center of Legal
Medicine Geneva and Lausanne, Centre Hospitalier Universitaire Vaudois and
Evangelia Ralli • NMR Laboratory, Chemistry Department, University of Crete,
Rawi Ramautar • Leiden Academic Center for Drug Research, Leiden University, Leiden,
The Netherlands
Samantha Riccadonna • Computational Biology Unit, Research and Innovation Centre,
Fondazione E. Mach, Trento, Italy
Lucie Řimnáčová • Biology Centre, Institute of Entomology, Analytical Biochemistry &
Contributors xi
Serge Rudaz • School of Pharmaceutical Sciences, University of Geneva, University of

Lausanne, Geneva 4, Switzerland; Swiss Center of Applied Human Toxicology (SCAHT),
University of Basel, Basel, Switzerland
Martial Saugy • Center of Research and Expertise in Anti-Doping Sciences,
Jan Schripsema • Grupo Metabolômica, Universidade Estadual do Norte Fluminense,
Campos dos Goytacazes, RJ, Brazil
Petr Šimek • Biology Centre, Institute of Entomology, Analytical Biochemistry &
Aikaterina Siopi • School of Physical Education and Sport Science at Thessaloniki,
Aristotle University of Thessaloniki, Thessaloniki, Greece; Department of Physical
Education and Sport Science at Thermi, Aristotle University of Thessaloniki, Thessaloniki,
Greece
Apostolos Spyros • NMR Laboratory, Chemistry Department, University of Crete,
Zdeněk Švagera • Institute of Laboratory Diagnostics, Department of Biochemistry,
University Hospital Ostrava, Ostrava, Czech Republic
Sofia Tachtalidou • NMR Laboratory, Chemistry Department, University of Crete,
Georgios A. Theodoridis • Laboratory of Forensic Medicine and Toxicology, Department
of Chemistry, Aristotle University of Thessaloniki, Thessaloniki, Greece
Catherine G. Vasilopoulou • Metabolic Engineering and Systems Biology Laboratory,
Institute of Chemical Engineering Sciences, Foundation for Research & Technology -
Hellas (FORTH/ICE-HT), Patras, Greece; Human and Animal Physiology Laboratory,
Department of Biology, University of Patras, Patras, Greece
Christina Virgiliou • Laboratory of Forensic Medicine and Toxicology, Department of
Chemistry, Aristotle University of Thessaloniki, Thessaloniki, Greece
Panagiotis A. Vorkas • Section of Biomolecular Medicine, Division of Computational
and Systems Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial
College London, London, UK
Elizabeth J. Want • Computational and Systems Medicine, Imperial College London,
London, UK
Ian D. Wilson • Department of Surgery and Cancer, Imperial College London, London, UK
Michael Witting • Research Unit Analytical BioGeoChemistry, Helmholtz Zentrum
München – German Research Center for Environmental Health, Neuherberg, Germany;
Chair of Analytical Analytical Food Chemistry, Wissenschaftszentrum Weihenstephan für
Ernährung, Landnutzung und Umwelt, Technische Universität München, Freising,
Germany
Helena Zahradníčková • Biology Centre, Institute of Entomology, Analytical
Biochemistry & Metabolomics, Czech Academy of Sciences, České Budějovice,
Czech Republic
Part I
Fundamentals
Chapter 1
Metabolic Profiling: Status, Challenges, and Perspective

Helen G. Gika, Georgios A. Theodoridis, and Ian D. Wilson
Abstract
Metabolic profiling has advanced greatly in the past decade and evolved from the status of a research topic
of a small number of highly specialized laboratories to the status of a major field applied by several hun-
dreds of laboratories, numerous national centers, and core facilities. The present chapter provides our view
on the status of the remaining challenges and a perspective of this fascinating research area.
Key words Metabolomics, Metabonomics, Biomarker, Metabolite identification, MetID, Biochemical

pathway
1 Introduction
The field of untargeted metabolic profiling, also known as metabo-

nomics/metabolomics [1, 2] or metabotyping [3], involves the
study of the small molecule complement of samples such as bio-
logical fluids (plasma, serum, urine) cells, organs, or whole organ-
isms. The earliest examples of the use of “holistic,” untargeted,
and hypothesis-free metabolic phenotyping can perhaps be traced
back to the work of Dent and Dalgliesh [4, 5] who, in the late
1940s, used two-dimensional paper chromatography for the dis-
covery of new disease biomarkers. This early work was followed by
studies by Pauling and colleagues based on the use of gas chroma-
tography to profile urinary volatiles for disease diagnosis [6–8].
Pioneering work on metabolic fingerprinting based on the use of
liquid chromatography was also being undertaken at this time
(e.g., [9, 10]).
It can be argued that such studies comfortably precede genomic
(and proteomic) profiling by some years and falsify the repeated,
and erroneous, statement that metabolic phenotyping is the latest
addition to the “omics field,” rather than the first! However, it was
only with the availability of analytical systems that combined rapid,
multianalyte detection with an element of structural information
that, when used in combination with multivariate statistical a nalysis,
Georgios A. Theodoridis et al. (eds.), Metabolic Profiling: Methods and Protocols, Methods in Molecular Biology, vol. 1738,
https://doi.org/10.1007/978-1-4939-7643-0_1, © Springer Science+Business Media, LLC, part of Springer Nature 2018
3
4 Helen G. Gika et al.
enabled the relatively rapid detection and identification of the

potential biomarkers hidden in the profiles. These techniques, cen-
tered on 1H nuclear magnetic resonance (NMR) spectroscopy and
mass spectrometry (MS), with the latter often hyphenated to a
separation such as liquid or gas chromatography (LC, GC), per-
mitted the full development of the field and resulted in a rapid
increase in the number of research groups active in metabolic phe-
notyping. This increase in research activity was accompanied by a
dramatic, and still increasing, rise in the volume and sophistication
of publications on the topic (see Fig. 1) and, while still lagging
somewhat in number behind those of the genomics and pro-
teomics, has now reached many hundreds per year.
However, with this expanding application of metabolic pheno-
typing in virtually all areas of the life sciences, there has been an
increasing realization of the need for careful study design and stan-
dardization of methodology. This has also been accompanied by a
better understanding of the advantages and limitations of the
“holistic” approach to biomarker discovery and to the rise of more
“targeted” approaches. So, untargeted methods have the major
advantage that they are unencumbered by the preconceptions of
the investigator and are therefore more likely to discover novel,
and unexpected, metabolites. They are therefore ideally hypothesis-
free, but hypothesis generating. However, a problem with this
approach can be that many metabolites are seen to change in the
test population relative to the control group that are not specific to
the condition under investigation but rather represent a general
response to effects such as stress or environmental factors.
Disentangling these general, nonspecific changes from the direct
effects on metabolites more directly/mechanistically involved in
the process(es) under investigation can sometimes require signifi-
cant effort.
Another related concern is how comprehensive the coverage of
the metabolome can be using the techniques available. This again
is by no means a trivial question as, unlike, e.g., the human genome
which has been fully sequenced and the number of genes esti-
mated, it cannot yet be claimed that the extent of the metabolome
has been fully mapped. Also, the range of concentrations that the
metabolites encompass from, presumably, zeptomoles to micro-
moles makes demands on the available analytical techniques that
are, to say the least, challenging. While there have been very sig-
nificant, and continuing, advances in analytical technologies, these
have, in some ways, only served to highlight their limitations.
Indeed, while the minimum specification for the “perfect metabo-
typer” is easy to produce, with such instruments required to be
universally sensitive, nonspecific, unbiased, rapid, robust, and sta-
ble, quantitatively possessed of a large dynamic range and provid-
ing enough information on each of the components of the sample
Metabolic Profiling: Status, Challenges, and Perspective 5
9000 number of publications per year

8000
7000
6000
5000
4000
3000
2000
1000
0
2000 2002 2004 2006 2008 2010 2012 2014 2016
Genomics Proteomics Transcriptomics Metabolomics
Fig. 1 Publication trends in the areas of genomics/transcriptomics, proteomics, and metabolic phenotyping
from 2000 to 2016 generated from references contained within SCOPUS using the search terms genomics,
proteomics, metabolomics or metabonomics or metabolic profiling, and transcriptomics (search made June
2017)
as to provide unequivocal identification, comparison with what is

currently available for this type of work reveals a large gap in capa-
bility. And indeed, although several technologies have some of the
characteristics of the hypothetical “perfect metabotyper,” there is
no single analytical solution/platform that provides the means of
obtaining comprehensive metabolic profiles.
For this reason attempts at ensuring the maximum recovery of
metabolite information currently depend on the use of several ana-
lytical platforms and methods. In the case of 1H NMR spectros-
copy, the use of 600 MHz instruments provides a good compromise
between field strength, sensitivity, dynamic range, and affordabil-
ity. Such instruments offer a means for the rapid analysis of biofluid
samples such as urine and plasma and provide informative spectral
data permitting structural characterization and identification
(helped by the availability of good spectral libraries) [11]. In terms
of the ease of use for the analysis of liquid samples, 1H NMR spec-
troscopy usually requires minimal sample preparation other than
filtering and pH adjustment. Solid, or semisolid samples such as
tissues, can be analyzed either following extraction into a suitable
solvent or in the native state using “magic angle spinning” NMR
spectroscopy (although uptake of this technology in metabolic
profiling applications has been limited) [11]. A typical example of
the information available from the use of 1H NMR spectroscopy in
a comparative toxicology study on acetaminophen (APAP) and its
related, less toxic, meta-isomer AMAP is shown in Fig. 2 for “aque-
ous” liver extracts obtained from control and treated mice [12].
These spectra can readily be used to distinguish between extracts
obtained from different treatments and show the presence of both
Fig. 2 Representative 1H-NMR spectra of hepatic extract metabolic profiles of the acetaminophen (APAP),
AMAP, and control groups at 1 h. Resonances assigned to drug-related molecules have been colored in red.
Key: APAP/AMAP-G APAP/AMAP glucuronide, APAP-SG APAP glutathionyl, APAP-NAC APAP-N-acetylcysteinyl,
APAP/AMAPNHCOCH3 APAP/AMAP N-acetyl resonance, GSH reduced glutathione, GSSG oxidized glutathione,
Phe phenylalanine, d-3-HB d-3-hydroxybutyrate, AMP adenosine monophosphate, overlapped resonances
from glucose/glycogen/maltose labeled. From reference [12] reprinted with permission
drug metabolites and endogenous metabolites present in the sam-

ples and give an insight into the metabolic changes that result from
the administration of the test compounds to this species. A
particular advantage of NMR spectroscopy-based methods is that
they are inherently reproducible, and, e.g., one-dimensional 1H
NMR spectra acquired, at the same field strength on the same sam-
ple, should give very similar results irrespective of the laboratory or
instrument manufacturer.
Mass spectrometry can, like NMR spectroscopy, be used for
the direct analysis of liquid (or gaseous) samples by the simple
expedient of directly infusing the samples into the ion source of the
instruments (“DIMS”) [13], and this approach, using either flow
injection analysis or specialist interfaces such as the “nanomate,”
has many advantages in terms of simplicity and speed. However,
particularly in the case of complex matrices such as urine- or blood-
derived samples, ion suppression can be problematic. In addition it
can be difficult to determine isobaric or isomeric compounds if
they are present as mixtures using this approach. This has led to the
adoption of the use of hyphenated techniques involving a chro-
matographic or electrophoretic separation prior to MS. Thus liq-
uid and gas chromatography-mass spectrometry (LC-MS, GC-MS)
and capillary zone electrophoresis-MS (CZE-MS or CE-MS) are
very widely used for metabolic phenotyping, with LC-MS-based
techniques currently in the ascendancy. GC-MS-based methods
are, of course, particularly well suited to the analysis of volatile
analytes present in, e.g., breath [14]. However, as will be evident
from even a brief survey of the metabolic phenotyping literature,
GC-MS-based methods have found a widespread use in metabolic
phenotyping for involatile metabolites following chemical modifi-
cation, via carefully optimized “derivatization” protocols, to make
the analytes volatile [15]. This derivatization step should be stud-
ied meticulously since different metabolites react with different
rates and the reaction conditions applied can affect the outcome of
the analysis [16]. Procedures are available for the preparation of
volatile derivatives of most analyte classes such as amino acids, sug-
ars, polar acids, etc., and these can be applied to suitable extracts
for a broad range of sample types (plasma, urine, plant extracts,
food, etc.). Clearly, the need for fairly extensive sample preparation
and derivatization required for the analysis of involatile metabolites
means that getting to the point of actually performing the GC-MS
analysis can be quite time-consuming and labor intensive. However,
GC-MS with electron impact (EI) is a mature and robust technol-
ogy, supported with a numerous databases of spectral data (NIST,
Fiehn, etc.) to support the identification of potential biomarkers
[14].
In the case of LC-MS-based metabolic phenotyping, the cur-
rent state of the art employs methods based on so-called ultra
(high) performance LC separations (UPLC, UHPLC) based on

the use of high operating pressures and stationary phases formed
from sub 2 μm-sized particles [17–19]. Chromatography on such
phases provides excellent chromatographic efficiency enabling
high-resolution separations to be obtained in reasonable analysis
times (typically 5–15 min). Chromatography is performed using
solvent gradient-based reversed-phase (RP) separations on C-18-
bonded (or similar) phases on columns of 5–15 cm in length and
flow rates from 200 to 900 μL/min. An example of the type of
data that can be acquired using gradient RP-UPLC-MS is shown
in Fig. 3 [19]. As this figure shows, while a large number of com-
pounds elute in the first few minutes of the analysis, by careful
optimization of the solvent gradient used metabolites can be spread
throughout the whole of the analysis time. Sample preparation for
liquid samples such as urine can often be limited to dilution and
centrifugation to remove particulates. For protein-containing sam-
ples such as serum or plasma, it is first necessary to remove the
proteins as these would otherwise irretrievably damage the col-
umn, but this is usually easily achieved via protein precipitation,
typically via the addition of organic solvents. These reversed-phase
separations have been shown to be suitable for medium polar to
nonpolar analytes but are less good for polar ionic species, and for
these types of analytes, the use of an alternative such as hydropho-
bic interaction liquid chromatography (HILIC) provides a partial
Fig. 3 Representation of a 3D mass chromatogram obtained from the reversed-

phase analysis of rat urine on UPLC-TOF-MS. Reprinted from reference [19] with
permission
solution being suitable for some, but not all, such compounds [19,
20]. For polar/ionic compounds that are unsuitable for analysis by
HILIC-MS, the only remaining option may be to use ion pair (IP)
LC where a suitable charged molecule, e.g., tributyl ammonium, is
added to the mobile phase as an oppositely charged counter ion to
“pair” with the oppositely charged analytes [21]. But, while effec-
tive, the use of IPLC generally requires the (effectively) permanent
dedication of the system to this mode of operation thereafter, as
decontaminating the instrumentation to remove all traces of the
IP-reagent can be challenging. An alternative to LC-based meth-
ods for polar ionic compounds is, of course, to employ a capillary
electrophoresis for the separation, and CE-MS methods have
indeed shown utility in this role [22].
The upshot of all of this is that, in order to obtain the most
comprehensive metabolite profile of a sample set possible, it may
require more than one chromatographic system and analysis in
both positive and negative modes of ionization (generally electro-
spray ionization (ESI), but also possibly APCI).
Having developed a suitable separation, a number of chal-
lenges remain in order to ensure that the analytical data that are
obtained are useful. Unlike NMR spectroscopy-based methods,
which are generally very robust, those utilizing a hyphenated MS
have a number of challenges that need to be addressed. These
result from the tendency of the analytical system (column and
detector) to become modified over the course of the analysis. This
can lead to minor changes in retention time, sensitivity, and (less
often) mass accuracy. The existence of such effects requires the use
of careful quality control procedures that can be used to monitor
the analysis and correct for analytical drift of whatever sort. Various
methods have been proposed for ensuring the validity of the data,
of which one of the most common is the use of so-called quality
control or QC samples. These are most often generated by making
a representative bulk pool sample from aliquots of the samples to
be analyzed. Typically it is first necessary to equilibrate the LC
system by the repeated injection of a number of QC samples, which
results in stable retention times. After this the QC sample is injected
at regular intervals throughout the sample analysis [23]. After the
run is completed, the data from these QC samples can be analyzed
using multivariate statistical procedures such as principal compo-
nent analysis (PCA), which provides a powerful tool to reveal
trends in data that would indicate time-related (or other) effects
that compromise the outcome. Assuming that the data passes such
preliminary scrutiny, further measures to optimize it, e.g., peak
alignment, can be performed, and the data can be examined for the
presence of potential biomarkers. This part of the process relies
heavily on the correct choice and the correct function of software
tools. Some software still operate as black boxes, not providing
much information or offering much freedom in the selection of

parameters. This kind of software also performs different levels of
multivariate statistics, such as PCA, and offers options for advanced
visualization plots. Recently the applicability of open-source soft-
ware (including web-based data treatment servers) has increased to
a great extent. Such tools necessitate basic-level knowledge of pro-
gramming and use of software language in R or MATLAB environ-
ment, but they offer unparalleled freedom in optimizing and
tailoring the data treatment and data scrutiny process. They also
offer advanced control in the visualization of the findings and the
generation of plots, tables, and illustrations. Such tools can provide
impressive outputs; however, statistical analysis is not yet proof and
the famous Benjamin Disraeli quote is still timely [24]. Hence
researchers are advised to pay exceptional attention to verify the
validity of their findings by the use of different statistical analysis
tools. Researchers who are new to the field are advised to resort to
the assistance of fellow researchers who are more experienced to
the specific topic of metabolome statistical analysis. Indeed the
concept and the needs are different from those applied in statistics
for genomics. Hence the statistical analysis tools to be selected
and/or their fine-tuning for effective data treatment in metabolo-
mics may vary to a great extent from the treatment of genomic
data.
Having found peaks that appear to be correlated with the con-
dition being studied, the next task is identification. This is often by
no means trivial. For techniques such as NMR spectroscopy and
GC-MS, there are large databases of spectral information that can
be interrogated that may provide clues or even positive identifica-
tions of the metabolites of interest. In the case of LC-MS, the cur-
rent situation is less promising as reliable databases for ESI-based
spectra are still under development. Even when it seems possible to
identify target metabolites from such databases, it can often be dif-
ficult to obtain authentic standards to confirm these tentative iden-
tifications. In such circumstances the complete characterization of
the unknowns using a range of MS techniques is required, and
modern instruments can enable high-resolution MS data to be
obtained, including scan analyses such as MS-MS/MSn experi-
ments in data-dependent acquisition mode, MS/MS with different
collision energies (e.g., MSE), and so forth. By combining differ-
ent levels of data, e.g., accurate mass with isotope ratio measure-
ment from both precursor and product ions to obtain molecular
formulae, and, e.g., comparison of the predicted fragmentation
patterns for the tentatively identified metabolite with those
obtained by experiment, it may be possible to increase confidence
in the putative identification of unknowns. In the end, however, it
may be necessary to isolate and identify unknowns using comple-
mentary methods, such as NMR spectroscopy, or alternatively syn-

thesize standards when feasible.
Having identified potential biomarkers, there are a number of
further considerations that require addressing. The first of these is
biological context and plausibility—how likely is it that these mol-
ecules are involved mechanistically in the phenomenon under
investigation? Directly? Or are they merely changes as part of a
“global” response by the system as it tries to maintain homeostasis,
etc. Indeed biochemical pathway analysis remains another key
challenge in the development and maturation of omics-based bio-
marker discovery. The last few years have seen increasing efforts
being invested in the development and application of software for
pathway analysis. These software, either commercial- or Internet-
based free web servers, accept lists of compound names or tables
with metabolite concentrations, and their outcome lists indicate
the most affected biochemical pathways. Such tools provide differ-
ent levels of sophistication of their visualization tools. The overall
aim is to utilize well-documented meta-analysis tools and data
from public databases such as the Human Metabolome Database
(HMDB) and/or KEGG (Kyoto Encyclopedia of Genes and
Genomes) in order to offer tools able to systematically highlight
the metabolites (and their corresponding pathways) with the most
important perturbations in the studied data set [25].
Having obtained some level of confidence that there is a direct
link to the system under investigation, there is a need to accurately
quantify these putative biomarkers (and possibly related molecules
in the same pathway) using targeted, validated assays. The use of
such methods to reanalyze the sample set can confirm not only that
the targeted molecules have indeed responded in the way seen in
the untargeted assay but can also provide quantitative concentra-
tion data, rather than mere “fold change” results. Having achieved
this level, the next step should be to analyze samples from other
studies to confirm that the findings of the initial investigation are
indeed valid.
This whole path is not an easy endeavor; however, it is in our
view a necessary process that the research community needs to
undertake to promote scientific knowledge in various aspects of
the human activity, from farming and agriculture to environmen-
tal, medical, and the life sciences. This is because the metabo-
lome directly reflects the current status of the biological system
under investigation. Another reason is the fact that changes in the
metabolome are the final result of the gene and protein function
and as such are expressed (multiplied) to a large scale in relation to,
e.g., the single nucleotide polymorphisms which caused their per-
turbation. Yet another reason is that (e.g.) in certain food products
such as olive oil, wine, or honey, there is hardly anything else to
look for: the content of such a sample is massively dominated by
small metabolites. For these and many other reasons, we believe

that the field of metabolic profiling will continue to increase. When
the technological issues are finally overcome, we could expect
stronger growth, rapid maturity, and larger application in routine
analysis and operation. In the meantime additional research efforts
and investments are necessary.
References
1. Nicholson JK, Lindon JC, Holmes E (1999) and its less toxic meta-isomer. Arch Toxicol
“Metabonomics”: understanding the meta- 90:3073–3085. https://doi.org/10.1007/
bolic responses of living systems to patho- s00204-015-1655-x
physiological stimuli via multivariate statistical 13. Han J, Danell RM, Patel JR et al (2008)
analysis of biological NMR spectroscopic data. Towards high-throughput metabolomics using
Xenobiotica 29:1181–1189 ultrahigh-field Fourier transform ion cyclotron
2. Fiehn O, Kopka J, Dörmann P et al (2000) resonance mass spectrometry. Metabolomics
Metabolite profiling for plant functional 4:128–140
genomics. Nat Biotechnol 18:1157–1161 14. Theodoridis G, Gika HG, Wilson ID (2011)
3. Gavaghan CL, Holmes E, Lenz E et al (2000) Mass spectrometry-based holistic analytical
An NMR-based metabonomic approach to approaches for metabolite profiling in systems
investigate the biochemical consequences of biology studies. Mass Spectrom Rev 30:884–
genetic strain differences: application to the 906. https://doi.org/10.1002/mas.20306
C57BL10J and Alpk:ApfCD mouse. FEBS 15. Kopka J (2006) Current challenges and devel-
Lett 484:169–174 opments in GC–MS based metabolite profiling
4. Dent CE (1952) Lectures on the scientific basis technology. J Biotechnol 124:312–322
of medicine, vol 2. Athlone Press, London 16. Moros G, Chatziioannou AC, Gika HG et al
5. Dalgliesh CE (1956) Two-dimensional paper (2017) Investigation of the derivatization con-
chromatography of urinary indoles and related ditions for GC–MS metabolomics of biological
substances. Biochem J 64:481–485 samples. Bioanalysis 9:53–65
6. Teranishi R, Mon TR, Robinson AB et al 17. Gika HG, Theodoridis GA, Plumb RS et al
(1972) Gas chromatography of volatiles from (2014) Current practice of liquid chroma-
breath and urine. Anal Chem 44:18–20 tography–mass spectrometry in metabolomics
7. Pauling L, Robinson AB, Teranishi R et al and metabonomics. J Pharm Biomed Anal
(1971) Quantitative analysis of urine vapor and 87:12–25
breath by gas-liquid partition chromatography. 18. Rainville PD, Theodoridis G, Plumb RS et al
Proc Natl Acad Sci U S A 68:2374–2376 (2014) Advances in liquid chromatogra-
8. Robinson AB, Pauling L (1974) Techniques phy coupled to mass spectrometry for meta-
of orthomolecular diagnosis. Clin Chem bolic phenotyping. TrAC Trends Anal Chem
20:961–965 61:181–191. https://doi.org/10.1016/j.
9. Scott CD, Chilcote DD, Lee NE (1972) trac.2014.06.005
Coupled anion and cation-exchange chroma- 19. Theodoridis GA, Gika HG, Want EJ et al
tography of complex biochemical mixtures. (2012) Liquid chromatography–mass spec-
Anal Chem 44:85–89 trometry based global metabolite profiling: a
10. Scott CD, Chilcote DD, Katz S et al (1973) review. Anal Chim Acta 711:7–16
Advances in the application of high resolution 20. Gika HG, Wilson ID, Theodoridis GA (2014)
liquid chromatography to the separation of LC–MS-based holistic metabolic profiling.
complex biological mixtures. J Chromatogr Sci Problems, limitations, advantages, and future
11:96–100 perspectives. J Chromatogr B Analyt Technol
11. Lenz EM, Wilson ID (2007) Analytical strat- Biomed Life Sci 966:1–6
egies in metabonomics J. Proteome Res 21. Michopoulos F, Whalley N, Theodoridis G
6:443–458 et al (2014) Targeted profiling of polar intra-
12. Kyriakides M, Maitre L, Stamper BD et al cellular metabolites using ion-pair-high perfor-
(2016) Comparative metabonomic analysis mance liquid chromatography and-ultra high
of hepatotoxicity induced by acetaminophen performance liquid chromatography coupled
to tandem mass spectrometry: applications to
serum, urine and tissue extracts. J Chromatogr an HPLC−MS-based method for metabo-
A 1349:60–68 nomic analysis: application to human urine.
22. Ramautar R, Nevedomskaya E, Mayboroda J Proteome Res 6:3291–3303
OA et al (2011) Metabolic profiling of 24. h t t p s : / / e n . w i k i p e d i a . o r g / w i k i /
human urine by CE-MS using a positively Lies,_damned_lies,_and_statistics
charged capillary coating and comparison with 25. Xia J, Wishart DS (2010) MetPA: a web-based
UPLC-MS. Mol BioSyst 7:194–199. https:// metabolomics tool for pathway analysis and
doi.org/10.1039/c0mb00032a visualization. Bioinformatics 26:2342–2344.
23. Gika HG, Theodoridis GA, Wingate JE https://doi.org/10.1093/bioinformatics/
et al (2007) Within-day reproducibility of btq418
Chapter 2
Quality Control and Validation Issues in LC-MS

Metabolomics
Olga Begou, Helen G. Gika, Georgios A. Theodoridis, and Ian D. Wilson
Abstract
Global metabolic profiling (untargeted metabolomics) of different and complex biological matrices aims
to implement an holistic, hypothesis-free analysis of (potentially) all the metabolites present in the analyzed
sample. However, such an approach, although it has been the focus of great interest over the past few
years, still faces many limitations and challenges, particularly with regard to the validation and the quality
of the obtained results. The present protocol describes a quality control (QC) procedure for monitoring
the precision of the analytical process involving untargeted metabolic phenotyping of urine and plasma/
serum. The described/suggested methodology can be applied to different biological matrices, such as
biological biofluids, cell, and tissue extracts.
Key words Quality control, Untargeted metabolomics, Biological samples
1 Introduction
Metabolomics or metabonomics, two terms interwoven with each

other, represent an expanding research discipline, dealing with the
holistic analysis of metabolites (small molecules with molecular
weight typically lower than 1500 Da). Basically, holistic profiling
aims to provide a snapshot of the metabolic phenotype (metabo-
type) and to monitor changes of the endogenous profile of living
systems in response to biological stimuli or genetic manipulation
[1].
The field of metabolomics exhibited significant development
in the last decade, especially due to the advancement of new tech-
nology platforms such as mass spectrometry (MS) and NMR
(mainly H1 NMR) spectroscopy [2]. In combination with each
other, as well as with other technologies, e.g., liquid and gas chro-
matography (LC and GC) and advanced, sophisticated, multivari-
ate statistical tools, these analytical techniques allow the
simultaneous measurement of hundreds of endogenous com-
pounds in different matrices, such as blood, urine, cells and various
15
16 Olga Begou et al.
tissues [3, 4]. However, the comprehensive analysis, simultaneous

monitoring and relative quantification of numerous compounds
face challenges due to the different physicochemical properties,
chemical classes, and concentration ranges of the metabolites pres-
ent and the complexity of the biological matrix [5–8].
Metabolomic studies can follow two distinct paths as either
untargeted or targeted approaches. Targeted methods (including
semi-targeted methods) are often hypothesis-driven analyses and as
such can be focused on the measurement of specific metabolites or
specific metabolic pathways (as described in more detail in Chapters
5 and 6 of the present book). In targeted analyses, Quality Control
can be applied, with method evaluation criteria and the analytical
figures of merit such as repeatability, analytical accuracy, and preci-
sion (for a review on targeted metabolomics, see ref. 9). Validation
of targeted methods, although not trivial, is easier to perform,
compared to untargeted methods, because predefined metabolites
are analyzed and standards/reference solutions can be used. In
contrast, untargeted metabolomics, by its very nature, focuses on
the unbiased analysis of the sample of interest with the aim of the
discovery of uncharacterized/unexpected biomarkers [10]. In
such a case, validation and quality assurance of the analysis is much
harder, especially when mass spectrometry is used coupled to a
chromatographic technique, where factors such as retention time
and MS-detection sensitivity may change during the course of a
run [4, 11]. Hyphenated techniques, such as LC and GC-MS,
inevitably generate high-dimensional data requiring extensive and
complex data processing [12]. As a result, and despite its wide
applicability, only a small number of articles have reported valida-
tion strategies for global metabolic profiling (e.g., see ref. 13).
A tested partial solution to this problem is the use of “quality
control” (QC) samples whose role is to evaluate the stability and
precision of the analysis [2, 14–16]. The use of QC samples pro-
vides information about the analytical performance of the overall
system and also strengthens the analyst’s confidence in the quality
of the acquired data. Usually, QC samples are used for analytical
platform equilibration, for monitoring the analytical signal allow-
ing for intra-/inter-day precision evaluation, for signal correction
(normalization), and method standardization [2, 15, 16]. QC data
can be used as quantitative indicators of random errors or fluctua-
tions during the analytical run. As a result, analysis of QC samples
in metabolomic studies has a greater value than just the evaluation
of chromatographic and mass spectrometry performance.
The idea for the QC samples is to prepare them by mixing
small and equal aliquots from the real samples of interest so that
the resulting pool sample (QC) will contain a mean concentration
of all the metabolites present in the real samples. Such an approach
may function for small- to medium-scale studies of up to ca 200
Quality Control and Validation Issues in LC-MS Metabolomics 17
samples. For larger-scale studies, such schemes may not be possi-

ble; for practical reasons, a bulk sample of the same matrix may be
substituted [17, 18], while sometimes both a bulk matrix and the
study samples are used to provide a double QC, with the former
used to enable analyses to be compared over long time periods/
different batches [19]. The utility of this concept, or variations of
it, has gained recognition over the past few years and has been
widely applied for the analysis of endogenous compounds in a vari-
ety of different matrices, such as urine, plasma, serum, cells, and
tissues [11, 15, 16, 20–22].
In the present protocol, we describe the use of QC samples
during the analytical procedure involved in the reversed-phase
LC-MS-based global metabolic profiling of biological samples.
With adaption, similar procedures can be employed for other types
of chromatography.
2 Materials
All solvents (methanol, acetonitrile, formic acid) used should be of

LC-MS analytical grade. Water should be of Millipore quality
(18.0 MΩ, at 25 °C). Standards are of analytical or higher grade.
All standards should be stored at −20 °C or −80 °C (see Note 1).
2.1 Stock, Working,

and Calibration Stock solutions of all metabolites should be prepared at a concen-
Solutions tration of 1 mg/mL (or appropriate concentration depending on
analyte solubility) in methanol or mixtures of methanol with water.
Working standards at a concentration of 10 μg/mL are prepared
from stock solution by dilution with ultrapure water. From the
stock solutions of the standards, prepare a mixture (concentration
1–5 μg/mL) of the compounds of interest by mixing appropriate
volumes.
All solutions should be stored at −20 °C.
2.2 Mobile Phase –– Mobile Phase A: Water + 0.1 vol.% formic acid: add 1 mL of
formic acid to 1 L of Millipore water.
–– Mobile Phase B: Acetonitrile + 0.1 vol.% formic acid: add
100 μL of formic acid to 1 L of LC-MS grade acetonitrile. In
the case of serum samples use methanol instead of acetonitrile
or mixture of methanol with acetonitrile as mobile phase B.
–– Wash Solvent: Use a “strong” solvent acetonitrile/water
80:20 v/v for post-injection cleaning cycles and a weak solvent
water/acetonitrile 80:20 v/v for pre-injection washes (see
Note 2).
2.3 Chromato- Chromatographic analysis can be performed on a HSS T3 C18

graphic Materials column (Waters, 2.1 mm × 150 mm, <2 μm particle size) or simi-
and Instrumentation lar. Pre-columns filters and guard columns of the same material as
the analytical column should be also used.
An ultrahigh performance liquid chromatography (U(H)PLC)
system coupled to a high-resolution mass spectrometer with an
ESI source can be used.
2.4 Software
Appropriate software include but are not limited to the following:
Data acquisition and processing software (e.g., Excalibur,
MassLYnx, Analyst, or other). Special software for peak picking
such as Marker Lynx, Sieve, MarkerView (or other vendor soft-
ware), or open-source/free software (XCMS, MzMine, MetAlign
or other). Microsoft Excel, Statistica, and other advanced spread-
sheet software. SIMCA-P or other software for multivariate statis-
tical analysis. MATLAB software and programming language or
associated software packages: The programming language R is a
popular and easy solution for data analysis, statistical computing,
and graphics support.
3 Methods
3.1 Analytical 1. Before starting, ensure that the mass spectrometer is in a suit-
System Preparation able condition for the analysis of the samples. Check the mass
accuracy of the mass spectrometer; if necessary, calibrate the
mass spectrometer following the appropriate procedure recom-
mended by the vendor to achieve maximum mass accuracy and
resolution. It is advisable that the calibration procedure should
take place every 3–4 months and that the temperature of the
laboratory should not vary significantly.
2. Load the required liquid chromatographic method by setting
parameters such as flow rate, column and autosampler tempera-
ture, wash cycles, and gradient elution program.
3. Allow the system to equilibrate and run a no-injection gradient
in order to aid column equilibration. Carefully examine the
results for this blank run for evidence of system/solvent con-
tamination etc., and decontaminate if necessary.
4. Depending on the matrix of the samples being analyzed, run a
suitable number of QC samples (prepared as described below)
to achieve system stability. It has been observed that for urine
analysis, 5 QC replicates are needed and for serum or plasma
10–20 injections (see Notes 3 and 4).
3.2 Sample Handling 1. It is advisable to always divide all samples on collection into an
appropriate number of sub-aliquots for later storage and also in
order to avoid unnecessary freeze/thaw cycles.
2. If possible, store sub-aliquots in different freezers, in case a
freezer malfunction takes place. Ideally, samples should be
stored in freezers as soon as possible after sampling/collection
and at the lowest available temperature, at least at −20 °C but
preferably at −80 °C.
3. Make sure all samples, stock, and working solutions are cor-
rectly and fully labeled. Thaw stock/working solutions and real
samples shortly before use and sample preparation,
respectively.
3.3 Sample 1. Vigorously vortex every sample, after thawing at room tem-
Preparation perature, prior to sample preparation.
2. Transfer an appropriate volume (e.g., 50–200 μL) of each sam-
ple into 1.5 mL Eppendorf tubes.
3. Depending on the matrix (urine, plasma, serum), a different
sample preparation procedure is recommended (see Notes 5
and 6) as explained below in Subheadings 3.3.1–3.3.3.
3.3.1 For Urine Samples
1. Dilute the sample in ratio 1:3 v/v with Millipore quality water.
2. Vortex for 5 min and centrifuge at 12,000 × g or higher speeds
for 5–10 min at 4 °C.
3. Transfer the clear supernatant into an LC-MS vial (n.b. 96 well
plates or similar can be substituted for LC-MS vials).
4. Place the vial (well plate) into a pre-cooled (4 °C) autosampler
for analysis. If analysis is not being undertaken immediately,
store the vial (well plate) in a fridge (0–4 °C) if analysis will start
shortly or a freezer (−20/80 °C) if analysis will be delayed for
longer than a few hours.
3.3.2 For Plasma or
Serum Samples 1. Add three times the sample volume of ice cold methanol or
acetonitrile for protein precipitation (see Note 7).
2. Vortex for 5 min and centrifuge at 12,000 × g or higher speeds
for 5–10 min.
3. Transfer the clear supernatant into Eppendorf tubes.
4. Evaporate to dryness and reconstitute with water/acetonitrile
95:5 v/v to the initial volume.
5. Repeat step 2.
6. Transfer the clear supernatant into an LC-MS vial (well plate).
7. Place the vial (well plate) into a pre-cooled (4 °C) autosampler
for analysis. If analysis is not taking place immediately, follow
the procedure described for urine in Subheading 3.3.1, step 4.

When taking plasma/serum samples out of the freezer, it is
advisable to centrifuge samples again before analysis.
3.3.3 For Quality Control
Samples (QC) 1. Create a pool sample by mixing equal volumes (e.g. 20 μl) of
each sample analyzed. Make an appropriate number of aliquots
(a minimum of one QC for each ten samples to be analyzed and
possibly up to one for every five; see below), and store them at
a freezer.
2. Handle each QC as a real sample and follow the sample prepa-
ration procedure developed for the type of matrix to be
analyzed.
3. If the number of the samples being analyzed is large (e.g.,
>1000) and the creation of a pooled sample from all the sam-
ples is impractical, consider making QC sample from only the
samples contained within each analytical run being analyzed to
provide a within-batch QC and have a separate “bulk matrix”
QC sample also analyzed within every batch to enable between
batch comparisons and normalization to be conducted
[17–19].
3.3.4 For Test
Mix Sample 1. Spike a QC sample with an appropriate volume of the methano-
lic mixture mentioned in Subheading 2.1 above, in order to
achieve a final concentration of 5–10 μg/mL. It is advisable to
perform this procedure by evaporating the methanolic stock
solution and then reconstituting by dissolution in the QC
matrix.
3.4 Analytical 1. Make sure that the order of the samples is randomized, to avoid
Sequence Preparation introducing bias due to changes with individual analytical runs
and between run “batch” effects. Randomization of large sam-
ple sets can be performed using the specific commands in
spreadsheet programs such as Excel. QC samples should not be
randomized but inserted regularly in the run sequence (e.g.,
one QC sample every five to ten real samples).
2. The number of QC samples placed in a sequence depends on
the total number of samples analyzed and on the duration of
the analysis. For a small number of samples (<100), one QC
every five sample injections is recommended, and for a larger
number QC, samples should represent at least 10% of the total
analyzed.
3. After column equilibration, in the middle of the sequence and
at the end, insert the spiked test mix solution. Standard solution
injections should be avoided within the batch of test samples, as
the system can be disequilibrated by the injection of non-matrix
solutions (see Notes 3 and 8). Similarly blank sample injections

should be avoided.
4. A good way to improve quantification aspects in such analysis is
to implement a sequence of injections of serial dilutions of the
QC sample. This enables observations on saturating peaks,
which would be better detected in dilute samples (e.g.,
1:10 v/v). At the same time, this series of analyses facilitates a
study of the response of non-saturating peaks in response to
dilution.
In untargeted metabolomic studies, data analysis strategies

require some steps involving raw data acquisition, normalization,
scaling, and feature and peak detection in order to finally reach
biomarker detection and identification.
3.5 Data Analysis

From the raw data acquired, observe the peak width, retention
time, peak intensity/signal, mass accuracy, and noise intensity from
3.5.1 Data Processing indicative chromatographic peaks located at the beginning, mid-
dle, and end of the chromatogram. These can help researchers to
define carefully the optimum parameters for the software (XCMS,
MassLynx, etc.) to be used for peak alignment, peak peaking, and
integration.
Data from the QC samples are used to evaluate the quality of
analysis via the following steps: Treat QC samples as a separate
group and process them alone using exactly the same parameters
selected for processing the whole sample set. Peaks that appear in
less than 70% or 80% of the QC samples should be omitted from
the data set [14]. Apply principal component analysis (PCA), using
SIMCA-P or other statistical software, on the processed data from
the QC samples in order to observe any trends, such as time-related
drift during the run, on these specific samples.
More details on data pre-processing can be found in Chapters
3 and 4 of this book.
3.5.2 Data Quality
Evaluation For data quality evaluation, there are some steps that should be
taken into consideration. It is advisable to subject the data to these
tests against preset criteria of quality as explained below.
Firstly, assess the performance of the analytical run by inspect-
ing an overlay of full-scan chromatograms of all the replicates of
the test mix samples, placed at the beginning, middle, and end of
the sequence. Calculate chromatographic data, such as the reten-
tion time, peak area, and height, and determine the precision (RSD
%) of each value. The acceptable limits should be less than 2% for
retention time variation and less than 20–30% for peak area varia-
tion. In case that these conditions are not met, reexamine the data
carefully to find any trend that might indicate system underperfor-

mance or malfunction.
The next step is to assess the data acquired from QC samples
regularly placed among the samples, in order to strengthen and
corroborate the initial assessment of the performance of the ana-
lytical run. Evaluate the full-scan chromatograms of all QC sam-
ples. It is expected that all QC samples show good consistency,
reproducibility, and small variability, with the possible exception of
the first five to ten injections (conditioning QCs). An overlay of
full-scan chromatograms can provide only an estimation of analyti-
cal repeatability and system stability. A thorough check should be
done on extracted ion chromatograms of various peaks distributed
along the whole length of the analysis. Examined peaks should be
spread in both retention time and detected mass. The precision of
the retention time, the mass detected, and the signal intensity of
different peaks located across the chromatogram should be per-
formed with preset criteria (e.g., RSD% should be less than 2% for
retention times, less than 20% for the area of abundant peaks and
less than 30% for low intensity peaks) as mentioned above. Peaks
that fail such criteria should be excluded from the data set (includ-
ing data from the test samples). Evidence of poor repeatability for
the QC samples should be taken as a possible reason for not accept-
ing a run.
If there are any problems concerning the reproducibility of the
data, try to estimate where the inconsistency is appearing. If it is
located in a specific time window, investigate if it is worth exclud-
ing this window in the peak picking and alignment process. If the
data is poor and fails the acceptance criteria, investigate the reason
for the failure and repeat the analysis.
Providing that the run passes the predetermined acceptance
criteria, perform principal component analysis (PCA) on all sam-
ples and QC replicates, in order to observe variations in the
acquired data. PCA will immediately show similarities or differ-
ences among real and QC samples. The latter should ideally form
a tight cluster (assuming that there are differences between the
samples). For untargeted data, investigate the effect of different
scaling modes. Applying Pareto scaling and different visualization
tests helps in identifying potential underlying issues. The following
steps are suggested:
1. Color QC and real samples according to analytical run order
and try to identify time trends in the score plots.
2. Check for any QC outliers or for batch effect due to run order
3. Examine the QC data for any unexpected trends. If such trends
are found, study additional visualization plots such as time series
plots, and look for samples that are out of 2SD or 3SD thresh-
olds. Test samples appearing as such “outliers” should not be
discarded outright but rather be thoroughly scrutinized. QC

samples out of 2SD indicate that the data from the neighboring
test samples should be thoroughly scrutinized and probably dis-
carded with the samples themselves being reanalyzed.
Thereinafter, in the exported peak table data, look for any sim-
ilarities in the pairs (retention time and mass) among the peaks
reported, in order to assess the success of peak alignment. Then,
using appropriate statistical software, create graphs with the data
exported from the QC samples to evaluate the analytical perfor-
mance and precision.
Figure 1 presents a schematic of the proposed validation
scheme. It starts with the QC preparation, by mixing equal vol-
umes of all the real samples (urine, plasma/serum, etc.) followed
by determing the run order sequence and then injecting the QC
sample periodically, e.g., every five to ten real samples during anal-
ysis. From the resulting data set, an extracted ion chromatogram
(XIC) of all QC samples can be created, via data processing or data
mining, where data is collected only for specific m/z values of the
compounds of interest. Those results can also be used for the for-
mation of an analysis chart, showing total sequence order (both
QC and real samples). Data processing is complemented by statis-
tical analysis. Repeatability and run order trends can be assessed by
making QC quality control charts and data tables, evaluating %RSD
values of QC’s ion intensities. Features presented in most samples
(~70%) with %RSD values <30% indicate a good data set. Finally,
PCA score plots can be created from the examined results, provid-
ing information on the accuracy and precision of the analysis, as
well as for any time trends related to the order of analysis of both
QC and test samples.
4 Notes
1. The storage period for each standard in the freezer is dependent

on the nature of the analyte and the solvent used for dissolu-
tion. Certain standards dissolved in water are stable for up to
20 days, or longer, at −20 °C.
2. Make sure mobile phases are filtered and sonicated to degas
them properly.
3. Avoid making injections with standard solutions during the
sequence so the system isn’t deconditioned.
4. Always monitor and record the exact column pressure (bar or
psi) at the beginning and end of a run. Rising pressure can be
an indication that a column is becoming blocked and so provides
an indication of the need for either replacement or cleaning.
Fig. 1 Schematic of a validation scheme from QC and sequence preparation (top), to data processing (middle)
to statistical analysis (bottom). From top left: QC samples are prepared by initially mixing equal volume of all
the real samples. Data processing includes scrutiny of data as extracted ion chromatogram (XIC) and data
processing or data mining. Trends in data should be checked across the run order. Repeatability and run order
trends can be assessed by making QC quality control charts and data tables, evaluating %RSD values of QC’s
ion intensities. PCA score plots should exhibit tight cloud of QC samples (red in the presented plot) within the
whole sample set (black dots). In the bottom of the figure, a control chart (left) shows features showing poor
repeatability in red (CV >30% in QC samples) and repeatable features in green (CV <30% in QC samples).
These charts are providing information for the precision of the analysis, as well as for any time trends related
to the order analysis of both QC and real samples (bottom right)
5. Apply proper cleanup steps during sample preparation. Filtering

samples improves the medium-/long-term integrity of the sep-
aration system and as a result may safeguard high-quality results
if necessary.
6. During sample preparation, all necessary safety precautions

must be taken. Make sure gloves and goggles are used and that
samples are prepared under fume hoods.
7. It has been noticed that using ice cold solvents (methanol, ace-
tonitrile) at three times the sample volume helps to provide bet-
ter protein precipitation than solvents at ambient temperature.
8. In the case of large numbers of samples being analyzed, divide
them into randomized batches. Carefully monitor the system
performance during each batch and have appropriate cleaning
cycles in between (column, cone, source, etc.). Before starting
a new sequence, make sure that the system is re-equilibrated.
References
1. Nicholson JK, Lindon JC, Holmes E (1999) of day-to-day reproducibility and robustness of
“Metabonomics”: understanding the meta- LC-MS methods for global metabolite profiling
bolic responses of living systems to patho- in metabonomics/metabolomics. Bioanalysis
physiological stimuli via multivariate statistical 4:2239–2247. https://doi.org/10.4155/
analysis of biological NMR spectroscopic data. bio.12.212
Xenobiotica 29:1181–1189. https://doi. 9. Begou O, Gika HG, Wilson ID et al (2017)
org/10.1080/004982599238047 Hyphenated MS-Based Targeted approaches in
2. Dunn WB, Wilson ID, Nicholls AW, metabolomics. Analyst 142(17):3079–3100
Broadhurst D (2012) The importance of 10. Theodoridis G, Gika HG, Wilson ID (2011)
experimental design and QC samples in large- Mass spectrometry-based holistic analytical
scale and MS-driven untargeted metabolomic approaches for metabolite profiling in sys-
studies of humans. Bioanalysis 4:2249–2264. tems biology studies. Mass Spectrom Rev
https://doi.org/10.4155/bio.12.204 30:884–906. https://doi.org/10.1002/
3. Griffiths WJ, Koal T, Wang Y et al (2010) mas.20306
Targeted metabolomics for biomarker dis- 11. Gika HG, Zisi C, Theodoridis G et al (2016)
covery. Angew Chem Int Ed 49:5426–5445. Protocol for quality control in metabolic pro-
https://doi.org/10.1002/anie.200905579 filing of biological fluids by U(H)PLC-MS. J
4. Wilson I (2016) Methods and techniques for Chromatogr B Analyt Technol Biomed Life
metabolic phenotyping. Bioanalysis 9:1–3. Sci 1008:15–25. https://doi.org/10.1016/j.
https://doi.org/10.4155/bio-2016-4985 jchromb.2015.10.045
5. Gika HG, Wilson ID, Theodoridis GA (2014) 12. Gorrochategui E, Jaumot J, Lacorte S et al
LC–MS-based holistic metabolic profiling. (2016) Data analysis strategies for targeted and
Problems, limitations, advantages, and future untargeted LC-MS metabolomic studies: over-
perspectives. J Chromatogr B 966:1–6. https:// view and workflow. TrAC Trends Anal Chem
doi.org/10.1016/j.jchromb.2014.01.054 82:425–442. https://doi.org/10.1016/j.
6. Gika HG, Theodoridis GA, Earll M et al trac.2016.07.004
(2010) Does the mass spectrometer define 13. Naz S, Vallejo M, García A et al (2014)
the marker? A comparison of global metabo- Method validation strategies involved in non-
lite profiling data generated simultaneously via targeted metabolomics. J Chromatogr A
UPLC-MS on two different mass spectrom- 1353:99–105. https://doi.org/10.1016/j.
eters. Anal Chem 82:8226–8234. https://doi. chroma.2014.04.071
org/10.1021/ac1016612 14. Bijlsma S, Bobeldijk I, Verheij ER et al (2006)
7. Rainville PD, Theodoridis G, Plumb RS, Wilson Large-scale human metabolomics studies: a
ID (2014) Advances in liquid chromatogra- strategy for data (pre-) processing and valida-
phy coupled to mass spectrometry for meta- tion. Anal Chem 78:567–574. https://doi.
bolic phenotyping. TrAC Trends Anal Chem org/10.1021/ac051495j
61:181–191. https://doi.org/10.1016/j. 15. Want EJ, Wilson ID, Gika H et al (2010)
trac.2014.06.005 Global metabolic profiling procedures for urine
8. Gika HG, Theodoridis GA, Earll M, Wilson ID using UPLC-MS. Nat Protoc 5:1005–1018.
(2012) A QC approach to the determination https://doi.org/10.1038/nprot.2010.50
16. Gika HG, Theodoridis GA, Wingate JE for precision large scale urinary metabolic phe-
et al (2007) Within-day reproducibility of notyping. Anal Chem 88:9004–9013. https://
an HPLC-MS-based method for metabo- doi.org/10.1021/acs.analchem.6b01481
nomic analysis: application to human urine. 20. Veselkov KA, Vingara LK, Masson P et al
J Proteome Res 6:3291–3303. https://doi. (2011) Optimized preprocessing of ultra-
org/10.1021/pr070183p performance liquid chromatography/mass
17. Zelena E, Dunn WB, Broadhurst D et al spectrometry urinary metabolic profiles for
(2009) Development of a robust and repeatable improved information recovery. Anal Chem
UPLC−MS method for the long-term metab- 83:5864–5872. https://doi.org/10.1021/
olomic study of human serum. Anal Chem ac201065j
81:1357–1364. https://doi.org/10.1021/ 21. Michopoulos F, Theodoridis G, Smith CJ,
ac8019366 Wilson ID (2010) Metabolite profiles from
18. Dunn WB, Broadhurst D, Begley P, Human dried biofluid spots for metabonomic stud-
Serum Metabolome (HUSERMET) Consortium ies using UPLC combined with oaToF-MS. J
(2011) Procedures for large-scale metabolic pro- Proteome Res 9:3328–3334. https://doi.
filing of serum and plasma using gas chromatog- org/10.1021/pr100124b
raphy and liquid chromatography coupled to mass 22. Ramautar R, Nevedomskaya E, Mayboroda
spectrometry. Nat Protoc 6:1060–1083. https:// OA et al (2011) Metabolic profiling of
doi.org/10.1038/nprot.2011.335 human urine by CE-MS using a positively
19. Lewis MR, Pearce JTM, Spagou K et al charged capillary coating and comparison with
(2016) Development and application of ultra- UPLC-MS. Mol BioSyst 7:194–199. https://
performance liquid chromatography-TOF MS doi.org/10.1039/c0mb00032a
Chapter 3
Data Treatment for LC-MS Untargeted Analysis

Samantha Riccadonna and Pietro Franceschi
Abstract
Liquid chromatography-mass spectrometry (LC-MS) untargeted experiments require complex chemo-
metrics strategies to extract information from the experimental data. Here we discuss “data preprocess-
ing”, the set of procedures performed on the raw data to produce a data matrix which will be the starting
point for the subsequent statistical analysis. Data preprocessing is a crucial step on the path to knowledge
extraction, which should be carefully controlled and optimized in order to maximize the output of any
untargeted metabolomics investigation.
Key words Preprocessing, Peak picking, Retention time correction, Metadata, Quality check, Missing
values
1 Introduction
Liquid chromatography-mass spectrometry (LC-MS) is an estab-

lished and widely used analytical technique and combines the sepa-
ration potential of LC with the ability of MS to quantify the ion
intensity [1]. From a data scientist point of view, LC-MS produces
“three-dimensional” datasets where each ion generated in the ion-
ization source is characterized by m/z value, retention time (rt)
and intensity.
While in targeted applications selected ions are used to quan-
tify the analytes of interest, untargeted investigations deal with the
full set of ionic signals trying to characterize all the detectable ana-
lytes including chemical unknowns [2]. This potential, however,
comes at a price because the knowledge extraction process is not
straightforward and heavily relies on bioinformatics [3].
A schematic representation of the data analysis workflow maxi-
mizing the process of knowledge mining in an untargeted LC-MS
investigation is presented in Fig. 1: the path involves a constant
feedback between experimental design (DoE), data processing and
statistical analysis. An extensive treatment of all the steps goes
beyond the scope of this chapter and here we will focus on the
27
28 Samantha Riccadonna and Pietro Franceschi
Question Preprocessing
Definition
Data conversion
Data generation Peak picking
(open file format)
Design of
Grouping and retention time correction
Experiment
(DoE)
Metadata
Quality check
organization
Imputation of missing values
Explorative Data
Analysis transformation
Data Modeling &

Annotation
Annota
Statistical Analysis
Result Validation
& Interpretation
Fig. 1 Data analysis workflow of a typical LC-MS untargeted experiment: the path from data to knowledge is
usually represented as linear, but it is more likely to have many feedback loops (dashed lines). Only the red
boxes are detailed in the text
s o-called data preprocessing. The objective of this phase is to sum-

marize the raw analytical data into a two-dimensional matrix with
samples on the rows and ions—usually indicated as features—on
the columns. Our discussion will also touch data organization,
because this step plays a fundamental role in guaranteeing the
interoperability and reproducibility of the whole data analysis pro-
cess [4]. Considering the high pace of evolution of bioinformatics,
a pure “how-to” guide would be outdated quite fast. Thus we try
to include also some general considerations valid in a wider variety
of contexts. In parallel, to give practical guidance to the researcher,
we illustrate how the workflow can be implemented in the statisti-
cal environment R [5], which is obviously not the only possible
solution [6–11]. We believe that a general understanding of the
key aspects of the problem can be of value beyond the specific solu-
tion chosen to perform data analysis.
Data Treatment for LC-MS Untargeted Analysis 29
2 Software
1. Metadata organization. Spreadsheets to inspect/fill tables.

The ISA metadata model to encode the information [4, 12].
The Risa Bioconductor package [13] and the ISAcreator appli-
cation [14] to fill in and read ISA-Tab files.
2. Data conversion. Vendor software allowing data export in
open file format (see Note 1) or the cross-platform ProteoWizard
toolkit [15, 16].
3. Processing (peak picking, grouping and retention time cor-
rection, quality check, imputation of missing peaks). The
xcms [17–19] Bioconductor package, which is a popular solu-
tion for processing raw chromatography-MS data.
4. Example dataset. Spiked-apple dataset available from
MetaboLights (MTBLS59) [20], downloaded and unzipped in
a directory named “MTBLS59” within the R working directory
[20]. Briefly, the dataset contains the results of the analysis of
40 apple extracts divided in four classes that were spiked with a
set of known compounds. The extracts were analysed by UPLC-
QTOF—MS in positive and negative mode.
3 Methods
3.1 Metadata The term “experiment metadata” refers to the set of information
Organization describing the samples and the analytical pipeline (see Note 2).
Organizing and storing a detailed record of the sample metadata
(see Note 3) is crucial to perform a reliable interpretation of the
results, evaluating the presence (and the impact) of possible con-
founding factors.
1. Collect and organize (possibly with the help of a spreadsheet)
all the important properties of the samples (e.g. treatment type,
day of collection, origin, etc.). Keep track also of the sample
specific analytical details.
2. Run ISAcreator to generate your ISA-Tab files (see Note 4).
The tool will support the description, through commonly
accepted “ontologies”, of the general aim of your experiment
(“Study”), the analytical platform (“Study Assay”), the DoE
(“Study Design Descriptors”, “Study Factors”), the analytical
information (“Study Protocols”) and the details about each
sample (“Sample Definitions”). Handling large datasets with
ISAcreator may not be easy, in particular if one wants to auto-
matically fill some of the required fields. In this case, an alterna-
tive choice is to use ISAcreator only to generate the backbone
of the ISA-Tab files and to fill in the sample details either by a
spreadsheet or a scripting language (e.g. R, python, etc.). In R,
ISA-Tab files can be managed directly by using the Risa pack-

age, which is based on the definition of the ISAtab-class. If the
ISA-Tab files are stored into a folder named “ISA_info” in the
R working directory, they can be read into R by typing
mydata<−readISAtab(path=“ISA_info”)
The sample information can then be modified using the func-
tion updateAssayMetadata, and, once finished, the ISA-Tab
files can be saved into a new folder “ISA_exp_description”
with the command (see Note 5):
write.ISAtab(mydata,path=“ISA_exp_description”)
3. Perform the final consistency checks in ISAcreator.
4. Create the final archive containing the raw data and their
description with ISAcreator.
3.2 Data Conversion In general, the machine software can be used to perform this task
(See Note 6) (see Note 7). At present, the open source alternative is the
ProteoWizard “msconvert” (see Note 8), which is available as a
command line tool or as a point-and-click application for Windows
(MSconvertGUI) (see Note 9).
1. Open a terminal or a command window and run msconvert
into the raw data directory. As an example, to convert vendor-
specific raw data (here .RAW data files) into mzML files and
store them into the “open_data” directory use:
msconvert *.RAW -o open_data.
Alternatively, the use of the GUI application is self-explanatory.
Also in this case, batch mode conversion of a set of files is
allowed.
3.3 Processing Each metabolite produces at least one peak in the (m/z, rt) plane. To
detect metabolites, an automatic algorithm has to be able to identify
such peaks and to distinguish them from electrical or chemical noise
[7, 21]. This process of peak identification (peak picking) has to be
performed automatically on all the samples under analysis.
Subsequently, the peak lists identified in each sample have to be
joined together in a final list of features. To do that, it is necessary to
account for m/z and rt shifts from sample to sample [22]. The mea-
sured m/z values can be different because even the most advanced
mass spectrometer is not absolutely precise. Rt shifts, instead, are
unavoidable characteristics of chromatography. Technically speak-
ing, a “retention time correction” step is implemented to compen-
sate for the shifts in rt, and, afterwards, the features found in the
different samples are “grouped” (across samples) in a final consensus
list. All the software solutions (commercial or open source) imple-
ment different strategies to perform these steps [23, 24].
3.4 Preprocessing: The characteristics that distinguish a peak from the noise are inten-
Peak Picking sity, width and shape. Different algorithms handle them in differ-
ent ways [7, 21]. Most importantly, not only the algorithm but
also the specific parameters of each algorithm greatly affect the
results (see Note 10).
1. Choose the peak-picking algorithm on the basis of the charac-
teristics of the analytical pipeline [25, 26]. In xcms several
choices are already available (see Note 11). The type of instru-
mental setting used to analyse MTBLS59 suggests using the
centWave algorithm [18].
2. Identify the critical parameters of the algorithm and possibly
link them with the characteristics of the analytical pipeline. For
centWave, for example, the most important parameters to be
considered are ppm, which is determined by the mass accuracy
and is used to determine the “reliability” of each mass trace;
prefilter, which has to be set considering the minimum intensity
(I) of a true signal (at the single spectrum level) and also in how
many consecutive scans (k) that signal should be detected;
peakwidth, which deals with the range of the acceptable chro-
matographic peak-widths (in seconds); and snthresh, which
defines a cut-off for distinguishing signal from noise (signal-to-
noise ratio).
3. Check on a representative sample the results of the peak picking
(see Note 12). For instance, to run centWave with default
parameters on a specific file, issue
test_sample <- xcmsSet(files="MTBLS59/apple_
control_neg_001.CDF", method="centWave")
To visually inspect the outcomes of this step, plotting the peaks
in the m/z, rt plane, issue
mypeaks<-peaks(test_sample)
plot(mypeaks[,c("rt","mz")])
4. Optimize the peak-picking parameters until the result is satis-
factory (see Note 13).
5. Perform the peak picking on the whole dataset with the selected
set of parameters. The final xcmsSet object (see Note 11) will
contain the full list of the peaks detected in all the samples, with
some additional information on the mass and retention time
ranges. The complete analysis (with the parameters set as sug-
gested in [26]) can be run issuing:
cdffiles <- list.files(path="MTBLS59",pattern=
”CDF”,
full.names = TRUE)
myfiles<-grep("neg",cdffiles, value=TRUE)[1:40]
xs<-xcmsSet(files=myfiles,method="centWave",
ppm = 15, prefilter = c(0,0), peakwidth =
c(5,20), snthresh=1)
3.5 Preprocessing: 1. Choose the retention time correction algorithm among the
Retention Time available options (see Note 11). In this example we use obiwarp
Correction [26], which is based on dynamic time warping, an algorithm
and Grouping (See used to find the best stretching of the time dimension of two
Note 14) time series to make them as similar as possible [27]. LC-MS
data are multidimensional, so in obiwarp the similarity is deter-
mined taking into account the full spectral information.
2. Determine the more relevant parameters of the algorithm. In
obiwarp the relevant parameters are the choice of the reference
sample for the rt warping (center); the m/z bins used for
retrieving the spectra (profStep); the spectral similarity function
(distFunc); and the penalizations on the warping optimization
(gapInit and gapExtend). With the default parameters (which
work reasonably well in most of the cases), the reference sample
is the one containing the largest number of peaks. The reten-
tion time correction can be performed with the following
command.
xsr <− retcor(xs, method=“obiwarp”)
3. Choose the “grouping” algorithm or, in other terms, the pro-
cedure which matches peaks with similar m/z and rt found in
different samples. Among the possible choices, we rely on the
default density-based solution [17].
4. Choose the grouping parameters according to the analytical
pipeline: bw defines the retention time window used for the
density estimation and mzwid accounts for small differences in
the m/z value of the corresponding peaks detected in the differ-
ent samples. To avoid the inclusion in the final list of features
peaks detected only in a small number of samples (which could
be, e.g. artefacts), it is possible to keep only groups which con-
tain peaks detected in at least minsamp samples (in at least one
sample class). Again, we set the parameters as suggested in [26]
and minsamp equal to 2:
xsr <− group(xsr, bw = 2, mzwid = 0.025,
minsamp = 2)
3.6 Preprocessing: At the end of the preprocessing, it is extremely important to check

Quality Check the outcomes of the overall process. This can be done i mplementing
a series of QC plots (see Chapter 2). Firstly, a global evaluation of
the retention time correction procedure helps in detecting
unwanted extreme deformations of the original time scale.
Secondly, compounds known to be present in the samples should
be correctly detected and aligned on the whole dataset.
1. Visually inspect the retention time correction. In xcms this can
be done using the dedicated function.
plotrt(xsr)
Retention Time Deviation vs. Retention Time
Retention Time Deviation

4
2
0
−4
−8
0 200 400 600 800
Retention Time
Fig. 2 Outcome of the retention time correction performed on the MTBLS59 data as obtained by plotrt, after
minor visual parameters customization. Each line represents one sample and is coloured by sample class. The
retention time correction is different for each sample, and it is not linear in time
The situation for the MTBLS59 data is presented in Fig. 2 which

shows the retention time deviation (in seconds) for each
sample along the chromatographic retention time. Usually,
the first and the last part of the chromatography require the
heaviest corrections. However, in the example, the greatest
correction has been applied to one of the red samples and is
of about 8 s. Considering that here we are dealing with
UPLC, the correction is acceptable indicating a satisfactory
reproducibility in the chromatography. As a rule of thumb,
the maximum correction should be comparable to the chro-
matographic peak width in that particular rt range.
2. Inspect the extracted ion traces (EICs) associated with com-
pounds known to be present in the samples. In presence of a
good preprocessing, they should look nicely aligned across the
samples. In addition, a corresponding feature (group) has to be
present in the final list of features. In the case of the apple
extract in MTBLS59, possible metabolites to check are
quercetin-3-glucoside and quercetin-3-galactoside, both show-
ing a strong ionic signal at m/z 463.087 (in negative ion mode)
[20] at around 415 s. The EICs in the vicinity of their peak can
be extracted by issuing:
mzrange<−cbind(“mzmin”=463.087–0.005,
“mzmax”=463.087+0.005)
rtrange<−cbind(“rtmin”=405,"rtmax”=430)
myeic_corrected<−getEIC(xsr,mzrange=mzrange,
rtrange=rtrange,rt=“corrected”)
plot(myeic_corrected)
The position of the corresponding feature in the groups can be
extracted with the following commands:
Extracted Ion Chromatogram: 463.08 − 463.09 m/z
1500
1000
Intensity
500
0
405 410 415 420 425 430
Retention Time (seconds)
Fig. 3 Quality check on the processing procedure using the quercetin-3-glucoside and quercetin-3-galactoside
in negative ionization, which are expected to be present. Minor visual parameters customization has been
performed compared to the base command reported in the text. Samples are coloured by sample class. The
black dashed line corresponds to the identified the peak group
mygroups<−groups(xsr)
mycompound<−mygroups[mygroups[,"rtmed”]>405
& mygroups[,"rtmed”]<430
& mygroups[,"mzmed”]>463.082
& mygroups[,"mzmed”]<463.092,"rtmed”]
abline(v=mycompound)
The resulting plot is shown in Fig. 3. It is clear that the EICs are
correctly aligned (see Note 15). In this case, however, xcms
finds only one feature associated with both compounds
because of the initial choices of the peak-picking algorithm
(see Note 16).
3.7 Preprocessing: It could happen that peaks belonging to a subset of samples are
Imputation of Missing missing from “their” groups. This is correct if the correspond-
Peaks ing analyte is not present in a specific sample, but it could also
be the result of an error in the peak-picking phase. These miss-
ing peaks produce missing values in the data matrix, which are
usually tricky to handle during statistical analysis. Several
options exist to “impute” missing values (which are common
also in other -omic technologies). The clever solution available
in xcms is implemented by the function fillPeaks, which goes
back to the raw data integrating the signal measured in the area
corresponding to the “missing” peak. If the peak was wrongly
missed, fillPeaks will recover it, otherwise that missing value
will be filled with the sum of the noise.
1. Run the imputing function issuing:

xsf <− fillPeaks(xsr)
3.8 Data Matrix Extract the final two-dimensional data matrix with samples on the
Extraction rows and features on the columns, using as intensity the integral of
the area under the peak:
mat<-t(groupval(xsf,value=“into”))
The m/z and rt details about the features are encoded in the
mygroups object generated at Subheading 3.6, step 2.
4 Notes
1. Raw data are usually stored in vendor proprietary formats that

can normally be read only by proprietary software. A close
data format hampers data comparison and sharing, which is
instead made possible adopting “open” data formats with
publicly available specifications. Examples of open formats
available for LC-MS are mzML [28], which is supported by
the proteomics standards initiative, mz5 [29], mzDB [30] and
netCDF.
2. Sample metadata often mirror the factors included in the
experimental design (DoE) (type of treatment, gender, etc.),
but they also include key characteristics of the samples. They
have to be considered during the design of the analytical work-
flow, in order to ensure proper randomization of the samples
[31–37].
3. The use of a standardized language and/or ontologies is
essential to profit as much as possible from the experiment,
allowing result comparison, integration and sharing [4, 12].
The compliance with ISA standards is also a requirement for
submitting experimental data (both raw data and metadata) to
MetaboLights, an open-access and general purpose repository
for metabolomics data [38].
4. An ISA-Tab investigation comprises a set of tab-separated files
which record all the relevant metadata about the samples and
the analytical pipeline. A step-by-step tutorial on how to use
ISAcreator can be found on the MetaboLights website
(https://www.ebi.ac.uk/metabolights/help).
5. For additional arguments and details on the functions, please
refer to the package documentation or to its vignette. (https://
www.bioconductor.org/packages/release/bioc/html/Risa.
html)
6. It is worth mentioning that the data conversion to an open
data format is an optional step. In particular, it is not manda-
tory when the analyses are performed with the vendor soft-
ware. On the contrary, it is mandatory when an open source

solution is preferred. It is important to know that not all the
relevant analytical metadata are usually included in the open
source version of the data (for instance, the analytical column
temperature). To avoid information loss, it is then necessary to
store a copy of the original “closed” files which can be
inspected only with the vendor software. This fact has to be
taken into consideration in the planning of the data storage
infrastructure.
7. Examples of data conversion can be found in [9].
8. The list of open and closed formats handled by ProteoWizard
is available on the project website (http://proteowizard.
sourceforge.net/tools.shtml).
9. To perform the conversion, the vendor-specific Windows
dynamic link libraries (DLLs) are needed for reading the raw
data files and the sample metadata. Such DLLs are distributed
either as part of the specific vendor software or within the
Windows version of the ProteoWizard suite.
10. Automatic peak detection usually leads to suboptimal solu-
tions when compared to visual inspection (e.g. it could hap-
pen that a peak is instead just noise). Nevertheless, visual
inspection will introduce subjectivity and not reproducible
results (besides the fact that it is a very time-consuming task).
The advantage of automatic systems is that they always apply
the same “approach” impartially. Thus given the “approach”
(or, more precisely, the algorithm and its parameters), the out-
come of the procedure is determined. The algorithms used to
perform all the preprocessing tasks should be flexible, but an
excess of flexibility can produce unwanted artefacts (wrongly
matched features). Therefore to get the maximum from the
data, it is necessary to match this flexibility with the real ana-
lytical characteristics of the experiment (which should be also
optimized!). Given the importance of the parameter tuning
phase, automatic optimization tools has also been developed
[39].
11. The characteristics of the different algorithms are discussed in
the xcms vignette and manual, both available from the
Bioconductor website (https://bioconductor.org/packages/
release/bioc/html/xcms.html).
12. Since the chosen values for the peak-picking parameters will be
applied to the whole dataset, it is important that the reference
is representative of the sample complexity. The more represen-
tative the sample is, the better the final outcome of the analysis
will be. If quality control (QC) samples are a pool of samples,
they are usually a good choice.
MTBLS59
0 200 400 600
default set1
2500
2000
1500
m/z
1000
500
0 200 400 600
rt
Fig. 4 Peak-picking outcome on the MTBLS59 dataset: (left) the peak detection is performed using the default
settings for centWave (right) or the set of parameters defined in Methods Subheading 3.4, step 5
13. The signature of a good peak picking is a structured organiza-

tion of points in the space defined by the retention time (rt)
on the abscissa and m/z on the ordinate. Since each chemical
produces several m/z’s signals at the same retention time, the
points should be arranged in “vertical stripes”. Dots spread
uniformly in the plane or clear horizontal patterns are the
symptoms of a problem in peak picking or in the chromatog-
raphy itself. The number of detected peaks can also be an
interesting indicator of a sensible choice of the parameters, as
shown in Fig. 4. The situation on the left panel of Fig. 4 is
obviously not satisfactory.
14. In some cases an additional grouping step can be necessary
before retention time correction.
15. Actually a minor retention time correction has been applied
here: the compound elutes within 405–430 s which corre-
spond to a correction smaller than 2 s can be noted from
Fig. 3. The EICs for the uncorrected measures can be obtained
using
myeic_raw<−getEIC(xsr,mzrange=mzrange,rtrang
e=rtrange,
rt=“raw”)
16. The choice of the peak-picking algorithm and of its parameters
obviously limits the range of compounds that can be detected.
In many cases, the samples will contain analytes eluting with
diverse chromatographic peak shapes. An automatic algorithm
will never be able to tackle all the possible analytical situations.

The optimal preprocessing finds the “best” and less critical
compromise between flexibility and reliability. The case pre-
sented in Fig. 3 is particularly critical because the two isomers
elute almost at the same time and their chromatographic peaks
are not well separated. The presence of a single feature for
these distinct metabolites would make impossible to rely on
these settings to study a biological process which affect the
balance between these two metabolites.
References
1. Ardrey RE (2003) Liquid chromatography – mics measurements. BMC Bioinformatics

mass spectrometry: an introduction. John 9:375
Wiley & Sons, Chichester, UK 12. Rocca-Serra P, Salek RM, Arita M et al (2016)
2. Patti GJ, Yanes O, Siuzdak G (2012) Data standards can boost metabolomics
Innovation: metabolomics: the apogee of the research, and if there is a will, there is a way.
omics trilogy. Nat Rev Mol Cell Biol Metabolomics 12:14
13:263–269 13. González-Beltrán A, Neumann S, Maguire E
3. Alonso A, Marsal S, Julià A (2015) Analytical et al (2014) The Risa R/bioconductor pack-
methods in untargeted metabolomics: state of age: integrative data analysis from experimental
the art in 2015. Front Bioeng Biotechnol 3:23 metadata and back again. BMC Bioinformatics
4. Sansone S-A, Rocca-Serra P, Field D et al 15(Suppl 1):S11
(2012) Toward interoperable bioscience data. 14. Rocca-Serra P, Brandizi M, Maguire E et al
Nat Genet 44:121–126 (2010) ISA software suite: supporting
5. R Core Team (2016) R: a language and envi- standards-compliant experimental annotation
ronment for statistical computing. R and enabling curation at the community level.
Foundation for Statistical Computing, Vienna, Bioinformatics 26:2354–2356
Austria. www.R-project.org 15. Kessner D, Chambers M, Burke R et al (2008)
6. De Vos RCH, Moco S, Lommen A et al (2007) ProteoWizard: open source software for rapid
Untargeted large-scale plant metabolomics proteomics tools development. Bioinformatics
using liquid chromatography coupled to mass 24:2534–2536
spectrometry. Nat Protoc 2:778–791 16. Chambers MC, Maclean B, Burke R et al
7. Rafiei A, Atefeh R, Lekha S (2014) Comparison (2012) A cross-platform toolkit for mass spec-
of peak-picking workflows for untargeted liq- trometry and proteomics. Nat Biotechnol
uid chromatography/high-resolution mass 30:918–920
spectrometry metabolomics data analysis. 17. Smith CA, Want EJ, O’Maille G et al (2006)
Rapid Commun Mass Spectrom 29:119–127 XCMS: processing mass spectrometry data for
8. Yu T, Park Y, Johnson JM, Jones DP (2009) metabolite profiling using nonlinear peak align-
apLCMS–adaptive processing of high- ment, matching, and identification. Anal Chem
resolution LC/MS data. Bioinformatics 78:779–787
25:1930–1936 18. Tautenhahn R, Böttcher C, Neumann S (2008)
9. Gorrochategui E, Jaumot J, Tauler R (2015) A Highly sensitive feature detection for high res-
protocol for LC-MS metabolomic data pro- olution LC/MS. BMC Bioinformatics 9:504
cessing using chemometric tools. Protocol 19. Benton HP, Want EJ, Ebbels TMD (2010)
Exchange. https://doi.org/10.1038/ Correction of mass calibration gaps in liquid
protex.2015.102 chromatography-mass spectrometry metabolo-
10. Katajamaa M, Oresic M (2007) Data process- mics data. Bioinformatics 26:2488–2489
ing for mass spectrometry-based metabolo- 20. Franceschi P, Masuero D, Vrhovsek U et al
mics. J Chromatogr A 1158:318–328 (2012) A benchmark spike-in data set for bio-
11. Lange E, Tautenhahn R, Neumann S, Gröpl C marker identification in metabolomics: spike-in
(2008) Critical assessment of alignment proce- metabolomics apple data set. J Chemom
dures for LC-MS proteomics and metabolo- 26:16–24
21. Zhang J, Gonzalez E, Hestilow T et al (2009) 30. Bouyssié D, Dubois M, Nasso S et al (2015)
Review of peak detection algorithms in liquid- mzDB: a file format using multiple indexing
chromatography- mass spectrometry. Curr strategies for the efficient analysis of large
Genomics 10:388–401 LC-MS/MS and SWATH-MS data sets. Mol
22. Aberg KM, Alm E, Torgrip RJO (2009) The Cell Proteomics 14:771–781
correspondence problem for metabonomics 31. Krzywinski M, Altman N (2014) Points of sig-
datasets. Anal Bioanal Chem 394:151–162 nificance: designing comparative experiments.
23. Smith R, Ventura D, Prince JT (2015) LC-MS Nat Methods 11:597–598
alignment in theory and practice: a compre- 32. Krzywinski M, Altman N (2014) Points of sig-
hensive algorithmic review. Brief Bioinform nificance: analysis of variance and blocking. Nat
16:104–117 Methods 11:699–700
24. Koch S, Bueschl C, Doppler M et al (2016) 33. Krzywinski M, Altman N, Blainey P (2014)
MetMatch: a semi-automated software tool for the Points of significance: nested designs. Nat
comparison and alignment of LC-HRMS data Methods 11:977–978
from different metabolomics experiments. Meta. 34. Krzywinski M, Altman N (2014) Points of sig-
https://doi.org/10.3390/metabo6040039 nificance: two-factor designs. Nat Methods
25. Brodsky L, Moussaieff A, Shahaf N et al (2010) 11:1187–1188
Evaluation of peak picking quality in LC-MS 35. Altman N, Krzywinski M (2014) Points of sig-
metabolomics data. Anal Chem nificance: sources of variation. Nat Methods
82:9177–9187 12:5–6
26. Patti GJ, Tautenhahn R, Siuzdak G (2012) 36. Blainey P, Krzywinski M, Altman N (2014)
Meta-analysis of untargeted metabolomic data Points of significance: replication. Nat Methods
from multiple profiling experiments. Nat 11:879–880
Protoc 7:508–516 37. Altman N, Krzywinski M (2015) Points of sig-
27. Prince JT, Marcotte EM (2006) nificance: split plot design. Nat Methods
Chromatographic alignment of ESI-LC-MS 12:165–166
proteomics data sets by ordered bijective inter- 38. Haug K, Salek RM, Conesa P et al (2013)
polated warping. Anal Chem 78:6140–6152 MetaboLights–an open-access general-purpose
28. Martens L, Chambers M, Sturm M et al (2011) repository for metabolomics studies and associ-
mzML–a community standard for mass spec- ated meta-data. Nucleic Acids Res
trometry data. Mol Cell Proteomics 41:D781–D786
10:R110.000133 39. Libiseller G, Dvorzak M, Kleb U et al (2015)
29. Wilhelm M, Kirchner M, Steen JAJ, Steen H IPO: a tool for automated optimization of
(2012) mz5: space- and time-efficient storage XCMS parameters. BMC Bioinformatics
of mass spectrometry data sets. Mol Cell 16:118
Proteomics 11:O111.011379
Chapter 4
Bio- and Chemoinformatics Approaches for Metabolomics

Data Analysis
Michael Witting
Abstract
Metabolomics data analysis includes several repetitive tasks, including data sorting, calculation of exact
masses or other physicochemical properties, or searching for identifiers in different databases. Several of
these tasks can be automated using command line tools or short scripts in different scripting languages like
Perl, Python, or R. This chapter presents simple solutions and short scripts written in R that can be used
for the interaction with specific web services or for the calculation of physicochemical properties or molec-
ular formulae.
Key words R, isotope pattern, Formula calculation, Physicochemical properties, Command line, Web
service, Identifier conversion
1 Introduction
Metabolomics data analysis is very diverse in its nature and ranges

from primary data pre-processing, e.g., alignment, peak picking,
and normalization, to biological interpretation of obtained statisti-
cal results. Automation in metabolomics is a key point for a large-
scale analysis. While the instrumental part today is largely
automated, and peak pickers in current software can handle several
hundreds to thousands of LC-MS runs, the final data analysis still
has to be performed manually to a certain degree. However, differ-
ent steps in data preparation or annotation can be performed in
an automated manner using scripting languages like Perl, Python,
or R. These languages often require only a minimal understanding
of programming to perform basic tasks of data manipulation. R is
a prominent example, extensively used in metabolomics, e.g., the
processing package XCMS was developed in R [1, 2] (see Chapter
3). However, automated data preparation should not only focus on
the initial steps of pre-processing but should also be included in
later ones, especially for time-consuming, repetitive tasks. In
metabolite identification, different physicochemical properties like
41
42 Michael Witting
exact mass or the water/octanol partition coefficient can be calcu-

lated for the comparison with measured masses and the stratifica-
tion of potential candidate structures. For example, a molecule
with a very low logP value cannot be a candidate structure for a
metabolite eluting late in a reversed-phase separation. This
approach has been used by several groups and allows a fast reduc-
tion of candidate lists [3, 4]. Different online tools or freeware
solutions are available for the calculation of different molecular
properties and are usually used for single molecules drawn or
uploaded one at a time, which in case of several hundreds of
metabolites is tedious and time-consuming. Furthermore, metabo-
lite identification is facing a new frontier, de novo description of
potentially hundreds to several thousands of new molecules with
unknown or only partially known structure. The usual first step is
the calculation of molecular formulas from exact mass following
specific rules, known as the seven golden rules [5]. Several algo-
rithms for the calculation of molecular formulas and their valida-
tion using isotope patterns have been described and are used in the
field [6–8]. At a certain stage, new lead structures for novel metab-
olites have to be generated, which need the enumeration of all
possible chemical structures from a single molecular formula.
Different software tools for this tasks exist, e.g., MOLGEN and
the Open Molecule Generator (OMG) and its derivative Parallel
Molecule Generator (PMG) [9–11].
Lastly, reporting of metabolite structures in an automatically
searchable manner is extremely important. Current metabolite
databases contain several hundreds of thousand metabolite struc-
tures, each one with a unique identifier in each individual database.
In order to avoid overlap between different database searches or
not to report duplicate structures, unique identifiers are impor-
tant, whereas the chemical structure of a metabolite is the most
unique identifier possible. Therefore, it should always be reported
with metabolite identifications. However, images of structures can-
not be read by automatic systems. Different string representations
of molecules have been invented, including the SMILES and
InChI system. For larger molecules, these strings can be quite
lengthy. In case of InChI, a short representation of the InChI key
has been invented, which represents a hashed version of the long
InChI code. Fiehn and Kind reported that reporting of chemical
structures in databases, as well as in publications, is an important
step to unify approaches in metabolomics [12]. However, machine
readable structure representation like SMILES or InChI keys is
hardly human readable. The same is true for IUPAC names of mol-
ecules if the molecule is getting larger. Therefore, often trivial
names or identifiers of common biological databases, like KEGG,
HMDB, Lipid Maps, or ChEBI, are used. In contrast to struc-
tures, they can be ambiguous. For example, acetate is usually used
as a name for acetic acid, since it is the form in which it is present
Bio- and Chemoinformatics Approaches for Metabolomics Data Analysis 43
in cells. In KEGG, acetate and acetic acid are stored under the
same KEGG id but have different accession numbers in
ChEBI. Cross mapping of different database identifiers is therefore
important and has to be updated on a regular basis and manually
annotated and corrected. The chemical translation service is an
open tool available to the metabolomics community, which offers
highly curated mappings between identifiers from databases that
are important for metabolomics [13].
This chapter presents approaches for the calculation of physi-
cochemical parameters, molecular formulae from exact masses,
generation of molecular structures, mapping of different database
identifiers and their conversion, as well as mapping of metabolites
onto pathways. Much emphasis is laid on (semi)automatic process-
ing using command line tools and short R scripts, based on previ-
ously published packages. Simple examples show the possibilities
and are also easy to use even for beginners.
2 Materials
All presented calculations and scripts were performed on a Windows

7 or 10 64bit machine using R 3.2.5 within RStudio Version
0.99.903, Java 1.8.0_111 64bit, and ChemAxon JChem Suite
16.11.28.0 64 bit. The Parallel Molecule Generator was down-
loaded from SourceForge (https://sourceforge.net/projects/
openmg/). All R libraries were downloaded from the
Comprehensive R Archive Network.
2.1 Usage This chapter uses two different types of command line tools. Usage
of Command Line requires a few simple commands to handle the command line. The
Tools command line shell can be called in windows by using the windows
key and typing cmd and pressing the enter button. This should
open a command line in windows similar to the one shown in
Fig. 1. This chapter will solely focus on the Windows version of
commands.
First, if installed correctly, the ChemAxon tools should be
available as environmental variables and can be called by simply
typing cxcalc or molconvert. This can be tested by simply typing
cxcalc. If the following message appears, the environmental vari-
ables are not correctly set: “cxcalc is not recognized as an internal
or external command, operable program or batch file.” If this is
the case, please check the ChemAxon online manual on how to
integrate manually the environmental variables. The tools can be
also called directly from their installation folder, usually C:\Program
Files\ChemAxon\JChemSuite\bin.
The second possibility used in this chapter is to call java pack-
ages in a .jar file to perform calculations. For this, Java has to be
installed correctly and integrated in the environmental variables. .jar
44 Michael Witting
Fig. 1 (a) Example for the use of command line tools on a Windows system. The first lines demonstrate the use
of the cxcalc command from the ChemAxon JChem Suite for calculation of an exact mass, molecular formula,
and logP. The following lines show how to call the Parallel Molecule Generator java file to calculate all possible
structures for the molecular formula of leucine. (b) Screenshot of RStudio showing all basic windows. In the
scripting area, all commands can be collected in order to allow repetitive analysis without the need to recall all
commands. The global environment contains all currently active variables and allows their inspection for type
and size, while the output console allows to run single commands and prints all outputs
files are executed by calling it with the command java –jar XXX.jar
args, whereby XXX.jar refers to the .jar file to be used and args is
replaced by the argument that shall be passed to the program. An
example is shown in the lower part of Fig. 1. An important command
to remember is how to change between directories; this is done by
using the cd command, e.g., cd <pathToDirectory> changes into
the respective directory, while cd .. changes to next higher/lower
level in the directories, e.g., from C:\Users\TestUser\ to C:\Users.
2.2 R Scripts All scripts can be used directly in R or more comfortably with an
IDE like RStudio, which allows more convenient work without the
need to remember all variables, since they are displayed in a sepa-
rate window and also allow manual inspection (Fig. 1b). The pre-
sented code snippets work with different packages which are not
part of the R base distribution and have to be therefore installed
manually. Specific packages are jsonlite, XML, rcdk, RCurl, curl,
rcdk, OrgMassSpecR, and CHNOSZ with their dependencies.
The following script will install all needed packages from CRAN
()https://cran.r-project.org/.
#list of packages needed
packages <- c("jsonlite", "XML", "rcdk",
"RCurl", "curl", "OrgMassSpecR", "CHNOSZ")
install.packages(packages, dependencies = T)
For the work with R scripts, a basic understanding of R func-
tionalities and data structures is needed (see Note 1). Most results
in the example scripts are returned as a list of specific datatypes,
e.g., formula generation with rcdk returns a list of the formal class
cdkFormula. Positions in this list are accessed via [i] or [[i]], where
i is the index of interest and specific information like the string
containing the molecular formula are stored in slots and can be
accessed via the @ operator.
#get formula string
formulae[[1]]@string
Another example is the use of JavaScript Object Notation,
short JSON, for transfer of structured data transfer, e.g., from web
services. This data format is used by many REST web services for
result representation. One example is the Chemical Translation
Service offered by the Fiehn Lab. A specific JSON result could
look like this.
[
{
"fromIdentifier": "inchikey",
"searchTerm": "QNAYBMKLOCPYGJ- REOHCLBHSA- N",
"toIdentifier": "Chemical Name",
"result":
[
"L-Alanine",
"L-2-Aminopropanoic acid"
]
}
]
The package jsonlite or the content() function from the httr
package allows a direct conversion of the results in a JSON string
to specific datatypes for a more convenient access, which is
described in a later section of this chapter.
46 Michael Witting
3 Methods
3.1 Calculation Different physicochemical properties are often employed in data

of Exact Mass, logD, analysis of metabolomics data, with exact mass being the most
and Others Using important in mass spectrometry-based metabolomics. The water/
ChemAxon octanol partition coefficient logP is often used as a value of mole-
cule polarity and proxy for retention in LC-MS and can help in
rejection of false-positive annotations [3].
One package often employed for this task offered by the com-
pany ChemAxon is called JChem and includes several command
line utilities based on their proprietary Java tools. These command
line tools can be used for different purposes. The most employed
command for the needs in metabolomics is cxcalc used for the cal-
culation of different physicochemical properties. The second most
important command is molconvert, which allows the interconver-
sion of different chemical structure formats. Lastly, for commands
not available via cxcalc directly, evaluate can be used. Examples
below explain the different use of these commands and how they
can be combined in useful blocks (see Notes 2 and 3).
1. A file containing several structures for testing can be prepared
using Microsoft Excel or a text editor. The header should con-
tain #SMILES and Name. The # is important since it labels the
first line as header. Structures and names should be separated by
a tab. Following entries are used in this paragraph.
#SMILES Name
Cn1cnc2c1c(=O)n(c(=O)n2C)C Caffeine
CC(C)C[C@@H](C(=O)O)N L-Leucine
CC[C@H](C)[C@@H](C(=O)O)N L-Isoleucine
C([C@@H]1[C@H]([C@@H]([C@H]([C@@H](O1)O)O)O)
O)O D-Glucose
c1cc(ccc1c2c(c(=O)c3c(cc(cc3o2)O)O)O)O
Kaempferol
2. The command cxcalc is used for calculating different properties
of molecules, whereby the most important are exact mass, for-
mula, and logP. Following command can be used for the simple
calculation of all three properties for a single molecule directly
in the command line.
cxcalc exactmass formula logP Cn1cnc2c1c(=O)
n(c(=O)n2C)C
This function delivers an output on the console with following
fields separated by a tab: id, exact molecular weight, formula, logP.
id Exact molecular weight Formula logP
1 194,080375578 C8H10N4O2 -0,55
However, if you want to store the result in a file, you have to
add the following tag –o<yourPathToResultFile>, e.g.,
cxcalc –o E:\cxCalcResult.txt exactmass for-

mula logP
Cn1cnc2c1c(=O)n(c(=O)n2C)C
In the similar way, all structures stored in test.smiles can be
converted and stored in a file.
cxcalc –o E:\cxCalcResult.txt exactmass for-
mula logP E:\inputSmiles.txt
3. Using the commands above, an identifier is reported instead of
the original SMILES string. This could be overcome by using
the molconvert function and conversion to .sdf format before
calculation. The .sdf format allows to store properties in a tag
for each molecule. Afterwards the .sdf file can be back con-
verted to SMILES together with the tags. In order not to run
three single commands but one, the different commands can be
combined by using the pipe | command, which works on both
Linux and Windows. The part -T “*” tells the molconvert com-
mand to store all calculated properties in tags.
molconvert sdf Cn1cnc2c1c(=O)n(c(=O)n2C)C
| cxcalc -S exactmass formula | molconvert
smiles -T "*"
#SMILES name EXACTMASS FORMULA
Cn1cnc2n(C)c(=O)n(C)c(=O)c12 194,080375578
C8H10N4O2
This creates the following output, which can be stored again in
a file using the –o tag. The {smiles} behind the input file path
defines the input as SMILES, which is necessary for molconvert if
data is read from a text file.
molconvert sdf E:\inputSmiles.txt{smiles} |
cxcalc -S exactmass formula logP | molcon-
vert smiles -T "*" -o E:\cxCalcResult.txt
4. SMILES is a proprietary representation of chemical structures
and is not open, which lead to several implementations includ-
ing an open version. Several SMILES can represent the same
molecule. In contrast to this, the InChI representation was
from the beginning of an open project, which leads to one
implementation. Therefore, InChI representations are unique
and should be used for the reporting of molecular structures.
One major drawback is that InChIs are even harder to read by
humans than SMILES. For very efficient reporting, so-called
InChI keys, a hashed version of InChIs are used. Both can be
generated with the command evaluate -e molString(‘inchikey’)
or evaluate -e molString(‘inchi:AuxNone’) (note: AuxNone
suppresses the ChemAxon auxiliary information). The com-
mand below first converts all SMILES stored in test.smiles to .
sdf format and adds InChI and InChI keys with the tags INCHI
48 Michael Witting
and INCHIKEY. Then exact mass and formula are calculated

and finally everything is back converted to SMILES
representation with all additional fields and stored in a tab-sep-
arated text file.
molconvert sdf E:\inputSmiles.
txt{smiles} | evaluate -e "molString('in
chi:AuxNone')" -S -t INCHI | evaluate -e
"molString('inchikey')" -S -t INCHIKEY | cx-
calc -S exactmass formula logP | molconvert
smiles -T "*" -o E:\cxCalcResult.txt
3.2 Calculation For people who would not like to use ChemAxon or are not
of Exact Mass, logD, allowed to use it, the Chemistry Development Kit (CDK) is a rich
and Others Using open-source alternative [14, 15]. However, it requires knowledge
RCDK on programming in Java, and no command line tools are available.
The package rcdk provides all CDK functionalities in R and can be
accessed easily. An important point is the correct installation of the
rJava package which is used to interface java functionalities with R
(see Notes 4 and 5).
1. First the required library is loaded.
#load required libraries
library(rcdk)
2. Next, a molecule to work with is generated by parsing a SMILES
string, in this case caffeine.
#parse smiles
smile <- 'Cn1cnc2c1c(=O)n(c(=O)n2C)C'
mol <- parse.smiles(smile)[[1]]
3. Before the molecule can be passed to the functions for the cal-
culation of exact mass and logP, it has to be prepared. This
includes detection of aromaticity, type of atoms, and configura-
tion of isotopes.
#prepare molecule
do.aromaticity(mol)
do.typing(mol)
do.isotopes(mol)
4. After the molecule has been prepared, the get.exact.mass(), get.
alogp(), get.xlogp(), get.tpsa(), and get.volume() functions can
be used for calculation of exact mass, logP, total polar surface
area, and the molecular volume.
#do the calculation
get.exact.mass(mol)
get.alogp(mol)
get.xlogp(mol)
get.tpsa(mol)
get.volume(mol)
5. Further descriptors can be calculated with the rcdk package and

used for different purposes, e.g., QSRR modeling [16]. Similar
to the example above, the molecules have to be parsed and pre-
pared. Molecular descriptors are grouped into different catego-
ries, which can be retrieved with the get.desc.categories()
function. From each category, the descriptor names are avail-
able via the get.desc.names() function.
#get all possible descriptor categories
descriptorCategories <- get.desc.catego-
ries()
descriptorCategories
#get descriptor names of all topological de-
scriptors
descriptorNames <- get.desc.
names(descriptorCategories[3])
descriptorNames
6. These names can be used then to tell rcdk which descriptors
have to be calculated. Calculation is carried out using the eval.
desc() function. If all possible descriptors should be calculated,
“all” can be used as argument instead of specific descriptors
names. The first example calculates only topological descrip-
tors, whereas the second calculates all descriptors. The type of
the result is in both cases a list.
#get descriptor names of all topological de-
scriptors
descriptorNames <- get.desc.
names(descriptorCategories[3])
descriptorNames
#do the calculation
topologicalDescriptors <- eval.desc(mol, de-
scriptorNames)
topologicalDescriptors
#calculate all descriptors
allDescriptors <- eval.desc(mol, get.desc.
names("all"))
allDescriptors
3.3 Formula rcdk offers possibilities useful for mass spectrometry-based metab-
and Isotope Pattern olomics, e.g., calculation of molecular formulas from masses, vali-
Calculation Using dation of molecular formulae, or calculation of isotope patterns. A
RCDK usual first step in metabolite identification is the calculation of
molecular formula from a measured exact mass and the compari-
son of theoretical and measured isotope patterns. Different soft-
ware tools exist for this purpose. The formula calculation and
validation is based on the seven golden rules proposed by Kind and
Fiehn [5].
50 Michael Witting
The following examples show how rcdk can be used for the
calculation of molecular formulae from a given exact mass.
1. First, the required library is loaded. In this first part, only rcdk
is required; however, the packages OrgMassSpecR and
CHNOSZ provide useful functionalities used later.
library(rcdk)
library(OrgMassSpecR)
library(CHNOSZ)
2. The R script starts from a given isotopic pattern measured
([Glucose + Na]+). First, the monoisotopic mass is located (in
this case lowest m/z value) and is used as input for the generate.
formula() function of the rcdk package. This function has sev-
eral important parameters to set; first the search window has to
be set correctly; otherwise, the search for potential sum formu-
las is too exhaustive. Second, as stated in the seven golden rules,
the number of elements has to be restricted. Since the exact
mass used as input corresponds to a [M+Na] + adduct, sodium
has to be included; additionally, the parameter charge has to set
correct, in this case to +1.
#get measured isotope pattern ([Glucose +
Na]+)
isoPatternMeasured <- data.frame(mz =
c(203.052609, 204.056051, 205.057227,
206.060394, 207.061845),
int = c(100.000, 6.856, 1.433, 0.087,
0.009))
#get monoisotopic mass
exactmass <- isoPatternMeasured$mz[1]
#calculate all possible formulae
formulae <- generate.formula(exactmass, win-
dow = 0.001,
elements = list(c("C",0,10),c("H",0,50),
c("N",0,5),c("O",0,50),
c("Na",0,1)),
validation = T,
charge = 1)
3. In the first case, only one valid chemical formula is produced
and returned as a list of cdkFormula objects. Different values
can be accessed via the @ operator.
#get formula string
formulae[[1]]@string
#get charge
formulae[[1]]@charge
#get mass
formulae[[1]]@mass
4. If more than one result is produced, the different results can be

accessed by iterating through the returned list. The following
example uses the isotope pattern of glutathione as input.
Glutathione contains an S atom and therefore has a very specific
M+2 pattern. A resolution of 50,000 was used, and glutathione
was detected as [M+H] + adduct.
#get measured isotope pattern ([Glutathione
+ H]+)
isoPatternMeasured <- data.frame(mz =
c(308.091083, 309.089103, 309.094509,
310.087005, 310.096054, 311.090415,
311.098745),
int = c(100.000, 1.886, 11.251, 4.606,
1.896, 0.524, 0.168))
#get monoisotopic mass
exactmass <- isoPatternMeasured$mz[1]
5. Similar to the previous example, the generate.formula() func-
tion is employed to produce potential formulae. In this case, it
is important to consider sulfur in the calculation.
#calculate all possible formulae
formulae <- generate.formula(exactmass, win-
dow = 0.001,
elements = list(c("C",0,20),c("H",0,50),
c("N",0,5),c("O",0,50),
c("S",0,5)),
validation = T,
charge = 1)
6. The list that is returned contains two different possible molecu-
lar formulae. A for loop allows to iterate through this list and
access single cdkFormula objects.
for(formula in formulae) {
print(formula)
}
7. In the next step, the get.isotopes.pattern() function is used to
simulate an isotopic pattern. The two parameters for this func-
tion are a valid cdkFormula object and the minimal abundance
of an isotope. At the current stage, the isotope pattern predic-
tion in rcdk does not take into account the charge. Therefore,
masses of the predicted isotope pattern have to be corrected for
the mass of an electron.
#simulate isotopic pattern
isoPatternCalculated <- get.isotopes.
pattern(formula, minAbund = 0.001)
#check for neutral???
#rcdk currently does not support charged
formulas in isotope pattern
52 Michael Witting
#correction for the mass of a electron need-

ed
isoPatternCalculated[,1] <- isoPatternCalcu-
lated[,1] - 5.48579909070 * 10^-4
8. These simulated isotopic patterns have to be compared with the
measured for confirmation. Different metrics are available for
this. In this example, a simple dot product is used for calculat-
ing similarity, offered by the SpectrumSimilarity() function of
the OrgMassSpecR package, which returns the cosine similarity
value between two mass spectra. Since comparison of all differ-
ent predicted isotope pattern with the measured will yield scores
close to 0.9, for a better comparability dissimilarity is used,
which is calculated by the following formula: (1—similarity) *
1000. The smaller this value is, the better two isotopic patterns
match. Together with the calculation, the function produces a
mirror plot of the two (measured and predicted) isotope pat-
terns. The final dissimilarity score is printed but can be also
stored in a list or a data frame.
#iterate through result list
for(formula in formulae) {
print(formula)
#simulate isotopic pattern
isoPatternCalculated <- get.isotopes.
pattern(formula, minAbund = 0.001)
#check for neutral???
#rcdk currently does not support charged
formulas in isotope pattern
#correction for the mass of a electron
needed
isoPatternCalculated[,1] <- isoPatternCal-
culated[,1] - 5.48579909070 * 10^-4
#compare measurend and calculated isotopic
pattern
similarity <- SpectrumSimilarity(isoPattern
Measured, isoPatternCalculated,
t = 0.001, b = 0.01,
top.label = "measured isotope pattern",
bottom.label = "simulated isotope pattern",
xlim = c(305, 315))
#label peaks (aesthetics for plotting)
plot.window(xlim = c(305,315), ylim = c(-
150, 150))
text(310, 50, paste0("dissimilarity = ",
round((1-similarity)*1000, 2)), cex = 0.75)
mtext(formula@string, line = 1)
print((1-similarity)*1000)
}
9. Lastly, if further filtering of formulae should be performed, the

formula string can be casted into a list with elements and their
respective atom counts using the makeup() function from the
CHNOSZ package.
#isolate formula string
formulaString <- formulae[[1]]@string
#parse into named list
elementList <- makeup(formulaString)
elementList #named vector
#get number of carbon atoms
elementList["C"]
3.4 Molecular One major corner stone in identification of unknown metabolites

Structure Generation is to calculate potential formulae for a given exact mass as described
above. However, this is only the first step. Behind a single sum
formula, several isomeric structures can be found. Generation of
potential lead structures is therefore a key issue. An open-source
tool for structure generation from molecular formulae has been
published by Peironcely et al. called Open Molecule Generator
(OMG) [17]. A command line tool or a GUI is available from
https://sourceforge.net/projects/openmg/, including the
Parallel Molecule Generator, which takes advantage of multicore
capabilities. A simple example shows how this tool can be used,
whereby advanced filtering of structures is performed utilizing
ChemAxon tools.
1. The first example calculates all possible structures for C6H13NO2,
the sum formula of l-Leucine, and stores the results in a .sdf file
E:\PMG>java -jar PMG_1.0.jar C6H13NO2 -o
leucinePMGExample.sdf
2. This generates 32,395 possible molecules. PMG allows valences
of 4 for carbon; 2 for oxygen; 3 or 5 for nitrogen; 2, 4, or 6 for
sulfur; and 3 or 5 for phosphor. Therefore, the result list may
also include weird-looking structures. These results can be
cleaned by valence checks, e.g., cxcalc offers the possibility to
clean the data using Markush enumerations. This filters the
molecules based on valences usually found in organic mole-
cules, e.g., 3 for nitrogen.
cxcalc -o leucinePMGExampleEnum.sdf enumera-
tions -v -f sdf leucinePMGExample.sdf
With this filtering, still 24,843 possible molecules remain,
which indicates that further possibilities to restrict the possible
chemical space are needed (Fig. 2).
3. The generated, very exhaustive list of possible molecules can
be further narrowed down by using the –goodlist and –badlist
options, which allows to specify substructures to be included
or excluded. Note that the Chemistry Development Kit (cdk)
54 Michael Witting
Fig. 2 (a) Enumeration of chemical structures using from structures generated by PMG containing a
cxcalc from the JChem tools allows to remove chemi- valence 5 nitrogen were removed, reducing possible
cal invalid structures, e.g., containing nitrogen atoms chemical structures from 32,395 to 24,843
of valence 5. In this particular example, all results candidates
has to be installed to make use of these options. For this lists of

potential substructures identified by a MS/MS experiment can
be included, e.g., a neutral loss of 44 can indicates the presence
of a carboxylic acid. In this case, the –goodlist option would
include a carboxylic acid. For the leucine example, an amino
acid moiety as good list fragment for molecular structure gen-
eration is included, and the command is executed.
java -jar PMG_1.0.jar C6H13NO2 -goodlist
leucineFrag.sdf -o leucineGoodList.sdf
4. These results contain again molecules with nitrogen of valence
5 and can be filtered using the enumeration option as described
above.
cxcalc -o leucinePMGExampleGoodListEnum.sdf
enumerations -v -f sdf leucinePMGExample-
GoodList.sdf
From 56 possible formulas containing an amino acid group,
47 remain after the enumeration of valences. These structures can
be used as input for in silico fragmentation tools like MetFrag or
retention time prediction [18, 19].
3.5 Identifier The most unique identifier of a molecule is its structure. However,
Conversion using it is hardly human readable, especially for larger molecules. Still, its
the Chemical uniqueness makes it number one for reporting identified metabo-
Translation Service lites. SMILES and InChIs are quite long; however, the InChI keys
represent an elegant way for structure reporting and can be used in

the Chemical Translation Service (CTS), offered by the Fiehn Lab
for retrieval of ids of different databases. The most used databases
in the field of metabolomics are the Kyoto Encyclopedia of Genes
and Genomes, Human Metabolome Database, Lipid Maps, and
MetaCyc [20–26]. CTS is available under http://cts.fiehnlab.
ucdavis.edu/ and offers single and batch id conversion [13].
However, this is limited to maximum 100 identifiers in one batch.
If larger amounts have to be processed, conversion has to be
repeated several times. Additionally, if CTS should be integrated
into a workflow or several hundreds of identifiers need to be pro-
cessed, the offered web services are good alternatives. These can be
used to convert different database identifiers into each other or
search for chemical names from InChI keys. The following exam-
ples show that this web service can be invoked in the R environ-
ment (see Note 6).
1. First, the two required libraries for the following examples are
loaded.
library(httr)
library(jsonlite)
2. The first example uses the GET() function from the httr pack-
age to retrieve data via a get request from the constructed query
URL. Details on the construction of different query URLs can
be found on the CTS webpage (http://cts.fiehnlab.ucdavis.
edu/moreServices/index). First the identifier from and to
which should be converted is defined together with a query.
#define from and to identifier
from <- 'inchikey'
to <- 'Chemical Name'
#generate query
queryString <- "QNAYBMKLOCPYGJ- REOHCLBHSA- N"
3. Next, a query URL is constructed from the basic URL for con-
version, the identifiers that should be employed, and the query
itself. This string has to be encoded as URL, since its specific
character is not supported in URLs (e.g., blanks are represented
as %20). The URLencode() function converts specific characters
to representation used in URL. Finally, the query is executed
using GET(). This returns all details of the query including the
result, which can be accessed with the content() function from
the same package, either processed as text, which returns a
JSON string, or parsed, which returns a list containing the
parsed results.
#construct url for GET request
baseUrl <- "http://cts.fiehnlab.ucdavis.edu/
service/convert"
56 Michael Witting
queryUrl <- paste0(baseUrl, "/", from, "/",

to, "/", queryString)
#use httr GET to retrieve results
queryResult <- GET(URLencode(queryUrl))
#access to content of queryResult
result <- content(queryResult, "text")
result f<- content(queryResult, "parsed")
4. Easier access to the web service and the final result can be done
by using the fromJSON function from the jsonlite package,
which directly performs the request and returns a parsed result
list.
#use jsonlint fromJSON
queryResult <- fromJSON(URLencode(queryUrl))
5. The CTS web services do not allow multiple conversions or
mixed input identifiers at the moment. The next example shows
what to do if you are dealing with a mixed list of identifiers and
you would like to produce a list of one common identifier. Each
database identifier has its unique structure, e.g., KEGG com-
pound identifiers consist of a C followed by a five-digit number.
This can be used to identify the type identifier based on regular
expressions. The MIRIAM registry hosted at the European
Bioinformatics Institute contains several hundred identifiers
and offers preconfigured regular expressions that can be used
(http://www.ebi.ac.uk/miriam/main/mdb?section=intro)
[27]. In the next example, different queries are tested with reg-
ular expressions for their type and the respective from identifier
is used.
#define from and to identifier
from <- ''
to <- 'inchikey'
#generate queries
queries <- c('QNAYBMKLOCPYGJ- REOHCLBHSA- N',
'C00041', 'CHEBI:16977')
#base url for construction of query url
service/convert"
for(queryString in queries) {
if(grepl("^[A-Z]{14}-[A-Z]{10}(-[A-Z])?",
queryString, perl = T)) {
from <- 'inchikey'
} else if(grepl("^CHEBI:\\d+$", quer-
yString, perl = T)) {
from <- 'ChEBI'
} else if(grepl("^C\\d+$", queryString,
perl = T)) {
from <- 'KEGG'
} else {
}
#construct url for request
service/convert"
queryUrl <- paste0(baseUrl, "/", from, "/",
to, "/", queryString)
#use jsonlint fromJSON
queryResult <-
fromJSON(URLencode(queryUrl))
print(queryResult$result)
}
3.6 Manual Both two big pathway servers MetaCyc and KEGG offer web APIs
and Automated to interact with them. The KEGG database utilizes a simple REST
Mapping on KEGG web service for interaction. Different functions allow the retrieval
Pathways of information on pathways, enzymes, and their linked compounds.
This allows querying the DB within a workflow. The functionalities
include a mapper, which highlights compounds of interest with
default or user-defined colors.
1. This is a simple example how to retrieve a .png file manually
using the KEGG Search & Color function (http://www.
genome.jp/kegg/tool/map_pathway2.html). In this example,
all metabolites of the upper glycolysis pathway are colored in
red and all of the lower glycolysis pathway are in blue.
C00267 red C00074 blue
C00110 red
C00236 red
C00197 red
C00631 red
2. These values can be copied and pasted into the field “Enter
objects one per line followed by bgcolor, fgcolor:” (compound
input in Fig. 3). If pathway maps of a specific organism should
be colored, this organism has to be selected under “Search
Against” (organism input in Fig. 3). After pressing the “Exec”
button, a list of links to the colored pathway maps is available in
the browser. The number in brackets behind each pathway
name indicates the number of compounds that have been
mapped to this specific pathway.
3. Since manual mapping of several hundreds of pathways
can be tedious, an automated version would be useful. This can
be achieved by a small R script. The first example just performs
simple highlighting of compounds on a single pathway map.
58 Michael Witting
Fig. 3 Manual mapping of compounds is performed with the KEGG Mapper (http://www.genome.jp/kegg/tool/
map_pathway2.html). KEGG IDs of compounds of interest are pasted into the text field marked with compound
input; optional fore and background color can be defined. An organism can be selected by entering the respec-
tive three letter code in the field marked as organism input. After pressing the Exec button, a list with all
pathways containing minimum one of the compounds is shown. By clicking on the respective pathway, the
map is shown
The link returns a .html page, but a downloadable .png is hid-

den in the <img> html tag. We utilize the XML package to
isolate specifically the html tag that contains the link to the .png
file, which is downloaded to the working directory using curl_
download(). A detail description of the construction of KEGG
web link can be found under http://www.kegg.jp/kegg/rest/
weblink.html.
#generate list with KEGG IDs and bgcolor
compoundList <- list(id = c("C00267",
"C00103", "C00668", "C05345", "C05378",
"C00118", "C00110", "C00236",
"C00197", "C00631", "C00074", "C00036",
"C00022", "C00186", "C00024", "C00033"))
#id of pathway on which metabolites will be
mapped
pathwayId <- "map00010"
baseUrl <- "http://www.kegg.jp/pathway"
#create query url
queryUrl <- paste0(baseUrl, "/", pathwayId,

"+", paste(compoundList$id, collapse="+"))
#get html content
htmlContent <- getURL(queryUrl)
htmlDoc <- htmlParse(htmlContent)
img <- xpathSApply(htmlDoc, "//img[@
name='pathwayimage']/@src")
#download picture
curl_download(url = paste0("http://www.
kegg.jp", img), destfile = paste0(pathwayId,
".png"))
4. Finally, more complicated mapping with self-defined colors can
be performed (e.g., defining back and foreground colors). This
requires a list of compounds and their respective colors. In the
example below, the same coloring as for manual mapping is
used. Compound identifiers have to be linked with their colors
by a+, and different compound/color pairs are separated by a
new line separator “\n.” A for loop can be used to iterate
through the list of compounds and construct the correct string
for the query.
#generate list with KEGG IDs and bgcolor
compoundList <- list(id = c("C00267",
"C00103", "C00668", "C05345", "C05378",
"C00118", "C00110", "C00236",
"C00197", "C00631", "C00074", "C00036",
"C00022", "C00186", "C00024", "C00033"),
bg = c("red", "red", "red", "red", "red",
"red", "red", "red",
"red", "red", "blue", "blue", "blue",
"blue", "blue", "blue"))
#create string of compounds and color ac-
cording to KEGG API
compoundString <- ""
for(i in 1:length(compoundList[[1]])) {
if(i == 1) {
compoundString <-
paste0(compoundList$id[i],
"+",compoundList$bg[i])
} else {
clipboard <- paste0(compoundList$id[i],
"+",compoundList$bg[i])
compoundString <- paste0(compoundString,
"\n", clipboard)
}
}
5. The query URL is constructed by combining a base URL with
a pathway identifier and the string containing the compounds
60 Michael Witting
and colors. URLencode() is used to convert the string to its

URL representation.
#id of pathway on which metabolites will be mapped
pathwayId <- "map00010"
baseUrl <- "http://www.kegg.jp/kegg-bin/show_pathway?map="
#create query url
queryUrl <- URLencode(paste0(baseUrl, pathwayId, "&multi_
query=",compoundString))
#get html content
htmlContent <- getURL(queryUrl)
htmlDoc <- htmlParse(htmlContent)
img <- xpathSApply(htmlDoc, "//img[@name='pathwayimage']/@src")
#download picture
curl_download(url = paste0("http://www.kegg.jp", img), destfile
= paste0(pathwayId, ".png"))
4 Notes
1. Consult a book or web page on working with R if not familiar,

e.g., https://cran.r-project.org/doc/
manuals/r-release/R-intro.pdf
2. The command line tools can be invoked into any scripting lan-
guage which can handle command line tools, e.g., R or Perl.
For integration, in Java libraries, all mentioned calculations are
available. Please check license from ChemAxon for exact use.
3. The molconvert and cxcalc command line tools need the cor-
rect installation of JChem and setting of the respective environ-
mental variables. Please refer to the installation manual.
4. Please check the rJava CRAN package. (Re)Installation of the
latest Java version followed by (re)installation of rJava may help
with problems.
5. CDK can also use the KNIME workflow environment [28].
6. Once the principle of REST and JSON is clear, the concept can
be applied in different web services in metabolomic databases.
Lipid Maps offers a REST web service for interaction with their
data.
References
1. Benton HP, Wong DM, Trauger SA et al (2008) ment, matching, and identification. Anal Chem
XCMS2: processing tandem mass spectrometry 78:779–787
data for metabolite identification and structural 3. Müller C, Dietz I, Tziotis D et al (2013)
characterization. Anal Chem 80:6382–6389 Molecular cartography in acute chlamydia
2. Smith CA, Want EJ, O'Maille G et al (2006) pneumoniae infections—a non-targeted
XCMS: processing mass spectrometry data for metabolomics approach. Anal Bioanal Chem
metabolite profiling using nonlinear peak align- 405:5119–5131
4. Stanstrup J, Gerlich M, Dragsted LO et al phy mass spectrometry and its use for peak
(2013) Metabolite profiling and beyond: annotation in metabolomics.
approaches for the rapid processing and anno- Metabolomics:1–11
tation of human blood serum mass spectrome- 17. Peironcely JE et al (2012) OMG: Open
try data. Anal Bioanal Chem Molecule Generator. J Cheminformatics
405(15):5037–5048 4:1–13
5. Kind T, Fiehn O (2007) Seven golden rules for 18. Wolf S et al (2010) In silico fragmentation for
heuristic filtering of molecular formulas computer assisted identification of metabolite
obtained by accurate mass spectrometry. BMC mass spectra. BMC Bioinformatics 11:148
Bioinformatics 8:105 19. Gerlich M, Neumann S (2013) MetFusion:
6. Tziotis D, Hertkorn N, Schmitt-Kopplin P integration of compound identification strate-
(2011) Kendrick-analogous network visualisa- gies. J Mass Spectrom 48:291–298
tion of ion cyclotron resonance Fourier trans- 20. Kanehisa M, Goto S (2000) KEGG: Kyoto
form mass spectra: improved options for the encyclopedia of genes and genomes. Nucleic
assignment of elemental compositions and the Acids Res 28:27–30
classification of organic molecular complexity.
Eur J Mass Spectrom 17:415–421 21. Kanehisa M et al (2006) From genomics to
chemical genomics: new developments in
7. Witting M, Lucio M, Tziotis D et al (2015) KEGG. Nucleic
DI-ICR-FT-MS-based high-throughput deep Acids Res 34(suppl 1):D354–D357
metabotyping: a case study of the Caenorhabditis
Elegans–Pseudomonas Aeruginosa infection 22. Wishart DS et al (2012) HMDB 3.0—the
model. Anal Bioanal Chem 407:1059–1073 human metabolome database in 2013. Nucleic
Acids Res 41((Database issue)):D801–D807.
8. Treutler H, Neumann S (2016) Prediction, https://doi.org/10.1093/nar/gks1065
detection, and validation of isotope clusters in
mass spectrometry data. Meta 6:E37 23. Wishart DS et al (2009) HMDB: a knowledge-
base for the human metabolome. Nucleic Acids
9. Kerber A et al (1998) MOLGEN 40 Match- Res 37(Database):D603–D610
communications in mathematical and in com-
puter. Chemistry 37:205–208 24. Sud M et al (2007) LMSD: LIPID MAPS
structure database. Nucleic Acids Res 35(suppl
10. Peironcely JE et al (2012) OMG: Open 1):D527–D532
Molecule Generator. J Cheminformatics 4:21
25. Caspi R et al (2008) The MetaCyc database of
11. Jaghoori MM et al (2013) PMG: multi-core metabolic pathways and enzymes and the
Metabolite Identification. Electronic Notes in BioCyc collection of pathway/genome data-
Theoretical Computer Science 299:53–60 bases. Nucleic Acids Res 36(suppl
12. Kind T, Scholz M, Fiehn O (2009) How large 1):D623–D631
is the metabolome? A critical analysis of data 26. David S. Wishart, Yannick Djoumbou
exchange practices in chemistry. PLoS One Feunang, Ana Marcu, An Chi Guo, Kevin
4:e5440 Liang, Rosa Vázquez-Fresno, Tanvir Sajed,
13. Wohlgemuth G et al (2010) The chemical Daniel Johnson, Carin Li, Naama Karu, Zinat
translation service—a web-based tool to Sayeeda, Elvis Lo, Nazanin Assempour, Mark
improve standardization of metabolomic Berjanskii, Sandeep Singhal, David Arndt,
reports. Bioinformatics 26:2647–2648 Yonjie Liang, Hasan Badran, Jason Grant,
14. Steinbeck C et al (2003) The chemistry devel- Arnau Serra- Cayuela, Yifeng Liu, Rupa
opment kit (CDK): an open-source java library Mandal, Vanessa Neveu, Allison Pon, Craig
for chemo- and bioinformatics. J Chem Inf Knox, Michael Wilson, Claudine Manach,
Comput Sci 43:493–500 Augustin Scalbert; HMDB 4.0: the human
15. Willighagen EL, Mayfield JW, Alvarsson J, Berg metabolome database for 2018, Nucleic Acids
A, Carlsson L, Jeliazkova N, Kuhn S, Pluskal T, Research, gkx1089, https://doi.
Rojas-Chertó M, Spjuth O, Torrance G, Evelo org/10.1093/nar/gkx1089
CT, Guha R, Steinbeck C (2017) The Chemistry 27. Juty N, Le Novère N, Laibe C (2012)
Development Kit (CDK) v2.0: atom typing, Identifiers.org and MIRIAM registry: commu-
depiction, molecular formulas, and substructure nity resources to provide persistent identifica-
searching. J Cheminform 9:33. https://doi. tion. Nucleic Acids Res 40:D580–D586
org/10.1186/s13321-017-0220-4 28. Beisken S et al (2013) KNIME-CDK: work-
16. Cao M et al (2014) Predicting retention time flow-driven cheminformatics. BMC
in hydrophilic interaction liquid chromatogra- Bioinformatics 14:257
Part II
Methods
Chapter 5
HILIC-MS/MS Multi-Targeted Method for Metabolomics

Applications
Christina Virgiliou, Helen G. Gika, and Georgios A. Theodoridis
Abstract
Metabolomics aims at the identification and quantification of key-end point metabolites, basically polar, in
order to study changes in biochemical activities in response to pathophysiological stimuli or genetic modi-
fications. Targeted profiling assays have enjoyed a growing popularity during the last years with LC-MS/
MS as a powerful tool for development of such (semi-) quantitative methods for a large number of metab-
olites. Here we describe a method for absolute quantification of ca. 100 metabolites belonging to key
metabolite classes such as sugars, amino acids, nucleotides, organic acids, and amines with a hydrophilic
interaction liquid chromatography (HILIC) system comprised of ultra (high) performance liquid chroma-
tography (UHPLC) with detection on a triple-quadrupole mass spectrometer operating in both positive
and negative electrospray ionization modes.
Key words HILIC-MS/MS, Metabolic profiling, Targeted metabolomics, Polar analytes
1 Introduction
Metabonomics or metabolomics, often described as the holistic

metabolic profile of complex matrices such as biological fluids, tis-
sue, and cell extracts, represents with genomics and proteomics the
major platforms in system biology [1]. Developments in analytical
chemistry and particular advances in separation and spectroscopic
technologies made metabolomics a rapidly developed research field
over the past few decades [2, 3]. Although mass spectrometry and
NMR are the most popular analytical platforms used for
metabolomics-based studies, emphasis is lately placed on LC/MS
approaches due to a wide range of coverage of metabolites and its
high efficiency [4]. Typically MS-studies for metabolite profiling
can be categorized as targeted and untargeted/holistic approaches.
65
66 Christina Virgiliou et al.
Targeted analysis is usually a hypothesis-driven strategy. It focuses

mainly on the measurement (identification and quantification) of
selected metabolites with known chemical properties, thus sample
preparation can be adapted in order to minimize limitations
including matrix effect [5]. On the other hand, the holistic

approach investigates the whole metabolic complement of the ana-
lyzed sample. However, it has been realized that there are a num-
ber of issues that need to be resolved for untargeted MS-studies
including standardization, robustness, and reproducibility that
affect the validity of the results [6–8]. Limitations of MS-methods
and current developments on triple-quadrupole instrumentation
along with advanced software capabilities have resulted in the devel-
opment of tailor-made targeted metabolomics methods able to
(semi)-quantify tens of analytes of specific interest in a single injec-
tion and provide solid, quantitative, and unambiguous data
[9–11].
The development of a comprehensive method for targeted
metabolomics represents a challenge. The samples of interest often
contain highly polar metabolites with different physicochemical
properties, coexisting in samples over different concentration
ranges. Additional efforts are required for fine-tuning of all analyti-
cal parameters in order to find the optimum for most of the ana-
lytes measured within the method. According to the literature, a
number of multi-analytes methods use hydrophilic interaction liq-
uid chromatography (HILIC) separations for simultaneous analy-
sis of large numbers of polar metabolites [12, 13]. Reversed phase
(RP) chromatography is not able to retain polar metabolites, while
although the ion-pair chromatography has been reported as the
most powerful option ion-pair reagent contaminates the MS sys-
tem to a major extent [14, 15].
In the present protocol, a procedure for the identification and
quantification of ca. 100 metabolites via (HILIC) UPLC-MS/MS
method is described. The aim of the method is the identification
and quantitation of key-end point metabolites known to exist in
biological fluids (serum, plasma, urine, amniotic fluid) in order to
study their metabolic profile. Mass spectrometer and chromato-
graphic condition were optimized in order to achieve satisfactory
detection, quantification, maximum peak capacity, and retention
for as many metabolites as possible in a single run [16, 17].
Absolute quantitation can be performed using the standard addi-
tion approach. Matrix effect and recovery can be estimated, and the
sample preparation procedure is optimized in order to reduce
matrix effects particularly for blood samples.
HILIC-MS/MS Multi-Targeted Method for Metabolomics Applications 67
2 Materials
All solvents used should be of LC/MS analytical grade. Use puri-

fied water (18.2 MΩ, at 25 °C). All used standards should be of
analytical or higher grade. Stock and working standard solutions
should be kept at −20 °C, and solvents containing >50% of water
should be replaced after 20 days (see Note 1).
2.1 Stock Classify compounds in different concentration groups: group A,

and Working Standard group B, and group C (details in Table 1) (see Notes 2 and 3).
Solutions Stock solutions of the analytes should be prepared in concen-
trations of (group A) 1000, (group B) 5000, and (group C)
10,000 mg/L in methanol(MeOH)/water, 1:1 (v/v) or water,
depending on analyte solubility (see Note 4). Prepare working
standards mixtures from the stock solution by appropriate dilution
with acetonitrile(MeCN)/water, 95:5 (v/v).
Calibration standards (9 standards) should be prepared by
serial dilution of the highest concentration standard.
For the standard mixture with the highest concentration (std
9): Mix 0.0032 ml of each standard in group A with 0.052 ml of
each standard in group B and 0.152 ml of each standard in group
C. In order to reach the final volume of 8 ml dilute with 95:5
MeCN:H2O. Repeat this step depending on the desired final vol-
ume of std. 9.
2.2 Liquid Mobile phase A, stock buffer solution: In a 50 ml beaker, weigh

Chromatography 0.631 g ammonium formate and add H2O (<50 ml) (MilliQ).
Sonicate until salt is fully dissolved. Pour the sonicated solvent into
a 50 ml volumetric flask and add H2O MilliQ till the calibration
mark. Final ammonium formate buffer concentration: 200 mM.
Mobile phase A, 95:5 MeCN:H2O, 10 mM ammonium for-
mate: Pour the 50 ml buffer into a clean 1000 ml Schott bottle.
Add gradually 950 ml MeCN. Shake and sonicate for homogeniza-
tion and place in ultrasonic bath for degassing (see Note 5).
Mobile phase B, stock buffer solution: In a 50 ml beaker,
weigh 0.2254 g ammonium formate and add water (<250 ml)
(MilliQ). Sonicate until salt is fully dissolved. Pour dissolved buffer
into a 250 ml volumetric flash and add H2O MilliQ till the calibra-
tion line. Final ammonium formate buffer concentration: 14.2 mM.
Mobile Phase B, 30:70 MeCN:H2O, 10 mM ammonium for-
mate: With the use of a clean volumetric cylinder measure 210 ml
Table 1
Group concentration and nominal concentration of each group in standards 1–9 used for the construction of calibration curves
Group Group Group Group

Meatbolites concentration Meatbolites concentration Meatbolites concentration Meatbolites concentration Standard Mixture mg/L
2-Hydroxyisobutyric acid A Cytidine A Lactose B Sorbitol B STD 1 A 0.01

2-Hydroxyisovaleric acid A Cytosine A Leucine A Suberic acid B B 0.09
2-MethylHippuric acid A Dimethylamine A Lysine B Sucrose A C 0.475
3-Methylhistidine B Folic acid A Malic acid A Taurine B STD 2 A 0.02

4-Hydroxyphenyllactate B Fructose B Malonic acid B Theobromine A B 0.18
α-Ketoglutaric acid B Fucose B Maltose B Thiamine A C 0.95
Acetylcarnitine A Fumaric acid A Mannitol B Threonine B STD 3 A 0.08

Adenine A γ-Aminobutyric A Methionine A Thymidine A B 0.72
acid
Adenosine A Galactosamine A Methylamine A Thymine B C 3.8
Adipic acid A Galactose B Monoisoamylamine A Trimethylamine A STD 4 A 0.2

Alanine B Glucose C N-AcetylAspartate A Trimethylamine- A B 1.8
n-oxide
Arabitol B Glutamic acid B Nicotinamide A Tryptamine A C 9.5
Arginine A Glutamine B Nicotinic acid A Tryptophan B STD 5 A 0.5

Ascorbic acid A Glycine B Ornithine B Tyrosine B B 4.5
Asparagine B Guanine B Pantothenic acid A Uracil C C 23.75
Aspartic acid B Hippuric acid B Phenylalanine B Uric acid C STD 6 A 1

Benzoic acid A Histamine A Picolinic acid B Uridine A B 9
Betaine A Homocysteine B Proline B Valine B C 47.5
Biotin A Hypotaurine A Putrescine A Vitamin B12 A STD 7 A 1.2

Cadaverine A Hypoxanthine A Pyridoxine A Xanthine B B 10.8
Caffeine A Inosine A Pyroglutamic acid B Xanthurenic acid A C 57
Choline A Inositol B Pyruvic acid B Xylitol B STD 8 A 1.5

Cotinine A Isoleucine A Ribose B Xylose B B 13.5
Creatine A Itaconic acid A Riboflavine A C 71.25
Creatinine A Kynurenic acid A Sarcosine A STD 9 A 2

Cystine B Lactic acid C Serine B B 18
C 95
buffer and pour into a 500 ml Schott bottle. Measure 90 ml MeCN

and pour it gradually to the buffer. Shake and sonicate for homog-
enization and degassing.
Wash, weak solvent: Measure 420 ml water and pour into a
clean 1 L Schott bottle, measure 180 ml of methanol, and add to
water. Finally add 0.1% formic acid (600 μl).
Purge, strong solvent: Measure 380 ml acetonitrile and pour
into a clean 1 L Schott bottle, measure 20 ml of methanol and add
to water. Finally add 0.1% Formic acid (400 μl).
Seal wash solvent: Measure 900 ml of water and pour into a
1 L Schott bottle, measure 100 ml of MeCN, and add to water.
Mix thoroughly and degas shortly in ultrasonic bath.
Equipment and column: A Waters Acquity H Class UPLC sys-
tem coupled to Xevo TQD MS-spectrometer under control of
MassLynx 4.1 is used for the present protocol. In cases where
the sample volume is limited, total recovery vials (or similar type)
with pre-slit silicone septa screw caps 9 mm were used.
Chromatography is performed on an Acquity BEH Amide Column
(2.1 mm i.d. × 150 mm, 1.7 μm) protected by an Acquity UPLC
Van-Guard pre-column.
3 Methods
Methods for LC-MS/MS are prepared and stored in the centrally

shared hard drive.
3.1 UPLC Method With regard to UPLC conditions, the parameters for binary sol-
vent manager are as follows: flow rate is kept constant throughout
the whole analysis at 0.5 ml/min and the following gradient is
programmed: 4 min isocratic step at 100% A, then rising to 40% B
linearly over the next 21 min and finally reaching 85% B over
5 min. The column is equilibrated for 10 min in the initial condi-
tions. Regarding the sample manager, the flow through needle sys-
tem is applied in the present protocol. Injection volume is set at
5 μl and sample temperature at 6 °C. Injection system is subjected
to two washing cycles with a strong solvent and a weak solvent
prior to injection and one cycle of 6 s with the strong solvent for
post-wash. Column temperature is set to 40 °C.
3.2 MS Method In order to edit the MS method, find the optimum parameters for
the MRM transition of each metabolite. For this protocol, manual
optimization is performed in order to find precursor and product
ions and the optimum cone voltage and collision energy. With
regard to capillary voltage, the best possible for most metabolites

detected in positive and negative ionization mode is applied. All
the other parameters were set according to the tune page and the
linked calibration file.
Apply multiple reaction monitoring (MRM) mode for the
detection and quantification of all the compounds. Operate elec-
trospray ionization at polarity switching mode. Set capillary volt-
age at +3.5 kV or −3.5 kV, block and desolvation temperatures at
150 °C and 350 °C, respectively. Set desolvation gas flow rate at
650 L/h and cone gas at 50 L/h. Optimum cone voltage and col-
lision energy for each analyte after direct infusion and optimum
time window and dwell times are presented in Table 2.
System start-up and pretests.
Follow each step here to ensure high quality of your data. For
the system start-up, the system is prepared for analysis and checked
for system suitability.
1. Place beakers with mobile phase A and B, wash, purge, and seal
wash solvents to the corresponding channels and in the two left
channels, install neat acetonitrile and water.
2. Prime eluents for 3 min each.
3. Connect the BEH Amide column.
4. Set the column temperature to 40 °C and the temperature of
sample manager to 6 °C.
5. Flush the column with 95% MeCN and 5% water for 15 min at
flow of 0.2 ml/min.
6. Increase flow rate gradually to 0.5 ml/min.
7. Flush column until equilibration (psi delta <20) (see Note 6),
8. Load UPLC method.
For MS, load the appropriate Acquity DB file with the exten-
sion .ipr and the most recent calibration file.
3.3 Sample Extraction of samples may vary between different matrices. So far
Preparation the described method or its variants is tested with serum, urine,
amniotic fluid, intra-/extracellular content, feces, and various types
of animal tissue but also in foods such as honey, muscle tissue, and
flour. The sample preparation procedure for blood serum samples
is presented below.
1. Allow samples to thaw at room temperature.
2. Mix 50 μl of sample with 130 μl MeCN, 10 μl H2O, and 10 μl
MeOH in an eppendorf vial of 1 ml by the use of variable vol-
ume pipette 20–200 μl (see Note 7).
Table 2
Analytes that can be detected and monitored with the HILIC-MS/MS method
Cone
Monoisotopic Precursor Product voltage Collision Rt Molecular Dwell
A/A Metabolites Formula mass ion ion (V) energy (V) Polarity (min) weight time
1 2-Hydroxyisobutyric acid C4H8O3 104.04 103 57 30 10 − 8.0 104.10 0.005
2 2-Hydroxyisovaleric acid C5H10O3 118.06 117 71 30 12 − 6.0 118.13 0.005
3 2-MethylHippuric acid C10H11NO3 193.20 192 148 35 12 − 8.2 193.20 0.005
4 3-Methylhistidine C7H11N3Ο2 169.09 170 109 30 10 + 19.0 169.18 0.003
5 4-Hydroxyphenyllactate C9H10O4 182.05 181 63 33 12 − 10.0 182.17 0.02
6 a-Ketoglutaric acid C5H6O5 146.01 145 101 20 9 − 16.0 146.11 0.02
7 Acetylcarnitine C9H17NO4 203.12 204 85 30 10 + 14.4 203.23 0.003
8 Adenine C5H5N5 135.05 136 119 40 20 + 3.6 135.13 0.003
9 Adenosine C10H13N5O4 267.10 268 136 20 15 + 4.4 267.24 0.003
10 Adipic acid C6H10O4 146.06 145 101 25 12 − 16.0 146.14 0.005
12 Alanine C3H7NO2 89.05 90 44 20 10 + 16.0 89.09 0.005
13 Arabitol C5H12O5 152.07 151 89 25 10 − 9.9 152.14 0.02
14 Arginine C6H14N4O2 174.11 175 70 30 19 + 21.9 174.20 0.005
15 Ascorbic acid C6H8O6 176.03 176 70 20 15 + 3.6 176.12 0.02
16 Asparagine C4H8N2O3 132.05 133 74 20 14 + 18.2 132.11 0.02
17 Aspartic acid C4H7NO4 133.04 134 74 18 16 + 21.8 133.11 0.02
HILIC-MS/MS Multi-Targeted Method for Metabolomics Applications
18 Benzoic acid C7H6O2 122.04 121 77 25 11 − 1.8 122.12 0.032

19 Betaine C5H11NO2 117.07 118 59 38 18 + 12.5 117.15 0.005
(continued)
71
Table 2
72
(continued)
Cone
20 Biotin C10H16N2O3S 244.09 245 227 25 14 + 8.1 244.31 0.003
21 Cadaverine C5H14N2 102.12 103 86 15 8 + 20.4 102.17 0.003
22 Caffeine C8H10N4O2 194.08 195 138 38 18 + 0.9 194.19 0.032
Christina Virgiliou et al.
23 Choline C5H14NO 104.11 104 60 40 22 + 7.0 104.17 0.003

25 Cotinine C10H12N2O 176.09 117 80 30 20 + 1.1 176.22 0.032
26 Creatine C4H9N3O2 131.07 132 90 28 10 + 16.3 131.11 0.003
27 Creatinine C4H7N3O 113.06 114 88 30 10 + 4.8 113.11 0.003
29 Cystine C6H12N2O4S2 240.02 241 152 26 12 + 24.6 240.30 0.02
30 Cytidine C9H13N3O5 243.09 244 112 15 10 − 11.0 243.22 0.005
31 Cytosine C4H5N3O 111.04 112 95 40 14 + 7.5 111.10 0.003
32 Dimethylamine C2H7N 45.06 46 30 30 30 + 8.2 45.08 0.005
33 Folic acid C19H19N7O6 441.14 442 295 22 13 + 22.4 441.39 0.005
34 Fructose C6H12O6 180.06 181 140 5 8 + 11.8 180.16 0.01
35 Fucose C6H12O5 164.06 163 59 22 12 − 7.0 164.16 0.005
36 Fumaric acid C4H4O4 116.01 115 71 25 8 − 20.3 116.07 0.02
37 g-Aminobutyric acid C4H9NO2 103.06 104 69 22 15 + 17.3 103.12 0.005
38 Galactosamine C6H13NO5 179.07 180 72 18 18 + 17.8 179.17 0.01
39 Galactose C6H12O6 180.06 179 89 15 8 − 11.9 180.15 0.02
Cone
40 Glucose C6H12O6 180.06 179 59 25 16 − 14.6 180.15 0.02
41 Glutamic acid C5H9NO4 147.05 130 84 25 16 + 21.0 147.13 0.02
42 Glutamine C5H10N2O3 146.07 148 84 20 15 + 17.8 146.14 0.005
43 Glycine C2H5NO2 75.03 76 30 35 6 + 17.0 75.06 0.02
44 Guanine C5H5N5O 151.05 152 135 35 17 + 10.0 151.13 0.005
45 Hippuric acid C9H9NO3 179.06 178 134 32 11 − 9.4 179.17 0.005
46 Histamine C5H9N3 111.08 112 95 23 12 + 13.7 11.15 0.003
48 Homocysteine C4H9NO2S 135.03 134 88 10 8 − 16.4 135.19 0.005
49 Hypotaurine C2H7NO2S 109.02 110 92 22 18 + 15.8 109.14 0.015
50 Hypoxanthine C5H4N4O 136.04 137 110 40 18 + 4.8 136.11 0.003
51 Inosine C10H12N4O5 268.08 269 137 15 10 + 9.2 268.22 0.005
52 Inositol C6H12O6 180.06 181 109 15 10 + 17.6 180.16 0.02
53 Isoleucine C6H13NO2 131.09 132 86 25 12 + 13.3 131.17 0.003
54 Itaconic acid C5H6O4 130.03 129 85 20 8 − 14.4 130.09 0.02
55 Kynurenic acid C10H7NO3 189.04 190 172 32 12 + 9.8 189.16 0.005
56 Lactic acid C3H6O3 90.03 89 43 30 10 − 11.7 90.08 0.02
57 Lactose C12H22O11 342.12 343 163 10 10 + 18.5 342.30 0.02
58 Leucine C6H13NO2 131.09 132 86 20 10 + 13.4 131.17 0.003
59 Lysine C6H14N2O2 146.11 147 84 14 14 + 22.5 146.19 0.01
(continued)
73
Table 2
74
(continued)
Cone
60 Malic acid C4H6O5 134.02 133 115 22 10 − 20.2 134.08 0.02
61 Malonic acid C3H4O4 104.01 103 59 15 9 − 14.4 104.06 0.02
62 Maltose C12H22O11 342.12 341 161 25 8 − 18.1 342.30 0.02
63 Mannitol C6H14O6 182.08 183 69 15 11 + 13.5 182.17 0.02

65 Methionine C5H11NO2S 149.05 150 104 22 9 + 14.0 149.21 0.005
66 Methylamine CH5N 31.04 32 32 25 3 + 10.0 31.05 0.005
67 Monoisoamylamine C5H13N 87.10 88 43 18 11 + 4.6 87.10 0.003
68 N-AcetylAspartate C6H9NO5 175.04 176 134 18 10 + 20.1 175.14 0.005
69 Nicotinamide C6H6N2O 122.05 123 96 40 15 + 1.1 122.12 0.032
70 Nicotinic acid C6H5NO2 123.03 124 80 38 18 + 10.5 123.10 0.005
71 Ornithine C5H12N2O2 132.09 133 70 45 10 + 22.7 132.16 0.02
72 Pantothenic acid C9H16NO5 218.10 220 90 25 13 + 12.2 219.23 0.005
73 Phenylalanine C9H11NO2 165.08 166 120 22 12 + 12.6 165.19 0.003
74 Picolinic acid C6H5NO2 123.03 124 106 25 10 + 28.9 123.10 0.005
75 Proline C5H9NO2 115.06 116 70 20 20 + 14.5 115.13 0.02
76 Putrescine C4H12N2 88.10 89 72 15 8 + 21.0 88.15 0.003
77 Pyridoxine C8H11NO3 169.07 170 152 28 12 + 2.0 169.18 0.003
78 Pyroglutamic acid C5H7NO3 129.04 130 84 30 12 + 15.0 129.11 0.005
Cone
79 Pyruvic acid C3H4O3 88.02 89 48 20 12 + 7.1 88.06 0.02
81 Ribose C5H10O5 150.05 149 89 22 10 − 4.3 150.13 0.005
82 Riboflavine C17H2ON4O6 376.14 377 243 38 22 + 9.6 376.36 0.005
83 Sarcosine C3H7NO2 89.05 90 44 20 8 + 15.3 89.09 0.005
84 Serine C3H7NO3 105.04 106 60 20 10 + 17.9 105.09 0.005
85 Sorbitol C6H14O6 182.08 181 101 35 10 − 13.5 182.17 0.02
87 Suberic acid C8H14O4 174.09 173 111 35 12 − 10.7 174.20 0.005
88 Sucrose C12H22O11 342.12 341 179 40 14 − 16.8 342.29 0.02
89 Taurine C2H7NO3S 125.01 126 108 25 10 + 14.4 125.14 0.005
90 Theobromine C7H8N4O2 180.06 181 163 35 17 + 1.1 180.16 0.032
91 Thiamine C12H17N4OS 265.11 265 122 22 12 + 11.8 300.81 0.003
92 Threonine C4H9NO3 119.06 120 74 20 10 + 10.0 119.11 0.005
93 Thymidine C10H14N2O5 242.09 243 127 11 9 + 1.6 242.22 0.032
94 Thymine C5H6N2O2 126.04 127 110 40 19 + 1.2 126.11 0.032
95 Trimethylamine C3H9N 59.07 60 45 31 10 + 5.6 59.11 0.005
96 Trimethylamine-n-oxide C3H9NO 75.07 76 59 28 10 + 13.0 75.10 0.003
97 Tryptamine C10H12N2 160.10 161 144 15 11 + 4.1 160.21 0.003
98 Tryptophan C11H12N2O2 204.09 205 146 20 15 + 12.7 204.22 0.005
99 Tyrosine C9H11NO3 181.07 182 136 22 13 + 14.5 181.19 0.005
(continued)
75
Table 2
76
(continued)
Cone
100 Uracil C4H4N2O2 112.03 113 70 40 15 + 1.9 112.08 0.02
101 Uric acid C5H4N4O3 168.03 169 141 35 15 + 16.3 168.11 0.02
102 Uridine C9H12N2O6 244.07 243 110 35 16 − 4.7 244.12 0.005
103 Valine C5H11NO2 117.08 118 72 20 10 + 14.4 117.15 0.003

104 Vitamin B12 C63H89CoN14O14P 1355.58 678 147 42 27 + 19.4 1335.37 0.005
105 Xanthine C5H4N4O2 152.03 153 136 33 15 + 7.4 152.11 0.01
106 Xanthurenic acid C10H7NO4 205.04 206 188 35 12 + 12.8 205.17 0.02
107 Xylitol C5H12O5 152.07 151 89 30 10 − 9.9 152.15 0.02
108 Xylose C5H10O5 150.13 149 89 25 7 − 8.0 150.13 0.005
3. Vortex for 10 min and centrifuge at 5480 × g-force for 10 min.

4. Transfer supernatant to an LC-MS with glass insert vial.
5. Immediately transfer to precooled sample manager.
For standard addition approach, sample preparation of spiking

samples is as follows (see Note 8):
1. Mix 50 μl of sample with 100 μl of standard calibration mixture
1 and 50 μl of MeCN in an eppendorf vial of 1 ml by the use of
a variable volume pipette 20–200 μl.
2. Vortex for 10 min and centrifuge for 10 min.
3. Transfer supernatant to an LC-MS with glass insert vial.
4. Repeat steps 1–3 using increased concentration standard cali-
bration mixtures (at least 3 in total).
5. Immediately transfer to precooled sample manager.
In case of external calibration curve transfer standards to

LC-MS vial with glass insert (see Notes 9 and 10).
For quality control sample (QC sample), in order to evaluate
systems stability, mix equal volumes of all samples of the dataset
and follow the sample preparation procedure.
3.4 UPLC-MS/MS The following steps describe how to set up a sample table for data
Analysis acquisition. The following sample order has been evaluated to be
optimal for both throughput and quality controls:
1. Run a gradient without injection in order to evaluate column
performance and have a measure of minimum and maximum
pressure during analysis (see Note 11).
2. Perform six replicate injections of a standard mixture in order to
pretest systems performance (retention time, signal) and to
equilibrate system (see Note 12).
During analysis with external calibration approach:

1. Run a before-batch calibration curve.
2. Run injections of a QC sample in order to perform matrix equil-
ibration of the system. The number of equilibration injections
depends on the analyzed matrix. For cell media and urine five
injections may be adequate, for blood and tissue samples higher
numbers may be necessary, depending also on the use of the
column (new columns need more injections to saturate active
sites/equilibrate).
3. Samples are injected in random order in blocks of ten samples.

4. After each block of ten samples, injection of a standard mixture
and a QC sample is performed.
5. Repeat steps 2 and 3 for maximum 100 samples (see Note 13).
6. Run an after-batch calibration curve.
During analysis with standard addition approach

1. Run injections of a QC sample in order to perform matrix equil-
ibration of the system.
2. Samples followed by their paired spiking samples are injected in
order in block of ten samples.
3. After each block of ten samples, injection of a standard mixture
and a QC sample is performed.
4. Repeat steps 2 and 3 for maximum 100 samples.
After finishing sample batch analysis column and the MS have
to be cleaned for further use.
1. Flush the column with 50% MeCN and 50% H2O for 50 min,
column temperature 50 °C, flow 0.2 ml/min.
2. Flush the column with 95% MeCN and 5% H2O for 30 min,
column temperature 40 °C, flow rate 0.5 ml/min.
3. Clean MS cone as recommended.
3.5 Data Treatment- Quantitation can be performed in both manual and automated
Quantification ways using vendor software. In the present protocol TargetLynx
of Metabolites (Waters) is used. A quantify method must be created before inte-
gration or quantification can be performed. Related software from
other vendors includes MultiQuant (Sciex), Xcalibur (Thermo
Fischer), and MassHunter WorkStation-Quantitative Analysis
(Agilent).
TargetLynx data can be saved as .qld files for further manipula-
tion. Complete summary of the results and .qld files (area, response,
concentration, S/N, SD, measured concentration, etc.) can be
exported as .txt files and open with excel Microsoft program for
further treatment.
In case of external calibration curve approach, once you find
the optimum method parameters, calibration of standards and
quantification can be performed directly with only one process (see
Note 14).
When standard addition method is applied, automated quanti-

tation is not available. Integrated results for each sample and
spiked/fortified samples together with corresponding spiking lev-
els must be imported in spreadsheet program (excel or similar)
where unknown concentrations of metabolites in samples can be
calculated as intercept/slope.
4 Notes
1. This will help to avoid the growth of bacteria.

2. Classification of compounds in the current report is based on
their reference concentration in serum according to the litera-
ture and HMDB database.
3. Compounds are present in biological fluids at different con-
centration ranges so their grouping is important in terms of
quantitation.
4. In certain cases addition of minor amounts of base NaOH,
KOH (for riboflavine, uric acid, xanthine, threonine, aspar-
tate), or HCl acid (for 2-hydroxyisobutyric, inositol, 3-meth-
ylhistidine, cystine, xylose, tyrosine) and heating (for thymine,
uracil) and/or sonication (for cysteine) is needed to assist
dissolution.
5. Due to high concentration of buffer, addition of MeCN turns
the solvent cloudy. Ultrasonic at room temperature assists in
dissolvation.
6. Initial column pressure should not exceed 5300 psi; otherwise,
system may be overpressure when eluent B reach 85%.
7. The choice of extraction solvent is based on average solvent
content in MeCN, MeOH, and H2O of standard mixtures.
8. Since there is no available analyte free matrix, the standard
addition approach is performed in order to avoid limitations
such as matrix effect, in cases were absolute concentration is
required.
9. External calibration curve can be applied only when compari-
son between samples of the same or different batches and not
absolute quantitation is required.
10. For the external calibration standards, standard 9 should be
diluted 1:1 with 95:5 MeCN:H2O and then proceed to serial
dilutions.
11. Optimum column pressure ranges during the gradient: mini-

mum 5000–5300 psi, maximum 11,000–12,500 psi. In case of
higher pressures at the top end of the ranges clean the column
as recommended.
12. According to our group tests that have been performed for
system equilibration, four injections are the minimum for
retention time and signal stabilization for serum.
13. According to tests that have been performed for system perfor-
mance, 100 serum samples were considered as an upper limit
in order to avoid variations due to column and cone contami-
nations from the matrix.
14. Always use mean calibration curve (before and after-batch cali-
bration curve) for quantification. In cases when TargetLynx is
used, just select standard samples of both curves together with
real samples, QCs, and standard mixtures, and perform inte-
gration and quantitation by a single process.
References
1. Nicholson JK, Lindon JC, Holmes E (1999) 9. Griffiths WJ, Koal T, Wang Y et al (2010)
‘Metabonomics’: understanding the meta- Targeted metabolomics for biomarker discov-
bolic responses of living systems to patho- ery. Angew Chem Int Ed Engl 49:5426–5445
physiological stimuli via multivariate statistical 10. Michopoulos F, Whalley N, Theodoridis G
analysis of biological NMR spectroscopic data. et al (2014) Targeted profiling of polar intra-
Xenobiotica 29:1181–1189 cellular metabolites using ion-pair-high per-
2. Nicholson JK, Lindon JC (2008) Systems biol- formance liquid chromatography and -ultra
ogy: Metabonomics. Nature 455:1054–1056 high performance liquid chromatography
3. Nicholson JK, Connelly J, Lindon JC et al coupled to tandem mass spectrom. : appli-
(2002) Metabonomics: a platform for studying cations to serum, urine and tissue extracts.
drug toxicity and gene function. Nat Rev Drug J Chromatogr A 1349:60–68
Discov 1:153–161 11. Roberts LD, Souza AL, Gerszten RE et al
4. Theodoridis GA, Gika HG, Wilson ID (2011) Targeted metabolomics. Curr Protoc Mol Biol
Mass spectrometry-based holistic analytical 98:302.1–302.24
approaches for metabolite profiling in sys- 12. Gika HG, Theodoridis GA, Vrhovsek U
tems biology studies. Mass Spectrom Rev et al (2012) Quantitative profiling of polar
30:884–906 primary metabolites using hydrophilic inter-
5. Zhou B, Xiao JF, Tuli L, Ressom HW (2012) action ultrahigh performance liquid chro-
LC-MS- based metabolomics. Mol BioSyst matography-tandem mass spectrometry.
8:470–481 J Chromatogr A 1259:121–127
6. Fiehn O, Robertson D, Griffin J et al (2007) 13. Schiesel S, Lämmerhofer M, Lindner W
The metabolomics standards initiative (MSI). (2010) Multitarget quantitative metabolic
Metabolomics 3:175–178 profiling of hydrophilic metabolites in fer-
7. Theodoridis GA, Gika HG, Want EJ et al mentation broths of β-lactam antibiotics
(2012) Liquid chromatography-mass spec- production by HILIC-ESI-MS/MS. Anal
trometry based global metabolite profiling: a Bioanal Chem 396:1655–1679
review. Anal Chim Acta 711:7–16 14. Coulier L, Bas R, Jespersen S et al (2006)
8. Verpoorte R, Choi YH, Kim HK (2010) Simultaneous quantitative analysis of metabo-
Metabolomics: will it stay? Phytochem Anal lites using ion-pair liquid chromatography-
21:2–3 electrospray ionization mass spectrometry.
Anal Chem 78:6573–6582
15. Buescher JM, Moco S, Sauer U et al metabolomics applications. Electrophoresis

(2010) Ultrahigh performance liquid 36:2215–2225
chromatography–tandem mass spectrometry 17. Sampsonidis I, Witting M, Koch W et al
method for fast and robust quantification of (2015) Computational analysis and ratiomet-
anionic and aromatic metabolites. Anal Chem ric comparison approaches aimed to assist
82:4403–4412 column selection in hydrophilic interaction liq-
16. Virgiliou C, Sampsonidis I, Gika HG et al uid chromatography–tandem mass spectrom-
(2015) Development and validation of a etry targeted metabolomics. J Chromatogr A
HILIC-MS/MS multitargeted method for 1406:145–155
Chapter 6
Ion Pair Chromatography for Endogenous Metabolites

LC-MS Analysis in Tissue Samples Following Targeted
Acquisition
Filippos Michopoulos
Abstract
A protocol for the preparation of tissue extracts for the targeted analysis of ca. 150 polar metabolites,
including those involved in central carbon metabolism is described, using a reversed-phase ion pair U(H)
PLC-MS method. Data collection enabled by multiple-reaction monitoring provides highly specific, sensi-
tive acquisition of metabolic intermediates with a wide range of physicochemical properties and pathway
coverage. Technical aspects are discussed for method transfer along with the basic principles of sample
sequence setup, data analysis, and validation. General comments are given to help the assessment of data
quality and system performance.
Key words Metabonomics, Targeted analysis, Mass spectrometry, Metabolites, Ion pair
chromatography
1 Introduction
Late 1980s investigations that would lead to the formation of a

new omic technology, subsequently called metabonomics, was
established predominately based on 1H NMR spectroscopy [1, 2].
The unparalleled reproducibility and quantitative nature of NMR
spectroscopy, combined with its inherently high structural infor-
mation content in combination with pattern recognition infor-
matic tools opened up new horizons into understanding biological
phenomena. Despite the relative lack of sensitivity that limits the
use of the NMR technology to relatively highly abundant metabo-
lites, the technique remained the workhorse of the fast-growing
metabonomics field until early in the twenty-first century [3–7].
Advances in analytical instrumentation and particularly at the
interface between mass spectrometry with separation techniques
(liquid or gas chromatography, capillary electrophoresis, etc.)
raised scientific interest in these hyphenated technologies as they
83
84 Filippos Michopoulos
offered increased coverage of the metabolome and generally much

greater sensitivity than 1H NMR spectroscopy [8–12]. The oppor-
tunities offered by MS in combination with the plethora of avail-
able chromatographic substrates and formats open a new horizon
in metabolite detection producing information-rich data that
required the development of new bioinformatic tools to deconvo-
lute the resulting complex data sets and visualize metabolic pheno-
types [13–15]. Hypothesis-free, untargeted, metabolic profiling
approaches greatly benefited from these technological advances in
these hyphenated analytical tools which enabled the establishment
of differential metabolic phenotypes across many research areas.
However, successful deployment of LC-MS-based tools had to
overcome challenges such as structure elucidation, metabolite ID
determination, standardization, and reproducibility in untargeted
metabonomics.
In the last decade, an increasing number of publications have
reported the development of targeted methodologies for metabo-
lite detection [16–19]. These assays utilize tandem mass spectrom-
eters to selectively detect hundreds of known metabolic
intermediates and are reproducible, amenable to standardization
without requiring the often time-consuming structure elucidation
required by untargeted methodologies. The ability to customize
the selection of the metabolites to target specific biological process
and pathways (e.g., central carbon metabolism) has enabled ana-
lysts to better optimize chromatographic parameters and improve
the detection of highly polar metabolites such phosphorylated
compounds and carboxylic acids. The depth of information pro-
vided from such “tailor-made” approaches has been driven by the
need to address more mechanistic biological questions. In the case
of targeted approaches for polar hydrophilic analytes, reversed-
phase chromatography is poorly suited to their analysis, and chro-
matographic strategies employing hydrophobic interaction liquid
chromatography (HILIC) [19–21] and ion-pair chromatography
using [16–18] and to lesser extend derivatization protocols [22,
23] have been widely utilized.
Here we describe a targeted protocol using ion-pair chro-
matographic separation to resolve ~150 metabolic intermediates
on a C18 column. The protocol has been optimized to provide
semiquantitative data on tissue specimens although it has also
been used to analyze urine, plasma, serum, and samples obtained
from cell-based in vitro studies. The proposed methodology can
also be adapted to address the demands for absolute quantifica-
tion of analytes of particular interest, but for such a purpose, we
advise the development of a separate assay to quantify just these
metabolites.
Ion Pair Chromatography for Endogenous Metabolites LC-MS Analysis in Tissue… 85
2 Materials
Water of chromatographic grade purity (18.2 MΩ), organic sol-

vents (acetonitrile, methanol, isopropanol), and mobile phase
additives (acetic acid, tributylamine) of HPLC grade or higher
should be used. The same quality criteria must be applied to the
analytical standards to obtain the highest available purity.
2.1 Preparation For all the analytical standards listed in Table 1, the appropriate
of Analytical amount is weighed in a 1.5 mL Eppendorf tube to result in a 1 mL
Standards, Test 50 mM solution in MeOH/H2O 50/50 v/v (stock solution). For
Mixture, and Infusion amino acids and nucleotides, solubility is improved by the addition
Solution of small amounts of HCl, while other classes of metabolite may
require small amounts of 1 M NaOH. Stock solutions must be
diluted 1/100 v/v with HPLC grade water to produce a solution
(dilution A) with nominal concentration of 500 μM which will be
further diluted 1/50 with either HPLC water (dilution B) or
MeOH/H2O 50/50 v/v (infusion solution) to result final stan-
dard concentration of 10 μM. Stock solutions, dilutions A and B
should be transferred to 1.8 mL cryovials and stored at −20 °C as
reference material for future use.
An aliquot of 30 μL of the appropriate stock solution standards
(see Table 1) can be mixed in a 15 mL Falcon tube and concentra-
tion adjusted to 100μΜ with HPLC water before dividing into
10 μL aliquots (test mixture 1–6) and storage at −20 °C for use for
batch validation.
2.2 Solvent In a 15 mL Falcon tube, add 0.960 mL of acetic acid, 2.360 mL

Preparation for Liquid tributylamine, and 1.0 mL of MeOH. After a quick manual shake,
Chromatography the content of the tube is added to 990 mL of HPLC water in a
1 L solvent bottle, and then the falcon tube is rinsed with 5.68 mL
2.2.1 Mobile Phase A
of HPLC grade water before it is added to the mobile phase A
bottle. Shake mobile phase A bottle manually for 1 min to ensure
good solvent homogenization before connecting to the chromato-
graphic system.
2.2.2 Mobile Phase B In a 1 L solvent bottle mix 800 mL of MeOH with 200 mL of
isopropanol and manually shake for 1 min before connecting to
chromatographic system.
2.2.3 Syringe In a 0.5 L solvent bottle, mix equal volumes of isopropanol with
and Needle Wash acetonitrile and shake to ensure mixing.
2.2.4 Seal Wash Solvent In a 0.5 L solvent bottle, mix 400 mL of HPLC water with 100 mL
of isopropanol and shake to ensure mixing (see Notes 1–3).
2.3 Instrumentation Sample analysis accordingly is performed on Thermo Ultimate

3000 RS pump combined with an Ultimate 3000 autosampler
operating at 4 °C or instrumentation of similar specifications using
Table 1
List of metabolites measured with the current methodology, mass spectrometer parameters, and text
mixtures composition
Q1 mass Q3 mass RT DP EP CE CXP Test

Metabolite name (Da) (Da) (min) (V) (V) (V) (V) mixture
Pyruvic acid 87 43 6.4 −30 −10 −12 −1 1
Lactic acid 89 43 4.7 −40 −10 −19 −1 1
Glyoxylic acid 91 73 11.1 −18 −10 −13 −5 5
Malonic acid 103 59 9.7 −27 −10 −14 −5 3
Serine 104 74 0.8 −40 −10 −18 −5 1
Cytosine 110 67 0.8 −30 −10 −19 −5 4
Uracil 111 42 0.9 −26 −10 −30 −5 6
Creatinine 112 68 0.8 −55 −10 −25 −4 2
Proline 114 68 0.8 −55 −10 −18 −4 2
Maleic acid 115 71 9.9 −35 −10 −10 −3 3
Fumaric acid 115 71 11.3 −31 −10 −14 −5 4
Valine1 116 70 0.8 −70 −10 −15 −5 4
Succinic acid 117 73 9.9 −45 −10 −16 −5 1
Threonine 118 74 0.8 −45 −10 −15 −5 2
Benzoic acid 121 77 11.9 −30 −10 −19 −5 4
Nicotinic acid 122 78 9.5 −35 −10 −19 −5 3
Thymine 125.1 42 1.5 −35 −10 −22 −5 6
Pyroglutamic acid 128 84 5 −75 −10 −16 −5
Mesaconic acid 129 85 11.3 −35 −10 −12 −5 2
Itaconic acid 129.1 85 10.6 −33 −10 −15 −5 6
Leucine 130 84 1.3 −55 −10 −18 −3 4
Isoleucine 130 84 1.2 −55 −10 −18 −3 2
Creatine 130 88 0.8 −35 −10 −14 −4 4
Asparagine 131.1 95 0.8 −31 −10 −19 −5 4
Glutaric acid 131.1 87 10.4 −35 −10 −19 −5 5
Aspartic acid 132 88 3 −40 −10 −19 −5 4
Malic acid 133 115 10.7 −30 −10 −16 −9 2
Adenine 134 107 1.2 −70 −10 −25 −5 2
Salicylic acid 137.1 65 13 −48 −10 −41 −5 6
Nitrophenol 138.1 108 10.4 −59 −10 −20 −5 6
(continued)
Table 1 (continued)

α-ketoglutaric acid 145 57 11.1 −30 −10 −18 −7 1
Adipic acid 145 83 10.8 −35 −10 −18 −5 3
α-ketoglutaric acid b 145 101 11.1 −35 −10 −12 −7 1
Glutamine 145.1 127.1 0.8 −45 −10 −16 −5 5
Glutamic acid 146 102 2.6 −40 −10 −19 −5 5
Hydroxy glutaric acid 147 85 10.7 −40 −10 −22 −5 4
Citramalic acid 147 87 10.6 −50 −10 −20 −5 6
Methionine 148 47 0.9 −35 −10 −25 −7 4
Guanine 150.1 133.1 1.6 −45 −10 −20 −5 5
Hydroxy phenyl acetic acid 151 107 9.9 −30 −10 −10 −5 2
Xanthine 151 108 1 −50 −10 −23 −10
Histidine 154 93 0.8 −45 −10 −25 −4 4
Orotic acid 155 111 6.2 −30 −10 −18 −9 3
Pimelic acid 159 97 11.5 −40 −10 −20 −5 1
Pimelic acid 2 159 115 11.5 −40 −10 −20 −5 1
Indole-2-carboxylic acid 160.1 116.1 13.5 −40 −10 −20 −5 5
Mercapturic acid 162 84 9 −30 −10 −15 −10
Coumaric acid 163.1 119.1 10.8 −45 −10 −21 −5 6
Phenylalanine 164.1 147.1 2.5 −48 −10 −19 −5 6
Phthalic acid 165 121 12.5 −40 −10 −16 −5 6
3-(2-Hydroxyphenyl)propionic 165.1 121.1 13 −44 −10 −21 −5 5
acid
Methylxanthine 165.1 122.1 2 −47 −10 −27 −5 6
Quinolinic acid 166 122 11.9 −25 −10 −15 −5
Phosphoenolpyruvic acid 167 79 11.8 −35 −10 −20 −5 1
Glyceraldehyde-3-phosphate 169 97 11.4 −22 −10 −16 −5 5
Dihydroxyacetone phosphate 169 97 13.7 −40 −10 −15 −5
cis-Aconitic acid 173 85 11.9 −33 −10 −18 −5 3
Arginine 173.1 131.1 0.8 −40 −10 −19 −5 4
Shikimic acid 173.1 93 3.4 −40 −10 −25 −5 6
Citrulline 174.1 131.1 0.8 −30 −10 −21 −5 4
(continued)

Tyrosine 180 163 1.1 −45 −10 −20 −5 6
Sorbitol/mannitol 181 89 0.8 −58 −10 −19 −7 3
Pserine 184 97 6.7 −35 −10 −18 −5 1
Glycerate 3 phosphate 185 79 11.4 −35 −10 −44 −3 1
2-Trans-3-indolepyruvic acid 186.1 142.1 13.3 −30 −10 −29 −5 4
Kynurenic acid 188 144 11.8 −40 −10 −25 −9
Citric acid 191 111 12 −30 −10 −18 −7 5
Citric acid specific 191 87 12 −35 −10 −26 −5 5
Isocitric acid specific 191 73 12.1 −35 −10 −32 −5 3
Glucuronic acid 193.1 113 3.7 −31 −10 −19 −5 5
Ferulic acid 193.1 134 11.1 −46 −10 −22 −5 6
Tryptophan 203.1 116 3.8 −50 −10 −25 −5 6
Xanthurenic acid 204 160 11.5 −45 −10 −20 −10
Kynurenine 207 190 1.8 −40 −10 −12 −5
p-Creatine 210 79 10.5 −35 −10 −35 −5 2
Pantothenic acid 218 88 9.6 −46 −10 −20 −5 3
N-Acetylglucosamine 220.2 119 0.8 −44 −10 −11 −5 6
Dinitrosalicylic acid 227 183 14.9 −50 −10 −22 −5 5
Ribose 5 P 229 97 4.7 −50 −10 −22 −5 1
Ribulose 5 P 229 97 5.8 −50 −10 −22 −5 2
Xylulose 5 P 229 97 5.8 −50 −10 −22 −5 5
Melatonin 231 216 11.2 −65 −10 −22 −5
Cystine 239 120 0.8 −35 −10 −17 −5 2
Uridine 243 110 1.1 −60 −10 −25 −5 3
Palmitic acid 255.4 237.4 17.8 −80 −10 −29 −5
Glucosamine-6-P 258.1 97 0.8 −35 −10 −26 −5 5
Glucose-6-P 259 97 4.3 −55 −10 −27 −5 1
Mannose-6-P 259 79 4.4 −55 −10 −70 −5 2
Galactose-1-P 259 241 4.7 −55 −10 −20 −5 4
Fructose-6-P 259 169 4.9 −55 −10 −17 −5 6
Glucose-1- P 259 241 5.3 −55 −10 −20 −5 3
Fructose-1-P 259 97 6.1 −55 −10 −65 −5 5
(continued)
Table 1 (continued)

Glycerate 1,3 bisphosphate 265 167 13.3 −40 −10 −20 −5 1
Adenosine 266 134.3 3.15 −38 −10 −18 −12
Inosine 267 135 1.5 −80 −10 −37 −10 3
6-Phosphogluconic acid 275 79 11.3 −60 −10 −66 −5 1
Guanosine 282 150 1.6 −80 −10 −25 −5 3
Ophthalmic acid 288 195 6.4 −70 −10 −25 −5
Arginosuccinate 289 132 2.75 −50 −10 −26 −12
2′-Deoxycytidine monophosphate 306.1 79 7.6 −45 −10 −70 −5 5
Glutathione red 306.3 143.1 5.9 −50 −10 −26 −5 5
2′-Deoxyuridine monophosphate 307 111 8.9 −55 −10 −35 −5 1
Thymidine monophosphate 321 195 9.6 −55 −10 −25 −5 4
Cytidine monophosphate 322 97 6.6 −50 −10 −39 −5 3
Uridine monophosphate 323 97 7.9 −52 −10 −33 −5 3
Adenosine-3′5′- cyclic 328 134 10.3 −80 −10 −40 −5 1
monophosphate
2′-Deoxyadenosine 330.1 79 10 −55 −10 −60 −5 2
monophosphate
Fructose 1,6 bisphosphate 339 97 11.8 −60 −10 −28 −5 1
Guanosine-3′5′-cyclic 344 150 9.8 −67 −10 −39 −5 4
monophosphate
2′-Deoxyguanosine 346 79 9.3 −70 −10 −62 −5 6
monophosphate
Adenosine monophosphate 346.2 79 9.6 −70 −10 −62 −5 1
Inosine monophosphate 347 79 8.7 −55 −10 −85 −5 3
Guanosine monophosphate 362 79 8.7 −50 −10 −70 −5 3
S-5-Adenosyl-l-cysteine 369 282 1.3 −60 −10 −18 −5
S-5-Adenosyl-l-cysteine 1 369 134 1.3 −60 −10 −35 −5
Riboflavin 375.1 255.1 9.9 −55 −10 −26 −5 5
S-5Adenosyl-l-homocysteine 383 188 2.1 −40 −10 −24 −5
2′-Deoxycytidine diphosphate 386 79 11.4 −50 −10 −88 −5 2
(continued)
Table 1 (continued)

Thymidine diphosphate 401 79 11.9 −70 −10 −80 −5 3
Uridine diphosphate 403 159 11.5 −60 −10 −41 −5 3
2′-Deoxyadenosine diphosphate 410 79 12.1 −70 −10 −85 −5 4
Butyryl CoA 418 79 15.1 −50 −10 −80 −5 5
Acetoacetyl CoA 424.8 382.8 14.1 −35 −10 −16 −4 5
Isovaleryl CoA 424.9 134 15.6 −50 −10 −45 −5 2
Malonyl CoA 425.5 404 14.3 −28 −10 −12 −4 2
Adenosine diphosphate 426.1 79 11.9 −80 −10 −88 −3 2
Methyl malonyl CoA 432.5 410.5 14.4 −30 −10 −10 −5 4
Folic acid 440 311 11.9 −75 −10 −32 −7 3
Guanosine diphosphate 442 79 11.5 −66 −10 −87 −5 5
2′-Deoxycytidine triphosphate 466 79 13.2 −80 −10 −105 −5 4
2′-Deoxyadenosine diphosphate 467 159 13.4 −60 −10 −50 −9 5
Thymidine triphosphate 481 159 13.5 −70 −10 −48 −5 2
2′-Deoxyadenosine triphosphate 490 79 13.5 −75 −10 −105 −5 1
Adenosine triphosphate 506.1 79 13.4 −90 −10 −106 −1 4
Guanosine triphosphate 522.1 79 13.3 −70 −10 −105 −5 2
ADP ribose 558.1 346.1 11.6 −80 −10 −33 −5 4
UDP glucose 565 323 11 −80 −10 −33 −5 3
UDP glucuronic acid 579 403 13.1 −60 −10 −32 −5 3
Glutathione ox 611.6 306.3 10.4 −80 −10 −31 −5 6
Nicotinamide adenine dinucleotide 662.3 540.1 7.8 −50 −10 −22 −9 1
Nicotinamide adenine dinucleotide 664.3 79 12 −110 −10 −120 −3 2
reduced
Nicotinamide adenine dinucleotide 742.2 620.1 11.7 −60 −10 −22 −11 5
phosphate
Nicotinamide adenine dinucleotide 744.3 79 13.4 −110 −10 −118 −3 1
phosphate reduced
Coenzyme A (CoA) 766 79 14 −150 −10 −130 −5
Flavin-adenine-dinucleotide 784.5 97 13.3 −85 −10 −53 −5 6
Acetyl CoA 808.3 79 14.2 −85 −10 −53 −5 1
(continued)
Table 1 (continued)

Proprionyl CoA 822 79 14.6 −130 −10 −120 −5 6
Crotonyl CoA 834 79 14.4 −150 −10 −90 −5 4
Isobutyryl CoA 836.1 408 15.1 −155 −10 −55 −5 3
OH Butyryl CoA 852 79 14.1 −150 −10 −90 −5 6
Succinyl CoA 866.2 79 14.5 −130 −10 −120 −5 6
Hydroxymethyl glutaryl CoA 910.2 79 14.4 −150 −10 −120 −5 6
1
Valine very weak ionization in negative ESI
Ribose-5-P, ribulose-5-P, xylulose-5-P, glucose-6-P, and so forth correspond to ribose-5-phosphate, ribulose-5-
phosphate, xylulose-5-phosphate, glucose-6-phosphate
DP declustering potential, EP entrance potential, CE collision energy, CXP collision exit potential
a Acquity HSS T3 UPLC column (Waters Corp, 2.1 × 100 mm,

1.8 μm particle size) with column temperature maintained at
60 ± 0.5 °C. Spectrometric data is acquired on ABSCIEX 4000
triple quadrupole instrument operation in negative electrospray
ionization acquisition mode.
3 Methods
3.1 U(H)PLC Analysis is performed using a 5 μL injection of sample with gradi-

ent elution at a flow rate of 400 μL/min with a binary solvent
mixing schedule of Mobile phase B, over Mobile phase A: 0 min,
0% B; 0.5 min, 0% B; 4 min, 5% B; 6 min, 5% B; 6.5 min, 20% B;
8.5 min, 20% B; 14 min, 55% B; 15 min, 100% B; 17 min, 100% B;
18 min, 0% B; and 21 min 0% B.
3.2 Mass To obtain the optimum detection parameters, each metabolite

Spectrometry (infusion solution) is infused via a syringe pump (Harvard
Apparatus 22) at flow rate of 10 μL/min, and the optimized
parameters for declustering potential, entrance potential, collision
energy, and collision exit potential are given in Table 1. Interface
source parameters should be optimized at the flow rate (0.4 mL/
min) of the intended chromatographic separation. For this purpose
10 μL/min of each metabolite (dilution A) should be infused via a
syringe pump to a T-connector where it should be combined with
0.39 mL/min 90% mobile phase A via the UHPLC system deliver-
ing to the ESI source a final metabolite concentration of 12.5μΜ.
The optimal ESI source parameters should be as follows: ion spray
voltage −4.0 kV, temperature 550 °C, collision gas 5, curtain gas
30, ion source gas1 60, and ion source gas2 50. Gas values are
arbitrary units.
3.3 Tissue Extraction Tissue samples must be processed and extracted from frozen to
and Preparation reduce endogenous metabolite degradation. For the present pro-
tocol, we can use a combined extraction homogenization approach
that is performed using a Precellys 24 system with an attached tem-
perature control unit. For soft tissue samples, we advise the use of
CK14 tubes, while for harder tissue, the CK28R format is more
appropriate. Extraction and homogenization is achieved in the
appropriate tube format from a 50 mg of frozen tissue with 1 mL
ACN/MeOH/H2O 40/40/20 v/v/v. Tubes must be shaken at
5000 rpm for 20 s, and the process must be repeated three times
with an intermittent pause of 30 s between each repeat. While the
extraction and homogenization is taking place, the extraction
chamber unit must be at the lowest possible temperature and not
more than 10 °C. A clear supernatant is obtained after centrifuga-
tion at 10691 × g for 5 min at 0 °C, and this is transferred to cryo-
vial for storage at a minimum −20 °C until analysis. The
extraction-homogenization procedure is repeated with fresh sol-
vent as described above, and the resulted clear supernatant is com-
bined in the same cryovial.
3.4 Pre-Analytical Prior to LC-MS analysis, a minimum of a 1 in 10 dilutions with

Considerations HPLC water must be performed to avoid excess carryover of
highly abundant metabolites across injections. For different tis-
sues, an exploratory dilution experiment can be employed using
different dilution extract/total sample volume ratios (1/10,
1/20 v/v, etc.) to define the optimal dilution for a given study.
Samples after water dilution must be quickly vortexed and centri-
fuged at 2250 × g for 10 min at 4 °C.
For batch validation and metabolite confirmation, a pooled
sample (QC) is prepared by mixing equal aliquots of the individual
extracts, and this is treated the same way as the study sample
extracts. The QC sample is also used to condition the analytical
platform at the beginning of the sample sequence as well as being
injected at regular intervals during the analytical sequence to assess
metabolite analytical reproducibility.
Metabolite identification and retention time confirmation is
obtained by cross comparison between text mixture, spiked QC,
and individual samples. The test mixture is an aqueous sample con-
sisting of test mixtures 1–6 diluted 1/20 v/v with HPLC water.
The spiked QC sample is a test mixture sample prepared in the
matrix (QC) of interest.
3.5 Sample The sample sequence should be set up appropriately to provide a

Sequence clear measure of the data quality and system performance. Gradient
blank, solvent blank, test mixture, conditioning samples, quality
controls (QC), and spiked QC are additional to individual sample
injections that help in assessing the analysis quality. The use of

these injections and a brief description of the sample sequence are
given here:
1. A gradient blank at the beginning of the sample sequence helps
to identify any signals/impurities/contaminants related to the
solvents used for chromatography or prolonged carry-over in
the system. The gradient blank is an acquisition where no injec-
tion is performed and the chromatographic system runs the
gradient solvent schedule (see Notes 4 and 5). As contaminants
can be specific to different batches of solvent, this is an impor-
tant pre-analysis check as poor quality solvents can adversely
affect the analysis (see Notes 1–3).
2. The use of two solvent blank injections after the gradient blank
helps the identification of signals related to solvent in which the
samples have been reconstituted/diluted. Consecutive injec-
tions of solvent blank injections enable the assessment of the
severity and persistence of any carry-over.
3. System conditioning, which is essential to stabilize analyte
retention times, is performed with the same matrix as that of
the analysis sample set. The pooled sample, described in the tis-
sue extraction and preparation section above, is recommended
to be used unless volume limitations due to small sample size
prohibit its use. In these circumstances, a matrix sample of the
same origin from a different study can be substituted. System
conditioning improves reproducibility of detected signals and
the number of conditioning injections is matrix and instrument
saturation related. A minimum of 7–10 conditioning injections
is recommended (see Note 6).
4. Test mixture and spiked QC injections are the first steps of
batch quality validation. After system conditioning, two injec-
tions of the test mixture followed by a spiked QC injection
enables retention time confirmation and assessment of the over-
all system performance by examination of the resulting metabo-
lite resolution in the ion chromatogram. The first text mixture
injection does not always provide the optimal chromatographic
separation so two to three consecutive injections are
recommended.
5. Due to the relatively high metabolite concentrations (5 μΜ for
each) in both test mixture and spiked QC, which may result in
some modest carry-over on the following injection, a further
two to three system conditioning injections (as per point 3
above) are recommended to re-equilibrate the chromatographic
system for analysis of the sample set under investigation.
6. QC injections are the foundation of the statistical analysis and
the assessment of the analytical reproducibility for each metabo-
lite detected. The sample analysis should start and finish with a
QC injection to enable quality assessment at the beginning and

end of the individual sample analysis. A minimal of five QC
injections are required to obtain sufficient robustness for the
statistical analysis. Therefore QC injections are performed at
regular intervals, interspersed between the individual test sam-
ples (see Note 7).
7. Samples are analyzed in random order to reduce data bias, in
blocks of five to ten, sandwiched between two consecutive QC
injections. The number of injections in a sample set defines the
length of the sample block required to achieve the minimum of
five QC injections needed for robust statistical results.
8. Upon completion of the last QC injection, a spiked QC and a
test mixture injection should be obtained to assess any metabo-
lite retention time drift and intensity loss during the analysis
along the sample sequence.
9. Finishing the sample sequence with two injections of the sol-
vent blank and a blank gradient helps to explore carry-over
across the analytical batch as well as any increase or decrease on
the background signal (threshold) essential for peak
integration.
3.6 Data Analysis Raw spectrometric data are processed with MultiQuant software to
obtain peak areas for each of the detected metabolites across the
sample set. The first step prior to peak integration is the visual con-
firmation of the retention time for each metabolite peak by
comparing the trace obtained for the test mixture, spiked QC, and
the individual samples. Given the inability to obtain biological
matrix free from the endogenous metabolites being determined,
the solvent blank injection data can be used to define the back-
ground threshold values for each metabolite to ensure signals are
associated with actual metabolite presence and not the background.
Smoothing factor and peak splitting are two additional parameters
that must be customized using the sample injections in order to
obtain accurate and reproducible peak integration. The peak inte-
gration report can be exported as a text file from the MultiQuant
software and further processed in customer optimized visualization
and statistical software packages. For example, in Excel univariate
statistical analysis (f-test, t-test, ANOVA) can be performed to vali-
dated significant differences across meaningful group comparison
on metabolite level very simply. It is also essential to extend the
univariate analysis and combine with analytical reproducibility data
(coefficient of variation values, CV) to reinforce statistical signifi-
cance (see Note 8). Data normalization is also essential to reduce
bias and trends due to analysis order or loss of signal intensity since
no internal standards are used in this analysis protocol. Median
value normalization is highly regarded as an adequate approach for
this type of data set. Typical validation criteria for detecting
Fig. 1 Typical ion chromatogram of a text mixture injection obtained following the proposed chromatographic
separation
important differences in metabolite concentrations across two

groups of samples can be regarded as the complete accomplish-
ment of the following three: (1) absolute natural logarithmic fold
change >0.5, coefficient of variation <30%, and p-value < 0.05.
Figure 1 provides a representative ion chromatogram trace of a
text mixture injection following the gradient elution sequence of
the proposed protocol.
4 Notes
1. Prepare fresh mobile phase A before every analytical batch of

samples. This helps in reducing background signals as well as
avoiding bacterial growth in the aqueous LC solvent.
2. Replace mobile phase B every month to avoid retention time
shift due to evaporation-induced changes in solvent
composition.
3. Monitor pressure buildup across the analytical batch to avoid
deterioration of chromatographic performance.
4. Always prime solvent lines before commencing a new batch
analysis. This helps to reduce the formation of air bubbles in the
LC system.
5. Ensure enough needle and syringe wash is available to reduce
carry-over between sample injections.
6. Clean the ESI source after prolonged periods of use to restore
mass spectrometer sensitivity.
7. Data quality assessment is essential prior to data analysis steps.

Visual cross-comparison of ion chromatograms (overlay them)
of two test mix injections: one acquired at the beginning and
the second one at the end of the sample sequence must be uti-
lized to assess signal intensity loos or retention time drift and
reproducibility of the chromatographic resolution. Similar
examination must be performed using the first and last QC
sample injections. If signal intensity drops by more than 30%, it
is advised not to proceed further the data analysis. Similarly,
measures must be taken if retention time drifts more than 20 s
across the analysis time. Signal loss or retention time drift must
not be assessed using conditioning injection data.
8. Data visualization with multivariate statistical tools such as prin-
cipal component analysis (PCA) can be utilized to more com-
prehensively assess data quality by examining QC group
injections forming a tight cluster on the scores plot.
References
1. Bales JR et al (1998) Metabolic profiling of 8. Gika HG et al (2010) Does the mass spectrom-
body fluids by proton NMR: self-poisoning eter define the marker? A comparison of global
episodes with paracetamol (acetaminophen). metabolite profiling data generated simultane-
Magn Reson Med 6(3):300–306 ously via UPLC-MS on two different mass
2. Nicholson JK, Wilson ID (1989) High resolu- spectrometers. Anal Chem 82(19):8226–8234
tion proton magnetic resonance spectroscopy 9. Dunn WB, Ellis DI (2005) Metabolomics: cur-
of biological fluids. Prog Nucl Magn Reson rent analytical platforms and methodologies.
Spectrosc 21(4–5):449–501 TrAC Trends Anal Chem 24(4):285–294
3. Barzilai A et al (1991) Phosphate metabolites 10. Lenz EM, Wilson ID (2007) Analytical strate-
and steroid hormone receptors of benign and gies in Metabonomics. J Proteome Res
malignant breast tumors. A nuclear magnetic 6(2):443–458
resonance study. Cancer 67(11):2919–2925 11. Lindon JC, Nicholson JK (2008) Analytical
4. Gavaghan CL et al (2000) An NMR-based technologies for metabonomics and metabolo-
metabonomic approach to investigate the bio- mics, and multi-omic information recovery.
chemical consequences of genetic strain differ- TrAC Trends Anal Chem 27(3):194–204
ences: application to the C57BL10J and 12. Theodoridis G, Gika HG, Wilson ID (2008)
Alpk:ApfCD mouse. FEBS Lett LC-MS-based methodology for global metab-
484(3):169–174 olite profiling in metabonomics/metabolo-
5. Kurhanewicz J et al (1995) Citrate as an in vivo mics. TrAC Trends Anal Chem
marker to discriminate prostate cancer from 27(3):251–260
benign prostatic hyperplasia and normal pros- 13. Shulaev V (2006) Metabolomics technology
tate peripheral zone: detection via localized and bioinformatics. Brief Bioinform
proton spectroscopy. Urology 45(3):459–466 7(2):128–139
6. Lynch MJ, Nicholson JK (1997) Proton MRS 14. Smith CA et al (2006) XCMS: processing mass
of human prostatic fluid: correlations between spectrometry data for metabolite profiling
citrate, spermine, and myo-inositol levels and using nonlinear peak alignment, matching, and
changes with disease. The Prostat identification. Anal Chem 78(3):779–787
30(4):248–255 15. Xia J et al (2012) MetaboAnalyst 2.0—a com-
7. Marx A et al (1996) Determination of the prehensive server for metabolomic data analy-
fluxes in the central metabolism of sis. Nucleic Acids Res 40(W1):W127–W133
Corynebacterium glutamicum by nuclear mag- 16. Lu W, Bennett BD, Rabinowitz JD (2008)
netic resonance spectroscopy combined with Analytical strategies for LC–MS-based targeted
metabolite balancing. Biotechnol Bioeng metabolomics. J Chromatogr B
49(2):111–129 871(2):236–242
17. Buescher JM et al (2010) Ultrahigh perfor- hydrophilic metabolites in fermentation broths

mance liquid chromatography− tandem mass of β-lactam antibiotics production by HILIC–
spectrometry method for fast and robust quan- ESI–MS/MS. Anal and Bioanal Chem
tification of anionic and aromatic metabolites. 396(5):1655–1679
Anal Chem 82(11):4403–4412 21. Yuan M et al (2012) A positive/negative ion-
18. Michopoulos F et al (2014) Targeted profiling switching, targeted mass spectrometry-based
of polar intracellular metabolites using ion- metabolomics platform for bodily fluids, cells,
pair-high performance liquid chromatography and fresh and fixed tissue. Nat Protoc
and-ultra high performance liquid chromatog- 7(5):872–881
raphy coupled to tandem mass spectrometry: 22. Kloos D et al (2014) Analysis of biologically-
applications to serum, urine and tissue extracts. active, endogenous carboxylic acids based on
J Chromatogr A 1349:60–68 chromatography-mass spectrometry. TrAC
19. Gika HG et al (2012) Quantitative profiling of Trends Anal Chem 61:17–28
polar primary metabolites using hydrophilic 23. Kloos D et al (2012) Derivatization of the tri-
interaction ultrahigh performance liquid chro- carboxylic acid cycle intermediates and analysis
matography–tandem mass spectrometry. by online solid-phase extraction-liquid chro-
J Chromatogr A 1259:121–127 matography–mass spectrometry with positive-
20. Schiesel S, Lämmerhofer M, Lindner W (2010) ion electrospray ionization. J Chromatogr A
Multitarget quantitative metabolic profiling of 1232:19–26
Chapter 7
LC-MS Untargeted Analysis

Elizabeth J. Want
Abstract
LC-MS untargeted analysis is a valuable tool in the field of metabolic profiling (metabonomics/metabolomics),
and the applications of this technology have grown rapidly over the past decade. LC-MS offers advantages over
other analytical platforms such as speed, sensitivity, relative ease of sample preparation, and large dynamic range.
As with any analytical approach, there are still drawbacks and challenges to overcome, but advances are con-
stantly being made regarding both column chemistries and instrumentation. There are numerous untargeted
LC-MS approaches which can be used in this ever-growing research field; these can be optimized depending on
sample type and the nature of the study or biological question. Some of the main LC-MS approaches for the
untargeted analysis of biological samples will be described in detail in the following protocol.
Key words LC-MS, Untargeted, Mass spectrometry, Liquid chromatography, Metabolic profiling
1 Introduction
Metabolic profiling involves the measurement of low molecular

weight metabolites, typically <1 kDa in a biological sample, such as
a biofluid (e.g., urine, serum) or tissue. These measurements can
offer valuable insights into responses to therapeutic interventions,
disease diagnosis and progression, as well as the effects of ageing,
diet, and exercise on an individual [1–3].
The complexity of biological samples means that no single ana-
lytical approach will provide complete metabolome coverage.
Typically, a combination of NMR spectroscopy, GC-MS, and
LC-MS will be employed to enrich the metabolome information
recovered [4–6]. Improvements in instrument sensitivity and reso-
lution, as well as ever-growing databases for the structural identifi-
cation of metabolites, mean that greater metabolome coverage can
now be achieved [7].
LC-MS is a sensitive tool for metabolic profiling. Separation of
metabolites in a sample prior to mass spectrometric (MS) analysis
reduces the complexity of the sample and hence the risk of ion sup-
pression. Chromatographic separation can also aid in discriminating
99
100 Elizabeth J. Want
between many isobaric species, which, when combined with high

accuracy MS measurements, can improve considerably the struc-
tural elucidation of potential biomarkers. The breadth of LC col-
umn chemistries and types of mass spectrometers available mean
that LC-MS is a versatile technique for both untargeted metabolic
profiling and targeted metabolite analyses. Further, the advent of
ultra(high) performance liquid chromatography (UPLC or
UHPLC) has significant improved chromatographic resolution
and hence metabolite coverage [8–11]. Here, we will focus on the
application of LC-MS for the untargeted analysis of biofluids and
tissues.
Whereas targeted metabolite analysis focuses on a particular
class of molecules or pathway, offering sensitive, accurate, and
quantitative measurements [12, 13], untargeted LC-MS analyses
can be invaluable for hypothesis generation and biomarker discov-
ery [14]. The keys to these untargeted approaches are (a) unbiased
sample preparation, i.e., not favoring the extraction of a particular
type or class of molecule, and (b), where possible, unbiased LC-MS
analysis, where metabolites spanning a range of classes and polari-
ties can be separated and detected in one or a handful of chromato-
graphic analyses.
Typically, untargeted LC-MS studies employ reversed-phase
(RP) chromatographic techniques [15–19]. RP chromatography
enables the separation and detection of a range of moderately polar
to nonpolar metabolites. However, many biofluids or tissue samples
contain a plethora of more polar molecules, which will not be
retained and detected using RP approaches. Hydrophilic interaction
liquid chromatography (HILIC) has grown in applications and value
for studies including metabolic profiling [20, 21]. HILIC, akin to
normal phase chromatography, enables the separation and detection
of polar and nonpolar molecules, with greater retention of polar
molecules than RP technologies. Thus, HILIC can be considered a
complementary technique to RP chromatography and to this end is
often used in parallel for metabolic profiling studies.
Mass spectrometers of choice for untargeted LC-MS studies
are time-of-flight (ToF) or quadrupole-time-of-flight instruments
(Q-ToF). Advantages of these mass spectrometers include high
mass resolution and mass accuracy, sensitivity, fast scan speed, and
large dynamic range. The ability of Q-ToF mass spectrometers to
perform MS/MS experiments to aid in structural elucidation of
potential biomarkers makes them invaluable in untargeted meta-
bolic profiling.
Here, untargeted LC-MS protocols are described for the anal-
ysis of biological samples using both RP and HILIC chromatogra-
phy. Extraction of metabolites from different biological samples is
described. The analysis of tissue samples is covered in more detail
in Chapter 17. The importance and implementation of data quality
assessment, such as through the use of quality control samples, are
introduced but covered in more detail in Chapter 2.
LC-MS Untargeted Analysis 101
2 Materials
2.1 Samples In metabolic profiling, biological samples are commonly analyzed,

e.g., urine, serum/plasma, tissue, cell extracts, and cell media (see
Notes 1–3).
2.2 Standards 1. System suitability/test mix.

and Chemicals (a) It is advised to include a mixture of compounds, which can
be termed “system suitability mix or test mix” (see
Subheading 3.3) in order to assess chromatographic and
mass spectrometric performance. This can be commer-
cially obtained or designed in-house to encompass metab-
olite classes of interest.
2. Metabolite extraction solvents.
(a) Plasma/serum: methanol, acetonitrile.
(b) Urine (HILIC analysis): acetonitrile.
(c) Tissue: methanol/dichloromethane/MTBE/water.
3. UPLC-MS mobile phases.
(a) LC-MS grade water, acetonitrile, methanol, isopropanol—
exact compositions depend on nature of analysis (see
Subheading 3.3; LC gradients) (see Note 10).
(b) Mobile phase additives: formic acid, ammonium acetate,
ammonium formate.
4. Lockmass and calibration solutions: leucine-encephalin, sodium
formate.
2.3 Common 1. Pipettes.

Equipment 2. Pipette tips.
3. Eppendorf tubes.
4. Bead beater polypropylene tubes.
5. MS vials.
6. MS well plates.
7. Sealing cap mats.
8. Glass bottles.
2.4 LC-MS Systems 1. LC or UPLC chromatography system, e.g., Acquity Ultra

Performance LC.
2. Analytical columns—specific to chromatographic method, e.g.,
C18, C8, HILIC, HSS.
3. Mass spectrometer, e.g., Q-ToF.
2.5 Other 1. Benchtop centrifuge.

Instrumentations 2. Precellys bead beater or similar.
3. Vacuum evaporator.
4. Ultrasonic bath.
5. Vortex mixer.
2.6 Data Processing 1. Vendor-supplied software or freeware (e.g., XCMS, MZMine

Software 2) for preprocessing of raw data. This includes peak detection,
alignment, and normalization (Subheading 3.6).
2. R and associated software packages.
3. Excel or similar.
4. SIMCA, MATLAB, or similar for multivariate analysis.
5. PRISM or JMP or similar for univariate analysis.
3 Methods
LC-MS untargeted analysis can be divided into multiple key steps

(Fig. 1) as follows: sample preparation, chromatographic separa-
tion, and mass spectrometric detection, followed by data analysis.
Sample preparation to extract metabolites is dependent on sample
type. The most common samples used in untargeted LC-MS analy-
ses are biofluids, e.g., urine, serum, plasma, as well as tissue (liver,
heart, brain, etc.), cell extracts, and cell media. Metabolites are
extracted from these samples and the extracts analyzed by LC-MS
or UPLC-MS, often using a Q-ToF mass spectrometer, ideally
employing both positive and negative electrospray ionization. Key
to sample preparation is the removal of particulates and proteins
from the samples as these can affect chromatographic performance
and may even result in column blockage. Sample preparation for
untargeted LC-MS analysis should not favor specific types of mol-
ecules but rather should aim for a broad extraction of metabolite
classes, unless there is a specific interest in a class of molecules. It
may be that the preparation method extracts polar metabolites into
an aqueous fraction and nonpolar metabolites into an organic frac-
tion, which are then analyzed using separate chromatographic
methods, as in the case of tissue samples. This has benefits of reduc-
ing sample complexity and may improve detection, quantification,
and identification.
Important aspects of an untargeted LC-MS analysis are:
System Suitability or Test Samples: Typically, 10–15 compounds
will be used in a system suitability mix, which will be run at the
start of the analysis (see step 8, Subheading 3.3) to assess instru-
ment performance (e.g., chromatographic peak shape and reten-
tion time, mass accuracy, and detector response) before the analysis
of study samples. Parameters such as detector response are prone
Fig. 1 LC-MS untargeted workflow showing the main steps of the process. These steps are described in this
protocol
to change over the analytical run, and so this mix can also be run
at the end of the sample analysis to assess these changes. For a
more detailed description, the reader is directed to Chapter 2.
Quality Control Samples: Quality control (QC) samples are the
key to successful, robust untargeted LC-MS analyses. These sam-
ples are representative of the study sample set and are usually made
by mixing small aliquots (e.g., 10–50 μL) from all study samples.
QC samples are used for (a) conditioning of the LC column and
(b) assessment of data quality. Subheading 3.3 step 9 describes the
setup for QC samples in the run. For a more detailed description,
the reader is directed to Chapter 2.
Sample Randomization: Study samples should be randomized
within a batch to avoid bias which may arise from changes in the
system over the course of the run, e.g., decrease in detector sensi-
tivity and drifts in chromatographic retention times or mass accu-
racy. Samples can be randomized using in-house scripts, online
software (https://www.randomizer.org/; https://www.random.
org/lists/), or randomized block design in the case of larger stud-
ies. For a more detailed description, the reader is directed to
Chapter 2.
3.1 Sample 1. Urine [22]

Preparation
(a) RP LC-MS:
–– Aliquot an appropriate volume of urine, e.g.,
50–100 μL, into a labeled Eppendorf tube.
–– Dilute urine sample with an appropriate volume of

LC-MS grade water (1:1 for human samples, 1:3 for
rodent and canine samples).
–– Centrifuge at 10,000 × g for 10 min to remove any
precipitate.
–– Aliquot sample into clean Eppendorf tube.
–– Prepare quality control sample as described in step 5.
–– Aliquot an appropriate volume of sample into well
plate or MS vials.
(b) HILIC LC-MS:
–– Aliquot appropriate volume of urine, e.g., 50–100 μL,
into a labeled Eppendorf tube.
–– Dilute urine sample with appropriate volume of aceto-
nitrile (1:1 for human samples, 1:3 for rodent and
canine samples).
–– Centrifuge at 10,000 × g for 10 min to remove
precipitate.
–– Aliquot sample into clean Eppendorf tube.
–– Prepare quality control sample as described in step 5.
–– Aliquot appropriate volume of sample into well plate
or MS vials.
2. Serum/plasma
(a) The key to extracting metabolites from serum/plasma is
the precipitation of proteins. Here, a simple method involv-
ing protein precipitation with methanol is suggested [23].
Method 1: Preparation for RP LC-MS using methanol
extraction
(b) Aliquot an appropriate volume of serum/plasma into a
labeled Eppendorf tube.
(c) Add cold methanol in a ratio of 3:1, e.g., 150 μL metha-
nol:50 μL plasma.
(d) Leave at −20 °C for at least one hour; it is acceptable to
leave samples at this temperature overnight.
(e) Centrifuge at 10,000 × g for 10 min.
(f) Remove supernatant.
(g) Dry in vacuum concentrator.
(h) Resuspend in water or RP starting mobile phase (see Note 6).
3. Tissue sample preparation for RP and HILIC analysis [18, 24, 25]
(a) Weigh 50 mg (+/− 5 mg) tissue section and place in a)
2 mL polypropylene bead-beater tubes if using a Precellys
bead beater (or similar) or b) 2 mL Eppendorf tubes if
using a Qiagen tissue lyser (or similar). Keep tubes on ice

before, during, and after the extraction process.
(b) Homogenize the sample with 1.5 mL prechilled metha-
nol/water (1:1) using either a Precellys bead beater or
Qiagen tissue lyser following the instructions below. The
number of bead beating cycles will vary depending on the
tissue type (see Notes 4–5).
Method Requirements Instruction

Precellys bead beater Add 1 mm zirconium 40 s per cycle, cool on
beads—100 μL in tube ice; 40 s for
subsequent cycles
Qiagen tissue lyser Stainless steel beads 25 Hz speed, 5 min
cycle
(c) Centrifuge the sample at 4 °C for 10 min at 10,000 × g.

(d) T
o prepare the aqueous extracts, aliquot the supernatant
into a clean Eppendorf tube. Retain the pellet in the bead-
beater tube on ice for subsequent organic extraction (see
step 9) (see Note 7).
(e) Take separate 250 μL aliquots for RP and HILIC LC-MS
analyses.
(f) Take 150 μL of each sample to form the QC sample.
(g) Dry aqueous extracts in a Savant vacuum concentrator for
~180 min at 45 °C using the V-AQ (vacuum aqueous) mode.
(h) The sample can then be treated in one of the following
ways:
–– Resuspension in starting mobile phase or similar and
immediate analysis.
–– Storage at −40 °C or lower until analysis.
(i) To prepare the organic extracts, add 1.5 mL prechilled
solution of dichloromethane/methanol (1:3) to the pellet
from step 4.
(j)
Homogenize the sample using either a Precellys bead
beater or Qiagen tissue lyser, following the instructions in
step 2 (see Notes 8 and 9).
(k) Take separate 250 μL aliquots for RP (lipid) and HILIC
LC-MS analyses.
(l) Take 150 μL of each sample to form the QC sample.
(m) Store at −40 °C or lower until analysis.
3.2 Preparation Prepare a fresh stock solution of sodium formate; 0.1 mg/mL in
of LC-MS Calibration water. Add 1 mL of stock solution to 9 mL of isopropanol (IPA) to
Solution obtain a working solution of 0.01 mg/mL. This solution can be
stored at 4 °C for a number of weeks. Note that an alternative cali-

bration solution can be used instead, and many instrument vendors
may recommend one.
3.3 LC-MS Analysis 1. Transfer samples to LC-MS vials or polypropylene 96-well

plates. If samples have been stored in vials or well plates prior
to analysis, centrifuge at 1350 × g for 5 min at 4 °C (ideally)
or room temperature.
2. Place vials (see note 14) or well plates in the LC autosampler.
The autosampler temperature (see Notes 12 and 13) should
be maintained at 4 °C for all analyses except for lipid analysis,
where the temperature should be 8–10 °C to avoid precipita-
tion of the lipids. Samples should be stable in the autosampler
for 24–48 h (see note 15), but the stability of individual
metabolites over this time period will naturally vary. Stability
of metabolites can be assessed in the different biological sam-
ple types, e.g., by using the QC samples.
3. Create the sample analysis list. Choose the LC gradient and
MS settings as described in Subheadings 3.3 and 3.4, respec-
tively. Column temperature will be dependent on the analysis
and will vary from 40 °C for analysis of aqueous metabolites in
urine samples to 55 °C for lipid analysis. Column temperature
has been optimized based on the mobile phases employed for
the LC gradient.
4. Select electrospray ionization mode.
5. Ensure that the instrument is set up for the optimum accurate
mass by infusing a solution of a reference compound, such as leu-
cine encephalin into the mass spectrometer. Follow the instru-
ment-specific instructions for setup and acceptance criteria.
6. Calibrate the instrument prior to analysis using a solution of
the appropriate concentration of sodium formate (Subheading
3.2) or similar calibration solution. Follow the instrument spe-
cific instructions for calibration and acceptance criteria.
Calibration points should be spread as evenly as possible over
the mass range that is being acquired for the analysis (typically
50–1000 or 1200 m/z for metabolic profiling). In general,
the residual mDa for each individual calibration point should
be <1.5 mDa, with the majority of calibration points having
residuals of <0.5 mDa. With most instruments, this can be
performed using a setup wizard or manually depending on the
preferences of the instrument operator.
7. Start with a blank sample to assess the condition of the column.
This blank sample can be water or a solvent mixture similar to
the starting mobile phase conditions. If the LC column appears
“dirty,” i.e., shows a high background signal, then a sequence
of blank samples is recommended. Background signal levels
can be monitored until an acceptable level is reached.
8. Run the system suitability/test mix to assess peak intensity,

retention times, and mass accuracy.
9. Inject quality control (QC) samples (see Note 11) in order to
condition the column before study sample analysis. At least 10
QC samples are recommended for RP analysis of urine and
plasma/serum. This may need to be increased to ~15 QC
samples for RP analysis of tissue samples and for HILIC analy-
sis in general. These QC samples can be assessed by visually
overlaying and monitoring peak intensity, retention times, and
mass accuracy. Conditioning QC samples can also be used to
assess the appropriate injection volume needed in order to
obtain peaks that are not overloaded and, in the same manner,
sample dilution factors. An example of overlaid QC sample
chromatograms is shown in Fig. 2.
10. In addition, set up untargeted MS/MS experiments, e.g.,
DDA experiments using at least one conditioning QC sample
(see Subheading 3.5).
11. Randomize the study samples, add to the sample analysis list,
and inject a QC sample every 5–10 samples. Aim for at least
10 QC samples per study batch in order to be able to perform
appropriate statistics. End the sample batch with a QC sample,
followed by a blank sample to assess carryover and a final sys-
tem suitability/test mix injection to assess system changes
over the run.
12. Injection volume will depend on sample type and instrumen-
tation but usually ranges from 2 to 10 μL.
13. When satisfied with the performance of the system, using the
system suitability mix and the conditioning QC samples, start
the analysis of the study samples, including within run QC
samples.
3.3.1 LC Gradients 1. Examples of UPLC gradients for urine, serum, and tissue sam-
ples by both RP and HILIC analysis are shown in the following
tables: Tables 1a, 1b, 1c, 1d, 1e, 1f, and 1g. The column dimen-
sions and details are also provided. Note that these gradients
will require optimization for different column chemistries,
lengths, and particle sizes.
3.4 MS Settings Example MS settings, including cone and capillary voltage, are
shown in Table 2 below. Note that these settings are for Waters
Xevo Q-ToF mass spectrometers and will require optimization for
each specific mass spectrometer. However, they can be used as
guidelines, based on an ultra-performance liquid chromatography
(UPLC) setup with a flow rate of 400–500 μL/min and a Q-ToF
mass spectrometer.
Fig. 2 Selected portion of a BPI chromatogram showing serum QC samples overlaid. Inset is an example of
monitoring ion intensity of a specific metabolite over the QC samples
Table 1a
Example gradient for the RP analysis of urine samples
Time (min) A (%) B (%) Comment

0 99 1
1 99 1
3 85 15
6 50 50
9 5 95
10 5 95 Column washing
10.1 99 1 Column
re-equilibration
12 99 1
Column = UPLC Acquity HSS T3. Flow rate = 0.5 mL/min. Mobile phases; A = 0.1%
formic acid in water, B = 0.1% formic acid in acetonitrile. Taken from Ref. 21
3.5 MS/MS DDA experiments—data-dependent or data-directed experiments:

Experiments Tandem mass spectrometry experiments can be set up, where a
specific number of precursor ions are selected using predetermined
rules and thresholds, such as exceeding a certain ion intensity.
Table 1b
Example gradient for the HILIC analysis of urine samples

0 99 1
1 99 1
12.1 99 1 Column
re-equilibration
15 99 1
Column = UPLC Acquity HILIC. Flow rate = 0.4 mL/min. Mobile phases; A = 95%
acetonitrile, 5% water + ammonium acetate (10 mM final concentration), B = 50% ace-
tonitrile, 50% water + ammonium acetate (10 mM final concentration). Taken from
reference 21
Table 1c
Example gradient for the RP analysis of plasma/serum samples

0 99.9 0.1
2 99.9 0.1
6 75 25
10 20 80
12 10 90
21 0.1 99.9 Column washing
23 0.1 99.9
24 99.9 0.1 Column
re-equilibration
26 99.9 0.1
formic acid in water, B = 0.1% formic acid in methanol. Taken from reference 22
These MS/MS experiments can provide information which may

aid in the structural identification of discriminatory metabolites.
Although it is often the case that further MS/MS experiments
need to be performed, once such potential biomarkers have been
selected from data analysis, performing DDA experiments at this
time does not impact negatively on the study setup.
Table 1d
Example gradient for the HILIC analysis of plasma/serum samples and
aqueous tissue extracts

0 99 1
2 99 1
8 45 55
9 1 99
9.1 1 99 Flow rate increased to
0.8 mL/min
11.1 99 1 Column re-equilibration
19 99 1
19.1 99 1 Flow rate decreased to
0.4 mL/min
23 99 1
Flow rate = 0.4 mL/min unless otherwise stated. Column = UPLC Acquity
HILIC. Mobile phases; A = acetonitrile/water (95:5), B = acetonitrile/water (50:50).
Taken from reference 25
Table 1e
Example gradient for the RP analysis of tissue samples—aqueous extract

0 99.9 0.1
2 99.9 0.1
6 75 25
10 20 80
12 10 90
21 0.1 99.9 Column washing
23 0.1 99.9
24 99.9 0.1 Column
re-equilibration
26 99.9 0.1
formic acid in water, B = 0.1% formic acid in methanol. Taken from reference 23
Table 1f
Example gradient for the RP analysis of tissue samples—organic extract

0 60 40
2 57 43
2.1 50 50
12 46 54
12.1 30 70
18 1.0 99 Column washing
18.1 60 40 Column
re-equilibration
20 60 40
Column = UPLC Acquity CSH. Flow rate = 0.4 mL/min. Mobile phases; A = acetoni-
trile (ACN)/water (60:40); B = isopropanol/ACN (90:10). In both mobile phases
ammonium formate was diluted to 10 mM and formic acid to 0.1%. Taken from refer-
ence 25
Table 1g
Example gradient for the HILIC analysis of tissue samples—aqueous
extracts

0 99 1
2 99 1
8 45 55
9 1 99
9.1 1 99 0.8 mL/min column
washing
11 1 99 0.8 mL/min column
washing
11.1 99 1 0.8 mL/min column
washing
19 99 1 Column re-equilibration
23 99 1
Column = Acquity HILIC.Temperature = 35 °C. Flow rate = 0.4 mL/min unless oth-
erwise stated. Mobile phases; A = acetonitrile (ACN)/water (95:5); B = ACN/water
(50:5). In both A and B, the concentration of ammonium acetate is 10 mM and formic
acid is present at 1%. Taken from reference 25
Table 2
Example MS settings for positive and negative mode ESI analysis using a
Q-ToF mass spectrometer. The main parameters are shown. These will
need to be optimized for each instrument and ESI mode and are to some
extent dependent on solvent system and mobile phase flow rate
Parameter Setting
Capillary voltage 1–3 kV electrospray
(ESI)+,1–2.5kVESImode-
Cone voltage 30 V
Source temperature e.g., 120 °C
Desolvation temperature e.g., 350 °C
Cone gas flow 25 L/h
Desolvation gas flow 900 L/h
3.6 Data It is beyond the scope of this protocol to provide a detailed descrip-
Preprocessing tion of data analysis. However, the key data analysis steps pertain-
ing to LC-MS untargeted analysis for metabolic profiling studies
are described briefly in steps 1–7. There are many different com-
mercial and freeware available to perform some or all of these steps.
These are listed in the Materials section of this protocol.
1. The first step in data preprocessing of untargeted LC-MS data
is peak picking. Define chromatographic processing regions—it
may be that it is not desirable to include the solvent front, e.g.,
0–1 min, and the re-equilibration portion of the data. Some
software may allow the user to omit these regions from further
analysis.
2. This is followed by alignment of the peaks to correct for any
retention shifts that have occurred between samples during the
analysis.
3. Peaks are then integrated using peak area through a selected
algorithm depending on the software used.
4. Data is often normalized during this process. A common
approach is to use median fold change [26] to account for dif-
ferences in sample dilutions. These are particularly common in
biological samples such as urine samples.
5. QC filtering is key in untargeted LC-MS metabolic profiling
studies. This is explained in Subheading 3.7.
6. Output of metabolite feature table for multivariate analysis.
3.7 Data Analysis Once the metabolite table has been produced, this can be exported
into appropriate software, e.g., excel for further analysis, e.g., QC
CV filtering. This will be covered in more detail in Chapter 2.
3.8 Database Still a bottleneck in LC-MS untargeted analysis, structural identifi-

Searching cation is key in metabolic profiling studies. Databases such as
for Structural HMDB (http://www.hmdb.ca/), METLIN (https://metlin.
Identification scripps.edu/index.php), and LIPIDMAPS (http://www.lipid-
maps.org/) are growing in size as research groups and individual
researchers add to them. These databases can be searched in a few
ways in order to attempt to identify candidate biomarkers, such as
MS searches, MS/MS searches, or text searches. Readers are
directed to reviews in this area [27, 28].
The standard way to proceed with structural identification is to
carry out MS/MS experiments in order to produce fragmentation
patterns. This will provide additional high mass accuracy informa-
tion to that obtained from MS spectra alone. These MS/MS spec-
tra can be used to search databases and will enable the list of
potential candidates to be narrowed significantly from solely using
MS data. Untargeted MS/MS data (e.g., DDA experiments) can
be collected on QC conditioning samples or any QC sample
injected throughout the run, although it would be recommended
to use a sample before or after the main study sample analysis.
Targeted MS/MS is usually performed on selected samples con-
taining the metabolite feature(s) of interest in high abundance.
These targeted assays would be performed after univariate or mul-
tivariate analysis has been performed and potential biomarkers
selected. Many of the commonly used databases allow for upload-
ing of MS/MS fragment ion information, e.g., HMDB and
METLIN. HMDB will output predicted MS/MS spectra which
can be visualized against the experimental MS/MS spectra obtained
during a study.
4 Notes
1. Exercise caution when preparing samples from humans, as there

is a risk of infection. Samples should be handled and prepared
according to appropriate biosafety protocols.
2. Tissues should be collected onto ice and snap frozen as soon as
possible after collection or processed for analysis immediately.
This is to avoid further metabolism occurring or metabolite
losses.
3. Note that tissue samples can be prone to contamination from
sources such as anesthetics, surgical tubes, or instruments, e.g.,
cleaning solutions.
4. Tissue samples should be stored at the lowest possible tempera-

ture, ideally −80 °C but otherwise −40 °C. This will help to
minimize further metabolite losses.
5. At the time of collection, sub-aliquot samples, including tissue
samples, are placed into appropriate storage containers to avoid
freeze-thaw issues.
6. Exercise caution when handling solvents for sample preparation
and the preparation of mobile phases.
7. For the extraction of tissue samples, the volume of extraction
solvent stated in this protocol was optimized for 50 mg
(+/−5 mg) of tissue. This volume can therefore be scaled down
if the amount of tissue is less. It must be noted that the bead-
beater tubes or Eppendorf tubes are 2 mL, and so if >1.5 mL of
liquid is placed into them along with the tissue sample, there is
a risk of leaking through the lid of the tube during the extrac-
tion process.
8. The number and speed of the bead beating cycles can be
adjusted according to the nature of the tissue. For example,
fibrous tissue such as veins and placental samples may need
more cycles operated at a higher frequency, e.g., three cycles at
6500 rather than 4000 mHz.
9. There is a risk with the addition of stainless steel beads to
organic solvents in Eppendorf tubes that the tubes can degrade
during the tissue lysis process, resulting in damage and/or sam-
ple leakage.
10. When preparing the mobile phases, it is important to note that
the addition of salts, particularly to mobile phases with a high
organic content, will require more careful preparation. This is
the case for both HILIC and lipid mobile phases. The key to
successful preparation is to:
(a) Prepare in advance, e.g., the day before sample analysis
commences.
(b) Sonicate the bottles containing the mobile phases during
preparation to ensure salts have dissolved, ideally at
>40 °C. Allow the mobile phases to sit at room temperature
for at least 1 h before use; check no precipitation has occurred.
11. QC Samples. Alternatively, or perhaps in some cases, addition-
ally, a commercially purchased external QC sample could be
used. This has been used in the case of serum[15]. Although this
sample would not be representative of the study samples, it
offers value in terms of conditioning the column and assessing
instrument drift.
12. Column temperature must be adjusted depending on the
mobile phase composition—mixtures with methanol results in
higher viscosity than when using acetonitrile and therefore

higher backpressure. Increasing column temperature, e.g., to
50–55 °C, will help to reduce column backpressure.
13. When storing and analyzing organic extracts, 96-well plates
are best avoided as there is the potential of contaminants
leaching into the solvent and affecting the analysis. It is advised
to store the samples in glass MS vials.
14. When using MS vials, where possible, use vials from the same
batch (there are 100 vials in a box) as there may be differences
between the batches in terms of purity, which can affect the
analysis.
15. With larger studies, e.g., >100 samples, the analytical run time
is likely to be more than 24 h. Therefore, samples may be in
the autosampler at 4 °C for up to a few days, resulting in
potential degradation of less stable metabolites. It is important
that sample stability is assessed during analysis through the use
of quality control samples and perhaps also the system suit-
ability or test mix.
References
1. Holmes E, Loo RL, Stamler J et al (2008) by MS. Bioanalysis 8(9):981–997. https://

Human metabolic phenotype diversity and its doi.org/10.4155/bio-2015-0010
association with diet and blood pressure. 8. Wilson ID, Nicholson JK, Castro-Perez J et al
Nature 453(7193):396–400. https://doi. (2005) High resolution “ultra performance”
org/10.1038/nature06882 liquid chromatography coupled to oa-TOF
2. Newgard CB (2017) Metabolomics and meta- mass spectrometry as a tool for differential
bolic diseases: where do we stand? Cell Metab metabolic pathway profiling in functional
25(1):43–56. https://doi.org/10.1016/j. genomic studies. J Proteome Res
cmet.2016.09.018 4(2):591–598
3. Brennan L (2016) Metabolomics in nutrition 9. Nassar AF, Wu T, Nassar SF, Wisnewski AV
research-a powerful window into nutritional (2017) UPLC-MS for metabolomics: a giant
metabolism. Essays Biochem 60(5):451–458 step forward in support of pharmaceutical
4. Saurina J, Sentellas S (2017) Strategies for research. Drug Discov Today 22(2):463–470.
metabolite profiling based on liquid chroma- https://doi.org/10.1016/j.
tography. J Chromatogr B 1044- 1045:103– drudis.2016.11.020
111. https://doi.org/10.1016/j. 10. Zhao YY, Lin RC (2014) UPLC-MS(E) appli-
jchromb.2017.01.011 cation in disease biomarker discovery: the dis-
5. Haggarty J, Burgess KE (2016) Recent coveries in proteomics to metabolomics. Chem
advances in liquid and gas chromatography Biol Interact 215:7–16. https://doi.
methodology for extending coverage of the org/10.1016/j.cbi.2014.02.014
metabolome. Curr Opin Biotechnol 43:77–85. 11. Wang X, Sun H, Zhang A (2011) Ultra-
https://doi.org/10.1016/j. performance liquid chromatography coupled
copbio.2016.09.006. to mass spectrometry as a sensitive and power-
6. Markley JL, Brüschweiler R, Edison AS et al ful technology for metabolomic studies. J Sep
(2016) The future of NMR-based metabolo- Sci 34(24):3451–3459. https://doi.
mics. Curr Opin Biotechnol 43:34–40. org/10.1002/jssc.201100333
https://doi.org/10.1016/j. 12. Siskos AP, Jain P, Römisch-Margl W et al
copbio.2016.08.001 (2016) Interlaboratory reproducibility of a tar-
7. Chen Y, Xu J, Zhang R, Abliz Z (2016) geted metabolomics platform for analysis of
Methods used to increase the comprehensive human serum and plasma. Anal Chem
coverage of urinary and plasma metabolomes 89(1):656–665
13. Michopoulos F, Whalley N, Theodoridis G metabolic profiling in toxicological studies.

et al (2014) Targeted profiling of polar intra- Anal Chem 83(1):382–390. https://doi.
cellular metabolites using ion-pair-high perfor- org/10.1021/ac102523q
mance liquid chromatography and -ultra high 21. Virgiliou C, Sampsonidis I, Gika HG et al
performance liquid chromatography coupled (2015) Development and validation of a
to tandem mass spectrometry: applications to HILIC- MS/MS multi-targeted method for
serum, urine and tissue extracts. J Chromatogr metabolomics applications. Electrophoresis
A 1349:60–68. https://doi.org/10.1016/j. 36:2215–2225
chroma.2014.05.019 22. Want EJ, Wilson ID, Gika H et al (2010)
14. Monteiro MS, Carvalho M, Bastos ML et al Global metabolic profiling procedures for urine
(2013) Metabolomics analysis for biomarker using UPLC-MS. Nat Protoc 5(6):1005–
discovery: advances and challenges. Curr Med 1018. https://doi.org/10.1038/
Chem 20(2):257–271 nprot.2010.50
15. Dunn WB, Broadhurst D, Begley P et al (2011) 23. Want EJ, Coen M, Masson P et al (2010)
Human serum metabolome (HUSERMET) Ultra performance liquid chromatography-
consortium. Procedures for large-scale meta- mass spectrometry profiling of bile acid
bolic profiling of serum and plasma using gas metabolites in biofluids: application to
chromatography and liquid chromatography experimental toxicology studies. Anal Chem
coupled to mass spectrometry. Nat Protoc 82(12):5282–5289. https://doi.
6(7):1060–1083. https://doi.org/10.1038/ org/10.1021/ac1007078
nprot.2011.335. 24. Want EJ, Masson P, Michopoulos F et al
16. Gray N, Lewis MR, Plumb RS et al (2015) (2013) Global metabolic profiling of animal
High-throughput microbore UPLC-MS meta- and human tissues via UPLC-MS. Nat Protoc
bolic phenotyping of urine for large-scale epi- 8(1):17–32. https://doi.org/10.1038/
demiology studies. J Proteome Res nprot.2012.135.
14(6):2714–2721. https://doi.org/10.1021/ 25. Vorkas PA, Shalhoub J, Isaac G et al (2015)
acs.jproteome.5b00203 Metabolic phenotyping of atherosclerotic
17. Wilson ID (2015) Metabolic phenotyping by plaques reveals latent associations between free
liquid chromatography-mass spectrometry to cholesterol and ceramide metabolism in ath-
study human health and disease. Anal Chem erogenesis. J Proteome Res 14(3):1389–1399.
87(5):2519. https://doi.org/10.1021/acs. https://doi.org/10.1021/pr5009898
analchem.5b00409. No abstract available 26. Veselkov KA, Vingara LK, Masson P et al
18. Vorkas PA, Isaac G, Anwar MA et al (2015) (2011) Optimized preprocessing of ultra-per-
Untargeted UPLC-MS profiling pipeline to formance liquid chromatography/mass spec-
expand tissue metabolome coverage: application trometry urinary metabolic profiles for
to cardiovascular disease. Anal Chem improved information recovery. Anal Chem
87(8):4184–4193. https://doi.org/10.1021/ 83(15):5864–5872. https://doi.
ac503775m org/10.1021/ac201065j
19. Gika HG, Theodoridis GA, Plumb RS et al 27. Dias DA, Jones OA, Beale DJ, (2016) Current
(2014) Current practice of liquid and future perspectives on the structural iden-
chromatography-mass spectrometry in metab- tification of small molecules in biological sys-
olomics and metabonomics. J Pharm Biomed tems. Metabolites. 6(4). pii: E46.
Anal 87:12–25. https://doi.org/10.1016/j. 28. Bocker S (2016) Searching molecular structure
jpba.2013.06.032 databases using tandem MS data: are we there
20. Spagou K, Wilson ID, Masson P et al (2011) yet? Curr Opin Chem Biol 36:1–6. https://
HILIC-UPLC-MS for exploratory urinary doi.org/10.1016/j.cbpa.2016.12.010.
Chapter 8
NMR-Based Metabolic Profiling Procedures for Biofluids

and Cell and Tissue Extracts
Dimitra Benaki and Emmanuel Mikros
Abstract
Metabolomic studies offer a wealth of information on cells, tissues, and biofluids. The phenotype repre-
sentation through the metabolic profiling is a valuable tool for direct diagnosis, therapeutic strategies, and
system’s biology studies. Nuclear magnetic resonance (NMR) spectroscopy provides a nondestructive and
extremely reproducible method allowing simultaneous detection of a large number of known and unknown
chemical substances.
Sample collection and preparation and experimental conditions are critical for the reliability of the
subsequent analysis. The pre-analytical phase is decisive as it could generate biased spectral data misleading
the following analysis. The formulation of standard operating procedures is thus of crucial importance in
order to access meaningful samples and results. In this protocol, we provide standardized operations and
routine procedures from sample preparation to determine the measurement details for the acquisition of
NMR spectra highlighting major methodological issues.
Key words NMR spectroscopy, Metabolic profiling, Biofluids, Cell cultures, Tissues
1 Introduction
Metabolic profiling offers a comprehensive exploration of the

metabolome in biological systems encompassing advanced analyti-
cal techniques and multivariate statistical analyses. Solution-state
nuclear magnetic resonance (NMR) spectroscopy is a proven and
versatile tool of choice to simultaneously detect, in a rapid and
reproducible way the constituents of complex mixtures like bio-
logical samples.
NMR spectroscopy-based metabolic profiling provides a holis-
tic view of the alterations of a living system, monitoring in an
untargeted manner a variety of low molecular weight (MW)
metabolites, fluctuating under pathophysiological conditions or
genetic modifications. NMR-based metabolomics is applied in dis-
ease diagnosis, toxicology, system’s biology, and functional genom-
ics, as well as in studies that investigate how living organisms
117
118 Dimitra Benaki and Emmanuel Mikros
interact with their environment. Metabolites can be considered as

the terminal endpoint of biochemical processes influenced by vari-
ous external factors like diet (nutrimetabolomics), therapeutic
intervention (pharmacometabolomics), lifestyle, or others. A wide
range of biological samples such as urine, plasma, serum, cerebro-
spinal fluid (CSF), saliva, synovial fluid, semen, exhaled breath
condensate (EBC), as well as cell and tissue (organ, tumor, muscle)
homogenates can be analyzed with high-throughput automatic
routines coupled with powerful data analysis methods. The access
to metabolomic analysis along with other -omic methodologies
like genomics, transcriptomics, and proteomics has been character-
ized to be the “tour de force no. 1” of personalized medicine in
the future [1].
The main advantages of NMR spectroscopy for metabolic pro-
filing of biological samples are:
1. Rapid information recovery in a nondestructive manner. One-
dimensional NMR spectra can be acquired in a few minutes,
and the sample is fully recovered and can be reused.
2. Small sample volumes (about 500 mL).
3. Minimum sample preparation.
4. No pretreatment is required for samples like blood, plasma/
serumor urine.
5. Simultaneous detection of large number of small MW
metabolites.
6. No need for preselection of analytical conditions as required in
the chromatographic techniques, where specific experimental
parameters (column, detector, solvent) are required.
7. Detection over an extended concentration range.
8. Excellent reproducibility that makes 1H NMR spectroscope as
the method of choice for large epidemiological studies.
9. Detection of metabolites with no restrictions relating to volatil-
ity, polarity, and the presence of specific chromophores.
The main limitation of NMR is sensitivity as the detection limit
is in the micromolar range. However, higher magnetic fields allow
for better sensitivity and resolution of NMR spectra, while the use
of modern cryogenic probe heads (cryoprobes) can increase the
sensitivity several fold. Furthermore, integration of the system with
robotic sample changers and increased stability of conditions dur-
ing measurements (e.g., temperature) provide high accuracy and
reproducibility. Finally, NMR can be hyphenated with chromato-
graphic techniques and provide structural information comple-
mentary to MS. The most common analysis in NMR-based
metabolomics consists of simple 1D 1H spectra that can be acquired
in a few minutes per sample; however, in complex mixtures, 2D
NMR-Based Metabolic Profiling Procedures for Biofluids and Cell and Tissue Extracts 119
homonuclear and heteronuclear NMR can be employed giving the

possibility to identify unknown metabolites in an accurate
manner.
NMR is very sensitive to the pre-analytical processing of the
biosamples (collection, transportation, storage, sample prepara-
tion). In the case of biofluid collections, the daily routine of clinical
facilities may impose major obstacles. Well-standardized operations
of the pre-analytical phase are prerequisite for the reliability of the
following analysis. It is thus of crucial importance the formulation
of feasible SOPs in order to access meaningful samples and results.
2 Materials
2.1 Sample 1. Phosphate-buffered saline (PBS).

Collection 2. Ultrapure water.
and Metabolite
3. Chloroform, analytical grade, ≥99%.
Extraction
4. Methanol, LC-MS grade, ≥99.8%.
Dry ice and liquid N2 are required for the extraction procedure
and bovine serum albumin (BSA) and 1 M NaOH for the protein
content determination.
2.2 NMR Sample 1. KH2PO4 or Na2HPO4 × 7H2O or Na2HPO4 anhydrous >99%.

Preparation 2. KOH pellets or HCl.
3. Sodium azide (NaN3).
4. 3-(Trimethylsilyl)propionic-2,2,3,3-D4 acid sodium salt (TSP),
>98 atom % D.
5. Tetramethylsilane (TMS), >99%.
6. D2O, 99.9 atom % D.
7. CDCl3, 99.8% D.
2.3 Laboratory 1. Cell scraper, 18 mm blade.

Equipment 2. High precision pipettes.
3. Eppendorf tubes, 1.5 mL.
4. Centrifuge tubes, 15 mL.
5. Tissue homogenizer or mortar and pestle.
6. Lowry assay.
7. Refrigerated centrifuge.
8. Ultrasound bath.
9. Sample freeze drier or centrifugal vacuum concentrator.
10. −80 °C freezer.
11. Heparin blood collection tubes for blood sample collection.
12. Sterile urine collection tubes.
2.4 Instrumentation: 1. 600 MHz NMR spectrometer is advisable for biological sam-
NMR Equipment ples as a compromise in terms of resolution and sensitivity vs
cost.
2. Probe of 5 mm diameter with Z gradient (e.g., BBI, TXI for
Bruker Biospin).
3. Temperature control unit; N2 supply provides higher stability
compared to dry air.
4. Automatic sample changer (B-ACS60, Bruker Biospin, or
similar).
5. Software to control sample loading, temperature equilibration
and stability, matching and tuning, shimming, pulse calibration,
acquisition, and processing (ICONNMR, Bruker Biospin, or
similar).
2.5 Buffers Buffer A (50 mL): weigh 10.2 g KH2PO4, 50 mg TSP, and 6.5 mg
Preparation NaN3, add 40 mL D2O, and dilute in ultrasonic bath. Adjust the
pH to 7.4 with KOH pellets (for a volume of 50 mL around 33–35
pellets are required). Fill up with D2O to 50 mL volume and mix
very well. Aliquot and store at 4 °C.
Buffer B (500 mL): weigh 10.05 g Na2HPO4 × 7H2O (or
5.32 g Na2HPO4), 0.4 g TSP, and 0.2 g NaN3, add 380 mL ultra-
pure water, and dilute in ultrasonic bath. Adjust the pH at 7.4 with
1 M HCl. Fill volume up to 400 mL with ultrapure water. Add
100 mL D2O and mix well. Aliquot and store at 4 °C.
Check always the pH before use.
3 Methods
3.1 Sample The initial treatment step on the biological system has to be acute
Collection and effective (to stop all the enzymatic activity) in order to freeze
3.5.1 Cells
the existing state as well as to ensure system integrity and to gener-
ate reproducible results.
It is generally accepted that approximately 107 cells produce
NMR samples of the necessary quality for metabolomic studies.
Sample collection in monolayer cultures
1. Aspirate the cell medium or collect and store at −80 °C for fur-
ther analysis of the excreted metabolites (see Note 1).
2. Place the culture dishes on ice and wash twice with ice-cold PBS
to remove residual traces of culture medium.
3. Add ice-cold methanol up to the necessary volume to cover the
culture surface (2 mL for the 10-cm-diameter culture dish,
4 mL for 75 cm2 flask or adjust the volume according to the
culture vessel). Immediate quench of the cell metabolism is

decisive for the quality of the study.
4. Leave on ice for 5 min.
5. Use a scraper for the mechanical detachment of the cells (see
Note 2).
6. Transfer the suspension into a 15 mL centrifuge tube and pro-
ceed to the metabolite extraction (see Subheading 3.2.1.). To
store the cells, ice-cold PBS is used for cell quenching instead of
methanol (step 3). The cell quench is achieved solely by the low
temperature because methanol causes cell membrane disrup-
tion and consequent intracellular metabolite leakage [2]. The
suspension (PBS and cells) is centrifuged at 230 × g for 5 min,
the supernatant is aspirated, and the cell pellet is stored at
−80 °C till the metabolite extraction procedure.
Sample collection in cell suspension
1. Aliquot the desired number of cells (around 107) into a clean
15 mL centrifuge tube, and centrifuge at 230 × g for 5 min.
2. Discard the supernatant or store at −80 °C for further analysis.
3. Suspend the cells twice in ice-cold PBS (centrifuged at 230 × g
for 5 min, discard, and repeat).
4. Store the cell pellet at −80 °C or resuspend the cells in 2 mL of
ice-cold methanol and proceed to the metabolite extraction.
3.1.1 Tissues Tissue Collection

General issues: (1) animals should be handled humanely in
accordance with the European Union Directive 86/609 (or
updated). (2) The animals should be lightly anesthetized before
sacrifice with compounds that do not affect organism-organ
metabolome. (3) All anesthetic agents must be listed on an
approved Animal Protocol (IACUC Guidelines). (4) Treatment
with compounds that are detectable through NMR (e.g., ether
anesthesia for animal models) or that induce metabolic modifica-
tions [3, 4] should be avoided. A recently published study on male
C57BL/6J mice using untargeted and targeted metabolomics
reveals dramatic tissue-specific impacts of various collection strate-
gies [5]. (5) Organs and tissues exhibit topological heterogeneity,
e.g., ischemic and nonischemic areas in the heart and necrotic and
oxygenated regions in tumors. When part of the organ will be sub-
jected to metabolomic analysis, the same topological region should
be used across the entire study. (6) Use gloves and alcohol swabs
to clean the instruments after each operation.
1. Tissue collection has to be accomplished in a short time period.
2. Rinse the organs promptly with ultrapure water to remove
blood in order to avoid sample contamination.
3. Dab dry and proceed to the tissue dissection in order to avoid

degradation.
4. Place the tissue in prechilled, labeled containers, and immerse
directly in liquid N2.
5. Stored at −80 °C.
Tissue Homogenization
Blenders, beads shakers, or cryogenic mortar and pestle are
used for tissue homogenization. Regardless of the homogenization
method, all the accessories are prechilled in order to avoid tissue
degradation and have to be cleaned between samples to avoid con-
taminations (see Note 3).
The following procedure is described for homogenization
using acryogenic mortar and pestle.
1. Store the mortar and pestle overnight at −80 °C.
2. Place the mortar on dry ice along the whole procedure.
3. Poor liquid N2 into the mortar and transfer the frozen tissue.
4. At the end of the homogenization, allow the excess of liquid N2
to evaporate (continuous mixing).
5. Use a cold spatula (kept chilled on dry ice) to transfer the pul-
verized tissue promptly into chilled, pre-weighed, and labeled
eppendorfs or cryovials (see Note 4).
6. Reweigh to determine the tissue mass.
7. Store at −80 °C till the extraction procedure (see Note 5).
3.1.2 Biofluids Biofluid collection is easier than other biological samples; however,
processing and time intervals are crucial for system integrity and
consequent sample variability.
Blood collection is usually performed in prefilled vials with
anticoagulant supplements as EDTA, citrate, or heparin. EDTA
shows strong and broad resonance lines that obscure extended and
important areas of the 1H NMR spectra of the samples [6]. The
use of non-deuterated citrate can change the endogenous metabo-
lite concentration and/or saturate NMR receiver.
Plasma collection
1. Collect whole blood into heparin tubes in the morning pre-
prandially, after overnight fasting.
2. Centrifuge at 1500 × g for 10 min at 4 °C to separate blood
cells from plasma within 30 min of collection.
3. Collect the supernatant, aliquot, and store at −80 °C.
Serum collection
1. Allow blood to clot without anticoagulant on ice for 30 min.
2. Centrifuge at 1500 × g for 15 min at 4 °C.
3. Collect the serum, aliquot, and store at −80 °C.

Both procedures (plasma and serum collection) should not
exceed 2 h to minimize any metabolite degradation.
Urine collection
1. Collect midstream urine in sterile containers, after overnight
fasting.
2. Keep on ice or at 4 °C.
3. Centrifuged at 1850 × g for 10 min at 4 °C to remove debris.
4. Collect the supernatant, aliquot, and store at −80 °C.
3.2 Metabolite To identify different sources of contaminations: (1) check the

Extraction purity of the solvents to be used for the extraction procedure with
NMR, i.e., reduce to dryness the same, or better double the vol-
ume, to be used and record an NMR spectrum in the correspond-
ing deuterated solvents (methanol, chloroform, deionized water);
and (2) use blank extraction samples and record NMR spectra as
for the original samples [7].
In the case of the cell endo-metabolome (intracellular metabo-
lites) and tissue homogenates, a metabolite extraction step is
required to separate polar from nonpolar metabolites and isolate
insoluble macromolecular entities (proteins and disrupted mem-
branes). High MW biomolecules introduce broad, unresolvable,
signals into the 1H NMR spectra, disrupt the flat baseline, and
make the integration of metabolite signals and comparison impos-
sible. The three-solvent (methanol-water-chloroform) extraction
protocol is widely used both in cell and tissue metabolomic studies
[7–9].
3.2.1 Cell Extraction 1. For cell suspensions quenched by 2 mL ice-cold methanol, an

equal volume of ice-cold chloroform is added (methanol/
chloroform 1:1, adjust accordingly).
2. Sonicate for 5 min (keep the temperature of the bath low by
adding ice).
3. Add 1.8 mL (adjust accordingly) of cold ultrapure water.
4. Vortex for 1 min to create a homogenous emulsion.
5. Keep on ice for 15 min (helps the separation of the two
phases).
6. Centrifuge at 18,900 × g for 20 min at 4 °C.
7. Transfer the two layers into clean tubes, and store at 4 °C till
the second round of extraction is accomplished (step 8).
8. Repeat the extraction procedure (add methanol/chloroform/
water 2:2:1.8 and go to step 4).
9. Collect the two phases and pool with the first extraction
fractions.
10. Dry the extracts overnight in a vacuum centrifuge or under a
nitrogen stream and store at −80 °C.
3.2.2 Tissue Extraction In the case of cell pellets stored at −80 °C, the samples are left
to thaw on ice, add 2 mL ice-cold methanol, vortex to disrupt the
pellet, and follow the extraction procedure described above.
1. Add ice-cold CHCl3/MeOH solution (2:1, 1 mL total vol-

ume/100 mg tissue) to frozen pulverized tissue.
2. Mix vigorously for the cell enzymatic function deactivation.
Use a needle to disrupt any remaining large frozen
formations.
3. Sonicate in ice-water bath for 5 min.
4. Add an equivalent volume (1 mL) of cold ultrapure water.
5. Vortex vigorously for 1 min.
6. Leave on ice for 15 min.
7. Centrifuge at 18,900 × g for 20 min at 4 °C (see Note 6).
8. Collect the two phases in clean cryovials, and store at 4 °C till
the second extraction run (step 9).
9. Repeat the extraction procedure from step 1.
10. Collect the two phases and pool with the first extraction
fractions.
11. Dry overnight in a vacuum centrifuge or in N2 stream and
store at −80 °C.
When different tissue weights are accessible, adjust the required
solvent volumes in order to generate samples of similar concentra-
tion. The collection of equal volumes from each sample extracts
(step 8) makes easier the NMR sample preparation. For additional
normalization, dry the insoluble residual formed in the biphasic
interface, weigh, and solubilize 2 mL/g tissue (or in the case of
cells 200 μL/pellet) to determine the protein of content using a
Lowry assay [10].
3.3 NMR Sample The use of a buffer solution is necessary to ensure the stable and
Preparation constant pH of the samples. Proton resonances can be strongly
dependet on the pH and small fluctuations result in signal shifts,
hampering further analysis. The buffers used in NMR studies are
the sodium or potassium phosphates, as their capacity covers the
physiological pH range. It is advisable to generate quality control
(QC) samples by pooling 5 μL of each sample.
3.3.1 Cell and Tissue The dried cell and tissue extracts are reconstituted in 10% buffer A
Extracts: Polar Metabolites and 90% D2O (see Note 7).
1. Prepare the necessary volume of 10% buffer A in D2O 99.9%.
2. Leave the samples to thaw at room temperature.
3. Add 650 μL of the solution (step 1), and mix properly (use
vortex and/or ultrasonic bath) to ensure the resuspension of all
the contents.
5. Transfer 550 μL of the supernatant to a clean 5-mm-diameter
NMR tube.
3.3.2 Cell and Tissue 1. Prepare the necessary volume of CDCl3 0.03% v/v TMS solu-
Extracts: Nonpolar tion. During sample reconstitution, place the bottle on ice, and
Metabolites close the cup immediately after use to prevent, as much as pos-
sible, TMS evaporation.
2. Leave the samples to thaw at room temperature.
3. Add 650 μL of the solution (step 1), and mix properly (use
vortex and/or ultrasonic bath) to ensure the resuspension of all
the content.
5. Transfer 550 μL of the supernatant in a clean 5-mm-diameter
NMR tube.
3.3.3 Biofluids For urine samples, buffer A is used.

1. Leave the samples to thaw on ice.
2. Transfer 630 μL in an eppendorf and add 70 μL buffer A.
3. Vortexed for 60 s.
5. Transfer 550 μL of the supernatant in a clean 5-mm-diameter
NMR tube.
For other biofluids as EBC and CSF, as well as the extracellular
cell culture medium, the NMR samples are prepared as urine
samples. Additional precautions should be taken in the case of CSF
samples as exposure to the atmosphere causes CO2 evaporation,
and pH can reach high values (9.0–9.7). Phosphates do not have
any buffering capacity at this pH range, and solutions conditioning
can be aided by the addition of DCl to lower the pH to around 7
and then add 10% buffer A.
Blood plasma and serum contain proteins and should be
treated with care. Buffer B is added up to 50% of the total volume,
no vortex is applied, and mixing is achieved through gentle pipet-
ting (see Note 8).
1. Leave the samples to thaw on ice.

2. Remove particulates using a needle.
3. Transfer 350 μL to an eppendorf and add 350 μL buffer B and
mix very gently (pipette up and down).
4. Transfer 550 μL in a clean 5-mm-diameter NMR tube.
3.4 NMR A major challenge of fundamental importance in NMR metabolo-

Experiments mics spectra, is to remove water resonance (or HDO signals in the
and Parameters case of cell and tissue extract solutions) in a smooth and effective
way in order not to affect the nearby resonances.
Another important issue is the presence of large MW entities, as
such proteins and lipids, in certain biofluids (plasma, serum).
Molecules with short tumbling time give rise to high-intensity
broad signals in the 1H NMR spectra that obscure the analysis of
small MW entities. A pulse sequence that filters proton resonances
according to T2 relaxation rate is the Carr-Purcell-Meiboom-Gill
(CPMG) resulting in a flat baseline proton spectrum. The CPMG
pulse sequence offers the opportunity to depict resolved signals of
the low MW metabolites without discarding the proteinaceous con-
tent. Protein content transfers valuable information (e.g., LDL and
HDL relative concentration in blood -derived samples) which can
be accessed through a diffusion-editing pulse sequence with bipo-
lar-gradient pulses that allow obtaining a reasonable amount of dif-
fusion of the lipoprotein signals. Alternatively, the proteins can be
removed by an additional ultrafiltration or precipitation step [11].
Before the setup of NMR experiments:
1. Confirm the high-quality performance of the spectrometer fol-
lowing the control SOPs defined by the manufacturer using
appropriate standard solutions.
2. Perform a temperature calibration (see Note 9).
3. Optimize field homogeneity and water presaturation (prior to
biofluid samples runs) using the manufacturer’s standard
solution (sucrose 2 mM, 0.5 mM DSS, 2 mM NaN3 in H2O/
D2O 90/10) (part of the SOP).
4. Randomize sample analysis.
All NMR spectral parameters are optimized for a single sample,
and the automation program uses this set for all the samples of the
run (see Note 10). The sample of choice can be a QC sample,
being the most representative; while not a real specimen, a QC
sample contains the fingerprint information from the whole sample
collection. If no QC samples are available, a quite concentrated
one is selected (in urine samples, the color is indicative).
Water frequency determination is crucial as this value will be
used for the water signal suppression in all the spectra acquired in
a run.
Relaxation delay is different for the various fluids and should

be optimized according to the phase distortion of the residual
water signal (see Note 11).
The number of scans in the case of biofluids as urine, plasma,
and serum samples is adjusted to 32. More scans, 128, 192, or
even 256, are required for extracted metabolites (cells and
tissues).
Receiver gain. A parameter that is optimized and fixed is the
receiver gain value (e.g., adjusted at 90.5 for 600 MHz Bruker
instruments).
3.4.1 Specific Pulse 1. The proton 1D experiment, holding the major metabolomic
Sequence Parameters information (with the exception of samples with high content
of macromolecules), is acquired using the NOESY-presaturation
pulse sequence with gradients (noesygppr1d, Bruker library)
offering the optimum water suppression (see Note 12). A spec-
tral width of 20 ppm is required with a sampling of 64k points
resulting in an acquisition time of 2.7 s. The mixing time at
10 ms for the NOESY sequence is optimum in combination
with the presaturation and provides the best compromise for
water suppression and relaxation effect suppression thus affect-
ing peak quantification to a lesser extent. Plasma and serum
spectral width is adjusted to 30 ppm, and sampling is increased
up to 96k points resulting thus in an acquisition time of 2.7 s.
The 90 deg. pulse width should be optimized for each sample
separately (part of the acquisition automation routine) and kept
constant for all the spectra of the same sample. The processing
includes zero filling and exponential multiplication, phase cor-
rection, and axis calibration.
2. J-resolved pdeudo 2D experiments are very fast (5–10 min,
depending on the analysis and number of scans) and of utmost
importance offering the possibility to resolve overlapped sig-
nals. A spectral width of 16 ppm in the 1H axis, while 70–80 Hz
for the J coupling, is enough, with 12k points, 40 increments,
and 4–8 scans (4×n). Processing includes zero filling in both
dimensions (to 16 k and 256 for F2 and F1, respectively), line
broadening multiplication with a factor of 0.3 Hz, baseline cor-
rection, an additional tilt by 45 deg. step, and symmetrization
about the J 0 Hz line.
Additional spectra are required in the case of plasma and
serum samples (and generally for biofluids with proteins), either
to reduce the disturbance created by the macromolecular enti-
ties contribution (CPMG) or specific to derive the information
carried by large biomolecules (diffusion edited).
3. The T2 filter (relaxation edited) using the Carr-Purcell-
Meiboom-Gill sequence is applied to suppress the high MW
contribution (cpmgpr1d pulse sequence, Bruker library, with
presaturation during relaxation delay) in combination with

the presaturation scheme. The spectral width is set to 20 ppm,
with 72k points for data sampling resulting in a 3.1 s acquisition
time and 32 scans. Combination of 128 loops for the T2 filter-
ing with spin-echo delay of 300 μs results in a total echo time of
77 ms. For the processing, zero filling is implemented adding
up to 128k data points and a line broadening of 1 Hz is used
prior to Fourier transformation.
4. The diffusion-edited pulse sequence (molecular diffusion coeffi-
cient edited, ledbpgppr2s1d pulse sequence, Bruker library,
with presaturation during relaxation delay) uses a bipolar pulse
pair-longitudinal eddy current delay to analyze mainly the lipid
content of plasma lipoproteins. A Spectral width of 30 ppm is
applied using 96k data points and 2.7 s acquisition time. Typical
optimized values are diffusion delay (big delta, Δ) 120 ms, 3 ms
(little delta, δ), delay for the eddy current decay 5 ms, diffusion
gradient length that allows editing 1.5 ms, and spoil gradient
length 600 μs. Processing includes zero filling to 131k data
points and a line broadening of 1 Hz.
In plasma and serum samples, the glucose alpha-anomeric
proton resonance at 5.23 ppm is used as an internal standard for
the chemical shift axis calibration, due to interaction of the TSP
with protein molecules (see Note 13).
5. Simple 1H 1D NMR spectrum, i.e., without solvent suppres-
sion, using an optimized 90 deg. pulse is acquired for lipid
metabolite profiling. Preferentially, a 30 deg. pulse is used in
order to the improve signal-to-noise ratio at the same time.
Spectra are acquired with a spectral width of 20 ppm, 64k data
points, 2.73 s acquisition time, and 64 transients. Usually spec-
tra are recorded at 295 K. Processing includes zero filling to
131k data points and a line broadening of 1 Hz.
Although J-resolved spectrum is very important for the anal-
ysis of overlapped regions of the spectra, 2D experiments are
always performed on selected samples for the unambiguous
identification of spin systems (TOCSY spectrum) and 1H-13C
correlations (HSQC-DEPT edited spectrum).
6. 2D TOCSY experiments (dipsi2phpr pulse sequence, Bruker
library) are recorded with a spectral width of 16 ppm, 2k time
domain points, and 256 increments. Receiver gain and number
of scans are adjusted according to sample concentration.
7. 2D HSQC (hsqcedetgpsisp2.3 pulse sequence, Bruker library)
with multiplicity editing is recorded for a spectral width of
16 ppm for the 1H dimension and 190 ppm for 13C, 2k time
domain points, and 256 increments. Receiver gain and the num-
ber of scans are adjusted according to the sample
concentration.
4 Notes
1. Cell culture medium contains sugars and amino acids as nutri-

tion supplements and metabolites secreted into the medium
related to cellular function.
2. The use of a scraper for the mechanical detachment of cells is
preferred compared to standard trypsinization to avoid the use
of additional compounds. However, in a recent study, Kapoore
and coworkers [12] reported that leakage of metabolites dur-
ing trypsinization and scraping treatments may be dependent
on cell membrane architecture, affecting the recovery of dif-
ferent metabolite classes for different cell lines. Based on these
data, mechanical detachment is safer when the extraction pro-
cedure directly follows cell collection, and methanol with the
cells will be used in the extraction. For cell storage, trypsiniza-
tion can be used, but the supernatant should be checked for
metabolite leakage.
3. For cleaning, blenders are disassembled, and all compartments
are rinsed with deionized water and methanol, while a paper
roll can be used to clean thoroughly mortar and pestle.
Blenders may introduce temperature increase in the specimen
depending on the time necessary for the homogenization (soft
tissues as the liver, hard tissues as certain solid tumors).
4. Handle eppendorfs from the cup as any contact elevates the
temperature and might cause tissue defrost. The eppendorfs
are immersed in liquid N2 before and immediately after the
transfer of the ground tissue.
5. Eppendorfs containing pulverized tissue, when transferred
from liquid N2 to −80 °C, should be left with open cups for
about 1 h, to allow N2 evaporation and avoid the development
of high pressure that may spread the pulverized tissue. Tissue
homogenates are stored at −80 °C.
6. If an emulsion is still present, a 2 min centrifugation is usually
enough to separate the two phases.
7. It is advisable to prepare the reconstitution solution for all the
samples in the same run, in order to minimize errors of pipet-
ting (small fluctuations of the volume affect the quality of the
automated optimization of field homogeneity before spectral
acquisition).
8. Pipette with care to prevent foam formation due to the pro-
teinaceous content. Remove particulate matter with the help
of a needle.
9. Cell and tissue extracts, as well as biofluids spectra, are recorded
at 300 K, except for plasma and serum spectra which are
recorded at 310 K. For this temperature range, the calibration
standard is deuterated methanol solution, 99.8% D, in sealed

NMR tubes, ideally provided by the manufacturer.
10. For the field homogeneity, the internal standard provides the
quality control. The half-height line width (HHLW) value of
the TSP signal at 0.00 ppm in different biofluids can be used.
In cell and tissue extracts, and in urine samples, the HHLW
value should be <1 Hz, and Si satellites should be clearly visi-
ble, while in samples with protein content, such as blood
plasma and serum, it is expected to be onefold higher.
11. In urine, plasma, serum, and cell extract samples, a value of 4 s
is optimum, while in CSF samples and tissue extracts, a higher
value (up to 10 s) is required.
12. For water suppression, the pulse sequence of choice is a com-
bination of the soft presaturation scheme with the a 1D
NOESY pulse sequence with a short mixing time, preferably
10 ms. In more recent pulse sequences, spoil gradients have
been introduced. The power of the water suppression pulse is
as low as necessary, in order to introduce the least possible
distortion to adjacent signals. More forceful pulse schemes,
such as Watergate and sculpting, although more effective,
affect resonances close to the water signal, i.e., the anomeric
proton of sugars, the Cα proton of several amino acids, as well
as baseline quality. The pulse power for water presaturation
(pl9, Bruker instruments) is 25 Hz for H2O/D2O samples
(urine, CSF, EBS, plasma) or even less e.g., 5–10 Hz in the
case of 100% D2O samples (cell and tissue extracts).
13. In urine samples, the creatinine resonance can be used as an
internal standard for quantitation purposes.
Acknowledgments
The authors wish to acknowledge Eberhard Humpfer and Manfred

Spraul (Bruker BioSpin, Karlsruhe) for useful help and advice on
NMR parameters optimization.
References
1. Chen R, Mias GI, Li-Pook-Than J et al (2012) 2. Dietmair S, Timmins NE, Gray PP et al (2010)
Personal omics profiling reveals dynamic Towards quantitative metabolomics of mam-
molecular and medical phenotypes. Cell malian cells: development of a metabolite
148(6):1293–1307. https://doi. extraction protocol. Anal Biochem 404:155–
org/10.1016/j.cell.2012.02.009. Cohen 164. https://doi.org/10.1016/j.
J (March, 2012). Examining his own body, ab.2010.04.031
stanford geneticist stops diabetes in its tracks. 3. Collinet H, Renault D (2012) Metabolic
News.sciencemag.org. Retrieved from http:// effects of CO2 anaesthesia in Drosophila
www.sciencemag.org/news/2012/03/ Melanogaster. Biol Lett 8:1050–1054.
examining-his-own-body-stanford-geneticist- https://doi.org/10.1098/rsbl.2012.0601
stops-diabetes-its-tracks
4. Ghini V, Unger FT, Tenori L et al (2015) metabonomic procedures for NMR spectros-
Metabolomics profiling of pre-and post- copy of urine, plasma, serum and tissue extracts.
anesthesia plasma samples of colorectal patients Nat Protoc 2(11):2692–2703. https://doi.
obtained via Ficoll separation. Metabolomics org/10.1038/nprot.2007.376
11:1769–1778. https://doi.org/10.1007/ 9. Sapcariu SC, Kanashova T, Weindl D et al
s11306-015-0832-5 (2014) Simultaneous extraction of proteins
5. Overmyer KA, Thonusin C, Qi NR et al (2015) and metabolites from cells in culture. MethodsX
Impact of anesthesia and euthanasia on metab- 1:74–80. https://doi.org/10.1016/j.
olomics of mammalian tissues: studies in a mex.2014.07.002
C57BL/6J mouse model. PLoS One 10. Le Belle JE, Harris NG, Williams SR et al
10(2):e0117232. https://doi.org/10.1371/ (2002) A comparison of cell and tissue
journal.pone.0117232 extraction techniques using high-resolution
6. Nicholson JK, Buckingham MJ, Sadler PJ 1
H-
NMR spectroscopy. NMR Biomed
(1983) High resolution 1H NMR studies of 15:37–44
vertebrate blood and plasma. Biochem 11. Gowda NGA, Raftery D (2014) Quantitating
J 211(3):605–615 metabolites in protein precipitated serum using
7. Keun HC, Athersuch TJ (2011) Nuclear magnetic NMR spectroscopy. Anal Chem 86(11):5433–
resonance (NMR)-based metabolomics. In: Metz 5440. https://doi.org/10.1021/ac5005103
TO (ed) Metabolic profiling, Methods in molecu- 12. Kapoore RV, Coyle R, Staton CA et al (2015)
lar biology, vol vol 708. Springer Protocols, Cell line dependence of metabolite leakage in
Humana Press, New York, pp 321–334. https:// metabolome analyses of adherent normal and
doi.org/10.1007/978-1-61737-985-7 cancer cell lines. Metabolomics 11:1743–1755.
8. Beckonert O, Keun HC, Ebels TMD et al https://doi.org/10.1007/
(2007) Metabolic profiling, metabolomic and s11306-015-0833-4
Chapter 9
Untargeted GC-MS Metabolomics

Matthaios-Emmanouil P. Papadimitropoulos, Catherine G. Vasilopoulou,
Christoniki Maga-Nteve, and Maria I. Klapa
Abstract
Untargeted metabolomics refers to the high-throughput analysis of the metabolic state of a biological
system (e.g., tissue, biological fluid, cell culture) based on the concentration profile of all measurable free
low molecular weight metabolites. Gas chromatography-mass spectrometry (GC-MS), being a highly sen-
sitive and high-throughput analytical platform, has been proven a useful tool for untargeted studies of
primary metabolism in a variety of applications. As an omic analysis, GC-MS metabolomics is a multistep
procedure; thus, standardization of an untargeted GC-MS metabolomics protocol requires the integrated
optimization of pre-analytical, analytical, and computational steps. The main difference of GC-MS metab-
olomics compared to other metabolomics analytical platforms, including liquid chromatography-MS, is
the need for the derivatization of the metabolite extracts into volatile and thermally stable derivatives, the
latter being quantified in the metabolic profiles. This analytical step requires special care in the optimiza-
tion of the untargeted GC-MS metabolomics experimental protocol. Moreover, both the derivatization of
the original sample and the compound fragmentation that takes place in GC-MS impose specialized
GC-MS metabolomic data identification, quantification, normalization and filtering methods. In this
chapter, we describe the integrated protocol of untargeted GC-MS metabolomics with both the analytical
and computational steps, focusing on the GC-MS specific parts, and provide details on any sample depend-
ing differences.
Key words Untargeted metabolomics, Gas chromatography-mass spectrometry (GC-MS) metabolo-

mics, Metabolic profiling, Metabolic network analysis, Primary metabolism
1 Introduction
In the era of systems biology, high-throughput biomolecular

(omic) analyses have been used extensively in order to obtain a
global perspective of the molecular physiology of biological sys-
tems. Untargeted metabolomics pursues the metabolic physiology
and concerns the high-throughput analysis of the metabolic state
regulation of a tissue, biological fluid, or cell culture, through the
quantification of the concentration profile of all its measurable free
low molecular weight metabolites, i.e., its metabolic profile, under
various physiological conditions [1–3]. Thus, metabolomics should
133
134 Matthaios-Emmanouil P. Papadimitropoulos et al.
not be viewed as a chemometric technique but as a physiological

analysis leading to biologically relevant conclusions. As such, it is a
multistep procedure, and its standardization requires the inte-
grated optimization of pre-analytical, analytical, and computational
stages [4]. The pre-analytical steps include the experimental design
that needs to take into consideration both the biological and the
analytical constraints of the study and of the involved omic analy-
sis/es and the sample collection, handling, quenching, and storage
protocols. While special care should be given to the sample collec-
tion and handling protocol in any omic analysis to avoid major
perturbations of the samples physiological state, the optimization
of this step becomes even more important in metabolomics, taking
into consideration that metabolism is a very dynamic cellular pro-
cess. Interestingly, in metabolomics, quenching and metabolite
extraction, the first analytical step, could take place simultaneously
in the case of polar metabolites, as methanol used for quenching is
an extraction agent of polar metabolites too. In most cases, for
consistency between extraction protocols and different omic analy-
ses, it is preferable that the quenching is separated from the extrac-
tion step. Especially for tissues, quenching is carried out by fast
freezing of the samples in liquid nitrogen and stored at −80 °C
until further analysis. If transfer is required, the quenched samples
should be transported on dry ice (see Note 1).
There is no extraction reagent for the entire metabolome;
methanol- and chloroform-based protocols are mainly used for the
extraction of polar and nonpolar metabolites, respectively, and
optimized protocols have been proposed for a variety of systems
[2]. Concerning the analytical techniques for the acquisition of the
metabolic profile, in the case of untargeted metabolomic analyses,
mass spectrometry (MS) in conjunction with gas (GC) or liquid
chromatography (LC) is sometimes preferred over nuclear mag-
netic resonance (NMR) spectroscopy, because they are more sensi-
tive and high throughput [2, 5]. However, mass spectrometry-based
metabolic profiling contain a large number of unidentified peaks,
which cannot be related to the metabolic physiology of the inves-
tigated system [2, 5]; thus, educated methods for peak identifica-
tion are required. Between GC-MS and LC-MS, the advantages of
the former include the higher chromatographic resolution in the
gas compared to the liquid phase and the larger databases of identi-
fied peaks, because of its longer use in clinical chemistry practice.
Moreover, GC-MS can measure diverse classes of compounds with
one type of column, and the electron ionization mode involves
fragmentation of the molecules according to their structure, which
contributes to their identification and lessens the need for tandem
mass spectrometry. For all these reasons, GC-MS is an integral
equipment of a mass spectrometry metabolomics facility. It has
been widely used for untargeted metabolomic analyses of the
primary metabolism activity, as it can quantify compounds of

molecular weight smaller than 600 a.u.
Untargeted GC-MS Metabolomics 135
However, GC-MS metabolomics imposes an additional analytical

step, which concerns the derivatization of the extracted metabo-
lites into volatile and thermally stable derivatives [2]. This is a cru-
cial procedure that needs to be carefully standardized, especially in
untargeted metabolomics, and affects both the analytical and the
computational protocols of the analysis [2, 6]. This is due to the
fact that the derivatization kinetics is different between the com-
pounds and depends on the composition of each metabolite extract
(matrix effects) [2, 6]. Thus, the metabolic profile acquisition of all
derivatized samples in an experiment should be carried out at or
after the derivatization time at which all original metabolites in any
sample have been fully transformed into a derivative. This time
may differ between samples and should be identified through spe-
cific experiments [2, 6]. Moreover, there exists no derivatization
method that leads to one derivative per metabolite for all com-
pound classes [2, 6]. Some metabolites may form multiple deriva-
tives, the relative concentration of which changes with the duration
of the derivatization [2, 6]. The most widely used derivatization
method in untargeted GC-MS metabolomics, as it covers a wide
range of metabolite classes, is a two-step procedure, involving (a)
the methoximation of the ketone group-containing metabolites
into two stable methoximes of constant concentration ratio, and
(b) the silylation of all metabolites, including the methoximes
formed in the previous step, into their trimethylsilyl (TMS) deriva-
tives [2, 6]. In this case, the amine group-containing metabolites
can form multiple derivatives produced serially as the derivatiza-
tion progresses [2, 6]. The constant ratio between the two methox-
ime derivatives of ketone group-containing metabolites between
samples could be used as a quality control criterion that all samples
have been run under the same derivatization and equipment acqui-
sition conditions. It is underlined that it is the only such criterion
that can be applied post-experimentally, not requiring quality con-
trol samples [2, 6]. Moreover, estimating the effective derivative
peak area of each amine group containing metabolite that is directly
proportional to the concentration of the original metabolite in the
samples requires specialized normalization methods [2, 6] that
need to be appropriately used in the computational part of the
protocol [7]. However, the application of these normalization
methods requires the availability of certain measurements; thus,
the experimental protocol should be appropriately modified so that
the profiles of certain samples could be quantified at multiple
derivatization times [2, 6]. In this way, the metabolomic analysis
presented here is metabolite-centric and not feature- or
derivative-centric.
Taking all these special characteristics of untargeted GC-MS
metabolomics into consideration, we present a standardized proto-
col that can be applied to any biological system, with certain adap-
tations depending on the type of the sample. Such adaptations are
indicated either in the protocol and/or the accompanying notes.

The protocol integrates the experimental and the computational
parts, with the former including the polar metabolite extraction,
the derivatization, and the GC-MS metabolic profile acquisition
and then the metabolite derivative identification/quantification,
data normalization and filtering, the multivariate statistical analysis
and the metabolic network reconstruction. Our research group has
already used this protocol to acquire the metabolic profiles of plant
[8–10] and animal [11, 12] tissues, cell cultures [13, 14], and
blood plasma samples [5, 15].
2 Materials
2.1 Consumables 1. Tips for automatic pipettes.

2. Glass tips.
3. 2 mL microcentrifuge tubes (Eppendorf type)—for low-volume
samples.
4. Pellet micropestles (for the homogenization of low-volume
samples in Eppendorf-type tubes).
5. 15 mL centrifuge tubes (Falcon type)—for high-volume
samples.
6. 1.5 mL high-recovery autosampler glass vials with PTFE crimp
caps.
2.2 Reagents 1. HPLC grade water.

2. HPLC grade methanol.
3. Adonitol/ribitol (as internal standard).
4. U-13C-d-Glucose (as internal standard).
5. Pyridine anhydrous.
6. Methoxylamine hydrochloride (98+%).
7. N-methyl-N-(trimethylsilyl)trifluoroacetamide (MSTFA)
(97+%).
2.3 Wet Lab 1. Glass homogenizer with Teflon pestle (Thomas Scientific,
Equipment Swedesboro, NJ, USA).
2. Automatic pipettes.
3. Pipettes with glass tips (Drummond Scientific, Broomall, PA,
USA).
4. Water bath.
5. Cooling centrifuge (Thermo Fisher Scientific, Waltham, MA,
USA).
6. Vacuum centrifuge/concentrator (Thermo Fisher Scientific,

Waltham, MA, USA).
7. Benchtop shaker incubator (New Brunswick Scientific, Einfield,
CT, USA).
2.4 GC-MS 1. 3800 series GC (Varian, Palo Alto, CA, USA—now Bruker,
Equipment Billerica, MA, USA).
and Consumables 2. Saturn 2200 series MS ion trap (Varian, Palo Alto, CA, USA—
now Agilent, Santa Clara, CA, USA)—the protocol provided
can be applied to any GC-MS equipment.
3. CP-8410 autosampler (Bruker, Billerica, MA, USA).
4. GC capillary column: Zebron, ZB-50, 30 m × 0.25 mm
ID × 0.25 μm (Phenomenex, Torrance, CA, USA).
5. Helium 99.999% carrier gas (Air Liquide Hellas, Athens,
Greece).
2.5 Software 1. MS workstation v. 6.5 (Varian, Palo Alto, CA, USA—now

Agilent, Santa Clara, CA, USA)—any other software for peak
identification compatible with the available GC-MS equipment
could be used.
2. The National Institute of Standards and Technology (NIST)
mass spectral search program for the NIST/EPA/NIH mass
spectral library.
3. Microsoft Excel 2010 (Microsoft, Redmond, WA, USA).
4. M-IOLITE GC-MS metabolomic data repository, normaliza-
tion, and unknown peak identification software platform (v1
beta)—available upon request for academic users at miolite.
iceht.forth.gr [7].
5. TM4/MeV omic data analysis open-source software, v.4.9.0
[16, 17].
3 Methods
3.1 Analytical 1. (In the case of tissue or cell pellets available in large quantity, if
Protocol this step is not carried out before quenching)
Intact or Lyophilized Weigh the intact tissue [11, 12] or the selected amount of
Tissues or Cell Pellets lyophilized tissue [8–10] or cell pellet (if a large quantity is
available, see industrial cell culture application [13]) rapidly
3.1.1 Extraction of Polar before it thaws; the weight is required for the addition of the
Metabolites from appropriate amount of methanol/water and internal standards
below.
(In the case of cell pellet from a certain culture volume, e.g.,
from one petri dish culture—small quantity available)
The amount of cells in each sample is estimated from the

quantified protein content (need for the application of a protein
content quantification method)—if all samples originate from
the same culture volume, it is usually opted to add the same
amount of internal standards and methanol/water in excess to
all and then normalize the profiles based on the respective
protein content at the normalization stage (see Note 2).
2. Transfer the tissue or cell pellet (available in large quantity)
sample to a glass homogenizer, and add ice-cold HPLC grade
methanol in a ratio of 22 mL per g of sample (e.g., add 2.2 mL
of cold methanol to 100 mg of tissue) [8, 9].
(In the case of cell pellet from a certain culture volume, e.g.,
from one petri dish culture—small quantity available)
Add 0.5–1 ml methanol to the cell pellet from one petri
dish of mammalian cell cultures (protein content ~5 mg pro-
tein) directly into the Eppendorf-type tube (as transfer in
these cases is not preferable), and use a pellet micropestle for
its homogenization (step 5 below).
3. Add internal standards in a ratio of 0.1 μg adonitol/ribitol and
0.2 μg of U-13C-d-glucose per 1 mg of tissue/cell pellet/pro-
tein content.
4. Homogenize the sample thoroughly.
5. In the case that the glass homogenizer has been used, transfer
the homogenate to a 2 mL Eppendorf- or a 15 mL Falcon-
type tube (depending on the sample volume).
6. Incubate in water bath at 70 °C for 15–20 min.
7. Add HPLC grade water in equal volume to the methanol ini-
tially added in step 2, and mix gently, turning the tubes upside
down to enable the mixing of both phases.
8. Centrifuge the samples at 10,000 × g for 10 min at 4 °C.
9. Transfer the supernatant of each sample into a clean 2 mL
Eppendorf- or a 15 mL Falcon-type tube (depending on the
supernatant volume).
10. Repeat centrifugation as many times as required until no pre-
cipitate is visible.
11. Transfer all the supernatant into a high-recovery autosampler
glass vial (if the supernatant volume is small), or, if possible,
divide it into 2–4 replicates.
12. Vacuum dry the supernatant samples.
13. Cap each vial with a PTFE crimp cap and store at 4 °C until
further analysis.
Blood Plasma or Other 1. Add 200 μL of blood plasma into a 2 mL Eppendorf tube con-
Liquid (e.g., Culture taining 0.6 mL of ice-cold methanol and 2 μg of adonitol/
Medium) Samples ribitol and 4 μg of U-13C-d-glucose as internal standards (see
Note 3).
2. Mix the sample.

3. If transfer is required, the samples could be transported in ice as
proteins have been denatured by methanol.
4. Follow steps 6–13 as described previously in the section for
intact or lyophylized tissues or cell pellets.
3.1.2 TMS Derivatization 1. Vacuum dry each stored at 4 °C vial for 30 min to remove any
remaining humidity (see Note 4).
2. Add 50 μL of 20 mg/mL solution of methoxylamine hydro-
chloride in pyridine (see Notes 5 and 6) using glass pipette tips,
and mix gently (see Notes 4 and 6); the volume of the solution
to be added may vary from 30 μL for very small amounts of dry
extracts to 150 μL for larger samples—preliminary experiments
should be carried out to determine the optimal volume for a
particular sample type balancing between the need for in excess
availability of the derivatization reagents and avoiding any large
dilution of the dry extract [2, 6].
3. Incubate in a shaker incubator at 40 °C for 90 min.
4. Add MSTFA at twice the volume of the methoxylamine hydro-
chloride solution added in step 2 using glass pipette tips and
mix gently.
5. Incubate at 40 °C for at least 6 h. This is the derivatization time
at which all metabolites in a sample are estimated to have been
fully transformed into at least one of their TMS derivatives [2,
6]. This time may vary between biological systems and could be
optimized through a preliminary experiment at which the met-
abolic profile of a certain system is measured multiple times
from 15 min to 10 h of derivatization duration [2]. However,
based on our group’s experience with various systems, 6 h of
derivatization could be considered a “universal” time for most
biological sample types for the standardization of this step of
the protocol.
6. Place the vial on the autosampler for the sample to be used for
metabolic profile acquisition.
Significant note: In an optimized GC-MS metabolic profile
acquisition protocol, taking into consideration the constraint of
step 5, four [4] samples can be quantified per day with three
repetitions per sample and three runs of the solvent (pyridine) to
clean the column between samples (see the acquisition section
below). In this optimized protocol, the incubation time of step 5
is equal to 9 h.
3.1.3 GC-MS Metabolic 1. Set the injector temperature to 230 °C and the detector transfer
Profile Acquisition line, trap, and manifold temperatures to 250 °C, 220 °C, and
70 °C, respectively. These parameters can be appropriately
adapted in a GC-quadrupole MS.
2. Set the helium carrier gas flow rate to 1 mL/min.

3. Program the GC oven temperature gradient as follows: set ini-
tial oven temperature at 70 °C, hold for 5 min, and then increase
temperature up to 310 °C at a rate of 5 °C/min. Finally, hold
for 3 min. The total run time is 56 min.
4. Set the split ratio to the value optimized according to the
expected metabolite concentration range in the samples under
investigation. A split ratio between 1:40 and 1:25 is suitable for
most sample types; it requires optimization through prelimi-
nary tests.
5. Operate the electron ionization source at −70 eV.
6. Operate the mass spectrometer in scan mode over a mass range
of 50–600 m/z; start mass spectrum acquisition 4 min after the
initialization of the run to allow for the solvent to be eluted.
7. Inject 1 μL from a sample to acquire its metabolic profile (see
Note 7).
8. Acquire the metabolic profile of each sample at least twice in
two consecutive runs. At least two runs of solvent (pyridine)
should be carried out between biological samples to clean the
column. Figure 1 shows the profiles of a baby hamster kidney
(BHK) cell culture [14], C57BL/6J male mouse cortex, and
tomato leaf sample at 11 h of derivatization.
9. The metabolic profile of at least one sample per experi-
mental batch (optimally one per each week of runs) should be
quantified at least five times at derivatization durations from 9 h
to 15 h. These measurements are required for the application of
specialized normalization methods on the profiles of the amine
group-containing metabolites (see below).
3.2 Computational 1. Collect the files storing the MS-reconstructed chromatograms

Protocol of the metabolic profiles of the various samples, and proceed to
peak identification and quantification using a relevant software,
3.2.1 Peak Identification usually associated with the available GC-MS equipment.
and Quantification Special note : A “metabolite-centric” compared to a “feature-
centric” approach in the analysis of the MS-reconstructed
chromatograms is recommended [2, 18]. It is more biologically
relevant, enabling the identification and exclusion of artifacts at the
normalization and filtering step (see below). In the “feature-
centric” approach, one identifies and quantifies every ion peak in
the MS-reconstructed chromatogram, while in the “metabolite-
centric” approach, the profile is scanned based on a standardized
database of peaks, and each metabolite derivative is characterized
and quantified based on one of its ion peaks, called “marker or
quantifying ion.” One can use feature-scanning software to auto-
matically scan the profile and then filter the data based on the
metabolite database or scan the profile only for the peaks in the
Fig. 1 Total ion count MS-reconstructed gas chromatograms for polar metabolite extracts (left panel) and the
respective mass spectrum of the lactate 2TMS derivative (right panel) of samples from (a) industrial-scale BHK
cell culture [14], (b) C57BL/6 J male mouse cortex, and (c) tomato leaf. The straight arrow shows the peak of
the lactate 2TMS derivative in each chromatogram. The metabolites corresponding to some major peaks are
also denoted. In all three samples, the internal standard ribitol has been added in the same relative concentra-
tion with respect to the sample weight (see text)
database based on an accordingly developed method. Involvement

of an expert user understanding the biochemistry of the investi-
gated biological system and the specifics of the GC-MS analysis is
required at this stage to appropriately guide and supervise the peak
identification and quantification process.
2. Merge all profiles into one file that can be easily presented and
computationally processed. It is advisable that the data are sub-
mitted to standardized repositories to enable meta-analysis
studies. MetaboLights (http://www.ebi.ac.uk/metabolights/)
emerges as the reference repository for metabolomic data.
3.2.2 Data Normalization 1. Investigate whether all profiles were acquired at the same ana-
and Filtering lytical process conditions by estimating the ratio of the two
methoxime peaks of [U-13C]—glucose used as internal stan-
dard. If available, the ratio of the two methoxime peaks of
other metabolites could also be estimated. If these ratios

remain constant among the samples, subject only to random
variance, then constant conditions can be validated through-
out the experimental process, and the profiles are directly
comparable [2, 6]. If large deviation is indicated in any pro-
files, then these are excluded from further analysis. This qual-
ity control criterion does not require the availability of any
quality control (QC) samples and can be used post-experi-
mentally to evaluate the comparability of profiles that have
been acquired at different experimental batches.
2. Normalize the marker ion peak area of the detected com-
pounds with the marker ion peak area of the internal standard.
Ribitol has been proven a reliable internal standard for most
biological systems and applications, as it is not endogenously
produced or in very small quantities that do not affect the
quantification of the exogenously provided standard [2, 6, 8–
15]; the latter statement requires verification in every new sys-
tem under investigation. Among the two marker ions of ribitol
(217 or 319), the latter is usually preferred at this normaliza-
tion step as its smaller peak area does not bias the estimation of
the relative peak areas (RPAs) of the least abundant metabo-
lites toward very small values that can “skew” the subsequent
data analysis.
3. Estimate the cumulative (effective) peak area of the known
amine group-containing metabolites as the weighted sum of
their derivative peak areas [2, 6]. The weights are estimated
from the profiles of the samples that were quantified at least five
times within 9 h and 15 h of derivatization (see step 9 in GC-MS
metabolic profile acquisition), based on the algorithm described
in [2, 6]; see Fig. 2.
4. Filter out of the dataset technical and mathematical artifacts to
avoid skewing the subsequent data analysis. Specifically, exclude:
(a) Peaks with small signal-to-noise ratio
(b) Peaks with significant carryover
(c) Technical artifacts from column bleeding and/or reagents
(see Note 7)
eaks that are inconsistently detected among samples
(d) P
(with large number of missing values)
eaks with mean variation among technical replicates in all
(e) P
samples that is greater than 30%
B
11 1
10
0.9
9
with respect to internal standard ribitol
0.8
8
Relative Peak Area (RPA)
7 0.7
6
0.6
5
4 0.5
3
0.4
2 Aspartate_3TMS
Aspartate effective 0.3
1
Aspartate_2TMS
0 0.2
9 11 13 15 17 19 21
Derivatization duration (h)
Fig. 2 (a) The algorithm for the estimation of the weights for the derivative peak areas of an amine group
containing metabolite required for the estimation of its effective peak area in each sample (see equation at the
bottom) [2, 6]; (b) the profile of the relative peak areas (RPAs) of the two derivatives and of the estimated effec-
tive RPA of aspartate over ribitol (internal standard) in a Balbc/J male mouse cortex sample over eight [8]
different derivative durations. In (a), the algorithm is shown for a metabolite with three [3] derivatives (derivi)
and a normalization sample, the profile of which was acquired at five different derivatization durations (ti)—the
algorithm should be appropriately adjusted for different conditions. The “b” constant is defined in the same
order of magnitude either of the largest derivative or the sum of derivatives (if different) in the investigated
profiles, as explained in [2, 6]
eaks corresponding to unknown amine group containing

(f) P
metabolites, as it is not possible to estimate their effective
peak area—see point 3 above—and are subject to
derivatization biases [2, 6]
he peak of the least abundant methoxime derivative of any
(g) T
ketone group containing metabolite of known identity, if
both methoximes have been quantified; as the abundances of
the two methoximes of a ketone group containing metabo-
lite are linearly dependent, inclusion of both peak areas in
the final dataset will introduce bias in data analysis [2, 6].
After this normalization and filtering step, the final dataset
concerns peak areas of metabolites and not metabolite derivatives
(the only relevant deviation may be introduced from the presence
of unknown ketone group containing metabolites for which both
methoxime peaks have been identified; however, such pairs are
usually detectable despite unknown identity, and filtering of one of
the two occurs in step g above). Our group has recently developed
a software platform, called M-IOLITE [7], including a module for
GC-MS metabolomic data normalization and filtering as previ-
ously described (available upon request for academic users at mio-
lite.iceht.forth.gr).
3.2.3 Multivariate 1. The final metabolite profile dataset after normalization and fil-
Statistical Analysis tering can be used for further analysis to extract biologically
and Metabolic Network relevant conclusions. As for other omic profiles, multivariate
Reconstruction statistical analysis methods can be applied to identify correla-
tions between metabolites and/or physiological conditions.
Specifically, supervised or unsupervised clustering algorithms
can group metabolites and/or samples of similar profiles based
on a variety of profile distance metrics (e.g., hierarchical clus-
tering analysis (HCL) [19]). Moreover, methods like principal
component analysis (PCA) [20] or partial least squares (PLS)
regression [21] can be used to lower the dimensionality of the
problem, enabling the visualization of the differences between
the profiles in a 3-D space, providing also information about
the metabolites that contribute significantly to these differ-
ences. The metabolites with significantly differential concen-
tration between sets of metabolic profiles can be identified
with the multivariate significance analysis algorithm, called
significance analysis of microarrays (SAM) [22]. SAM has been
appropriately tuned for omic data, since it does not require
that they follow a particular distribution (as the t-test or F-test)
and estimated the false discovery rate for each threshold of
significance.
There exist many software platforms, publicly or commer-
cially available, which incorporate many of these bioinformatics
and data mining algorithms, some built around an interface that
is friendly for a nonexpert in computational analysis. The open-

source software TM4 Multiple Experiment Viewer (MeV)
(http://mev.tm4.org) provides a user-friendly platform with a
big gallery of multivariate statistics algorithms. MATLAB
(https://www.mathworks.com/products/matlab/) has also
developed an extensive bioinformatics toolbox, which com-
bines the functionality of the MATLAB environment and lan-
guage. Many developers and computational users opt for the
use of relevant modules in R language (R Project for Statistical
Computing; https://www.r-project.org/), most biologically
relevant having been incorporated in the Bioconductor open-
software suite (https://www.bioconductor.org/).
2. A major step in metabolomic data analysis is the interpretation
of the results in the context of metabolic network structure and
regulation. Fortunately, compared to biomolecular networks at
the transcriptional and translational levels, there exists extensive
knowledge about the structure and function of metabolic path-
ways. This information enables the estimation of the relative
change between pathway fluxes even in the cases that not all
metabolites in the pathways are quantified. Information for the
reconstruction of the metabolic pathways for different organ-
isms or tissues can be obtained from metabolic databases, e.g.,
KEGG (Kyoto Encyclopedia of Genes and Genomes)—
GenomeNet (http://www.genome.jp/kegg/); ExPASy: SIB
Bioinformatics Resource Portal (https://www.expasy.org/);
and BioCyc Database Collection (http://biocyc.org/), or
metabolomic databases, e.g., HMDB (Human Metabolome
Database) (http://www.hmdb.ca/). Moreover, there exist
genome-scale reconstructed metabolic networks for various
organisms, such as Recon 2 for the human (http://www.ebi.
ac.uk/biomodels-main/MODEL1603150001), the mouse
(http://www.ebi.ac.uk/compneur-srv/biomodels-main/
MODEL1507180055), and various tissues, such as the brain
[2, 11, 12].
4 Notes
1. Samples can be stored at −80 °C for at least 6 months; actually

the maximum storage duration that excludes metabolite degra-
dation and changes in the acquired metabolic profiles is cur-
rently under investigation, especially in connection with
biobanking. Avoid any sample thawing before the samples are
processed for extraction. It would be preferable if the samples
are extracted and the metabolite extracts can be stored for lon-
ger times at 4 °C.
2. For fibrous samples, like plant leaves, grinding the sample to

powder with mortar, pestle, and liquid nitrogen prior to extrac-
tion is advised.
3. EDTA (and not heparin) should be used as anticoagulant for
blood sample acquisition; it is considered to affect the composi-
tion of the sample less and its peak can be isolated from the rest
of the metabolic profile without affecting the measurements of
other metabolites. Citrate vacutainers should not be used.
4. MSTFA is not active for derivatization in the presence of water.
Please make sure that the metabolite extracts are appropriately
dried and MSTFA has been appropriately stored. Moreover, all
reagents used for the derivatization of the extracts should be
anhydrous.
5. Pyridine and MSTFA are volatile and hazardous. They should
be handled quickly in a fume hood applying proper
precautions.
6. Pyridine reacts with plastic. Avoid any use of plastic consum-
ables for handling and storing the samples in all protocol steps
after the initialization of pyridine use. The caps of the autosam-
pler vials should contain only PTFE and not silicone, as the
latter can react with pyridine and contaminate the samples.
7. Before each experiment, quantify the profile of a negative con-
trol sample to identify the peaks and their areas that originate
from the reagents and the consumables used in the process
(including column bleeding), so that they can be excluded from
the metabolic profile and further analysis.
References
1. Fiehn O (2002) Metabolomics–the link samples from mice on low- and high-fat diets.
between genotypes and phenotypes. Plant Mol J Chromatogr B 879(17–18):1467–1475.
Biol 48(1–2):155–171. https://doi. https://doi.org/10.1016/j.
org/10.1007/978-94-010-0448-0_11 jchromb.2011.01.028
2. Kanani H, Chrysanthopoulos PK, Klapa MI 6. Kanani HH, Klapa MI (2007) Data correction
(2008) Standardizing GC-MS metabolomics. strategy for metabolomics analysis using gas
J Chromatogr B 871(2):191–201. https:// chromatography-mass spectrometry. Metab
doi.org/10.1016/j.jchromb.2008.04.049 Eng 9(1):39–51. https://doi.org/10.1016/j.
3. Patti GJ, Yanes O, Siuzdak G (2012) ymben.2006.08.001
Innovation: metabolomics: the apogee of the 7. Maga-Nteve C, Klapa MI (2016) Streamlining
omics trilogy. Nat Rev Mol Cell Biol GC-MS metabolomic analysis using the
13(4):263–269. https://doi.org/10.1038/ M-IOLITE software suite. IFAC-
nrm3314 PapersOnLine 49(26):286–288. https://doi.
4. Vasilopoulou CG, Margarity M, Klapa MI org/10.1016/j.ifacol.2016.12.140
(2016) Metabolomic analysis in brain research: 8. Dutta B, Kanani H, Quackenbush J, Klapa MI
opportunities and challenges. Front Physiol (2009) Time-series integrated “omic” analyses
7:183. https://doi.org/10.3389/ to elucidate short-term stress-induced
fphys.2016.00183 responses in plant liquid cultures. Biotechnol
5. Spagou K, Theodoridis G, Wilson I et al (2011) Bioeng 102(1):264–279. https://doi.
A GC-MS metabolic profiling study of plasma org/10.1002/Bit.22036
9. Kanani H, Dutta B, Klapa MI (2010) Individual early indications for predisposition to latent
vs. combinatorial effect of elevated CO2 condi- insulin resistance in children conceived by
tions and salinity stress on Arabidopsis thaliana ICSI. PLoS One 9(4):e94001. https://doi.
liquid cultures: comparing the early molecular org/10.1371/journal.pone.0094001
response using time-series transcriptomic and 16. Saeed AI, Bhagabati NK, Braisted JC et al
metabolomic analyses. BMC Syst Biol 4:177. (2006) TM4 microarray software suite.
https://doi.org/10.1186/1752-0509-4-177 Methods Enzymol 411:134–193. https://doi.
10. Tooulakou G, Giannopoulos A, Nikolopoulos org/10.1016/S0076-6879(06)11009-5
D et al (2016) “Alarm photosynthesis”: cal- 17. Saeed AI, Sharov V, White J et al (2003) TM4:
cium oxalate crystals as an internal CO2 source a free, open-source system for microarray data
in plants. Plant Physiol 171(4):2577–2585. management and analysis. BioTechniques
https://doi.org/10.1104/pp.16.00111 34(2):374–378
11. Constantinou C, Chrysanthopoulos PK, 18. Allwood JW, Erban A, de Koning S et al (2009)
Margarity M, Klapa MI (2011) GC-MS metab- Inter-laboratory reproducibility of fast gas
olomic analysis reveals significant alterations in chromatography-electron impact-time of flight
cerebellar metabolic physiology in a mouse mass spectrometry (GC-EI-TOF/MS) based
model of adult onset hypothyroidism. plant metabolomics. Metabolomics 5(4):479–
J Proteome Res 10(2):869–879. https://doi. 496. https://doi.org/10.1007/
org/10.1021/pr100699m s11306-009-0169-z
12. Maga-Nteve C, Vasilopoulou CG, 19. Eisen MB, Spellman PT, Brown PO, Botstein
Constantinou C et al (2017) Sex-comparative D (1998) Cluster analysis and display of
study of mouse cerebellum physiology under genome- wide expression patterns. Proc Natl
adult-onset hypothyroidism: the significance of Acad Sci U S A 95(25):14863–14868. https://
GC-MS metabolomic data normalization in doi.org/10.1073/pnas.95.25.14863
meta-analysis. J Chromatogr B 1041- 20. Raychaudhuri S, Stuart JM, Altman RB (2000)
1042:158–166. https://doi.org/10.1016/j. Principal components analysis to summarize
jchromb.2016.12.016 microarray experiments: application to sporula-
13. Chrysanthopoulos PK, Goudar CT, Klapa MI tion time series. Pac Symp Biocomput
(2010) Metabolomics for high- resolution 2000:455–466
monitoring of the cellular physiological state in 21. Maitra S, Yan J (2008) Principle component
cell culture engineering. Metab Eng analysis and partial least squares: two dimen-
12(3):212–222. https://doi.org/10.1016/j. sion reduction techniques for regression. In:
ymben.2009.11.001 2008 Casualty actuarial society discussion
14. Vernardis SI, Goudar CT, Klapa MI (2013) paper program–applying multivariate statistical
Metabolic profiling reveals that time related models. Casualty Actuarial Society, Quebec,
physiological changes in mammalian cell perfu- pp 79–90
sion cultures are bioreactor scale independent. 22. Tusher VG, Tibshirani R, Chu G (2001)
Metab Eng 19:1–9. https://doi. Significance analysis of microarrays applied to
org/10.1016/j.ymben.2013.04.005 the ionizing radiation response. Proc Natl Acad
15. Gkourogianni A, Kosteria I, Telonis AG et al Sci U S A 98(9):5116–5121. https://doi.
(2014) Plasma metabolomic profiling suggests org/10.1073/pnas.091062498
Chapter 10
Rat Fecal Metabolomics-Based Analysis

Olga Deda, Helen G. Gika, and Georgios A. Theodoridis
Abstract
Fecal metabolomics-based analysis indisputably constitutes a very useful tool for elucidating the biochem-
istry of digestion and absorption of the gastrointestinal system. Fecal samples represent the most suitable,
non-invasive, specimen for the study of the symbiotic relationship between the host and the intestinal
microbiota.
It is well established that the balance of the intestinal microbiota changes in response to some stimuli,
physiological such as gender, age, diet, exercise and pathological such as gastrointestinal and hepatic dis-
ease. Fecal samples have been analyzed using the most widespread analytical techniques, namely, NMR
spectroscopy, GC-MS, and LC-MS/MS. Rat fecal sample is a frequently used and particularly useful sub-
strate for metabolomics-based studies in related fields. The complexity and diversity of the nature of fecal
samples require careful and skillful handling for the effective quantitative extraction of the metabolites
while avoiding their deterioration. Parameters such as the fecal sample weight to extraction solvent vol-
ume, the nature and the pH value of the extraction solvent, and the homogenization process are some
important factors for the optimal extraction of samples, in order to obtain high-quality metabolic finger-
prints, using either untargeted or targeted metabolomics.
Key words Metabolomics, Sample preparation, Fecal samples, Rats, NMR, GC-MS, LC-MS/MS,
Fecal extract
1 Introduction
Fecal sample is a useful bio-specimen in chemical/biochemical

point of view. Fecal sample analysis is a powerful tool in prognosis
and diagnosis for a plethora of gastrointestinal diseases [1, 2]. Raw
fecal samples require appropriate preparation in order to produce
reliable conclusions [3].
Sample preparation is a very critical aspect in metabolomics-
based analysis and can have significant effects on the quality of the
obtained results [4–6]. The peculiarities of the fecal sample in con-
trast with common samples include complexity, heterogeneity, and
a high concentration of nondigested macromolecules [3, 7, 8].
149
150 Olga Deda et al.
The composition of the fecal sample is largely influenced by

dietary factors and could reflect the nutrient’s metabolism by gut
microbiota [9]. Metabolomics-based analysis provides a detailed
picture of the fecal sample metabolome, which is derived from co-
metabolism of gut microbiota and the host organism. Fecal sam-
ples should undergo special treatment, due to their nature, for the
avoidance of negative effects on the analytical systems.
Fecal sample preparation is a determining factor for the quality
of the analysis, regardless of the analytical technique subsequently
applied. Inappropriate sample preparation could lead to ineffec-
tive, non-repeatable, and poor extraction, thus affecting the
obtained metabolic profile. The golden rule is a fast and repeat-
able, generic (nonselective) sample preparation process, capable of
extracting metabolites of different chemical classes. Each step of
this process is of great importance.
Undoubtedly, the ratio of fecal sample weight to extract sol-
vent volume has the greatest impact [8, 10, 11]. However, this
does not necessarily mean that denser samples lead to better results,
because there is a possibility that high density will create interfer-
ences in the analytical system.
Solvent extraction buffer is also a very strong factor affecting
the fecal metabolic profiling obtained by the most commonly used
analytical techniques, namely, nuclear magnetic resonance (NMR)
spectroscopy, gas chromatography-mass spectrometry (GC-MS),
and liquid chromatography-mass spectrometry (LC-MS/MS).
The nature of the extraction solvent and the pH value of the extrac-
tion solvent strengthen or weaken the extraction of particular
metabolites. For example, inappropriate choice of extraction sol-
vent solution pH value may lead to the deterioration of fecal
extracts through hydrolysis of some metabolites and affects the
extraction capability of ionizable metabolites.
Furthermore, homogenization, filtration, and centrifugation
are also parameters which can prove critical [12]. Homogenization
by sonication or mechanical smashing, by using a mechanical
crusher, or TissueLyser leads to better results, especially in NMR-
based metabolomics [11, 12]. Disaggregating fecal sample prior to
the homogenization step could also enhance extraction efficacy [8].
Although filtration is not as important a parameter as the pre-
viously mentioned processes, with regard to affecting the fecal
metabolic profiling, it is a proper (if not necessary) practice for
removing particulates thus protecting the analytical system, espe-
cially in LC-MS/MS-based metabolic profiling [13]. In order to
remove particulates, usually, two cycles of centrifugation are
required: the first in the early steps of the process and the second
immediately before the subsequent analysis.
Another practice, rather controversial, in fecal sample prepara-
tion is lyophilization. Although lyophilization eliminates the water
content [14], its use is not recommended as it has been proven to
Rat Fecal Metabolomics-Based Analysis 151
affect the obtained fecal metabolic profiling. For example, it could

lead to reduced ammounts of short-chain fatty acids (SCFAs) and
potentially of other volatile compounds [12, 15], thus editing the
profile especially in GC-MS mode.
In addition, more than one extraction cycle could be per-
formed, and the combination of extracts could take place [11].
This practice is not recommended for the sake of simplicity of the
sample preparation process.
Finally, depending on the technique by which fecal samples are
analyzed, other conditions and parameters may affect the analyti-
cal performance. For example, in NMR spectroscopy-based metab-
olomics analysis, PBS buffer preparation (preparation in D2O, pH
value, and salt addition) appreciably reflects the quality of the col-
lected spectra. The pH value of PBS may significantly affect the
behavior of the analytes and the chemical shift reference used.
Undoubtedly, for GC-MS based analysis the derivatization
process of the fecal sample is a major determining factor. Critical
factors such as residual moisture, insufficient derivatization
reagents, and shorter or longer incubation, at lower or higher tem-
perature, greatly affect the obtained chromatograms [16, 17].
Selection of an extraction solvent, similar to the mobile phase,
could result in better peaks in liquid chromatography, while very
concentrated fecal extract could block the column and prevent fur-
ther analysis.
2 Materials
2.1 Reagents 1. 1-propanol (LC-MS grade).

2. Acetonitrile (LC-MS grade).
3. Methanol (LC-MS grade).
4. Methanol (HPLC grade) (see Note 1).
5. Deuterium oxide (D2O) 99.96%.
6. Sodium dihydrogen phosphate monohydrate (H2O·NaH2PO4).
7. Disodium hydrogen phosphate anhydrous (Na2HPO4) (see
Note 2).
8. Sodium chloride, granular/USP/FCC (see Note 3).
9. Pyridine anhydrous 99.8%.
10. Methoxyamine hydrochloride (MeOX).
11. N-methyl-N-(trimethylsilyl)trifluoroacetamide (MSTFA).
12. Trimethylchlorosilane (TMCS).
13. Trimethylsilylpropanoic acid (ΤSP) (sodium salt of deuterated
molecule).
14. Ammonium formate, for MS ≥99.0% (see Note 4).

15. Formic acid eluent additive for LC-MS/MS (see Note 5).
16. Η2Ο (LC-MS/MS grade).
2.2 Laboratory 1. −80 °C freezer.

Equipment 2. Millipore purification system for water.
3. Electronic analytical balance.
4. Mini shaker laboratory vortex.
5. TissueLyser.
6. Ultrasonic tissue processor.
7. Micro refrigerated centrifuge.
8. Pure nitrogen gas generator.
9. Adjustable variable volume pipettes of 1000, 100, 20, and
10 mL and pipette tips.
10. Eppendorf tubes of 2 mL.
11. Screw cap glass vials of 1.5 mL.
12. Syringes of 2.5 mL.
13. 25 mm diameter sterile syringe filters PTFE with 0.22 μm
pore size.
14. NMR sample tubes, 5 mm, 7″ length (or appropriate size for
the spectrometer probe used).
15. GC-MS vials, caps, and insert vials 250 μL.
16. LC-MS/MS glass vials, caps with PTFE/silicone septa, and
low-volume inserts.
2.3 Instrumentation 1. 500 MHz (or higher) NMR spectrometer equipped with a
Software 5 mm triple resonance probe at 300 K (or similar)—the appro-
priate software to control acquisition of fecal sample 1H NMR
spectra (matching, tuning, shimming), to set pulse calibration
parameters, and to process.
2. Agilent 7890A GC coupled to a 5975C inert XL EI/CI MSD
with triple-axis detector MS and a CTC-CH 4222 autosampler
with an Agilent HP-5ms (29 m × 250 μm × 0.25 μm) in split/
splitless mode. MSD ChemStation (Agilent Technologies,
California, USA) to acquire and process GC-MS data.
3. ACQUITY UPLC coupled to a Xevo TQD MS system (Waters,
Massachusetts, USA) with an ACQUITY HILIC, BEH amide
column (2.1 × 150 mm, 1.7 μm). Waters MassLynx® software
to collect and process LC-MS/MS data.
3 Methods
3.1 Sample 1. Scientists should use personal protective equipment when

Collection and Storage handling experimental animals.
2. Each animal should be placed in a separate cage during sample
collection, in case where the animals are kept housed in non-
individual cage (see Note 6).
3. Fecal pellets should be collected immediately after defecation
to limit sample deterioration.
4. Fecal pellets should be collected using tweezers, which should
be cleaned from sample to sample.
5. More than one fecal pellet should be collected, in order to
subsequently receive the necessary volume of fecal extract, to
pass through the filter, in sample preparation.
6. Fecal pellets should be placed in 2 mL Eppendorf tube.
7. Sodium azide could be added in order to protect fecal samples
from microbial contamination.
8. Immediately after collection, the fecal samples should be fro-
zen at −80 °C, for long-term storage.
9. Fecal pellets could be subjected to snap freezing with liquid
nitrogen, as this process efficiently inhibits biochemical
interactions.
10. In cases where sample collection is directly from the intestine,
rats should be anesthetized and sacrificed in accordance to the
Helsinki Declaration for keeping and handling experimental
animals.
3.2 Sample 1. 500 mg of smashed stool material is placed in a 2 mL Eppendorf

Preparation tube.
3.2.1 Analysis Using 2. 1 mL phosphate-buffered saline (PBS 1.9 mM Na2HPO4,
NMR Spectroscopy 8.1 mM NaH2PO4, 150 mM NaCl, pH 7.4) is added, in a ratio
of 1:2 fecal sample weight to extraction buffer volume (see Note
7) [8]. PBS could be prepared either in D2O or in distilled
water. D2O should be added at least in a ratio of 1:5 (v/v) and
then made up to final volume with distilled water. NaCl is added
to facilitate the extraction of fecal samples. When preparing PBS
the dissolution οf the salts requires vortexing and sonication.
3. Vortex-mixing is performed for 2 min.
4. The mixture is homogenized using an ultrasonic homogenizer
for 15 min.
5. The fecal slurry is centrifuged for 20 min at 4 °C (18,000 × g).
6. 400 μL of the clear supernatant is diluted with 150 μL of D2O.
7. 50 μL of 0.1% 3-(trimethylsilyl)propionic-2,2,3,3-d4 acid

sodium salt (TSP) in D2O is added to the extract.
8. Centrifuge for 18 min at 4 °C (15,000 × g).
9. Finally 550 μL of the clear extract is placed in 5 mm NMR tube
for analysis using methods such as those described in Chap. 8
(Benaki and Mikros), in the present book.
3.2.2 Analysis Using 1. 500 mg of smashed stool material is placed in a 2 mL Eppendorf

GC-MS tube.
2. 1 mL of MeOH-CHCl3 1:1 (v/v) is added, in a ratio of 1:2
fecal sample weight to extraction buffer volume.
3. Vortex-mixing is applied for 2 min.
4. The mixture is homogenized by sonication for 10 min.
5. The organic fecal extract is centrifuged for 20 min at 4 °C
(18,000 × g).
6. 100 μL of the supernatant is placed in a glass GC-MS vial and
evaporated under a stream of nitrogen.
7. 20 μL methoxyamine hydrochloride in pyridine (40 mg/mL)
and 180 μL MSTFA [1% trimethylchlorosilane (TMCS)] are
added followed by incubation for 90 min at 28 °C and for
30 min at 37 °C, respectively (see Note 8).
8. 5 μL of a 1 mM solution of 2-fluorobiphenyl is added as an
internal standard. An alternative option is the addition of 5 μL
pentadecane (in pyridine). Before the addition of internal stan-
dard the vial should be allowed to fall to ambient temperature
(see Note 9).
9. The gas chromatographic analysis (GC-MS) of the fecal extracts
is performed as described in detail, in a previously published
paper [8].
3.2.3 Sample 1. 250 mg of smashed stool material is placed in a 2 mL Eppendorf

Preparation for HILIC-MS tube.
2. Aqueous extraction solvent, water with 1-propanol or acetoni-
trile (1:1 v/v), is added, in a ratio of 1:4 fecal sample weight to
extraction solvent volume (see Note 10) [8, 13]. Acetonitrile
extracts a great wide of metabolites, but 1-propanol preferably
extracts amino acids related to bowel disease.
3. Vortex-mixing is applied for 2 min.
4. The mixture is homogenized by sonication for 10 min.
5. Ultra-centrifugation of the samples is performed 30 min at 4 °C
(20,000 × g).
6. Supernatants are filtered through syringe filters (PTFE 0.22um).
Fig. 1 Rat fecal sample preparation protocols for metabolomics-based analysis using NMR spectroscopy,
GC-MS, and LC-MS/MS
7. The filtrate should be collected in an Eppendorf tube and not

immediately into the LC-MS vial, which has a narrower
opening.
8. The required sample volume, depending on the number of sub-
sequent injections, is added to the LC-MS vial insert prior to
analysis.
9. Analyze by targeted HILIC-MS/MS with a multi-analyte
method [e.g., as the one described in Chap. 5 (Virgiliou et al.),
in the present book and in a previous published paper [18]] or
an untargeted method [e.g., as described in Chap. 7 (Want), in
the present book].
The applied protocols for each analytical technique used are pre-
sented briefly in Fig. 1.
4 Notes
1. Used only for cleaning the analytical equipment.

2. Preparation of PBS, as already described in Chap. 8 (Benaki
and Mikros) and 14 (Spyros et al), in the present book.
3. 25 mg NaCl could be added for every 10 mL PBS buffer to
improve extraction efficacy.
4. Used for the preparation of mobile phases in LC-MS/MS sys-
tem as described in Chap. 5, in the present book.
5. Used for the preparation of wash solvent and purge solvent of
LC-MS/MS systems as described in Chap. 5, in the present
book.
6. The rats in each cage should be appropriately marked so that
they are effectively identified, which is necessary for the collec-
tion of their samples. Herbal hair dye can be used for the
marking and should be renewed at regular intervals (approxi-
mately every 15 days).
7. Higher extraction ratios can be used (such as 1:3) in cases of
insufficient fecal sample quantity, in order to ensure a suffi-
cient aliquot of fecal extract.
8. For a more effective derivatization process, the sample should
be placed in a glass insert in GC-MS glass vial.
9. Quick handling is required during the addition of the internal
standard, since the volatile derivatives could be evaporated
while opening the cap.
10. Lower extraction ratios can be used, such as 1:3 or 1:2, in
combination with supervision of the analytical system in order
to prevent problems in case of analysis of concentrated extracts.
References
1. Guinane CM, Cotter PD (2013) Role of the gut ing of grapes: solvent extraction protocol opti-
microbiota in health and chronic gastrointestinal misation. Metabolomics 8(2):175–185
disease: understanding a hidden metabolic 5. Gika H, Theodoridis G (2011) Sample prepa-
organ. Ther Adv Gastroenterol 6(4):295–308 ration prior to the LC-MS-based metabolo-
2. Holmes E, Li JV, Athanasiou T, Ashrafian H, mics/metabonomics of blood-derived samples.
Nicholson JK (2011) Understanding the role Bioanalysis 3(14):1647–1661
of gut microbiome-host metabolic signal dis- 6. Gika HG, Wilson ID, Theodoridis GA (2014)
ruption in health and disease. Trends Microbiol LC–MS-based holistic metabolic profiling.
19(7):349–359 Problems, limitations, advantages, and future
3. Deda O, Gika HG, Wilson ID, Theodoridis perspectives. J Chromatogr B 966:1–6
GA (2015) An overview of fecal sample prepa- 7. Bollard ME, Stanley EG, Lindon JC et al
ration for global metabolic profiling. J Pharm (2005) NMR-based metabonomic approaches
Biomed Anal 113:137–150 for evaluating physiological influences on bio-
4. Theodoridis G, Gika H, Franceschi P et al fluid composition. NMR Biomed
(2012) LC-MS based global metabolite profil- 18(3):143–162
8. Deda O, Chatziioannou AC, Fasoula S et al 14. Bezabeh T, Somorjai RL, Smith IC (2009)
(2017) Sample preparation optimization in ICP MR metabolomics of fecal extracts: appli-
fecal metabolic profiling. J Chromatogr B cations in the study of bowel diseases. Magn
1047:115–123 Reson Chem 47(S1):S54–S61
9. Hooper LV, Midtvedt T, Gordon JI (2002) 15. Monleon D, Garcia-Valles R, Morales JM et al
How host-microbial interactions shape the (2014) Metabolomic analysis of long-term spon-
nutrient environment of the mammalian intes- taneous exercise in mice suggests increased lipoly-
tine. Annu Rev Nutr 22:283–307 sis and altered glucose metabolism when animals
10. Lamichhane S, Yde CC, Schmedes MS et al are at rest. J Appl Physiol 117(10):1110–1119
(2015) Strategy for nuclear-magnetic-16. Gao X, Pujos-Guillot E, Martin J-F et al (2009)
resonance-based metabolomics of human feces. Metabolite analysis of human fecal water by gas
Anal Chem 87(12):5930–5937 chromatography/mass spectrometry with ethyl
11. Wu J, An Y, Yao J, Wang Y, Tang H (2010) An chloroformate derivatization. Anal Biochem
optimised sample preparation method for 393(2):163–175
NMR-based faecal metabonomic analysis. 17. Gao X, Pujos-Guillot E, Sébédio J-L (2010)
Analyst 135(5):1023–1030 Development of a quantitative metabolomic
12. Saric J, Wang Y, Li J et al (2008) Species varia- approach to study clinical human fecal water
tion in the fecal metabolome gives insight into metabolome based on trimethylsilylation
differential gastrointestinal function. derivatization and GC/MS analysis. Anal
J Proteome Res 7(1):352–360 Chem 82(15):6447–6456
13. Deda O, Gika H, Panagoulis T et al (2017) 18. Virgiliou C, Sampsonidis I, Gika HG, Raikos N,
Impact of exercise on fecal and cecal metabo- Theodoridis GA (2015) Development and vali-
lome over aging: a longitudinal study in rats. dation of a HILIC-MS/MS multitargeted
Bioanalysis 9(1):21–36 method for metabolomics applications.
Electrophoresis 36(18):2215–2225
Chapter 11
GC-MS Metabolomic Profiling of Protic Metabolites

Following Heptafluorobutyl Chloroformate Mediated
Dispersive Liquid Microextraction Sample Preparation
Protocol
Petr Hušek, Zdeněk Švagera, Dagmar Hanzlíková, Iva Karlínová,
Lucie Řimnáčová, Helena Zahradníčková, and Petr Šimek
Abstract
A simple analytical workflow is described for gas chromatographic-mass spectrometry (GC-MS)-based
metabolomic profiling of protic metabolites, particularly amino-carboxylic species in biological matrices.
The sample preparation is carried out directly in aqueous samples and uses simultaneous in situ heptafluo-
robutyl chloroformate (HFBCF) derivatization and dispersive liquid-liquid microextraction (DLLME),
followed by GC-MS analysis in single-ion monitoring (SIM) mode. The protocol involves ten simple
pipetting steps and provides quantitative analysis of 132 metabolites by using two internal standards. A
comment on each analytical step and explaining notes are provided with particular attention to the GC-MS
analysis of 112 physiological metabolites in human urine.
Key words Metabolomic profiling, GC-MS, Dispersive liquid-liquid microextraction, Chloroformate

derivatization, Urine, Quantitative analysis
1 Introduction
Comprehensive metabolomic analysis of small protic metabolites

possessing amino, carboxy, thio, or hydroxy groups in complex
biological matrices has been a demanding task because of the fre-
quent occurrence of structurally close and isomeric structures that
are difficult to separate and detect by at present prevailing liquid
chromatographic-mass spectrometry (LC-MS) techniques. As a
result, GC-MS combined with an efficient sample preparation
strategy involving metabolite derivatization (as a prerequisite) has
still been a popular, cost-effective tool in the analysis of amino and
organic acids and other protic metabolites that play important bio-
chemical roles in central metabolism. Current GC-MS-based
metabolomics relies mainly on two derivatization strategies: (1)
159
160 Petr Hušek et al.
CH3 O HFBCF O CH3 O

F F
pyridine F F
HO OH F O O O F
NH2 F O NH F
F F F F
F F
O
+ CO2
F F
Threonine F F Threonine-HFB
C4H9NO3 F F C18H12F21NO7
MM=119.0582 MM=753.0278
F
Fig. 1 Reaction scheme for the threonine protic functional groups with the HFBCF reagent. The carboxyl group
yields a HFB ester and the amino group a corresponding HFB carbamate, while the hydroxyl is transformed into
a HFB carbonate. MM = Monoisotopic mass
oximation with silylation and (2) reaction with alkyl chlorofor-

mates. The former approach requires strictly anhydrous condition
for silylation and has proved useful for profiling of polyhydroxylic
metabolites such as sugars [1, 2], steroids [3], sterols, and tocoph-
erols [4] but much less effective for metabolites bearing protic
nitrogen functional groups [5–7]. The latter approach has been
complementary and even more attractive; it can be applied in situ
in a complex biological matrix, and liquid-liquid microextraction
proceeds simultaneously under pyridine catalysis and, importantly,
with simultaneous carbon dioxide evolvement [8–11]. The arising
CO2 is partly dissolved in the whole sample medium; it enhances
the effective surface area between the immiscible organic and aque-
ous phase and renders thus a powerful dispersive phase enabling
the system to reach the final equilibrium in 5 s [4, 12].
Fluoroalkyl chloroformates (FCFs) possess some advanced fea-
tures over the traditionally used alkyl chloroformates (RCFs) pro-
viding highly volatile and much less polar derivatives extractable
into nonpolar hydrocarbon solvents. The reaction scheme for reac-
tion of heptafluorobutyl chloroformate (HFBCF) with threonine,
an amino acid possessing 3 protic functional groups, is shown in
Fig. 1.
The reaction product is formed directly in an aqueous environ-
ment with high yields in less than 5 s. In this way, extraordinary
clean protic metabolite extracts have been obtained from complex
biological matrices and have been successfully applied to metabo-
lite profiling of amino acids and steroids in human serum [4, 13]
or amino-carboxylic metabolites in human urine [12]. Moreover,
the clean extracts enabled chiral GC-MS analysis of 35 amino acid
enantiomeric pairs in human serum [14].
The workflow of the GC-MS-based metabolomic analysis is
depicted in Fig. 2.
GC-MS Metabolomic Profiling of Protic Metabolites 161
Fig. 2 The workflow for the GC-MS metabolomic analysis of protic metabolites in aqueous biological
matrices
Here, we describe a HFBCF-based GC-MS profiling method

for quantification of protic metabolites possessing amino, carboxy,
activated hydroxy, and thiol functional groups [12]. The elabo-
rated sample preparation protocol is simple and fast and follows a
procedure described in detail for urinary analysis in reference [12]
which should be consulted whenever necessary for getting more
comprehensive knowledge. It involves gradual pipetting of uni-
form small volumes of a sample and necessary liquid media in ten
steps:
(1) A sample
(2) An internal standard solution
(3) A reducing medium
(4) A pH adjustment
(5) An organic reaction medium containing the HFBCF reagent
(6 and 7) A repeated addition of a catalytic medium with pyridine
(8) An organic extraction medium
(9) An acidification medium
(10) An upper extraction phase transfer into a GC autosampler vial
(see Fig. 3) and, finally, the sample extract injection into a
GC-MS spectrometer.
Although the protocol was primarily developed for metabolo-
mic GC-MS analysis of protic metabolites in urine normalized to
creatinine [12], the procedure can directly be applied to any aque-
ous biological material with low protein content. If the content of
biopolymers is high (>2 mg/mL), then a prior precipitation step
must be adapted to the described workflow (see Note 1). The pro-
tocol may be modified for some applications, for instance, the
reducing step 3 can be omitted without any change, if the reduc-
tion of disulfide bonds is not required.
Fig. 3 A view on a 6 × 50 mm culture tube containing an aqueous sample (here urine) or an aqueous sample
extract and gradually added media during the sample preparation process: (a) an aqueous sample; (b) an aris-
ing two-phase system after performing steps 2–5, before the reaction initiation; (c) a turbid upper phase after
the first addition of the catalytic medium, step 6; occasionally visible CO2 bubbles can appear; (d) the organic
upper phase is clarified after step 7, which indicates that the reaction was completed; (e) increasing the
sample extract volume and its acidification in steps 8–9 enables an easy organic upper phase into an autos-
ampler vial in step 10 and final GC-MS analysis
2 Materials
2.1 Samples Samples containing no or little protein and cell residues (urine, cell
culture media) or cell and tissue extracts.
2.2 Chemicals, 1. Chemicals of analytical grade should be used. All solutions,

Solutions, except those containing the HFBCF reagent, should be pre-
and Reaction Media pared in distilled deionized water (DI water, <1.5 μS/cm,
25 °C) and stored at 4 °C (unless otherwise indicated).
2. Tris(3-hydroxypropyl)phosphine reducing agent (THP, 80%,
Merck): prepare a 5% stock solution by transferring 0.625 mL
of the THP liquid to a 10 mL volumetric flask, and adjust by
DI water up to 10 mL. Prepare a 0.5% working solution by
dilution 1:9 in a 10 mL volumetric flask (see Note 2).
3. 100 mM NaHCO3 (99.998% purity, Alfa Aesar) solution: dis-
solve 840 mg in 100 mL of DI water.
4. 1 M NaOH (99.99% purity, Alfa Aesar) solution: dissolve 4 g
in 100 mL of DI water.
5. Heptafluorobutanol (HFBOH) (99% purity, Pragolab, Prague,
Czech Republic).
6. Heptafluorobutyl chloroformate (HFBCF, 98%, Pragolab)
(see Note 3).
7. Solvents: isooctane (2,2,4-trimethylpentane, 99.5%), pyridine
(p.a., 99.0%), isopropanol (2-propanol, 99.5% purity) (all
Sigma-Aldrich).
8. The organic reaction medium: prepare isooctane, HFBCF,
and HFBOH in a volume ratio 15:4:1 (v/v/v) in a
eflon-
T capped, well-tightened 4 mL glass vial. Store in a
refrigerator, where the mixture remains stable for several
months.
9. The catalytic medium: mix 1 M NaOH with pyridine in a vol-
ume ratio of 24:1 (v/v).
10. The artificial urine solution: prepare the following chemicals
in DI water to final 10 g/L urea, 1 g/L creatinine, 7 g/L
NaCl, and 3 g/L K2SO4 [12].
11. The certified urine standard (product ORG-01) containing
diagnostic organic acids (ERNDIM Foundation (http://cms.
erndimqa.nl/). For the analyte concentrations, refer to the
website. Order the latest available batch.
2.3 Analytical 1. Internal standard solution (IS): 4-phenylbutyric acid (4PB,

Standards and Stock Sigma-Aldrich, MW = 164.2 g/mol); homophenylalanine (hF,
Solutions Sigma-Aldrich, MW = 179.2 g/mol). Prepare a stock solution
in 100 mM NaHCO3 with a final concentration of 200 μmol/L,
i.e., 5 nmol in 25 μL of the applied internal standard solution.
2. Protein amino acid (AA) standard solution (Sigma-Aldrich,
P/N AAS-18) in 0.1 M HCl containing alanine, glycine, valine,
leucine, isoleucine, threonine, serine, proline, aspartic acid,
methionine, glutamic acid, phenylalanine, lysine, histidine,
tyrosine, and cystine at a concentration of 2.5 and 1.25 mmol/L,
respectively. Alternatively, the protein AA mixture can be pre-
pared from stock solutions of particular AAs in 0.1 M HC. For
the complete metabolite list, refer to Table 1.
3. Non-protein amino acid, biogenic amine (Sigma-Aldrich)
standard solutions: prepare a stock solution of each metabolite
in 100 mM HCl with a final concentration of 100 μmol/L
(Table 1). Store at 4 °C.
4. Organic acid standards (Sigma-Aldrich); less common carboxylic
acids 2-hydroxysebacic, 3-hydroxyadipic, 3- hydroxypropionic,
3-hydroxyvaleric, and 5-hydroxyhexanoic acid and glycine con-
jugates hexanoylglycine, methylcrotonylglycine, and tiglylgly-
cine can be purchased from Dr. E. Brunet, Dept. Organic
Chemistry, University Autonoma de Madrid (Madrid, Spain),
http://www.uam.es/gruposinv/lumila/list.pdf; a racemate of
2-methyl citric acid (90%, C/D/N Isotopes, P/N X-4176).
5. The organic acid stock solutions data are summarized in Table 1
(see Note 4).
2.4 GC-MS 1. Agilent 7890A GC system equipped with G4513A autosampler

Instrumentation (Agilent), multimode injector (MMI), equipped with a 10 μL
syringe (CTC Analytics, P/N PAL3-SYH-207807).
Table 1
The list of protic metabolites determined by the described GC-MS protocol in human urine. The traditional name, the metabolite product after the HFBCF
derivatization, retention data, monoisotopic mass of the arising HFBCF derivatized product, diagnostic SIM ions, the used internal standard, the observed medium
metabolite concentration in urine [12]
Diagnostic Stock
GC ions Internal L3 solution Metabolite database coding
The profiled
No. metabolites as the RT MM m/ c Stock
No. [12]a Traditional name HFBCF derivatives (min) product zq1 m/zq2 Standardb (μM)c solution c (mM) HMDB PubChem KEGG
1 108 1-Methylhistidine 1-Methylhistidine 27.2 577.1 95 350 hF 40 0.1 M HCl 10 HMDB00001 92105 C01152
2 101 2,4-Diamino 2,4-Diamino 24.74 752.0 282 256 hF 40 DI water 10 HMDB02362 470
butyric acid butyrate
3 36 2-Aminobutyric 2-Amino 13.61 511.0 284 84 hF 20 DI water 10 HMDB00452 80283 C02356
acid butyrate
4 99 2-Amino 2-Aminopimelate 24.6 765.1 338 138 hF 40 DI water 10 HMDB34252 101122
heptanedioic-
acid
5 27 2-Aminoiso- 2-Aminoiso 12.14 525.1 284 241 hF 40 DI water 50 HMDB01906 6119 C03665
butyric acid butyrate
6 22 2-Hydroxy-3- 2-Hydroxy-3- 11.16 526.0 55 299 PB 20 DI water 10 HMDB00407 99823
methylbutyric methylbutyrate
acid (isovalerate)
7 34 2-Hydroxy-3- 2-Hydroxy-3- 13.06 540.1 284 484 PB 20 DI water 10 HMDB00317 10796774
methylpentanoic methylvalerate
acid
Diagnostic Stock
The profiled
8 19 2-Hydroxybutyric 2-Hydroxybutyrate 10.49 512.0 285 241 PB 20 DI water 10 HMDB00008 11266 C05984
acid
9 112 2-Hydroxy 2-Hydroxysebacate 27.34 808.1 95 381 PB 20 Acetonitrile 10 HMDB00424 128458
decanedioic acid
10 65 2-Hydroxyglutaric 2-Hydroxyglutarate 18.34 738.0 283 239 PB 100 DI water 100 HMDB02307 439340 C00894
acid
11 11 2-Hydroxyisobutyric 2-Hydroxyisobutyrate 8.74 512.0 241 285 PB 40 DI water 10 HMDB00729 11671
acid
12 82 2-Hydroxyphenylacetic 2-Hydroxyphenylacetate 20.52 560.0 91 333 PB 20 DI water 50 HMDB00669 11970 C05852
acid
13 26 2-Hydroxyvaleric 2-Hydroxyvalerate 12.12 526.0 55 299 PB 20 DI water 10 HMDB01863 98009
acid
14 3 2-Ketoisovaleric 2-Ketoisovalerate 5.81 298.0 71 113 PB 20 DI water 10 HMDB00019 49 C00141
acid
15 68a 2-Methylcitric 2-Methylcitrate-4 18.46 552.0 152 334 PB 40 DI water 10 HMDB00379 515
acid (lactone)
16 46 3-Aminoisobutanoic 3-Aminoisobutyrate 15.02 511.0 256 112 hF 100 0.1 M HCl 100 HMDB03911 64956 C05145
acid
17 67 3-Hydroxyadipic 3-Hydroxyadipate 18.41 752.0 85 127 PB 100 0.1 M HCl 50 HMDB00345 151913
acid
18 24a 3-Hydroxybutyric 3-Hydroxybutyrate-2 11.93 512.0 268 69 PB 100 DI water 10 HMDB00357 441 C01089
acid
19 4a 3-Hydroxyisovaleric 3-Hydroxyisovalerate- 6.46 300.1 59 85 PB 100 DI water 10 HMDB00754 69362
acid 1(OH)
(continued)
Table 1
(continued)
Diagnostic Stock
The profiled
20 41 3-Hydroxy 3-Hydroxy-3-methyl 14.37 752.0 85 285 PB 100 DI water 10 HMDB00355 1662 C03761
methylglutaric glutarate
acid
21 91 3-Hydroxy 3-Hydroxyphenyl 22.61 560.0 333 277 PB 20 0.1 M HCl 100 HMDB00440 12122 C05593
phenylacetic acid acetate
22 89a 3-Hydroxyproline 3-Hydroxyproline-2 21.9 765.0 521 538 hF 20 0.1 M HCl 10 HMDB02113 11137200 C04397
23 79a 3-Hydroxysebacic 3-Hydroxysebacate- 20.01 582.1 71 271 PB 20 DI water 10 HMDB00350 3017884
acid 1(OH)
24 9a 3-Hydroxyvaleric 3-Hydroxyvalerate- 8.21 300.1 71 271 PB 40 DI water 10 HMDB00531 107802
acid 1(OH)
25 131 3-Methoxytyramine 3-Methoxytyramine 32.36 845.1 319 376 hF 20 0.1 M HCl 10 HMDB00022 1669 C05587
26 5 3-Methyl-2- 2-Keto-3- 7.34 312.1 57 85 PB 20 DI water 10 HMDB00491 47 C03465
Oxovaleric acid methylvalerate
27 47 3-Methyladipic acid 3-Methyladipate 15.53 524.1 325 55 PB 20 DI water 50 HMDB00555 6999745
28 76 3-Methylcrotonyl Methylcrotonyl 19.54 339.1 83 82 hF 40 DI water 10 HMDB00459 169485
glycine glycine
29 116 3-Methylhistidine 3-Methylhistidine 28.38 577.1 95 150 hF 40 0.1 M HCl 10 HMDB00479 64969 C01152
30 111 4-Aminobenzoic 4-Aminobenzoate 27.29 545.0 146 345 hF 20 Ethanol 10 HMDB01392 978 C00568
acid
Diagnostic Stock
The profiled
31 87 4-Hydroxybenzoic 4-Hydroxybenzoate 21.41 546.0 303 347 PB 20 Ethanol 100 HMDB00500 135 C00156
acid
32 38 4-Hydroxybutyric 4-Hydroxybutyrate 14.05 512.0 227 269 PB 100 DI water 100 HMDB00710 10413 C00989
acid
33 115 4-Hydroxycinnamic 4-Hydroxycinnamate 27.98 572.0 329 572 PB 20 DI water 10 HMDB02035 637542 C00811
acid
34 110 4-Hydroxymandelic 4-Hydroxymandelate 27.25 802.0 575 347 PB 20 DI water 10 HMDB00822 328 C11527
acid
35 96 4-Hydroxy 4-Hydroxy 23.31 560.0 289 333 PB 40 0.1 M HCl 100 HMDB00020 127 C00642
phenylacetic acid phenylacetate
36 94a 4-Hydroxyproline 4-Hydroxyproline-2 23.12 765.0 294 521 hF 20 0.1 M HCl 50 HMDB06055 69248 C01015
37 63 4-Phenylbutyric 4-Phenylbutyrate 17.95 346.1 104 147 0.1 M 100
acid (4 PB, I.S.) NaHCO3
38 97 5-Aminolevulinic 5-Aminolevulinate 23.36 539.0 283 256 hF 40 0.1 M HCl 10 HMDB01149 137 C00430
acid
39 81 5-Aminopentanoic 5-Aminovalerate 20.41 525.1 256 269 hF 40 0.1 M HCl 10 HMDB03355 138 C00431
acid
40 53 5-Hydroxyhexanoic 5-Hydroxyhexanoate 16.67 540.1 227 113 PB 100 DI water 10 HMDB00525 170748
acid
41 144 5-Hydroxyindo 5-Hydroxyin 35.16 599.0 372 599 hF 20 DI water 10 HMDB00763 1826 C05635
leacetic acid doleacetate
42 138b 5-Hydroxylysine 5-Hydroxylysine 34.12 1022.1 269 256 hF 40 0.1 M HCl 50 HMDB00450 3032849 C16741
(isomers)
(continued)
Table 1
(continued)
Diagnostic Stock
The profiled
43 37a Acetylglycine N-Acetylglycine- 13.88 525.0 256 483 hF 200 DI water 10 HMDB00532 10972
2(NH)
44 54 Aconitic Aconitate 16.78 720.0 321 492 PB 200 DI water 100 HMDB00958 444212 C02341
acid(trans)
45 45 Adipic acid Adipate 14.95 510.1 282 311 PB 40 Ethanol 100 HMDB00448 196 C06104
46 29 Alanine Alanine 12.35 497.0 270 70 hF 40 0.1 M HCl 100 HMDB00161 5950 C00041
47 93 Aminoadipic 2-Aminoadipate 22.88 751.0 124 282 hF 40 0.1 M HCl 100 HMDB00510 469 C00956
acid
48 70 Asparagine Asparagine 19.1 522.0 295 95 hF 100 0.1 M HCl 100 HMDB00168 6267 C00152
49 64 Aspartic acid Aspartate 18.3 723.0 254 496 hF 40 0.1 M HCl 100 HMDB00191 5960 C00049
50 86 Azelaic Azelaate 21.23 552.1 353 152 PB 20 Ethanol 10 HMDB00784 2266 C08261
acid
51 20 Benzoic Benzoate 10.74 304.0 105 304 PB 40 Ethanol 100 HMDB01870 243 C00180
acid
52 44 Beta-alanine 3-alanine 14.91 497.0 270 113 hF 40 0.1 M HCl 10 HMDB00056 239 C00099
53 16 Citraconic Citraconate 10.29 494.0 295 267 PB 40 DI water 10 HMDB00634 643798 C02226
acid
54 21a Citramalic Citramalate-1 10.8 312.0 85 285 PB 100 DI water 10 HMDB00426 1081 C00815
acid (lactone)
55 60a Citric acid Citrate-2 (OH) 17.79 738.0 311 269 PB 400 DI water 100 HMDB00094 311 C00158
Diagnostic Stock
The profiled
56 132 Cystathionine Cystathionine 32.42 1038.0 328 282 hF 40 0.1 M HCl 100 HMDB00099 439258 C02291
57 90 Cysteine Cysteine (total)** 22.22 755.0 328 285 hF 100 0.1 M HCl 100 HMDB00574 5862 C00097
58 124a Diaminopimelic 2,6-Diaminopimelate 29.45 1006.1 308 536 hF 40 DI water 10 HMDB01370 439283 C00666
acid
59 135 DOPA 3,4-Dihydroxy 32.63 1057.0 149 388 hF 40 0.1 M HCl 10 HMDB00609 836 C00355
phenyl alanine
60 12 Ethylmalonic Ethylmalonate 8.78 496.0 297 468 PB 40 DI water 10 HMDB00622 11756
acid
61 130 Ferulic acid 4-Hydroxy-3- 30.64 602.0 602 375 PB 20 Ethanol 100 HMDB00954 445858 C01494
(trans) methoxycinnamate
62 15 Fumaric acid Fumarate 9.53 480.0 281 253 PB 40 DI water 10 HMDB00134 444972 C00122
63 83 Glutamic acid Glutamate 20.82 737.0 310 282 hF 40 0.1 M HCl 10 HMDB03339 23327 C00217
64 107 Glutamine Glutamine 26.64 554.1 84 282 hF 100 DI water 100 HMDB00641 5961 C00064
65 30 Glutaric acid Glutarate 12.51 496.0 227 297 PB 40 DI water 10 HMDB00661 743 C00489
66 58 Glyceric acid Glycerate (2,3- 17.68 740.0 113 497 PB 40 DI water 10 HMDB00139 439194 C00258
Dihydroxy
propionate)
67 35 Glycine Glycine 13.56 483.0 256 212 hF 200 DI water 100 HMDB00123 750 C00037
68 14 Glycolic acid Glycolate 9.35 484.0 285 213 PB 40 DI water 100 HMDB00115 757 C00160
69 125 Glycylproline Glycylproline 29.83 580.1 70 153 hF 40 0.1 M HCl 10 HMDB00721 79101
70 39a Hexanoylglycine Hexanoylglycine-1 14.07 155.1 99 71 hF 20 DI water 10 HMDB00701 99463
(cyclic)
(continued)
Table 1
(continued)
Diagnostic Stock
The profiled
71 78a Hippuric acid Hippurate-1 (cyclic, 19.86 161.0 105 161 hF 400 Ethanol 100 HMDB00714 464 C01586
60%)
72 106 Histamine Histamine 26.48 563.1 308 320 hF 40 0.1 M HCl 10 HMDB00870 774 C00388
73 114a Histidine Histidine-2 (NR) 27.9 789.0 307 362 hF 200 0.1 M HCl 100 HMDB00177 6274 C00135
74 100 Homocysteine Homocysteine 24.68 769.0 282 342 hF 40 MeCN 10 HMDB00742 778 C05330
(total)**
75 103 Homophenylalanine Homophenylalanine 25.98 587.1 91 283 0.1 M 100
(hF, I.S.) NaHCO3
76 104 Homovanillic acid Homovanillate 26.07 590.0 107 590 PB 20 0.1 M HCl 50 HMDB00118 1738 C05582
77 32 Hydroxyisocaproicacid 2-Hydroxyisocaproate 12.79 540.1 296 113 PB 20 DI water 10 HMDB00746 83697
78 118 Hydroxyphenyllactic 4-hydroxy- 28.54 816.0 572 345 PB 20 DI water 10 HMDB00755 9378 C03672
acid Phenyllactate
79 52 Hydroxypropionic acid 3-Hydroxypropionate 16.06 526.0 255 298 PB 100 DI water 10 HMDB00700 68152 C01013
(dimer)
80 113 Indolacetate Indolacetate 27.61 357.1 130 357 hF 20 MeCN 10
81 80 Isocitric acid Isocitrate 20.18 964.0 465 321 PB 100 DI water 100 HMDB00193 1198 C00311
82 50 Isoleucine Isoleucine 15.93 539.1 283 312 hF 40 0.1 M HCl 10 HMDB00172 6306 C00407
Diagnostic Stock
The profiled
83 59 Isovalerylglycine Isovalerylglycine 17.77 341.1 85 525 hF 40 DI water 10 HMDB00678 546304
84 6 Ketoleucine 2-Ketoisocaproate 7.43 312.1 85 57 PB 20 DI water 10 HMDB00695 70 C00233
85 126 Kynurenic acid Kynurenate 29.99 597.0 371 354 hF 40 NaHCO3 10 HMDB00715 3845 C01717
86 143 Kynurenine Kynurenine 35.07 842.1 146 372 hF 40 0.1 M HCl 10 HMDB00684 161166 C00328
87 13 Lactic acid Lactate 9.14 498.0 271 255 PB 100 DI water 100 HMDB00190 107689 C00186
88 49 Leucine Leucine 15.84 539.1 312 270 hF 40 0.1 M HCl 50 HMDB00687 6106 C00123
89 119a Lysine Lysine-2 (N,N-R) 28.79 780.1 310 256 hF 100 0.1 M HCl 100 HMDB00182 5962 C00047
90 51 Malic acid Malate 16.04 724.0 281 253 PB 40 DI water 10 HMDB31518 92824 C00497
91 7 Malonic acid Malonate 7.51 468.0 269 407 PB 40 Ethanol 10 HMDB00691 867 C00383
92 74 Mandelic acid Mandelate 19.45 560.0 289 333 PB 20 DI water 10 HMDB00703 439616 C01984
93 88 Methionine Methionine 21.65 557.0 61 357 hF 40 0.1 M HCl 10 HMDB00696 6137 C00073
94 117 Methioninesulfone Methionine sulfone 28.51 589.0 282 82 hF 40 0.1 M HCl 10
95 71 Methylcysteine S-Methylcysteine 19.13 543.0 61 300 hF 20 0.1 M HCl 10 HMDB02108 24417
96 33 Methylglutaric acid 3-Methylglutarate 12.93 510.1 311 282 PB 40 DI water 10 HMDB00752 12284
97 8 Methylmalonic acid Methylmalonate 7.54 482.0 283 438 PB 40 DI water 10 HMDB00202 487 C02170
98 18 Methylsuccinic acid Methylsuccinate 10.38 496.0 297 268 PB 40 DI water 50 HMDB01844 10349 C08645
99 66 N-Acetyl-asparticacid N-Acetylaspartate 18.38 539.0 270 312 hF 200 0.1 M HCl 50 HMDB00812 65065 C01042
100 28 Nicotinic acid Nicotinate 12.17 305.0 106 78 hF 40 0.1 M HCl 10 HMDB01488 938 C00253
(continued)
Table 1
(continued)
Diagnostic Stock
The profiled
101 109 Ornithine Ornithine 27.25 766.1 296 256 hF 40 0.1 M HCl 50 HMDB00214 6262 C00077
102 2 Oxalic acid Oxalate 5.38 454.0 113 183 PB 200 DI water 10 HMDB02329 971 C00209
103 42b Oxoglutaric acid 2-Ketoglutarate- 14.6 510.0 283 284 PB 200 DI water 100 HMDB00208 51 C00026
2(80%)
104 121 Palmitic acid Palmitate 28.91 438.2 255 438 PB 20 Ethanol 50 HMDB00220 985 C00249
105 31 Phenylacetic acid Phenylacetate 12.65 318.0 91 318 PB 20 Ethanol 10 HMDB00209 999 C07086
106 98 Phenylalanine Phenylalanine 23.39 573.1 91 330 hF 40 0.1 M HCl 100 HMDB00159 6140 C00079
107 84 Phenyllactic acid 3-Phenyllactate 20.92 574.0 330 131 PB 20 DI water 10 HMDB00779 3848 C01479
108 92 Phenylpyruvic acid Phenylpyruvate 22.84 572.0 118 329 PB 40 Ethanol 10 HMDB00205 997 C00166
109 61 Phthalic acid Phthalate 17.79 530.0 331 332 PB 20 DI water 10 HMDB02107 1017 C01606
110 56 Pimelic acid Pimelate 17.09 524.1 296 325 PB 20 Ethanol 10 HMDB00857 385 C02656
111 55 Proline Proline 16.84 523.0 296 297 hF 40 0.1 M HCl 10 HMDB00162 145742 C00148
112 142 Prolylhydroxyproline Prolylhydroxy 34.6 862.1 296 297 hF 100 0.1 M HCl 10 HMDB06695 11902892
proline
113 48 Propionylglycine Propionylglycine 15.55 313.1 57 56 hF 100 DI water 10 HMDB00783 98681
114 10 Propyl pentanoate 2-Propylvalerate 8.61 326.1 255 284 PB 40 DI water 10 HMDB40296 67328
115 69 Pyroglutamic acid Pyroglutamate 18.9 537.0 310 84 hF 200 DI water 100 HMDB00267 7405 C01879
116 73 Salicylic acid Salicylate 19.41 546.0 120 303 PB 20 DI water 10 HMDB01895 338 C00805
Diagnostic Stock
The profiled
117 120 Salicyluric acid 2-Hydroxy 28.82 603.0 120 403 hF 40 0.1 M HCl 10 HMDB00840 10253 C07588
hippurate
118 25 Sarcosine Sarcosine 12.07 497.0 270 226 hF 20 0.1 M HCl 10 HMDB00271 1088 C00213
119 95 Sebacic acid Sebacate 23.16 566.1 98 367 PB 40 Ethanol 10 HMDB00792 5192 C08277
120 77 Serine Serine 19.6 739.0 268 295 hF 100 DI water 100 HMDB00187 5951 C00065
121 134 Stearic acid Stearate 32.5 466.3 255 466 PB 20 Ethanol 50 HMDB00827 5281 C01530
122 72 Suberic acid Suberate 19.2 538.1 339 138 PB 20 MeCN 10 HMDB00893 10457 C08278
123 17 Succinic acid Succinate 10.32 482.0 283 55 PB 40 DI water 100 HMDB00254 1110 C00042
124 75 Thioproline Thioproline 19.46 541.0 314 287 hF 20 0.1 M HCl 10
125 57a Threonine Threonine-1(OH) 17.42 527.0 100 283 hF 100 0.1 M HCl 100 HMDB00167 6288 C00188
126 23 Tiglylglycine Tiglylglycine 11.27 339.1 83 55 hF 40 DI water 50 HMDB00959 6441567
127 139 Tryptamine Tryptamine 34.18 612.1 130 386 hF 20 0.1 M HCl 10 HMDB00303 1150 C00398
128 141 Tryptophan Tryptophan 34.44 612.1 130 131 hF 40 NaOH 100 HMDB00929 6305 C00078
129 127 Tyramine Tyramine 30.06 589.1 346 333 hF 20 Ethanol 10 HMDB00306 5610 C00483
130 129 Tyrosine Tyrosine 30.58 815.0 333 289 hF 40 0.1 M HCl 100 HMDB00158 6057 C00082
131 102 Urocanic acid Trans-urocanate 25.39 546.0 347 546 hF 20 DI water 10 HMDB34174 1549103
132 40 Valine Valine 14.1 525.1 298 283 hF 40 0.1 M HCl 100 HMDB00883 6287 C00183
133 128 Vanillactic acid 4-Hydroxy-3- 30.38 846.0 375 561 PB 20 DI water 10 HMDB00913 160637
methoxy
phenyllactate
134 122 Vanillylmandelic Vanillylmandelate 28.92 832.0 832 377 PB 20 0.1 M HCl 10 HMDB00291 736172 C05584
acid
MM = a molecular mass of each observed metabolite derivative; m/z = diagnostic ions in the EI spectrum of each metabolite; q1, q2 = diagnostic (quantitation and qualifier) fragment ions used for
the quantitative GC-SIM-EI-MS analysis
a
Numbering, traditional name, and metabolite product names according to the reference 12
b
Calibration against the internal standard; hF homophenylalanine, PB 4-phenylbutyric acid
c
Calibration level L3 (the observed average concentration in urine) [12]
2. Sky® 4 mm I.D. cyclo double taper inlet liner (Restek, P/N

23310).
3. ZB-XLB type, 30 m × 0.25 mm ID, 0.25 μm film thickness
(Phenomenex, P/N 7HG-G019-11).
4. Single quadrupole mass triple-axis detector (5975 MSD Inert
XL, Agilent) equipped with an inert EI ion source.
5. Autosampler 2 mL vials (12 × 32 mm) with 9 mm PP open hole
caps (Labicom, P/N 5310F-09) and 0.040″ PTFE/silicone/
PTFE Septa (Labicom, P/N 604060-09).
6. Inert conical glass insert, 200 μL volume (Chromacol, P/N
02-MTV).
2.5 Additional 1. Sample preparation glass culture tubes 6 × 50 mm; material:

Equipment sodium-potassium silicate
2. Glass (Merci, P/N Z1632000605010) or borosilicate glass
(Kimble-Kontes, P/N 73500-650).
3. Common screw cap Teflon-lined 2 and 4 mL amber vials for the
reagent solutions.
4. An adjustable 50 and 100 μL Transferpettor pipette with a glass
capillary (Brand, P/N 701868 and 701873) for manipulation
with the reagents and their mixtures in isooctane. The pipette
tips with 25 mm capillary (gel-loading type, VWR Int.) for aspi-
rating the upper organic phase in the 6 × 50 mm vial.
5. Alternatively, a common pipette (10–100 μL) such as a Biohit
Proline® mechanical pipette (Sartorius, P/N 720050) equipped
with an Optifit tip 200 (P/N 4059.9002) can be used.
6. A common vortex for sample mixing and a minicentrifuge such
as mySPIN 6 (Thermo Scientific, 2000 × g) for a complemen-
tary separation of immiscible layers in sample vials.
7. A commercial assay kit for creatinine analysis. For instance, a
creatinine kit (Dialab, P/N D95595).
8. A common spectrophotometer capable of measuring creatinine
at 490–510 nm in urine, such as Specord® Plus (Jena Analytik),
by an appropriate kit (see Subheading 2.5, item 7) used for
clinical applications.
3 Methods
Using appropriate laboratory wear, glasses, and other personal

protective equipment is recommended; follow standard laboratory
3.1 Sampling precautions and local guidelines, especially waste disposal regula-
and Storage tions. Use a functional fume hood for the sample workup.
and Normalization
of Samples
1. Serious attention should be paid to sample collection, trans-

port, and storage, because any omission may result in false
results. For urine, collection of the morning second-void sam-
ples is the most common practice and the easiest sampling
method.
2. For urine analysis, store freshly collected samples at 4 °C within
2 h. For longer than 48 h storage, freeze the samples and keep
at −20 °C (see Note 5).
3. If urine is a subject of study, measure creatinine concentration
in each collected urine aliquot used in the analysis data for nor-
malization of each determined metabolite (see Note 6).
3.2 Preparation 1. Prepare a calibration mixture following the procedure described

of Calibration in ref. 12 by adding an appropriate volume of each stock solu-
Solutions tion standard (Table 1) into a volumetric flask. Adjust to a final
volume with the artificial urine solution.
2. The concentration of each individual analyte denotes its average
level (here, level 3 = L3) measured in a pooled urine sample of
normal morning urine, Table 1 [12]. Prepare the lower (L1 and
L2) and higher (L4, L5) calibration points, i.e., 10 times, 2.5
times diluted, and 2.5 times and 10 times increased to the
respective medium L3 level. For the metabolites with highest
abundance, prepare level L6 (25 times higher than L3).
3. Distribute the calibration solution into appropriate aliquots
before freezing.
4. Prepare an appropriate pooled sample by mixing an equal small
volume of all samples included in the study for the verification
of the average EI MS response for each target metabolite and
for the quality control (QC) analysis (see also Chapter 2).
3.3 Sample 1. Transfer 25 μL of aqueous sample into a 6 × 50 mm culture

Preparation Protocol tube.
2. Spike the sample with 25 μL of the internal standard
solution.
3. Add 25 μL 0.5% THP reducing solution, and mix the contents
gently for 1–2 s and leave to stand for 1 min.
4. Adjust pH to ca 9 with 25 μL 100 mM NaHCO3 solution and
vortex gently.
5. Add 50 μL of the organic reaction medium (isooctane,
HFBCF, and HFBOH, 15:4:1, v/v/v) (see Note 7).
6. Add 25 μL of the catalytic medium (1 M NaOH-pyridine,
24:1, v/v), and vortex the content for ca 3 s leaving the
organic phase milky.
7. Add a second portion (25 μL) of the catalytic medium, and
shake the content for 5 s until the milky phase becomes
clarified.
Table 2
GC-EI-SIM-MS operating conditions
GC Injector Mode Pulsed splitless (elevated head pressure from

110 to 220 kPa)
Liner Sky® 4 mm I.D. cyclo double taper inlet liner
(Restek, P/N 23310)
Temperature 220 °C
Injection volume 1 μL
Temperature 220 °C
GC oven Initial temperature 75 °C
Ramp 6 °C/min to 150 °C, 8 °C/min to 190 °C,
12 °C/min to 250 °C and at 20 °C/min to
300 °C hold for 3 min
Run time 25 min
GC column Capillary column DB-XLB type, 30 m × 0.25 mm ID, 0.25 μm
film thickness (Agilent, P/N 122-1232)
Carrier gas Helium
Flow rate 1.2 mL/min
Mode Constant flow
Outlet pressure Vacuum
GC-MS transfer line Temperature 250 °C
MS quadrupole Ion source temperature 230 °C
Full scan mode m/z 50–800 Da
SIM mode Metabolite SIM m/z ions; see Table 1
Software MSD ChemStation (version E.02, Agilent)
8. Add 50 μL of the isooctane extraction medium, and mix for

about 1–2 s.
9. Add of 25 μL of 1 M aqueous HCl and vortex briefly; if the
phases are not well separated, a minicentrifuge may be a con-
venient option (see Note 8).
10. Aspirate 70–80 μL of the upper organic phase into a vial insert
(150–200 μL volume).
11. Inject a sample extract aliquot (1 μL) by using a pulsed split-
less injection into a GC injector, and start the GC-MS acquisi-
tion (see Note 9).
3.4 GC-MS Analysis 1. The instrument GC-MS conditions are summarized in Table 2.
2. First, analyze the standard mixtures to check the separa-
tion performance, retention times, the analyte peak shape, and
acquisition of the employed fragment ions in the obtained EI

spectra (Table 1).
3. Using a single quadrupole MS analyzer, single-ion monitoring
(SIM) mode is commonly used for quantification. Use the char-
acteristic m/z ions (a quantifier and a qualifier) listed in Table 1.
The SIM scanning should be arranged into convenient time
sequence groups (windows) associating closely co-eluting ana-
lytes m/z ions thus enabling an appropriate SIM dwell time for
each detected analyte [12].
4. Prepare an appropriate analysis sequence for a sample series
consisting of repeated blank, standard, the QCs (see Subheading
3.2, step 2), calibration, and real sample extracts. Measure the
blanks, QC samples, and standards regularly (at least every ten
sample runs). Analyze samples in a random order to avoid sys-
tematic errors.
5. Change the GC injector liner after approximately 150 samples
depending on the sample matrix. Before use, condition each
new liner by running the following sequence: solvent blank,
standard mixtures, the pooled QC sample extract (twice), and
solvent blank (twice).
3.5 Data Analysis 1. Peak area for quantifier and qualifier ion of each metabolite is
integrated. Their ratio is calculated to test for potential
interferences.
2. The peak area of each quantifier is normalized by the peak area
of the corresponding internal standard: amino acids and bio-
genic amines against homophenylalanine, compound No. 75;
organic acids against 4-phenylbutyric acid, compound No. 37,
Table 1 [12].
3. Use appropriate vendor data processing software for data cali-
bration and metabolite quantification.
4. Check metabolite responses in the QC samples measured regu-
larly throughout the whole sample set. If the analyte’s relative
response to the IS fluctuates with RSD >30%, then even a semi-
quantitative measurement of such metabolite is difficult, and it
should be excluded from the metabolomic study (see Note 10).
5. Once the metabolite levels have been determined and met, pre-
defined acceptance criteria normalize appropriately the mea-
sured metabolite concentrations relative to creatinine. For urine
recalculate the metabolite levels to creatinine or other suitable
reference factors. Export the analytical data matrix into a
Microsoft Excel® spreadsheet or other formats suitable for fur-
ther chemometric analysis.
6. Use an appropriate statistical software to recognize differences
among the studied metabolite sample sets. The calculation of
p-values by means of a t-test helps to determine significance of

the obtained results. The data set can be conveniently examined
graphically by means of box plots that display patterns of quan-
titative data and thus facilitate interpretation of observed
metabolite changes in the studied organism.
4 Notes
1. For serum/plasma, (lipo)proteins must be precipitated with a

suitable medium prior to the application of this protocol. The
workup requires first an internal solution addition (step 2),
which may be followed by the THP reduction of disulfide
bonds (step 3). However, the protein precipitation must pre-
cede step 4 pH adjustment; it can be carried out with perchlo-
ric acid [13], trichloroacetic acid, or an organic solvent [15,
16]. Note that the selected precipitation conditions for plasma
or serum may affect the obtained metabolite profile [15] and
thus the described protocol must always be adapted and vali-
dated to a particular demand.
2. The 5% stock THP solution can be kept in a freezer for a year
and the 0.5% working solution in a refrigerator for a month.
3. The 2,2,3,3,4,4,4-heptafluorobutyl chloroformate (HFBCF)
reagent is a liquid with a boiling point 105–107 °C and den-
sity 1.6 g/cm3 [14]. The reagent must be stored in tightly
closed Teflon-lined cap glass vials at 4 °C and thus is stable for
at least 24 months.
WARNING! Manipulation with HFBCF must be per-
formed in a well-ventilated area (fume hood).
4. A list of metabolites covered by the protocol is summarized in
Table 1. For the analytical purposes, the compounds can be
sorted into six groups, i.e., protein amino acids, non-protein
amino acids, dicarboxylic acids, hydroxycarboxylic acids,
organic (aromatic) acids, and metabolites sensitive to storage
conditions. The last group comprised of oxoacids, lactate,
5-hydroxyindol- and indolacetate, 3- and
4-hydroxyphenylacetate, 4-hydroxymandelate,
4-hydroxyphenyllactate, kynurenate, vanillylmandelate, kyn-
urenine, glycylproline, prolylhydroxyproline, and
3-methylcrotonylglycine; the stock and working solutions of
this group were kept in freezer at −20 °C.
5. Urine like other important biological matrices is a metabolite-
rich mixture. Be careful and always take into account proper-
ties of each metabolite of interest in the studied matrix. Check
carefully the metabolite stability by sample measurement
within a convenient time period before making final conclu-

sions from the measured data.
6. In metabolomics, normalization of samples is an important
practice [17]. For urinary analysis, creatinine is the commonly
used reference that indicates the urine concentration [18].
Urine of healthy women and men typically contains around
5–16 mmol/L of creatinine [12, 19]. If its concentration is
highly increased (more than 3–4 times), the sample should be
diluted with DI water before analysis maintaining thus the
urine composition closer the average creatinine abundance.
7. The used HFBC reagent volume (10 μL, 60 μmol) is efficient
for workup of urine volumes below 50 μL. In contrast with
classical alkyl chloroformates, the corresponding heptafluoro-
butyryl alcohol is not necessary in the reaction medium.
Nevertheless, a small 5% aliquot facilitates esterification of
polycarboxylic acids such as citrate.
8. The acidification step substantially further decreases the pyri-
dine catalyst content in the arising upper organic layer and
thus contamination of the GC-MS system. Consequently, the
liner change typically follows after 120–150 samples, less fre-
quently than in earlier methods [20].
9. If the prepared organic sample extracts are not measured
immediately, they can be stored in tightly closed Teflon-lined
autosampler vials for up to 2 weeks at −20 °C. A slow degra-
dation of some metabolites was observed, in particular kynure-
nate, tiglylglycine, 3-hydroxybutyrate, fumarate and
prolylhydroxyproline, histidine, 1-methyl- and
3-methylhistidine, and isocitrate.
10. Note that demands on metabolomic analysis do not always
conform to strict guidelines requested, for instance, by guide-
lines in drug analysis, and data showing a higher uncertainty
may be useful in the study. Moreover, metabolite concentra-
tions observed between two studied models rarely change by
more than one order of magnitude, and thus narrower calibra-
tion ranges can be used throughout a metabolomic study with
respect to the amount estimated in the pooled QC sample.
Acknowledgments
This work was supported by the Czech Science Foundation, proj-

ect No. 17-22276S.
References 10. Smart KF, Aggio RBM, Van Houtte JR et al

(2010) Analytical platform for metabolome
1. Andrews MA (1989) Capillary gas- analysis of microbial cells using methyl chloro-
chromatographic analysis of monosaccharides – formate derivatization followed by gas chroma-
improvements and comparisons using tography-mass spectrometry. Nat Protoc
trifluoroacetylation and trimethylsilylation of 5(10):1709–1729. https://doi.org/10.1038/
sugar o-benzyl-oximes and o-methyl-oximes. nprot.2010.108
Carbohydr Res 194:1–19. https://doi. 11. Wachsmuth CJ, Hahn TA, Oefner PJ et al
org/10.1016/0008-6215(89)85001-3 (2015) Enhanced metabolite profiling using a
2. Kostal V, Zahradnickova H, Simek P et al redesigned atmospheric pressure chemical ion-
(2007) Multiple component system of sugars ization source for gas chromatography coupled
and polyols in the overwintering spruce bark to high-resolution time-of-flight mass spec-
beetle, Ips typographus. J Insect Physiol trometry. Anal Bioanal Chem 407(22):6669–
53(6):580–586. https://doi.org/10.1016/j. 6680. https://doi.org/10.1007/
jinphys.2007.02.009 s00216-015-8824-x
3. Hill M, Parizek A, Kancheva R et al (2010) 12. Husek P, Svagera Z, Hanzlikova D et al (2016)
Steroid metabolome in plasma from the umbil- Profiling of urinary amino-carboxylic metabo-
ical artery, umbilical vein, maternal cubital vein lites by in-situ heptafluorobutyl chloroformate
and in amniotic fluid in normal and preterm mediated sample preparation and gas
labor. J Steroid Biochem Mol Biol 121(3– chromatography-mass spectrometry.
5):594–610. https://doi.org/10.1016/j. J Chromatogr A 1443:211–232. https://doi.
jsbmb.2009.10.012 org/10.1016/j.chroma.2016.03.019
4. Rimnacova L, Husek P, Simek P (2014) A new 13. Simek P, Husek P, Zahradnickova H (2008)
method for immediate derivatization of Gas chromatographic-mass spectrometric anal-
hydroxyl groups by fluoroalkyl chloroformates ysis of biomarkers related to folate and cobala-
and its application for the determination of ste- min status in human serum after
rols and tocopherols in human serum and dimercaptopropanesulfonate reduction and
amniotic fluid by gas chromatography- mass heptafluorobutyl chloroformate derivatization.
spectrometry. J Chromatogr A 1339:154–167. Anal Chem 80(15):5776–5782. https://doi.
https://doi.org/10.1016/j. org/10.1021/ac8003506
chroma.2014.03.007 14. Simek P, Husek P, Zahradnickova H (2012)
5. Simek P, Heydova A, Jegorov A (1994) High- Heptafluorobutyl chloroformate-based sample
resolution capillary gas-chromatography and preparation protocol for chiral and nonchiral
gas-chromatography mass-spectrometry of amino acid analysis by gas chromatography. In:
protein and nonprotein amino-acids, amino- Alterman MA, Hunziker P (eds) Amino acid
alcohols, and hydroxycarboxylic acids as their analysis: methods and protocols, Methods in
tert-butyldimethylsilyl derivatives. J High molecular biology, vol 828, pp 137–152.
Resoult Chromatogr 17(3):145–152 h t t p s : / / d o i .
6. Kanani HH, Klapa MI (2007) Data correction org/10.1007/978-1-61779-445-2_13
strategy for metabolomics analysis using gas 15. Husek P, Svagera Z, Hanzlikova D et al
chromatography-mass spectrometry. Metab (2012) Survey of several methods deprotein-
Eng 9(1):39–51. https://doi.org/10.1016/j. izing human plasma before and within the
ymben.2006.08.001 chloroformate-mediated treatment of amino/
7. Villas-Boas SG, Smart KF, Sivakumaran S, carboxylic acids quantitated by gas chroma-
Lane GA (2011) Alkylation or silylation for tography. J Pharm Biomed Anal 67–68:159–
analysis of amino and non-amino organic acid 162. https://doi.org/10.1016/j.
by GC-MS? Meta 1(1):3–20. https://doi. jpba.2012.04.027
org/10.3390/metabo1010003 16. Svagera Z, Hanzlikova D, Simek P et al (2012)
8. Husek P (1997) Urine organic acid profiling by Study of disulfide reduction and alkyl chloro-
capillary gas chromatography after a simple formate derivatization of plasma sulfur amino
sample pretreatment. Clin Chem acids using gas chromatography-mass spec-
43(10):1999–2001 trometry. Anal Bioanal Chem 402(9):2953–
2963. https://doi.org/10.1007/
9. Husek P, Simek P (2006) Alkyl chloroformates in
s00216-012-5727-y
sample derivatization strategies for GC analysis.
Review on a decade use of the reagents as esterify- 17. Wu YM, Li L (2016) Sample normalization
ing agents. Curr Pharm Anal 2(1):23–43. methods in quantitative metabolomics.
https://doi.org/10.2174/ J Chromatogr A 1430:80–95. https://doi.
157341206775474007 org/10.1016/j.chroma.2015.12.007
18. Bouatra S, Aziat F, Mandal R et al (2013) The 47(3):294–302. https://doi.org/10.1002/

human urine metabolome. PLoS One 8(9). jms.2952
h t t p s : / / d o i . o r g / 1 0 . 1 3 7 1 / j o u r n a l . 20. Dettmer K, Stevens AP, Fagerer SR et al (2012)
pone.0073076 Amino acid analysis in physiological samples by
19. Cimlova J, Kruzberska P, Svagera Z et al GC-MS with propyl chloroformate derivatiza-
(2012) In situ derivatization-liquid liquid tion and iTRAQ-LC-MS/MS. In: Alterman
extraction as a sample preparation strategy for MA, Hunziker P (eds) Amino acid analysis:
the determination of urinary biomarker prolyl- methods and protocols, Methods in molecular
4-hydroxyproline by liquid chromatography- biology, vol 828, pp 165–181. https://doi.
tandem mass spectrometry. J Mass Spectrom org/10.1007/978-1-61779-445-2_15
Chapter 12
Sheathless Capillary Electrophoresis-Mass Spectrometry

for the Profiling of Charged Metabolites in Biological
Samples
Rawi Ramautar
Abstract
Capillary electrophoresis (CE) is well suited for the profiling of highly polar and charged metabolites as
compounds are separated on the basis of their charge-to-size ratio. The protocol presented here is based
on using a recently developed sheathless interfacing design, i.e., a porous tip interface, for coupling CE to
electrospray ionization mass spectrometry (MS). It is demonstrated that sheathless CE-MS employing a
bare fused-silica capillary at low-pH separation conditions can be used for the profiling of both cationic
and anionic metabolites by only switching the MS detection and electrophoretic separation voltage polar-
ity. The proposed sheathless CE-MS protocol allows efficient and sensitive profiles to be obtained for a
broad array of charged metabolites, including amino acids, organic acids, nucleotides, and sugar phos-
phates, in various biological samples, such as urine and extracts of the glioblastoma cell line.
Key words Capillary electrophoresis, Mass spectrometry, Sheathless interface, Metabolomics,

Biological samples, Cationic metabolites, Anionic metabolites
1 Introduction
A key aim of metabolomics is to obtain an answer to a specific

research question, which may be of biological or clinical origin [1].
To achieve this goal, advanced analytical separation techniques are
often used for the global profiling of (endogenous) metabolites in
biological samples [2]. At present, the profiling of endogenous
metabolites is generally performed with MS in combination with
an online front-end chromatographic separation method [3].
Regardless of important developments in liquid chromatography
column technology and methodology, the selective and efficient
analysis of highly polar and charged compounds is still highly chal-
lenging. Capillary zone electrophoresis, referred to here as CE,
separates compounds on the basis of differences in their intrinsic
electrophoretic mobility, which is dependent on the charge and
size of the analyte. Therefore, CE is highly suited for the analysis
183
184 Rawi Ramautar
of polar and charged metabolites. Moreover, as the separation

mechanism of CE is fundamentally different from chromatographic-
based separation techniques, a complementary view on the compo-
sition of metabolites present in a given biological sample is
provided.
Currently, the use of CE-MS in the field of metabolomics is
relatively low as compared to other analytical techniques [4].
CE-MS is still considered a technically challenging approach by the
overall scientific community, suffering from a relatively poor repro-
ducibility and sensitivity. However, important to stress here is that
CE-MS has been used for the global profiling of native peptides and
endogenous metabolites in a clinical context for more than a decade
now. For example, Mischak and coworkers have analyzed peptides
in more than 20,000 human urine samples at different laboratories
with an acceptable interlaboratory reproducibility [5, 6].
Soga and coworkers were the first to show the utility of CE-MS
for the global profiling of metabolites in biological samples [7, 8].
CE is generally coupled to MS via a sheath-liquid interfacing tech-
nique [9, 10]; however, due to dilution of the capillary effluent by
the sheath liquid, the detection sensitivity is intrinsically
compromised.
Recently, it was demonstrated that the use of a sheathless inter-
face significantly improved the detection coverage of metabolites
present in various biological samples as compared to CE-MS
employing a classical sheath-liquid interface [11–14]. The sheath-
less interface used was based on a porous tip emitter, which was
invented by Moini [15], allowing the effective use of the intrinsi-
cally low-flow property of CE in combination with nano-ESI-MS.
In order to demonstrate the usefulness of sheathless CE-MS
and to further expand the role of this method in metabolomics, a
protocol is presented describing how this approach can be used for
the analysis of charged metabolites in various biological samples,
exemplified here for an extract of the human glioblastoma cell line
and human urine. Some of the procedures outlined in this protocol
has also been demonstrated in a visual manner recently [16]. Here,
it is shown that by only switching the MS detection and electro-
phoretic separation voltage polarity, a single sheathless CE-MS
method can be used for the profiling of cationic and a wide range
of anionic metabolites using exactly the same capillary and separa-
tion conditions, thereby reducing analysis time for the global pro-
filing of charged metabolites.
2 Materials
Prepare all solutions using ultrapure water (prepared by purifying

deionized water to obtain a sensitivity of 18 MΩ-cm at 25 °C) and
analytical-grade reagents.
CE-MS for Profiling Charged Metabolites in Biological Samples 185
2.1 Solutions 1. Background electrolyte (BGE) solution: 10% (v/v) acetic acid,
and Samples pH 2.2. Add 9.0 mL of water into a 10 mL glass vial, and add
for Analysis 1.0 mL of acetic acid to the water in a fume hood. Mix the solu-
tion thoroughly using a vortex. Store at 4 °C.
2. Metabolite standard mixture: dissolve 50 μL of a 50 μM cation
standard mixture containing the basic twenty l-amino acids
into 50 μL of water, and mix the solution thoroughly (see Note
1). Store at −80 °C when not in use. Dissolve 50 μL of a 50 μM
anion standard mixture containing 17 anionic metabolites into
50 μL of water, and mix the solution thoroughly (see Note 1).
Store at −80 °C when not in use. The anionic metabolites
include (1) 2-naphthol-3,6-disulfonic acid; (2) d(+)2-
phosphoglyceric acid; (3) d-ribose-5-phosphate; (4) d-glucose-
1-phosphate; (5) d-glucose-6-phosphate; (6)
d-fructose-6-phosphate; (7) inosine 5′-monophosphate; (8)
guanosine 3′,5′-cyclic monophosphate; (9) guanosine 5′-mono-
phosphate; (10) citric acid; (11) trimesic acid; (12) isocitric
acid; (13) gluconic acid; (14) adenosine 3′,5′-cyclic monophos-
phate; (15) 2-hydroxybutyric acid; (16) b-diphosphopyridine
nucleotide (NAD+); and (17) 3-hydroxybutyric acid.
2.2 Analytical 1. The protocol reported here can only be performed with a com-
Equipment mercially available sheathless CE equipment, also known as
CESI 8000 (Sciex, A98089). Dependent on the type of MS
instrument, a dedicated nanospray source is required for
hyphenating sheathless CE to MS, information which can be
obtained from the vendor.
2. For the electrophoretic separations, commercially available
fused-silica capillaries (dimensions, 30 μm ID × 90 cm total
length) are used (Sciex, B07367).
3 Methods
The protocol described here for the use of sheathless CE-MS for
metabolic profiling studies is for laboratory use only. Prior to using
this protocol, consult all relevant material safety data sheets
(MSDS). Please use all appropriate laboratory safety procedures,
including safety glasses, lab coat, and gloves, when performing the
experiments described in this protocol.
3.1 Setting 1. Place a new bare fused-silica cartridge with a porous tip emitter
Up the CE System (30 μm ID × 90 cm total length) in the CE instrument.
2. Check for flow of liquid through the capillary by performing a
forward rinse at 50 psi for 15 min using 100% methanol (see
Note 2). Carry out also a rinse in the opposite direction at
186 Rawi Ramautar
50 psi for 5 min using 100% methanol to check the flow of liq-
uid through the conductive capillary (see Note 3).
3. Rinse the separation capillary with water at 50 psi for 10 min by
keeping the porous tip section, that is, the sprayer tip, in a
50 mL Falcon tube containing 5 mL of water.
4. Rinse the separation capillary with 0.1 M NaOH at 50 psi for
10 min, then by water at 50 psi for 10 min, and finally with
BGE at 50 psi for 10 min.
3.2 Coupling 1. Remove the sprayer tip of the fused-silica cartridge from the
Sheathless CE to MS water tube, and install it in the nanospray source adapter for
coupling to the MS instrument (see Note 4). The ESI voltage is
set to 0 during this step.
2. Ensure that the height of the BGE vials in the CE instrument
matches the height of the sprayer tip.
3. Check for flow of liquid through the conductive capillary by
rinsing with BGE at 50 psi for 5 min (see Note 5).
4. Rinse the separation capillary with BGE at 50 psi for 10 min in
the forward direction (see Note 2).
5. Position the porous tip emitter at the entrance of the MS inlet
at a distance of circa 2–3 mm. Apply a voltage of 30 kV using a
ramp time of 1 min, and start acquiring MS data in the m/z
range from 65 to 1000 m/z for metabolic profiling studies
using first an ESI voltage of 0 (see Note 6).
6. Set the ESI voltage to 1000 V while continue recording data.
Increase the ESI voltage with increments of 100 V until a con-
stant background signal is observed.
7. Optimize the porous tip emitter position with respect to the
center of the MS inlet by moving it in the x, y, or z-direction in
order to see which position provides the maximal and most sta-
ble MS signal (see Note 7).
8. After determining the optimal ESI voltage, set the ESI voltage
to 0, and decrease the CE voltage from 30 kV to 1 kV using a
ramp time of 5 min (see Note 8).
9. Create on the basis of the optimized parameters a CE-MS
method for the analysis of metabolite standards and biological
samples.
3.3 Preparation 1. After culturing a human glioblastoma cell line U-87 MG

of Extracts (ATCC HTB-14) in Dulbecco’s Modified Eagle Medium
from the Human (Invitrogen) supplemented with 5% fetal calf serum, 4 mM of
Glioblastoma Cell Line l-glutamine, 25 mM of d-glucose and 1 mM of sodium pyru-
vate, and 100 μg/mL of penicillin/streptomycin (see Note 9),
wash the adherent human U-87 MG glioblastoma cells with
ice-cold saline solution (0.9%; w/v) to remove the remaining

residues of culture medium.
2. Add ice-cold methanol/water solution (8/2, v/v) to quench
the cellular metabolism.
3. Scrape the adherent cells using a 25 cm cell scraper.
4. Collect the methanol/water solution in a tube and snap-freeze
in liquid nitrogen.
5. Add chloroform to the methanol/water fraction (final ratio
8/8/2, v/v/v), and centrifuge the sample for 10 min at 4 °C
and 16,100 × g.
6. Collect the methanol/water layer and evaporate this fraction
using a SpeedVac concentrator. Reconstitute the dried material
in 50 μL water for analysis by sheathless CE-MS. When not in
use, store the sample at −80 °C.
3.4 Preparation 1. Collect human urine samples of healthy subjects, pool the sam-
of Human Urine ples, and store at −80 °C prior to usage.
Samples 2. Prior to sheathless CE-MS analysis, mix the pooled urine sam-
ple with the BGE (1:1, v/v), and centrifuge for 10 min at 4 °C
and 16,100 × g.
3.5 Analysis 1. Add 20 μL of the anionic or cationic metabolite standard mix-

of Metabolite ture into an empty 100 μL microvial (PCR vial) which fits into
Standards a CE vial, and put this vial in the inlet sample tray.
and Biological 2. Rinse the separation capillary with BGE at 50 psi for 3 min fol-
Samples lowed by sample injection at 2.0 psi for 60 s (~20 nL corre-
sponding to circa 3% of the capillary volume). Then perform
BGE injection at 1.0 psi for 10 s.
3. Start MS data acquisition and apply a voltage of −30 kV (ramp
time of 1.0 min) and a pressure of 0.5 psi at the inlet for anionic
metabolic profiling or only +30 kV (ramp time of 1.0 min) for
cationic metabolic profiling for 30 min (see Note 10). After the
30 min electrophoretic separation, stop MS data acquisition,
and decrease the CE voltage to −1 or +1 kV using a ramp time
of 5 min.
4. Between sample injections, rinse the capillary with water, 0.1 M
sodium hydroxide, and BGE each at 30 psi for 3 min.
5. Evaluate the recorded data by determining the migration times
and the signal intensity of the analyzed anionic and cationic
metabolite mixtures. Check whether the anionic metabolite
standards appear in the region between 10 and 28 min (Fig. 1).
Check also whether three structurally related isomers, i.e.,
d-glucose-1-phosphate, d-glucose-6-phosphate, and d-
fructose-6-phosphate, are partially separated (see Note 11).
188 Rawi Ramautar
Fig. 1 Multiple extracted ion electropherograms obtained for the analysis of anionic metabolite standards
metabolite (25 μM) with sheathless CE-MS in negative ion mode using a sheathless porous tip sprayer. Peaks:
(1) 2-naphthol-3,6-disulfonic acid; (2) d(+)2-phosphoglyceric acid; (3) d-ribose-5-phosphate; (4) d-glucose-1-
phosphate; (5) d-glucose-6-phosphate; (6) d-fructose-6-phosphate; (7) inosine 5′-monophosphate; (8) guano-
sine 3′,5′-cyclic monophosphate; (9) guanosine 5′-monophosphate; (10) citric acid; (11) trimesic acid; (12)
isocitric acid; (13) gluconic acid; (14) adenosine 3′,5′-cyclic monophosphate; (15) 2-Hydroxybutyric acid; (16)
b-Diphosphopyridine nucleotide (NAD+); (17) 3-Hydroxybutyric acid. Experimental conditions: BGE, 10% acetic
acid (pH 2.2); separation voltage, −30 kV (+0.5 psi applied at the inlet of the CE capillary); sample injection,
2.0 psi for 60 s (Reproduced from ref. 13 with permission from the authors)
6. Assess whether the cationic metabolite standards appear in

the region between 8 and 22 min. Check whether isoleucine
and leucine are migrating between 15 and 15.5 min, and deter-
mine if the resolution is circa 0.5 (Fig. 2).
7. Use the procedures described in Subheading 3.3, steps
1–6, for anionic and cationic metabolic profiling of extracts of
the glioblastoma cell line and human urine samples. A typical
profile obtained for cationic metabolites in an extract from the
glioblastoma cell line (cell density: ~20 cells/nL) by sheathless
CE-MS is shown in Fig. 3.
8. After analysis of the biological samples, analyze anionic
and cationic metabolite standard mixtures to assess whether the
performance of the sheathless CE-MS system is still adequate in
terms of expected migration times, peak shapes, and detection
sensitivity (see Note 12).
9. After the analyses or when not in use, rinse the capillary with
water at 50 psi for 15 min, and store the inlet part of the capil-
lary in a vial containing water and the porous section (outlet
part) in a tube also containing water (see Note 13).
Fig. 2 Extracted ion electropherogram obtained for the analysis of isoleucine and leucine (25 μM) with sheath-
less CE-MS in positive ion mode. Experimental conditions: BGE, 10% acetic acid (pH 2.2); separation voltage,
+30 kV; sample injection, 2.0 psi for 60 s
Fig. 3 Metabolic profile (total ion electropherogram) obtained for an extract of a glioblastoma cell line (cell
density ~20 cells/nL) with sheathless CE-MS in positive ion mode. Experimental conditions: BGE, 10% acetic
acid (pH 2.2); separation voltage, +30 kV; sample injection, 2.0 psi for 60 s (Reproduced from ref. 13 with
permission from the authors)
190 Rawi Ramautar
4 Notes
1. The cationic and anionic metabolite standard mixtures are

stable for at least 3 months when stored properly.
2. If no drop formation is observed at the end of the capillary
porous tip emitter, then repeat this step at a pressure of
100 psi. If no drop is observed under these conditions, then a
new capillary needs to be installed. For a visual overview of
this procedure, we would like to refer to reference [16].
3. Rinsing with methanol is only required when installing a new
capillary cartridge.
4. Prior to coupling the CE capillary to MS, ensure that the MS
instrument has been calibrated and connected to the CE
system.
5. During this rinsing step, a drop formation at the base of the
ESI sprayer needle should be observed. If not, repeat the pro-
cedure at a pressure of 100 psi. Install a new capillary when no
drop formation is observed under these conditions.
6. The mass spectrum should be void of signal as an electrospray
voltage is not applied.
7. The sheathless CE-MS method is based on a porous tip emit-
ter which allows the effective use of the intrinsically low-flow
property of CE. Obtaining a stable ESI signal under these
conditions is a prerequisite for reproducible metabolic profil-
ing studies. Therefore, careful positioning of the porous tip
emitter with respect to the entrance of the MS inlet is critical
in order to obtain a stable/constant background signal.
8. We have observed that a gradual decrease of the CE voltage
after the electrophoretic separation improves the durability of
the porous tip capillary emitter for reasons not clear at this
stage.
9. Use T75 cm2 cell culture flasks at 37 °C under 5% CO2 in an
incubator for growing the cells.
10. Ensure that the MS instrument is used in negative ion mode
for anionic metabolic profiling and in positive ion mode for
cationic metabolic profiling.
11. Concerning the analysis of the isomers d-glucose-1-phosphate,
d-glucose-6-phosphate, and d-fructose-6-phosphate (peaks 4,
5, and 6, respectively, in Fig. 1), the resolution between the
first two peaks is circa 0.75 and of the last two peaks circa
0.50.
12. The analytical performance of the sheathless CE-MS method
for metabolic profiling studies needs to be evaluated daily
using metabolite standard mixtures. Under the same experi-
mental conditions, consistent migration times, i.e., variation

below 3% for within-day (n = 10) and between-day (n = 5)
using a 20 nL injection of a metabolite standard mixture
(25 μM), peak areas (variation below 20%) and plate numbers
(ranging between 50,000 and 400,000) are typically obtained.
Detection limits in the nanomolar range are obtained for most
metabolite standards under these conditions. In case these
data are not obtained for the metabolite standards, the MS
instrument needs to be tuned and recalibrated, or the porous
tip capillary emitter needs to be changed.
13. When the sheathless CE-MS method is not in use, it is impor-
tant to disconnect the separation capillary and to store the
inlet side of the capillary in water and the outside submerged
with the protective sleeve in a tube also containing water to
prolong capillary lifetime. When proper rinsing and storage
conditions are used in combination with a suitable sample pre-
treatment procedure, a single porous tip capillary can be used
on average for 100 analyses.
Acknowledgment
Dr. Rawi Ramautar would like to acknowledge the financial sup-

port of the Veni and Vidi grant scheme of the Netherlands
Organization for Scientific Research (NWO Veni 722.013.008 and
Vidi 723.016.003).
References
1. Ramautar R, Berger R, van der Greef J et al ery and clinical application. Proteomics Clin
(2013) Human metabolomics: strategies to Appl 9(3–4):322–334
understand biology. Curr Opin Chem Biol 7. Soga T, Ueno Y, Naraoka H et al (2002)
17(5):841–846 Simultaneous determination of anionic inter-
2. Kuehnbaum NL, Britz-McKibbin P (2013) mediates for Bacillus subtilis metabolic path-
New advances in separation science for metab- ways by capillary electrophoresis electrospray
olomics: resolving chemical diversity in a post- ionization mass spectrometry. Anal Chem
genomic era. Chem Rev 113(4):2437–2468 74(10):2233–2239
3. Theodoridis GA, Gika HG, Want EJ et al 8. Soga T, Ohashi Y, Ueno Y et al (2003)
(2012) Liquid chromatography-mass spec- Quantitative metabolome analysis using capil-
trometry based global metabolite profiling: a lary electrophoresis mass spectrometry.
review. Anal Chim Acta 711:7–16 J Proteome Res 2(5):488–494
4. Ramautar R (2016) CE-MS in metabolomics: 9. Maxwell EJ, Chen DD (2008) Twenty years
status quo and the way forward. Bioanalysis of interface development for capillary
8(5):371–374 electrophoresis-electrospray ionization-mass
5. Pejchinovski M, Hrnjez D, Ramirez-Torres A spectrometry. Anal Chim Acta
et al (2015) Capillary zone electrophoresis on- 627(1):25–33
line coupled to mass spectrometry: a perspec- 10. Bonvin G, Schappler J, Rudaz S (2012)
tive application for clinical proteomics. Capillary electrophoresis-electrospray
Proteomics Clin Appl 9(5–6):453–468 ionization-
mass spectrometry interfaces: fun-
6. Pontillo C, Filip S, Borras DM et al (2015) damental concepts and technical develop-
CE-MS-based proteomics in biomarker discov- ments. J Chromatogr A 1267:17–31
192 Rawi Ramautar
11. Bonvin G, Veuthey JL, Rudaz S et al (2012) 14. Hirayama A, Tomita M, Soga T (2012) Sheathless
Evaluation of a sheathless nanospray interface capillary electrophoresis-mass spectrometry with a
based on a porous tip sprayer for CE-ESI-MS high-sensitivity porous sprayer for cationic metab-
coupling. Electrophoresis 33(4):552–562 olome analysis. Analyst 137(21):5026–5033
12. Ramautar R, Busnel JM, Deelder AM et al 15. Moini M (2007) Simplifying CE-MS opera-
(2012) Enhancing the coverage of the urinary tion. 2. Interfacing low-flow separation tech-
metabolome by sheathless capillary niques to mass spectrometry using a porous
electrophoresis-mass spectrometry. Anal Chem tip. Anal Chem 79(11):4241–4246
84(2):885–892 16. Zhang W, Gulersonmez MC, Hankemeier T
13. Gulersonmez MC, Lock S, Hankemeier T et al et al (2016) Sheathless capillary electrophoresis-
(2016) Sheathless capillary electrophoresis- mass spectrometry for metabolic profiling of
mass spectrometry for anionic metabolic profil- biological samples. J Vis Exp (116).
ing. Electrophoresis 37(7–8):1007–1014 doi:https://doi.org/10.3791/54535
Part III
Plant/Food Applications
Chapter 13
Two-Phase Extraction for Comprehensive Analysis

of the Plant Metabolome by NMR
Jan Schripsema and Denise Dagnino
Abstract
Metabolomics is the area of research, which strives to obtain complete metabolic fingerprints, to detect
differences between them, and to provide hypothesis to explain those differences [1]. But obtaining com-
plete metabolic fingerprints is not an easy task. Metabolite extraction is a key step during this process, and
much research has been devoted to finding the best solvent mixture to extract as much metabolites as
possible.
Here a procedure is described for analysis of both polar and apolar metabolites using a two-phase
extraction system. D2O and CDCl3 are the solvents of choice, and their major advantage is that, for the
identification of the compounds, standard databases can be used because D2O and CDCl3 are the solvents
most commonly used for pure compound NMR spectra. The procedure enables the absolute quantifica-
tion of components via the addition of suitable internal standards. The extracts are also suitable for further
analysis with other systems like LC-MS or GC-MS.
Key words Two-phase extraction, NMR, Metabolic fingerprints, Plants, Identification, Quantification
1 Introduction
Good planning is essential in any research project, even more so in

metabolomic experiments due to the large numbers of samples
typically analyzed. Every step should be made clear, from the
question(s) to be answered to the methods and analytical tech-
niques selected to achieve the goal. The whole procedure should
be able to extract the maximum amount of information in a time-
and cost-effective way. Once defined, the procedure should be
strictly followed in order to allow the proper comparison of many
samples [1].
NMR and MS, coupled to various chromatographic tech-
niques, are today the main analytical methods used in metabolo-
mics. Both methods have specific advantages and disadvantages
and continue to develop and find useful applications.
195
196 Jan Schripsema and Denise Dagnino
Advantages of NMR- over MS-based metabolomics are the

reproducibility of NMR spectra and the possibility of direct quan-
tification [2]. Further a large dataset exists, available in the litera-
ture or specific data banks, containing spectra of pure compounds.
Nevertheless, since the spectra vary according to the solvent used,
direct comparison is only possible if the spectra to be compared
have been obtained in the same solvent.
The extraction of the samples is a critical step in the metabolo-
mics experiment, and much literature has been devoted to this sub-
ject, e.g. [3–5].
Most of the literature concerning NMR-based metabolomics
uses mixtures of solvents in order to maximize the number of
extracted metabolites [6–8]. But, when using this strategy, it is not
possible to compare the spectra obtained directly with the readily
available literature or databases.
In the present protocol, we focus on the extraction of plant
material, taking into account the need to obtain complete meta-
bolic fingerprints of the samples and the qualitative and quantita-
tive determination of the compounds therein. We demonstrate the
usefulness of a two-phase extraction system using water and chlo-
roform. Both solvents are commonly used for pure compound
NMR spectroscopic identification. Thus, the spectra obtained in
metabolomic experiments can be readily compared with the avail-
able literature for the identification of the compounds therein. The
spectra obtained with the two-phase method suggested are com-
pletely complementary, and very rarely does a compound appear in
both phases. The separation in two phases diminishes superposi-
tion of signals (Fig. 1) and makes identification and quantification
of the compounds in the extracts easier and more precise. It is a
quick and extremely simple extraction procedure that can be car-
ried out in targeted or untargeted approaches, and if necessary, the
extracts can be further analyzed by LC-MS or GC-MS.
After obtaining the NMR spectra, the next steps in the metab-
olomic experiment involve the preprocessing of the obtained
experimental results to make them suitable for the subsequent
analysis that generally involves multivariate analysis. An important
step in the preprocessing is the alignment of the signals in the dif-
ferent NMR spectra. To simplify the alignment, the process of bin-
ning or bucketing is generally used. Multivariate analysis will reveal
the differences between sample groups, indicating the peaks or sig-
nals which are specifically increased or decreased in certain
datasets.
In the final step of the experiments, these results should be
interpreted and the compounds responsible for the signals
identified.
In fact, one of the main difficulties in metabolomics is the dis-
covery of the identity of the compounds detected. Peaks or signals
are found which are specifically increased or decreased in certain
Two-Phase Extraction for Comprehensive Analysis of the Plant Metabolome by NMR 197
TMSP
D2O phase
T T
T
PPM 9.0 8.0 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0
C C
CDCI3 phase
P C
P P
PPM
9.0 8.0 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0
Fig. 1 1H NMR spectra from the (a) D2O and (b) CDCl3 extracts of coffee powder obtained with the two-phase
extraction. In the CDCl3 phase, signals from caffeine (marked with C) and pyridine (marked with P) are indi-
cated. In the D2O phase signals from TMSP and trigonelline (marked with T) are indicated
datasets, but linking those peaks to specific compounds needs a lot

of care, and frequently additional experiments are required. For
instance, in metabolomics based on LC-MS, major difficulties arise
in the comparison of chromatograms (because of shifts in retention
times) and/or spectra (reproducibility of fragmentation), when
data obtained by different instruments or research groups are com-
pared and when databases are consulted. In NMR spectroscopy
the reproducibility is not such a problem, which makes the com-
parison of datasets easier.
2 Materials
2.1 Internal A known amount of internal standard solution is added to each of

Standards the phases so as to allow the quantification of the compounds
extracted. For the aqueous phase, TMSP [3-(trimethylsilyl)
proprionic-2,2,3,3-d4 acid, sodium salt, also reffered to as TSP] is
used and for the organic phase pyridine. The TMSP spectrum has
only one signal at 0.00 ppm making it suitable for both quantifica-
tion and as a reference for the chemical shift.
For the organic phase pyridine is suggested since it provides
some advantages. The spectrum of pyridine contains three signals
of which two (at 8.61 ppm and 7.69 ppm) are clearly separated
from other signals. The third signal (7.29 ppm) is close to the
residual solvent signal and is more difficult to integrate. TMS,
often used as chemical shift reference in chloroform, is not advised
for quantification purposes, because it is more volatile and its signal
is quite narrow. Furthermore its relaxation time is higher than
most other compounds [9].
In metabolomics experiments, many signals are not identified,
but the peak areas should be corrected to permit quantitative com-
parison between samples. First of all instead of the areas, the rela-
tive areas in relation to the area of the IS are used (area/area IS
signal). This relative area should then be corrected for eventual
deviations of the actual quantities of the IS added to the tube, the
quantity of plant material extracted, and the quantity of solvent
used to extract the plant material and of the liquid transferred to
the NMR tube. This leads to the following multiplication factor
(MF):
MF = ( Actual Weight IS sol. / Weight 100 …lIS sol.)
(1000 mg of plant material / actual Quant.plant material )
( Quant.extraction solvent / Quant.extraction solvent transferred ) .
This factor is the same for all signals in the NMR spectrum of
a sample. In this way direct comparison between samples is
possible.
If the compound of interest has well-resolved signals which
can be integrated, the absolute quantity of the compound can be
calculated using the following formula:
Quant.IS in 100 …lIS sol. (mg)
( area compound signal / area IS siggnal )
MF ( no.Hs IS signal / no.Hs compound signal )
( mol.weight compound / mol.weight IS) .
This formula provides the absolute quantity per gram of plant
material. These calculations are illustrated in Table 1 for specific
signals in the spectra shown in Fig. 1.
2.2 Preparation Deuterated solvents should be at least 99.8% deuterated. The

of the Internal internal standards should be of analytical grade or higher. Weighing
Standard Solution of standards and solvents should be carried out with maximum
(ISS) precision.
Table 1
Calculations with the NMR data shown in Fig. 1. For each spectrum the calculation of the absolute
quantity of a specific compound is illustrated
D2O sample CDCl3 sample

Trigonelline Caffeine
9.11 ppm 3.59 ppm
IS 0.00 ppm IS 8.61 ppm
IS solution 47.0 mg/100.0 g 151 mg/99.73 g
Actual weight IS sol. 110 mg 170 mg
Weight 100 μl IS sol. 110 mg 150 mg
Actual quant. plant material 105 mg 105 mg
Quant. extraction solvent 875 mg 1206 mg
Quant. extr. solvent transferred 524 mg 532 mg
Multiplication factor 15.90331 24.46831
Quant. IS in 100 μl IS sol. (mg) 0.0517 mg 0.227 mg
Area compound signal 62.54 252.29
Area IS signal 78.11 159.66
Number of Hs IS signal 9 2
Number of Hs compound signal 1 3
Molecular weight compound 137.14 194.19
Molecular weight IS 172.27 79.1
Abs. quant. (mg/g) 4.72 14.36
Internal standard solution 1 (ISS1): weigh 150 mg of pyridine

(MW 79.10 g/mol) in an empty bottle and add 100 g CDCl3.
Internal standard solution 2 (ISS2): weigh 50 mg of TMSP
(MW 172.27 g/mol) in an empty bottle and add 100 g D2O.
Internal standard solutions can be kept for up to 1 year at
room temperature, if properly handled. Flasks should be kept
tightly closed when not in use. In case of handling a large number
of samples, distribute the prepared ISS in smaller bottles so as to
minimize the time each flask is open.
3 Methods
3.1 Extraction The extraction described here has been used for dried plant mate-
and Analysis rial (freeze dried) and also for the analysis of dried bacteria and
food stuffs such as teas, coffee, butter, and cheese. The procedure
Fig. 2 Schematic representation of the two-phase extraction for comprehensive analysis of the plant metabo-
lome by NMR
is schematically presented in Fig. 2. All pipetting should be carried

out on a precision balance and the weights registered (see Note 1).
When pipetting CDCl3 with conventional pipettes, before the
actual pipetting, aspire the liquid a few times to avoid dripping
during pipetting.
1. Weigh 100 mg of dried plant material in a 2 ml Eppendorf
tube (see Notes 2 and 3).
2. Add 0.80 ml of D2O to the plant material, and make sure all
material has been wetted (see Note 4).
3. Add 0.80 ml of CDCl3 to the wetted plant material (see
Note 5).
4. Thoroughly vortex the mixture.
5. Speed the extraction by ultrasonic mixing for 10 min.
6. Centrifuge at 13,800 × g for 5 min.
7. Take out the Eppendorf tubes carefully from the centrifuge.
8. Add 0.10 ml of ISS1 to an NMR tube.
9. Take out 0.5 ml of the D2O phase (upper phase) from the
Eppendorf tube, and transfer it to the same NMR tube.
10. Seal the NMR tube, and vortex to mix the extract and the
ISS1.
11. Take another NMR tube and add to it 0.10 ml of ISS2.
12. Take out 0.5 ml of the CDCl3 phase (lower phase) with the
pipette adjusted to 0.800 ml, and add to the same NMR tube
(see Notes 6–10).
13. Seal the NMR tube, and vortex to mix the extract and the
ISS2.
14. Acquire the 1H NMR spectra (100–128 scans) of the extracts
in both NMR tubes. For D2O use presaturation of the residual
solvent signal (see Notes 11–14).
4 Notes
1. Weighing increases the precision since pipetting is less precise,

especially when chloroform is pipetted using conventional
pipettes.
2. The proportion of sample to solvent can be changed depend-
ing on the concentration of the extracted compounds. The
amount of sample chosen should preferably not saturate the
solvents.
3. To determine the best proportion of sample to solvent vol-
umes, re-extract the sample with the same procedure, and
determine whether the extract contains around the predicted
value. If not the proportion of sample to solvent can be
adapted.
4. When the material is well wetted before the addition of CDCl3,
the extraction efficiency is increased.
5. CDCl3 is added after the water to avoid evaporation.
6. Plant material stays between the two phases. When taking out
the lower phase, the layer should be carefully traversed with
the tip of the pipette. It is usually possible to displace the pellet
formed between the phases by pushing it with the pipette tip
and thus gain direct access to the lower phase.
7. When taking out the lower layer, chloroform vapors decrease
the actual volume of the liquid. One should therefore adjust
the pipette to 0.800 ml and take out as much of CDCl3 as pos-
sible without taking along any of the aqueous phase.
8. When the other phase enters the pipette care should be taken
not to transfer this to the NMR tube. The actual quantity
which is transferred is determined by the weight of the liquid
transferred to the NMR tube.
9. If a drop of the CDCl3 extract falls on the balance plate, just
allow it to evaporate (usually very quick) and register the
weight after.
10. Minimum amount of volume in the NMR tube for a good
quality measurement is 0.50–0.60 ml. If for any reason the
volume is insufficient, record the weight of the extract and just
add more deuterated solvent to complete the volume.
11. 100 or 128 scans are generally sufficient to obtain good-quality
spectra, but the quantity can be increased (or decreased)
according to the need.
12. After obtaining the spectra, they can be processed manually or
automatically. This involves phasing and calibration. The peak
shape should be verified to check if the sample was correctly
shimmed. In the water samples, the TMSP signal can be
checked for symmetry. In the chloroform samples, the TMS or

residual solvent signal can be used for this purpose. If not ade-
quate the samples should be remeasured or discarded.
13. In the measurement of the aqueous extracts, presaturation of
the residual solvent signal is used. In our experiments the
Bruker program zgcppr was used. Typical parameters include
the relaxation delay of 5.0 s. Acquisition of 64 K data points
and 100 or 128 scans.
14. To analyze the extracts by mass spectrometry methods, it is
necessary to dilute the extract in the deuterated solvents at
least by a factor of 1000 with non-deuterated solvent. This will
also eventually result in the exchange of H with D.
References
1. Schripsema J, Dagnino D (2015) Metabolomics. 5. Heyman HM, Meyer JJM (2012) NMR-based
In: Hostettmann K, Stuppner H, Marston A, metabolomics as a quality control tool for
Chen S (eds) Handbook of chemical and bio- herbal products. S Afr J Bot 82:21–32
logical plant analytical methods, 1st edn. Wiley, 6. Kim HK, Verpoorte R (2010) Sample prepara-
New York tion for plant metabolomics. Phytochem Anal
2. Schripsema J (2010) Application of NMR in 21:4–13
plant metabolomics: techniques, problems and 7. Beltran A, Suarez M, Rodríguez MA et al
prospects. Phytochem Anal 21:14–21 (2012) Assessment of compatibility between
3. Deda O, Gika HG, Wilson IA, Theodoridis GA extraction methods for NMR and LC/MS-based
(2015) An overview of fecal preparation for metabolomics. Anal Chem 84:5838−5844
global metabolic profiling. J Pharm Biomed 8. Kim HK, Choi YH, Verpoorte R (2010) NMR-
Anal 113:137–150 based metabolomic analysis of plants. Nat
4. Kim H-S, Park SJ, Hyun S-H et al (2011) Protoc 5:536–549
Biochemical monitoring of black raspberry 9. Schripsema J (2008) Comprehensive analysis of
(Rubus coreanus Miquel) fruits according to polar and apolar constituents of butter and mar-
maturation stage by 1H NMR using multi- garine by Nuclear Magnetic Resonance, reflect-
ple solvent systems. Food Res Int 44: ing quality and production processes. J Agric
1977–1987 Food Chem 56:2547–2552
Chapter 14
NMR Spectroscopy Protocols for Food Metabolomics

Applications
Evangelia Ralli, Maria Amargianitaki, Efi Manolopoulou, Maria Misiak,
Georgios Markakis, Sofia Tachtalidou, Alexandra Kolesnikova,
Photis Dais, and Apostolos Spyros
Abstract
NMR spectroscopy has become an indispensable tool for the metabolic profiling of foods and food prod-
ucts. In the present protocol, we report an analytical approach based on liquid-state NMR for the deter-
mination of polar and nonpolar metabolites in some common liquid (wine, spirits, juice) and solid (cheese,
coffee, honey) foods. Although the diversity of foods precludes the use of a single protocol, with small
modifications, the proposed methodologies can be adapted to a broader range of foodstuffs.
Key words NMR spectroscopy, Metabolite profiling, Food analysis, Authentication, Quality control
1 Introduction
High-resolution liquid-state NMR spectroscopy is a powerful ana-

lytical spectroscopic technique that can provide quantitative infor-
mation on the chemical composition of liquid or solid but soluble
mixtures of organic compounds. Compound identification and
verification can be achieved either by comparison with literature
data or public NMR spectroscopy databases. Even unknown com-
pounds can be detected and identified by performing more com-
plex 1D and 2D NMR experiments prior to quantification.
NMR spectroscopy has a long tradition in food analysis [1, 2]
and in recent years has had a very significant contribution in the
explosion of multivariate statistical analysis applications exploring
the rich information inherent in the chemical profile (low MW
organic compounds, usually secondary metabolites) of foods [3].
Together with mass spectrometry, NMR spectroscopy helped form
a powerful pair of modern analytical techniques that became the
workhorse and defined and shaped the field of modern food
metabolomics [1, 4].
203
204 Evangelia Ralli et al.
Foods can be liquid or solid, and their chemical composition

diversity necessitates the use of a broad array of pre-analysis sample
preparation procedures in order to extract useful metabolomic
information from the analysis of NMR spectra. Liquid foods, con-
taining a high concentration matrix of water or water/ethanol,
represent the simpler analytical case scenario for NMR analysis
(wine [5–7], spirits [8], juices [9]). These food samples can either
be directly analyzed or, if increased sensitivity is necessary, the
water/ethanol matrix can be easily removed via evaporation or
freeze-drying (Subheading 3.1). Solid foods that are completely
soluble in water or other solvents, such as honey, may also be
directly analyzed by liquid-state NMR spectroscopy [10, 11]. In
the case of solid insoluble foods, one can in principle resort to
solid-state NMR methodologies utilizing high-resolution magic-
angle spinning (HR-MAS) for food metabolomics applications, an
approach that is gaining interest in recent years [12, 13]. However,
for foods containing a significant amount of lipids, such as cheese
[14–16] or coffee [17, 18], access to the water-soluble metabo-
lome is limited by lipid interference, and vice versa, necessitating
the application of more complicated extraction protocols for
obtaining both the lipid and polar metabolite profiles (Subheading
3.2).
In the present protocol, we report two experimental protocols
that can be used for the successful metabolomic analysis of several
important foodstuffs. In the first protocol, simpler methodology
proposed (Subheading 3.1), liquid foods can be studied directly
(juice, alcoholic beverages, honey) or after water/ethanol removal
(wine). In the second methodology proposed (Subheading 3.2),
suitable extraction procedures are performed prior to the NMR
spectroscopic determination of both the water-soluble metabolite
and lipid profiles of the foods. We have applied methods in
Subheading 3.2 to several types of hard and soft cheeses and sev-
eral types of coffee (see Fig. 1) and cocoa grains, but it can be used
with slight modifications for other complex and even whole foods.
2 Materials
2.1 Direct Analysis 1. Screw cap glass vials (2 mL).

of Liquid or 2. A laboratory ultrasound sonicator.
Water-Soluble Foods
3. A laboratory micro-centrifuge.
4. Analytical balance.
5. Variable volume pipette (100–1000 μL).
6. Ultrapure water.
7. 5 mm diameter glass NMR tubes (or appropriate size for the
spectrometer probe used).
NMR Spectroscopy Protocols for Food Metabolomics Applications 205
Fig. 1 1H NMR spectra of coffee obtained by applying method in Subheading 3.2: lipid profile in CDCl3 (top) and
polar metabolite profile in D2O (bottom), both spectra obtained at a proton frequency of 500.13 MHz
8. Freeze dryer (Telstar Cryodos).

9. Deuterium oxide (D2O) containing 0.05% TMSP (trimethyl-
silyl propionic acid sodium salt, internal standard).
10. Oxalate buffer (pH = 4).
11. Phosphate buffer (pH = 7.4).
2.2 Analysis 1. Screw cap glass vials (4 mL).

of Solid Foods 2. Eppendorf vials (1.5 mL).
3. A laboratory ultrasound sonicator.
4. A laboratory micro-centrifuge.
5. Analytical balance.
6. Variable volume pipette (100–1000 μL).
7. Pestle and mortar.
8. Liquid nitrogen (0.5 L).
9. Ultrapure water.
10. Glass flasks (50 or 100 mL).
11. 5 mm diameter glass NMR tubes (or appropriate size for the
spectrometer probe used).
12. Freeze dryer (Telstar Cryodos).

13. D2O containing 0.05% TMSP (trimethyl-silyl propionic acid
sodium salt-d4, internal standard).
14. CDCl3 containing 0.03% TMS (tetramethylsilane, internal
standard).
15. NMR spectrometer with high-resolution liquid-state probe.
2.3 Buffer Oxalate buffer (pH = 4): transfer 0.0595 g of oxalic acid and
Preparation 0.1795 g of sodium oxalate in a 20 mL volumetric flask, dissolve in
D2O (99.9 atom % D), and leave in ultrasound sonicator for
15 min.
Phosphate buffer (pH = 7.4): prepare phosphate buffer
(pH 7.4) by weighing 0.721 g Na2HPO4, 0.131 g NaH2PO4,
1 mM TSP (0.025 g), and 3 mM NaN3 (0.542 g) into a 25 mL
volumetric flask. Add 5 mL of D2O and fill up to 25 mL with
water. Shake thoroughly, and leave in a sonicator at 40 °C, inter-
spersed by shaking the flask, until the salts are dissolved.
3 Methods
3.1 Direct Analysis 1. Store 2 mL of sample in a 4 mL screw cap vial at −18 °C for
of Liquid or 12 h.
Water-Soluble Foods 2. Lyophilize for at least 12 h.
3.1.1 Analysis of Wine 3. Add 400 μL D2O containing 0.05% TMSP and 200 μL oxalate
buffer.
4. Centrifuge in an Eppendorf vial at 13, 1 48 × g for 10 min.
5. Transfer the supernatant into a 5 mm NMR tube.
6. Run NMR experiment Protocol A.
3.1.2 Analysis 1. Transfer 600 μL of sample to a 5 mm NMR sample tube.

of Alcoholic Beverages
2. Add 100 μL of D2O containing 0.05% TMSP.
(Spirits)
3. Vortex for 1 min.
4. Run NMR experiment Protocol A.
3.1.3 Analysis of Honey 1. Weigh ~150 mg of sample in a 4 mL screw cap vial.

2. Add 600 μL of D2O containing 0.05% TMSP.
3. Vortex for 1 min and transfer into a 5 mm NMR sample tube.
4. Allow to equilibrate for 2–3 days at room temperature in the
dark (see Note 1).
5. Run NMR experiment Protocol B.
3.1.4 Analysis of Juice 1. Transfer 600 μL of juice to a 1.5 mL Eppendorf vial.

2. Centrifuge at 13,148 × g for 10 min.
3. Transfer 300 μL of the supernatant to a 5 mm NMR sample
tube.
4. Add 200 μL of phosphate buffer and 100 μL of D2O containing
0.05% TMSP.
6. Vortex for 1 min.
3.2 Analysis 1. Store 5 g of sample at −18 °C for 24 h.

of Solid Foods 2. Cut frozen sample to small pieces and weigh.
3. Place sample in 100 mL glass flask, connect to freeze dryer, and
3.2.1 Sample freeze-dry for 16 h.
Pretreatment
4. Weigh samples again to calculate the amount of moisture.
5. Grind sample with a pestle and mortar under liquid nitrogen
(see Note 2).
3.2.2 Polar Metabolite 6. Weigh 3× 0.30 g of ground sample into three Eppendorf vials
Extraction (total 0.90 g).
7. Add 1 mL of ultrapure water in each vial by a variable volume
pipette 100–1000 μL.
8. Seal vials with laboratory film, place into sonicator bath for
30 min, and centrifuge at 13,148 × g for 10 min (see Note 3).
9. Carefully remove the aqueous phase via pipette, and transfer
into a tarred 50 mL glass flask.
10. Re-extract the remaining pellet two more times (repeat steps
7–9).
11. After the extraction in triplicate, store the three Eppendorf
vials at −18 °C (to be used for the extraction of lipids in step
17) (see Note 4).
12. Freeze at −18 °C the glass flask with collected extracts and
freeze-dry for 16 h.
13. After freeze-drying, weigh the precipitates, and transfer
approx. half of the precipitate into a 4 mL screw cap glass vial.
14. Add 700 μL D2O containing 0.05% TMSP via a variable vol-
ume pipette in the glass vial, and place into ultrasound sonica-
tor bath for 30 min (see Note 5).
15. Filter the polar extracts through glass wool tightly packed into
a Pasteur pipette, directly into a 5 mm NMR tube.
3.2.3 Nonpolar
Metabolite Extraction 17. Add 1 mL chloroform in each Eppendorf vial obtained in step
11.
18. Seal the three Eppendorfs with laboratory film, place into
ultrasound bath for 30 min, and centrifuge at 10,000 rpm
(6,708 × g) for 10 min.
19. Carefully remove the liquid phase and transfer into a glass flask
of 50 mL.
20. Re-extract the remaining pellet two more times with chloro-
form (repeat steps 17–19).
21. Evaporate the chloroform from the glass flask in a rotary
evaporator.
22. Add 700 μL CDCl3 containing 0.03% TMS (by the use of vari-
able volume pipette 100–1000 μL) in the glass flask with the
dried extracts, and place flask into ultrasound bath for 5 min.
23. Filter extracts carefully through glass wool directly into a
5 mm NMR tube.
24. Run NMR experiment Protocol C.
3.3 NMR Set the probe temperature to 298 K, and wait (5–10 min) until the
Spectroscopy sample temperature is equilibrated. Lock, tune, and shim the sam-
Experimental ple according to standard NMR spectrometer procedures (see Note
Protocols 6).
3.3.1 Protocol A. 1H NMR
Spectroscopy WET 1. Obtain a 1H NMR spectrum (Bruker, zg30); integrate and save
Experiment the spectral regions containing the water and ethanol peaks to
be suppressed per automation program directions (see Note 7).
2. Load a standard WET solvent multisuppression (Bruker/WET)
pulse program with default spectrometer parameters.
3. Record a WET multisuppressed 1H NMR spectrum with param-
eters SW = 20 ppm, NS = 256 scans, DS = 8 dummy scans,
AQ = 3.3 s, and D1 = 1 s.
4. Perform Fourier transformation, phase correction, and baseline
correction according to standard NMR spectrometer (or pro-
cessing software) procedures.
3.3.2 Protocol B. 1H NMR 1. Obtain a 1H NMR spectrum (Bruker, zg30), and record the
Spectroscopy Water exact frequency of the residual water proton peak.
Presaturation Experiment
2. Load a standard solvent presaturation (Bruker, zgpr) pulse pro-
gram with default spectrometer parameters, and set the fre-
quency of the residual water signal exactly on resonance.
3. Record a water-suppressed 1H NMR spectrum with parameters
SW = 12 ppm, NS = 256 scans, DS = 4 dummy scans,
AQ = 5.45 s, and D1 = 1 s.

3.3.3 Protocol C. 1H NMR 1. Load a standard proton NMR experiment (Bruker, zg30).
Spectroscopy Standard 2. Record a 1H NMR spectrum with parameters SW = 12 ppm,
Experiment TD = 64 K, NS = 256 scans, DS = 4 dummy scans, AQ = 3.3 s,
and D1 = 1 s.
3.4 NMR Data The use of an internal standard in the form of TMSP allows the
Analysis quantitative determination of polar metabolites in the food sam-
ples from the integration of proton signals in the 1H NMR spectra
obtained. The 1H NMR spectra of the nonpolar fraction of solid
3.4.1 Metabolite Profiling foods (such as cheese, coffee, cocoa) can be used to quantify the
fatty acid profile of the lipid fraction of these foods. In either case,
identification of metabolites can be achieved through spiking
experiments (when the metabolite is expected or suspected) and
verified through 2D NMR spectroscopy. Other means of metabo-
lite identification include the use of publicly available NMR spec-
tral databases, such as FoodDB, HMDB, BMRB, etc. Multivariate
statistical analysis methods can be applied directly on the quantita-
tive NMR metabolite profiles, in order to develop models for
studying food authentication, quality control, cultivar, pedocli-
matic effects, etc.
3.4.2 Metabolite For metabolite fingerprinting applications, the whole NMR spec-
Fingerprinting trum is used for statistical analysis and development of multivariate
metabolomic models. The first step is to convert the NMR spectra
into ASCII files and/or use suitable software (AMIX) to segment
the NMR spectra into buckets, for variable reduction purposes.
The size (width) of the buckets is user-defined, usually between
0.005 and 0.01 ppm for 1H NMR data, and the spectra are nor-
malized to the total sum of integrals of all buckets. Regions of the
spectra that do not contain any signal or contain solvent signal
(e.g., water or ethanol) can be omitted from the bucketing proce-
dure. If the NMR spectra need alignment, due to pH, tempera-
ture, or other sample variations, this should be performed using
specialized software, before the bucketing procedure [19].
In the data matrix obtained, the different food samples make
up the rows, while the buckets make up the matrix columns,
respectively. The dataset can be subjected to multivariate statistical
analysis using a variety of commercial and academic/public soft-
ware packages, including STATISTICA, SIMCA, R, etc. Analysis
of variance (ANOVA) and unsupervised principal component
analysis (PCA) should be undertaken initially, in order to examine

the dataset for potential outliers and remove them from subse-
quent analysis.
PCA can also be used for the observation of possible trends
and relationships between the different categories/classes of food
samples. In order to enhance the separation among the different
classes of samples, supervised methods such as partial least squares
discriminant analysis (PLS-DA) or orthogonal partial least squares
discriminant analysis (OPLS-DA) can be used, using class mem-
bership as a dependent variable. PLS-DA and OPLS-DA models
also offer the advantage of producing robust models that can be
validated externally using test samples (food samples that have not
been used for model development) and then used for class predic-
tion of unknown samples. In all metabolomics models described,
the loading plots obtained by various statistical methods indicate
compound/spectral regions characteristic for particular traits of
the food samples that are strong contributors to the obtained class
separation/classification.
4 Notes
1. Full equilibration in water of the sugar molecules is necessary to

avoid spectral differences due to nonequilibrium forms between
samples.
2. For samples already in the ground form (coffee, cocoa), this
step is omitted.
3. Three phases are formed in the Eppendorf vial, a white solid
lipid phase at the bottom, an aqueous phase in the middle, and
a thin nonpolar liquid phase at the top.
4. If lipid metabolite profiling is not required, the Eppendorf vials
may be discarded.
5. For ground coffee/cocoa, an extra centrifugation of the sample
solution (10 min) may be performed after step 14.
6. It is of utmost importance that all NMR experiments in a
metabolomic dataset should be conducted under exactly the
same conditions, i.e., NMR sample volume, pH, temperature,
and pulse sequence parameters (receiver gain, number of scans,
relaxation delay, etc.).
7. The WET experiment is included in standard Bruker spectrom-
eter automation routines.
References using 1H-NMR profiling. Food Chem 189:60–

66. https://doi.org/10.1016/j.foodchem.
1. Spyros A, Dais P (2012) NMR spectroscopy in 2014.11.099
food analysis. RSC Food Analysis Monographs. 11. Kazalaki A, Misiak M, Spyros A et al (2015)
Cambridge Identification and quantitative determination
2. Uryupin AB, Peregudov AS (2013) Application of carbohydrate molecules in Greek honey by
of NMR techniques to the determination of employing 13C NMR spectroscopy. Anal
the composition of tobacco, coffee, and tea Methods 7(14):5962–5972. https://doi.
products. J Anal Chem 68(12):1021–1032.
org/10.1039/c5ay01243k
https://doi.org/10.1134/ 12. Valentini M, Ritota M, Cafiero C et al (2011)
S1061934813120125 The HRMAS-NMR tool in foodstuff charac-
3. Monakhova YB, Kuballa T, Lachenmeier DW terisation. Magn Reson Chem 49(Suppl
(2013) Chemometric methods in NMR spec- 1):S121–S125. https://doi.org/10.1002/
troscopic analysis of food products. J Anal mrc.2826
Chem 68(9):755–766. https://doi. 13. Ritota M, Casciani L, Failla S et al (2012)
org/10.1134/S1061934813090098 HRMAS-NMR spectroscopy and multivariate
4. Spyros A (2016) Application of NMR in food analysis meat characterisation. Meat Sci
analysis. In: Ramesh V (ed) Specialist periodi- 92(4):754–761. https://doi.org/10.1016/j.
cal reports: nuclear magnetic resonance, vol meatsci.2012.06.034
45. The Royal Society of Chemistry, 14. Gianferri R, Maioli M, Delfini M et al (2007) A
Cambridge, pp 269–307. https://doi. low-resolution and high-resolution nuclear
org/10.1039/9781782624103-00269 magnetic resonance integrated approach to
5. Košir IJ, Kidrič J (2001) Identification of investigate the physical structure and metabolic
amino acids in wines by one-and two- profile of Mozzarella di Bufala Campana
dimensional nuclear magnetic resonance spec- cheese. Int Dairy J 17(2):167–176
troscopy. J Agric Food Chem 49(1):50–56 15. Lamanna R, Piscioneri I, Romanelli V et al
6. Lee JE, Hwang GS, Van Den Berg F et al (2008) A preliminary study of soft cheese
(2009) Evidence of vintage effects on grape degradation in different packaging conditions
wines using 1H NMR-based metabolomic by 1
H-NMR. Magn Reson Chem
study. Anal Chim Acta 648(1):71–76 46(9):828–831
7. Son HS, Ki MK, Van Den Berg F et al (2008) 16. Schievano E, Pasini G, Cozzi G et al (2008)
1
H nuclear magnetic resonance-based metabo- Identification of the production chain of
lomic characterization of wines by grape variet- Asiago d’Allevo cheese by nuclear magnetic
ies and production areas. J Agric Food Chem resonance spectroscopy and principal compo-
56(17):8007–8016 nent analysis. J Agric Food Chem
8. Fotakis C, Kokkotou K, Zoumpoulakis P et al 56(16):7208–7214
(2013) NMR metabolite fingerprinting in 17. Bosco M, Toffanin R, De Palo D et al (1999)
grape derived products: an overview. Food Res High-resolution 1H NMR investigation of cof-
Int 54(1):1184–1194. https://doi. fee. J Sci Food Agric 79(6):869–878
org/10.1016/j.foodres.2013.03.032 18. Consonni R, Cagliani LR, Cogliati C (2012)
9. Belton PS, Colquhoun IJ, Kemsley EK et al NMR based geographical characterization of
(1998) Application of chemometrics to the 1H roasted coffee. Talanta 88:420–426
NMR spectra of apple juices: discrimination 19. Savorani F, Tomasi G, Engelsen SB (2010)
between apple varieties. Food Chem Icoshift: a versatile tool for the rapid alignment
61(1–2):207–213 of 1D NMR spectra. J Magn Reson
10. Spiteri M, Jamin E, Thomas F et al (2015) Fast 202(2):190–202
and global authenticity screening of honey
Chapter 15
Direct Injection Analysis of Fruit VOCs by PTR-ToF-MS:

The Apple Case Study
Brian Farneti
Abstract
The instrumental characterization of volatile organic compounds (VOCs) is essential to have a precise,
reliable, and reproducible estimation of food aroma and, therefore, of the overall product quality. In this
report, we introduce four analytical approaches based on PTR-MS (proton transfer reaction-mass spec-
trometry) technology suitable to fully investigate the complexity of apple aroma. In our opinion, these
proposed methodologies can be applied, with slight modification, to every kind of fruit for destructive and
nondestructive rapid VOC fingerprinting.
Key words PTR-ToF-MS, Profiling, VOCs, Aroma, Autosampler, Artificial chewing
1 Introduction
Studies focused on food quality improvement have to consider the

complexity of flavor profile, its evolution in time, and the interac-
tion with consumers. Volatile organic compounds (VOCs) are pro-
duced and released in most stages of the food production chain,
“from farm to fork”; therefore they play a relevant role in agro-
industrial processes.
The monitoring of VOCs produced by fruits and vegetables
needs analytical techniques that are capable of dealing with chal-
lenging issues: (1) the need for separating and quantifying VOCs
in complex gas mixtures; (2) the need to detect concentrations that
may span a large range, from trace levels to parts per million; and
(3) the need to track concentrations that rapidly change over time
[1]. Because of these experimental constraints, the ideal methodol-
ogy for VOC monitoring should be highly selective, with high sen-
sitivity and dynamic range and with high time resolution.
Non-chromatographic techniques, based on direct injection
mass spectrometric VOC assessment, are receiving great interest
mainly (1) because of their capacity to carry out rapid, high-
throughput measurement of large sample sets without affecting
213
214 Brian Farneti
samples and without interfering with the VOC production process

and (2) because of the possibility of rapid process monitoring [1].
VOC monitoring for the abovementioned applications not only
relies on high time resolution but also takes advantage of the high
sensitivity provided by most direct injection mass spectrometry
(DIMS) technologies. Rapid monitoring does not allow for pre-
concentration, and the compounds of interest may well be present
in the gas phase in the sub-ppt range nevertheless inducing a rele-
vant effect.
Besides its technological performances (e.g., sensitivity and
selectivity), advanced DIMS is also increasingly being used because
of its stability since the mass/charge ratio does not vary with the
experimental conditions. However, the greatest difficulty arising in
DIMS, due to the lack of chromatographic separation, is the need to
identify the compounds that generate the observed peaks, since the
latter can be the results of overlapping signals from the mix of differ-
ent VOCs present in the sample. A number of real-time mass spec-
trometric techniques are available for food VOC analysis, including
MS-e-noses, atmospheric pressure chemical ionization mass spec-
trometry (APCI-MS), selected ion flow tube-MS (SIFT-MS), and
proton transfer reaction-mass spectrometry (PTR-MS).
In the present protocol, we reported four analytical approaches
based on PTR-MS technology suitable to fully investigate the
complexity of apple aroma. In our opinion, these proposed meth-
odologies can be applied, with slight modification, to every kind of
fruit or vegetable.
1.1 PTR-MS PTR-MS is an emerging technique, developed in the mid-1990,

Technology that has already found a wide range of applications in ecological
and environmental monitoring, agriculture and food sciences, and
medical diagnostics [2, 3]. PTR-MS is a form of soft chemical ion-
ization mass spectrometry based on proton transfer from a proton-
ated reagent, most commonly H3O+. The fundamental ionization
process can be written as
H3O+ + R ® RH + + H 2O
Protonated water (H3O+) interacts with the trace gas molecule

(R). During this interaction, a proton transfers from the hydro-
nium to the trace gas molecule, which leads to a protonated and
therefore an ionized molecule (RH+) and a neutral water molecule
(H2O). This proton transfer reaction is energetically favorable for
all VOCs with a proton affinity higher than that of water (691 kJ/
mol). This criterion excludes the major components of air such as
N2, O2, and CO2 (consequently not interfering with the
measurement) but includes many trace gases including most vola-
tile organic compounds (VOCs).
Direct Injection Analysis of Fruit VOCs by PTR-ToF-MS: The Apple Case Study 215
The main constituents of a PTR-MS apparatus are the ion

source, the reaction region (drift tube), the mass analyzer, and the
ion detector.
In most apparatus, the H3O+ primary ion beam is produced by
a hollow cathode ion source. A promising development of the hol-
low cathode ion source is the use of primary parent ions other than
H3O+ to allow the detection of compounds with proton affinities
larger than that of H2O. This possibility was already investigated
by Yeretzian, Jordan, and Lindinger [4] using NH4+ ions to bracket
compounds for coffee aroma analysis. More recently, other ions
have been studied (e.g., O2+ or NO+), leading to the development
of a marketed commercial system allowing for rapid switching
between different ions: O2+, NO+, Kr+, and Xe+. Selected ion flow
technique (SIFT) provides this unique advantage as well, but the
better selection of the parent ions is paid by a lower sensitivity [5].
The second part of a PTR-MS apparatus is the drift region,
where the parent ions are driven by an electric field and eventually
interact with the VOC to be detected. The process is controlled by
drift-tube temperature, pressure, and electric potential. Decreasing
the drift-tube volume increases the maximum obtainable time res-
olution increasing reaction efficiency (higher sensitivity).
The last part of a PTR-MS system is the mass analyzer. Linear
quadrupoles were used in most instruments. They are robust and
relatively cheap, but have time resolution of the order of seconds, and
typically provide only unit mass resolution. An alternative to over-
come these limitations of the quadrupole mass analyzer is the cou-
pling of PTR with a time-of-flight (ToF) mass analyzer. The great
advantage of ToF analyzers, built upon the observation that at the
same kinetic energy, heavier ions fly more slowly than lighter ones, is
the enhanced analytical information provided. In fact, they may reach
a resolution up to 7000 (m/Δm) for commercial PTR-MS instru-
ments, thus allowing the separation of many isobaric compounds and
the simultaneous monitoring of multiple peaks at the same nominal
mass. Moreover, if proper mass calibration is applied to ToF spectra,
it is also possible to identify the compound sum formula. The whole
spectrum is acquired in a split second, while a scan with a quadrupole
can take a minute or more to get the same sensitivity.
1.2 Nondestructive Headspace VOC fingerprint by PTR-MS provides a potential tool

Analysis of Intact Fruit for discriminating fruit and vegetable not only based on genetic
differences but also based on origin and maturity stages. The non-
destructive application of PTR-MS may be not only important for
cultivar characterization but also in product origin separation and
physiological experiments related to fruit ripening and quality.
1.3 Destructive The correlation between VOC emitted by an intact fruit and its
Analysis of Fresh Fruit internal content is still commonly accepted. Indeed, most of the
Tissue investigations carried out on VOC composition in fruit, and related
216 Brian Farneti
with physiological and genetic studies, are based on static head-

space analysis of intact fruit.
However, the blend of apple aroma compounds is subjected to
important modifications of severalfold change due to fruit cutting.
Therefore, the destructive analytical assessment of fresh fruit tissue
is essential in particular for studies related to fruit quality.
1.4 Dynamic The flavor of a product is not a stable trait, but it can rapidly and
Analysis by In Vitro drastically change during time, for instance, during fruit consump-
Mastication tion. Differences in VOC release behaviors may influence the
human aroma perception during food consumption since VOCs
are released from the matrix and then transported to mouth and
nose receptors.
In order to describe the release kinetics of VOCs while the
food matrix is being chewed, we developed an analytical system
based on an artificial chewing device coupled with the PTR-
ToF-MS. This system allowed a precise dynamic VOC fingerprint-
ing while the food is processed.
1.5 Automated The high biological variability between samples is one of the main
Analysis of Frozen risks that has to be considered during the design of an experiment
Fruit Tissue focused on VOC assessment. Therefore, a high number of biologi-
cal replicates are fundamental for a statistically correct experimen-
tal design. The possibility to couple a PTR-ToF-MS instrument
with a multipurpose sampler allowed the analysis of more than 200
samples a day with high reproducibility and with reduced lab labor.
The limiting factor of this methodology is, however, the restricted
volume of the vials that have to be used (20 ml). This restriction
can be exceeded by using powdered frozen tissue of the fruit (i.e.,
cortex or peel).
2 Materials
2.1 Nondestructive 1. Sealed glass jars with a volume of 1000 ml. Each jar lid needs
Analysis of Intact Fruit two entries (one inlet and one outlet) with the possibility to be
tightly closed.
2. A laboratory water bath.
3. An analytical balance.
4. A zero air producer.
5. PTR-ToF-MS 8000 instrument (Ionicon Analytik GmbH,
Innsbruck, Austria).
The PTR-ToF-MS drift tube is set with the following condi-
tions: 110 °C drift-tube temperature, 2.25 mbar drift pressure, and
550 V drift voltage. This leads to an E/N ratio of about 140
Townsend, where E corresponds to the electric field strength and
N to the gas number density.
The sampling time per channel of ToF acquisition was 0.1 ns,
amounting to 350,000 channels for a mass spectrum ranging up to
m/z = 400. Every single spectrum is the sum of about 28,600
acquisitions lasting 35 μs each, resulting in a time resolution of 1 s.
2.2 Destructive 1. Sealed glass jars with a volume of 250 ml. Each jar lid needs two
Analysis of Fresh Fruit entries (one inlet and one outlet) with the possibility to be
Tissue tightly closed.
2. A laboratory water bath.
3. A cork borer (diameter of 1.70 cm).
4. Homemade six-blade knife (1 cm distance between blades).
5. A zero air generator.
Innsbruck, Austria). The PTR-ToF-MS setting is the same
described in Subheading 2.1.
2.3 Dynamic 1. A laboratory water bath.

Analysis by In Vitro 2. A cork borer (diameter of 1.70 cm).
Mastication
5. Artificial chewing device [6, 7]. This homemade device is com-
posed of a cylindrical glass cuvette (800 ml) sealed with a cap
and a manual notched plunger (Fig. 1). All the device’s ele-
ments are made of polytetrafluoroethylene. The cuvette cup has
two entries: one inlet connected to the zero air generator and
one outlet connected to PTR-ToF-MS.
2.4 Automated 1. A knife.

Analysis of Frozen 2. A basic analytical mill with grinding chamber resistant to liquid
Fruit Tissue nitrogen.
3. Deionized water.
4. Antioxidant solution (sodium chloride, citric acid, and ascorbic
acid).
5. Glass vials (volume of 20 ml) equipped with PTFE/silicone
septa.
7. GC autosampler (MPS Multipurpose Sampler, GERSTEL).
218 Brian Farneti
Fig. 1 Schematic representation of the chewing device: cylindrical glass cuvette of 800 ml sealed with a cap
and a notched plunger controlled manually. The heat map indicates the dynamic VOC fingerprinting of apple
fruit assessed by PTR-ToF-MS coupled with the artificial chewing device. Three examples of VOC release dur-
ing the dynamic analysis are shown (m/z 61.028, 81.07, 45.033). Each graph is divided by a line, at time 0, in
two phases, respectively, before and after the chewing moment
3 Methods
3.1 Nondestructive 1. A whole intact fruit, without any damages, is weighted and
Analysis of Intact Fruit placed into a 1000 ml glass jar and incubated for 30 min at
30 °C into the water bath (see Note 1).
2. After the incubation, the headspace of the samples is directly
connected to the PTR-ToF-MS instrument via a heated PEEK
tube (110 °C. 0.055″ diameter) and sampled at a flow rate of
40 standard cm3 per min (sccm). At the same time, the jar is
continuously flushed with zero air in order to avoid air contami-
nation and under pressure (see Notes 2–9).
3. Sampling measurement was performed over 60 cycles resulting
in an analysis time of 60 s/sample.
3.2 Destructive 1. A cylindrical portion of cortex tissue is sampled with the cork
Analysis of Fresh Fruit borer along the vertical lengthwise plane of the fruit, avoiding
Tissue the core portion with seeds.
2. From this cylinder, five identical disks with a diameter of
1.70 cm and 1 cm thick are cut with a homemade six-blade
knife. Adopting this strategy, we avoid any possible influence of
fruit size, and we have tissue samples of different fruit regions.
3. These five disks are placed into a 250 ml glass jar and incubated
for 30 min at 30 °C into the water bath (see Note 1).
4. After the incubation, the headspace of the samples is directly

connected to the PTR-ToF-MS instrument via a heated PEEK
tube (110 °C. 0.055″ diameter) and sampled at a flow rate of
40 sccm. At the same time, the jar is continuously flushed with
zero air in order to avoid air contamination and under pressure
(see Notes 2–9).
5. Sampling measurement was performed in 60 cycles resulting in
an analysis time of 60 s/sample.
3.3 Dynamic 1. A cylindrical portion of cortex tissue, with a diameter of 1.70 cm

Analysis by In Vitro and a length of 5 cm, is sampled with the cork borer along the
Mastication vertical lengthwise plane of the fruit, avoiding the core portion
with seeds.
2. This cortex tissue is placed inside the artificial chewing device,
constantly maintained at 36 °C into the water bath.
3. Before crushing the headspace VOC concentration of the apple
flesh cylinder was measured for 60 s. The chewing was per-
formed by pressing the notched plunger five times within 10 s.
VOC analysis continued for 120 s following mastication.
4. The headspace of the samples is directly connected to the PTR-
ToF-MS instrument via a heated PEEK tube (110 °C. 0.055″
diameter) and sampled at a flow rate of 40 sccm. At the same
time, the jar is continuously flushed with zero air in order to
avoid air contamination and under pressure inside the jar (see
Notes 2–9).
3.4 Automated 1. The antioxidant solution is prepared mixing 100 ml of deion-

Analysis of Frozen ized water with 40 g of sodium chloride, 0.5 g of ascorbic acid,
Fruit Tissue and 0.5 g of citric acid.
2. Fruit tissue (i.e., cortex or peel) is rapidly cut in small pieces
with a sharp knife and immediately frozen into liquid nitrogen.
This material can be stored at −80 °C for several weeks before
the PTR-ToF-MS analysis.
3. Before being analyzed, this frozen material has to be ground
using an analytical mill with the grinding chamber maintained
at low temperature by using liquid nitrogen.
4. One gram of powdered frozen sample is immediately inserted
into 20 ml glass vials also maintained at low temperature by
liquid nitrogen (see Note 1).
5. One milliliter of the antioxidant solution is added to this
sample.
6. These prepared samples can be preserved at 4 °C till the analysis
(maximum 24 h).
7. Sampling measurement was performed over 60 cycles resulting
in an analysis time of 60 s/sample. Each measurement was con-
220 Brian Farneti
ducted automatically after 20 min of sample incubation at 40 °C

by using an adapted GC autosampler (MPS Multipurpose
Sampler, GERSTEL), and it lasted for about 2 min. During
measurements 100 sccm of zero air was continuously injected
into the vial, through a needle; the outflow was instead deliv-
ered via Teflon fittings to the PTR-ToF-MS through a second
heated needle (40 °C) (see Notes 2–9).
3.5 Data Analysis The analysis of PTR-ToF-MS spectral data (Fig. 2) proceeded cor-
recting offline the count loss (due to the ion detection dead time)
through Poisson statistics, following the method reported by
Cappellin et al. [8], while the internal calibration was performed
according to the procedure described in Cappellin et al. [9] (see
Notes 10–20). Such approach allowed a mass accuracy higher
than 0.001 Th to be achieved sufficient for the sum formula deter-
mination in our case. Compound annotation was carried out com-
paring the spectral profile with fragmentation data of reference
standards. Noise reduction, baseline removal, and peak intensity
extraction were performed according to Cappellin et al. [9], using
modified Gaussians to fit the peak shapes. Absolute headspace
VOC concentrations, expressed in ppbv (parts per billion by vol-
ume), were calculated from peak intensities according to the for-
mula described by Lindinger et al. [2]. A constant reaction rate
coefficient of 2 × 10−9 cm3/s was used in the calculations, intro-
ducing a systematic error of up to 30% that can be accounted for if
the actual rate coefficient is known [10].
Fig. 2 Example of PTR-ToF-MS spectrum of apple headspace. The main figure represents a sample spectrum
in the mass region between 1 and 420 Th, while the second figure enlarges the region around a selected nomi-
nal mass peak (m/z 31)
4 Notes
Τo measure with PTR-ToF-MS:

1. Put glass vials and caps in oven at 50 °C for at least 2 days to
remove volatiles from a silicon/PTFE septum.
2. Check if you have enough source of primary ion for long
experiments.
3. Check if the flow inside the vial would not raise a fine powder
or liquid from the surface of the sample.
4. Be sure that each sample VOC would not provoke primary ion
depletion. If yes, reduce the quantity, dilute the sample with
liquid (i.e., water), or mix the sample headspace with an inert
gas (Ar or N2).
5. Wait enough time between one measurement and another in
case of sticky compounds (i.e., α-farnesene for apple fruit). You
can even raise the flow in order to clean the PEEK tube with
clean air. If you measure with autosampler, you should make
some trials before.
6. Control the drift pressure and avoid under- or overpressure in
the sample that you have to measure. The wrong drift pressure
can cause the problems with VOC protonation.
7. Do not leave the instrument with the ion source switched on if
you do not measure anything. It will dirty ion source.
8. Do not smoke or use perfume when you prepare samples or
measure the lab air.
9. If it is possible, follow the experiment remotely when you are
measuring with autosampler.
Τo analyze PTR-ToF-MS data:
10. HDF5 file (.h5) is the standard of data acquisition from
PTR-ToF-MS.
11. Prepare a log file which consists of “Sample_name,” “File_
name,” “First_spectrum,” and “Last_spectrum.” The begin-
ning and the end of an experiment can be recorded manually
during online measurements. However, if some automation
was applied (autosampler, multiple valve control, etc.), it is
possible to retrieve these intervals also from HDF5 files in an
automatic way.
12. Determine a list of calibration peaks. To become a calibration
peak, a mass peak should be single in the window m/z ± 0.5
and have a good intensity (102–103 counts). We also need to
know the theoretical m/z of this mass peak. It can be pro-
duced by the instrument (H3O+, NO+, etc.) or derived from
the matrix (acetone, acetaldehyde, etc.).
222 Brian Farneti
13. Determine a list of test peaks which can be equal to calibration

peaks or can be enlarged by other peaks that you want to test.
14. Calibrate files according to the log file and calibration and test
peaks.
15. Check the calibration errors which are calculated as the abso-
lute maximum difference of each measurement between the
theoretical m/z and the m/z obtained during calibration.
16. If the calibration errors are under the threshold (0.001), it is
possible to proceed with the extraction of mass peak concen-
trations. If not, the calibration peaks should be changed,
reduced, or added with new ones, and the calibration should
be rerun.
17. For this extraction, it is necessary to have a log file, peaks used
for optimization of a general peak shape and another list of
peaks for calculation of resolution. These peaks are usually
chosen from the calibration list. Optionally it can be down-
loaded also the peak structure which contains the list of all
mass peaks which were extracted for a similar experiment.
18. There are two options of concentration calculation such as
online, where concentration for each mass peak is recorded for
each spectrum, and average (so-called headspace), where the
average concentration of each mass peak for the interval deter-
mined in a log file is calculated.
19. It is very useful to optimize and select the mass peaks which
will be extracted.
20. It is possible to extract the counts per second for each mass
peak and also the concentration in ppbv. For the latter one, it
is important to specify the primary ion which was used for the
experiment.
References
1. Biasioli F, Yeretzian C, Märk TD et al (2011) transfer-reaction mass-spectrometry. Int J Mass

Direct-injection mass spectrometry adds the Spectrom 223–224:115–139
time dimension to (B)VOC analysis. TrAC 5. Jordan A, Haidacher S, Hanel G et al (2009)
Trends Anal Chem 30:1003–1017 An online ultra-high sensitivity Proton-
2. Lindinger W, Hansel A, Jordan A (1998) On-line transfer-reaction mass-spectrometer combined
monitoring of volatile organic compounds at with switchable reagent ion capability (PTR +
pptv levels by means of proton-transfer-reaction SRI − MS). Int J Mass Spectrom 286:32–38
mass spectrometry (PTR-MS) medical applica- 6. Farneti B, Khomenko I, Cappellin L et al
tions, food control and environmental research. (2015) Dynamic volatile organic compound
Int J Mass Spectrom 173:191–241 fingerprinting of apple fruit during processing.
3. Biasioli F, Gasperi F, Yeretzian C, Märk TD Lebenson Wiss Technol 63:21–28
(2011) PTR-MS monitoring of VOCs and 7. Farneti B, Algarra Alarcón A, Cristescu SM
BVOCs in food science and technology. TrAC et al (2013) Aroma volatile release kinetics of
Trends Anal Chem 30:968–977 tomato genotypes measured by PTR-MS fol-
4. Yeretzian C, Jordan A, Lindinger W (2003) lowing artificial chewing. Food Res Int
Analysing the headspace of coffee by proton- 54:1579–1588
8. Cappellin L, Biasioli F, Granitto PM et al eters by a novel dead time correction. Rapid

(2011) On data analysis in PTR-TOF-MS: Commun Mass Spectrom 25(1):179–183
from raw spectra to data mining. Sensors 10. Cappellin L, Soukoulis C, Aprea E et al (2012)
Actuators B 155(1):183–190 PTR-ToF-MS and data mining methods: a new
9. Cappellin L, Biasioli F, Schuhfried E et al tool for fruit metabolomics. Metabolomics
(2011) Extending the dynamic range of proton 8(5):761–770
transfer reaction time-of-flight mass spectrom-
Chapter 16
LC-MS Untargeted Protocol for the Analysis of Wine

Panagiotis Arapitsas and Fulvio Mattivi
Abstract
This chapter describes a protocol for the analysis of the metabolomic fingerprint of wine by liquid
chromatography-mass spectrometry. The straightforward, optimized sample preparation procedure is lim-
ited to a single-step dilution with water or acetonitrile. The separation of wine analytes is carried out by
two columns with orthogonal selectivity, including both reversed-phase (C18) and hydrophilic interaction
(HILIC) chromatography, while the detection is assured by a high-resolution quadrupole time-of-flight
mass spectrometer operating in negative and positive electrospray ionization mode, in order to obtain four
different chromatograms for each sample. This validated protocol, or parts of it, could be applied in several
oenological topic experimental designs, including wine quality and wine authenticity.
Key words Vitis vinifera, Grape, Food, Holistic, Metabolomics, Mass spectrometry, Liquid chroma-
tography, HILIC
1 Introduction
The ultimate scope of analytical chemistry is to obtain the quanti-

tative characterization of the chemical composition and the eluci-
dation of the structure of all chemical entities present in a given
matrix. This is further complicated considering that in living
beings, the chemical composition is a dynamic concept, modulated
in time.
For many decades, targeted methods have been widely used
and taught in university courses of analytical chemistry. Their appli-
cation is very advanced, and the knowledge and expertise for
method development-validation are well established. However,
these methods are limited to a small number of known, predefined
analytes. Metabolomics developed and evolved as a consequence of
the need to obtain comprehensive characterization of the organic
molecules in any biological system [1]. In contrast to the targeted
methods, in metabolomics, the aim is to achieve the widest possible
metabolic coverage in an unsupervised manner, including the com-
pounds so far unknown. In the targeted approach, the majority of
225
226 Panagiotis Arapitsas and Fulvio Mattivi
the metabolites present in the matrix are ignored or d iscarded dur-

ing the cleanup, the method being entirely focused on the best
quantitation of the target compounds, while a basic assumption in
the untargeted approach is the desire to monitor as many com-
pounds as possible (ideally all). In consequence, the measured
metabolites are by definition not predefined, and the method devel-
opment and validation follow a different workflow in respect to the
targeted analysis [2–5].
Due to the current instrumental limitations and the large
chemical diversity, it is not possible to register all the metabo-
lome of one sample with a single analysis, so often complemen-
tary platforms are used. The LC-MS techniques are more widely
used because they can offer a broader coverage, by combining
reversed (semipolar metabolites) with normal-phase (polar
metabolites) chromatography and positive with negative elec-
trospray ionization mode in MS. Such an approach could be
also further complemented with a GC-MS protocol to cover the
volatiles and an adequate LC-MS method to cover the apolar
(lipids) metabolites [3, 4, 6, 7].
For the development of a metabolomic fingerprint method, the
critical issue to manage is to register the maximum number of fea-
tures (chromatographic peaks or pairs of m/z with retention time)
by maintaining the same sensitivity, retention time, and peak shape
during the analysis of all the samples of a sequence. This requires
substantial changes in how an analytical chemistry protocol is built,
since the usual process of calibration against one or more chemical
standard is usually error prone or ineffective in untargeted metabo-
lomics. This is due (1) to the huge chemical diversity of the analytes
and (2) to the heavy load of contaminants entering the instruments
when the aim of the analysis is the whole metabolome.
Delicate topics to be considered are also the analysis in the
same sequence of all the samples of a specific experiment (if possi-
ble), the biological variability, and the use of the quality control
samples. Important parameters for the targeted methods, such as
limits of detection and quantification, matrix effect, and linear
range, are difficult—if not impossible—to validate for the develop-
ment of an untargeted method [5]. We do not want to suggest that
they are impossible to achieve within an untargeted method, rather
that more robust quantitative data can be obtained developing a
targeted quantitative method for a selected number of metabolites,
usually defined as putative biomarkers, previously identified via
untargeted experiment. The aim of this chapter is to describe how
to setup an untargeted analytical protocol which can be used to
explore the wine metabolome, producing a validated dataset suit-
able for further data analysis, in search of the putative biomarkers.
This protocol describes the various steps of the sample prepa-
ration, sample analysis, and data analysis in order to obtain four
complementary chromatographic traces from wine samples:
Wine Metabolome 227
Fig. 1 Typical BPI LC-MS chromatogram of the same wine sample analyzed in four different modes: (a) HILIC
UPLC-QTOF MS ESI–, (b) HILIC UPLC-QTOF MS ESI+, (c) reversed-phase UPLC-QTOF MS ESI+, and (d)
reversed-phase UPLC-QTOF MS ESI–. From these chromatograms, it is clear that each mode covers different
parts of the wine metabolome, although some metabolites would be registered from more than one platform
reversed-phase (RP) and hydrophilic interaction (HILIC)

ultrahigh-performance liquid chromatography (UPLC) coupled to
a high-resolution quadrupole time-of-flight mass spectrometer
(QTOF MS) in negative (ESI–) and positive (ESI+) electrospray
ionization modes (Fig. 1). Since for metabolomics analysis it is
crucial to analyze all the samples within the same sequence, under
a randomized order and with the use of quality control (QC) sam-
ples, this protocol is applicable to a group of samples and not just
one sample. The number of samples, i.e., the size of the experi-
ment, is usually a compromise between the desire to increase, in
the “fat” data matrix, the number of samples which is always lower
than the number of variables measured and at the same time to
prevent the column contamination due to injection of multiple
samples within the sequence.
This protocol was already validated and used in wine-
related projects [8–11], and here we suggest it could be applied
in experimental designs dealing almost with any oenological
Fig. 2 (a) PCA plot of a reversed-phase LS-MS analysis, produced with the described protocol. The experimen-
tal design was based in 12 different wines produced by six grape varieties (five Pinot gris, three Grillo, one
Chardonnay, one Traminer, one Muller, and one Inzolia wines). For each wine eight different bottles were ana-
lyzed. The QC sample injections form a tight cluster in the center of the plot, indicating the quality of the data
set. Injections of the same wine biological replicates cluster together, indicating that this method could be used
for wine authenticity. (b) PCA plot of a HILIC-MS analysis, produced with the described protocol. Also here
injections belonging in the same wine clustered together. The numbers which tagged the QCs indicate the
order of the injection in the beginning of the sequence. In this experiment, eight injections were necessary to
equilibrate the LC-MS; that number is expected for a HILIC-MS untargeted experiment
topic/problem, including wine quality and wine authenticity

(Fig. 2). It can be also applied, with small adaptations, to other
liquid samples.
Wine Metabolome 229
2 Material
All solvents (methanol, acetonitrile, isopropanol, chloroform),

mobile phase additives (formic acid, ammonia) used should be of
LC/MS grade. Water should be ultra-purified (18.2 MΩ), and all the
other chemicals used should be of the highest purity (see Note 1).
2.1 Analytical The LC system used is a Waters Acquity UPLC, which included a
Instrumentation quaternary solvent manager, a sample manager, and a column
manager. For the reversed-phase LC, the column used is a Waters
Acquity UPLC 1.8 μm 2.1 × 150 mm HSS T3 column and for the
normal phase LC/HILIC, a Waters Acquity BEH amide column
2.1 mm × 100 mm × 1.7 μm (HILIC type), equipped with an
Acquity UPLC BEH Amide 1.7 μm VanGuard pre-column. The
MS system is a Waters Synapt HRMS QTOF MS.
2.2 Solutions Mobile phase A: a 1 L volumetric flask is half filled with water, then
1 mL of formic acid is added, and finally the flask is filled to volume
with water.
Mobile phase B: a 1 L volumetric flask is half filled with metha-
nol, then 1 mL of formic acid is added, and finally the flask is filled
to volume with methanol.
A stock solution of HCOONH4 4 M is prepared in advance
and stored at 4 °C. In detail, 12.6 g of HCOONH4 is weighed in
a beaker, and about 50 mL of water is added. The solution is stirred
for 4 h and then transferred to a 100 mL volumetric flask and filled
to volume with water.
For the mobile phase C, 5 mL of the HCOONH4 4 M stock
solution is added to a 1 L volumetric flask containing 950 mL of
water, 1 mL of NH4OH is added, and then the flask is filled to
volume (HILIC-LC/MS ESI-). For the mobile phase D, a 1 L
volumetric flask is half filled with acetonitrile, then 1 mL of
NH4OH is added, and finally the flask is filled to volume with ace-
tonitrile (HILIC-LC/MS ESI–) (see Notes 2 and 3).
For the mobile phase E, 5 mL of the HCOONH4 4 M stock
solution is added to a 1 L volumetric flask containing 950 mL of
water, 1 mL of formic acid is added, and then the flask is filled to
volume (HILIC-LC/MS ESI+). For the mobile phase F, a 1 L
volumetric flask is half filled with acetonitrile, then 1 mL of formic
acid is added, and finally the flask is filled to volume with acetoni-
trile (HILIC-LC/MS ESI+)) (see Notes 2 and 3).
Seal wash solvent is a mixture of 500 mL of water and 500 mL
of methanol.
Weak needle wash for the reversed-phase LC (and strong nee-
dle wash for the normal phase/HILIC) is a mixture of 100 mL
methanol and 900 mL water. Strong needle wash for the reversed-
phase LC (and weak needle wash for the normal phase/HILIC) is
a mixture of 250 mL methanol, 250 mL acetonitrile, 250 mL iso-

propanol, and 250 mL water.
Seal wash solvent is a mixture of 500 mL of water and 500 mL
of methanol.
3 Methods
3.1 QTOF MS The MS data are collected using different runs in positive and neg-
Parameters ative ESI mode over a mass range of 50–2000 amu for RP-LC and
30–1000 for HILIC with scan duration of 0.3 s in centroid mode.
The instrument operated in W mode. The transfer collision energy
and trap collision energy are set at 6 and 4 V. The source parame-
ters are set as follows: capillary 3 kV for positive scan and 2.5 kV
for negative scan, sampling cone 25 V, extraction cone 3 V, source
temperature 150 °C, desolvation temperature 500 °C, desolvation
gas flow 1000 L/h, and nebulizer gas 50 L/h.
External calibration of the instrument is performed at the
beginning of each batch of analysis by direct infusion of a sodium
formate solution (10% formic acid/0.1 M NaOH/CH3CN at a
ratio of 1/1/8) by controlling the mass accuracy from 50 to
2000 m/z for RP-LC and from 30 to 1000 m/z for HILIC (less
than 5 ppm) and mass resolution (over 14,000 FWHM). Lock
mass calibration is applied using a solution of leucine enkephalin
(0.5 mg/L, m/z 556.2771 for ESI+ and 554.2620 for ESI- mode)
at 0.1 mL/min. LC and MS instruments of similar specifications
can be used.
3.2 LC Parameters The column used is held at 40 °C during the analysis, the injection
volume is 10 μL, the flow rate is 0.28 mL/min, and the samples
3.2.1 RP-LC Instrumental are kept at 4 °C throughout the analysis. The multistep linear gra-
Parameters dient used is as follows (mobile phases A and B): 0–1 min, 100%
mobile phase A isocratic; 1–3 min, 100–90% A; 3–18 min, 90–60%
A; 18–21 min, 60–0% A; 21–25.5 min, 0% A isocratic; 25.5–
25.6 min, 0–100% A; and 25.6–28 min 100% isocratic (see Note
3).
3.2.2 HILIC Instrumental
Parameters The column and pre-column are held at 50 °C during the analysis,
the injection volume is 10 μL, the flow rate is 500 μL/min, and the
samples are kept at 4 °C throughout the analysis. For the ESI- MS,
the mobile phases C and D are used, and the multistep gradient
elution is as follows: 0–1 min, 88% mobile phase C; 1–2 min,
88–80% of C; 2–10 min, 80–72% C; 10–11 min, 72–50% C;
11–13 min, 50% C; 13–13.1 min, 50–88% C; and 13.1–18 min,
88% C. For the ESI+ MS, the mobile phases E (instead of C) and
F (instead of D) are used, under the same multistep gradient (see
Note 3).
Wine Metabolome 231
3.3 Sample Once the sampling is completed and all the samples are in the labo-
Preparation ratory, wines should be codified according to a randomized
sequence, so the sample preparation and analysis can be completed
following this randomized sequence (see Note 4).
Wines are uncorked under nitrogen atmosphere (see Note 5),
and an aliquot is transferred into a 15 mL amber vial (filled to
capacity). Then, again under nitrogen atmosphere, a QC pooled
sample is prepared using 0.5 mL of each sample, and this is treated
in the same way as the study samples. The QC can be used to opti-
mize the optimum dilution for the LC-MS system used and the
number of the samples (see Notes 6 and 7). For a Waters Acquity
UPLC-Synapt QTOF MS system used for this protocol, the fol-
lowing sample preparation is optimized for experimental designs
including 60–200 commercial wines (see Notes 8 and 9).
3.3.1 Sample Under N2 atmosphere, in a 2 mL Eppendorf microtube, 1 mL of

Preparation for RP-LC/MS each wine is diluted with 1 mL water (see Note 10), and 20 μL of
the internal standard is then added (10 mg o-coumaric acid in
10 mL of CH3OH). Then the solution is filtered with 0.2 μm
PTFE filters into a 2 mL amber vial (MS certificated) prior to LC/
MS analysis. The same procedure is followed for the blank, but
instead of wine 1 mL of water is used.
3.3.2 Sample Under N2 atmosphere, in a 2 mL Eppendorf microtube, 0.6 mL of

Preparation for HILIC-MS each wine is diluted with 1.2 mL acetonitrile, and 30 μL of the
internal standard is then added (14 mg of xanthosine and 25 mg of
nicotinic acid in 10 mL of CH3OH:H2O (1:1)). Then the solution
is filtered with 0.2 μm PTFE filters into a 2 mL amber vial (MS
certificated) prior to LC/MS analysis. The same procedure is fol-
lowed for the blank, but instead of wine 0.6 mL of water is used.
3.4 LC-MS System 1. Load mobile phases and wash solutions.

Start-Up and Pretest 2. Clean the ESI source according to the vendor protocol.
3. Prime the mobile phases for 5 min.
4. Connect the column.
5. Flush the column for 10 min at 0.05 mL/min, with 100% B
for reversed-phase LC or 100% D/F for normal phase LC.
6. Increase the column oven temperature at 40 °C for RP-LC
and 50 °C for HILIC.
7. Set the mobile phase at 50% B—50% A for RP-LC or 80%
D/F—20% C/E for HILIC.
8. Calibrate the MS and control the mass accuracy according to
the vendor protocol.
9. Bring gradually first the mobile phase % and then the flow rate
at the initial LC condition.
10. Load the UPLC method and wait until the back pressure
equilibration (psi delta <20).
3.5 Sample The samples should be injected according to the randomized order
Sequence established before the sample preparation (see Subheading 3.3).
Then the analysis should be made according to the following
sequence (see Notes 3, 6–9, 11–15):
1. First run a blank sample injection.
2. Then run four QC injections for the reversed phase and ten QC
injections for the normal phase LC, in order to equilibrate the
column (Fig. 2).
3. Now you can start the injection of the real samples, but for
every six real sample injections, one QC sample should be
injected.
4. The sequence should end with one QC sample and finally one
blank sample injection.
3.6 Data Analysis Data analysis in metabolomics follows a specific but very wide
workflow [11, 12] that should include the following steps:
1. Data preprocessing, divided in deconvolution, peak picking, fil-
tering (optional), alignment, and bucketing (binning).
2. Data pretreatment, subdivided into normalization (optional),
centering (optional), scaling, managing missing values, and
managing outliers.
3. Data processing, employing the use of statistical tools such as
multivariate supervised and unsupervised analysis.
4. Data visualization, such as PCA plots (Fig. 2).
5. Data quality validation.
6. Marker detection, annotation, and interpretation.
The researcher can use various informatics tools (one or more)
to accomplish the above workflow, based in his/her experience/
knowledge and financial support. For example, XCMS online [12]
is a valid, free, and widely used tool. In this protocol, we propose
the use of the commercial and user-friendly software Progenesis QI
(Waters, Nonlinear Dynamics). Other useful open source
informatics tools are mzMine [13], MetaDB [11], MetAlign [14],
and MetaboAnalyst [15].
Between the objects of this protocol, the marker detection and
validation, metabolite annotation, or the hypothesis generation is
not included. For this reason, the data analysis will be concluded
with the data visualization (Fig. 2) which is important to validate
the quality of the data set produced by the instrumental analysis
(see Note 15).
Progenesis QI steps:
Wine Metabolome 233
1. Make a new experiment, and choose a name for it and a direc-

tory to save the files created by Progenesis QI.
2. Set the analysis parameters according to your instrument, the
data format (centroided with 17,000 resolution in our case),
and ionization polarity of the analysis (positive or negative).
3. Import the raw files of the injections to Progenesis QI (decon-
volution and peak picking).
4. Start automatic processing (alignment), by selecting a QC from
the middle of the sequence as reference, untick the blank sam-
ples (never use blanks or standard mixes for the alignment), and
ignore ions eluting before the 1st minute and after the 22nd
minute of the chromatography.
5. Inspect visually the alignment but without performing any
manual changes.
6. In the experimental design page, divide your samples with
respect to the groups of the study (i.e., treatment vs. control or
each wine variety in a different group or wines from different
zones in different groups).
7. Do the peak picking after the experimental design set up, with-
out changing the parameters. This peak picking is applied to the
compounds (the software is grouping isotopic peaks and
adducts belonging to one compound) and not to ions as before.
8. After that go directly to the Compound Statistic page to visual-
ize the PCA plot. This is the first quality control of the analysis
(data). For a good quality data, the QC samples should group
together to a tight cluster (Fig. 2).
4 Notes
1. Please consult all relevant material safety data sheets (MSDS)

before use. Some of the chemicals used in this protocol are
acutely toxic and carcinogenic. Please use all appropriate safety
practices when performing the extraction including the use of
engineering controls (fume hood, glovebox) and personal
protective equipment (safety glasses, gloves, lab coat, full-
length pants, closed-toe shoes).
2. For an efficient hydrophilic interaction chromatography
(HILIC) method, analytes often exhibit the strongest reten-
tion when they are ionized, thus bases at low to mid-pH and
acids at mid- to high pH. This is the reason why we propose,
for the HILIC analysis, to combine acidic mobile phases with
ESI+ data acquisition and basic mobile phases with ESI– data
acquisition.
3. Prepare all the necessary mobile phases 1 day before the analy-
sis (especially for HILIC), and load all the volume in the LC
bottles from the beginning. Avoid adding mobile phase dur-
ing the analysis.
4. It is important to collect, register, and keep track of the details
and the characteristics of all samples analyzed. MetaDB [11] is
a useful tool for this aim. Register all the meta-data informa-
tion of the samples according to the minimum reporting stan-
dards for plant biology context information in metabolomics
studies [16].
5. N2 atmosphere is important to avoid possible O2-driven
reactions.
6. If the QC sample is a pooled sample, prepared by mixing equal
aliquots of each individual sample of the sample set, can pro-
vide import help for (a) training, (b) method development/
adaptation, (c) column equilibration, (d) data quality control,
and (e) marker quality evaluation.
7. Surrogate QC samples can be used in long-term studies which
include more than one sequence. Standard mixes are not con-
sidered good QC solution in the metabolomics field.
8. For commercial light-bodied white wines, it is possible to per-
form 200 injections without losing sensitivity. For full-bodied
red wines, the maximum injection suggested for this sample
preparation/instrumental setup is around 80 injections.
9. In case of a large number of samples (over 200), consider the
possibility to divide the sample set in smaller subsets. In this
case instrumental and data analysis should be made separately,
and then compared for common markers. All sample subsets
should have equally divided groups, i.e., same number of con-
trol and treatment condition samples, and not one subset with
the control samples and one subset with all the treatment
samples.
10. Degas the water used for the wine dilution.
11. Clean the LC tubing and pumps with a solution of 25% meth-
anol, 25% acetonitrile, 25% isopropanol, and 25% water every
2 weeks for reversed phase and every 5 days for HILIC.
12. If standard mixes are included in the sequence, inject them (a)
in the beginning of the sequence after the blanks and before
the QCs and (b) in the end of the sequence after the last QC
and before the last blank injection.
13. If blanks or standard mixes are included during the injections
of the real samples, run a minimum of two QC injections
before continuing with the real sample injections.
Wine Metabolome 235
14. Monitor during the sequence (a) the LC back pressure, (b)
retention time of specific metabolites, (c) mass error for spe-
cific metabolites, and (d) the area for specific metabolites.
15. Use data visualization with unsupervised multivariate statisti-
cal tools, such as PCA plot, to control the data quality, a step
which should be frequently monitored during the sequence of
analysis, by examining the QC distribution/clustering in the
PCA plot and/or possible outliers in the PCA plot (Fig. 2). If
you have to reinject—or re-prepare and reinject—more
sample(s), it is better to include it/them in the end of the
sequence before the last QC.
References
1. Nicholson JK, Lindon JC (2008) Systems biol- 10. Mattivi F, Arapitsas P, Perenzoni D et al (2015)
ogy: metabonomics. Nature 455:1054–1056 Influence of storage conditions on the compo-
2. Gika HG, Theodoridis GA, Vrhovsek U et al sition of red wines - advances in wine research -
(2012) Quantitative profiling of polar primary ACS symposium series. In: ACS symposium
metabolites using hydrophilic interaction ultrahigh series. ACS Publications, Washington, DC,
performance liquid chromatography-tandem mass pp 29–49
spectrometry. J Chromatogr A 1259:121–127 11. Franceschi P, Mylonas R, Shahaf N et al (2014)
3. Theodoridis G, Gika H, Franceschi P et al MetaDB a data processing workflow in untar-
(2011) LC-MS based global metabolite profil- geted MS-based metabolomics experiments.
ing of grapes: solvent extraction protocol opti- Front Bioeng Biotechnol 2:72
misation. Metabolomics 8:175–185 12. Smith CA, Want EJ, O' Maille G et al (2006)
4. Theodoridis GA, Gika HG, Want EJ et al XCMS: processing mass spectrometry data for
(2012) Liquid chromatography–mass spec- metabolite profiling using nonlinear peak
trometry based global metabolite profiling: a alignment, matching, and identification. Anal
review. Anal Chim Acta 711:7–16 Chem 78(3):779–778
5. Naz S, Vallejo M, García A et al (2014) Method 13. Katajamaa M, Miettinen J, Oresic M (2006)
validation strategies involved in non-targeted MZmine: toolbox for processing and visual-
metabolomics. J Chromatogr A 1353:99–105 ization of mass spectrometry based molecular
6. Buscher JM, Czernik D, Ewald JC et al (2009) profile data. Bioinformatics 22:634–636
Cross-platform comparison of methods for 14. Lommen A (2009) MetAlign: interface-driven,
quantitative metabolomics of primary metabo- versatile metabolomics tool for hyphenated
lism. Anal Chem 81:2135–2143 full-scan mass spectrometry data preprocess-
7. Cajka T, Fiehn O (2016) Toward merging ing. Anal Chem 81:3079–3086
untargeted and targeted methods in mass 15. Xia J, Wishart DS et al (2016) Using MetaboAnalyst
spectrometry- based metabolomics and 3.0 for comprehensive metabolomics data analysis.
Lipidomics. Anal Chem 88:524–545 Curr Protoc Bioinformatics 55:14.10.1–14.10.91.
8. Arapitsas P, Speri G, Angeli A et al (2014) The https://doi.org/10.1002/cpbi.11
influence of storage on the “chemical age” of 16. Sumner LW, Amberg A, Barrett D et al
red wines. Metabolomics 10:816–832 (2007) Proposed minimum reporting stan-
9. Arapitsas P, Ugliano M, Perenzoni D et al dards for chemical analysis chemical analysis
(2016) Wine metabolomics reveals new sulfo- working group (CAWG) metabolomics stan-
nated products in bottled white wines, pro- dards initiative (MSI). Metabolomics
moted by small amounts of oxygen. 3:211–221
J Chromatogr A 1429:155–165
Part IV
Life Science Applications

Chapter 17
Tissue Multiplatform-Based Metabolomics/Metabonomics

for Enhanced Metabolome Coverage
Panagiotis A. Vorkas, M. R. Abellona U, and Jia V. Li
Abstract
The use of tissue as a matrix to elucidate disease pathology or explore intervention comes with several
advantages. It allows investigation of the target alteration directly at the focal location and facilitates the
detection of molecules that could become elusive after secretion into biofluids. However, tissue metabolo-
mics/metabonomics comes with challenges not encountered in biofluid analyses. Furthermore, tissue
heterogeneity does not allow for tissue aliquoting. Here we describe a multiplatform, multi-method work-
flow which enables metabolic profiling analysis of tissue samples, while it can deliver enhanced metabo-
lome coverage. After applying a dual consecutive extraction (organic followed by aqueous), tissue extracts
are analyzed by reversed-phase (RP-) and hydrophilic interaction liquid chromatography (HILIC-) ultra-
performance liquid chromatography coupled to mass spectrometry (UPLC-MS) and nuclear magnetic
resonance (NMR) spectroscopy. This pipeline incorporates the required quality control features, enhances
versatility, allows provisional aliquoting of tissue extracts for future guided analyses, expands the range of
metabolites robustly detected, and supports data integration. It has been successfully employed for the
analysis of a wide range of tissue types.
Key words Metabolomics, Metabonomics, Metabolic profiling, Metabolic phenotyping, Lipidomics,

Tissue, Extraction, Metabolome, Lipidome, Coverage, Multiplatform, NMR, UPLC-MS, MSE,
HILIC
1 Introduction
Metabolomics/metabonomics applications in tissue, as a matrix for

studying disease or intervention, come with several advantages, all
of which derive from the fact that the target pathology/alterations
are directly examined. Firstly, this allows for easier and more rele-
vant mechanistic elucidation, which in turn can provide targets for
pharmacotherapeutic intervention. Additionally, differentially pro-
duced metabolites can be more easily identified in the correspond-
ing tissue, as they may be metabolized, excreted, diluted or
confounded by the complexity of the biofluid after secretion by the
tissue. After being identified as differentially produced within the
tissue, subsequent follow-up using targeted methods that have
239
240 Panagiotis A. Vorkas et al.
higher sensitivity may be applied in biofluids as a less invasive option.

Tissue candidate biomarkers can also be explored in the context of
in vivo imaging. Lastly, tissue analysis allows for the mapping of
organ-to-organ interactions and crosstalk, which has been proven
to be very informative when systemically studying disease [1].
The use of tissue as a matrix for metabolomics/metabonomics
applications has increased in the recent years. Papers employing tissue
metabonomics have tripled in the past 5 years. Although metabolite
extraction methods employ a solvent-based approach that was
described several decades back [2, 3], the current trend is diverging
from conventional approaches and moving toward more high-
throughput settings. Homogenization techniques such as the mortar
and pestle are now abandoned, and techniques such as bead beating
are favored [4]. At the same time, the traditional bilayer extraction
methodology is replaced by consecutive monolayer extractions using
solvents with different physicochemical properties. This can provide
higher repeatability [5]. Extensive studies in a comprehensive metab-
olomic setting aiming to identify the optimal solvent systems for
organic and aqueous extractions have also been conducted [6].
Here we describe a multiplatform, multi-method workflow
which enables metabolic profiling analysis of tissue samples and
covers an extended part of the metabolome [7]. This workflow
utilizes dual consecutive monolayer extractions. Tissue extracts are
subsequently analyzed by untargeted reversed-phase (RP-) and
hydrophilic interaction liquid chromatography (HILIC-) ultra-
performance liquid chromatography coupled to mass spectrometry
(UPLC-MS) and nuclear magnetic resonance (NMR) spectros-
copy. RP-UPLC-MS is employed for the analysis of organic extracts
(lipid profiling) and HILIC-UPLC-MS and NMR for the profiling
of aqueous extracts (polar metabolite profiling). This pipeline
incorporates the required quality control features, enhanced versa-
tility, repeatability, and reproducibility. It allows provisional ali-
quoting of tissue extracts for future analyses (which could be
guided by the results of the initial untargeted analyses), supports
data integration, and covers an extended part of the metabolome
with potential for detecting unknown molecules [8]. It has been
successfully employed for the analysis of several tissue types from
humans, rodents, and bovines, such as liver, brain, spleen, kidney,
breast (normal and cancerous), adipose, arteries and veins, athero-
sclerotic plaques, tissue segments covering the whole range of the
gastrointestinal track, skeletal muscle, heart, and also model organ-
isms such as C. elegans [6–12].
2 Materials
All solvents used should be LC-MS grade. All solvent mixtures are
described as volume parts per solvent. Efforts should be taken to
minimize the presence of contaminants in containers used (see
Note 1).
Tissue Multiplatform-Based Metabolomics/Metabonomics for Enhanced Metabolome… 241
2.1 Tissue 1. Dry ice.

Dissection 2. Weighing boats (large for cutting and small for weighing).
3. Balance (≤0.1 mg readability).
4. Single-use scalpels and forceps (see Note 2).
2.2 Tissue Extraction 1. Bead beating: bead beating tubes (2 mL), filled with 1 mm zir-
conium beads. Zirconium beads should cover just above the
curved bottom of the tube (see Notes 3 and 4).
2. Bead beater (Bertin Technologies).
3. Solvent mixture for organic extraction: methanol/methyl tert-
butyl ether (MTBE) (1:3).
4. Solvent mixture for aqueous extraction: water/methanol
(MeOH) (1:1)
5. Centrifuge (e.g., Eppendorf).
6. Eppendorf tubes 1–2 mL (see Note 5).
7. Solvent mixture evaporation using vacuum concentrator work-
ing at 30 °C (2 h for organic and 2.5 h for aqueous extraction)
(see Note 6).
2.3 Analysis 1. Reconstitution solvent mixture: water/acetonitrile (ACN)/

isopropanol (IPA) (1:1:3).
2.3.1 RP-Lipid Profiling 2. Multi-sample vortex mixer.
3. Centrifuge (Eppendorf).
4. Vials: deactivated clear glass 12 × 32 mm screw neck total
recovery vial, large glass 20 mL vials.
5. Mobile phase A: ACN/water (60:40), 10 mM ammonium
formate, and 0.1% formic acid. For 2 L of mobile phase A, add
800 mL of water followed by the addition of 1.26 g of ammo-
nium formate. Swirl and sonicate until the salt has been fully
dissolved. Proceed with the addition of 2 mL of formic acid.
Swirl briefly. Add 1200 mL of acetonitrile and degas for 5 min
by sonication.
6. Mobile phase B: IPA/ACN (90:10) and 0.1% formic acid. For
2 L of mobile phase B, add 1800 mL of isopropanol, 200 mL
of acetonitrile, and 2 mL of formic acid. Degas by brief
sonication.
7. UPLC separation is performed using an Acquity UPLC System
(Waters Corp., USA).
8. Column: Acquity UPLC CSH C18 2.1 × 100 mm, 1.7 μm,
column (Waters Corp, USA).
9. Weak wash solvent mixture: ACN/water (60:40).
10. Strong wash solvent mixture: IPA/ACN (90:10).
11. Seal wash: add 125 mL of IPA to 2.5 L water.

12. MS: Xevo G2 QTof (Waters MS Technologies, UK) with an
electrospray ionization (ESI) source.
2.3.2 HILIC-MS 1. Reconstitution solvent mixture: acetonitrile (ACN)/water

of Aqueous Extract (95:5).
2. Multi-sample vortex mixer.
3. Centrifuge (e.g., Eppendorf).
4. Sonicator.
5. Vials: LCGC certified clear glass 12 × 32 mm screw neck vial,
150 μL inserts (deactivated).
6. Mobile phase A: acetonitrile (ACN)/water (95:5), 10 mM
ammonium acetate, and 0.1% formic acid. For 2 L of mobile
phase A, add 100mL of water, followed by 1.54 g of ammo-
nium acetate. Sonicate until the salt has been fully dissolved.
Add 2 mL of formic acid and briefly swirl the bottle. Proceed
by gradually adding 100 mL of acetonitrile and sonicating for
10 min. Then, add another 200 mL of acetonitrile and 5 min
sonication. Continue with 400 mL acetonitrile/5 min sonica-
tion, and lastly add 1200 mL of acetonitrile and sonicate for a
final 5 min.
7. Mobile phase B: ACN/water (50:50), 10 mM ammonium
acetate, and 0.1% formic acid. For 2 L of mobile phase B, in
1 L of water, add 1.54 g of ammonium acetate. Sonicate until
the salt has been fully dissolved. Add 2 mL of formic acid and
swirl briefly. Proceed by gradually adding 1 L of acetonitrile
and sonicate for 5 min.
8. UPLC separation is performed using an Acquity UPLC System
(Waters Corp., USA).
9. Column: Acquity UPLC BEH HILIC 2.1 × 100 mm, 1.8 μm,
column (Waters Corp, USA).
10. Weak wash solvent mixture: ACN/water (95:5).
11. Strong wash solvent mixture: ACN/water (50:50).
12. Seal wash: add 125 mL of IPA to 2.5 L water.
13. MS: Xevo G2 QTof (Waters MS Technologies, UK) with an
electrospray ionization (ESI) source.
2.3.3 NMR Spectroscopy 1. Reconstitution buffer: for 1 L of buffer, mix 500 mL of high
purity water with 21.73 g Na2HPO4 (anhydrous), 2.63 g
NaH2PO4 (anhydrous), and 0.1 g sodium 3-trimethylsilyl-1-
[2,2,3,3,-d4] propionate (TSP) (0.01% m/v). Mix well and
then add 200 mL of D2O. Mix until salts are completely dis-
solved. Sonicate the solution or use magnetic spinner if neces-
sary. Adjust to pH 7.4 with NaOH crystals or HCl (high
concentration). Transfer the mix to a 1 L volumetric flask and

fill up to 1 L with high purity water. Readjust the pH to 7.4.
Consider the addition of a bacteriostatic agent (see Note 7).
2. Tubes: NMR tubes with an outer diameter of 5 mm.
3. NMR instrument: a 600 MHz NMR spectrometer (Bruker
BioSpin).
2.4 Data Centroiding is performed using MassLynx (Waters). The

Deconvolution DataBridge software (MassLynx, Waters) is used for conversion of
and Processing files from MassLynx (UPLC-MS analysis) to NetCDF format. The
2.4.1 UPLC-MS XCMS package (version 1.46.0) using R (version 3.2.2) is
employed for data deconvolution.
2.4.2 NMR Spectroscopy MATLAB (version 2014a) programming language is utilized for
importing and processing of NMR data, using in-house developed
scripts.
2.5 Statistical For multivariate statistical analysis, such as principle component

Analysis analysis (PCA) and orthogonal projection to latent structures
(OPLS), SIMCA version 14.0.0 is used. For t-test and fold change
calculations, MS Excel is used. Spearman correlations are calcu-
lated using R version 3.2.2. Correlation networks are constructed
using Cytoscape version 3.4.0 (Cytoscape Consortium) [13].
2.6 Pathway For pathway mapping the KEGG database [14] and Ingenuity
Mapping Pathway Analysis software (QIAGEN) is used.
3 Methods
An overview of the procedure is summarized in Fig. 1.

According to the origins of the tissue used, severe biohazards
might exist. Take every required precaution.
3.1 Randomization The MS Excel random number generating function, RAND, is

used for sample order randomization. After generating the random
number using the function, sample labels were sorted according to
the generated random number.
3.2 Tissue Handling 1. Tissues samples must be stored at −80 °C.

and Dissection 2. Maintain tissue samples in dry ice when outside of the freezer
(during transporting, handling etc).
3. Use single-use, disposable scalpels and forceps (see Notes 2 and
8).
4. Cut each tissue on a new/clean large container (large weighing
boat). It is recommended to place the container with the tissue
on top of dry ice, in order to maintain the tissue sample at a low

temperature. Transfer and weigh the tissue sample in a new clean
weighing boat (use the empty weighting boat to tare the balance,
prior to loading the tissue sample). One aliquot should corre-
spond to 20–25 mg of wet tissue mass. One aliquot should be
considered the optimum material for most MS applications and
two aliquots for NMR spectroscopy (although as little as half the
recommended mass can be used). The total amount of the
required wet tissue should correspond to (be multiplied by) the
number of aliquots required for the methods/platforms employed
(20–25 mg x required aliquots). Samples should not differ by
more than 5% of their total weight. In order to follow the proto-
col described here, 60–75 mg ± 5% of wet tissue is required (3
aliquots). Other options are available for difficult to cut tissue or
when cutting should be avoided altogether (see Note 9).
5. Transfer the dissected/weighed tissue into the bead beating
tube (preloaded with beads).
6. Include blank samples to test for contamination induced by the
extraction procedure (extraction blanks). Set up bead beating
tubes without tissue sample. The number of recommended
extraction blanks used should be ≥7 and ≥5% of the total num-
ber of samples.
3.3 Tissue Extraction 1. Randomize samples.

2. Add 1500 μL of the organic extraction solvent to the bead beat-
3.3.1 Organic Extraction ing tubes loaded with the weighed tissue and/or beads (sam-
(Round 1) ples/blanks) (see Notes 9–12).
3. Load the bead beater and initiate the vibration sequence at
6500 Hz for 40 s. Repeat for two to four cycles. Each cycle
should be separated by placing the samples in dry ice for 5 min
(see Note 13).
4. Centrifuge at 20,000 rcf for 30 min at 4 °C.
5. Decant the number of aliquots corresponding to the amount of
tissue used into Eppendorf tubes. For 75 mg of wet tissue, the
organic extraction solvent should be divided into three aliquots
of 400 μL.
6. Produce extraction pooled samples by combining a small
amount of the remaining tissue extracts by decanting an equal
volume from each bead beating tube in a large container.
Fig. 1 An overview of the tissue metabolomics/metabonomics multiplatform procedure. The procedure is com-
prised of seven basic steps: (1) tissue dissection, (2) metabolite extraction from the tissue sample (two con-
secutive extractions: organic, round 1, followed by aqueous, round 2), (3) data acquisition (NMR, HILIC-, and
RP-UPLC-MS), (4) data deconvolution and processing, (5) statistical analysis, (6) metabolite structure assign-
ment, and (7) data integration and biological interpretation. MeOH, Methanol; MTBE, methyl tert-butyl ether
Proceed by aliquoting the same volume as with individual sam-

ples (step 5) into Eppendorf tubes. If 75 mg of tissue is used,
then the pooled extracts should be aliquoted in 400 μL (see
Note 14).
7. Proceed to complete solvent evaporation of the aliquots using a
vacuum concentrator working at 30 °C, for 2 h (see Notes
15–17).
8. Ensure the bulk of the organic extraction solvent has been com-
pletely removed from the bead beating tube before performing
the aqueous extraction.
3.3.2 Aqueous Extraction 1. Randomize samples (optional; see Note 18).

(Round 2) 2. Add 1500 μL of the aqueous extraction solvent to the bead
beating tubes loaded with weighed tissue and/or beads (sam-
ples/blanks) (see Note 10–12).
3. Load in bead beater vibrating at 6500 Hz for 40 s, for two to
four cycles. Cycles are separated by freezing of the samples in
dry ice until the extraction solvent mixture freezes (~5 min) (see
Note 19).
4. Centrifuge at 20,000 rcf for 30 min at 4 °C.
5. Decant the number of aliquots corresponding to the amount of
tissue used into Eppendorf tubes. For 75 mg of tissue, the
aqueous extraction solvent should be divided in three aliquots
of 400 μL. Two aliquots should be considered the optimal
amount for NMR analysis of aqueous extracts (see Notes 12
and 20).
6. Produce extraction pooled samples by combining a small
amount of the remaining solvent (of the homogenized tissue)
by decanting an equal volume from each bead beating tube in a
large container. Proceed by aliquoting the same volume as with
individual samples into Eppendorf tubes (as with step 5). For
75 mg of tissue, 400 μL should be aliquoted (see Note 14).
7. Proceed to complete solvent evaporation of the aliquots using a
vacuum concentrator working at 30 °C, for 2.5 h (see Notes 15
and 16) .
3.4 UPLC-MS 1. Add 200 μL of the lipid profiling reconstitution solvent mixture
Analysis and Data to the Eppendorf tubes of the extracts and blanks (see Note
Treatment 21).
3.4.1 Reconstitution
3. Centrifuge for 20 min at 20,000 rcf at 4 °C.
Lipid Profiling
4. Decant the supernatant into an LC-MS total recovery vial.
5. Combine 50 μL of the reconstitute to produce the QC pooled
sample (see Notes 14 and 22).
6. Use the QC or extraction pooled samples to perform a dilution

series by diluting using the reconstitution solvent mixture (see
Note 23) .
HILIC 1. Add 120 μL of the HILIC reconstitution solvent mixture in the

Eppendorf tubes of the extracts and blanks (see Note 21).
3. Sonicate for 5 min.
5. Centrifuge for 20 min at 20,000 rcf at 4 °C.
6. Combine approximately 20 μL of the reconstitute to produce
the QC pooled sample (see Notes 14 and 22).
7. Decant the remainder of the supernatant into an LCGC vial
loaded with deactivated inserts.
8. Use the QC or extraction pooled samples to perform a dilution
series by diluting with the reconstitution solvent mixture (see
Note 23) .
3.4.2 UPLC-MS RP-lipid profiling
UPLC Conditions The suggested binary solvent manager parameters are weak
wash volume, 1000 μL; strong wash volume, 1000 μL; seal wash
frequency, 1 min; column temperature, 55.0 °C; autosampler tem-
perature, 8.0 °C; partial loop with needle overfill option; needle
overfill flush, 4 μL; and injection volume 3 μL for positive polarity
and 10 μL for negative polarity mode (see Note 21).
The elution gradient is set as follows: 60–57% A(0.0–2.0 min),
57–50% A (2.0–2.1 min; curve 1), 50–46% A(2.1–12.0 min),
46–30% A (12.0–12.1 min; curve 1), 30–1%A (12.1–18 min),
1–60% A (18.0–18.1 min), and 60% A (18.1–20.0 min). Flow rate
is maintained at 0.4 mL/min.
Characteristic chromatograms with detection in positive and
negative polarity mode are illustrated in Fig. 2.
HILIC
The suggested binary solvent manager parameters are weak
wash volume, 600 μL; strong wash volume, 200 μL; seal wash fre-
quency, 1 min; column temperature, 40.0 °C; autosampler tem-
perature, 4.0 °C; partial loop with needle overfill option; needle
overfill flush, 4 μL; and injection volume 5 μL for both polarity
modes (see Note 21).
The elution gradient and flow rate are set as follows:
99% A (0.0–2.0 min; 0.4 mL/min)
99–45% A (2.0–8.0 min; 0.4 mL/min)
45–1% A (8.0–9.0 min; 0.4 mL/min)
1% A (9.0–9.1 min; 0.4–0.6 mL/min)
1% A (9.1–11.0 min; 0.6 mL/min)

1–99% A (11.0–11.1 min; 0.6 mL/min)
99% A (11.1–17.0 min; 0.6 mL/min)
99% A (17.0–17.1 min; 0.6–0.4 mL/min)
99% A (17.1–21.0 min;0.4 mL/min)
Characteristic chromatograms with detection in positive and
negative polarity mode are illustrated in Fig. 2.
MS Conditions For RP-lipid profiling, the mass range is set between 50 to

2000 m/z, while for HILIC between 50 to 1200 m/z. The MSE
mode (continuum mode) is used for MS acquisition. Three func-
tions (parallel acquisition channels) are employed for acquisition:
function 1, low collision energy; function 2, high collision energy;
and function 3, lock mass acquisition channel. For both low and
high collision energy functions, survey scan time is set to 0.2 s. For
high collision energy acquisition, collision energy is ramped for
low masses 20–40 V and linearly increased up to a 30–50 V ramp,
for high masses. Leucine enkephalin (2 ng/μL, 50% ACN, 0.1%
FA) is used for lock mass correction. Lock mass data were collected
every 30 s for 0.2 s. Other MS parameters were set as follows: cone
voltage 30 V, capillary voltage 2 KV, source temperature 120 °C,
desolvation temperature 550 °C, and desolvation gas 900 L/h.
UPLC-MS Sample Pre-run steps:

Analysis Setup
1. Perform an initial conditioning of the system by running the
gradient either without sample injection or with an injection
of a pooled sample or other type of complex mixture (e.g.,
animal tissue extracts or authentic standards mixture). Inject
10–20 times (see Notes 24 and 25).
2. Run a test sample. Use this to assess adduct formation, mass
accuracy, chromatographic peak width/resolution, and
retention time shifting. A standards mix or a biological sample
of known metabolite composition (as a more cost-efficient
approach), can be used. For tissue types of unknown consis-
tency, a range of commonly detected compounds, which can
be used for this task, can be found in Table 1. A sample with
pre-determined compounds of known concentration/inten-
sity can also be used to assess that the instrument/method is
functioning at a satisfactory sensitivity level (see Note 26).
3. Second round of conditioning is performed using QC pooled
sample. Inject three times. Assess instrument stability and
intensity. If signal intensity is low, more sample volume can be
injected. See next step.
4. Inject 1:2 dilution and assess intensity response to dilution.
Reduce injection volume so that signal intensity is decreased
Fig. 2 Characteristic chromatograms/spectra from RP-UPLC-MS analysis of organic extracts (lipid profiling) in
(a) positive and (b) negative modes, HILIC-UPLC-MS analysis of aqueous extracts in (c) positive and (d) nega-
tive modes, and NMR spectra of aqueous extracts acquired using the standard sequence (gray line; prior to
baseline correction) and CPMG sequence (black line)
along with the corresponding dilution, and repeat this step.

Further dilute the samples if required (see Note 27).
5. Inject a pooled sample and selected samples from each group
or pooled samples of each group, using a data-dependent
Table 1
250
A range of compounds that are regularly detected in tissue samples, using the described UPLC-MS methods. Signals from these compounds can be used
to assess adduct formation, mass accuracy, chromatographic peak width/resolution, and retention time shifting
Mol mass – positive mode Mol mass – negative

(most frequently occurring mode (most frequently Peak width at half
Mol formula Ret timea (min) adduct) occurring adduct) max (min)
HILIC L-carnitine C7H15NO3 7.32 162.1130 [M+H]+ – 0.043
+
Adenosine C10H13N5O4 2.09 268.1046 [M+H] 266.0895 [M-H]−, 0.053
302.0656 [M+Cl]−
Panagiotis A. Vorkas et al.
Creatinine C4H7N3O 3.14 114.0667 [M+H]+ 112.0511 [M–H]− 0.062

Taurine C2H7NO3S 5.80 126.0225 [M+H]+ 124.0068 [M–H]− 0.037
l-leucineb C6H13NO2 5.83 176.0663 [M+2Na–H]+ 130.0868 [M–H]− 0.052
b + −
l-isoleucine C6H13NO3 5.96 176.0663 [M+2Na–H] 130.0868 [M–H] 0.035
+ −
Glycerophosphocholine C8H20NO6P 8.32 258.1106 [M+H] 242.0793 [M–CH3] 0.035
Lipid Stearic acid C18H36O2 3.90 – 283.2637 [M–H]− 0.045
profiling
SM(d18:2/16:0) C39H77N2O6P 5.20 701.5598 [M+H]+ 745.5496 [M+FA–H]− 0.065
TG(16:0/18:1/18:1) C55H102O6 16.29 876.8020[M+NH4]+ – 0.061
+
CE(18:2) C45H76O2 16.16 666.6189 [M+NH4] – 0.052
LysoPC(16:0)c C24H50NO7P 1.40 496.3403 [M+H]+ 540.3301 [M+FA–H]− 0.048
LysoPC(0:0/16:0)c C24H50NO7P 1.31 496.3403 [M+H]+ 540.3301[M+FA–H]− 0.048
−
PI(18:0/20:4) C47H83O13P 6.56 – 885.5493 [M–H] 0.078
CE Cholesteryl ester, FA formic acid, SM sphingomyelin, lysoPC lysophosphatidylcholine, PI phosphatidylinositol, TG triacylglycerol

a
RT may slightly vary depending on instrument setup
b
l-leucine and l-isoleucine can be used for ensuring adequate chromatographic resolution in the HILIC method. These two peaks should be fully resolved
c
LysoPC(16:0) and lysoPC(0:0/16:0) can be used for ensuring adequate chromatographic resolution in the lipid profiling method. These two peaks should be resolved at least at 50%
acquisition (DDA) MS method. Increase the injection vol-

ume, if necessary, in order to obtain tandem-MS spectra with
higher intensity.
6. Inject a pooled sample and selected samples from each group
or pooled samples of each group, using a modified (increase
survey scan time to 0.5 s) MSE method. Increase the injection
volume, if required, to obtain tandem-MS spectra with higher
intensity.
7. Third round of conditioning using pooled QC sample. Inject
five times. Assess instrument stability by comparing repeated
injections from this step.
8. Inject reconstitution blank (reconstitution solvent mixture)
and assess carry-over. Inject three times.
9. Inject extraction blanks (×1 each) followed by dilution series
from low to high concentration (at least ×3 each).
10. Final round of conditioning using pooled QC samples. Inject
three times.
Within run:
1. Randomize samples.
2. Include a QC pooled sample injection at least every ten (or less)
samples (see Note 28).
Post-run:
1. Inject the pooled sample dilution series from high to low (at
least four times for each dilution of the series) (see Note 29).
2. Inject extraction blanks (inject each sample once).
3. Inject pooled QC sample three times.
4. Inject the reconstitution blank three times (reconstitution sol-
vent mixture) and assess carry-over.
3.4.3 Data Deconvolution 1. Centroid the UPLC-MS files using MassLynx.

2. Convert centroided files to the NetCDF file format using
DataBridge, and use only the first function for further analysis.
Alternatively, use MSConvert (ProteoWizard 3.0.11567) to
convert to mzXML file format (no further modifications to the
succeeding workflow are required).
3. Use the XCMS software to apply:
(a) Chromatographic peak picking (centWave algorithm).
(b) Retention time (RT) correction (obiwarp algorithm).
(c) Peak grouping (density algorithm).
(d) Fill peaks that resulted in zero intensity, with the intensity
of the regional background (fillPeaks algorithm).
(e) Export data matrix.
3.4.4 Data Normalization Normalize the data using total area or median fold change normal-
ization (see Note 30). There are cases where normalization might
not be necessary.
3.4.5 Statistical Analysis 1. Import data in SIMCA along with demographic data.
and Quality Control 2. Transform and scale the data. For UPLC-MS data a logarithmic
transformation followed by scaling is usually recommended.
The exploration of various transformation and scaling options
according to methodology, instrumentation, and acquired data
should be considered.
3. Use OPLS-DA to remove features attributed to the procedure
or contamination, by comparing samples against blanks, as pre-
viously described [7]. Typically, a cutoff of P(corr) > 0.5 should
be used in order to exclude the features attributed to
contaminants.
4. Use PCA to evaluate QC samples repeatability. Remove con-
tamination of the QC samples, if present, using PCA or
OPLS-DA (Fig. 3). If the latter method is used, a cutoff of
P(corr) > 0.8 is typically selected. Remove only features that are
exclusively present in the QC samples (see Note 31).
5. Use PCA for QC dilution feature assessment. Remove features
that do not respond to dilutions as previously described [7]. It
is recommended to only remove features that elute with the
solvent front, or correspond to lipid signals in areas of high
coelution of lipids of the same class in HILIC (these areas can
be found in previously described in-house databases of anno-
tated lipid species of the HILIC method) [7]. The QC dilution
series can also be retrospectively employed to further assess the
quality of the discovered putative markers of the studied disease
or intervention.
6. Proceed to further statistical analyses and group comparisons.
3.4.6 Metabolite Perform metabolite identification as follows:

Structure Assignment
1. Perform accurate mass database search to obtain possible hits.
2. Match collision-induced dissociation fragments or spectrum to
spectra registered in databases.
3. Match authentic standard RT and mass-to-charge ratio to the
unknown compound, acquired under identical experimental
conditions (see Note 32).
4. Match an authentic standard MS/MS spectrum to the unknown
compound, acquired under identical experimental conditions
(see Note 32).
5. Spike sample(s) with authentic standards if necessary.
Fig. 3 The quality control (QC) samples may be containing additional contamination, as compared to samples,
due to differences between sample/QC preparation. A simple procedure can be followed to remove the fea-
tures associated with this contamination using multivariate statistics. (a) The PCA scores plot demonstrates a
contamination in the QC samples, which is not present in the individual samples. (b) The loadings plot can
assist in pinpointing the contaminating features. (c) Feature intensity per sample can help identify features that
are only present in QC samples and constitute a contaminant. (d) OPLS-DA scores plot. OPLS-DA can be
employed as an alternative method for identifying and removing contaminants. (e) The S-plot can be used to
visualize the features for exclusion. (f) The PCA scores plot after features corresponding to contaminants have
been removed, showing the QC samples clustering in the center of the individual samples
Additional steps that could aid metabolite structure

assignment:
1. Use the isotopic pattern for deciphering elemental composi-
tion. This can be done using manual approaches or software
(Elemental Composition version 4.0, MassLynx, Waters).
2. Use the RT to assess whether the possible candidate compound

structures are plausible (see Note 33).
3. Use in silico fragment generation (MassFragment, Waters), if
corresponding tandem-MS spectra are not available in
databases.
4. Use pairwise correlations (see Subheading 3.6.2), in combina-
tion with RT proximity. Correlations can prove useful for deter-
mining structurally related features (isotopes, adducts,
fragments) but also metabolites with close biological/(bio)
chemical proximity.
3.5 NMR Analysis Two aliquots (corresponding to 50 mg of tissue sample) are rec-
and Data Treatment ommended for the NMR acquisition:
3.5.1 Reconstitution
1. Add 700 μL of sodium phosphate buffer into the dried tissue
for Aqueous Extracts
extract.
2. Vortex for 1 min followed by sonication for 5 min.
3. Centrifuge for 20 min at 20,000 × g at 4 °C.
4. If two aliquots are used, transfer the supernatant from the first
tube to the second, and repeat steps 2 and 3.
5. Transfer 600 μL of the supernatant into an NMR tube.
6. Perform the same procedure for extraction blanks.
7. Load a small number of tubes with the buffer only, to test for
contamination.
3.5.2 NMR Data Standard one-dimensional NMR sequence [recycle delay

Acquisition (RD)−90°−t1−90°−tm−90°−acquire free induction decay (FID)]
is used. If standard 1D spectra of the samples contain broad signals
which raise up the spectral baseline and interfere with the visualiza-
tion of the signals derived from the small molecules, the Carr-
Purcell-Meiboom-Gill (CPMG) sequence [RD−90°−(τ−180°−τ)
n−FID] can also be employed for acquisition (Fig. 2). A total of
512 scans are accumulated into 64 k data points, with a spectral
width of 20 ppm. Detailed NMR acquisition parameters can be
found in Beckonert et al. [15] .
3.5.3 Data Deconvolution The following steps are performed in MATLAB using in-house
and Processing developed scripts:
1. Perform phasing, baseline correction, and calibration using the
TSP signal.
2. Remove the TSP and water signals from the spectra.
3. Normalize if necessary. Normalize the data using total area or
median fold change normalization (see Note 30).
4. Align spectra if necessary.
3.5.4 Statistical Analysis

and Quality Control 1. Import data in SIMCA along with demographic/external data.
2. Scale the data. Scaling to unit variance or Pareto can be used.
Logarithmic transformation prior to scaling can also be
employed.
3. Use OPLS-DA to remove features attributed to procedure/sol-
vent contamination, by comparing samples against blanks, as
previously described [7]. Typically, a cutoff of P(corr) > 0.5
should be used in order to exclude features attributed to
contaminants.
4. Use PCA to evaluate QC samples repeatability. Remove con-
tamination of the QC samples if present using PCA or
OPLS-DA. If the latter method is used, a cutoff of P(corr) > 0.8
is typically selected. Remove only features that are exclusively
present in the QC samples.
5. Proceed to further statistical analyses and group comparisons.
3.5.5 Metabolite Metabolite identification is performed as follows:

Structure Assignment
1. Perform statistical total correlation spectroscopy (STOCSY) on
the signals of interest to search for highly correlated signals,
which could be deriving from the same molecule.
2. Check 1D NMR spectra and J-resolved spectra for signal
multiplicity.
3. Match the signal chemical shifts and multiplicities to the pub-
lished literature [16–18], databases such as AMIX (Bruker) and
Human Metabolome Database (HMDB), and/or software
such as the Chenomx Profiler (Chenomx Inc.).
4. Use 2D NMR spectra, such as 1H-1H COSY (correlation spec-
troscopy), 1H-1H TOCSY (total correlation spectroscopy), 1H-
13
C HSQC (heteronuclear single-quantum correlation) and
1
H-13C HMBC (heteronuclear multiple bond correlation), to
identify the connectivity of the signals.
5. Use hyphenated techniques such as LC-NMR, LC-SPE-NMR,
or LC-SPE-NMR/MS to reduce sample complexity, increase
the concentrations of the molecules of interest and gain both
MS and NMR structural data for metabolite identification.
6. Spike in the samples with the authentic standards of putatively
assigned molecules if required.
3.6 Biological For the processes of correlation network construction and pathway
Interpretation mapping, a table of metabolites that demonstrated statistically sig-
nificant alterations should be produced, along with statistical
power, fold change, and values of each metabolite per sample. It is
also recommended to assign to each metabolite its corresponding
KEGG ID to facilitate importing in pathway mapping software.
3.6.1 Data Integration Several approaches can be employed for data integration depend-
ing on the biological question and techniques/methods used:
1. Combine two or more data matrices by using multiblock multi-
variate methods, such as O2PLS and OnPLS [19], where
extraction of predictive components can be achieved.
2. Perform correlation analysis (see Subheading 3.6.2) to obtain
associations between entities from the same or different experi-
ments/datasets. Collectively, these associations could lead to
inferences assisting in the biological interpretation of the dys-
regulation or intervention studied.
3. Use appropriate informatics tools (described in Subheading
3.6.3) to perform metabolite enrichment of altered metabolites
detected from the same or different experiments. Additionally,
metabolite/gene enrichment can be combined if such data are
available.
3.6.2 Correlation Construct correlation networks after calculating the Spearman cor-
Analysis relation coefficients. Correlation coefficients along with their cor-
responding entity pairs, which could be paired combinations of
metabolites, proteins, genes, demographics etc., can then be trans-
ferred to the Cytoscape software, where the correlation network
can be assembled. All presented correlations in the network should
have a p < 0.05 (after multiple testing correction), while the cutoff
of the correlation coefficient should be set based on the complexity
of the illustrated network. The correlation matrix can be demon-
strated either as a matrix table or as a heat map.
3.6.3 Pathway Mapping Import the list of metabolites/KEGG IDs in the KEGG mapper
and/or Ingenuity Pathways Analysis software to map the pathways
corresponding to the assigned discriminant metabolites. When
available, include data from additional biological levels, such as
protein and mRNA, in order to combine metabolite and gene
enrichment. This can result in higher confidence and statistical
power of results.
4 Notes
1. Rinse thrice with LC-MS-grade water followed by rins-

ing three times with an organic solvent (an organic solvent
used for rinsing should be the same as the organic solvent used
in the solution to be prepared).
2. Biological hazard: scalpels are extremely sharp and can easily
cut through gloves and skin. Dispose immediately after use,
and avoid any effort to clean and reuse them.
3. The amount of beads used should remain consistent between
tubes (±5% of bead mass). A time-efficient option is to use a
constant volume of beads (a utensil can be used in this case).

Aim for one or two layers of beads above the curved bottom
of the tube.
4. Use three steel beads (2.8 mm) for physically harder tissue. In
this case, use reinforced tubes. Avoid using solvent volumes
that are less than half the volume of the container to prevent
breakage of the tube during bead beating.
5. Eppendorf tubes should have the capability of ≥25,000 rcf
and should be of high quality to avoid contamination due to
the leaching of plasticisers and other impurities.
6. Lyophilisation and nitrogen evaporation can also be used for
drying down the extraction solvent mixture.
7. If a specific tissue type is expected to have a heavy bacterial
load (e.g., colon), then 0.04% NaN3 should be added in the
buffer as a bacteriostatic. The high toxicity of NaN3 implies
that it should be used only when it is absolutely essential.
8. Single-use scalpels and forceps are required to minimize the
risk of contamination.
9. Where tissue dissection can be difficult to perform, the analyst
can opt for an approach where the total weight of the sample
is determined and the extraction solvent is normalized propor-
tionally to the total tissue mass [7].
10. Extraction solvent volume must be accordingly adjusted when
large tissue samples are expected for bead beating, to accom-
modate the increase in total volume.
11. Avoid using proportions of solvent volume-to-tissue weight
lower than 4:1.
12. If aliquots for specific targeted methods are to be included in
the workflow, it would be optimal to add appropriate internal
standards prior to initiating extraction (if method to be used is
known beforehand).
13. Check if the tissue has been fully homogenized. Look for
completely pulped tissue with no presence of any structure at
the bottom of the bead beating tube. If full homogenization
has been achieved, avoid additional cycles. Never exceed four
cycles. If full homogenization has not been achieved after 4
cycles, then opt for the steel beads option.
14. Do not include blank samples in the preparation of the pooled
samples.
15. Ensure complete solvent evaporation. Residual extraction sol-
vents can alter dissolvation efficiency and analyte concentra-
tion. Residual methanol can cause a strong contamination
signal in NMR spectroscopy.
16. Evaporation time could vary depending on the extraction sol-

vent used, solvent volume, tissue type, and condition of the
vacuum concentrator.
17. There are cases where the presence of specific lipids in high
concentrations in the tissue sample would make the organic
extract appear more liquid, even after full evaporation of the
organic solvents.
18. Randomization for the second round of extraction (aqueous
extraction) is optional and may not be possible if samples are
forwarded for aqueous extraction before organic extraction
has been completed for every sample.
19. If tissue has been fully homogenized from the organic extrac-
tion step, apply only two beating cycles during the aqueous
extraction. Never exceed four cycles.
20. NMR spectroscopy is a non-destructive technique. Samples
after being run by NMR can be evaporated and stored for
further analysis.
21. Optimal reconstitution volumes may vary according to tissue
type and instrumentation. Adjust the reconstitution volume to
inject the minimum volume possible. Use the extraction
pooled samples for testing and optimizing the system.
22. Decant and combine enough volume to cover the required
QC injections.
23. Dilutions of 1:2, 1:3, 1:4, 1:6, and 1:8 of a pooled sample in
the reconstitution solvent mixture are suggested. If the extrac-
tion pooled sample is used for the dilutions, then a 2:1 con-
centration option is available and can be prepared by
reconstituting with half the volume.
24. Extracts from an easily accessible animal tissue can be used.
25. The number of injections here may depend on the state of the
column. New columns may require more injections to
condition.
26. Use annotated metabolites from the literature [7]. Use a range
of compounds with different masses and physicochemical
properties. A range of suggested compounds that are regularly
detected in tissue extracts and can be employed to assess
adduct formation, mass accuracy, chromatographic peak
width/resolution, and retention time shifting can be found in
Table 1.
27. If after reducing the injection volume, the ion signal is main-
tained at detector saturation levels, consider further dilution
of the samples. Consider the overall profile, and avoid making
decisions based on the (saturated) signals of a small set of
ions. It is acceptable to maintain a small number of peaks at
saturation intensities, if this allows for further signals to rise

above noise level and become detectable.
28. If a small number of samples (<60) is being run, include at
least seven QCs through the run, as this is the minimum
required number of repeated injections for robust statistics
such as calculating reproducibility using CV%.
29. Inject repeated injections from high to low concentrations to
avoid losing column conditioning. The first injection of each
dilution can be discarded if necessary. The described methods
have demonstrated minimal carry-over [7].
30. Systematic instrument signal reduction due to source contam-
ination would usually require normalization. Total area nor-
malization would typically resolve this issue. Occasionally,
differences in tissue composition (normal and diseased) would
require normalization procedures such as median fold change
[20].
31. The QC sample contamination may be induced due to minor
differences in sample handling, such as multiple injections
from the same vial or differences employed to accommodate
for the larger volume of the QC sample.
32. The same experimental conditions demand running sample
and authentic standard one after the other in the same analy-
sis/run. Changes in mass due to differential adduct formation
as compared to prior analyses and minor changes in retention
time may occur.
33. The log P value (P: partition coefficient) can be used as a mea-
sure of lipophilicity and can provide a significant aid in the task
of comparing between the level of partitioning of compounds
in the chromatographic system and their subsequent differ-
ences in RT.
Acknowledgments
This research was supported by the Royal Society of Chemistry

and National Institute for Health Research (NIHR) Biomedical
Research Centre (BRC) based at Imperial College Healthcare
NHS Trust and Imperial College London. The views expressed are
those of the authors and not necessarily those of the NHS, the
NIHR, or the Department of Health. MRAU is funded by the
Imperial College President’s PhD Scholarship and the Stratified
Medicine Graduate Training Programme in Systems Medicine and
Spectroscopic Profiling (STRATiGRAD).
References
1. Lamour SD, Veselkov KA, Posma JM et al rotic plaques relate to stroke risk: an explor-
(2015) Metabolic, immune, and gut microbial atory study. Eur J Vasc Endovasc 52:5–10.
signals mount a systems response to Leishmania https://doi.org/10.1016/j.ejvs.2016.01.022
major infection. J Proteome Res 14:318–329. 12. Anwar MA, Adesina-Georgiadis KN, Spagou K
https://doi.org/10.1021/pr5008202 et al (2017) A comprehensive characterisation
2. Folch J, Lees M, Sloane Stanley GH (1957) A of the metabolic profile of varicose veins; impli-
simple method for the isolation and purifica- cations in elaborating plausible cellular path-
tion of total lipides from animal tissues. J Biol ways for disease pathogenesis. Sci Rep 7:2989.
Chem 226:497–509 https://doi.org/10.1038/
3. Bligh EG, Dyer WJ (1959) A rapid method of s41598-017-02529-y
total lipid extraction and purification. Can 13. Shannon P, Markiel A, Ozier O et al (2003)
J Biochem Physiol 37:911–917 Cytoscape: a software environment for inte-
4. Geier FM, Want EJ, Leroi AM et al (2011) Cross- grated models of biomolecular interaction net-
platform comparison of Caenorhabditis elegans works. Genome Res 13:2498–2504. https://
tissue extraction strategies for comprehensive doi.org/10.1101/gr.1239303
metabolome coverage. Anal Chem 83:3730– 14. Kanehisa M, Goto S (2000) KEGG: Kyoto
3736. https://doi.org/10.1021/ac2001109 encyclopedia of genes and genomes. Nucleic
5. Masson P, Spagou K, Nicholson JK et al (2011) Acids Res 28:27–30
Technical and biological variation in UPLC- 15. Beckonert O, Keun HC, Ebbels TM et al
MS- based untargeted metabolic profiling of (2007) Metabolic profiling, metabolomic and
liver extracts: application in an experimental metabonomic procedures for NMR spectros-
toxicity study on galactosamine. Anal Chem copy of urine, plasma, serum and tissue extracts.
83:1116–1123. https://doi.org/10.1021/ Nat Protoc 2:2692–2703. https://doi.
ac103011b org/10.1038/nprot.2007.376.
6. Anwar MA, Vorkas P, Li JV et al (2015) nprot.2007.376 [pii]
Optimization of metabolite extraction of 16. Nicholson JK, Foxall PJ, Spraul M et al (1995)
human vein tissue for ultra performance liquid 750 MHz 1H and 1H-13C NMR spectros-
chromatography-mass spectrometry and copy of human blood plasma. Anal Chem
nuclear magnetic resonance-based untargeted 67:793–811
metabolic profiling. Analyst 140:7586–7597 17. Yap IK, Brown IJ, Chan Q et al (2010)
7. Vorkas PA, Isaac G, Anwar MA et al (2015) Metabolome-wide association study identifies
Untargeted UPLC-MS profiling pipeline to multiple biomarkers that discriminate north
expand tissue metabolome coverage: applica- and south Chinese populations at differing
tion to cardiovascular disease. Anal Chem risks of cardiovascular disease: INTERMAP
87:4184–4193. https://doi.org/10.1021/ study. J Proteome Res 9:6647–6654. https://
ac503775m doi.org/10.1021/pr100798r
8. Vorkas PA, Shalhoub J, Isaac G et al (2015) 18. Saric J, Wang Y, Li J et al (2008) Species varia-
Metabolic phenotyping of atherosclerotic tion in the fecal metabolome gives insight into
plaques reveals latent associations between free differential gastrointestinal function.
cholesterol and ceramide metabolism in ath- J Proteome Res 7:352–360. https://doi.
erogenesis. J Proteome Res 14:1389–1399. org/10.1021/pr070340k
https://doi.org/10.1021/pr5009898 19. Lofstedt T, Trygg J (2011) OnPLS-a novel
9. Ashrafian H, Li JV, Spagou K et al (2014) multiblock method for the modelling of pre-
Bariatric surgery modulates circulating and car- dictive and orthogonal variation. J Chemom
diac metabolites. J Proteome Res 13:570–580. 25:441–455. https://doi.org/10.1002/
https://doi.org/10.1021/pr400748f cem.1388
10. Anwar MA, Vorkas PA, Li J et al (2016) 20. Veselkov KA, Vingara LK, Masson P et al
Prolonged mechanical circumferential stretch (2011) Optimized preprocessing of ultra-
induces metabolic changes in rat inferior vena performance liquid chromatography/mass
cava. Eur J Vasc Endovasc 52:544–552. spectrometry urinary metabolic profiles for
https://doi.org/10.1016/j.ejvs.2016.07.002 improved information recovery. Anal Chem
11. Vorkas PA, Shalhoub J, Lewis MR et al (2016) 83:5864–5872. https://doi.org/10.1021/
Metabolic phenotypes of carotid atheroscle- ac201065j
Chapter 18
UHPLC-HRMS Analysis for Steroid Profiling in Serum

(Steroidomics)
Federico Ponzetto, Julien Boccard, Raul Nicoli, Tiia Kuuranne,
Martial Saugy, and Serge Rudaz
Abstract
The extraction and untargeted UHPLC-HRMS analysis of endogenous steroids in serum samples is
described in this protocol. The employed full-scan acquisition mode provides the adequate sensitivity to
highlight the main endogenous steroids present in blood, including mineralocorticoids, progestogens, and
androgens. Technical aspects for both chromatography and mass spectrometry are discussed in detail,
together with a proposition of setup for sample sequence and data analysis. Furthermore, general com-
ments are given to help the assessment of data quality and system performance.
Key words Steroids, Extended steroid profiles, Steroidomics, UHPLC-HRMS
1 Introduction
Steroids are structurally related hormones originating from choles-

terol synthesized in numerous organs, including the adrenal glands,
testis, ovaries, brain, placenta, and adipose tissue [1, 2], which
regulate many essential functions such as growth, metabolic rate,
sexual functions, and stress response. Because altered regulation of
steroid metabolism (due to genetic or environmental factors) is
often related to severe pathologies, the large-scale monitoring of
steroidogenesis provides crucial information for therapeutic pur-
pose [3]. Furthermore, the persistent misuse of anabolic andro-
genic steroids, which perturb testosterone metabolism leading to
better performance/recovery or health issues, raises the need of
such a monitoring also in anti-doping field [4–7].
Steroid analysis for the diagnosis of endocrine diseases was ini-
tially achieved by gas chromatography-based (GC) methods in the
1950s and since 1970 was then complemented by high-throughput
immunoassays (IA) [8]. However, as IA techniques often revealed
problems of cross-reactivity, particularly in the case of steroids, a
261
262 Federico Ponzetto et al.
standardization of steroid tests employing separation-based tech-

niques, such as chromatographic methods coupled to mass spec-
trometry, was recommended [9–11]. Coupling mass spectrometry
to either liquid chromatography (LC-MS) or gas chromatography
(GC-MS) is the current reference approach for measuring steroids;
in particular, the latter, even if limited by the need for a derivatiza-
tion step aiming at improving the volatility of the compounds prior
to analysis, constitutes the reference for anti-doping analyses [12].
More recently, untargeted profiling strategies combining the high
peak capacity provided by ultrahigh pressure liquid chromatogra-
phy (UHPLC) and the resolution offered by high-resolution mass
spectrometry (HRMS) are representing an appealing alternative
for simultaneously measuring hundreds of steroids in complex bio-
logical matrices. This steroidomic approach was first described by
Sjövall in 2004 and defined as the “characterization and quantifica-
tion of metabolic profiles of steroids” [13].
In this protocol an untargeted steroidomic analysis on serum
samples using a UHPLC-HRMS platform is presented. The overall
methodology has been optimized to provide high extraction recov-
eries for steroids in serum samples as well as good analytical perfor-
mance for both sensitivity and chromatographic separation, using
101 endogenous steroids as reference standards. The proposed
procedure could also be adapted for absolute quantification of ste-
roids of particular interest, but in general the development of a
separate assay is preferred for such a purpose.
2 Materials
Ultrapure water (18.2 MΩ, total organic carbon <5 ppb), organic
solvents (acetonitrile, ACN, and methanol, MeOH), and formic
acid (FA) of HPLC grade or higher purity should be used. The
same quality criteria must be applied to analytical standards to
obtain the highest available purity. Reference endogenous steroid
standards as well as deuterium-labeled internal standards could be
purchased on the market from various providers, such as Sigma-
Aldrich, Steraloids, and Cerilliant. Charcoal stripped human serum
(steroid depleted/negative serum) is purchased from Dunn
Labortechnik GmbH (Asbach, Germany).
2.1 Analytes Stock Endogenous steroids and labeled internal standards should be pur-
Solutions, Internal chased as calibrated stock solutions (1 mg/mL or 100 μg/mL in
Standard, and Quality MeOH or ACN) or as powder. For the latter, stock solutions are
Control Mixtures prepared by weighing the appropriate amount of powder and dis-
solving it in MeOH to obtain a final concentration of 1 mg/
mL. Stock solutions are finally diluted in MeOH to obtain
intermediate solutions at a concentration of 10 μg/mL. Both stock
Steroidomics 263
and intermediate solutions are stored at −80 °C and should be

thawed only when it is strictly mandatory to limit degradation.
Testosterone-d3, androsterone-d4,
17α-hydroxyprogesterone-d8, and cortisol-d4 stock solutions are
diluted in MeOH to obtain an Internal Standard Mix solution
(IS-mix) at a concentration of 3 ng/mL, 10 ng/mL, 5 ng/mL,
and 25 ng/mL, respectively. The IS-mix is stored at −20 °C and
should be thawed at least 30 min before its utilization during sam-
ple preparation.
Quality Control Mix (QC-mix) is prepared by mixing equal
aliquots of all serum samples that have to be analyzed in the
study (see also Chapter 2). It is important to prepare a sufficient
amount of QC-mix aliquots before starting the analyses and store
them at −20 °C. For each analytical batch, the required number of
QC-mix aliquots should be treated as all the other serum samples
and thawed overnight in refrigerator at 4 °C the day before
extraction.
2.2 Solvent 1. Mobile Phase A: In a 1 L solvent bottle, add 1 mL of FA to 1 L

Preparation for Liquid of ultrapure H2O and shake to ensure mixing. This solution is
Chromatography stored at room temperature and expires after 7 days (see Notes
1 and 3).
2. Mobile Phase B: Add 2.5 mL of FA to a 2.5 L bottle of ACN and
shake to ensure mixing. This solution is stored at room tem-
perature and expires after 4 months (see Notes 2 and 3).
3. Weak Wash Solution: In a 1 L solvent bottle, mix 980 mL of
ultrapure H2O with 20 mL of ACN and shake to ensure mixing.
This solution is stored at room temperature and expires after
7 days (see Note 3).
4. Strong Wash Solution: 1 L of pure ACN is used in an appropriate
bottle. This solvent is stored at room temperature and expires
after 4 months (see Note 3).
2.3 Instrumentation Sample preparation is performed using a PRESSURE+ 96 Positive

Pressure Manifold (Biotage, Uppsala, Sweden). Analyses are car-
ried out using a Waters Acquity UPLC system (Milford, MA, USA)
including a binary solvent manager, a sample manager equipped
with an external fixed loop of 20 μL, and a column manager. The
UPLC is coupled to a Q Exactive Plus mass spectrometer (Thermo
Fisher Scientific, Waltham, MA, USA). Data are acquired and pro-
cessed using Xcalibur PC software (version 3.1), and the data treat-
ment is performed with Progenesis QI software (Version 2.0,
64-bit, Nonlinear Dynamics, Newcastle upon Tyne, UK).
3 Methods
3.1 Serum Sample Supported liquid extraction (SLE) on ISOLUTE® SLE+ 400 μL
Extraction 96-well plates (Biotage, Uppsala, Sweden) is used to extract ste-
roid hormones from serum samples. The extraction procedure
consists of five different steps:
1. For each sample, an aliquot of 200 μL of serum is spiked with
20 μL of the IS-mix, diluted with 200 μL of water, and then
agitated for 15 min at 250 rpm.
2. Each well is then loaded with 400 μL of the previously diluted
sample, and a positive pressure of 3 psi is applied for 30 s to
facilitate sample loading and adsorption.
3. The elution is carried out, after a 5 min waiting period, by add-
ing 700 μL of dichloromethane to each well and applying a
pressure of 3 psi for 1 min.
4. Extracts are then collected in collection plates equipped with
1.5 mL glass inserts, evaporated to dryness for approximately
15 min at 40 °C under a gentle stream of nitrogen, and finally
reconstituted with 100 μL of a MeOH-H2O 50:50 (v/v) mix-
ture (reconstitution solvent).
5. After 15 min of gentle shaking (250 rpm), 10 μL of each extract
is injected into UHPLC-HRMS system for analyses.
3.2 UHPLC Liquid chromatography is performed with a Kinetex C18 column

(150 × 2.1 mm, 1.7 μm; Phenomenex, Torrance, CA, USA) set at
30 °C. The sample injection volume is 10 μL (partial loop mode),
and the flow rate is set at 300 μL/min. The mobile phase A is a
solution of 0.1% FA in H2O, and the mobile phase B is 0.1% FA in
ACN. The linear gradient starts from 5% of B and increases linearly
to 95% over 16.8 min, followed by 2 min of plateau at 95% B; the
column is then re-equilibrated for 7 min at initial conditions. The
total run time is 25.9 min (see Notes 4 and 5).
3.3 Mass HRMS analyses are performed in positive electrospray ionization

Spectrometry (ESI) in both full-scan (FS) and data-dependent MS/MS (ddMS2)
acquisition modes. Mass calibration (<3 ppm) is performed before
each analytical sequence using the Pierce® LTQ Velos ESI Positive
Ion Calibration standard mixture (Thermo Fisher Scientific) con-
taining n-butylamine, caffeine, MRFA (peptide of Met-Arg-Phe-
Ala acetate salt), and Ultramark 1621. The heated ESI source
(HESI II) is used with a probe heater temperature of 425 °C. The
sheath gas and auxiliary gas pressures are set to 50 and 15 arbitrary
units, respectively, and the sweep gas flow is set to 3 arbitrary units.
The ion spray voltage is set to 4.5 kV, the capillary temperature to
250 °C, and the S-Lens RF level to 55%. FS mass spectra are
Steroidomics 265
acquired in profile mode using a mass resolution of 70,000 (full

width at half maximum, FWHM) at m/z 200, with a maximum IT
fill time of 125 ms, and the automatic gain control (AGC target)
set to 3e6. The acquired mass range is from m/z 200 to 600. The
ddMS2 acquisition mode is performed using a mass resolution of
17,500 FWHM, with a maximum IT fill time of 64 ms and the
AGC target set to 5e4. The isolation window is set to 0.4 m/z, and
the number of different ions to be fragmented after each FS event
(loop count) is set to 5, using stepped normalized collision energy
values of 20, 40, and 60. For the data-dependent acquisition, a
minimum AGC target of 5e3 is set, resulting in an intensity thresh-
old of 7.8e4, above which the MS/MS analysis is triggered. For
improved selectivity, both apex trigger and dynamic exclusion fea-
tures are enabled (see Note 6). All detailed parameters of FS-ddMS2
experiment are presented in Fig. 1.
3.4 HRMS Prior to the injection of the extracted serum samples, it is recom-
Maintenance mended to carry out the cleaning of the ESI source followed by
and Calibration the calibration of the MS system. The MS cleaning procedure
involves the cone and the ion transfer tube that are removed from
the MS instrument after cooling the ESI source temperature to
room temperature. The instrument is then switched on standby
mode, and the two parts of the ion source are subjected to three
washing steps of 10 min sonication each in three different
solvents:
1. 10% FA in ultrapure H2O.
2. Ultrapure H2O.
3. HPLC grade MeOH.
Once the third step is completed, the cone and the ion transfer
tube are dried under a steam of argon and then assembled again on
the instrument (see Note 7). Then, the calibration MS tune file
(sheath gas flow rate set to 3 arbitrary units, both auxiliary and
sweep gas flow set to 0; ion spray voltage of 3.5 kV and capillary
temperature of 320 °C with the S-Lens RF level at 50%) is loaded
and the ESI source connected to the syringe filled up with the
Positive Ion Calibration standard mixture. The flow rate is set at
5 μL/min, and the system should be stabilized for 10 min; the cali-
bration is performed only when the TIC variation is constantly
lower than 10%.
3.5 In-House When setting up a steroidomic study, it is important not only to

Endogenous Steroids develop a suitable analytical procedure but also to measure ade-
Database quate reference material, with which data of the untargeted acqui-
sitions could be matched for the identification of most interesting
compounds. This protocol is based on the use of an in-house data-
base, which consists of up to 101 endogenous steroids, including
General
Runtime 1 to 17 min
Polarity positive
In-source CID 0.0 eV
Default charge state 1
Inclusion -
Exclusion -
Tags -
Full MS
Microscans 1
Resolution 70,000
AGC target 3e6
Maximum IT 125 ms
Number of scan range 1
Scan range 200 to 600 m/z
Spectrum data type Profile
dd-MS2 / dd-SIM
Microscans 1
Resolution 17,500
AGC target 5e4
Maximum IT 64 ms
Loop count 5
MSX count 1
TopN 5
Isolation window 0.4 m/z
Isolation offset 0.0 m/z
Fixed first mass -
NCE / stepped NCE 20, 40, 60
Spectrum data type Profile
dd Settings
Minimum AGC target 5.00e3
Intensity threshold 7.8e4
Apex trigger 2 to 6 s
Charge exclusion -
Peptide match -
Exclude isotopes -
Dynamic exclusion 6.0 s
Fig. 1 Full-scan and data-dependent MS/MS experiment parameters

Steroidomics 267
major androgens, progestogens, mineralocorticoids, and their

phase I metabolites, but the number of standards strongly depends
on the foreseen application. For this purpose, working solutions of
each analyte (see Table 1) at a concentration of 100 ng/mL are
prepared in the reconstitution solvent and injected in the UHPLC-
HRMS system using the analytical method described above. From
these injections, information on retention time and ionization
properties of all the target endogenous steroids has been retrieved
and summarised in Table 1.
3.6 Performance To monitor sample preparation and UHPLC-HRMS system per-

Monitoring formance, two different strategies based on IS-mix and QC-mix
are considered:
1. Regarding extraction, 20 μL of the IS-mix containing the four
deuterium-labeled internal standards of different endogenous
steroids (testosterone-d3 and androsterone-d4, androgens;
17α-hydroxyprogesterone-d8, progestogen; cortisol-d4, miner-
alocorticoid) are spiked in all samples. A noticeable reduced peak
area of IS-mix compounds could be the signal of a problem in the
sample preparation suggesting the re-extraction of a sample.
2. QC-mix is used in steroidomic studies for conditioning the ana-
lytical platform at the beginning of the sample sequence.
Furthermore, the injection of QC-mix at regular intervals dur-
ing the analytical sequence allows for the assessment of analyti-
cal variability and instrument performance (see Notes 8 and 9).
Chromatograms of selected identified compounds are presented in
Fig. 2.
RT: 9.39 - 9.80 SM: 7G RT: 10.87 - 11.21 SM: 7G RT: 11.16 - 11.55 SM: 7G RT: 9.95 - 10.18 SM: 7G
RT: 9.56 NL: 3.02E6 RT: 11.05 NL: 7.76E5 RT: 11.36 NL: 7.00E5 RT: 10.07 NL: 5.12E5
MA: 7228687 m/z= MA: 1863896 m/z= MA: 1725832 m/z= MA: 1100718 m/z=
100 289.21332-289.21910 100 273.21856-273.22402 100 273.21856-273.22402 100 287.19769-287.20343
F: FTMS + p ESI Full F: FTMS + p ESI Full F: FTMS + p ESI Full F: FTMS + p ESI Full
ms [200.00-600.00] ms [200.00-600.00] ms [200.00-600.00] ms [200.00-600.00]
MS MS MS MS
Relative Abundance
Relative Abundance
Relative Abundance
Relative Abundance
50 50 50 50
9.97
0 0 0 0
9.4 9.5 9.6 9.7 9.8 10.9 11.0 11.1 11.2 11.2 11.3 11.4 11.5 10.0 10.1
Testosterone Etiocholanolone Androsterone Androstenedione
100 291.22895-291.23477 100 271.20293-271.20835 100 361.19734-361.20456 100 347.21822-347.22516
ms [200.00-600.00] ms [200.00-600.00] ms [200.00-600.00] ms [200.00-600.00]
MS MS MS MS
Relative Abundance
Relative Abundance
Relative Abundance
Relative Abundance
50 50 50 50
0 0 0 0
10.6 10.7 10.8 10.9 10.0 10.1 10.2 7.0 7.1 8.0 8.1 8.2
DHT DHEA Cortisone Corticosterone
100 363.21297-363.22023 100 347.21822-347.22516 100 331.22346-331.23008 100 315.22871-315.23501
ms [200.00-600.00] ms [200.00-600.00] ms [200.00-600.00] ms [200.00-600.00]
MS MS MS MS
Relative Abundance
Relative Abundance
Relative Abundance
Relative Abundance
50 50 50 50
0 0 0 0
6.8 6.9 7.0 7.1 8.2 8.3 8.4 10.1 10.2 10.3 10.4 12.2 12.3
Cortisol 11-deoxycortisol 17α-hydroxyprogesterone Progesterone
Fig. 2 Chromatograms of 12 selected endogenous steroids extracted from a QC-mix sample

Table 1
268
Chemical formula, retention time, and most abundant ion m/z value of the 101 endogenous steroids of the in-house steroid database
Most abundant ion [m/z]

Chemical Retention time
Compound name formula Exact mass [Da] [min] [M+H]+ [M+Na]+ [M–H2O]+ [M–2H2O]+
11α-Hydroxyprogesterone C21H30O3 330.21950 8.82 331.22677
11β-Hydroxyandrostenedione C19H26O3 302.18820 8.24 303.19547
11β-Hydroxyandrosterone C19H30O3 306.21950 8.69 271.20564
Federico Ponzetto et al.
11β-Hydroxyepiandrosterone C19H30O3 306.21950 8.22 271.20564

11β-Hydroxyetiocholanolone C19H30O3 306.21950 8.56 271.20564
11β-Hydroxyprogesterone C21H30O3 330.21950 9.89 331.22677
11β-Hydroxytestosterone C19H28O3 304.20385 7.56 305.21112
11-Dehydrocorticosterone C21H28O4 344.19876 7.87 345.20604
11-Dehydrotetrahydrocorticosterone C21H32O4 348.23006 8.43 331.22677
11-Deoxycorticosterone C21H30O3 330.21950 9.79 331.22677
11-Deoxycortisol C21H30O4 346.21441 8.36 347.22169
11-Ketoetiocholanolone C19H28O3 304.20385 8.92 287.20056
11-Oxoandrosterone C19H28O3 304.20385 8.97 305.21112
16α-Hydroxyandrostenediol C19H30O3 306.21950 6.51 271.20564
16α-Hydroxyandrostenedione C19H26O3 302.18820 7.74 303.19547
16α-Hydroxyandrosterone C19H30O3 306.21950 9.10 289.21621
16α-Hydroxydehydroepiandrosterone C19H28O3 304.20385 7.62 287.20056
16α-Hydroxyetiocholanolone C19H30O3 306.21950 9.02 289.21621
16α-Hydroxytestosterone C19H28O3 304.20385 6.85 305.21112
16β-Hydroxydehydroepiandrosterone C19H28O3 304.20385 7.29 287.20056
17α,20α-Dihydroxyprogesterone C21H32O3 332.23515 9.07 333.24242
17α-Hydroxypregnenolone C21H32O3 332.23515 9.96 315.23186
19-Hydroxyandrostenedione C19H26O3 302.18820 7.07 303.19547
19-Hydroxytestosterone C19H28O3 304.20385 6.62 305.21112
20α-Cortolone C21H34O5 366.24063 6.99 331.22677
20α-Dihydrocortisone C21H30O5 362.20933 6.27 363.21660
20α-Dihydroprogesterone C21H32O2 316.24023 11.21 317.24751
20β-Cortol C21H36O5 368.25628 6.82 333.24242
20β-Cortolone C21H34O5 366.24063 7.11 331.22677
21-Deoxycortisol C21H30O4 346.21441 8.15 347.22169
21-Hydroxypregnenolone C21H32O3 332.23515 9.73 315.23186
Steroidomics
(continued)
269
Table 1
270
(continued)

3α,21-Dihydroxy-5α-pregnane-11,20- C21H32O4 348.23006 8.43 349.23734
dione
3α,5α-Tetrahydrocorticosterone C21H34O4 350.24571 8.46 315.23186
3α,5α-Tetrahydrocortisol C21H34O5 366.24063 7.41 331.22677
3α,5β-Tetrahydroaldosterone C21H32O5 364.22498 6.54 387.21420
3α,5β-Tetrahydrocorticosterone C21H34O4 350.24571 8.34 315.23186
3α,5β-Tetrahydrocortisol C21H34O5 366.24063 7.45 331.22677
3α,5β-Tetrahydrocortisone C21H32O5 364.22498 7.77 347.22169
5α,20α-Tetrahydroprogesterone C21H34O2 318.25588 12.47 301.25259
5α-Androstane-3α,17β-diol C19H32O2 292.24023 10.61 257.22638
5α-Androstan-3β-ol-7,17-dione C19H28O3 304.20385 7.36 287.20056
5α-Androstane-3β,12β,15α-triol C19H32O3 308.23515 5.69 273.22129
5α-Androstane-3β,7α,16β-triol C19H32O3 308.23515 6.73 273.22129
5α-Androstanedione C19H28O2 288.20893 11.35 289.21621
5α-Androstane-3β,17β-diol C19H32O2 292.24023 9.97 275.23694
5α-Dihydro-11-deoxycorticosterone C21H32O3 332.23515 10.98 333.24242
5α-Dihydrocorticosterone C21H32O4 348.23006 8.98 349.23734
5α-Dihydrocortisol C21H32O5 364.22498 7.54 365.23225
5α-Dihydrocortisone C21H30O5 362.20933 7.58 363.21660
5α-Dihydroprogesterone C21H32O2 316.24023 13.47 317.24751
5-Androsten-3β,17β-diol-16-one C19H28O3 304.20385 7.31 287.20056
5β-Androstane-3α,17β-diol C19H32O2 292.24023 10.36 275.23694
5β-Androstan-11α-ol-3,17-dione C19H28O3 304.20385 8.53 269.18999
5β-Androstan-11β-ol-3,17-dione C19H28O3 304.20385 9.13 305.21112
5β-Androstane-3α,17α-diol C19H32O2 292.24023 11.57 257.22638
5β-Androstane-3β,17α-diol C19H32O2 292.24023 10.74 257.22638
5β-Androstanedione C19H28O2 288.20893 11.45 289.21621
5β-Androstane-3β,17β-diol C19H32O2 292.24023 10.08 275.23694
5β-Dihydro-11-dehydrocorticosterone C21H30O4 346.21441 8.87 347.22169
5β-Dihydrocorticosterone C21H32O4 348.23006 9.02 349.23734
5β-Dihydrocortisol C21H32O5 364.22498 7.86 347.22169
5β-Dihydrocortisone C21H30O5 362.20933 8.10 363.21660
Steroidomics
5β-Dihydroprogesterone C21H32O2 316.24023 13.38 317.24751
(continued)
271
Table 1
272
(continued)

5β-Dihydrotestosterone C19H30O2 290.22458 10.85 291.23186
7α-Hydroxydehydroepiandrosterone C19H28O3 304.20385 7.00 287.20056

7α-Hydroxypregnenolone C21H32O3 332.23515 8.28 315.23186
7β-Hydroxydehydroepiandrosterone C19H28O3 304.20385 6.63 287.20056
Adrenosterone C19H24O3 300.17255 8.38 301.17982
Aldosterone C21H28O5 360.19368 6.48 361.20095
Allotetrahydrocortisol C21H34O5 366.24063 7.41 331.22677
Allotetrahydrodesoxycorticosterone C21H34O3 334.25080 10.65 317.24751
Androst-5-ene-3β,16α,17α-triol C19H30O3 306.21950 7.92 271.20564
Androstan-3β,5α,6β-triol C19H32O3 308.23515 9.92 273.22129
Androst-5-ene-3β,16β,17α-triol C19H30O3 306.21950 6.61 271.20564
Androstenedione C19H26O2 286.19328 10.12 287.20056
Androsterone C19H30O2 290.22458 11.41 273.22129
Corticosterone C21H30O4 346.21441 8.16 347.22169
Cortisol C21H30O5 362.20933 6.97 363.21660
Cortisone C21H28O5 360.19368 7.10 361.20095
Dehydroandrosterone C19H28O2 288.20893 10.55 271.20564
Dehydroepiandrosterone C19H28O2 288.20893 10.18 271.20564
Dihydrotestosterone C19H30O2 290.22458 10.79 291.23186
Epiandrosterone C19H30O2 290.22458 10.63 273.22129
Epitestosterone C19H28O2 288.20893 10.20 289.21621
Etiocholanolone C19H30O2 290.22458 11.12 273.22129
Pregnanediol C21H36O2 320.27153 12.10 285.25768
Pregnanetriol C21H36O3 336.26645 10.38 301.25259
Pregnanolone C21H34O2 318.25588 12.86 301.25259
Pregnenolone C21H32O2 316.24023 12.25 299.23694
Progesterone C21H30O2 314.22458 12.21 315.23186
Testosterone C19H28O2 288.20893 9.59 289.21621
Tetrahydro-11-deoxycortisol C21H34O4 350.24571 9.52 315.23186
Steroidomics
273
3.7 Analytical
The analytical sequence should be constructed carefully with the
Sequence
aim of simultaneously providing the best compromise between a
satisfying data acquisition and a clear measure of the data quality
and system performance. Solvent blank, negative serum samples,
conditioning samples, and QC-mix samples are added to the ste-
roidomic study serum sample injections helping the assessment of
the analysis quality. The proposal of a steroidomic analytical
sequence is described below:
1. A solvent blank injection at the beginning of the analytical
sequence helps the identification of background noise related to
the reconstitution solvent. If more than one solvent blank injec-
tion is performed, it is also possible to evaluate the presence of
any carry over effect.
2. Before starting the conditioning process, a minimum of two
injections of negative serum samples should be performed.
These injections are useful in evaluating the status of the ana-
lytical system, in particular in regard to the chromatographic
column. Indeed, the injection of negative serum samples could
help the assessment of any carry over of endogenous steroids
related to previous analyses, which is difficult to detect when
injecting the solvent blank.
3. After the negative serum sample injections, a minimum of five
QC-mix sample injections should be performed to stabilize the
chromatographic system. Furthermore, it allows conditioning
the HRMS system, because of a significant performance decrease
across the first injections of serum matrix. When QC-mix vol-
ume is limited due to scarce volume of study samples, it is also
possible to perform the system conditioning by injecting other
serum samples; the key aspect is to inject matrix-based samples
that are as similar as possible to the serum samples of the study.
For example, if the study is conducted on male subjects, it could
be possible to use, as conditioning samples, any other male
serum samples or a commercially available pool of male serum.
The system conditioning is mandatory to obtain reliable results
with adequate repeatability.
4. QC-mix sample injections are of crucial importance for the sta-
tistical analysis and for the evaluation of the instrument perfor-
mance in terms of detection and sensitivity of the analytes of
interest. The analysis of steroidomic study samples should
always start and end with a QC-mix sample injection to enable
quality assessment at the beginning and end of the analytical
batch; QC-mix injections could also be used during the data
treatment for normalization purposes. As a minimum of five
QC-mix injections are mandatory to obtain a sufficient robust-
ness for the statistical analysis, these injections should be per-
Steroidomics 275
formed at regular intervals (to be decided taking into account

the size of the study, the number of batches and therefore of the
length of the analytical sequence) across the sequence between
study samples.
5. Study samples are analyzed randomly in blocks of five in mini-
mum between a QC-mix and another in order to reduce the
bias of the data. Study sample injections are comprised between
two consecutive QC-mix injections, and the size of the block
should always be kept the same across all the sequences; the size
of the block is resulting from the total number of samples to be
analyzed, from the number of QC-mix necessary to have robust
statistical results and from the length of the sequence.
6. The end of sample sequence is identical to the beginning, i.e.,
with two injections of negative serum samples and one of sol-
vent blank. These injections help in evaluating the presence of
carry over as well as the increase or decrease of background
noise, which can affect the peak integration.
3.8 Data Treatment Raw data obtained from the UHPLC-HRMS analysis are pro-
cessed with appropriate software to obtain peak areas of all peaks
detected in the steroidomic study samples. Several solutions and
commercial and open-source-based strategies could be engaged;
however, the following steps should be achieved:
1. The first step of the data treatment is the alignment of the chro-
matograms. In most of the cases, it is possible to choose between
an automatic or manual alignment, in which the user should
define the peak of interest that should be used to align the chro-
matograms. We suggest, in the case of untargeted analyses,
where it is not possible to assess the nature of all the peaks in the
chromatogram and then decide which are important and which
are not, to use the automatic alignment, which uses various
peaks as a reference for the alignment.
2. The second and probably one of the most important steps of
the data treatment concerns the so-called peak picking. Here,
the user has to define where, and how, the software should
search for peaks in the aligned chromatograms. It could be
appropriate to perform the peak picking only in the retention
time window covering the period of the gradient, i.e., to exclude
all compounds that are too well retained and elute only in the
washing step or those which are not retained at all hence and
elute during the dead time. It is also necessary to define a sensi-
tivity threshold (the higher the sensitivity, the larger the num-
ber of peaks detected) and the type of adducts. For this step, we
suggest using the highest sensitivity in order to detect the maxi-
mum number of peaks, which could then be reduced employ-
ing statistical tools, and to include only the adducts typical of
steroidal compounds, such as [M+H]+, [M+H−H2O]+, and

[M+H−2H2O]+. Once the “peak picking” is performed, the
software will generate a list of features annotated by a mass
(m/z) and a retention time (min); for the analysis of serum
sample after SLE extraction, an average of 25,000 features is
considered as normal.
3. The last step of the data treatment is represented by the feature
annotation. As compound identification still constitutes a major
challenge when performing large-scale untargeted analyses for
providing biochemical interpretation, it is possible to achieve a
different level of annotation (Level 1–3) using both the previ-
ously created in-house database and dedicated databases includ-
ing endogenous and/or exogenous steroids which were
developed over the last years, e.g., the Human Metabolite
Database (HMDB), LipidMaps, and METLIN.
Firstly, a match with the in-house created database, including
the 101 endogenous steroids presented in Table 1, is performed
searching among the features generated by the “peak picking.”
When performing this match, it is mandatory to define the
range of uncertainty both for the mass (suggested 5 ppm) and
the retention time (suggested 1%). The features which show
the same retention time, m/z value, and isotopic pattern com-
pared to a reference endogenous standard present in the data-
base are annotated at Level 1. Then, by working with online
databases, it is also possible to predict the retention time of all
steroid compounds [14] of the selected database and search if
features in the list result in a match with the predicted retention
time, m/z value, and isotopic pattern, to achieve a Level 2
annotation. Finally, all features that have a match with the
online database but only for m/z and isotopic pattern criteria
are assigned with a Level 3 annotation.
When available, the comparison of MS/MS or MSn fragmen-
tation spectrum with entries of reference experimental MS/
MS spectra could also help the annotation process, even if in
the case of steroid compounds, which have highly similar
fragmentation patterns, the MS/MS information should be
preferably used as an exclusion criterion rather than an iden-
tification criterion.
4 Notes
1. Mobile phase A, as well as weak wash solution, should be pre-

pared fresh every week. This step is important to avoid bacterial
growth in aqueous solution used in the chromatographic
system.
Steroidomics 277
2. Mobile phase B should be replaced once each 4 months to

avoid retention time shifts caused by evaporation and conse-
quent changes in solvent composition.
3. All solvent lines should be primed before starting each analysis,
with the aim of eliminating air bubbles that could be formed in
the chromatographic system.
4. Chromatographic system pressure should be monitored during
the analytical batch and between different batches, to evaluate
chromatographic performance. Any significant drift in pressure
could influence retention times and therefore data
interpretation.
5. If the conducted steroidomic study is composed by more than
one analytical batch, it is important to clean the chromato-
graphic column between the batches setting the system to 80%
of mobile phase B for a minimum of 30 min. This step helps in
reducing the pressure of the system, which can increase after the
injection of several extracted serum samples.
6. When setting up the MS method, it could be useful to evaluate
in the MS tune software the presence of background noise ions
when conditioning the analytical column. Then, if necessary,
their m/z values should be added to an exclusion list to avoid
the frequent triggering of ddMS2 experiments resulting from
these interfering ions.
7. The ESI source should be cleaned before each analytical batch
as described in Subheading 3.4. This procedure ensures the best
sensitivity and repeatability, hence allowing the comparison
between serum samples of a same study injected in different
analytical batches.
8. QC samples are of crucial importance for evaluating the perfor-
mance of the instrument during an analytical sequence. It could
be useful to set up an automatic processing method in the
Xcalibur environment in order to integrate some target peaks/
compounds and monitor their areas and retention times in QC
samples across the whole sequence. Furthermore, when an ana-
lytical batch is part of a much larger study, the peak area and the
retention time values could also both represent a first approach
to compare the performance of the instrument between
sequences in different days.
9. Multivariate statistical analysis such as principal component
analysis (PCA) is the basic approach to assess the quality of the
acquired data when dealing with numerous detected features.
For example, it is desirable that in the score plot of PCA, all QC
samples are grouped in a unique cluster close to the center of
the graph.
References
1. Sanderson JT (2006) The steroid hormone reference intervals in healthy adults. Steroids
biosynthesis pathway as a target for endocrine- 76:244–253
disrupting chemicals. Toxicol Sci 94:3–21 9. Rosner W, Auchus RJ, Azziz R et al (2007)
2. Arukwe A (2008) Steroidogenic acute regula- Position statement: utility, limitations, and pit-
tory (StAR) protein and cholesterol side- falls in measuring testosterone: an endocrine
chain cleavage (P450scc)-regulated society position statement. J Clin Endocrinol
steroidogenesis as an organ-specific molecu- Metab 92:405–413
lar and cellular target for endocrine disrupt- 10. Handelsman DJ, Wartofsky L (2013)
ing chemicals in fish. Cell Biol Toxicol Requirement for mass spectrometry sex steroid
24:527–540 assays in the journal of clinical endocrinology
3. Shackleton CHL (2012) Role of a disordered and metabolism. J Clin Endocrinol Metab
steroid metabolome in the elucidation of sterol 98:3971–3973
and steroid biosynthesis. Lipids 47:1–12 11. Wartofsky L, Handelsman DJ (2010)
4. Kicman AT (2008) Pharmacology of anabolic Standardization of hormonal assays for the
steroids. Br J Pharmacol 154:502–521 21st century. J Clin Endocrinol Metab
5. World Anti-Doping Agency (WADA), 95:5141–5143
Montreal (2015) Anti-doping testing figures 12. World Anti-Doping Agency (WADA),
report. http://www.wada-ama.org. Accessed Montreal (2016) endogenous anabolic andro-
May 2017 genic steroids, measurement and reporting,
6. Basaria S (2010) Androgen abuse in athletes: technical document TD2016EAAS. http://
detection and consequences. J Clin Endocrinol www.wada-ama.org. Accessed May 2017
Metab 95:1533–1543 13. Sjovall J (2004) Fifty years with bile acids and
7. Büttner A, Thieme D (2010) Side effects of steroids in health and disease. Lipids
anabolic androgenic steroids: pathological 39:703–722
findings and structure-activity relationships. 14. Randazzo GM, Tonoli D, Strajhar P et al
Handb Exp Pharmacol 195:459–484 (2017) Enhanced metabolite annotation via
8. Fanelli F, Belluomo I, Di Lallo VD et al (2011) dynamic retention time prediction: steroido-
Serum steroid profiling by isotopic dilution– genesis alterations as a case study. J Chromatogr
liquid chromatography–mass spectrometry: B Analyt Technol Biomed Life Sci.
comparison with current immunoassays and 1071:11–18
Chapter 19
Metabolomics in Human Acute-Exercise Trials: Study

Design and Preparation
Aikaterina Siopi and Vassilis Mougios
Abstract
Metabolomics can be of great value in the study of exercise metabolism. However, because of the high
intraindividual and interindividual biological variability of the human metabolome, special considerations
should be taken into account when designing an acute-exercise metabolomic study. To study different
exercise parameters, e.g., different exercise modes, intensities, etc., a crossover study design, where each
participant acts as their own control, is preferable to a parallel design, one involving different groups of
participants. Moreover, the study should include a no exercise, control trial. Before each trial, participants
should follow carefully designed preparatory steps to control for possible confounding factors, i.e., main-
tain repeatable and constant conditions for all individual trials of the study to minimize variation due to
factors other than the one(s) being studied. This chapter focuses on the design of human metabolomic
studies, where the intervention is an acute metabolic challenge, such as an exercise bout or a test meal, and
presents some basic steps for screening potential participants, performing preliminary tests, preparing for
the trial day, and performing the trial.
Key words Metabolomics, Acute exercise, Study design, Humans
1 Introduction
Metabolomics can be of great value in the study of exercise metab-

olism. It can help to decipher the molecular mechanisms behind
exercise-induced metabolic responses and benefits. The response
of the metabolome to exercise may be indicative of an individual’s
fitness or disease status; therefore, exercise metabolomics could be
used for diagnostic/prognostic purposes [1]. Moreover, metabo-
lomics can prove very useful in assessing the effectiveness of exer-
cise interventions and training regimens. Ultimately, all these could
translate into personalized exercise prescription on the basis of an
individual’s metabolic profile at rest or in response to exercise.
Therefore, metabolomics have been increasingly applied in exer-
cise intervention studies during the past decade [1–7]. Currently,
279
280 Aikaterina Siopi and Vassilis Mougios
research efforts are focusing on recording a comprehensive chart

of the molecular changes caused by exercise.
When the aim is to study different exercise parameters, e.g.,
different exercise modes or intensities, a crossover design, where
each participant serves as their own control, seems to be preferable
to a parallel design, which includes different groups of participants
[8]. The reason is the very high interindividual variability of the
human metabolome. Let us assume that the study objective is to
compare the effects of three different exercise types on the metab-
olome. In a parallel study design, we would randomize the partici-
pants to three groups, and each group would perform exercise of
one type. In a crossover design, however, all of the participants
would perform all three exercises in a randomized order. The latter
design implies more visits for each participant, but it has the advan-
tage that their metabolic responses to one exercise type will be
compared to their own responses to another exercise type, not the
responses of other individuals. However, if the study objective is
the effect of a specific health condition or disease on the response
of the metabolome to exercise, a parallel design, with a carefully
matched control group, is inevitable.
In an acute-exercise metabolomic study, it is also necessary to
include a no exercise trial. This should be identical to the exercise
trials in all aspects except that the participants will rest instead of
exercising. At the end, the post-exercise metabolic profiles or fin-
gerprints will be compared to that of the resting trial in order to
isolate the effects of exercise from other confounding factors such
as diurnal variation or fasting. Moreover, the samples from the
resting trial can be used to correct for the batch effect, which hin-
ders metabolomic analysis, especially when based on LC-MS plat-
forms. As long as all samples from one participant are analyzed in
the same batch, normalizing the data for each metabolite in the
samples of the exercise trials to the respective value in the resting
trial could correct for unwanted batch effects [9].
Regarding the choice of biological matrix, blood plasma/
serum and urine are biofluids that have been traditionally used to
monitor health status, since they are easily collected and reflect the
global state of an individual. Urine may have an advantage over
blood for biomarker identification purposes, as it is under no
homeostatic mechanisms. Change is the most important aspect of
biomarker discovery. Metabolite changes in blood could actually
be detected in urine with higher sensitivity, since blood changes
tend to be quickly normalized by strict homeostatic mechanisms.
Urine is accessible noninvasively, so it is easier to collect in large
quantities and through multiple time points to study time-
dependent changes in metabolites. Moreover, urine is more stable
and less complex than other biofluids, therefore, an ideal source of
biomarkers [10, 11].
Design of Acute-Exercise Metabolomic Studies 281
In the following sections, we will present all the recommended

materials and procedures when preparing for an acute-exercise
metabolomic study in humans.
2 Materials
The following instruments are recommended:

●● Sphygmomanometer
●● Automatic biochemistry analyzer for preliminary biochemical
screening.
●● A device for body composition analysis such as skin calipers,
bioelectrical impendence analysis (BIA), or dual-energy X-ray
absorptiometry (DXA).
●● Indirect calorimetry metabolic cart—oxygen and carbon diox-
ide analyzers.
●● Treadmill or cycle ergometer.
●● Heart rate monitor.
●● Pedometer (step counter).
●● Nutritional analysis software.
3 Methods
3.1 Screening The screening process is necessary to determine whether an indi-

Potential Participants vidual meets the criteria to participate in the study. There should
be clearly defined inclusion and exclusion criteria. The inclusion
criteria depend on the particular target population of the study,
i.e., gender, age, lifestyle (e.g., sedentary or active), training status,
health status, etc. The exclusion criteria usually aim to eliminate
possible confounding factors and to decrease biological variation,
which is a big problem in metabolomic studies. Exclusion criteria
often include the presence of acute or chronic disease (other than
the one perhaps being tested), any contraindication for exercising,
use of medication or supplements, smoking, and dieting or recent
change in body weight (e.g., >2 kg within 6 months). The usual
steps for the screening process are:
1. Prescreening/phone assessment. Using a form that you will
have created for this purpose, record the individual’s personal
information, including contact information, information
regarding all inclusion and exclusion criteria, any previous par-
ticipation in other research studies, as well as availability (days
and hours) for participating in the study (see Note 1).
2. First visit. If a volunteer is eligible to continue, arrange a first

appointment for them to visit the research facilities and meet
the staff. During the meeting, explain all steps of the study
thoroughly (see Note 2), and have the volunteer read and sign
the necessary consent forms, while you answer any questions
they may have (see Note 3). Help the volunteer complete all
necessary paperwork, such as medical history, physical activity
questionnaires, and dietary questionnaires.
3. Medical examination and screening measurements. (This step
may or may not be carried out during the first visit, depending
on what is convenient.) Collect basic anthropometric data (e.g.,
height, weight, and waist circumference), vital signs (e.g., blood
pressure and heart rate), and a blood sample for biochemical
screening (under fasted conditions). Medical clearance for par-
ticipating in exercise trials should follow the established guide-
lines [12]. More than one visit may be necessary to complete
the screening process.
3.2 Preliminary Tests After the screening process and before the trials, participants usu-
ally need to go through some preliminary tests (see Note 4). These
can include:
1. Body composition measurements, such as lean body mass, total
fat mass, and visceral fat mass (see Note 5). Make sure the par-
ticipants have followed the necessary preparations for the
respective analysis. These measurements can be used as descrip-
tive characteristics of the study sample. Moreover, they can be
correlated with outcomes of the metabolomic analysis.
2. Measurement or estimation of resting energy expenditure (see
Note 6). This measurement will be used to design the dietary
plan of the participants during the study in order to decrease
intraindividual and interindividual variations.
3. Cardiorespiratory fitness assessment. Perform a maximal or
submaximal incremental test (depending on your study sample)
to assess maximal oxygen consumption (VO2max) and maximal
heart rate [12]. This measurement may be used to set the exer-
cise intensity for endurance exercise trials.
4. Muscular strength measurement. Perform a determination or
prediction (depending on your study sample) of one-repetition
maximum [13]. This measurement may be used to set the exer-
cise intensity for resistance exercise trials.
3.3 Preparation Arrange some extra time at the end of the preliminary tests appoint-
for the Exercise Trials ment to talk with each participant about the scheduling and prepa-
ration for the trials. Again, be thorough and provide all instructions
in writing as well as verbally for the participants to take home (see
Note 2). Schedule the next appointment (see Notes 4 and 7). A
Fig. 1 Overview of the preparation for each trial of the study (see Note 10)
usual practice is to space the trials 1 week apart. Do not exceed

2 weeks to avoid the risk of losing comparability among trials. For
premenopausal women, each trial should be scheduled at the same
phase of the menstrual cycle [14].
The preparation for each trial usually begins 2 days before
(Fig. 1). The aim is to control for possible confounding factors.
You have to maintain repeatable and constant conditions for all
individual trials of the study to minimize biological variation due
to factors other than the one being studied (i.e., exercise). This
step is critical for the success of an acute human metabolomic
study, where variation is a huge issue. Take the time to help partici-
pants comply with the instructions (see Note 8).
Participants need to record their diet [15] and physical activity
for the agreed period (1 or 2 days) before the first trial in order to
replicate it before the next trial(s), thus minimizing intraindividual
variance. To minimize interindividual variance, all participants
should have equivalent dietary intakes: they should follow isoener-
getic diets (i.e., diets matching energy intake to energy expendi-
ture) or, in case a dietary intervention is planned together with the
exercise intervention, cause the same energy deficit or surplus to all
participants. To this end, a qualified dietitian should design person-
alized dietary plans for at least the day before the metabolomic trial
(see Note 9). Participants should follow their plans but nevertheless
record what they actually ate. At the end of the study, all dietary
records must be analyzed to validate that there was no significant
nutritional difference that could have affected the results of the
metabolomic analysis. Unless otherwise dictated by the design of
the study, participants should be instructed to refrain from intense
exercise, caffeine intake, and alcohol consumption on the day before
the trial, as all these may have a reverberation on metabolism.
To prepare for the first trial:
1. Provide each participant with special forms to record their daily
dietary intake, physical activity, and step count. They usually need
to start recording all these 2 days before the trial (see Note 10).
2. Give each participant a pedometer and instruct them how to

use them properly. Step counts provide a rough estimate of the
habitual physical activity.
3. Go through the forms with the participant, give thorough
instructions, and make sure they have no unanswered
questions.
4. Provide the participant with printed detailed instructions as
well to take with them and refer to at all times. These instruc-
tions should include how to keep dietary and physical activity
records, how to measure food portions, and how to use the
pedometer. Remind the participant to avoid intense exercise,
caffeine intake, and alcohol consumption.
5. Provide printed detailed instructions for the night before and
the morning of the trial day. These instructions usually include
having the last meal 12 h before their morning appointment in
the lab (but no more than 14 h), avoiding any food consump-
tion at home in the morning, shedding of first morning urine if
there will be urine sampling, etc.
6. Provide each participant with contact information of a member
of the research team to refer to for any questions or problems
that may arise during the preparation for the trials.
7. Ask them to bring all records with them at all visits.
3.4 Trial Day After arrival of the participant to the lab and before starting the
trial:
1. Weigh them to make sure that there has been no considerable
change since the previous measurement.
2. Go with them through a checklist to make sure they have suc-
cessfully followed all necessary steps of the preparation
process.
3. Check their dietary and physical activity records of the preced-
ing days.
4. Take a 24-h dietary recall to cross-validate the dietary record of
the previous day and to make sure it has been completed prop-
erly (see Note 11).
5. If they have failed to follow any of the preparation steps and you
feel that the reliability of the process has been compromised,
you may have to reschedule the trial.
During the trial:
1. Have the participants remain seated and relaxed during the
non-exercising parts of the trial.
2. Mark their water consumption during the first trial and have
them replicate it during the next trial(s).
3. Maintain constant environmental conditions (room tempera-

ture and humidity).
4. Maintain a stress-free environment as possible (the room should
be tidy, quiet, as private as possible, not crowded, etc.).
At the end of the first trial, schedule the next trial and go
through the preparatory instructions with the participants (see Note
12). For all trials participants have to follow exactly the same prepa-
ratory steps as they did for the first trial (Fig. 1). If a participant’s
dietary intake for the day before the trial is even slightly different
from the prescribed dietary plan, they should repeat their actual
dietary intake (not the prescribed) as recorded in the respective
dietary record (and cross-validated with the 24-h dietary recall).
4 Notes
1. Even if there is compensation for the participants and a large

number of candidates, usually the necessary strict criteria make
recruitment difficult. Only a small portion of those initially
interviewed will continue through the screening process.
2. Be thorough when explaining the study process to the partici-
pants. Provide written information/instructions as well. If
possible, use figures and/or tables to give an overview of the
study. Nonetheless, remember to repeat the process/instruc-
tions at every step of the way to ensure maximal compliance.
It is your job to guide the participants throughout the process
and help them comply with the instructions. Compliance is
very important to ensure comparability among study trials or
groups and to control for the high intraindividual and interin-
dividual biological variance that complicates metabolic
profiling.
3. All procedures have to be approved by the institutional review
board and comply with the Helsinki declaration of 1975, as
revised in 2013.
4. After enrollment of a participant to the study, it is important
that they complete all trials as soon as possible. Any unneces-
sary delays increase the chances for dropping out, losing eligi-
bility to participate, or losing comparability of the trials.
5. The gold standard for body composition analysis is
DXA. However, this method requires expensive and non-por-
table equipment. A practical compromise can be the method
of BIA. Preparation for this analysis includes abstention from
drinking water or other liquids for the last 4 h; abstention
from eating for the last 12 h; abstention from exercising,
drinking caffeine, drinking alcohol, using a sauna, or using a
hot tub for the last 24 h; and abstention from diuretics for the
last 7 days prior to the analysis.
6. If possible, use the method of indirect calorimetry with the
canopy technique to measure resting energy expenditure.
Alternatively, you can use equations to estimate energy needs
[16].
7. In a crossover design, each participant must complete all trials
in a randomized order. You can use a random-number genera-
tor software (one of the many available online for free) to
obtain random sequences.
8. In human trials, the interpersonal communication skills of the
researcher/staff are very important.
9. The dietary plan should be carefully designed to maximize
compliance and avoid major effects of diet on the metabo-
lome. It should not include complicated recipes, expensive
ingredients, or food that the participant does not like, is not
used to, or is allergic to. Ideally, the research project should
provide catering for all meals of the participants.
10. Depending on how many sampling points there are through-
out the day of each trial, you may need to control diet and
physical activity on the day of the trial as well.
11. Dietary recalls need to be taken by a trained and experienced
professional to be accurate.
12. The best way to match acute-exercise bouts is to measure
exercise energy expenditure by indirect calorimetry.
References
1. Lewis GD, Farrell L, Wood MJ et al (2010) 5. Enea C, Seguin F, Petitpas-Mulliez J et al (2010)

Metabolic signatures of exercise in human (1)H NMR-based metabolomics approach for
plasma. Sci Transl Med 2(33):33ra37. https:// exploring urinary metabolome modifications
doi.org/10.1126/scitranslmed.3001006 after acute and chronic physical exercise. Anal
2. Daskalaki E, Easton C, Watson DG (2014) Bioanal Chem 396(3):1167–1176. https://doi.
The application of metabolomic profiling to org/10.1007/s00216-009-3289-4
the effects of physical activity. Curr Metabol 6. Nieman DC, Gillitt ND, Sha W (2015)
2(4):233–263 Metabolomics-based analysis of banana and
3. Pechlivanis A, Kostidis S, Saraslanidis P et al pear ingestion on exercise performance and
(2013) 1H NMR study on the short- and recovery. J Proteome Res 14(12):5367–5377.
long-term impact of two training programs of https://doi.org/10.1021/acs.
sprint running on the metabolic fingerprint of jproteome.5b00909
human serum. J Proteome Res 12(1):470– 7. Peake JM, Tan SJ, Markworth JF (2014)
480. https://doi.org/10.1021/pr300846x Metabolic and hormonal responses to isoener-
4. Pechlivanis A, Papaioannou KG, Tsalis G et al getic high-intensity interval exercise and con-
(2015) Monitoring the response of the human tinuous moderate-intensity exercise. Am
urinary Metabolome to brief maximal exercise J Physiol Endocrinol Metab 307(7):E539–
by a combination of RP-UPLC-MS and (1)H E552. https://doi.org/10.1152/
NMR spectroscopy. J Proteome Res ajpendo.00276.2014
14(11):4610–4622. https://doi. 8. Heinzmann SS, Merrifield CA, Rezzi S et al
org/10.1021/acs.jproteome.5b00470 (2012) Stability and robustness of human met-
abolic phenotypes in response to sequential 13. Dohoney P, Chromiak JA, Lemire D et al

food challenges. J Proteome Res 11(2):643– (2002) Prediction of one repetition maximum
655. https://doi.org/10.1021/pr2005764 (1-RM) strength from a 4-6 RM and a 7-8 RM
9. Siopi A, Deda O, Manou V (2017) Effects of submaximal strength test in healthy young
different exercise modes on the urinary meta- adult males. J Exercise Physiol 5:54–59
bolic fingerprint of men with and without met- 14. Wallace M, Hashim YZ, Wingfield M et al
abolic syndrome. Metabolites 7(1). https:// (2010) Effects of menstrual cycle phase on
doi.org/10.3390/metabo7010005 metabolomic profiles in premenopausal
10. Wu J, Gao Y (2015) Physiological conditions women. Hum Reprod 25(4):949–956.
can be reflected in human urine proteome and https://doi.org/10.1093/humrep/
metabolome. Expert Rev Proteomics deq011
12(6):623–636. https://doi.org/10.1586/1 15. Walsh MC, Brennan L, Malthouse JP (2006)
4789450.2015.1094380 Effect of acute dietary standardization on the
11. Li M (2015) Urine reflection of changes in urinary, plasma, and salivary metabolomic pro-
blood. Adv Exp Med Biol 845:13–19. https:// files of healthy humans. American. Am J Clin
doi.org/10.1007/978-94-017-9523-4_2 Nutr 84(3):531–539
12. American College of Sports Medicine (2016) 16. Harris JA, Benedict FG (1918) A biometric
ACSM's guidelines for exercise testing and study of human basal metabolism. Proc Natl
prescription, 10th edn. Wolters Kluwer, Acad Sci U S A 4(12):370–373
Philadelphia, PA
Index
A D
Acute-exercise��279 Data analysis workflow�� 27, 103, 232
Algorithm��30–32, 34, 36, 37, 42, 112, 142–144, 251 Data conversion��29, 35, 36, 43, 45
Analyte quantitation�� 21, 27, 66, 78, 84, 209, 225 Data integration�� 35, 240, 245, 256
Analytical batch��20, 94, 95, 274, 277 Data modeling��28
Androgens�� 261, 267 Data quality assessment�� 96, 100, 103
Anionic metabolite��184, 185, 187, 188, 190 Data treatment�� v, 10, 27–38, 78–79,
Annotation�� 41, 43, 46, 220, 232, 254, 258, 276 246–248, 251–255, 263, 274–276
Aroma�� 48, 178, 214–216 Data visualization��96
Artificial chewing��216–219 Deconvolution�� 84, 232, 233, 243, 245, 251, 252, 254
Atmospheric-pressure chemical ionization Derivatization�� 7, 84, 135, 139, 140, 142–144,
(APCI)��9, 214 146, 151, 156, 159, 164–173, 262
Authentication��209 Design of experiment (DoE)�� v, 27, 35, 134,
Authenticity�� v, 228 216, 227, 228, 231, 233
Direct infusion mass spectrometry (DIMS)��7, 214
B Dispersive liquid liquid microextraction
Background electrolyte (BGE)��185–189 (DLLME)�� 159–162, 164–173, 176
Biological database��42
E
Biological interpretation�� 41, 245, 255, 256
Biological variability�� 216, 226 Electrospray ionization (ESI)�� 9, 10, 18, 91, 95, 102,
Blood�� v, 7, 15, 66, 70, 77, 118, 119, 121, 112, 184, 186, 190, 227, 229–231, 233, 242, 264, 265
122, 125, 126, 130, 136, 138–139, 146, 280, 282 Endocrine diseases��261
Endogenous metabolites�� 7, 94, 183, 184
C Equilibration injections�� 77, 78, 80
Caenorhabditis elegans��240 Exact mass�� 42–44, 46–49, 53
Calibration curve��77–80 Exercise metabolomics��279–281
Calibration solutions��17, 101, 105–106, 175 Exercise modes��280
Capillary electrophoresis (CE)�� v, 9, 83, 188, 189, 250 External calibration curve��77–79
Carr-Purcell-Meiboom-Gill sequence (CPMG)��126, Extracted ion chromatogram (XIC)��22–24
127, 249, 254 Extraction�� 5, 27, 35, 70, 79, 92, 93, 100–102, 104,
Cationic metabolite�� 187, 188, 190 105, 114, 119, 121–124, 129, 134, 136–139, 145, 146,
Cell culture��125, 129, 133, 136–138, 150, 151, 153, 154, 156, 161, 176, 196, 199, 201, 204,
140, 141, 162, 190 207, 208, 220, 222, 230, 233, 240, 241, 245–247, 251,
Cerebrospinal fluid (CSF)�� 118, 125, 130 254, 256–258, 262–264, 267, 276
Chemical structure�� 42, 46, 47, 54
F
Chemoinformatics��v, 41–60
Chloroformate derivatization��160 Features��21, 23, 24, 28, 30, 33–36, 38, 112, 113,
Chromatogram�� 8, 21–24, 93, 95, 96, 107, 108, 135, 140, 160, 226, 240, 252, 253, 255, 265, 276, 277
140, 141, 151, 197, 227, 248, 249, 267, 275 Fecal sample��149–156
Chromatogram alignment�� 21, 275 Food��v, 7, 11, 70, 203, 205, 213,
Command line�� 30, 43, 44, 46, 48, 53, 60 214, 216, 284, 286
Confounding factors�� 29, 280, 281, 283 Food analysis�� 199, 203
Correlation analysis�� 144, 256 Fresh fruit tissue�� 215, 217, 218
https://doi.org/10.1007/978-1-4939-7643-0, © Springer Science+Business Media, LLC, part of Springer Nature 2018
289
Metabolic Profiling: Methods and Protocols
290 Index

G Metabolic profiling�� v, 3–12, 84, 99–101, 106, 112, 113,
117–130, 134, 150, 151, 185–188, 190, 240, 285
Gas chromatography (GC)��v, 3, 4, 7, 10, 15, 16, Metabolome coverage�� 99, 239, 245,
83, 99, 134, 150–152, 154–156, 159–162, 164–173, 249, 250, 253
176, 196, 217, 220, 226, 261 Metabolomics�� v, 3, 15, 23, 24, 35, 41, 44, 54,
Global metabolic profiling��v, 16, 17 58, 65, 117, 133, 149–156, 159–179, 183, 195, 203,
Gut microbiota��150 205, 225, 239–259, 279–286
Metabonomics��v, 3, 5, 15, 83, 84, 239, 245,
H
249, 250, 253
Headspace�� 215, 216, 218–222 Metabotype�� 4, 5, 15
Heptafluorobutyl chloroformate (HFBCF)�� 159–162, Metadata organization�� 29–30, 234
164–173, 176 Method validation�� 16, 225, 226
Heteronuclear single quantum correlation Methoximation��135
(HSQC)��128 Mineralocorticoids��267
High-resolution magic-angle spinning (HR-MAS)��204 Missing peaks�� 29, 34, 142, 232
Human metabolome��11, 55, 145, 255, 280 Molecular formula��10, 42–45, 49–51, 53
Human metabolome database (HMDB)�� 11, 42, 55, Mono-layer extractions��240
79, 113, 145, 209, 255, 276 Multi-method��240
Hydrophilic interaction liquid chromatography
(HILIC)��8, 66, 84, 100, 101, 104, 105, N
107, 109–111, 114, 152, 154–156, 227, 229–231, 233, Normalization�� 16, 20, 21, 41, 124, 135–138,
240, 242, 245, 247, 250, 252 140–144, 175, 179, 232, 252, 274
Nuclear magnetic resonance (NMR)��v, 4–7, 9–11, 15,
I
65, 83, 99, 117, 134, 150–155, 195, 197, 200, 203–210,
Identification�� v, 4, 5, 7, 10, 21, 30, 41, 42, 49, 240, 242, 243, 245, 246, 249, 254, 255, 257, 258
53, 66, 92, 93, 99, 102, 109, 113, 128, 134, 136, 137, Nuclear Overhauser effect spectroscopy
140, 141, 196, 203, 209, 252, 255, 265, 274, 276, 280 (NOESY )�� 127, 130
In-house database�� 265, 267–273, 276 Nutrimetabolomics��118
Ion pair chromatography�� 66, 83–96
Isotope pattern�� 42, 49–53 O
Oenological topic��227–228
J
Online databases��276
Java�� 43–46, 48, 60 Orthogonal projection to latent structures
(OPLS)�� 210, 243, 253, 255
K
P
Kyoto Encyclopedia of Genes and Genomes
(KEGG)�� 11, 42, 56–60, 145, 243, 255, 256 Participant��280–286
Pathway mapping�� 243, 255, 256
L Peak alignment��9, 21, 23, 41, 102,
Large-scale monitoring��261 112, 232, 275
LipidMaps��42, 55, 60, 113, 276 Peak picking�� 18, 22, 29–31, 34, 36, 37, 41,
Liquid chromatography (LC)�� 18, 27–38, 66–69, 112, 232, 233, 251, 275, 276
85, 100, 107 Pharmacometabolomics��118
LogD��46–49 Phase I metabolites��267
Physicochemical parameter��43
M Plants�� 7, 136, 146, 234
Pre-analytical considerations�� 92, 134
Mass spectrometry (MS)�� v, 4, 7–10, 15, 16, 23,
Preliminary tests�� 140, 282
24, 27–38, 46, 49, 54, 65–80, 83, 84, 91–92, 99–101,
Principal component analysis (PCA)�� 9, 10, 21–24,
104, 107–109, 112, 113, 115, 118, 134, 137, 140,
96, 144, 209–210, 228, 232, 233, 235, 243, 252, 253,
141, 150, 183–187, 190, 191, 195, 203, 213, 218,
255, 277
220, 226, 229–231, 240, 242, 243, 245, 248, 251,
Progestogens��267
255, 262, 264–266, 276
Proton transfer reaction-mass spectrometry
Matrix effect�� 66, 79, 135, 226
(PTR-MS)��v, 214, 215
Metabolic Profiling: Methods and Protocols
Index
291
Q Steroid profiling�� 160, 261, 266–273

Steroidomics�� 261, 266–273
Quality check��22, 24, 29, 32–34, 93, 177 Study design�� 4, 29, 279
Quality control (QC)�� v, 9, 15–24, 36, 77, 78, 92–94, Study preparation��175
96, 100, 103–108, 112–115, 124, 126, 130, 135, 142,
175, 177, 179, 209, 226–228, 231–235, 240, 247, 248, T
251–253, 255, 258, 259, 262, 263, 267, 274, 275, 277
Targeted metabolomics�� 16, 66, 121
Quantitation��21, 66, 78–80, 130, 226
Test mixture�� 85, 92–94
Quantitative methods��226
Tissue��v, 5, 16, 17, 65, 70, 77, 84, 92,
Questionnaire��282
93, 99–102, 104, 105, 107, 110, 113, 114, 118, 119,
R 121–127, 129, 130, 133, 136–138, 145, 152, 162,
215–219, 239–241, 243, 245, 246, 248, 250, 254,
Randomization�� 20, 35, 243, 258 257–259, 261
Raw data�� 21, 28, 30, 34–36, 94, 102, 275 Tissue aqueous extraction�� 110, 246
Reproducibility�� 22, 28, 33, 66, 83, 84, 92–94, Tissue organic extraction�� 111, 258
96, 118, 184, 196, 197, 216, 240, 259 Total correlation spectroscopy (TOCSY )�� 128, 255
Reversed phase liquid chromatography�� 3, 84, Trimethyl-silyl (TMS) derivatives�� 135, 139
100, 227, 240 Two-phase extraction��195–202
Robustness�� 66, 94, 274
R scripts��29, 43, 45, 50, 57 U
S Ultra (high) performance liquid chromatography
(UHPLC)�� 8, 18, 91, 100, 107, 227,
Sample preparation�� v, 5, 7, 8, 19, 24, 66, 70–77, 240, 261–277
100, 102–105, 118, 119, 124, 125, 149–151, 153–156, Untargeted metabolomics�� 3, 16, 21, 84,
159–162, 164–173, 176, 226, 231, 232, 234, 263, 267 100, 112, 117, 121, 133–146, 226
Scaling��21, 22, 232, 252, 255 Urine��3, 5, 7, 8, 15, 17–20, 23, 66, 70, 77, 84, 99,
Screening measurement��282 101–103, 106–109, 112, 118, 120, 123, 125, 126,
Sequence (analytical/sample)�� 21, 23, 92–96, 177, 210, 130, 160–175, 177–179, 184, 187, 188, 280, 284
226–228, 231–235, 245, 249, 254, 264, 267, 274, 275,
277, 286 V
Sheathless capillary electrophoresis-mass spectrometry
Validation��v, 15, 23, 24, 42, 49, 85,
(CE-MS)�� v, 7, 9, 183–191
92–94, 225, 232
Simplified molecular-input line-entry system
Volatile organic compounds (VOCs)��v, 213–222
(SMILES)�� 42, 46–48, 54
Spectrum�� 31, 123, 126, 128, 140, 141, 190, W
198, 200, 208, 209, 215, 217, 220–222, 252, 276
Standard addition method��79 Wine��v, 11, 204,
Standardization�� 4, 16, 66, 84, 134, 139, 262 206, 225–235
Statistical analysis��10, 18, 23, 24, 27, 34, 93, 94, 136, Wine authenticity��228
144–145, 203, 209, 243, 245, 252, 255, 274, 277 Wine quality��228

2018 Book MetabolicProfiling

Uploaded by

Copyright:

Available Formats

2018 Book MetabolicProfiling

Uploaded by

Document Information

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

2018 Book MetabolicProfiling

Uploaded by

Copyright:

Available Formats

Methods in

Molecular Biology 1738

For further volumes:

Methods and Protocols

ISSN 1064-3745 ISSN 1940-6029 (electronic)

Library of Congress Control Number: 2018931170

© Springer Science+Business Media, LLC, part of Springer Nature 2018

Printed on acid-free paper

Thessaloniki, Greece Georgios A. Theodoridis

1 Metabolic Profiling: Status, Challenges, and Perspective������������������������������������� 3

5 HILIC-MS/MS Multi-Targeted Method for Metabolomics Applications����������� 65

Part III Plant/Food Applications

13 Two-Phase Extraction for Comprehensive Analysis of the Plant

Part IV Life Science Applications

17 Tissue Multiplatform-Based Metabolomics/Metabonomics

M.R. Abellona U • Section of Biomolecular Medicine, Division of Computational

Alexandra Kolesnikova • NMR Laboratory, Chemistry Department, University of Crete,

Serge Rudaz • School of Pharmaceutical Sciences, University of Geneva, University of

Metabolic Profiling: Status, Challenges, and Perspective

Key words Metabolomics, Metabonomics, Biomarker, Metabolite identification, MetID, Biochemical

The field of untargeted metabolic profiling, also known as metabo-

enabled the relatively rapid detection and identification of the

9000 number of publications per year

Genomics Proteomics Transcriptomics Metabolomics

as to provide unequivocal identification, comparison with what is

drug metabolites and endogenous metabolites present in the sam-

(high) performance LC separations (UPLC, UHPLC) based on

Fig. 3 Representation of a 3D mass chromatogram obtained from the reversed-­

much information or offering much freedom in the selection of

mentary methods, such as NMR spectroscopy, or alternatively syn-

small metabolites. For these and many other reasons, we believe

Quality Control and Validation Issues in LC-MS

Key words Quality control, Untargeted metabolomics, Biological samples

Metabolomics or metabonomics, two terms interwoven with each

tissues [3, 4]. However, the comprehensive analysis, simultaneous

samples. For larger-scale studies, such schemes may not be possi-

All solvents (methanol, acetonitrile, formic acid) used should be of

2.1 Stock, Working,

2.3 Chromato-­ Chromatographic analysis can be performed on a HSS T3 C18

the procedure described for urine in Subheading 3.3.1, step 4.

solutions (see Notes 3 and 8). Similarly blank sample injections

In untargeted metabolomic studies, data analysis strategies

3.5 Data Analysis

carefully to find any trend that might indicate system underperfor-

discarded outright but rather be thoroughly scrutinized. QC

1. The storage period for each standard in the freezer is dependent

5. Apply proper cleanup steps during sample preparation. Filtering

6. During sample preparation, all necessary safety precautions

Data Treatment for LC-MS Untargeted Analysis

Liquid chromatography-mass spectrometry (LC-MS) is an estab-

Imputation of missing values

Data Modeling &

s­ o-­called data preprocessing. The objective of this phase is to sum-

1. Metadata organization. Spreadsheets to inspect/fill tables.

ISA-Tab files can be managed directly by using the Risa pack-

3.6 Preprocessing: At the end of the preprocessing, it is extremely important to check

Retention Time Deviation vs. Retention Time

Retention Time Deviation

0 200 400 600 800

The situation for the MTBLS59 data is presented in Fig. 2 which

Thessaloniki, Greece Georgios A. Theodoridis

1 Metabolic Profiling: Status, Challenges, and Perspective�� 3

5 HILIC-MS/MS Multi-Targeted Method for Metabolomics Applications�� 65

Fig. 3 Representation of a 3D mass chromatogram obtained from the reversed-

2.3 Chromato- Chromatographic analysis can be performed on a HSS T3 C18

s o-called data preprocessing. The objective of this phase is to sum-