Reiher M. Atomistic Approaches in Modern Biology (2007)

268
Topics in Current Chemistry
Editorial Board:
V. Balzani A. de Meijere K. N. Houk H. Kessler J.-M. Lehn
S. V. Ley S. L. Schreiber J. Thiem B. M. Trost F. Vgtle
H. Yamamoto

Recently Published and Forthcoming Volumes
Bioactive Conrmation I
Volume Editor: Peters, T.
Vol. 272, 2007
Biomineralization II
Mineralization Using Synthetic Polymers and
Templates
Volume Editor: Naka, K.
Vol. 271, 2007
Biomineralization I
Crystallization and Self-Organization Process
Volume Editor: Naka, K.
Vol. 270, 2007
Novel Optical Resolution Technologies
Volume Editors:
Sakai, K., Hirayama, N., Tamura, R.
Vol. 269, 2007
Atomistic Approaches in Modern Biology
From Quantum Chemistry
to Molecular Simulations
Volume Editor: Reiher, M.
Vol. 268, 2006
Radicals in Synthesis I
Methods and Mechanisms
Volume Editor: Gansuer, A.
Vol. 263, 2006
Molecular Machines
Volume Editor: Kelly, T. R.
Vol. 262, 2006
Immobilisation of DNA on Chips II
Volume Editor: Wittmann, C.
Vol. 261, 2005
Immobilisation of DNA on Chips I
Volume Editor: Wittmann, C.
Vol. 260, 2005
Prebiotic Chemistry
From Simple Amphiphiles to Protocell Models
Volume Editor: Walde, P.
Vol. 259, 2005
Supramolecular Dye Chemistry
Volume Editor: Wrthner, F.
Vol. 258, 2005
Glycopeptides and Glycoproteins

Synthesis, Structure, and Application
Volume Editor: Wittmann, V.
Vol. 267, 2006
Molecular Wires
From Design to Properties
Volume Editor: De Cola, L.
Vol. 257, 2005
Microwave Methods in Organic Synthesis

Volume Editors: Larhed, M., Olofsson, K.
Vol. 266, 2006
Low Molecular Mass Gelators

Design, Self-Assembly, Function
Volume Editor: Fages, F.
Vol. 256, 2005
Supramolecular Chirality
Volume Editors: Crego-Calama, M.,
Reinhoudt, D. N.
Vol. 265, 2006
Radicals in Synthesis II
Complex Molecules
Volume Editor: Gansuer, A.
Vol. 264, 2006
Anion Sensing
Volume Editor: Stibor, I.
Vol. 255, 2005
Organic Solid State Reactions
Volume Editor: Toda, F.
Vol. 254, 2005
Atomistic Approaches in Modern Biology

From Quantum Chemistry to Molecular Simulations
Volume Editor: Markus Reiher
With contributions by
L. Bertini P. G. Bolhuis M. Bruschi C. Dellago
M. Dittrich P. Fantucci L. de Gioia C. Greco C. Herrmann
B. Kirchner F. Neese W. Reckien M. Reiher K. Schulten
H. M. Senn S. Sinnecker J. Thar W. Thiel J. Yu
G. Zampella
123
The series Topics in Current Chemistry presents critical reviews of the present and future trends in
modern chemical research. The scope of coverage includes all areas of chemical science including
the interfaces with related disciplines such as biology, medicine and materials science. The goal of
each thematic volume is to give the nonspecialist reader, whether at the university or in industry,
a comprehensive overview of an area where new insights are emerging that are of interest to a larger
scientic audience.
As a rule, contributions are specially commissioned. The editors and publishers will, however, always
be pleased to receive suggestions and supplementary information. Papers are accepted for Topics in
Current Chemistry in English.
In references Topics in Current Chemistry is abbreviated Top Curr Chem and is cited as a journal.
Visit the TCC content at springerlink.com
ISSN 0340-1022
ISBN-10 3-540-38082-5 Springer Berlin Heidelberg New York
ISBN-13 978-3-540-38082-5 Springer Berlin Heidelberg New York
DOI 10.1007/978-3-540-38085-6
This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microlm or in any other way, and storage in data banks. Duplication of
this publication or parts thereof is permitted only under the provisions of the German Copyright Law
of September 9, 1965, in its current version, and permission for use must always be obtained from
Springer. Violations are liable for prosecution under the German Copyright Law.
Springer is a part of Springer Science+Business Media
springer.com
c Springer-Verlag Berlin Heidelberg 2007

The use of registered names, trademarks, etc. in this publication does not imply, even in the absence
of a specic statement, that such names are exempt from the relevant protective laws and regulations
and therefore free for general use.
Cover design: WMXDesign GmbH, Heidelberg
Typesetting and Production: LE-TEX Jelonek, Schmidt & Vckler GbR, Leipzig
Printed on acid-free paper 02/3100 YL 5 4 3 2 1 0
Volume Editor
Prof. Dr. Markus Reiher
Laboratorium fr Physikalische Chemie
ETH Zrich
Hoenggerberg Campus, Wolfgang-Pauli-Str.10
8093 Zrich, Switzerland
markus.reiher@phys.chem.ethz.ch
Editorial Board
Prof. Vincenzo Balzani
Prof. Jean-Marie Lehn
Dipartimento di Chimica G. Ciamician

University of Bologna
via Selmi 2
40126 Bologna, Italy
vincenzo.balzani@unibo.it
ISIS
8, alle Gaspard Monge
BP 70028
67083 Strasbourg Cedex, France
lehn@isis.u-strasbg.fr
Prof. Dr. Armin de Meijere
Prof. Steven V. Ley
Institut fr Organische Chemie

der Georg-August-Universitt
Tammanstr. 2
37077 Gttingen, Germany
ameijer1@uni-goettingen.de
University Chemical Laboratory

Lenseld Road
Cambridge CB2 1EW
Great Britain
Svl1000@cus.cam.ac.uk
Prof. Dr. Kendall N. Houk
Prof. Stuart L. Schreiber
University of California
Department of Chemistry and
Biochemistry
405 Hilgard Avenue
Los Angeles, CA 90024-1589
USA
houk@chem.ucla.edu
Chemical Laboratories
Harvard University
12 Oxford Street
Cambridge, MA 02138-2902
USA
sls@slsiris.harvard.edu
Prof. Dr. Joachim Thiem

Prof. Dr. Horst Kessler
TU Mnchen
Lichtenbergstrae 4
86747 Garching, Germany
kessler@ch.tum.de

Universitt Hamburg
Martin-Luther-King-Platz 6
20146 Hamburg, Germany
thiem@chemie.uni-hamburg.de
VI
Editorial Board
Prof. Barry M. Trost
Prof. Dr. Hisashi Yamamoto
Department of Chemistry
Stanford University
Stanford, CA 94305-5080
USA
bmtrost@leland.stanford.edu
Department of Chemistry
The University of Chicago
5735 South Ellis Avenue
Chicago, IL 60637
USA
yamamoto@uchicago.edu
Prof. Dr. F. Vgtle

Kekul-Institut fr Organische Chemie
und Biochemie
der Universitt Bonn
Gerhard-Domagk-Str. 1
53121 Bonn, Germany
voegtle@uni-bonn.de

Also Available Electronically
For all customers who have a standing order to Topics in Current Chemistry,
we offer the electronic version via SpringerLink free of charge. Please contact
your librarian who can receive a password or free access to the full articles by
registering at:
springerlink.com
If you do not have a subscription, you can still view the tables of contents of the
volumes and the abstract of each article by going to the SpringerLink Homepage, clicking on Browse by Online Libraries, then Chemical Sciences, and
nally choose Topics in Current Chemistry.
You will nd information about the
Editorial Board
Aims and Scope
Instructions for Authors
Sample Contribution
at springer.com using the search function.
Preface
This volume of Topics in Current Chemistry presents an overview of atomistic

theoretical methods applied to molecular biological systems. It thus repesents
a bottom-up view of chemistry on biology from a theoretical perspective. The
chapters are arranged such that important issues are considered starting from
a quantum mechanical perspective and proceeding to a molecular mechanics
and molecular dynamics description of the motion of the elementary particles
involved (i.e., of electrons and atomic nuclei and then of atoms and molecules),
which are responsible for the properties and function of biomolecules.
Depending on the length and time scales relevant for a given phenomenon
to be investigated, tailored theoretical methods are required to account for
these. If one is interested in large scale motions of molecules, a molecularmechnanics-based description will be appropriate. If, however, chemical reactions at local active sites within extended supramolecular ensembles such as
metalloproteins shall be described, a quantum chemical descriptions of electrons and atomic nuclei is most appropriate as such a type of modelling does
automatically adjust to any given chemical situation.
While the emphasis in each chapter is on the applicability and usefulness of
the various theoretical approaches for a given biological system, the authors
made an effort to also provide brief introductions to the foundations of these
approaches. It is hoped that each introduction helps to understand the origin
of the particular method under consideration. For instance, the rst chapter
provides a brief introduction to density functional theory and time-dependent
density functional theory in the appendix.
This volume is organized in a bottom-up spirit and starts with a review by
De Gioia et al. on how chemical reactions at the active site of the hydrogenase
metalloenzyme can be analyzed with quantum chemical methods, i.e., with
those based on the fundamental laws of quantum mechanics.
Quantum chemistry can also provide detailed knowledge on molecular
properties as probed by various spectroscopic techniques. These options are
discussed in the following two chapters. The rst of these by Sinnacker and
Neese deals with theoretical prediction and interpretation of resonance spectra
obtained, for instance, from electron spin resonance (ESR), nuclear magnetic
resonance (NMR), electronnuclear double resonance (ENDOR) as employed
in studies on active sites of metalloenzymes.
Preface
The second one by Herrmann and Reiher discusses techniques of vibrational

spectroscopy with a focus on how information on local structures within large
aggregates of molecules such as proteins can be extracted selectively based on
local vibrations or through the selective process of intensity uptake in different
vibrational spectroscopy techniques.
While the rst three chapters deal with the basic quantities provided by
quantum chemistry, namely (stationary) structures, properties and energetics in the electronic ground and excited states, the fourth chapter by Kirchner and collaborators introduces the reader to concepts of how to add the
time-dimension to quantum chemical studies. First-principles dynamics is
introduced in order to describe the Newtonian motion of atomic nuclei on
a potential energy surface, which is available on the y from the electronic
structure.
The coupling of quantum chemical and Newtonian mechanics methods is
subject of chapter 5 by Senn and Thiel. These so-called QM/MM methods aim
at a combination of the best of both worlds. Of course, since the elements of
both theories are by and large incompatible (compare the notions of orbitals
and positions) a coupling of both can be achieved via the general concept of
energy dened in both worlds.
A notorious problem of molecular dynamics is, however, the fact that the
simulation time of a single reactive aggregate is usually much too small to
observe spontaneous events that require a nonnegligible activation barrier.
To overcome this problem, simulation techniques for rare events have been
developed. The sixth chapter by Dellago and Bulthuis concentrates on one of
these techniques called transition path sampling.
Then, chapter 7 by Schulten and collaborators focuses on molecular mechanics only. Here, the entities described are atoms, united atoms or functional
groups. This reduces the number of degrees of freedom and allows one to study
the dynamical behaviour of huge protein complexes like ATPase.
Finally, I would like to thank R. Mastalerz for his help concerning formating
issues of this volume.
Zrich, July 2006
M. Reiher
Contents
Quantum Chemical Investigations of Reaction Paths

of Metalloenzymes and Biomimetic Models
The Hydrogenase Example
L. Bertini M. Bruschi L. de Gioia P. Fantucci C. Greco G. Zampella
Theoretical Bioinorganic Spectroscopy

S. Sinnecker F. Neese . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
First-Principles Approach to Vibrational Spectroscopy of Biomolecules

C. Herrmann M. Reiher . . . . . . . . . . . . . . . . . . . . . . . . . .
85
CarParrinello Molecular Dynamics Simulations

and Biological Systems
J. Thar W. Reckien B. Kirchner . . . . . . . . . . . . . . . . . . . . . 133
QM/MM Methods for Biological Systems
H. M. Senn W. Thiel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Transition Path Sampling Simulations of Biological Systems
C. Dellago P. G. Bolhuis . . . . . . . . . . . . . . . . . . . . . . . . . . 291
PcrA Helicase, a Molecular Motor Studied
from the Electronic to the Functional Level
M. Dittrich J. Yu K. Schulten . . . . . . . . . . . . . . . . . . . . . . . 319
Author Index Volumes 251271 . . . . . . . . . . . . . . . . . . . . . . 349
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Top Curr Chem (2007) 268: 146

DOI 10.1007/128_2006_080
Springer-Verlag Berlin Heidelberg 2006
Published online: 19 October 2006
Quantum Chemical Investigations of Reaction Paths

of Metalloenzymes and Biomimetic Models
Luca Bertini1 Maurizio Bruschi2 Luca de Gioia1 (u) Piercarlo Fantucci1
Claudio Greco1 Giuseppe Zampella1
1 Department
of Biotechnology and Biosciences,

Universita degli Studi di Milano-Bicocca, Piazza della Scienza, 2, 20126 Milan, Italy
Luca.degioia@unimib.it
2 Department of Environmental Sciences, Universita degli Studi di Milano-Bicocca,
Piazza della Scienza, 2, 20126 Milan, Italy
1
Investigation of Metalloenzymes with Quantum Chemical Methods . . . .
2
2.1
2.2
The Hydrogenase Example . . . . . . . . . . . . . . . . . . . . . . . . . . .

[NiFe] and [Fe] Hydrogenases . . . . . . . . . . . . . . . . . . . . . . . . .
Biomimetic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
5
7
3
3.1
3.2
. . . . . . . . .
. . . . . . . . .
9
9
. . . . . . . . .
10
3.4
Modeling Reactivity . . . . . . . . . . . . . . . . . . . . . .
Optimization of Molecular Structures . . . . . . . . . . . .
Investigation of Electronic Properties and Relative Stability
of Active Site Models . . . . . . . . . . . . . . . . . . . . .
Theoretical Investigations Related to the Reactivity
of Bio-Inspired Coordination Compounds . . . . . . . . .
Theoretical Investigations of the Enzyme Catalytic Cycle .
. . . . . . . . .
. . . . . . . . .
12
15
4
4.1
4.2
Modeling the Full H-Cluster . . . . . . . . . . . . . . . . . . . . . . . . . .

The Broken Symmetry Approach . . . . . . . . . . . . . . . . . . . . . . . .
Application of the Broken Symmetry Approach to the Entire H-Cluster . .
17
17
21
5
5.1
5.2
Modeling Photochemical Reaction Paths . . . . . . . . . . . . . . . . . . .

Introduction to Time-Dependent Density Functional Theory (TDDFT) . .
TDDFT Modeling of CO Photolysis
in the Carbon Monoxide Inhibited Form of [Fe] Hydrogenase . . . . . . .
26
26
Conclusions and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . .
32
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
Outline of Density Functional Theory . . . . . . . . . . . . . . . . . . . . .
33
Outline of Time-Dependent Density Functional Theory . . . . . . . . . . .
38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
3.3
28
Abstract Quantum chemical methods allow one to investigate chemical aspects that are
often difcult to evaluate using only experimental approaches. In particular, the continuous increase in reliability and speed of quantum chemical methods has recently allowed
L. Bertini et al.
the investigation of very complex molecular systems, such as biological macromolecules.

In this contribution, we present applications of quantum chemical methods to the investigation of reaction paths of metalloenzymes and related biomimetic models, using
hydrogenase models as a reference case. In particular, we discuss several examples from
the literature, emphasizing the possibilities (and limitations) offered by present theoretical approaches to study structures, electronic properties and reactivity of metalloenzyme
models. Some relevant aspects which have not yet been fully explored using theoretical
methods, such as the role of antiferromagnetic coupling and photochemical reactions in
[Fe] hydrogenases, are treated in more detail, with presentation and discussion of original
data recently obtained in our laboratory.
Keywords Coordination compounds DFT Hydrogenases Metalloenzymes
Quantum chemistry
Abbreviations
B3LYP Becke3-Lee-Yang-Parr DFT functional
BP86
Becke-Perdew 1986 DFT functional
CI
Conguration interaction
CIS
Conguration interaction singles
DFT
Density functional theory
dppe
1,2-bis(diphenylphosphino)ethane)
DTMA Di(thiomethyl)amine
EDT
1,2 Ethanedithiolate
EPR
Electron paramagnetic resonance
Iron atom of the binuclear cluster proximal to the [Fe4 S4 ] cluster in [Fe] hydroFep
genases
Fed
Iron atom of the binuclear cluster distal to the [Fe4 S4 ] cluster in [Fe] hydrogenases
G2
Gaussian-2 molecules set
GGA
Generalized gradient approximation
HF
HartreeFock
HOMO Highest occupied molecular orbital
LUMO Lowest unoccupied molecular orbital
KS
KohnSham
LDA
Local density approximation
MP2
MllerPlesset second order perturbation method
MOs
Molecular orbitals
CT
Charge transfer
o-xyldt Orto-xylenedithiolate
PBE
PerdewBurkeErnzerhof DFT functional
PDT
1,3-Propanedithiolate
RI
Resolution of identity
TDA
TammDancoff approximation
TDDFT Time-Dependent density functional theory
TDHF Time-dependent HartreeFock
TZVP A triple-zeta basis set
VWN
VoskoWilkNusair DFT functional
ZORA Zero-order regular approximation
BS
Broken symmetry
Quantum Chemical Investigations of Metalloenzymes
HS
High spin
COSMO Conductor-like screening model
1
Investigation of Metalloenzymes with Quantum Chemical Methods
Enzymes are interesting molecules not only due to their crucial biochemical and medical relevance, but also because the elucidation of the molecular
properties responsible of their often remarkable catalytic activity can drive
the design and synthesis of bio-inspired catalysts with potential technological applications. Several enzymes are known to bind one or more transition metal ions, which often play a key role in the catalytic mechanism.
The biological role of transition metals has, in turn, stimulated the synthesis and characterization of coordination compounds featuring structural
and/or functional features related to metalloenzymes and several examples
are known in which the investigation of biomimetic complexes has complemented the elucidation of the properties of the corresponding metalloenzymes [1].
The tremendous increase in computer power, coupled to the renement
of theories and algorithms, has allowed in recent years the application of
quantum chemical methods to the investigation of complex molecular systems containing atoms from almost all the periodic table, thereby disclosing
the possibility to investigate the reactivity of metalloenzymes and biomimetic
models using theoretical methods [27].
Present quantum chemical methods are suited to compute structures and
relative energies of reactants, products and intermediate species, as well as
transition states, allowing the dissection of reaction paths. Moreover, the possibility to characterize structural and electronic properties of both ground
and excited states gives the opportunity to investigate photochemical reactions and spectroscopic properties, which can often be directly compared
with experimental data.
In this contribution, the possibilities offered by quantum chemical
methods to the investigation of reaction paths of metalloenzymes and related
biomimetic models are presented using hydrogenases as reference example.
In fact, almost all theoretical investigations of coordination compounds related to the active site of hydrogenases have been carried out using quantum
chemical approaches based on Density Functional Theory (DFT) [811],
which is based on the fundamental theorem that proves the existence of
a functional of the electron density that contains all energy contributions,
including the so-called correlation contributions, which are related to the mutual interaction of groups of electrons [12]. For a thorough description of
DFT, the reader is referred to the seminal book of Parr and Yang [13], whereas
a short primer can be found here in Appendix I.
L. Bertini et al.
The different description of correlation effects is one of the fundamental

differences between DFT and the conventional HartreeFock (HF) theory. In
fact, the HF method neglects correlation effects because each electron is supposed to move in the mean eld provided by all other N-1 electrons, even
though the HF energy expression includes an exact term for the exchange
energy, which could be considered as a sort of correlation for electrons of
equal spin. In the past, several methods have been proposed and used to
introduce correlation effects via post-HF corrections, all of them including
to different extent the conguration interaction (CI) [14], which, in turn, is
based on the idea of approximating the exact wavefunction (in the limit of
the adopted basis set) building up a combination of N-electron conguration
state functions.
The HF method, among the so-called ab initio approaches, provides
the cheapest way to obtain a molecular wave function, whereas the CI expansion (in all possible variants) is usually extremely time consuming. On
the contrary, the computational time required by DFT calculations is only
slightly longer than for HF, giving at the same time a one-electron structure similar to HF, thus offering a straightforward interpretation of the results while incorporating the correlation effects. Moreover, the description
of open-shell systems is more balanced in DFT than in the HF, which generally enhances the importance of high-spin congurations (that are characterized by the highest exchange contributions). Finally, when open-shell
systems are treated at the unrestricted level (different MOs for different
spin) the DFT solution is generally less contaminated by high multiplicity
contributions than the HF one. DFT methods can also give excellent results for molecular systems characterized by near-degenerate states, which
are generally not adequately described by ab initio mono-determinantal
perturbation theories (for example MP2). However, limitations affect also
DFT methods. The accuracy of DFT results cannot be increased systematically, as it occurs with ab initio approaches based on variational principle,
which ensures that large CI expansions on large basis sets of N-electron
state functions can be arbitrarily close to the (non-relativistic) exact solution. Other limitations of the DFT approach are related to the adopted
exchange-correlation functional, for which the exact form is unknown. As
a consequence, functional reliability can be only established in a heuristic way by comparing DFT results with experiments or extremely accurate
results obtained from highly correlated ab initio methods. In this context,
coordination compounds represent a challenging case because very accurate ab initio results are often unavailable and also experimental results may
be scarce. In spite of the above limitations, DFT has become a very useful tool for the investigation of models of metal-containing proteins. In fact,
functionals based on GGA and so called hybrid methods, in which a portion of HartreeFock exchange energy is added to the exchange-correlation
energy, give accuracies similar or higher than ab initio MP2 methods. In add-
ition, efcient approaches developed to calculate the Coulomb energy, such

as the RI approximation [1518], have further improved the computational
efciency.
The hybrid three parameter B3LYP and the pure BP86 functionals have
been widely used in computational studies of models of the active site of enzymes. The hybrid B3LYP functional [1922], which was originally calibrated
on the G2 database of organic molecules, is now extensively used to study
also metal-containing molecules, due to increasing evidence that B3LYP can
predict accurately properties like bond dissociation energies and molecular
geometries when coupled to an appropriate basis set. Similar considerations
hold true for BP86 [23, 24]. Reaction energies can be generally evaluated
with a reasonable accuracy, even if they can be affected by the choice of the
exchange-correlation functional and basis sets [2, 25]. Other relevant experimental observables such as spin densities, EPR hyperne coupling constants,
g-tensors and vibrational frequencies can be presently computed within the
DFT framework with a sufcient accuracy to allow comparison with the corresponding experimental data [26].
2
Hydrogenases, which are widespread in prokaryotes and found also in some
unicellular eukaryotes, are enzymes that catalyze the reversible oxidation of
dihydrogen and therefore play a fundamental role in energy metabolism [26].
Most hydrogenases can catalyze the reaction in both directions in vitro, although in vivo they are usually involved either in H2 uptake or H2 evolution, depending on the metabolic state of the organism. Studies of isotope
exchange led to the conclusion that dihydrogen cleavage catalyzed by hydrogenases follows an heterolytic reaction path, which implies formation of
intermediate hydride species [27].
Experimental and theoretical studies of hydrogenases are not only driven
by the biological relevance of these enzymes [28, 29], but also by the possibility to cheaply produce large amount of hydrogen gas, which is a very
promising future energy carrier [30].
2.1
[NiFe] and [Fe] Hydrogenases
The two hydrogenase families more thoroughly investigated so far are [NiFe]
and [Fe] hydrogenases, which are phylogenetically distinct and named according to the metal ions bound in their active site [3139]. Remarkably,
another evolutionary distinct family of hydrogenases has been recently found
to contain an iron cofactor of functional importance [40].
L. Bertini et al.
The structural organization of [NiFe] and [Fe] hydrogenases, as well as

several peculiar features of the metal cofactors involved in H2 activation, have
been disclosed by X-ray diffraction and spectroscopic studies [4148]. In the
active site of [NiFe] hydrogenases a nickel ion is bound to the protein by four
cysteine residues, two of which are also coordinated to an iron ion (Fig. 1).
IR spectroscopy studies revealed that the iron ion binds also two CN and
one CO ligands [49]. In the catalytically inactive form of the enzyme an additional oxygen-containing ligand, which has been recently characterized as
a hydroperoxide moiety, bridges the two metal ions [50]. Several redox states
for [NiFe] hydrogenases have been characterized by kinetic and spectroscopic
studies. Four correspond to paramagnetic forms (usually referred to as NiA, Ni-B, Ni-C, Ni-L). The Ni-A form is slowly activated in the presence of
H2 , whereas Ni-B undergoes quick activation. The Ni-C form, which is two
electrons more reduced than Ni-B, is catalytically active and upon illumination forms the Ni-L species, which is stable at temperature below 100 K. EPR
silent forms (Ni-SU, Ni-SI and Ni-R) have also been identied and might be
involved in the catalytic cycle [51].
An unusual [Fe6 S6 ] cluster, referred to as H-cluster and formed by a classical [Fe4 S4 ] cluster bridged by a cysteine residue to a binuclear subcluster
([2Fe]H ), is bound in the active site of [Fe] hydrogenases (Fig. 2).
The two iron atoms of the [2Fe]H subcluster are coordinated by CO and
CN ligands and by a chelating S X3 S moiety, which has been proposed
to be either di(thiomethyl)amine (DTMA) or 1,3-propanedithiolate (PDT).
Spectroscopic investigations of [Fe] hydrogenases are consistent with a +2 oxidation state for the [Fe4 S4 ] cluster, both in the oxidized and reduced forms
Fig. 1 Structure of the bimetallic cluster found in the active site of [NiFe] hydrogenases
Fig. 2 Structure of the H-cluster found in [Fe] hydrogenases
of the H-cluster [52, 53]. Three redox states for the [2Fe]H subcluster have
been characterized spectroscopically [5456]. In the fully oxidized and fully
reduced forms the bimetallic cluster is EPR silent, whereas the partially oxidized form is paramagnetic. Moreover, in the oxidized inactive form of the
enzyme a CO group bridges the iron atoms of the [2Fe]H cluster and a coordination site on the iron atom (Fed ) distal relative to the cysteine residue
is occupied by an oxygen containing species which is displaced by CO in the
CO-inhibited form of the enzyme [57]. In addition, spectroscopic data are
consistent with movement of the -CO group from bridging to terminal position upon reduction of the enzyme [58].
2.2
Biomimetic Models
The possibility of using molecular hydrogen as a clean energy carrier in
the next future is stimulating the development of novel, cheap and ecocompatible processes for H2 production and oxidation. In this context, insight
into electronic and structural features of [NiFe] and [Fe] hydrogenases active
sites has driven the synthesis of coordination compounds that could mimic
the catalytic properties of the enzymes [5963] (Fig. 3).
The coordination compound [Fe2 (S2 C3 H6 )(CO)6 ] is the simplest synthetic
model of the [2Fe]H binuclear subsite found in [Fe] hydrogenases [6466].
Further synthetic efforts led to model complexes characterized by cyanide
and/or phosphine ligands [62, 67, 68] (Fig. 4). Recently Pickett and collaborators reported the synthesis of a exa-carbonyl Fe6 S6 complex closely resem-
L. Bertini et al.
bling the structure of the entire H-cluster [69], while other research groups
obtained dinuclear model containing isonitriles [70], carbenes [71] and new
DTMA derivatives [72]. Notably, most of the above-mentioned model compounds proved to be able to catalyze the electrochemical reduction of protons [73, 74].
Complexes resembling the cofactor of [NiFe] hydrogenases were initially
synthesized by Darensbourg [61] and Pohl [75]. NiFe complexes resembling
the enzyme active site were also reported by Evans and coworkers [76, 77], by
the Schroders group [7880] and by Sellmann and collaborators [81] (Fig. 5).
It is important to underline that the characterization of the chemical properties of synthetic models of hydrogenases has greatly beneted by the application of quantum chemical methods (see below). In this context, theoretical
studies have been mainly used to characterize transiently formed species,
evaluate the relative stability of isomers, dissect reaction mechanisms and
propose the synthesis of novel coordination compounds that are expected to
show catalytic activity.
Fig. 3 Structure of the simplest synthetic model of the [2Fe]H cluster
Fig. 4 Structures of recently reported synthetic models of the [2Fe]H cluster
Fig. 5 Structures of some functional and structural synthetic models of the [NiFe] hydrogenase bimetallic cluster
3
Modeling Reactivity
3.1
Optimization of Molecular Structures
The ability of quantum chemical methods to accurately reproduce the experimental geometry of coordination compounds is a fundamental prerequisite for the investigation of reactivity. While all quantum chemical methods
generally perform very well when used to predict the structure of organic
molecules, their performance can be less satisfactory when dealing with coordination compounds [7]. It is known that HF calculations generally lead
to too long metalligand bonds, whereas MP2 calculations generally overcorrects for the HF error when considering coordination compounds. Therefore,
CCSD(T) calculations are often necessary to obtain accurate ML distances.
The comparison of experimental and computed ML bond distances reveals
that, among DFT methods, those adopting the GGA scheme generally lead to
results in very close agreement with experiment. However, it has been noted
that GGA methods tend to overestimate ML bond lengths when L = halide or
chalconide. Bond distances obtained with the hybrid functional B3LYP generally compare well with those obtained with GGA methods [7].
As a consequence of the complexity of hydrogenases and experimental
limitations, there has been great interest in applying theoretical methods,
particularly density functional theory, to investigate structures of models of
the different redox states of hydrogenases. Most calculations reported so far
have been carried out to investigate models inspired to the active site of [Fe]
hydrogenases, for which more experimental data to be used as reference are
available. The theoretical investigation of the structural features of coordination compounds related to [Fe] hydrogenases has been carried out almost
exclusively using DFT methods. In particular, most results have been obtained using the pure BP86 and the hybrid B3LYP functionals. Bond distances
are generally reproduced accurately by both functionals. However, Ni S
distances are systematically overestimated by about 0.05
A by B3LYP [82],
whereas BP86 structures are affected by slightly smaller deviations when
compared to experimental data [83].
Structures of [Fe(CN)x (CO)y ] complexes have also been recently investigated at the BP86 level of theory, showing that computed Fe CO bond
distances are in very close agreement with X-ray data, whereas Fe CN bond
lengths are generally slightly overestimated [84].
The capability of B3LYP/TZVP and BP86/TZVP to accurately reproduce
experimental structures has been compared also for the coordination compound [Ni(NHPnPr3 )(S3 )], which is related to the active site of [NiFe] hydrogenases. Notably, both functionals reproduce accurately the peculiar attened
tetrahedral geometry of the complex [85].
10
L. Bertini et al.
FT-IR spectra are extremely useful for the characterization of intermediate species and redox states of [NiFe] and [Fe] hydrogenases, as well as of
related model complexes, due to the peculiar vibrational frequencies of the
CN and CO groups. In particular, the calculation of vibrational frequencies and their comparison with experimental data have been used extensively
to validate theoretical protocols and to predict the structural features of intermediate species relevant to [NiFe] and [Fe] hydrogenase chemistry (see
below). Note that the computation of vibrational frequencies is also necessary
to check the curvature of the potential energy surface around the stationary point, in order to exactly locate a true minimum (where all vibrational
frequencies are positive) or a transition state (one imaginary vibrational
frequency), and to calculate the zero point energy correction to the pure
electronic energy in order to obtain thermodynamic data. As for the accuracy of computed vibrational frequencies, it is well know that, due to
the harmonic approximation adopted in the calculation of force constants,
ab initio computed vibrational frequencies are systematically higher than
the experimental ones [25]. Therefore, empirical correction factors are often used to improve the agreement with the experiment. DFT vibrational
frequencies are generally more accurate than HF ones and similar to those
obtained with correlated ab initio methods. In the specic case of CO and
CN groups in [NiFe] and [Fe] hydrogenase models, wave numbers obtained
from BP86 calculations with harmonic approximation compare extremely
well with experimental data. Such an excellent agreement may be partially
due to error cancelation effects [86]. In fact, several groups have evaluated
the relationship between computed and experimental CO and CN vibrational frequencies in models related to the hydrogenases active sites, observing excellent correlation between calculated frequencies and experimental
spectra [87, 88].
3.2
Investigation of Electronic Properties and Relative Stability of Active Site Models
The evaluation of the relative stability of isomers, as well as insight into reaction pathways leading to their inter-conversion, is often crucial to address
chemically relevant problems in hydrogenase chemistry.
In general, computed free energy values are obtained from the electronic
SCF energy considering translational, rotational and vibrational contributions to the total partition function, under the assumption that the partition
function can be written as the product of such terms. To evaluate enthalpy
and entropy contributions, the values for temperature and pressure have to
be set according to the conditions that have to be modeled. Moreover, a scaling factor for the SCF wavenumbers is usually employed. Rotations are usually
treated classically and vibrational modes are generally described according to
harmonic approximations.
11
In the eld of coordination compounds related to [Fe] hydrogenases, the

systematic investigation of [Fe(CN)(PMe3 )(CO)]2 (-CO)(S2 Cn H2n ) isomers
was undertaken using DFT to rationalize experimental data [89]. Interestingly, it turned out that the most stable isomer does not correspond to the
crystallographic structure. The cause for the different stability ranking between experiment and DFT is to be ascribed to crystal packing forces, which
could not be taken into account in DFT calculations. In fact, four isomers
of [Fe(CN)(PMe3 )(CO)]2 (-CO)(S2 Cn H2n ) were characterized in solution by
spectroscopic techniques and the properties of the most abundant isomer
well compare with those computed for the most stable isomer predicted by
DFT calculations.
DFT was also used to predict the regiochemistry of protonation in models
of the active site of [Fe] hydrogenases, evaluating the energy difference
among the possible isomers of [HFe2 (S2 C3 H6 )(CN)(CO)4 (PMe3 )] [90].
The comparison between the [2Fe]H cluster in the active site of [Fe] hydrogenases and related synthetic model compounds has stimulated several
theoretical investigations aimed at disclosing the properties of the active
site responsible for the peculiar structure of the enzymatic cluster. In fact,
even though several synthetic models reproduce the main structural features of the [2Fe]H cluster well, one striking difference is evident comparing
Fe(I)Fe(I) synthetic models and the [2Fe]H cluster. The former are characterized by a vacant coordination site along the Fe Fe axis, whereas in the
enzyme a CO group coordinated to Fed approaches the other iron ion and
a vacant coordination site appears on Fed . To shed light on this issue, Darensbourg, Hall and coworkers investigated factors affecting the rotation of the
Fe(CO)3 group in the series of binuclear complexes [Fe2 (-SRS)(CO)6 ] in
which R = EDT, PDT or o-xyldt (orto-xylenedithiolate) [82]. The computation of stable species and transition states showed that Fe(CO)3 rotation
leading to -CO species is easy, leading to the suggestion that the protein environment could ne tune the geometry of the binuclear cluster. Electronic
factors affecting the structure of the [2Fe]H cluster were investigated, using
DFT calculations, also by Bruschi et al., who studied a series of Fe(I)Fe(I)
complexes differing for the nature of the ligand corresponding to the cysteine residue coordinated to the [2Fe]H cluster in the protein [91]. The results
highlight that increased rotation of the Fed (CO)2 (CN) group was observed
going from soft to hard ligands, leading to structures that closely resembled
the [2Fe]H cluster when the ligand is a strong electron-donor group. These
observations suggested that the proximal [Fe4 S4 ] cluster could play a role in
modulating the structural and electronic properties of the [2Fe]H cluster in
the enzyme.
The above-mentioned results nicely show how the investigation of very
simple models is not always a shortcoming of quantum chemical investigations. In fact, the possibility to neglect environmental effects and to investigate series of related compounds (which could be very difcult to obtain
12
L. Bertini et al.
experimentally) can allow one to highlight relationships that otherwise could

be difcult to unravel.
The investigation of structural and electronic properties of Fe(I)Fe(I)
models related to the reduced form of the [2Fe]H cluster has also allowed the
disclosure of key factors affecting H+ binding to the metal cofactor. Darensbourg, Hall and collaborators have used molecular orbital analysis to show
that, as a consequence of rotation of the Fe(CO)3 group, the highest occupied
molecular orbital (HOMO) of the [2Fe]H cluster changes drastically, resulting
in partial disruption of the Fe Fe bond density and development of a charge
dipole along the Fe Fe axis. Moreover, the iron center corresponding to Fed
in the [2Fe]H subcluster becomes more electrophilic [82].
Starting from the observation that most synthetic Fe(I)Fe(I) complexes fail
to reproduce the orientation of diatomic ligands around the iron ions that is
observed in the enzyme active site, and assuming that this difference could
be crucial to explain the different catalytic activity of synthetic and enzymatic systems, Darensbourg, Hall and collaborators have used DFT to design
coordination compounds that should be characterized by structural features
more closely resembling the enzyme active site [92]. To this end they explored
structures, electronic properties, and relative stabilities of a broad series of
binuclear complexes differing for the nature of the dithiolate linker and of the
other ligands.
As for [NiFe] hydrogenase models, DFT was recently used to evaluate the possible formation of a Ni Fe bond in the diamagnetic complex
[(dppe)Ni(-PDT)Fe(CO)3 ], keeping into account the possibility that Ni and
Fe fragments could have open-shell character. With this aim, Schroder, Neese
and colleagues carried out geometry optimizations both for singlet and
triplet states, starting from the assumption that geometry optimization for
ferromagnetic alignment is generally a good approximation to the geometry of antiferromagnetically coupled centers. It turned out that the geometry
of the closed-shell singlet state is in better agreement with experimental
data, providing strong evidence against open-shell character in [(dppe)Ni(PDT)Fe(CO)3 ] and suggesting that a strong two-center, two-electron bond is
formed between Ni and Fe ions [79].
3.3
Theoretical Investigations Related to the Reactivity
of Bio-Inspired Coordination Compounds
Quantum chemical investigations have been widely used to dissect reaction
pathways leading to bio-inspired synthetic models related to the active site
of hydrogenases. Indeed, theoretical approaches can well complement experimental investigations in this eld because it is often difcult to characterize
the structure and electronic properties of short-lived intermediates. In addition, quantum chemical methods may allow one to disclose the steric and
13
electronic factors at the basis of peculiar reactivity properties, as well as to

explore the effect of chemical modications not yet sampled experimentally.
In this context, it is also important to underline that the theoretical investigation of a reaction path not only implies the characterization of intermediate
species but also of transition state structures. The optimization of transition state structures by quantum chemical methods still depends very much
on the chemical intuition of the computational chemist. Several strategies
have been proposed to efciently locate transition states and the reader is
referred to specic texts for a thorough discussion of the topic [2, 25, 93].
In our laboratory, the optimization of transition state structures is generally
carried out according to a procedure based on a pseudo NewtonRaphson
method. Initially, geometry optimization of a guessed transition state structure is carried out constraining the distances corresponding to the reaction
coordinate. Then, vibrational analysis of the constrained minimum energy
structure is carried out and, if one negative eigenmode corresponding to the
reaction coordinate is found, the curvature determined at such a point is used
as starting point in the transition state search. Localization of the transition
state is carried out using an eigenvector-following search: the eigenvectors
in the Hessian are sorted in ascending order, the rst one being that associated with the negative eigenvalue. However, after the rst step the search is
performed by choosing the critical eigenvector with a maximum overlap criterion, which is based on the dot product with the eigenvector followed in the
previous step.
Hall, Darensbourg and collaborators have characterized reaction paths
related to substitution at di-nuclear carbonyl species, shedding light onto
the role of -CO and neighboring groups [82]. In particular, the cyanation
of the complex [Fe2 (-PDT)(CO)6 ] was investigated by DFT, showing that
the CN/CO substitution leading to [Fe2 (SCH2 CH2 CH2 S)(CO)4 (CN)2 ]2 takes
place in two steps characterized by associative mechanism. In particular, DFT
calculations allowed to show that the CN group initially reacts with the less
sterically hindered iron ion, leading to a -CO intermediate species. The rst
cyanation step is followed by ip of the PDT linker, leading to a species in
which the Fe(CO)3 moiety is sterically unhindered. Then, another CN group
reacts with the Fe(CO)3 unit, going through a transition state characterized
by a semibridging CO.
Zampella et al. [83] have used DFT to investigate the mechanism
of cyanation of the complex [Fe2 (CO)5 (MeSCH2 C(Me)(CH2 S)2 )], which
is the rst example of a coordination compound characterized by the
[Fe2 S3 ] ligation observed in the [2Fe]H cluster [94]. According to experimental results, the reaction of [Fe2 (CO)5 (MeSCH2 C(Me)(CH2 S)2 )] with
cyanide takes place at the Fe atom distal to the thioether ligand, leading
to [Fe2 (CO)5 (CN)(MeSCH2 C(Me)(CH2 S)2 )], which spontaneously loses CO.
The resulting complex can react again with CN , leading to a species which
was assigned to [Fe2 (CO)4 (-CO)(MeSCH2 C(Me)(CH2 S)2 )(CN)2 ]2 . How-
14
L. Bertini et al.
ever, peculiar features of the reaction pathways could not be fully addressed
on the basis of experimental results. In particular, the rst cyanation step was
about 10 000-times faster than the corresponding process for [Fe2 S2 ] species.
In addition, the second cyanation step on the [Fe2 S3 ] species was kinetically
hindered and spectroscopic data suggested the formation of long-lived -CO
species. The dissection of the reaction path within the DFT approach allowed
to show that cyanation of [Fe2 (CO)5 (MeSCH2 C(Me)(CH2 S)2 )], which implies
a transition state structure featuring a -CO group, is characterized by a low
energy barrier due to stabilization of the -CO group by the thioether ligand.
In addition, the second cyanation step is kinetically hindered due to deactivation of the Fe ion and hindered rotation of the reacting FeL3 group. Finally,
close correspondence between computed and experimental FT-IR data supported the assignment of [Fe2 (CO)4 (-CO)(MeSCH2 C(Me)(CH2 S)2 )(CN)2 ]2
to a -CO species.
DFT calculations have been used to complement experimental results also
by Rauchfuss, Benard and coworkers, which have shown that the peculiar
structural properties of the coordination compound [Fe2 (SCH2 NMeCH2 S)
(CO)6 ] are a consequence of the balance between two different factors: the
anomeric effect, which favors the axial disposition of the methyl group, and
steric repulsion between the axial methyl and one carbonyl ligand [95]. DFT
was also used to elucidate factors affecting the regiochemistry of protonation in bimetallic carbonyl thiolates, showing that protonation at the Fe Fe
bond and at CN ligands can be modulated by the nature of coligands, occurring more easily at the Fe Fe bond in [Fe2 (PDT)(CN)(CO)4 (PMe3 )] than in
[Fe2 (PDT)(CN)(CO)4 (PH3 )] or [Fe2 (PDT)(CN)(CO)4 (P(OMe)3 )] [90].
DFT calculations have been also carried out to elucidate the catalytic
properties of models of [NiFe]-hydrogenase active site. As an example,
Zampella et al. have recently used DFT to investigate the reaction between H2 and models of the complex [Ni(NHPnPr3 )(S3)] (where S3 =
bis(2-sulfanyl-phenyl)sulde2 ), with the aim of clarifying the most important factors contributing to its remarkable catalytic activity [85]. In fact,
[Ni(NHPnPr3 )(S3)] is one of the few known coordination compounds featuring a Ni thiolate core resembling the enzyme active site and capable of
catalyzing H2 cleavage. The DFT characterization of transient intermediate species and transition states revealed that formation of a non-classical
adduct between [Ni(NHPnPr3 )(S3)] and H2 implies rearrangement of the
Ni coordination geometry, which can take place due to the peculiar strongly
attened tetrahedral geometry of the parent complex. Moreover, the theoretical investigation showed that H2 cleavage is achieved according to an
oxidative addition step followed by hydrogen migration to one of the sulfur
ligands.
15
3.4
Theoretical Investigations of the Enzyme Catalytic Cycle
The most widely studied issue related to hydrogenase chemistry concerns the
investigation of the mechanism of H2 cleavage/production in the active site of
[NiFe] and [Fe] hydrogenases. However, the elucidation of structure-function
relationship in the protein poses formidable problems to the theoretical community. In spite of continuous increase in computing power and development
of efcient computer codes, the quantum chemical investigation of entire
proteins is still a long-term goal. Nevertheless, theoretical investigation of
simpler models of the hydrogenase active site has allowed to shed light on
several key factors relevant to understand the enzyme catalytic cycle.
Binding and activation of H2 on models of the [2Fe]H cluster found in
[Fe] hydrogenases has been investigated by several groups. Prompted by the
observation that Fed has been suggested to be the site of H2 binding and
activation, DFT was used to study H2 heterolytic cleavage taking place on
the model [(H2 )(CO)2 (CN)Fe(-PDT)Fe(CO)(CN)(SMe)] [88, 9699]. Theoretical characterization of several plausible intermediate species, as well as
comparison between computed and experimental IR frequencies, led to the
conclusion that heterolytic cleavage of H2 on a Fe(II)Fe(II) redox form with
concomitant protonation of a S atom or a CN group is kinetically or thermodynamically unfavorable. Notably, when PDT was replaced by DTMA, cleavage of
H2 , mediated by the N atom of DTMA, became kinetically and thermodynamically favorable. Consequently, on the ground of DFT results it was proposed
a catalytic cycle in which binding of H2 takes place at the vacant coordination site of a -CO Fe(II)Fe(I) species. Subsequent one-electron oxidation leads
to a Fe(II)Fe(II) species which is capable to activate H2 , leading to a formal
Fe(II)Fe(II) intermediate characterized by protonated DTMA and terminal
hydride coordination to Fed . Release of 2H+ and 2e closes the catalytic cycle.
DFT was also used to explore other plausible routes of H2 activation catalyzed by models of the [2Fe]H cluster, with the aim of better understanding
the chemistry of synthetic coordination compounds and possibly drive the
design of new catalysts inspired to the active site of [Fe] hydrogenase. In
particular, starting from theoretical [100, 101] and experimental [58] observations indicating that [2Fe]H models characterized by bridged and terminal CO can easily interconvert, Bruschi et al. have investigated H2 cleavage
catalyzed by [(CO)2 (CN)Fe(-PDT)Fe(CO)(CN)(CH3 S)] model complex, according to a path that implies formation of -H intermediates. Results showed
that H2 heterolytic cleavage can take place on Fep , leading to protonation of
a S atom of PDT and concomitant coordination of H to both metal centers. Zhou et al. have reached analogous conclusions [102, 103], showing also
that a -H Fe(II)Fe(II) intermediate cannot be easily reduced because all the
low-lying 3d bonding orbitals of the iron ions are occupied and the LUMO is
characterized by high energy.
16
L. Bertini et al.
Theoretical investigations based on quantum chemical approaches have

been aimed at investigating crucial aspects related to the catalytic cycle of
[NiFe] hydrogenases. Even though most quantum chemical studies published
so far were carried out using DFT, different computational strategies were
used to gain insight into key features of intermediate species formed in the
catalytic cycle [811]. Several groups have evaluated relative stabilities of possible intermediate species, as well as energy barriers associated to heterolytic
cleavage of dihydrogen, to propose plausible catalytic cycles [96, 99, 104106].
Others have compared experimental and computed spectroscopic properties,
such as g tensors and hyperne tensors, to discern among different structures
of the NiFe cluster [107114]. In particular, calculations have been carried
out within the ZORA [115, 116], using both local VWN [117] and gradient
corrected (BP86) functionals. Even though further improvements in the accuracy of computed magnetic properties is needed and might be expected
to arise when simultaneously considering spin-polarization and spin-orbit
coupling [118], theoretical results generally agree reasonably well with experimental data.
Quantum chemical results obtained studying [NiFe] hydrogenase models
converge on many details, such as the proposal that -H species should be
formed along the catalytic cycle. However, many issues, such as the site of
H2 activation, as well as the redox state of the metal ions in the intermediate species remain unclear. Another relevant and still controversial issue
of [NiFe] hydrogenase chemistry concerns the nature of the ground state of
Ni(II) in Ni-SI and Ni-R redox states, which have been proposed to be involved in the catalytic cycle. In fact, several spectroscopic data are compatible
with a diamagnetic (S = 0) ground state for the Ni(II) ion [119, 120]. However, L-edge soft X-ray data have pointed to possible formation of high-spin
(S = 1) Ni(II) species in the catalytic cycle [121, 122]. With the aim of shedding light on this crucial aspect, Hall and coworkers have carried out DFT
calculations, using the B3LYP functional, showing that the Ni coordination
geometry in high spin (triplet) models ts well to the distorted tetrahedral
structure of the bimetallic cluster observed in the enzyme [99]. However,
other aspects related to the relative stability of low and high spin states of
the Ni(II) ion in the catalytic cycle are less easily addressable by DFT calculations. In fact, it has been recently shown that the both hybrid and non-hybrid
functionals can predict the wrong multiplicity of the ground state for some
transition metal complexes [123126]. In the hydrogenase eld, Bruschi et al.
have recently compared the performance of three functionals (BP86, B3LYP
and B3LYP [86, 127]) in reproducing the structural and electronic properties of well-characterized high- and low-spin [Ni(II)S4] complexes [91].
Results led to the conclusion that BP86 is the most suited functional to reproduce the structures of [Ni(II)S4] complexes, whereas the prediction of
the ground state, as well as the relative stability of high and low spin states,
is more problematic and may be dependent on the adopted functional. It
17
was also concluded that both Ni-SI and Ni-R are potential spin-crossover
species.
4
Modeling the Full H-Cluster
4.1
The Broken Symmetry Approach
As discussed in the previous sections, much of the theoretical work carried out in order to understand the mechanism for hydrogen production
in [Fe] hydrogenases has been focused on the [2Fe]H cluster and related
models. However, in the enzyme the [2Fe]H unit is bridged by a cysteine
residue to a [Fe4 S4 ] cluster (Fig. 6), which could play an important role in
the modulation of the stereo-electronic properties of the [2Fe]H cluster, and
consequently, on its catalytic features.
Mssbauer studies carried out on the isolated forms of [Fe] hydrogenases
have shown that the [Fe4 S4 ] cluster is always in the formally diamagnetic oxidation state +2, indicating that the changes in the oxidation state occur only
at the [2Fe]H cluster [52, 53]. However, Fe atoms of the [Fe4 S4 ] cluster acquire
a paramagnetic character due to the exchange coupling between the two subunits of the H-cluster. These experimental observations further support the
hypothesis that the [Fe4 S4 ] cluster could play an important role in modulating redox and magnetic properties of the active site. Furthermore, as known
from the chemistry of other proteins, [Fe4 S4 ] clusters are often involved in
electron transfer. Therefore, transient species in which the [Fe4 S4 ] unit is reduced to the oxidation state +1 should also be considered in order to model
the reactivity of the full H-cluster.
Fig. 6 Schematic representation of the entire H-cluster
18
L. Bertini et al.
The theoretical modeling of the entire H cluster represents a difcult challenge, not only for the size of the system, but also because the [Fe4 S4 ] cluster
is composed by two layers of high spin Fe atoms coupled antiferromagnetically to give a global low spin ground state. For example, in the oxidized
[Fe4 S4 ]2+ cluster the two layers, composed by high spin Fe2+ (S1 = 2) and high
spin Fe3+ (S2 = 5/2) atoms with aligned spins (ferromagnetic interaction)
(Stot = S1 + S2 = 9/2), are coupled antiferromagnetically to give a diamagnetic (S = 0) ground state, while in the reduced [Fe4 S4 ]+ cluster there is one
Fe2+ Fe3+ (S1 = 9/2) and one Fe2+ Fe2+ layer (S2 = 4) coupled antiferromagnetically to give the S = 1/2 ground state. Such spin-coupled systems
are difcult to treat within the DFT approach because their ground state
wavefunctions typically correspond to linear combinations of multiple determinants.
To discuss energies and properties of [Fe4 S4 ] clusters, it is important to
describe the weak antiferromagnetic coupling at the same level of theory of
strong metal-metal and metal-ligand bonds. In the framework of the unrestricted formalism, these exchange interactions can be treated reasonably
well by employing the broken symmetry (BS) approach developed by Noodleman et al. [128, 129]. The BS approach consists in the localization of the
opposite spins on different parts of the molecule to give a mono-determinant
representation of the spin exchange interactions within the molecule. For example, in [Fe4 S4 ] clusters a BS wavefunction can be built by localizing the
electrons on one of the dimeric layers of Fe atoms with spin up and on the
other one featuring spin down. In order to localize the spins on different centers, space and spin wavefunctions are relaxed by reducing their symmetry
with respect to that of the nuclear framework. The BS wavefunction is not
a pure spin state (it is an eigenstate of Sz , but not of S2 ). In this context, the BS
wavefunction can be considered as a low spin wavefunction affected by a large
spin contamination. In spite of this restriction, Noodleman [128, 129] showed
that the BS state is a weighted average of pure-spin wavefunctions (where
the weights are the ClebschGordan coefcients relating the spin quantum
numbers for oppositely aligned spin vectors of the subunits) and that correspondingly the BS state energy is a weighted average of pure spin state
energies. The BS state energy can then be correlated to the spin state energies
derived from the phenomenological Heisenberg Hamiltonian, which for two
spin centers takes the form:

H = J SA SB ,
(1)
where SA and SB are the spin operators on center A and B, respectively, and
J is the exchange coupling constant, which represents the strength of the exchange interaction between the two spin centers. In the following, it will be
assumed that J has negative values for antiferromagnetic interactions, and
positive values for ferromagnetic interactions. In this case, the Heisenberg
Hamiltonian gives the spin state energies

E Stot = 1/2J Stot (Stot + 1) ,
19
(2)
and the difference between the energies of two successive spin states equals to

(3)
E Stot E Stot 1 = JStot ,
where Stot is the total spin ranging from |SA SB | to |SA + SB |. The determination of J from Eq. 2 allows one to construct the full ladder of pure spin states
energies. Notably, the J value can be determined from the BS wavefunction
and the pure high spin wavefunction which is easily calculated by aligning
ferromagnetically the spins of the iron sites. In fact, Noodleman showed that
the expectation values of the spin operator product dened in Eq. 1 can be
derived for the HS and BS states as [129]:
SA SB HS = + SA SB
SA SB BS = SA SB ,
(4)
(5)
from which the difference between HS and BS state energies can be written as:

(6)
E HS; Smax = SA + SB E BS; MS = |SA SB | = 2JSA SB .
Equation 6 can, therefore, be used to derive J, which in turn can then be used
in projecting the BS state energy on any pure spin state energy. In particular, the energy difference between the high spin and the low spin ground state
energy (E(Smax ) E(Smin )) can be derived from Eq. 2 as:

(7)
E Smax E Smin = 1/2J Smax (Smax + 1) 1/2J Smin (Smin + 1) ,
and the difference between this equation and Eq. 6 gives Espin = E(BS)
E(Smin ), which represents the energy correction of the BS state energy to the
pure spin ground state energy.
It should be noted that the use of the Heisenberg Hamiltonian is restricted
to weakly interacting dimers, for which unpaired electrons are well localized
onto the two metal centers. In the case of strongly interacting dimers the unpaired electrons are largely delocalized onto the ligands and E(BS) E(Smin ).
The description of the BS approach given above for a two spin centers system can be easily generalized to a system with more than two spin centers by
introducing the spin Hamiltonian operator

H=
Jij Si Sj ,
(8)
ij
where the summation is extended over all pairs of spin centers i and j bearing spin Si and Sj , and Jij is the exchange coupling constant between them.
In these multiple spin centers systems several BS states can be calculated by
aligning ferromagnetically or antiferromagnetically the spins of different centers. For N spin centers, 2N /2 non-equivalent combinations of spins can be
dened to obtain BS states. The Jij parameters can then be determined from
20
L. Bertini et al.
the HS state and a consistent number of BS states by applying the spin algebra of Eqs. 4 and 5. In order to reduce the number of coupling constants to be
determined, the approximation of restricting the summation in Eq. 8 to adjacent spin centers is generally assumed. Other simplifying assumptions, such
as the dimer approximation [130], have been proposed.
The electronic structure of the [Fe4 S4 (SCH3 )4 ]x cluster, going from the all
ferric (x = 0) to the all ferrous redox states (x = 4), have been extensively investigated by Noodleman et al. [131135]. Spectroscopic studies on [Fe4 S4 ]
clusters in synthetic complexes and in proteins [136140] have shown that
the four Fe atoms generally occur in equivalent pairs, thus supporting the use
of a spin Hamiltonian dened in terms of pairs of centers. For the highest
(x = 0) and lowest oxidation states (x = 4), the two Fe2+ Fe2+ or Fe3+ Fe3+
layers coupled antiferromagnetically are treated as single spin centers with
S = 4 and S = 5, respectively, and the correction for the energy of the diamagnetic pure spin state is readily found by applying the equations given above.
In the case of the [Fe4 S4 (SCH3 )4 ]2 cluster, which is composed by the two
mixed valence Fe3+ Fe2+ dimers, a further complication arises. Mssbauer
studies have shown that the four Fe atoms are equivalent, indicating delocalization of the valence electron distribution to give a proper Fe2.5+ Fe2.5+
oxidation state for each layer. This intralayer delocalization can be described
by introducing a non-Heisenberg term into the spin Hamiltonian operator
which produces a splitting of the spin states [141144]. The spin Hamiltonian
operator takes the form

H = J SA SB B S12 + 1/2 B S34 + 1/2 ,
(9)
where B is the resonance delocalization parameter and S12 and S34 are the
total spin of the two mixed-valence layers. The spin state energies obtained
from this Hamiltonian are given by

(10)
E(S) = 1/2J S(S + 1) B S12 + 1/2 B S34 + 1/2 .
It is interesting to note that the delocalization energy depends on the spin
quantum number (Sij ) of the mixed valence pair and that it increases with
the pair spin. Several authors have shown that the ground spin state of the
system is generally the result of the competition between the Heisenberg and
resonance delocalization of the Hamiltonian [142, 145, 146].
The [Fe4 S4 (SCH3 )4 ] clusters in the oxidation states 1 and 3 are composed by a mixed-valence Fe2.5+ Fe2.5+ layer and by a Fe3+ Fe3+ or
Fe2+ Fe2+ layer, respectively. In these cases a third parameter, which describes the possible interlayer resonance delocalization between the ferrous
or ferric layers and the mixed-valence layer, has to be added to the spin
Hamiltonian operator. This term has the same form of that used for the intralayer delocalization and leads to the expression of the operator

(11)
H = J SA SB B S12 + 1/2 B S + 1/2 ,
and to the corresponding energies

E(S) = 1/2J S(S + 1) B S12 + 1/2 B S + 1/2 ,
21
(12)
where B is the interlayer resonance delocalization parameter and S is the total

spin of the cluster. The values of J calculated for [Fe4 S4 ] clusters in different
redox states ranges from about 100 cm1 in the all-ferrous [Fe4 S4 (SCH3 )4 ]4
cluster to about 900 cm1 in the all-ferric [Fe4 S4 (SCH3 )4 ]0 cluster [133] while
the calculated values of B are in the range 700900 cm1 [132]. For a detailed
explanation of the approaches used for the calculation of the exchange parameters the reader is referred to the extensive work of Noodleman et al. In
particular, analytical expressions for the energy correction from the BS to
the pure spin state for the [Fe4 S4 (SCH3 )4 ] cluster in different redox states are
reported in [132, 133].
4.2
Application of the Broken Symmetry Approach to the Entire H-Cluster
The application of the BS approach to the H-cluster should account not only
for exchange interactions within the [Fe4 S4 ] cluster but also for the exchange
interaction between the [Fe4 S4 ] and [2Fe]H clusters, a requisite that further
complicates the analysis of this large system. Therefore, it is not surprising
that only one theoretical work on the entire H-cluster in combination with
the BS approach has been reported so far [147]. This study was focused on
the active form of the H-cluster [(CH3 S)4 (Fe4 S4 )(-PDT)(Fe2 )(CO)3 (CN)2 ]3
(3a) derived from the X-ray structure of Clostridium pasteurianum (CpI) in
which a CO ligand is bridged between the Fep and Fed atoms [48] (hereafter
the labels will indicate the total charge of the complex and the isomer form
of the [2Fe]H moiety; a for the isomer with a bridged CO ligand and b for the
isomer with all terminal CO ligands. The two isomers will be also referred as
bridged and open, respectively). In 3a the [Fe4 S4 ] cluster was found to feature the formal Fe(II)Fe(II)Fe(III)Fe(III) redox state and the [2Fe]H cluster
the formal Fe(I) Fe(II) redox state. Brunold et al. [147] also investigated the
molecular and electronic properties of the CO-inhibited form of the H cluster [(CH3 S)4 (Fe4 S4 )(-PDT)(Fe2 )(CO)4 (CN)2 ]3 (3a-CO) and of the transient
state generated by irradiation of the CO-inhibited form at cryogenic temperatures (3b). This unstable state results from the photodissociation of the
bridging CO, leading to an isomer of 3a featuring all terminal COs.
As discussed above, the exchange interactions in this paramagnetic active
form of the H cluster can be treated with a three spin model Hamiltonian,
which may be written in the general form as:

(13)
H = J12 S1 S2 + J13 S1 S3 + J23 S2 S3 .
As suggested by Brunold et al. [147] the interaction between the non-adjacent
spin centers 1 and 3 is negligible. This simplify the spin model Hamiltonian to
22
L. Bertini et al.
the form:

H = Jcube SA SB + JH SA SH ,
(14)
where Jcube represents the exchange interaction between the two Fe dimers of
the diamagnetic [Fe4 S4 ] cluster with SA = SB = 9/2 and JH corresponds to the
exchange interaction between the vicinal mixed-valence Fe dimer (A) and the
[2Fe]H cluster with S = 1/2 (Fig. 7). In this model, three different spin congurations of the [Fe4 S4 ] cluster have to be considered, depending on which Fe
atoms are coupled to form the two mixed-valence dimers [147]. In addition,
for each of these congurations the vicinal mixed-valence Fe dimer (A) can be
coupled to the unpaired electron of the [2Fe]H cluster in a ferromagnetic (F)
or antiferromagnetic (AF) way. This leads to two different BS solutions: one
with the unpaired electron of the [2Fe]H aligned with those of the vicinal
mixed-valence Fe dimer (F), and one with opposite spins (AF). Thus, a total
of six different BS congurations have been calculated and the resulting value
for JH has been obtained from (Eq. 6):

(15)
E F; MBS = 5 E AF; MBS = 4 = 2JH 9/2 1/2 = 9/2 JH .
Because of the three different spin congurations of the [Fe4 S4 ] cluster, three
distinct JH values can be obtained. A fourth value may be calculated by the
ferromagnetic and antiferromagnetic coupling of the unpaired electron of the
[2Fe]H cluster with the high spin state of the [Fe4 S4 ] cubane according to:

(16)
E F; MBS = 19/2 E AF; MBS = 17/2 = 2JH (9) 1/2 = (9)JH .
The average value of JH for the complex 3a is equal to + 15 10 cm1 , in very
good agreement with the experimental value of 20 cm1 [52, 53]. Notably, the
calculated value of JH in the CO-inhibited form of the H-cluster 3a-CO (150
50 cm1 ) is signicantly larger than the value calculated for 3a, in satisfactory
agreement with the experimental value of about 100 cm1 . The agreement between exchange parameters derived from theory and experiment shows that
the BS approach within the DFT method is a suitable procedure for complex magnetic systems like those examined here. The computed molecular
and electronic structure of the elusive species 3b is consistent with that suggested on the basis of the few structural and spectroscopic experimental data
available. The structure of the [2Fe]H cluster is, in fact, characterized by all
terminal CO ligands, with the unpaired electron remaining on this moiety,
and the [Fe4 S4 ] maintaining the formal + 2 oxidation state (Fig. 7b).
The results so far discussed assign to the [Fe4 S4 ] cluster a marked inuence on the electronic and molecular properties of the [2Fe]H cluster. In
particular, the unpaired electron is almost fully localized on the Fed atom
in 3a, while the addition of CO to the Fed atom signicantly alters the spin
density within the [2Fe]H cluster, leading to a more extended spin delocalization between the two Fe centers. This electronic structure description
accounts for the different values of the exchange parameter JH . In fact, the
23
Fig. 7 Schematic diagram of the three spin model for the bridged (a) and open (b) forms
of H-cluster. It should be noted that the unpaired electron in the [2Fe]H moiety is localized on the Fed and Fep atom for the bridged and open form, respectively
large amount of spin density on the Fep atom in 3a-CO should facilitate the
exchange interaction with the mixed-valence Fe dimer in the [Fe4 S4 ] cluster
inducing more paramagnetism into the latter, in agreement with Mssbauer
investigations [52, 53]. Notably, in the case of 3b the unpaired electron is almost entirely localized on the Fep atom indicating that transition between the
bridged (3a) and open (3b) isomers provides a transfer of the spin from the
Fed to the Fp atom.
Recently, Bruschi et al. have undertaken the study of the H-cluster as
a function of environment and redox state [148], investigating the 5, 4
and 3 redox states and considering both open and bridged isomers of the
[2Fe]H cluster. In order to explore the effects of the environment on the
24
L. Bertini et al.
electronic and molecular properties of the H-cluster model, geometry optimization and electronic structure calculations have also been carried out on
models soaked in the COSMO polarizable continuum medium [149] characterized by dielectric constants equal to 4 and 40 (from hereafter in the labels
of complexes a subscript is used to indicate the value of the environment),
which are values commonly used to describe protein environment [150152].
Geometry optimisations have been carried out with the TURBOMOLE suite
of programs [153] in connection with the resolution of the identity technique
(RI) [17, 18], using the BP86 functional [23, 24] and an all-electron valence
triple- basis set with polarization functions on all atoms [154]. The starting
structure was taken from the X-ray geometry of the H-cluster of CpI [Fe] hydrogenase [48]. Some relevant computed structural parameters are reported
in Table 1.
In the more reduced complex [(CH3 S)4 (Fe4 S4 )(-PDT)(Fe2 )(CO)3 (CN)2]5
5b, the unpaired electron is localized on the [Fe4 S4 ] cluster (Table 2) suggesting the formal redox state Fe(III)Fe(II)Fe(II)Fe(II) and correspondingly,
the Fe(I)Fe(I) redox state for the [2Fe]H cluster. In the open isomer the
Fed (CO)2 CN group is partially rotated, with a CO ligand oriented toward
the Fep atom. The Fe S Fep angle is equal to 159 degrees, a value signicantly larger than that observed in the X-ray diffraction structure (Table 1).
This could be an indication that the environment has a marked inuence
on 5a; in particular, the Fe S Fep angle is drastically reduced increasing
the dielectric constant of the environment to 4. In 5b40 the effect of the
environment is even more evident and the stable form corresponds to a strucTable 1 Relevant geometrical parameters of complexes 5a3a and 5b3b computed at
BP86/def-TZVP level of theory. The three values from the left refer to the parameters
calculated in gas-phase, and in the polarizable continuum medium with = 4 and 40,
respectively
5a
5b
4a
4b
3a
3b
Expt.
a
in degrees
in
A
c taken from [48]
d taken from [43]
e taken from [58]
b
Fe S Fep a
Fep Fed b
,
142.1, 125.0
159.2, 142.7,
154.4, 141.3, 130.4
153.4, 140.7, 129.1
134.3, 126.2, 121.7
146.8, 133.4, 121.7
118.0 c , 116.4 d , 119.5 e
,
2.601, 2.571
2.732, 2.596,
2.621, 2.599, 2.580
2.662, 2.593, 2.562
2.556, 2.551, 2.542
2.614, 2.588, 2.569
2.617 c , 3.555 d , 2.601 e
25
Table 2 Atomic spin densities of the Fe atoms in the H-cluster of the [Fe] hydrogenase
computed at BP86/def-TZVP level of theory. Fe1,2 and Fe3,4 refer to the Fe atoms in the
[Fe4 S4 ] cluster of the two layers coupled antiferromagnetically; Fep and Fed indicate the
proximal and distal Fe atoms of the [2Fe]H cluster, respectively. The asterisks specify the
atomic spin density of the Fe center coordinated to the bridging S atom
5b1
Fe1,2
Fe3,4
Fep
Fed
5a1
4b1
4a1
3b1
3a1
3.24 , 3.22 3.24 , 3.21 3.04 , 3.07 3.04, 3.06

3.07 , 3.07 3.10, 3.07
2.93, 2.89 2.89, 2.91 3.21, 3.15 3.14, 3.13 3.09, 3.06 3.08, 3.07
0.03
0.01
0.21
0.04
1.21
0.17
0.01
0.03
0.01
0.1
0.09
0.88
ture characterized by a -CO group. The gas-phase geometry optimization

of the [(CH3 S)4 (Fe4 S4 )(PDT)(Fe2 )(-CO)(CO)2 (CN)2 ]5 isomer 5a1 , reveals
that this species is unstable. However, the same complex is stable when the
optimization is carried out in presence of a polarizable continuum medium
5a4 and 5a40 . In this case the electronic structure of the [Fe4 S4 ] cluster is
very similar to that calculated for 5b4 , being still compatible with a formal
Fe(III)Fe(II)Fe(II)Fe(II) redox state and with the Fe(I)Fe(I) redox state of the
[2Fe]H cluster.
After mono-electron oxidation of the fully reduced species to give the 4
redox state of the H-cluster, both the open and bridged isomers have been
characterized in gas-phase, as well as in the polarizable continuum medium.
In both isomers the electron is removed from the [Fe4 S4 ] cluster. This leads to
an electronic structure which is compatible with the Fe(II)Fe(II)Fe(III)Fe(III)
redox state of the [Fe4 S4 ] cluster and with the Fe(I)Fe(I) redox state of the
[2Fe]H cluster. However, in the gas-phase a signicant spin density is localized on the Fep and Fed atoms in the open and bridged forms, respectively, indicating electron delocalization between the [Fe4 S4 ] and the [2Fe]H clusters.
This delocalization is reduced when the H-cluster is soaked in the polarizable
continuum medium. The environment signicantly affects also the geometry,
as observed in the case of the 5 redox state. In particular the Fe S Fep
angle is reduced from about 150 degrees to about 130 degrees, when moving from the gas-phase to the = 40. The molecular geometries of the [2Fe]H
cluster in the open and bridged isomers are very similar to those calculated
for the 5 redox state, indicating that the oxidation state of the [Fe4 S4 ] cluster
has only a minor effect on the geometry of the [2Fe]H cluster.
A further mono-electron oxidation of the [(CH3 S)4 (Fe4 S4 )(-PDT)(Fe2 )
(CO)3 (CN)2 ]4 complex leads to the 3 redox state, which corresponds to the
form investigated by Brunold and collaborators. In this case, the inuence of
the environment is less signicant than in the case of 5 and 4 complexes, as
expected due to the reduction of the total charge. However, also in this case
26
L. Bertini et al.
the Fe S Fep angle decreases from 146.8 to 121.7 degrees when moving
from the gas-phase to the polarizable continuum medium with = 40.
Several interesting observation can be pointed out in the light of the presented results. The large negative charge in the gas phase models leads to
a structure of the [2Fe]H subcluster signicantly different from that observed
experimentally in the protein. In this regard, the better agreement of the molecular structures of the solvated complexes may suggest that the H-cluster
in the protein environment is characterized by a lower effective charge. This
observation is intriguing since a detailed analysis of the protein structure
shows that there are only a few charged residues near the H-cluster, which
are, however, not sufcient to neutralize its negative charge. Furthermore,
this investigation shows that the protein environment can inuence the electronic and molecular characteristic of the binuclear cluster. In particular, the
modulation of the dielectric constant of the protein environment affects the
stability of the bridged and open isomers, providing a possible mechanism of
interconversion between the two forms.
5
Modeling Photochemical Reaction Paths
5.1
Introduction to Time-Dependent Density Functional Theory (TDDFT)
TDDFT has become the reference method for studying the excited state of
medium and large size molecular systems [155]. This approach has the same
advantages and limitations that DFT has in comparison with RHF and correlated methods for the ground state. TDDFT includes electronic correlation
effects through i) the approximated exchange-correlation functionals and
ii) the contribution of double excited congurations, which are not accounted
in the Conguration-Interaction Singles (CIS) method. Nevertheless, TDDFT
is a single-reference method, and therefore can not achieve high accuracy as
high-level correlated methods like multi-reference conguration-interaction
or equation-of-motion coupled-cluster. Finally, TDDFT computational cost
is of the same magnitude as ground state DFT, and when RI approximation
is applied with non-hybrid functionals, one can save extra CPU time up to
10100 times. Appendix II describes shortly the basic theory of TDDFT.
Regarding the vertical excitation energies, as far as low-lying excited states
with negligible double-excitation character are considered, the error that affects TDDFT results is around 0.1 to 0.5 eV, remembering that the chemical
accuracy is 0.05 eV (around 1 Kcal/mol). TDDFT results can be very poor for
excited state of molecular system with extended conjugated bonds [156], as
well as for Rydberg and charge-transfer (CT) excited states [155]. These facts
are mainly related to an unsatisfactory description of virtual orbitals because
27
of the non-correct long-range behavior of the exchange-correlation potentials [157]. The effects of the type of exchange-correlation functional and the
basis set are also a crucial issue in the TDDFT level of theory. As commented
above, the accuracy of TDDFT depends strongly on the type of excited state
considered.
Since we are dealing with biomimetic models of metalloenzymes, it is important to discuss the accuracy of TDDFT for transition metal complexes.
Like ground state DFT, also for TDDFT such systems pose more problems
than organic molecules, because of the multireference character of the wavefunction of the former. Gisbergen et al. [158] present a small benchmark of
excitation energies of transition metal (TM) compounds, from which it results a TDDFT non-uniform accuracy with respect to the high-level methods.
As far as the type of functional is concerned, Hummel et al. [159] claim that
B3LYP plus the inclusion of solvation effects by means of a continuum model
yield accurate excitation energies for M(CO)n carbonyl complexes. A recent
benchmark by Petit et al. [160] clearly shows that excitation energies do not
depend on the basis set and pseudopotential quality, and that the accuracy of
the PBE0 hybrid functional [161] is close to that of high-level methods.
Excited state energy gradients are crucial properties in studying photochemical reaction path, since they allow one to explore excited state potential
energy surfaces, going beyond applications related to the evaluation of the
vertical excitations. Recent laser spectroscopy developments have increased
our knowledge of the excited state properties (normal modes, dipole moment,
etc). With ultra-fast spectroscopy, it is possible in some simple cases to follow
whole photochemical reaction paths [162].
The development of the TDDFT gradient of the excited state energy is
recent. In 1999 Van Caillie and Amos used the so-called Z-vector method
previously developed by Handy and Schaefer [163], to implement LDA [164]
and GGA [165] excited state energy rst derivatives (CADPAC). A few years
later, Furche and Ahlrichs [166] implemented the gradients in the TURBOMOLE program suite [153], using the Lagrangian method [167]. Finally,
within a plane-wave basis set framework, Hutter [168] implemented the TDA
gradient using the Lagrangian methods, while Doltsinis and Kosov [169] used
an implicit differentiation scheme.
The reliability and effectiveness of excited state geometry optimization depends on many factors, as the size and the symmetry of the investigated
system and the topology of the energy surfaces (excited state intersections,
etc). In general, the computational cost of this optimization is similar to that
in the ground state, but unfavorable situations like uctuations of the excited state energies and gradients or singlet-triplet instability [170] can occur
The quality of the TDDFT excited state optimized geometry is promising. In
the case of small systems [166, 171] adiabatic excitation energies and bond
distances are in good agreement with experimental or high-level calculated
values.
28
L. Bertini et al.
5.2
TDDFT Modeling of CO Photolysis
in the Carbon Monoxide Inhibited Form of [Fe] Hydrogenase
The theoretical investigations of photochemical processes involving transition metals at TDDFT level are mainly devoted to the study of the vertical
excitations, with the assignment of the excitation levels, the simulation of the
electronic spectra, etc. The exploration of the excited state potential energy
surfaces (characterization of the stationary points as local/global minima
and rst-order saddle points) is still lacking, since TDDFT gradients have
been only recently developed and implemented in some quantum chemistry suite programs. Moreover, excited states wave functions of coordination
compounds often possesses multireference character, which is unavoidably
poorly described by a single reference methods as TDDFT. Nevertheless, in
the recent literature there are some examples of TDDFT study of photochemical processes in which bio-inorganic systems are involved. For example, in
the case of the CO photodissociation of CO-ligated hemoglobin [172], the
authors have characterized some potential energy surfaces by single-point
excitation energy computations along the Fe CO bond length reaction coordinate. As the authors pointed out, it is reasonable to expect that this picture
would be close to the one obtained from the true minimum-energy path of
hemoglobin. Another interesting example is the study of the photochemical
activation of N2 by dinuclear Ru and Fe complexes [173].
[Fe] hydrogenase gives an intriguing example of a photochemical reaction path which can be investigated by means of TDDFT. When the oxidized
form of the enzyme is exposed to CO, the [2Fe]H cluster binds a further CO
molecule which substitutes the H2 O molecule on the distal Fe atom [174
177]. In this form the enzyme (Hox CO hereafter) is inactive, but the active
form (Hox hereafter) can be restored by illumination with white light at cryogenic temperature. This process gives two different light-induced species as
a function of temperature:
h
Hox CO Hox I + CO
6 14 K
h
Hox CO Hox II + CO
40 K
IR spectra [178] have highlighted that Hox I is the result of the photolysis of
the exogenous CO, while Hox II is characterized by the loss of the CO group
bridging the iron centres.
The theoretical modeling of this photolysis process is still lacking,
but some interesting results have been recently reported in two papers
by Fiedler et al. [147, 179]. In the rst paper the valence excited states
of two biomimetic model of [2Fe]H , namely [Fe2 (S2 C3 H6 )(CO)6 ] and
[Fe2 (S2 C3 H6 )(CO)4 (CN)2 ]2 have been studied. Beside electronic and res-
29
onance Raman spectra simulations, the authors estimate the elongation of

Fe Fe bond length in the rst and third excited state of [Fe2 (S2 C3 H6 )(CO)6 ].
The second paper is devoted to the calculation of different properties on
a model of the full H-cluster in which the [Fe4 S4 ] cluster in explicitly considered. In this paper two models for Hox I and Hox II based on the comparison
between calculated and experimental [174, 177, 180] EPR parameters are also
proposed, corroborating the hypothesis made by Chen et al. [178].
The following section focuses on the possibilities supplied by TDDFT
exploration of the excited state potential energy surfaces in studying photochemical reaction paths, using CO photolysis of Hox CO as an example.
The attempt is to show which kind of information can be obtained from
a TDDFT study on a model system, using geometry optimization of the excite
state geometry structure. Computations are performed using TURBOMOLE
with the BP86 functional and TZVP basis set. RI approximation is used with
auxiliary basis set of the same TZ quality [181]. Excited state structures are
optimized until the excited state energy gradient norm reaches values at lower
A1 .
than 0.01 hartree
The modeling of the CO photolysis mechanism is carried out using the CO
inhibited [2Fe]H cluster biomimetic system [Fe2 (-PDT)(-CO)(CO)6 ] (hereafter a-CO). This binuclear complex has two FeI FeI atoms and terminal CO
groups can be distinguished between the two CO trans to the -CO (one anti
and one syn with respect to the carbon of CH2 group of the -PDT) and
the four CO trans to the -PDT. When a-CO loses a CO molecule one obtains
[Fe2 (-PDT)(CO)6 ] (hereafter a). a-CO and a have S0 1A ground state with
global Cs minimum geometry (Fig. 8).
The analysis of the optimized geometries for a-CO and a reveals some
important differences between the two structures. The Fe-Fe internuclear disA going from a-CO to a. Fe C distances relative
tance increases by 0.448
A and 0.062
A reto the two C atoms of syn and anti CO increase by 0.067
Fig. 8 Structure of a and a-CO ground states. All distances and angles in
A and degree
30
L. Bertini et al.
spectively, while the Fe C O bond angles decrease by 10.1 and 7.4 degrees.
These effects can be rationalized in light of the analysis of the highest occupied molecular orbital (HOMO) of the two model complexes [179]. The
rst bond length increase can be accounted for by observing that in a-CO
a d-orbitals Fe Fe antibonding combination is found, which becomes bonding in a. Regarding the Fe C bond length increase, the HOMO analysis does
not give a clear explanation. In both a-CO and a the HOMO has a d-Fe-p-C
orbital combination with antibonding character, which is stronger for a-CO,
according to the overlap population. This could be one of the reasons of the
observed structural changing.
In the investigation of the valence excited states, we consider the two single
a-CO excited states S1 1A and S3 1A , along with the corresponding states of
a S1 1A and S2 1A . The a excited states have the same electronic state of the
a-CO ones, and the processes
h
a-CO S1 1A a S1 1A + CO

h
a-CO S3 1A a S2 1A + CO

can be considered as two of the possible paths of the hypothetical CO photolysis of a-CO. In Table 3 the main features of the two excited states under
investigation for both systems are reported. The S1 1A and S3 1A d-CO excited states are characterized by the one-electron excitation HOMO LUMO
(37a ) and HOMO LUMO+1 (68a ), respectively. These virtual MOs possess antibonding combinations with respect to both Fe C bonds of two CO
trans to the -CO. In the case of the LUMO + 1 the combination in which
Table 3 Computed energies for a-CO and a (in hartree). Evertical is the energy of the excited state at the geometry of the optimized ground state; for the S0 ground state Evertical
is the energy at the optimized minimum geometry; the leading one-electron excitation
refers to the single-excitation contribution a i; Eopt is the excited state energy at its
own optimized geometry; Gradient corresponds to the norm of the excited state energy
gradient in hartree
A1
d-CO state Evertical
(hartree)
leading one-electron
excitations
Eopt (hartree)
Gradient
S0 1A
S1 1A
S3 1A
4236.3114
4236.2521
4236.2280
67 a 37a
67 a 68a
4236.2716
4236.2425
0.010
0.016
4122.9559
4122.8510
4122.8501
35 a 62a
60 a , 61a 62a
4122.8737
4122.8704
0.0010
0.0011
d state
S0 1A
S1 1A
S2 1A
31
the anti CO in trans has an higher weight, while the opposite happens for the
LUMO.
The optimized a-CO excited states structures are reported in Fig. 9. The
nal norm of the gradient is different from zero in a non-negligible way
(Table 3) and uctuation of energies and geometries are observed during optimization. Furthermore, we found for both optimized structures a critically
low value (around 0.01) of the lowest singlet instability matrix eigenvalue.
However, it is reasonable to consider these structures as two points along the
CO photolysis reaction path that belong to S1 and S3 potential energy surfaces. Within this assumption, we can characterize the reaction paths through
the analysis of the optimized structure.
In both a-CO optimized excited state structures, the -CO loses its bridging character, becoming a terminal CO. The labilized Fe C bond for S1 is the
one in trans position to the anti terminal CO, while for S3 the one in trans
Fig. 9 Structure of a-CO and a excited states. All distances and angles in
A and degree.
a,b a-CO S1 1A and S3 1A excited state structures, respectively; c,d a S1 1A and S2 1A
excited state structures, respectively
32
L. Bertini et al.
position to the syn terminal CO. Therefore we have characterized two channels in S1 and S3 potential energy surfaces along which the photolysis brings
to the loss of the -CO group, although it is not possible to surely assess which
is exactly the leaving CO group. Nevertheless for S1 and S3 the leaving CO
group is one of those bound to the Fe syn or anti to the carbon of -PDT,
respectively.
The excited state optimized structures for a are reported in Fig. 9. The gradient norms of optimized excited state energies are around 0.001 hartree
A1 ,
and therefore these are likely stationary structures. The Fe Fe internuclear
distances are remarkably increased with respect to the ground state, because
in both cases molecular orbitals with Fe Fe antibonding combinations are
populated.
The nal considerations we can make in light of the results obtained are
that a-CO S1 and S3 are involved in the photolysis of -CO group. Since a-CO
is a model for the CO inhibited H-cluster, we can argue that similar excited
states might be involved in the formation of Hox II, in the approximation of
considering the effect of the [Fe4 S4 ] cluster negligible.
6
Conclusions and Perspectives
The application of quantum chemical methods to the investigation of hydrogenases and related biomimetic models has nicely complemented experimental studies, allowing us to disclose key factors related to the catalytic
properties of the enzymes. In particular, the possibility to use theoretical
methods to study intermediate and transition state structures, reaction energies and activation barriers, as well as spectroscopic properties, has allowed
us to dissect the molecular mechanism of several reaction paths relevant
to hydrogenase chemistry. Nevertheless, several relevant issues remain unsolved. For example, further experimental and theoretical work is necessary
to corroborate the formation of high-spin species in the catalytic cycle of
[NiFe] hydrogenases. This will also need further studies to evaluate and improve the reliability of DFT functionals in properly describing relative stabilities of low and high spin species. As for [Fe] hydrogenases, the investigation
of the role of the [Fe4 S4 ] moiety on the catalytic properties of the [2Fe]H
cluster has started to be addressed only very recently and is still largely
incomplete.
The examples discussed in this review also show that both the investigation
of small and large models is necessary to obtain a clear picture of the chemistry of complex molecular systems such as hydrogenases. In general, the
investigation of simple models, which are computationally inexpensive, allows one to use sophisticated quantum chemical approaches and to compare
properties of several models. However, the investigation of more complex
33
models is mandatory when the effect of neighboring groups has to be taken

into account. Indeed, the adopted computational models are often a compromise between a realistic representation of the enzyme active site and the
necessity to use small models to maintain computational costs to a reasonable limit. In fact, even though calculations carried out on model systems
including also a few aminoacids not directly bound to the metal clusters have
been reported [112, 113], most theoretical investigations of hydrogenases active site have been carried out using relatively small models in which the
cysteine residues coordinated to the metal ions have been modeled by H or
CH3 groups.
The extension of small models to include effects due to steric constraints
and electrostatic effects imposed by the protein environment is expected to
be crucial to clarify key issues in hydrogenase chemistry. As an example,
aminoacids in the nearby of the metallic clusters of hydrogenases, such as
cysteine, aspartate, glutamate and histidine might be directly involved in H2
activation. To the best of our knowledge, only one theoretical study related to
hydrogenases and keeping explicitly into account the protein environment of
the active site has been reported so far [182]. Amara et al. employed a hybrid
QM/MM potential method for the investigation of some [NiFe] hydrogenase
redox states, reaching the conclusion that one of the cysteine residues is involved in the heterolytic cleavage of dihydrogen and that -H species can
be formed in the catalytic cycle. However, as emphasized by the authors,
owing to the expense of calculations an exhaustive investigation of all plausible structures and redox states could not be carried out, underlining some
of the problems that are still associated with the proper modeling of large
protein portions.
Appendix
A
Outline of Density Functional Theory
In density-functional theory [13] (DFT) methods, the problem of nding the
electronic structure of an atomic or molecular system is solved on the basis of
its three-dimensional electron density (r) as a basic variable. This approach
represents a simplication with respect to common methods of quantum
chemistry based on the 3N-dimensional (or 4N when spin variables are included) wave function, where N is the number of the electron involved. The
HohenbergKohn theorem [12] provides a rigorous framework for this approach. The rst theorem assesses that a one-to-one map between the ground
state wave function and the electron density exists through the external po-
34
L. Bertini et al.
tential
v(r) =
Z
,
|r R |
(17)
where runs over the number of nuclei of charges Z with coordinates R .

The total energy is written as a functional of the electron density (r)

E (r) = F (r) + v(r)(r) dr ,
(18)
where F[(r)] is the unknown functional of the density. Since F[(r)] does
not depend on the v(r), it must be an universal functional of the electron
density. The second theorem denes a variational principle for the electron
density: for a given trial density t (r) we have that E0 E[t (r)]. The energy functional is minimized according
to the variational principle with the

constraint of the N-representability (r) dr = N. In the language of the functional derivatives this means that the differential E[(r)] = 0

E (r)
(r) dr N = 0 ,
(19)
(r)
where is the undetermined Lagrange multiplier. From the denition of the
differential of a functionals, it turns out that

F (r)
+ v(r) .
(20)
=
(r)
All current DFT implementations in quantum chemistry are based on the
KohnSham (KS) approach which introduces in an ingenious way the molecular orbitals machinery to overcome the problems of the early ThomasFermi model in computing the kinetic energy [183]. The basic idea is the
introduction of the so-called non-interacting reference system, obtained by
turning off the electron-electron repulsion in the real interacting system as
as
follow. Let us dene the parameter dependent Hamiltonian operator H
=
T +
V +
Vee ,
H
(21)
where 0 1 is called interelectronic coupling-strength parameter. For

= 1 we have the real interacting system, and
V=1 is the usual electronN

nucleus potential such that V = i v(r i ). The = 0 case corresponds to
the non-interacting system in which
V=0 is such that the electron density
(r)=0 = (r)=1 .
The exact ground state kinetic energy T[(r)] of the real interacting system ( = 1) can be expressed as a natural orbitals (NO) expansion, but it can
be approximated with the exact ground state kinetic energy TS [(r)] of the
non-interacting system:

N
2

2

NO NO
i = TS (r) ,

i
T (r) =
ni i
i

2
2
i=1
35
(22)
i=1
where the natural orbitals iNO are eigenvectors of the one-electron density
matrix, with the orbital occupation numbers 0 ni 1 and with the total
number of electron N = i ni . The exact electron density (r) is then approximate as follows:
(r) =

i=1
N
NO 2

ni i
|i |2 ,
(23)
i=1
where i are the KS molecular orbitals. The exchange-correlation energy

functional is dened as the difference between the kinetic and electronelectron interaction energy of the real and reference systems:

Exc (r) = T (r) TS (r) + Vee (r) J (r) ,
(24)
where Vee [(r)] is the electron-electron interaction energy whose classical
part J[(r)] is given by

(r)(r )
dr dr .
(25)
J (r) =
|r r |
The Exc contains the non-classical contributions to the kinetic and electronelectron interaction energies, but it is mainly dominated by the exchange
contribution, since TS [(r)] is the dominant part of the true kinetic energy.
Finally the vKS (r) is

(r)
dr + vxc (r) ,
(26)
vKS (r) = v(r) +
|r r |
where vxc (r) is the exchange-correlation potential given by the functional
derivatives of Exc with respect to (r). The problem of nding the KS molecular orbitals i is then solved using the self-consistent-eld (SCF) machinery:
the i (r) orbitals are the lowest energy eigenfunctions of the KS one-electron
operator
hKS

2

hKS i (r) =
+ vKS (r) i (r) = i i (r)
(27)
2
where i are the KS orbital energies. The KS total energy is

N

2

i + v(r)(r) dr + J (r) + Exc (r) .

EKS =
i

2
(28)
i=1
From a computational point of view, the i KS MOs are conveniently expanded in a xed basis set of functions (usually called atomic orbitals)
such that i = p cpi p . In turn, p can be represented by a given number of
36
L. Bertini et al.
linearly combined functions (gaussian). The number of p basis functions

considered determines the variational exibility of the basis set and therefore
the quality of the results. For instance, for atoms having occupied orbitals 1s,
2s and 2p, a basis set composed by 4 of s type and 2 of p type is called
double zeta basis. Additional functions can be added to represents diffuse
states (negatively charged) or polarization effects (high angular momentum
functions).
Expanding the KS molecular orbitals in the atomic basis sets (LCAO approach) the total KS electronic energy becomes [184]
Nbasis
EKS =
pq
Nbasis

1
Ppq hpq +
Ppq Prs pq|r + Exc (P) ,
2 pqrs
(29)
where the indices p, q, r and s run over the number of atomic basis functions.
Ppq are the elements of the one-electron density matrix in the basis of the
atomic orbitals such that the (r)
Ppq =
N
(r) =
cpi cqi
i=1
N

i
i (r)i (r) =
(30)
N
basis
Ppq p (r)q (r) .
(31)
pq
Finally hpq and (pq|rs) are the one- and two-electron integrals, and Exc (P) underlines the P density matrix dependence of the exchange-correlation energy,
since (r) is computed according to Eq. 31. The KS orbital equation is solved
iteratively since vKS (r) is a functional of the (r) but (r) is determined by
vKS (r). A guess electron density is computed according to Eq. 31 and the Exc
functional is calculated to generate the vxc potential. Then one optimizes the
electron density by minimizing the KS energy.
The form of the exchange-correlation energy (Exc ) within the KohnSham
framework is the following

Exc (r) = f (r), (r), 2 (r), ... dr
(32)
and the vxc potential of the Eq. 26 is the functional derivative of Exc . For
instance if f depends only on (r) and its gradient, vxc is

Exc (r)
f
f
f
f
vxc (r) =
=
,
(33)
(r)
(r) x (r) y (r) z (r)
where x (r) = (r)/x. The f function in Eq. 32 is generally written as follows

f = (r)xc (r), (r), 2 (r), ... .
(34)
37
The exchange-correlation energy per unit volume xc is an analytic function

with some parameters that can be tted to some experimental data or determined according to some model systems. The form of xc represents the main
feature of a given DFT functional and therefore also its reliability and limitations. When xc is that of the homogeneous electron gas, one obtains the
simplest approximation to the exact Exc called local-density-approximation
(LDA) functional. In LDA the xc is a function only of the density, but this
approximation can be improved by including the explicit dependence on the
(r) (generalized gradient approximation, GGA) or on the kinetic energy
(meta-GGA functionals [185]). Among GGA functionals, the BeckePerdew
(BP86) exchange-correlation functional is one of the most used. This functional is composed by the Becke88 exchange [23] and the Perdew GGA correlation [24] terms.
Another important approach to built DFT functionals is the so-called hybrid method based on the adiabatic connection formula [186]. According to
the latter, the exact Exc functional is given by the following integral:
1
Exc =

|vxc ()| d ,
(35)
where is the parameter dened in the Hamiltonian operator in Eq. 21, while
is the antisymmetric wave function that minimizes the energy functional.
To cite Axel Becke from one of his seminal papers on hybrid functionals [20]:
This formula connects the non-interacting KohnSham reference systems to
the fully interacting real system through a continuum of partially interacting systems all which share a common density (the density of the real, fully
interaction system. The approximate Exc functional can be obtained by numerical integration. The simplest case is the two end-points approximation
in which the integrand Exc is evaluated at = 0 and = 1. The 0 |vxc (0)|0
is reduced to the HF exchange, since we are dealing with the non-interacting
system. The 1 |vxc (1)|1 remains unknown, but it can be approximated
using others DFT functionals. The functionals obtained in this simple way are
called EH+H
half-and-half functionals:
xc
1
1
= EHF
+ EDFT
.
(36)
EH+H
xc
2 x
2 xc
Becke proposed an improvement of the EH+H
xc , which is known in the quantum chemistry literature as Becke-3 parameters hybrid exchange-correlation
functional (B3):

LSDA
LSDA
PW91
EB3
+ a EHF
+ bEB88
.
(37)
xc = Exc
x + Ex
x + cEc
The empirical parameters a, b and c are determined by tting experimental data. In Eq. 37 LSDA refers to the local-spin-density approximation and EB88
[23] and EPW91
[22] are the gradient corrections to the
x
c
38
L. Bertini et al.
LSDA exchange and correlation. The most popular hybrid functional is the
B3LYP [21], which is similar to B3:

LSDA
VWN
+ bEB88
= ELSDA
+ a EHF
(38)
EB3LYP
xc
x
x + Ex
x + Ec
LYP

VWN
+ c Ec Ec
,
where LYP and VWN refer to the Lee-Yang-Parr [22] and Vosko-WilkNussair [187] correlation functionals.
GGA functionals correct some of the LDA shortcomings, meta-GGA ones
improve certain properties with respect to GGA, and nally hybrid functionals give signicant improvement over GGA for several molecular properties.
This seems to represent a constant improvement of the functionals but, as
a matter of fact, there is not a unique recipe to accurately reproduce molecular
properties using DFT, and the reliability of this approach critically depends
on the type of molecule considered and on the properties of interest.
The evaluation of the analytic DFT energy gradients is similar to that of
HF [184]:

pq|rs
Exc
Spq
Etot hpq
pq
=
Ppq
Wpq +
+
Ppq Prs ,
pq
pqrs
(39)
N c c is the energywhere Spq are the overlap integrals and Wpq = i=1
i pi qi
weighted density matrix. Analytic second derivatives are more complicated
since they involve the derivatives of the MO coefcients, which are computed
using the coupled-perturbed SCF equations.
B
Outline of Time-Dependent Density Functional Theory
TDDFT is based on the linear response theory [188]. The formal framework
of this approach was given by Runge and Gross who have extended the rst
HohenbergKohn theorem to the time-dependent case [189]. The linear response electron density of a given excited state is [190193]

(r,) =
Xia ()a (r)i (r) + Yai ()a (r)i (r) ,
(40)
ia
where indices i and a run over the occupied and virtual KS ground state
MOs, respectively. The frequency dependent coefcients Xia () are associated
with the single excitations i a , while Yai () are related to non-physical
de-excitations a i which introduce the correlation effects of doubleexcitations in (r,). X and Y matrices represent the linear response of the
ground state KS MOs to the time-dependent external perturbation v(t) intro-
39
duced in the time-dependent KS equation with potential vKS (r, t) given by

(r, t)
dr + vxc (r, t) ,
(41)
vKS (r,t) = v(t) +
|r r |
(r,t) is the Fourier transform of (r,), v(t) is the applied eld, and when it
is turned on very slowly (adiabatic approximation) the vxc (r,t) potential can
be evaluated as the functional derivative of the static Exc [(r)t ] with respect
to t (r), where the time-dependent density is evaluated at a xed time t.
To evaluate X and Y coefcients, one invokes the dependence of the variation of (r,) on the variation of the KS potential through the KS kernel
KS (r,r ,), which represents the linear response of the KS non-interacting
system:

(42)
(r,) = KS (r,r ,)vxc (r ,) dr dr

(r)a (r)i (r ) (r ) i (r) (r) (r )a (r )
i
a
a
i
KS (r,r , ) =
+
.
(

)
+
(
a
i
a i )
ia
(43)
Calculation of the excitation energy requires to nd the poles of KS (r,r ,).
This can be done solving the following non-hermitian eigenvalues-eigenvectors problem

A B X
1
0 X
=
(44)
B A Y
0 1 Y
Aai,bj = ab ij (a i ) + Kai,bj
Bai,bj = Kai,jb

Kai,bj =
a (r)i (r)
1
2 Exc
+
(r )j (r ) dr dr
|r r | (r)(r ) b
(45)
(46)
(47)
The second functional derivatives of the Exc with respect to (r,) represents
the exchange-correlation kernel f xc(r,r ,): the adiabatic approximation can
be seen as the static limit f xc(r,r , = 0), which is evaluated as the functional
derivative of the stationary potential vxc /. Since the frequency dependence of f xc is neglected, TDDFT cannot give accurate excitation energies
when excited states have strong double-excitation character (i.e., for organic
polyenes [194]). In fact it can be shown that exact f xc is strongly frequencydependent near the double excitation [195, 196].
It is worth mentioning the so-called Casidas formulation of the
TDDFT [197] in which the problem 21 is written in the Hermitian form
F i = 2i F i . Here the matrix elements are written as a function of the KS
orbital energies and the Kai,bj integrals, and F i eigenvector easily allow the excited state term assignment. Indeed a CIS wave function that corresponds to
40
L. Bertini et al.
the (r, ) response density can be calculated from the F i , which permits one
to determine the main excitations involved in a given excited state.
In order to understand the TDDFT level of theory with respect to other
methods, one has to point out that when the Exc functional derivative is
neglected and B = 0, than TDDFT is reduced to CIS. Therefore A elements introduce the correlation effects of the single excited congurations.
If one neglects Exc functional derivative in Eq. 46, TDDFT is reduced to
time-dependent HartreeFock (TDHF). Within TDHF, double-excited congurations are introduced through the B elements since they involve i a
excitation and b j de-excitation at the same time. This is mathematically equivalent
compute ground state/double-excited conguration matrix
to
ia

elements, 0 H jb where 0 is the single-determinant wave function that
corresponds to KS ground state density [198]. Finally if one neglects the
Yia () elements, the TammDancoff approximation (TDA) formulation of the
TDDFT is obtained [199]. This approach often gives results of the same quality of the full TDDFT approach and it is easier to implement. Besides, in
TDDFT one introduces additional correlation effects through the functional
derivative of Exc . It could be argued that the TDDFT level of theory is higher
than ground state DFT because of the effects of single/double excitations introduced through A and B matrix elements. On the other hand, the inclusion
of single and double congurations tries to account for electronic correlation
effects as Exc does, and therefore a problem of overestimation of the correlation energy could arise.
Excited state energy rst derivatives with respect to the nuclear positions
can be derived in an advantageous way through the denition of the variational Lagrangian of the excited state energy L[X, Y, C0 , , W, Z] [166], where
C0 are the KS ground state MOs coefcients and is the excitation energy.
The two extra parameters W and Z constrain the MOs to be orthonormal
solutions of the ground state KS equation. When a stationary conditions of
L with respect to all parameters is reached, the nal expression of the rst
derivative with respect to a generic nuclear position is given by

vxc
Spq
hpq
pq
=
Ppq
Wpq +
Ppq
(48)
pq

xc
pq|rs
fpqrs
pqrs +
(X + Y)pq (X + Y)rs
+
pqrs
xc are second functional derivatives of E
where i) fpqrs
xc integrals which are
found in Eq. 7; ii) P and are the relaxed one and two electron density matrices [164] and, together with the W energy weight relaxed density matrix,
represents the derivative of the excitation energy with respect to one-electron,
two-electron and overlap integrals (hpq , (pq|rs) and Spq , respectively). In
Eq. 24 no derivatives of the C0 coefcients are needed. Moreover, it depends
41
on hpq , (pq|rs) and Spq derivatives built on the atomic orbital basis set, which
are already implemented in the ground state gradients computation. For these
reasons, the computational cost of excited state and ground state gradients is
similar. Furthermore the recent application of the RI approximation to both
excited state energy and gradients calculations, allow the computation on
large system up to 4000 basis functions [200].
The TDDFT optimization of the geometrical structure of a given excited
state is carried out as follows. Using the KS ground state MOs, one calculates
the excited state energy and response density. Afterwards the W and Z elements are computed in order to evaluate the nuclear forces. Finally, the nuclear
coordinates are updated, and KS ground state MOs newly calculated.
References
1. Holm RH, Solomon EI (2004) Thematic issue: Biomimetic inorganic chemistry.
Chem Rev 104:347
2. Koch W, Holthausen MC (2002) A Chemists guide to density functional theory. Wiley, second edition
3. Niu S, Hall MB (2000) Chem Rev 100:353
4. Siegbahn PEM, Blomberg MRA (2000) Chem Rev 100:421
5. Lovell T, Himo F, Han W-G, Noodleman L (2003) Coord Chem Rev 238-239:211
6. Friesner RA, Baik M-H, Gherman BF, Guallar V, Wirstam M, Murphy RB, Lippard SJ
(2003) Coord Chem Rev 238-239:267
7. Ziegler T, Autschbach J (2005) Chem Rev 105:2695
8. Siegbahn PEM, Blomberg MRA, Wirstam Pavlov M, Crabtree RH (2001) J Biol Inorg
Chem 6:460
9. Fan H-J, Hall MB (2001) J Biol Inorg Chem 6:467
10. Stein M, Lubitz W (2002) Curr Opin Chem Biol 6:243
11. Bruschi M, Zampella G, Fantucci P, De Gioia L (2005) Coord Chem Rev 249:1620
12. Hohenberg P, Kohn W (1964) Phys Rev 136:B864
13. Parr RG, Yang W (1989) Density-Functional Theory of Atoms, Molecules. Oxford
University Press, Oxford
14. Levine I (1999) Quantum Chemistry. Prentice Hall, fth edition
15. Dunlap BI, Connolly JWD, Sabin JR (1979) J Chem Phys 71:3396
16. Baerends EJ, Ellis DE, Ros P (1973) Chem Phys 2:41
17. Eichkorn K, Treutler O, Ohm H, Haser M, Ahlrichs R (1995) Chem Phys Lett 240:283
18. Eichkorn K, Wigend F, Treutler O, Ahlrichs R (1997) Theor Chem Acc 97:119
19. Becke AD (1992) J Chem Phys 96:2155
21. Stevens PJ, Devlin FJ, Chablowski CF, Frisch MJ (1994) J Phys Chem 98:11623
22. Lee C, Yang W, Parr RG (1988) Phys Rev B 37:785
23. Becke AD (1988) Phys Rev A 38:3098
24. Perdew JP (1986) Phys Rev B 33:8822
25. Cramer CJ (2002) Essentials of computational chemistry Theories and models.
Wiley
26. Vignais P (2005) Coord Chem Rev 249:1677
27. Krasna AJ, Rittenberg D (1954) J Am Chem Soc 76:3015
42
L. Bertini et al.
28. Vignais PM, Billoud B, Meyer J (2001) FEMS Microbiol Rev 25:455
29. Lenz O, Friedrich B (2001) Biospektrum 6:515
30. Cammack R, Robson R, Frey M (1997) Hydrogen as a fuel learning from nature.
Taylor and Francis, London UK
31. Albracht SP (1994) Biochim Biophys Acta 167:1188
32. Graf EG, Thauer RK (1981) FEBS Lett 136:165
33. Adams MWW (1990) Biochim Biophys Acta 1020:115
34. Cammack R (1999) Nature 397:214
35. Nicolet Y, Lemon BJ, Fontecilla-Camps JC, Peters JW (2000) Trends Biochem Sci
25:138
36. Peters JW (1999) Curr Opin Struct Biol 9:670
37. Horner DS, Heil B, Happe T, Embley TM (2002) Trends Biochem Sci 27:148
38. Nicolet Y, Cavazza C, Fontecilla-Camps JC (2002) J Inorg Biochem 91:1
39. Armstrong FA (2004) Curr Op Chem Biol 8:133
40. Lyon EJ, Shima S, Buurman G, Chowdhuri S, Batschauer A, Steinbach K, Thauer RK
(2004) Eur J Biochem 271:195
41. Volbeda A, Charon MH, Piras C, Hatchikian EC, Frey M, Fontecilla-Camps JC (1995)
Nature 373:580
42. Volbeda A, Garcin E, Pieras C, De ALLacey, Fernandez VM, Hatchikian EC, Frey M,
Fontecilla-Camps JC (1996) J Am Chem Soc 118:12989
43. Garcin E, Vernede X, Hatchikian EC, Volbeda A, Frey M, Fontecilla-Camps JC (1999)
Structure 7:557
44. Higuchi Y, Yagi T, Yasuoka N (1997) Structure 5:1671
45. Higuchi Y, Ogata H, Miki K, Yasuoka N, Yagi T (1999) Structure 7:549
46. Ogata H, Mizoguchi Y, Mizuno N, Miki K, Adachi S, Yasuoka N, Yagi T, Yamauchi O,
Hirota S, Higuchi Y (2002) J Am Chem Soc 124:11628
47. Nicolet Y, Piras C, Legrand P, Hatchikian EC, Fontecilla-Camps JC (1999) Structure
7:13
48. Peters JW, Lanzilotta WN, Lemon BJ, Seefeldt LC (1998) Science 282:1853
49. Bagley KA, Duin EC, Roseboom W, Albracht SP, Woodruff WH (1995) Biochemistry
34:5527
50. Volbeda A, Martin L, Cavazza C, Matho M, Faber BW, Roseboom W, Albracht SP,
Garcin E, Rousset M, Fontecilla-Camps JC (2005) J Biol Inorg Chem 10:239
51. Volbeda A, Fontecilla-Camps JC (2005) Coord Chem Rev 249:1609
52. Pereira AS, Tavares P, Moura I, Moura JJG, Huynh BH (2001) J Am Chem Soc 123:2771
53. Popescu CV, Munck E (1999) J Am Chem Soc 121:7877
54. De Lacey A, Stadler C, Cavazza C, Hatchikian EC, Fernandez VM (2000) J Am Chem
Soc 122:11232
55. Pereira AS, Tavares P, Moura I, Moura JJ, Huynh BH (2001) J Am Chem Soc 123:2771
56. Bennet B, Lemon BJ, Peters JW (2000) Biochemistry 39:7455
57. Lemon BJ, Peters JW (1999) Biochemistry 38:12969
58. Nicolet Y, DeLacey AL, Vernede X, Fernandez VM, Hatchikian EC, FontecillaCamps JC (2001) J Am Chem Soc 123:1596
59. Evans DJ, Pickett CJ (2003) Chem Soc Rev 35:268
60. Rauchfuss TB (2004) Inorg Chem 43:14
61. Darensbourg MY, Lyon EJ, Smee JJ (2000) Coord Chem Rev 206207:533
62. Liu X, Ibrahim SK, Tard C, Pickett CJ (2005) Coord Chem Rev 249:1641
63. Bouwman E, Reedijk J (2005) Coord Chem Rev 249:1555
64. Lyon EJ, Georgakaki IP, Reibenspies JH, Darensbourg MY (1999) Angew Chem, Int
Ed 38:3178
43
65. Le Cloirec A, Best SP, Borg S, Davies SC, Evans DJ, Hughes DL, Pickett CJ (1999)
Chem Comm, p 2285
66. Schmidt M, Contakes SM, Rauchfuss TB (1999) J Am Chem Soc 121:9736
67. Dong W, Wang M, Liu X, Jin K, Li G, Wang F, Sun L (2006) Chem Comm, p 305
68. Schwartz L, Eilers G, Eriksson L, Gogoll A, Lomoth R, Ott S (2006) Chem Comm,
p 520
69. Tard C, Liu X, Ibrahim SK, Bruschi M, De Gioia L, Davies SC, Yang X, Wang L-S,
Sawers G, Pickett CJ (2005) Nature 433:610
70. Nehring JL, Heinekey DM (2003) Inorg Chem 42:4288
71. Tye JW, Lee J, Wang H-W, Mejia-Rodriguez R, Reibenspies JH, Hall MB, Darensbourg MY (2005) Inorg Chem 44:5550
72. Gao W, Liu J, Ma C, Weng L, Jin K, Chen C, Akermark B, Sun L (2006) Inorg Chim
Acta 359:1071
73. Artero V, Fontecave M (2005) Coor Chem Rev 249:1518
74. Best SP (2005) Coord Chem Rev 249:1536
75. Osterloh F, Saak W, Hasse D, Pohl S (1997) Chem Comm, p 979
76. Davies SC, Evans DJ, Hughes DL, Longhurst S, Sanders JR (1999) Chem Comm,
p 1935
77. Smith MC, Barclay JE, Cramer SP, Davies SC, Gu WW, Hughes DL, Longhurst S,
Evans DJ (2002) J Chem Soc, Dalton Trans, p 2641
78. Marr AC, Spencer DJE, Schroder M (2000) Coord Chem Rev 206207:1055
79. Zhu W, Marr AC, Wang Q, Neese F, Spencer DJE, Blake AJ, Cooke PA, Wilson C,
Schroder M (2005) Proc Natl Acad Sci 102:18280
80. Stenson PA, Marin-Becerra A, Wilson C, Blake AJ, Mc Master J, Schroder M (2006)
Chem Comm, p 317
81. Sellman D, Geipel F, Lauderbach F, Heinemann FW (2002) Angew Chem, Int Ed
41:632
82. Georgakaki IP, Thomson LM, Lyon EJ, Hall MB, Darensbourg MY (2003) Coord
Chem Rev 238239:255
83. Zampella G, Bruschi M, Fantucci P, Razavet M, Pickett CJ, De Gioia L (2005) Chem
Eur J, 11:509
84. Loschen C, Frenking G (2004) Inorg Chem 43:778
85. Zampella G, Bruschi M, Fantucci P, De Gioia L (2005) J Am Chem Soc 127:13180
86. Neugebauer J, Reiher M, Kind C, Hess BA (2002) J Comput Chem 23:895
87. Cao Z, Hall MB (2001) J Am Chem Soc 123:3734
88. Liu Z-P, Hu P (2002) J Am Chem Soc 124:5175
89. van der Vlugt JI, Rauchfuss TB, Whaley CM, Wilson SR (2005) J Am Chem Soc
127:1601
90. Gloaguen F, Lawrence JD, Rauchfuss TB, Benard M, Rohmer MM (2002) Inorg Chem
41:6573
91. Bruschi M, De Gioia L, Zampella G, Reiher M, Fantucci P, Stein M (2004) J Biol Inorg
Chem 9:873
92. Tye JW, Darensbourg MY, Hall MB (2006) Inorg Chem 45:1552
93. Jensen F (1998) Introduction to computational chemistry. Wiley
94. van der Vlugt JI, Rauchfuss TB, Whaley CM, Wilson SR (2002) Chem Eur J 8:4037
95. Lawrence JD, Li H, Rauchfuss TB, Benard M, Rohmer MM (2001) Angew Chem Int
Ed 40:1768
96. Niu S, Thomson LM, Hall MB (1999) J Am Chem Soc 121:4000
97. Fan H-J, Hall MB (2001) J Am Chem Soc 123:3828
98. Niu S, Hall MB (2001) Inorg Chem 40:6201
44
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.
133.
134.
135.
136.
L. Bertini et al.
Fan H-J, Hall MB (2002) J Am Chem Soc 124:394
Bruschi M, Fantucci P, De Gioia L (2002) Inorg Chem 41:1421
Bruschi M, Fantucci P, De Gioia L (2003) Inorg Chem 42:4773
Zhou T, Mo Y, Liu A, Zhou Z, Tsai KR (2004) Inorg Chem 43:923
Zhou T, Mo Y, Zhou Z, Tsai K (2005) Inorg Chem 44:4941
Pavlov M, Siegbahn PEM, Blomberg MRA, Crabtree RH (1998) J Am Chem Soc
129:548
De Gioia L, Fantucci P, Guigliarelli B, Bertrand P (1999) Inorg Chem 38:2658
De Gioia L, Fantucci P, Guigliarelli B, Bertrand P (1999) Int J Quantum Chem 73:187
Stein M, van Lenthe E, Baerends EJ, Lubitz W (2001) J Am Chem Soc 123:5839
Stein M, Lubitz W (2001) Phys Chem Chem Phys 3:5115
Stein M, Lubitz W (2001) Phys Chem Chem Phys 3:2668
Foerster S, Stein M, Brecht M, Ogata H, Higuchi Y, Lubitz W (2003) J Am Chem Soc
125:83
van Gastel M, Stein M, Brecht M, Schroder A, Lendzian F, Bittl R, Ogata H, Higuchi Y,
Lubitz W (2006) J Biol Inorg Chem 11:41
Stadler C, de Lacey AL, Hernandez B, Fernandez VM, Conesa JC (2002) Inorg Chem
41:4417
Stadler C, de Lacey AL, Montet Y, Volbeda A, Fontecilla-Camps JC, Conesa JC, Fernandez VM (2002) Inorg Chem 41:4424
Neese F (2003) Curr Opin Chem Biol 7:125
van Lenthe E, Baerends EJ, Snijders JG (1993) J Chem Phys 99:4597
Sadlej AJ, Snijders JG, van Lenthe E, Baerends EJ (1995) J Chem Phys 102:1758
Vosko SH, Wilk L, Nusair M (1980) Can J Phys 58:1200
Stein M, Lubitz W (2004) J Inorg Biochem 98:862
Dole F, Fournel A, Magro V, Hatchikian EC, Bertrand P, Guigliarelli B (1997) Biochemistry 36:7847
Wang CP, Franco R, Moura JJG, Moura I, Day EP (1992) J Biol Chem 267:7378
Wang H, Ralston CY, Patil DS, Jones RM, Gu W, Verhagen M, Adams M, Ge P, Riordan C, Marganian CA, Mascharak P, Kovacs J, Miller CG, Collins TJ, Brooker S,
Croucher PD, Wang K, Stiefel EI, Cramer SP (2000) J Am Chem Soc 122:10544
Gu W, Jacquamet L, Patil DS, Wang H-X, Evans DJ, Smith MC, Millar M, Koch S, Eichhorn DM, Latimer M, Cramer SP (2003) J Inorg Biochem 93:41
Reiher M, Salomon O, Sellmann D, Hess BA (2001) Chem Eur J 7:5195
Cavallo L, Jacobsen H (2003) J Phys Chem A 107:5466
Paulsen H, Duelund L, Winkler H, Toftlund HH, Trautwein AX (2001) Inorg Chem
40:2201
Poli R, Harvey JN (2003) Chem Soc Rev 32:1
Salomon O, Reiher M, Hess BA (2002) J Chem Phys 117:4729
Noodleman L, Norman JG Jr (1979) J Chem Phys 70:4903
Noodleman L (1981) J Chem Phys 74:5737
Xu Z, Thompson LK, Miller LK, Ruiz DO, Alvarez S (2003) Inorg Chem 42:1107
Noodleman L, Peng CY, Case DA, Mouesca J-M (1995) Coord Chem Rev 144:199
Mouesca J-M, Chen JL, Noodleman L, Bashford D, Case DA (1994) J Am Chem Soc
116:11898
Torres RA, Lovell T, Noodleman L, Case DA (2003) J Am Chem Soc 125:1923
Noodleman L, Norman JG Jr, Osborne JH, Aizman A, Case DA (1985) J Am Chem
Soc 107:3418
Aizman A, Case DA (1982) J Am Chem Soc 104:3269
Papaefhymiou V, Millar MM, Munck E (1986) Inorg Chem 25:3010
45
137. Carney MJ, Papaefhymiou GC, Spartalian K, Frankel RB, Holm RH (1988) J Am
Chem Soc 110:6084
138. Auric P, Gaillard J, Meyer J, Moulis JM (1987) Biochem J 242:525
139. Antanaitis BC, Moss TH (1975) Biochim Biophys Acta 405:262
140. Middleton P, Dickson DPE, Johnson CE, Rush JD (1988) Eur J Biochem 88:135
141. Anderson PW, Hasegawa H (1955) Phys Rev 100:675
142. Blondin G, Girerd JJ (1990) Chem Rev 90:1359
143. Noodleman L, Baerends EJ (1984) J Am Chem Soc 106:2316
144. Papaefhymiou V, Girerd JJ, Moura I, Moura JJG, Munck E (1987) J Am Chem Soc
109:4703
145. Noodleman L, Case DA (1992) Adv Inorg Chem 38:423
146. Borchch SA, Bominaar EL, Blondin B, Girerd JJ (1993) J Am Chem Soc 115:5155
147. Fiedler AT, Brunold TC (2005) Inorg Chem 44:9322
148. Bruschi M, Greco C, Fantucci P, De Gioia L, manuscript in preparation
149. Klamt A, Schrmann G (1993) J Chem Soc Perkin Trans 2 5:799
150. Klamt A (1995) J Phys Chem 99:2224
151. Warshel A, Naray-Szabo G, Sussman F, Hwang K (1989) Biochemistry 28:3629
152. Bottoni A, Lanza CZ, Miscione GP, Spinelli D (2004) J Am Chem Soc 126:1542
153. Ahlrichs R, Bar M, Haser M, Horn H, Kolmel C (1989) Chem Phys Lett 62:165
154. Schafer A, Huber C, Ahlrichs R (1994) J Chem Phys 100:5829
155. Dreuw A, Head-Gordon M (2005) Chem Rev 105:4009
156. Cai ZL, Sendt K, Reimers JR (2002) J Chem Phys 117:5543
157. Tozer DJ, Handy NC (1998) J Chem Phys 109:10180
158. Van Gisbergen SJA, Groeneveld JA, Rosa A, Snijders JG, Bearends EJ (1999) J Phys
Chem 103:6835
159. Hummel P, Oxgaard J, Goddard WA III, Gray HB (2005) Inorg Chem 44:2454
160. Petit L, Maldivi P, Adamo C (2006) J Chem Theory and Comput 1:953
161. Adamo C, Scuseria G, Barone V (1999) J Chem Phys 1:2889
162. Hyotcherl H, Cao J, Zewail AH (2001) Angew Chem, Int Ed 20:1532
163. Handy NC, Schaefer HF (1984) J Chem Phys 81:5031
164. Van Caillie C, Amos RD (1999) Chem Phys Lett 308:249
165. Van Caillie C, Amos RD (2000) Chem Phys Lett 317:159
166. Furche F, Ahlrichs R (2002) J Chem Phys 117:7433
167. Helgaker T, Jorgensen P (1989) Theor Chim Acta 75:111
168. Hutter J (2003) J Chem Phys 118:3928
169. Doltsinis NL, Kosov DS (2005) J Chem Phys 122:144101
170. Weiss H, Ahlrichs R, Haser M (1993) J Chem Phys 99:1262
171. Burcl R, Amos RD, Handy NH (2002) Chem Phys Lett 335:8
172. Dreuw A, Dunietz BD, Head-Gordon M (2002) J Am Chem Soc 124:12070
173. Reiher M, Kirchner B, Hutter J, Sellmann D, Hess BA (2004) Chem Eur J 10:4443
174. Adams MWW (1987) J Biol Chem 262:15054
175. Lemon BJ, Peters JW (1999) Biochemistry 38:12969
176. Lemon BJ, Peters JW (2000) J Am Chem Soc 122:3793
177. Bennett B, Lemon BJ, Peters JW (2000) Biochemistry 39:7455
178. Chen Z, Lemon BJ, Huang S, Swartz JD, Peters JW, Bagley KA (2002) Biochemistry
41:2036
179. Fiedler AT, Brunold TC (2005) Inorg Chem 44:1794
180. Albracht SPJ, Roseboom W, Hatchikian EC (2006) J Biol Inorg Chem 11:88
181. Bertini L, Greco C, Fantucci P, De Gioia L, manuscript in preparation
182. Amara P, Volbeda A, Fontecilla-Camps JC (1999) J Am Chem Soc 121:4468
46
183.
184.
185.
186.
187.
188.
189.
190.
191.
192.
193.
194.
195.
196.
197.
198.
199.
200.
L. Bertini et al.
Kohn W, Sham LJ (1965) Phys Rev 140:A1133
Johnson BG, Fisch MJ (1994) J Chem Phys 100:7429
Perdew JP, Kurth S, Zupan A, Blaha P (1999) Phys Rev Lett 82:2544
Harris J (1984) Phys Rev A 29:1648
Vosko S, Wilk L, Nussair M (1980) Can J Phys 58:1200
McWeeny R (1989) Methods of Molecular Quantum Mechanics, 2. edn. Academic
Press, London
Runge E, Gross EKU (1984) Phys Rev Lett 52:997
Gross EKU, Kohn W (1985) Phys Rev Lett 55:2850
Petersilka M, Gossmann UJ, Gross EKU (1996) Phys Rev Lett 76:1212
Marques MAL, Gross EKU (2004) Ann Rev Phys Chem 55:427
Bauernschmitt R, Ahlrichs R (1996) Chem Phys Lett 256:454
Hsu C-P, Haed-Gordon M (2001) J Phys Chem A 105:451
Gorling A (1998) Phys Rev A 57:3433
Aryasetiawan F, Gunnarsson O (2002) Phys Rev B 66:165119
Casida M (1995) Recent Advances in Density Functional Methods, chapter TimeDependent Density Functional Response Theory for Molecules. World Scientic,
Singapore, p 220
Stratmann RE, Scuseria GE, Frisch MJ (1998) J Chem Phys 109:8218
Hirata S, Head-Gordon M (1999) Chem Phys Lett 314:291
Rappoport D, Furche F (2005) J Chem Phys 122:1277
Top Curr Chem (2007) 268: 4783

DOI 10.1007/128_2006_081

Sebastian Sinnecker Frank Neese (u)
Institut fr Physikalische und Theoretische Chemie, Universitt Bonn, Wegelerstr. 12,
53115 Bonn, Germany
theochem@thch.uni-bonn.de
1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
2
2.1
2.2
2.3
2.4
2.5
Theoretical Methods for Bioinorganic Spectroscopy .

Magnetic Spectroscopy of the Ground State . . . . . .
Linear Response Theory for Ground State Properties .
Mssbauer Spectroscopy . . . . . . . . . . . . . . . .
Optical Spectroscopy . . . . . . . . . . . . . . . . . .
Environment Effects . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
50
50
53
60
62
63
3
3.1
3.1.1
3.1.2
3.1.3
3.2
3.3
Recent Case Studies . . . . . . . . . . . . . . . . . . . . . .

High Precision Studies of Organic Radicals . . . . . . . . .
Molecular g-Tensors of Phenoxyl Radicals . . . . . . . . .
Quinone Radical AnionSolvent Interactions . . . . . . . .
The Cysteine Radical and Its Unusual g-Tensor Anisotropy
The Spectroscopy of Blue Copper Proteins . . . . . . . . .
Mononuclear Iron Complexes
from Outstanding Model Systems to Fascinating Enzymes .
Characterization of an Elusive Fe(V) Species . . . . . . . .
Binuclear Mixed Valence Copper Complexes
ENDOR and HYSCORE Spectra Simulations from DFT . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
64
64
65
66
67
70
. . . . . . . . .
. . . . . . . . .
71
76
. . . . . . . . .
78
Conclusions and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
3.4
3.5
4
Abbreviations
ABS
absorption spectroscopy
CASSCF
complete active space SCF
CD
circular dichroism
DFT
density functional theory
ENDOR
electron-nuclear double resonance
EPR
electron paramagnetic resonance
ESEEM
electron spin echo envelope modulation
ET
electron transfer
EXAFS
extended X-ray absorption ne structure
HDvV
HeisenbergDiracvanVleck Hamiltonian
HF
HartreeFock
HFCs
hyperne coupling constants
HYSCORE hyperne sublevel correlation
MB
Mssbauer
.
.
.
.
.
.
.
.
.
.
.
.
48
MCD
MCSCF
MD
MO
MRCI
MRPT
NMR
NQCs
OZ
PCET
QM/MM
rR
SCF
SH
SOC
SOMF
SOMO
SORCI
SS
TauD
TDA
TD-DFT
XAS
XC
ZFS
S. Sinnecker F. Neese
magnetic circular dichroism
multicongurational self-consistent eld
molecular dynamics
molecular orbital
multi reference conguration interaction
multi reference perturbation theory
nuclear magnetic resonance
nuclear quadrupole coupling constants
orbital Zeeman
proton-coupled electron transfer
quantum mechanical/molecular mechanical
resonance-Raman
self-consistent eld
spin-Hamiltonian
spin-orbit coupling
spin-orbit mean-eld
semi occupied molecular orbital
spectroscopy oriented conguration interaction
spinspin
taurine/-ketoglutarate dioxygenase
time-dependent DFT
X-ray absorption
exchange-correlation
zero-eld splitting
1
Introduction
The past decade has witnessed an explosive activity in the application of
quantum chemical methods to problems of bioinorganic chemistry. It is now
almost commonplace to publish an experimental paper together with calculations to back up the conclusions that have been reached or to distinguish between those possibilities that the experimental results left open. In
this context the term quantum chemical methods has been applied almost
synonymously with density functional theory (DFT). Despite of or perhaps
even due to the enormous popularity of DFT in bioinorganic chemistry, one
frequently voiced complaint from the experimental community is that the
calculations often take too little or no notice of the available experimental
results. The majority of theoretical papers deal with the calculation of reaction mechanisms which are either proposed or rejected based on theoretically
calculated reaction- or transition state energies. Since there are seldom accurate and detailed experimental data available for these thermodynamic and
kinetic parameters, the accuracy of the calculations may sometimes be difcult to assess. In this respect it is important to connect the computations
to the vast body of experimental data that has been assembled by the ex-
49
perimental bioinorganic community over the past four decades. Traditionally,

a large part of this experimental data is of spectroscopic origin. The reason
for this central role that spectroscopy plays in bioinorganic chemistry is that
these methods allow one to selectively probe the active sites of metalloproteins and metalloenzymes. Secondly, spectroscopic methods can be applied to
short-lived intermediates, mixtures or only partially puried proteins, which
makes them a versatile tool to follow biochemical events or elementary reaction steps. Thirdly, spectroscopic methods do not only yield geometric
structure information but also information at subatomic resolution, e.g. electronic structure information. Since understanding electronic structures is at
the heart of understanding the unique reactivities of metalloenzymes, spectroscopy is an extremely useful complement to the X-ray crystallography. The
latter technique has the highest information content with respect to geometric structure. However, there are limitations too which concern short-lived
species, mixtures and the necessity to perform the experiments on single
crystals which may or may not be stable in the X-ray beam.
An important aspect of many active sites of metalloproteins and radical
enzymes is that they exist, at least transiently, in open-shell congurations.
The presence of unpaired electrons then gives rise to paramagnetism and
allows the spectroscopist to probe the active site with various magnetic techniques including (in order of increasing resolution) magnetic susceptibility, magnetic circular dichroism (MCD), electron paramagnetic resonance
(EPR), electron-nuclear double resonance (ENDOR), electron spin echo envelope modulation (ESEEM) or hyperne sublevel correlation spectroscopy
(HYSCORE). The EPR, ENDOR, ESEEM, and HYSCORE techniques are covered concisely in a recent book by Schweiger and Jeschke [1], susceptibility
is covered in many textbooks on molecular magnetism [24], and MCD
spectroscopy is covered in a number of references [58]. In most cases, the
metallic active sites or protein cofactors are also amenable to studies with
various optical spectroscopies including standard absorption spectroscopy
(ABS), circular dichroism (CD), MCD and also resonance-Raman (rR) spectroscopy [911]. Finally, the spectroscopy of inner shells of transition metal
atoms and ions is of increasing importance in bioinorganic chemistry. X-ray
based techniques such as X-ray absorption (XAS) and extended X-ray absorption ne structure (EXAFS) are heavily employed since they allow element
specic studies to be carried out [1214]. Finally, due to the abundance and
importance of iron in bioinorganic chemistry, the technique of Mssbauer
(MB) spectroscopy is of considerable utility and yields valuable local information about iron centers including their magnetic properties if the experiments
are carried out with an applied external magnetic eld [15]. It is impossible to
cover all of these techniques at any level of detail in the framework of a single
piece of work and therefore most of our examples are selected from the elds
of EPR and MB spectroscopy where the theoretical techniques have turned
out to be particularly useful.
50
The central theme of this work is to show how one can combine spectroscopic techniques with theoretical investigations in order to obtain insight
into the geometric and electronic structures of metalloproteins and model
complexes as well as biologically important radical species. The majority
of examples is taken from recent investigations in our own laboratory and
should be reasonably representative of the considerable activities that are now
underway in many research groups. An attempt will be made to provide the
reader with at least a rough idea of some of the methodological aspects that
underlies the theoretical approaches used and which are either based on DFT
or on simplied correlated ab initio methods. More comprehensive information on the various theoretical methods with a full coverage of the recent
literature may be found in recent reviews [1619].
2
Theoretical Methods for Bioinorganic Spectroscopy
2.1
Magnetic Spectroscopy of the Ground State
The active sites of metalloproteins are always of low-symmetry and therefore one can safely assume that the systems under investigation are orbitally
nondegenerate. As mentioned above, many metalloprotein active sites exist
in open-shell ground state congurations. The presence of unpaired electrons
is intimately related to the occurrence of a net molecular magnetic dipole
moment and consequently to a rich variety of molecular magnetic phenomena.
In general, a ground state with total spin S gives rise to 2S + 1 magnetic
sublevels with MS = S, S 1, ..., S. At the level of the BornOppenheimer
Hamiltonian, which covers the leading electrostatic interactions between
electrons and nuclei, these 2S + 1 sublevels are all degenerate and this does
not change upon the inclusion of the so-called scalar relativistic effects.
However, if the spinorbit coupling (SOC) and the direct dipolar spinspin
(SS) coupling as the leading spin-dependent relativistic effects are included in
the treatment, the degeneracy of the 2S + 1 sublevels is lifted. This lifting of
degeneracy is referred to as zero-eld splitting (ZFS) and is parameterized
by a tensor with symbol D.
If an external magnetic eld is applied, the energies of the magnetic sublevels are further modied. This is the molecular electronic Zeeman effect
and it is parameterized by the so-called molecular g-tensor (symbol g).
The g-tensor arises (mainly) from the SOC induced coupling of the electronic
spin- and orbital-moments to the external magnetic eld. It may be thought
of as the EPR analogue of the NMR chemical shift tensor. For a S = 1/2 system
the molecular Zeeman effect leads to a linear divergence of the energies of
51
the MS = 1/2 and MS = +1/2 states at a rate that is proportional to the effective g-value which depends in turn on the relative orientation of the molecule
to the external magnetic eld. However, while the chemical shift is specic
for a given nucleus, the g-tensor is a system-property since it measures the
distribution of orbital currents over the entire molecule which is closely related to the response of the spin-density distribution to the presence of the
external magnetic eld.
Inclusion of the nuclear spins leads to further magnetic effects which can
be studied by magnetic spectroscopies. First of all, the nuclear spin interacts
with the external magnetic eld, thus giving rise to the nuclear Zeeman effect
which is parameterized by the chemical-shift tensor ( ). It occupies a central
role in the interpretation of nuclear magnetic resonance (NMR) experiments.
Secondly, the nuclear spins interact with the electron spins via three distinct
mechanisms: (a) the well-known (isotropic) Fermi contact interaction which
is proportional to the probability of nding unpaired electrons at the position
of the magnetic nuclei, (b) an anisotropic (traceless) electron spin-nuclear
spin dipolar interaction and (c) the anisotropic (nontraceless) SOC induced
coupling of the nuclear spins with the orbital motions of the unpaired electrons. The sum of these interactions is parameterized by the hyperne coupling tensor (symbol A). Finally, the indirect coupling of nuclear spins leads
to very small effects which show up as spinspin couplings in NMR spectroscopy and are parameterized by a spinspin coupling tensor (symbol J).
Larger effects occur for nuclei with a nuclear spin I > 1/2. In this case, the
nuclear quadrupole moment may interact with the electric eld gradient produced by the surrounding charges. This is referred to as quadrupole coupling
and it is measured by the quadrupole tensor (symbol Q).
The very powerful concept of the SpinHamiltonian (SH) allows one to
treat all of these complicated interactions in a semi-phenomenological way
where a minimum of empirical parameters allows the deduction of the entire
spin-physics of the system under investigation. The SH is an effective Hamiltonian and contains only spin-variables of a ctitious electron spin S and
the nuclear spins IA , IB , ... All reference to the spatial part of the wavefunction and therefore to detailed molecular electronic and geometric structure
is implicitly contained in the SH parameters D, g, A, Q, and J which are
considered as adjustable parameters in the analysis of experiments. Thus, the
SH is a convenient place to rest in the analysis of experimental data by theoretical means [20]. The SH which includes the interactions covered above
is:

SPIN =
(A)

H
I (A) + N Bg (A)
SD
S + Bg
S+
(1)
SA(A)
N I
A
+
I (A) Q(A)
I (A) +

(A)

I J (AB)
I (B) ,
A<B
52
where the sum over A refers to the magnetic nuclei, B is the magnetic ux
density and and N is the electronic and nuclear Bohr magneton, respectively. The SH acts on a basis of product functions |SMS |I (A) MI(A) ...
(N)
|I (N) MI .
For not too many spins this basis is often small enough to allow exact diagonalization of the SH and therefore exact quantum mechanical treatments
of the spin-physics in the SH framework. For high-dimensional SH problems,
both, brute-force (Piligkos et al. 2006) and a variety of perturbation theoretical methods can be employed in order to arrive at exact or good approximate
solutions.
The role of theory is then to derive the connection of rst principle electronic structure approaches to the SH parameters and to devise practical
algorithms for the prediction of the values of these parameters. In this way,
calculations can help to develop the full information content of spectra and
thus allow conclusions about the geometric and electronic structure of the
system under investigation to be drawn (Fig. 1). A great deal of progress has
been made in recent years in the calculation of EPR parameters (summarized
in [2123]). The theory of NMR parameters is a well-studied eld in theoretical chemistry and mature procedures with well-dened accuracy for the
desired quantities have been established [21].
In many cases, it is sensible to divide the system under investigation into
distinct fragments with given electron spins SA , SB , ... The interaction of the
electron spins of the different spin-systems is usually referred to as exchange
coupling and is given by the symbol J [note that unfortunately the same symbol is usually used for the nuclear spinspin interaction (see e.g. Eq. 1). It
should, however, from the context be evident to which interaction J refers].
The interaction consists of the genuine isotropic exchange interaction
J which is purely electrostatic in origin and only arises from the antisymmetry
requirement of the N-electron wavefunction with respect to electron permutations (consequently, the notion of an exchange interaction is misleading.
In fact, such an interaction does not exist in nature. The exchange coupling
is a consequence of the interplay between electronelectron repulsion and the
Fermion nature of electrons). Unfortunately, there are several popular ways
to write the isotropic exchange Hamiltonian (HeisenbergDiracvanVleck
Hamiltonian, HDvV). Perhaps the most common way is to write:
HDvV = 2J
SB .
SA
H
(2)
In this formalism two interacting electrons give rise to a singlet-state (St = 0)

and a triplet state (St = 1) which are split in energy by |2J|. Positive values of
J lead to St = 1 below St = 0 (ferromagnetic coupling), while negative values
of J lead to the reverse situation (antiferromagnetic coupling). In addition
to the HDvV Hamiltonian, there exists the direct dipoledipole interaction
between the electron spins which also contributes in an anisotropic and trace-
53
Fig. 1 The relationship of theory and experiment. Electronic structure calculations solve
the Schrdinger equation for a system that is characterized by its molecular Hamiltonian
and provide the spin Hamiltonian (SH) parameters, e.g. g-tensors, hyperne coupling
constants or Mssbauer parameters. These data are also the result of the spectral simulations and therefore the intersection of theory and experiment. From a careful comparison
of simulated and calculated SH parameters, conclusions on electronic and even molecular
structures can be drawn
less way to J. Very interesting phenomena arise from the combination of the
exchange interaction with the on-site spin-Hamiltonians described in Eq. 1.
Entire books are devoted to the study of such interactions [2, 24].
2.2
Linear Response Theory for Ground State Properties
All SH parameters can be approached using the powerful language of linear
response or analytic derivative techniques. For the present purposes it is sufcient to focus on the time independent case. In order to appreciate the general
concepts that are involved, the linear response equations for a Self-Consistent
Field (SCF) ground state will be sketched below. This description is appropriate if the state of interest is well described by a HF (HartreeFock) or DFT
single determinant:
(x1 , x2 , ..., xN ) = |1 ...N | ,
(3)
where the {} constitute the set of occupied spin orbitals and xi denotes the
space and spin coordinates of the ith electron. The ground state energy is
written as:
1
E = ENN +
h i +
i
(ii| jj) cHF (ij|ij) + cDF EXC [] .
(4)
2
i
i,j
54
Here, ENN is the nuclearnuclear repulsion energy,

h is the one-electron operator, (ij|kl) is the electronelectron repulsion integral
1
k (x2 )l (x2 ) dx1 dx2
(5)
(ij|kl) =
i (x1 )j (x1 )r12
and EXC [] is the exchange-correlation energy functional with the ground
state density being given by:

(x) =
i (x)i (x) .
(6)
i
The parameters cHF and cDF are scaling parameters for the HF exchange
energy and the XC-energy, respectively. Thus, HF theory corresponds to
cHF = 1; cDF = 0, pure DFT corresponds to cHF = 0; cDF = 1 while hybrid DFT
methods choose 0 < cHF < 1. The orbitals are determined self-consistently
from the SCF equations:

)

E
[]
(x
XC

ii

dx cHF
K + cDF
(7)
h+
i (x) = i i (x) ,
|x x |
(x)
i
where the second term represents the Coulomb-potential, the third term the
HF exchange interaction with the nonlocal operator:
K ij |l = (il|kj) .
k |
(8)
We also dene the Coulomb operator:

k |
J ij |l = (ij|kl) .
(9)
The term in brackets is collectively referred to as

F , the pseudo-one-electron
SCF operator. The orbitals are usually approximated by a nite expansion in
a basis set {}:

i (x) =
ci (x)
(10)
which results in the following expression for the density:

P (x) (x)
(x) =
(11)
with the density matrix

P =
ci ci .
(12)
In order to approach molecular properties one needs to determine the derivatives of the total SCF energy with respect to external perturbations , , ...
Since all SH parameters are bilinear in external perturbations, the desired
quantity is the second partial derivative. For illustrating the concepts, it is
55
sufcient to consider the case where the basis functions are chosen to be independent on the external perturbations. To include such a dependence (as
is necessary for example for geometric or magnetic eld perturbations) is
straightforward but would lead to more lengthy equations which are not of
interest for the purpose of this work. Since the cs are determined in a variational procedure, one has:
E ci
=0.
ci
(13)
And therefore the rst derivative of the energy with respect to a perturbation
is:

E
=
P |
h | ,
(14)

=0
where:

h
|
h | =

(15)
if the basis functions are independent of the perturbation. Through straightforward differentiation the second partial derivative becomes:

P
2 E
=
P |
h | +
h | .
(16)
|

=0; =0
This important equation contains two contributions: the rst term is referred
to as a rst-order contribution since it only depends on the ground state density. The second contribution is a second-order contribution since it requires
the knowledge of the rst derivative of the density matrix with respect to an
external perturbation.
In order to calculate the derivative of the density matrix a number of techniques could be employed. The most straightforward approach is to start
from the condition of a SCF solution:
FPS SPF = 0 ,
(17)
where F is the matrix of

F in the AO basis, S is the overlap matrix and P the
density matrix. Differentiation with respect to yields the equations that
are necessary to solve for the perturbed density matrix P() . We will pursue a more pedestrian approach here and determine the perturbed orbitals
()
(0)
i |i() themselves in terms of the zeroth order orbitals i |i(0) .
Differentiation of the SCF equations yields:

()
()
() (0)

+
=0.
(18)
F
F (0) (0)
i
i
i
i
56
The perturbed orbitals are expanded as:

()
() (0)
=
Uai
a
i
(19)
(here and below labels i, j, k, l refer to occupied orbitals and a, b, c, d to unoccupied ones). The unitary matrix U has only occ/virt blocks in the case that
the basis functions do not depend on the perturbation. In order to determine
the unique elements of U one uses the perturbed SCF equations:

()
(0) (0) (0)
() () (0)
Ubi a(0)
F i b
F i i
+ a(0)
=0
(20)

()
(0)
(0) () (0)
F
(0)
+
a
=0.
= Uai

i
a
i
(21)
However,
F () depends on the perturbed orbitals. Therefore, ones needs to
take the derivative of the SCF operator carefully.
()
()
(x )
() jb
() ()
c
Ubj
(22)
F =h +
K bj + Ubj
K
HF

|x x |
jb
2 EXC [] ()
+ cDF
(x ) dx dx
(x)(x )
()
() jb
() bj
() jb
J bj + Ubj
J cHF Ubj
K cHF Ubj
K
=
h() +
Ubj
jb
fxc [] () (x) dx
()
(0)
(0)
()
bj
bj

Ubj
J cHF K + cDF fxc []b (x)j (x) dx
=h +
+ cDF
jb

() jb
+ Ubj
K jb + cDF fxc []j(0) (x)b(0) (x) dx ,
J cHF
where the XC-kernel fxc [] has been dened as the second functional
derivative with respect to and it has been tacitly assumed that for all functionals in use this yields a factor (x x ) which reduces the double integral to
a single integral. This yields the rst-order equations:

()
(0)
(0) () (0)
(0)
+
a
(23)
Uai
h

i
a
i
()

+
Ubj (bj|ai) cHF (ba|ji) + (ai|fxc |jb)
jb

()
+ Ubj (jb|ai) cHF (ja|bi) + (ai|fxc |jb) = 0 .
At this point it is useful to distinguish two different types of perturbation:

First, electric eld-like perturbations yield purely real a(0) |
h() |i(0) and
57
consequently, also purely real and symmetric U matrices. In this case one has:
A(E) U () = V () .
(24)
With:

(0)
(0)
+ 2(jb|ia) + 2cDF (ai|fxc |jb)

=
A(E)
ij
ab
a
i
ia,jb

cHF (ba|ji) + (ja|bi)

()
Vai
= a(0) |
h() |i(0) .
(25)
(26)
Note that the A-matrix (the electric hessian) is independent of the nature of
the perturbation and that the U and V matrices have been written as vectors
with a compound index (ai).
Secondly, magnetic eld like perturbation yield purely hermitian imaginary a(0) |
h() |i(0) and consequently also purely imaginary and hermitian U
matrices. This leads to:
A(M) U = V ()
(27)
with

(0)
(0)
+ cHF (ib|ja) (ba|ij) .

=
A(M)
ij
ab
a
i
ia,jb
(28)
Thus, magnetic eld-like perturbations yield much easier response (or

coupled perturbed) equations in which the contributions from any local
potential vanish. In fact, in the absence of HF exchange the A-matrix becomes diagonal and the linear equation system is trivially solved. This then
leads to a sum-over-orbital like equation for the second derivative which
resembles in some way a sum-over-states equation. One should, however,
carefully distinguish the sum-over-states picture from linear response or analytic derivative techniques since they have a very different origin. For electric
eld-like perturbations or magnetic eld-like perturbations in the presence
of HF exchange one thus has to solve a linear equation system of the size
N (occupied) N (virtual) which may amount to a dimension of several hundred thousand coefcients in large-scale applications. However, there are
efcient iterative techniques to solve such large equations systems without
ever explicitly constructing the full A-matrix. The key step of all existing procedures is the formation of a so-called sigma vector which is the product of
the A-matrix with a trial U-vector.
In any case, once the perturbed orbitals have been determined the perturbed density is easily found as:

()
() (0) (0)
() (0) (0)
P
=
Uai
ca ci + Uai
ci ca .
(29)
i
In order to illustrate the use of the equations above in a nontrivial example,

let us study the dominant contribution to the g-tensor. The g-tensor reads in
58
the SH formalism:
spin = Bg
S,
H
(30)
where is Bohrs magneton and B the magnetic ux density. First, it is

important to note that the order in which the perturbations are applied is immaterial owing to the interchange theorem; thus the more easily handled
perturbation is conveniently treated rst. In the case of the g-tensor, where
the two perturbations in question are the spin-independent orbital Zeeman
(OZ) operator and the spin-dependent SOC operator, it is more convenient to
treat the OZ operator as the and the SOC operator as the perturbation.
The SOC operator is taken in an accurate mean-eld approximation (SOMF)
developed by Hess et al. [25] and efciently implemented into the ORCA program recently [26]. The SOMF operator reads:

h SOMF =

z SOMF (i)
s (i) .
(31)
i
With the matrix elements of the kth component of the SOMF operator given
by:

SO

SOMF 1el-SO
=
zk
gk
(32)

P
hk
+

SO
SO
3

3

gk
gk
2
2
and:
2
3

hk1el-SO (r i ) =
ZA riA
liA;k
2 i A
(33)
2 3
(34)
lij;k rij .
2
Here, P is the total charge density matrix, ZA the nuclear charge of atom A, riA
is the position of electron i relative to nucleus A and
liA;k is the kth component of the angular momentum of the ith electron relative to atom A. Likewise,
lij;k is the kth component of
rij is the distance between electrons i and j and
the angular momentum of electron i relative to electron j. The OZ operator
referred to a global gauge origin (O) is:

l(O) (i) .
(35)
h OZ = B

gkSO (r i , rj ) =
Thus, one solves three sets of coupled-perturbed equations in order to obtain

three sets of U-coefcients for the three canonical directions of the magnetic
eld. In the spin-unrestricted case, each U-matrix falls into distinct spin-up
and spin-down sets as long as the perturbation is spin-independent which is
59
the case for the OZ operator. Finally, one inserts the solution into the secondorder part of Eq. 16 and makes use of the fact that within a given multiplicity
all matrix elements of the SOC operator can be written in terms of the
sz
component only. One arrives at:

()
2 E
1 P SOMF
z
=

(36)
B S =0; =0 2 B
()
1 P SOMF
z

2 B
1 P
=
2 B
()
SOMF
.
z

Thus, spin-dependent perturbations lead to the response spin-density matrix instead of the response electron density matrix which is present in the
electric-eld case. Comparing these matrix elements with the matrix of the
SH for spin S one nds:
(OZ/SOC)
g
1 P
=
2S B
()
SOMF

z

(37)
(note that both, the response spin density and the SOC matrix elements are
purely imaginary. In an actual implementation one operates with skew symmetric real-valued matrices in which case a factor of i2 = 1 arises in front of
the g-tensor equation).
In a similar way all terms in the SH can be calculated by analytic derivative techniques. Details of the actually involved procedures may be found
in the literature. Importantly, Eq. 16 is not only valid in the case of a SCF
ground state description but is of much wider applicability. In the case of correlated ab initio methods, the equations to be solved in order to determine the
effective density and its response naturally become more complicated than
the relatively simple coupled-perturbed SCF equations sketched in this work.
The general line of thought is, however, identical and merely the mechanics of the calculation become more involved. This statement holds for any
nonrelativistic variational or nonvariational methods (e.g. for conguration
interaction, MllerPlesset or coupled-cluster methods) but does not apply
to quasi-relativistic (two-component) methods which treat the SOC together
with the electron-electron repulsion. For such methods, rst-order perturbation theory is usually sufcient to calculate SH parameters. However, as long
as these methods are based on single determinants, severe problems with internal consistency and symmetry breakings occur the discussion of which is
outside the scope of this work. Note also, that these methods are also about an
order of magnitude more expensive than the nonrelativistic one-component
methods and it may therefore be speculated that they will be reserved for
60
rather special applications in which it turns out to be mandatory to treat the

SOC to innite order.
2.3
Mssbauer Spectroscopy
An important spectroscopy in a bioinorganic chemistry context is Mssbauer
spectroscopy. It is a method which selectively probes the charge and spin
distribution around iron centers [27, 28]. In zero-applied magnetic eld, the
Mssbauer spectrum of an isolated iron center consists of a single pair of
lines which are called quadrupole doublet. The center of the pair is referred
to as isomer shift and the splitting is known as quadrupole splitting. Both
properties are readily calculated if the electron density of the ground state is
known. The quadrupole splitting of the iron nucleus is given by

1 2
1
EQ = e0 Qq 1 + 2 ,
(38)
2
3
where q = Vmax is the numerically largest eigenvalue of the symmetric eldgradient tensor V, = [Vmid Vmin ]/Vmax is the asymmetry parameter and
e0 is the elementary charge. The proportionality constant Q is the quadrupole
moment of the 57 Fe nucleus in the I = 3/2 excited state which is not known
precisely. In this study a value of 0.16 barn has been assumed in keeping with
a number of previous calculations. This value was also veried as the best
choice in a recent calibration study on a series of iron complexes [29]. The
electric eld gradient tensor is (, = x, y, z):

(A)
(39)
V
=
P rA5 rA2 3rA; rA; ,
,
where rA = r RA is the position of the electron relative to nucleus A at position RA , P is the density matrix from a ground-state DFT calculation with the
basis set {}.
The Mssbauer isomer shift can be determined to good accuracy from
ground-state DFT calculations using the calibration procedure outlined
in [30]. Since the isomer shift is known from basic principles to be directly
proportional to the electron density at the iron nucleus [27], it can be determined from a linear regression analysis, e.g. a plot of the calculated electron
density at the nucleus versus the measured isomer shift. For our method of
choice, the B3LYP functional together with the exible core properties basis set CP(PPP) with an extended set of polarization functions, it can be
computed from the regression [30]
= 6.5481 0.3666[(0) 11800]
in mm s1 ,
(40)
61
where (0) is the nonrelativistically calculated electron density at the iron nucleus. Indeed, the calculations in Fig. 2 show a very good correlation between
the two quantities with a regression coefcient of 0.986.
It is important to note, however, that the calculated electron density at the
iron nucleus is far from the exact density. First of all, using Gaussian basis
functions, the basis set limit is difcult to approach since Gaussians have the
wrong behavior close to the nucleus. The nuclear cusp can be reasonably well
modelled with very steep basis functions. To get very close to the basis set
limit is nevertheless almost impossible given the nite accuracy with which
numbers can be represented by a digital computer. It has been our strategy to
use a basis set with sufcient exibility in the core region to allow for the distortion of the inner s-, and p-shells and have termed this basis set CP(PPP).
Secondly, different density functionals lead to grossly different electron densities at the nucleus and therefore a given calibration is only valid for a given
functional. This large functional dependence may be expected since present
day functionals do not only have the wrong asymptotic behavior in the long
range but also in the short range close to the nucleus. Third, it is important to ensure that the numerical integration accuracy is sufcient to lead to
converged results in the presence of steep, uncontracted basis functions in
the core region. Fourth, the neglect of scalar relativistic effects is serious and
leads to an underestimation of the electron density of more than 3000 au3 .
This seems unacceptable since the variation of the electron density over the
range of molecules studied is only 7 au3 . Recent scalar relativistic calculations using the ZORA method have greatly improved on the value of (0) but
the quality of the regression remained unchanged [29, 31]. The reason why
Fig. 2 The linear relationship between calculated electron densities at the iron nuclei (0)
and measured isomer shifts of 15 different complexes. The obtained parameters can be
used in further application studies employing the same basis set CP(PPP) and functional
B3LYP
62
Fig. 3 Relative contributions to the electron densities at the iron nuclei (0). The gure
clearly shows that changes in (0) can be attributed to changes in the 3s and valence
shells, while the contributions from the 1s and 2s shells are almost constant, independent
of the charge and spin states of the systems
the correlation with experiment is still as good as has been observed in the
nonrelativistic calculations is shown in Fig. 3. It is demonstrated that the contributions of the 1s- and 2s-orbitals to the total electron density at the iron
nucleus is very constant over the range of bonding environments typically
met in iron containing complexes. Thus, very large but very systematic errors
cancel out in the linear regression analysis and ensure that DFT is an accurate
and reliable tool in the study of Mssbauer isomer shifts provided that the
results are standardized with respect to functional and basis set. Given these
precautions, the regression does not depend on the charge or spin state of the
iron and has proven in many applications to be trustworthy [2933].
2.4
Optical Spectroscopy
Optical spectroscopy will only be touched upon in a cursory manner since it
will only be briey addressed in the application section. The application of
DFT methods to excited states is difcult at rst glance since the Hohenberg
Kohn theorems only apply to the ground state. However, response theory
once more offers a solution to the problem. To this end, one needs to extend the formalism developed above to the case of time-dependent oscillatory
perturbations such as the oscillatory electric eld of a monochromatic, linearly polarized light beam. The second-order response of the system to such
a perturbation is the frequency dependent polarizability. This quantity has
poles at the positions where the frequency of the external eld equals the
excitation frequency of the system. In practice, one nds these poles by the
63
diagonalization of the response matrix. The random-phase approximation

leads to a generalized eigenvalue problem in which both, the electric and
magnetic Hessian matrices appear. Solution of this generalized eigenvalue
problem then directly yields the excitation frequencies. From the eigenvectors
one can calculate the electric and magnetic transition densities and therefore various transition properties such as oscillator strengths and rotation
strengths of CD spectra. An interesting approximation is provided by the
so-called TammDancoff approximation (TDA) which amounts to the diagonalization of the electric Hessian A(E) [34]. The TD-DFT/TDA equations
very much resemble the conguration interaction problem with single excitations only (CIS) and can be efciently solved at the similar computational
expense. Under most circumstances TD-DFT/TDA, in particular in combination with hybrid density functionals, is substantially more accurate than CIS.
However, one should be aware of a number of limitations of this methodology. In particular, TD-DFT fails for charge transfer and covalent-to-ionic
transitions, does not describe multiplet effects correctly, does not produce
any double excitation and does not produce pure spin-eigenfunctions in
the spin-polarized case. If one is aware of these limitations good use can
still be made of this method since it is applicable to large molecules and
analytic gradients for geometry optimizations on excited states can be implemented with reasonable effort in a computationally efcient way [3537].
For more details, the reader is referred to a recent review by Head-Gordon
and Dreuw [38].
A radically different approach to excited states is offered by multireference ab initio methods. These usually start with a multicongurational
self-consistent eld (MCSCF) description of the desired states followed by
the treatment of dynamic correlation effects with conguration interaction
(MRCI) or perturbation theoretical (MRPT) methods. A particularly successful MRPT variant is the CASPT2 method developed and widely used by
Roos and co-workers [39]. A computationally affordable variant of MRCI is
the spectroscopy oriented conguration interaction (SORCI) procedure developed in this laboratory and successfully used in recent studies [40]. These
methods are reviewed in some detail elsewhere [16].
2.5
Environment Effects
Bioinorganic active sites and model complexes are in the vast majority of
cases not studied in the gas phase but in condensed media. Consequently,
environment effects are always present in the experimental data and should
be modelled at some level of sophistication in order to obtain reliable
results.
Continuum dielectric approaches are a very convenient way to model longrange interactions between the system under study and a continuum, which
64
is characterized by its dielectric constant [41, 42]. In these models, the solute is placed in a cavity and interacts with the unstructured, but polarizable
surrounding. These methods do not increase the degrees of freedom of the
calculations, which makes them routinely applicable also for larger systems.
Furthermore, it is possible to perform geometry optimizations of the solute in
the solvent. Within the last years interesting studies appeared that have been
devoted to the calculation of solvent shifts of spectroscopic parameters employing continuum models, including ABS [4345], NMR [4648], and EPR
data [4955]. It is clear that these approaches are especially well suited for
the study of solutesolvent interactions, but they also have been used to study
protein active sites [56]. Nevertheless, it must be mentioned that continuum
models are not able to describe short-range interactions like solvent to metal
ion coordination or hydrogen bonding effects.
A more rigorous way to model environment effects of protein active sites is
the use of mixed quantum mechanical/molecular mechanical (QM/MM) approaches [57, 58]. Here, the cofactor of interest is modeled by the QM method,
while the protein backbone is described by the much faster MM approach.
In consequence, these methods allow the treatment of very large systems
entire proteins can be considered in this way. QM/MM studies that include
spectroscopic applications are available in the literature to an increasing degree [18, 46, 59, 60].
More approximate than QM/MM methods is the supermolecule approach,
in which the active site or solute and a rather small part of the surrounding
are treated on equal footing employing QM methods. While it was shown that
this method can give accurate spectroscopic parameters, for example for radicals in protic solution [50, 61], the number of solvent molecules that can be
treated is limited and conformational problems can occur.
Finally, abinitio molecular dynamics (MD) simulations are feasible, but
can be computationally fairly expensive. However, the results for the calculation of spectroscopic parameters are very promising [62].
3
Recent Case Studies
3.1
High Precision Studies of Organic Radicals
An accurate description of the electronic structures of organic radicals with
a single unpaired electron is much easier to achieve than for transition metal
clusters. Molecular g-tensors, hyperne coupling constants (HFCs) and nuclear quadrupole coupling constants (NQCs) can be straightforwardly calculated. The high accuracy that can be achieved for these systems allows for
the study of very subtle details like small conformational changes or solvent
65
effects. This is demonstrated here by discussing three different case studies, including investigations on phenoxyl radicals, benzosemiquinone radical anionsolvent interactions, and the cysteine thiyl radical conformational
changes. In all cases it was possible to draw important conclusions on the molecular and electronic structures of the systems by comparing calculated and
measured spectroscopic parameters. It is however, extremely important to be
aware about the achievable accuracy of the employed methods. It must be ensured that the investigated effects are larger than the relative errors in the
calculations. Typically, this is claried by benchmark calculations on similar,
but experimentally well-understood systems.
3.1.1
Molecular g-Tensors of Phenoxyl Radicals
A recent study has underlined the high value of combining experimental
and theoretical work on molecular g-tensors [63]. Three different phenoxyl
radicals, hydrogen bonded to an imidazolium, were investigated. These systems serve as models of the TyrD tyrosyl radical in photosystem II. The
one electron oxidation of the parent neutral compounds to yield the radical cations may either take place via a proton-coupled electron transfer
(PCET), or via a pure electron transfer (ET) as displayed in Fig. 4. In order to
discriminate between the possible reaction products, a high-eld (W-band)
EPR study was undertaken. It was possible in these measurements to resolve the g-tensor anisotropies. The interpretation of the obtained data was
guided by DFT calculations. Complete geometry optimizations were performed employing the BP functional [6466] in combination with the TZVP
basis set [67]. The g-tensors were calculated from the coupled perturbed
KohnSham equations [68] employing the B3LYP hybrid functional [69, 70]
and the EPR-II basis set [71]. Furthermore, the COSMO dielectric continuum
model was employed in the geometry optimizations and single-point calculations [72]. Calculations were performed on both conceivable products,
the OH and NH species shown in Fig. 4. The analysis of the data al-
Fig. 4 One electron oxidation of the charge neutral phenol species can take place via
proton-coupled electron transfer (PCET) or electron transfer (ET). Here, R = Ph is considered as an example
66
lowed only one conclusion: the radical that was found in the experiments
must be the NH form. Especially the gx ( gmax ) value calculated for the OH
form (2.0034) deviates by more than 3000 ppm from the experimental value
of gx = 2.0065. This deviation is about one order of magnitude larger than
the typical accuracy of g-tensor predictions for such radicals. Furthermore,
the experimental gx -value is almost perfectly reproduced in the calculations
on the NH form (gx = 2.0066). In consequence, the calculations provided
strong evidence that the NH-form has been formed under the experimental conditions and that the one electron oxidation must occur via a PCET
mechanism.
3.1.2
Quinone Radical AnionSolvent Interactions
Magnetic properties can also be used as very valuable parameters to obtain information on the surrounding of a given open shell species. The
effect of H-bonding, certainly one of the most important intermolecular
interactions in biochemistry, was recently studied by means of the parabenzosemiquinone radical anion in water [61]. Different model systems were
tested, employing the radical anion and one, two, four or 20 solvent molecules
within the supermolecule approach. Complete geometry optimizations were
performed with the B3LYP hybrid functional in combination with a double basis set, augmented by polarization and diffuse functions. Again, the
spectroscopic parameters were obtained from B3LYP calculations with the
well-proven EPR-II basis set, including g-values, as well as HFCs and NQCs
of deuterium solvent nuclei that form hydrogen bonds to the carbonyl oxygen
atoms of the quinone radical anion. This type of interaction was also found in
biological systems [73].
In general, better results were obtained with increasing size of the model
systems. This shows that the second solvation shell has a non-negligible inuence on the calculated properties. Through the complete calculation of
all relevant SH parameters (g-values, HFCs, NQCs and Euler angles) it was
possible to simulate the ENDOR spectra from DFT (Fig. 5). Only the line
width parameters, which are not available from the theoretical work, were
adjusted. From the comparison with the experimental spectra two important conclusions could be drawn: (i) Present-day DFT calculations can be used
to generate high-quality initial guess parameters for otherwise underdetermined spectral simulations, and (ii) the calculated geometry of the system
must represent a realistic scenario. The second point is especially important
for the deduction of H-bond lengths. While H-bond angles can be obtained
from the experimental Euler angles, the hydrogen bond lengths have to be
estimated from the dipolar HFCs or NQCs of a bridging deuterium employing crude approximations or t equations. Hence, it is a tempting idea to
take H-bond lengths directly from DFT calculations. This requires, however,
67
Fig. 5 Simulation of 2 H ENDOR spectra from calculated HFCs, NQCs, g-values and Euler
angles for the benzosemiquinone radical anion in water (model system with 20 solvent
molecules). Comparison with Q-band pulse 2 H ENDOR (Mims) absorption spectra (rst
derivative)
that a consistent set of calculated spectroscopic parameters is available and in

good agreement with all experimental data.
3.1.3
The Cysteine Radical and Its Unusual g-Tensor Anisotropy
Our next example is a theoretical study of the cysteine thiyl radical [74]. Such
sulfur centred radicals play an important role in biochemistry. They are met,
68
for instance, as catalytic intermediates in all three classes of ribonucleotide

reductases. The large spinorbit coupling constant of sulfur in comparison
to oxygen, nitrogen or carbon leads to unique spectroscopic features of this
species. One example is the noticeable anisotropy of the molecular g-tensor
with principle values that resemble transition-metal complexes, rather than
organic radicals.
The aim of the study of van Gastel et al. was to obtain a correlation of
g-values and HFCs of the thiyl radical with its structure and the polarity
of its surrounding. The carboxyl and amino groups were modeled in their
charge neutral form since calculations on zwitterionic model systems produce artifacts in gas-phase computations. The effect of a protic environment
was included in this work by additional calculations, where one or two water molecules that form hydrogen bonds to the cysteine sulfur atom were
added to the model system. Again, geometry optimizations were performed
employing the B3LYP functional in combination with triple- basis sets, augmented by polarization and diffuse functions. The g-values were obtained
from coupled perturbed SCF theory as discussed above [68]. Furthermore, 1 H
and 33 S HFCs were calculated. All these properties were obtained from B3LYP
calculations employing the IGLO-III basis set [75]. In addition, excitation energies were computed employing the SORCI approach [40].
Three different conformers with respect to rotations around the C C
bond of the radical were found in the geometry optimizations and characterized as local minima with the help of frequency calculations (Fig. 6).
The obtained structures are close in energy with differences not larger than
630 cm1 . Transition energies to the rst excited state are given for all isomers
in Table 1. The results indicate that the involved orbitals are almost degenerated in energy for the conformer 2, while distinctly larger excitation energies
were obtained for the 1st and 3rd minimum. Calculated g-values are also given
in Table 1. A large gx shift was found for all conformers. It is mainly determined by electronic transitions from the SOMO-1 to the SOMO orbitals and
therefore nicely correlates with the calculated excitation energies.
Fig. 6 Fischer projection of the cysteine thiyl radical (left). DFT geometry optimizations
yielded three different conformers 13 with respect to rotations around the C C bond.
While their energies were found to be similar within 630 cm1 , noticeably different gtensors were calculated for the conformers. A molecular coordinate system was chosen
with gx parallel to the C S bond and with gz located in the C C S plane
69
In the next step, a polar environment was modeled by adding water

molecules to the system. Hydrogen bonds were found between the water proA. For
tons and the sulfur atom with S H distances between 2.39 and 2.60
all three conformers, every additional water molecule decreased the calculated gx value. This effect can be traced back to a lifting of the degeneracy of the energies of the sulfur 3pz (SOMO-1) and 3py orbitals (SOMO),
which increases the energy denominator. The results obtained with two water
molecules are also given in Table 1.
Hyperne coupling constants were calculated in this study for the two
-protons and the 33 S nucleus. The 1 H -HFCs were found to be mainly determined by the conformation of the cysteine radical, while the inuence of
the number of solvent molecules on the calculated HFCs was found to be
extremely small. Furthermore, a large isotropic 33 S HFC of about 40 MHz
was obtained with variations of about 10 MHz depending on the model
system.
The experimental g-values of various thiyl radicals were found to strongly
depend on the chosen species. While values between 2.10 and 2.49 were in
general reported for gx , numbers in the range of gx = 2.152.25 were obtained
for the majority of investigated thiyl radicals. These data were nicely reproduced in the study of van Gastel et al. [74]. It was concluded that radicals with
gx values of about 2.15 are present in the rst minimum with a H-bonded
sulfur atom, while the experimental gx values of about 2.25 were traced back
to the same conformer without H-bonds. Such ndings can be very valuable
with respect to the wide variety of spectroscopic parameters that have been
found so far for cysteine radicals.
Table 1 Calculated properties of the three conformers 13 of the cysteine thiyl radical
from Fig. 6. Excitation energies E01 (cm1 ) from the ground states to the rst excited
states were calculated from SORCI. Molecular g-values were obtained from DFT (second order) employing the pure radicals (gas phase) and the radicals ligated by two water
molecules that form H-bonds to the sulfur atom (2H2 O). Third order g-values are given
for the model systems without water molecules in parenthesis. They were found to be
necessary to reproduce the negative gz shifts from the experiments
1
gas phase
E01
gx
gy
gz
1732
2.2649 (2.2649)
2.0215 (2.0043)
2.0024 (1.9852)
1083
2.3692 (2.3692)
2.0214 (1.9877)
2.0023 (1.9686)
3429
2.1680 (2.1680)
2.0161 (2.0092)
2.0023 (1.9954)
2 H2 O
gx
gy
gz
2.1407
2.0212
2.0024
2.1952
2.0229
2.0025
2.0928
2.0156
2.0023
70
3.2
The Spectroscopy of Blue Copper Proteins
Transition metal active sites are the chief ingredient of bioinorganic chemistry. On one hand it is a highly interesting task to explore the properties
of these systems, but on the other hand it is still an outstanding challenge
to accurately calculate the spectroscopic properties of transition-metal complexes. In order to arrive at a correct microscopic description of their physical
properties one needs to be able to deal with near-degeneracy situations,
complicated multiplet problems, spin-state energetics, excited states of vastly
different characters, spin-polarization, large spinorbit coupling effects and
interfragment spin-couplings simultaneously. In DFT one encounters problems in the balanced description of the partially ionic, partially covalent
metalligand bonds, the correct balance between spin polarization and spin
contamination, the construction of spin-eigenfunction and multiplet states
and the accurate calculation of excitation energies. In addition, an appropriate approximation of the spinorbit coupling operator and adequate, often
large, basis sets are required. By contrast, conguration interaction-based
methods suffer from the high computational effort and from the fact that
the number of spectroscopic properties that can be calculated with these approaches is very limited in most of the available quantum chemical programs.
However, in principle, these methods offer a solution to all of the problems
referred to above which is a signicant motivation for their further development.
The inuence of the protein surrounding on magnetic, electric and optical spectroscopic parameters of metalloproteins was recently investigated in
a QM/MM study on copper plastocyanin [18]. In these proteins, the metal is
coordinated to two histidines, a cysteine and a methionine amino acid residue
with an outstanding long Cu SMet bond of 2.9
A. This system has been intensively studied over several decades by experimental and theoretical means
and its electronic structure and spectroscopy is well understood. Hence, it is
an ideal benchmark system for a theoretical study.
Starting with the pdb le of the protein, a solvent box of water molecules
was added to the protein in the theoretical work and a MD simulation was
performed. Subsequently, QM/MM geometry optimizations were carried out.
The obtained structures were used for single point calculations on the active
site (a) in the gas phase, (b) in a continuum, and (c) in the protein environment employing the MM point charges. Furthermore, nonrelativistic and
scalar relativistic results were compared. Best agreement between calculated
and measured g-tensors, HFCs and NQCs was obtained for the scalar relativistic calculation in the protein environment, which was also the highest
level of theory employed in this study. The inuence of the surrounding on
the calculated spectroscopic properties was found to be quite substantial and
can alter the properties by up to 30%. Interestingly, this effect can be at-
71
tributed in similar extents to the protein backbone atoms and to the water
cage molecules.
Much effort of this work was devoted to the calculation of the Cu HFCs.
The Fermi-Contact term was underestimated, which is a typical result for
transition-metal ions. A comparison with the results obtained for other
metals indicates a correlation with the number of unpaired electrons. The
spin-dipolar contribution is expected to be rather accurately predicted. The
huge spinorbit contribution was also distinctly underestimated, which was
attributed to the fact that the linear response of the density functionals is
too stiff with respect to the external perturbations. Such problems of presentday density functionals express themselves in the errors of many properties
like g-tensors, second-order contributions to the HFCs, or transition energies.
Hence, a very careful handling of the computed results is necessary.
3.3
Mononuclear Iron Complexes
from Outstanding Model Systems to Fascinating Enzymes
This paragraph is dedicated to high-valent Fe(IV) complexes. Such systems
have been known for a very long time to be highly important in biochemistry
in the form of the oxo-complexes (FeO)2+ . They were characterized or postulated as transient intermediates in a wide variety of heme- and nonheme
iron enzymes. Because of their transient nature, spectroscopy is the method
of choice for studying such high-valent systems. An additional complication
in interpreting the spectroscopic data obtained for Fe(IV) sites is the fact that
until recently there have been only a limited number of model systems available that could be used for calibration and comparison [76]. Furthermore, the
spin states of (FeO)2+ systems must be determined: While the (FeO)2+ species
found in native systems like taurine/-ketoglutarate dioxygenase (TauD) were
proposed to be in the S = 2 state, only model systems with S = 1 ground states
have been reported so far in the literature. The only exception we are aware
of is the [FeO(H2 O)5 ]2+ model compound studied by Pestovsky et al. [77] as
well as a non-oxo compound by Collins and co-workers [78].
Hence, it is helpful for the interpretation of the experimental data to
know, how the spin states inuence the spectroscopic parameters and which
geometric features determine the ground-state spin. This is a typical problem that can be solved by quantum chemical calculations, where the system
can be forced into the desired state. Such a study was performed on the
[Fe(IV)O(NH3 )5 ]2+ system, a prototypical but hypothetical (FeO)2+ model
system [79]. Ammonia ligands were chosen to obtain a model complex that is
small enough to allow the use of CI-based methods but that, at least crudely,
resembles the cyclam ligands of some of the real model systems. The iron,
oxygen and nitrogen atoms were described by basis sets of triple- quality
in the CI calculations and DFT geometry optimizations. Spectroscopic par-
72
ameters were obtained employing the core property basis set CP(PPP) for
iron [30], and the SOMF operator. Results from DFT (BP86 and B3LYP) were
compared with the results from CI (SORCI based on CASSCF reference wave
functions).
Starting with the results from DFT geometry optimizations for the complex in the S = 1 and S = 2 states, the S = 1 state was found to be slightly more
stable by 2 (B3LYP) and 6 kcal/mol (BP). SORCI single-point calculations at
the DFT geometries gave the opposite result: the ground state was assigned to
the S = 2 state, being 10.8 kcal/mol lower in energy than the S = 1 state. This
shows that it is dangerous to draw conclusions only on the basis of energy calculations. Electron and spin populations were found to be similar in the DFT
and ab initio calculations. By varying the ironligand distances it was demonstrated that the equatorial ligand eld determines the spin state of the system:
weaker ligands favor the S = 2 spin state.
In the next step, the spectroscopic parameters of the two spin states were
compared. Starting with the zero-eld splitting parameters, large and positive D values were found for both spin states employing DFT (Table 2). From
SORCI, a value of D = +10.1 cm1 was calculated for the more stable S = 2
state, which is a factor of 2 larger than the BP86 result. It is also expected
to be more accurate based on model calculations on small molecules and the
more realistic description of the d4 multiplet structure in the ab initio approach.
The calculated g-values were found to be rather close to the free electron
g-values and are therefore not well suited for discriminating between the two
spin states. The iron hyperne data revealed an interesting difference between
the S = 1 and S = 2 states: While a small, negative A|| value of 4.9 MHz was
Table 2 Comparison of calculated spectroscopic parameters (B3LYP) of the [Fe(IV)O
(NH3 )5 ]2+ model system in the S = 1 and S = 2 states: Zero-eld splitting parameters D
and E/D, g-values, hyperne coupling constants A, and Mssbauer parameters MB , Eq ,
and . Noticeably large differences between the spin states were found for A|| (57 Fe), and
the Mssbauer parameters
D (cm1 )
E/D
g
g||
A (57 Fe) (MHz)
A|| (57 Fe) (MHz)
MB (mm/s)
Eq (mm/s)
S=1
S=2
5.20
0.00
2.0181
2.0075
30.1
4.9
0.133
0.54
0.07
4.25
0.00
2.0139
2.0061
29.6
50.9
0.292
1.21
0.03
73
predicted for the S = 1 state, a larger HFC of A|| = 50.9 MHz was found for
the S = 2 state employing the B3LYP hybrid functional. It is not expected that
the calculated numbers are very accurate, but the large qualitative difference
should be solid. Hence, the iron HFC might be an important probe for experimental studies on the spin states of (FeO)2+ systems.
Considering the calculated MB parameters, it was found that the isomer
shift of the S = 2 species is similar to that of many low- or intermediate spin
Fe(III) isomers. A distinctly smaller value was obtained for the S = 1 species
and is typical for such systems which generally show isomer shifts around
0 mm/s. The larger value for the S = 2 state was mainly attributed to the
strongly increased equatorial bond distances and is similar to the value found
for the TauD reaction intermediate for which this spin state was proposed.
The computed quadrupole splittings strongly depend on the symmetry of the
systems. The computed change in the sign from positive quadrupole splittings for S = 1 to negative EQ values for S = 2 was also found in experimental
work: While a positive value was reported from Rohde et al. for a S = 1 model
system (Ques model compound), a negative quadrupole splitting was found
by Price et al. for the S = 2 intermediate from TauD (Fig. 7) [80, 81]. Finally,
the absorption spectra were calculated employing TD-DFT and SORCI with
a small CAS(4,5) reference space. It was found by both methods that the S = 1
systems features dd transitions of lower energy than the system in the S = 2
state, but the ordering of the excited states and the predicted transition energies are very different. Again, more accurate results are expected from the
SORCI approach.
We now switch to biological systems and start with the high-valent oxoferryl intermediate of TauD. These enzymes couple the decarboxylation of
-ketoglutarate with the hydroxylation of an unactivated C H bond of taurine and play therefore an important role in diverse biological processes [82].
The transient (FeO)2+ species was recently well characterized by Mssbauer
spectroscopy, but its exact molecular structure is still not known [81, 83]. The
proposed structure of this intermediate, which is called J, is given in Fig. 7.
The metal is bound to a His2 Asp facial triad and a succinate molecule. Fur-
Fig. 7 The proposed molecular structure of the oxo-ferryl intermediate J of TauD, the
structures of the Fe(IV) model compounds of Collins and Que, and the structure of the
Fe(IV)-oxo porphyrin radical cation intermediate from P450 (Compound I). The thick bar
represents the heme plane
74
thermore, it is strongly assumed that this species is in the quintet state, which
is supported by the calculations performed by Krebs et al. [83]: The measured
Mssbauer parameters and EQ were only well reproduced in calculations
on a trigonal-bipyramidal model system with an axial oxo ligand in the quintet state (Table 3). The accuracy of the employed methods was judged by test
calculations on the well known (FeO)2+ model compounds from Fig. 7. It is
interesting to compare the MB data of intermediate J with those calculated
for the model system [Fe(IV)O(NH3 )5 ]2+ (Table 2): The data obtained for the
S = 2 state of [Fe(IV)O(NH3 )5 ]2+ are in much better agreement with the MB
parameters of the TauD oxo-ferryl intermediate than the data of the S = 1
state.
While we have discussed so far only nonheme iron systems, iron porphyrins are certainly of outstanding relevance in bioinorganic chemistry and
we will now focus on a study of the high valent ferryl-oxo porphyrin radical
cation of cytochrome P450 as a representative case for this class (Fig. 7) [84].
The work that we have picked out includes also an extensive modeling of the
protein environment with QM/MM methods. It is divided in three parts that
deal with (a) small model systems containing the (FeO)2+ motif, (b) the interaction between a porphyrin radical and the (FeO)2+ moiety, and (c) the
treatment of the protein active site with and without surrounding. The use of
small reference systems made the application of ab initio methods possible,
while the workhorse was DFT, again.
DFT was employed for geometry optimizations, the calculation of magnetic hyperne tensors, g-tensor and MB properties, while the ab initio
methods were used for the calculation of excitation energies and exchange
coupling constants. The small, symmetric [Fe(O)(NH3 )4 (H2 O)]2+ model system (C4v ) was used to analyze the character of the Fe O bond and the MO
diagram in detail, but also optical spectra, g-tensors and zero-eld splittings of this species were provided. Ques model compound (Fig. 7) was also
Table 3 Comparison of measured and calculated Mssbauer parameters of the high-valent
iron-oxo intermediate from TauD. Calculated data from Krebs et al. [83]. Experimental
results from Price et al. (TauD) [81], Kostka et al. (Collins model compound) [95], and
Rohde et al. (Ques model compound) [80]
(mm/s)
EQ (mm/s)
TauD intermediate J
Exptl.
Calcd.
+0.31
+0.22
0.88
1.0
Collins model compound
Exptl.
Calcd.
0.04
0.08
0.89
0.76
Ques model compound
Exptl.
Calcd.
+0.17
+0.11
+1.24
+1.2
75
employed in this study to calibrate the computational methods. A comparison of calculated geometrical parameters with those from X-ray evidenced
again the suitability of the BP86 density functional for the optimization of
molecular structures, which is very efcient in combination with the resolution of identity (RI) approximation [85, 86]. However, the typical overestimation of the weaker nitrogenmetal bond distances was also found in this
work. Strong effort was dedicated to the question of the electronic ground
state of both model compounds, which can be either of triplet or quintet
symmetry.
The calculated MB properties of Ques model compound were found to be
very sensitive with respect to the employed geometries. This indicates that the
developed t parameters for the calculation of isomer shifts depend not only
strongly on the chosen functionals and basis sets, but also on the underlying
geometries. Typically, the t parameters were developed employing molecular structures from BP86 optimizations, or from X-ray crystallography.
The calculated 57 Fe and 17 O hyperne data were found to be very stable
with respect to different geometries. Axially symmetric tensors were obtained
in both cases. Provided that the strong underestimation of the Fermi-Contact
term is corrected (e.g. by scaling [29]), good agreement between theory and
experiment was found for the 57 Fe HFCs, while no experimental data are
available for the oxo-group. For both model systems, small g-shifts were
found, but very large D-values. From a comparison with the experimental
data it became clear that the employed DFT methods tend to noticeably underestimate the zero-eld splitting D.
The next challenge of this study was an appropriate description of the
S = 1 FeO unit ferromagnetically coupled to an S = 1/2 porphyrin radical.
An analysis of the spin populations has shown that the FeO moiety is only
marginally affected by the porphyrin radical with spin populations mainly located at the methane bridges, and to a lesser extent at the pyrrole nitrogen
atoms. It was possible to calculate accurate MB parameters and 57 Fe HFCs,
and exchange coupling constants for this system.
The successful studies on these model compounds gave a solid base for
the investigation of the Compound I. Employing the QM/MM methodology,
a huge inuence of the protein environment on the calculated data was found
by comparing results that were obtained with and without QM/MM point
charges. Especially, spin density is taken away from the sulfur atom of the
axial cysteine ligand and a decrease of charge at the iron center was found if
the protein environment was included. Since antiferromagnetic coupling was
observed in this species, the doublet ground state was studied in addition to
the quartet state employing broken symmetry DFT [87, 88]. The MB parameters were found to be very similar in both spin states. This indicates that
the electron density at the metal center is very similar. Hyperne data were
analyzed in detail for 57 Fe, 17 O, 14 N and 1 H, which are sensitive probes for
electronic and geometrical details of the system. Especially the 1 H superhy-
76
perne coupling constants of the cysteine ligand noticeably depend on the

conformation of the side chain.
The results of this work had signicance beyond providing a set of calibrated theoretical parameters which could be used for the identication of
compound I. A key nding was that the exchange coupling in compound I
is expected to be small and negative while the zero-eld splitting was large
and positive. In this situation the notation of a pure spin state partially loses
its meaning since at room temperature the system has three Kramers doublets that are almost equally populated and that are heavy mixtures of doublet
and quartet components lowest in energy. Thus, all three doublets could be
the starting point for further reactions. This has an important bearing for the
concept of two-state reactivity in which the potential energy surfaces of the
doublet and quartet states play a key role [89, 90]. A second key nding of this
work was that there is a second manifold of sextet and quartet states low in
energy which arise from spin coupling of the S = 2 state of the (FeO)2+ unit
with the porphyrin/thiyl radical. That these states are low lying and could also
be implicated in the observed reactivity is evident following the results of the
model studies. In fact, shortly after the appearance of [84] this possibility was
explored theoretically [91], and it was indeed found that the sextet/quartet
channel is competitive along the reaction coordinate.
3.4
Characterization of an Elusive Fe(V) Species
While the highest conclusively identied oxidation state of iron in biochemistry is Fe(IV), the possibility that Fe(V) might exist and play a role has
been proposed many times. However, chemical precedence was largely missing. In this respect, the synthesis and characterization of a genuine Fe(V)
species presents and important step in this eld. Wieghardt and co-workers
initially succeeded to generate a Fe(V)-nitrido species in immobilized samples by photolysis of an Fe(III)-azide precursor (Fig. 8). The main evidence
for the generation of a high-valent iron species came from the Mssbauer
isomer shift of the photoproduct which was observed to be slightly negative [92]. The spin state of the photogenerated species was plausibly assigned as S = 3/2 based on the observations made for analogous isoelec-
Fig. 8 Molecular structure of the [FeV (cyclam-ac)(N)]+ complex
77
tronic Mn(IV) and Cr(III) species. The photochemistry was recently reinvestigated experimentally and theoretically with the goal to obtain more
evidence that the proposed species is correct [93]. To this end, the Mssbauer measurements were complemented by X-ray absorption spectroscopy
at the iron K-edge. These experiments revealed a pre-edge peak which was
one eV higher in energy than observed in (FeO)2+ complexes which is consistent with the oxidation state Fe(V). Furthermore, analysis of the EXAFS
region revealed that a new short 1.61
A Fe X (X = N,O) bond is formed.
The experiments were supplemented by DFT calculations. Initially, the calculations were performed for the [Fe(cyclam-ac)(N)]+ complex (cyclam =
1,4,8,11-tetrazacyclotetradecane) and a S = 3/2 ground state. Neither the calculated isomer shift, nor the quadrupole splitting nor the optimized bond
A were found to be in any reasonable agreement with the exdistance of 1.75
perimental data. This could be taken as evidence either against the formation
of a high-valent Fe(V)-nitrido species or as a sign of a wrong ground state.
All attempts to reproduce the experimental data with protonated species or
other conceivable decay products remained without success. Therefore, the
S = 1/2 spin state was investigated theoretically. It quickly became evident
that not only was the energy of the S = 1/2 state much lower than that of
the S = 3/2 state but also the calculated Fe N distance, quadrupole splitting and isomer shift agreed very well with the experiments. In addition,
calculations on the Fe(III) azide precursor also gave good agreement with
the experimentally known structure and spectroscopic parameters. Thus,
the calculations provided strong evidence that the photolysis indeed led to
a high-valent Fe(V)-nitrido species and that this species had an unanticipated
ground state spin of S = 1/2. In order to conrm this result, magnetic susceptibility experiments were undertaken and revealed unambiguously that the
ground state spin is indeed S = 1/2. The reason for this immediately became
evident from an inspection of the calculated electronic structure. For an approximately octahedral site, the ve d-orbitals split into a sigma-antibonding
eg and a -antibonding t2g set. The interaction with the extremely strongly
-donating N3 ligand, however, strongly splits the t2g set in a low-lying
nonbonding dxy orbital and a nearly degenerate -antibonding dxz , dyz set.
2 )(d1
Energetically, spin-pairing in the dxy orbital to produce the (dxy
xz,yz )
(S = 1/2) conguration is simply much more favorable than the alternative
1 )(d1 , d1 ) (S = 3/2) conguration. It is noteworthy that the (d2 )(d1
(dxy
xz yz
xy
xz,yz )
(S = 1/2) is (almost) orbitally degenerate and JahnTeller active. Thus, there
are two low-lying Kramers doubles which explains the extremely unusual EPR
properties of the system. In fact, clean and reproducible EPR data could not
be obtained.
It should be evident from the preceding description how a close interaction
between theory and experiments can lead to essential insights on previously
elusive species.
78
3.5
Binuclear Mixed Valence Copper Complexes
ENDOR and HYSCORE Spectra Simulations from DFT
Our last example deals with a combined theoretical and experimental study
on a set of two delocalized class III mixed valence Cu1.5 /Cu1.5 , S = 1/2, complexes [94]. The structures of the two investigated systems are given in Fig. 9.
The investigated species serve as model systems for the active sites of nitrous
oxide reductases and cytochrome c oxidases.
From the experimental point-of-view, W- and X-band ENDOR, HYSCORE,
and 2D-TRIPLE were used to obtain 1 H, 2 H, and 14 N HFCs and signs. The
main goal of the accompanying DFT calculations was to provide help for the
spectral simulations and the assignment of the experimental parameters as
there was a substantial amount of spectral crowding and the analysis of the
experimental data turned out to be extremely challenging. In addition, this
study can be regarded as a further benchmark for the assessment of the accuracy that can be obtained in DFT calculations on transition-metal systems.
In order to probe the inuence of molecular structures on the calculated EPR
parameters, results from X-ray crystallography and geometry optimizations
employing the BP86 functional were compared.
Considering the electronic structures, similar SOMOs were calculated for
the two complexes. They are formed by the antibonding combination of the
two dz2 orbitals, but with covalent admixtures of the axial nitrogen ligands.
The spin densities of the two model systems are mainly determined by the
shape of the SOMOs. Noticeable spin populations were found on all eight nitrogen atoms of both complexes. Thus, a large number of superhyperne 14 N
and 1 H contributions to the spectra can be expected in addition to the molecular g- and Cu hyperne tensors. The calculated and measured g-values
and Cu HFCs of model complex Cu2 RGT are given in Table 4 and are in unexpected good agreement with the experiments. This was explained by the fact
that DFT works better for a delocalized system since the exchange hole is also
Fig. 9 The binuclear mixed valence CuI /CuII complexes with azacryptand ligands (left
and middle). The overlap of the two dz2 orbitals with three electrons results in a rather
strong doubly occupied Cu Cu bond (right), which denes also the z-axis of the system (g|| ). The single unpaired electron resides in the antibonding combination of the two
dz2 orbitals and is completely delocalized over the two metal centers
79
Table 4 Comparison of calculated and experimental g-values, HFCs and NQCs (MHz)
obtained for the model complex Cu2 RGT from Fig. 9
g
g||
Cu A
Cu A||
Nax A
Nax A||
Nax e2 Qq/h
Neq A
Neq A||
Nax e2 Qq/h
Calcd.
Exptl.
2.1378
2.0033
305.5
0.3
14.3
31.5
4.0
5.1
9.8
3.4
2.148
2.004
308
10
14.5
26.5
3.7
5.5
7
3.0
too delocalized in comparison to the exact one. Data of almost similar high
quality were obtained for the Cu2 GT species. Interestingly, the agreement between theory and experiment is typically much worse for mononuclear CuII
complexes. Nitrogen HFCs and NQCs were also well reproduced in the calculations (Table 4). These parameters were used for the calculation of W-band
and HYSCORE spectra of Cu2 RGT, which impressively underlines the possibilities of present-day electronic structure calculations. Finally, 1 H HFCs were
reported. Focusing on the N H protons, an overestimation of the anisotropic
HFCs was found, while the predicted Fermi-contact terms were too small.
Most importantly, however, is the fact that the DFT calculations gave
a qualitatively correct description of the relative sizes of the various HFCs in
the system. Perhaps even more importantly, the calculations provide numbers
and insight for the orientation of the various HFC and NQC tensors which are
required as input for the simulation routines. Even if the calculated numbers
are slightly in error, they provide an essential help for analyzing puzzlingly
complex experimental spectra which are typical of modern high-resolution
EPR experiments. In this respect, the study in [94] provided an important
benchmark study for showing how complicated experiments and advanced
theoretical methods can come together to arrive at a very detailed experimentally calibrated spin density distribution picture in a large and complex
molecule.
4
Conclusions and Outlook
Quantum chemical calculations have been proven to be an inherent part of
modern spectroscopy. Among them, DFT calculations can be performed rou-
80
tinely for systems with more than 100 atoms and provide a large number of
spectroscopic parameters at low computational cost. The strengths of the DFT
approach are the prediction of trends in properties, tensor orientations and
signs. Especially computed Euler angles can be highly relevant for experimentalists who want to simulate spectra with a large number of parameters.
Considering the accuracy of the computed numbers, careful calibration studies are necessary. It was shown that very accurate results can be obtained
for organic doublet radicals, while the investigation of transition-metal complexes is more complicated and the quality of the results depends strongly on
the chosen systems.
Considering the calculation of hyperne coupling tensors, basis sets are
required that are uncontracted and exible in the core region. Nevertheless,
DFT tends to underestimate the isotropic Fermi-contact contributions, while
the spin-dipole contributions are calculated quite well. Furthermore, secondorder contributions must be taken into account for transition metals and
heavier elements like sulfur. The predictions of these spinorbit contributions can be substantially in error, which is evident in calculated HFCs and
g-shifts. Considering the calculation of zero-eld splitting parameters, more
calibration studies are required.
In general, one can expect a more extensive use of CI-based methods for
the calculation of spectroscopic properties in future applications, especially
for transition-metal systems. The multideterminant character of the wave
functions can be directly considered and it is possible to use spin eigenfunctions. However, the implementation of property calculations based on these
ab initio methods is a challenging task but the attractive formal properties of
these methods make them an attractive target.
Above all technical details, we hope that the present piece of work has
provided a feeling of what can be achieved presently with theoretical spectroscopy approaches and how fruitful the interaction between theory and
experiments can be in the elucidation of the structure and properties of complex and unstable molecules. While this line of research has already produced
a substantial number of important contributions there is every reason to expect that the hand-in-hand use of experiments and calculations will become
more and more commonplace in the immediate future.
References
1. Schweiger A, Jeschke G (2001) Principles of Pulse Electron Paramagnetic Resonance.
Oxford University Press, Oxford
2. Kahn O (1993) Molecular Magnetism. VCH Publishers, Weinheim
3. Blundell S (2001) Magnetism in Condensed Matter. Oxford University Press, New
York
4. Jiles D (1998) Introduction to Magnetism and Magnetic Materials. CRC Press, Boca
Raton, Florida
81
5. Praneeth VKK, Neese F, Lehnert N (2005) Inorg Chem 44:2570

6. Stich TA, Buan NR, Brunold TC (2004) J Am Chem Soc 126:9735
7. Quintanar L, Gebhard M, Wang TP, Kosman DJ, Solomon EI (2004) J Am Chem Soc
126:6579
8. Kirk ML, Peariso K (2003) Curr Opin Chem Biol 7:220
9. Neese F, Solomon EI (1998) J Am Chem Soc 120:12829
10. Neese F, Zaleski JM, Zaleski KL, Solomon EI (2000) J Am Chem Soc 122:11703
11. Loeb KE, Zaleski JM, Westre TE, Guajardo RJ, Mascharak PK, Hedman B, Hodgson KO, Solomon EI (1995) J Am Chem Soc 117:4545
12. Robblee JH, Messinger J, Cinco RM, McFarlane KL, Fernandez C, Pizarro SA, Sauer K,
Yachandra VK (2002) J Am Chem Soc 124:7459
13. Garner CD (1991) Adv Inorg Chem 36:303
14. Solomon EI, Szilagyi RK, DeBeer George S, Basumallick L (2004) Chem Rev 104:419
15. Johnson CE (1975) In: Gonser U (ed) Mssbauer Spectroscopy (Topics in Applied
Physics). Springer, Berlin, p 139
16. Neese F, Petrenko T, Ganyushin D, Olbrich G (2006) Coord Chem Rev (in press)
17. Neese F (2006) J Biol Inorg Chem 11:702
18. Sinnecker S, Neese F (2006) J Comput Chem 27:1463
19. Neese F (2006) Special Per Rep EPR Spec (in press)
20. Grifth JS (1964) The Theory of Transition Metal Ions. Cambridge University Press,
Camebridge
21. Kaupp M, Bhl M, Malkin VG (2004) Calculation of NMR and EPR Parameters. Theory and Applications. Wiley, Weinheim
22. Neese F (2003) Curr Opin Chem Biol 7:125
23. Neese F, Solomon EI, Miller JS, Drillon M (2003) Magnetism: Molecules to Materials.
Wiley, Weinheim, p 345
24. Bencini A, Gatteschi D (1990) EPR of Exchange Coupled Systems. Springer, Berlin
25. Hess BA, Marian CM, Wahlgren U, Gropen O (1996) Chem Phys Lett 251:365
26. Neese F (2005) J Chem Phys 122:204107
27. Gtlich P, Link R, Trautwein A (1978) Mssbauer Spectroscopy and Transition Metal
Chemistry. Springer, Heidelberg
28. Gonser U (1975) In: Gonser U (ed) Mssbauer Spectroscopy (Topics in Applied
Physics). Springer, Berlin, p 1
29. Sinnecker S, Slep LD, Bill E, Neese F (2005) Inorg Chem 44:2245
30. Neese F (2002) Inorg Chim Acta 337:181
31. Zhang Y, Oldeld E (2003) J Phys Chem B 107:7180
32. Zhang Y, Mao JH, Godbout N, Oldeld E (2002) J Am Chem Soc 124:13921
33. Liu TQ, Lovell T, Han WG, Noodleman L (2003) Inorg Chem 42:5244
34. Hirata S, Head-Gordon M (1999) Chem Phys Lett 302:375
36. Rhrig UF, Frank I, Hutter J, Laio A, VandeVondele J, Rothlisberger U (2003)
Chemphyschem 4:1177
37. Burcl R, Amos RD, Handy NC (2002) Chem Phys Lett 355:8
38. Dreuw A, Head-Gordon M (2005) Chem Rev 105:4009
39. Roos BO, Andersson K, Flscher MP, Malmqvist PA, SerranoAndres L, Pierloot K,
Merchan M (1996) Adv Chem Phys 93:219
40. Neese F (2003) J Chem Phys 119:9428
41. Cramer CJ, Truhlar DG (1999) Chem Rev 99:2161
42. Tomasi J (2004) Theor Chem Acc 112:184
43. Caricato M, Mennucci B, Tomasi J (2004) J Phys Chem A 108:6248
82
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
Yan WZ, Xue Y, Zhu H, Zeng J, Xie DQ (2004) J Comput Chem 25:1833
Stoyanov SR, Villegas JM, Rillema DP (2004) J Phys Chem B 108:12175
Benzi C, Crescenzi O, Pavone M, Barone V (2004) Magn Reson Chem 42:S57
Pecul M, Ruud K (2004) Magn Reson Chem 42:S128
Cioni I (2004) Magn Reson Chem 42:S48
Cioni I, Reviakine R, Arbuznikov A, Kaupp M (2004) Theor Chem Acc 111:132
Rinkevicius Z, Telyatnyk L, Vahtras O, Ruud K (2004) J Chem Phys 121:5051
Cioni I, Adamo C, Barone V (2004) J Chem Phys 121:6710
Saracino GAA, Tedeschi A, DErrico G, Improta R, Franco L, Ruzzi M, Corvaia C,
Barone V (2002) J Phys Chem A 106:10700
Langgard M, Spanget-Larsen J (1998) J Mol Struct THEOCHEM 431:173
Barone V (1996) Chem Phys Lett 262:201
Sinnecker S, Rajendran A, Klamt A, Diedenhofen M, Neese F (2006) J Phys Chem A
110:2235
Siegbahn PEM, Blomberg MRA, Pavlov M (1998) Chem Phys Lett 292:421
Gao J (1995) In: Lipkowitz KD, Boyd DB (eds) Reviews in Computational Chemistry,
Vol 7. VCH Publishers Inc., New York, p 119
Mordasini TZ, Thiel W (1998) Chimia 52:288
Altun A, Thiel W (2005) J Phys Chem B 109:1268
Lin H, Schneboom JC, Cohen S, Shaik S, Thiel W (2004) J Phys Chem B 108:10083
Sinnecker S, Reijerse E, Neese F, Lubitz W (2004) J Am Chem Soc 126:3280
Asher JR, Doltsinis NL, Kaupp M (2004) J Am Chem Soc 126:9854
Benisvy L, Bittl R, Bothe E, Garner CD, McMaster J, Ross S, Teutloff C, Neese F (2005)
Angew Chem Int Ed Engl 44:5314
Becke AD (1988) Phys Rev A 38:3098
Perdew JP (1986) Phys Rev B 33:8822
Perdew JP (1986) Phys Rev B 34:7406
Schfer A, Huber C, Ahlrichs R (1994) J Chem Phys 100:5829
Neese F (2001) J Chem Phys 115:11080
Becke AD (1993) J Chem Phys 98:5648
Lee CT, Yang WT, Parr RG (1988) Phys Rev B 37:785
Barone V, Chong DP (1996) Recent Advances in Density Functional Methods. World
Scientic Publ Co, Singapore, p 287
Klamt A, Schrmann G (1993) Perkin Trans, p 799
Lubitz W, Feher G (1999) Appl Magn Reson 17:1
van Gastel M, Lubitz W, Lassmann G, Neese F (2004) J Am Chem Soc 126:2237
Kutzelnigg W, Fleischer U, Schindler M, Diehl P, Fluck E, Gnther H, Kosfeld R,
Seelig J (1990) NMR Basic Principles and Progress. Springer, Heidelberg, p 165
Costas M, Mehn MP, Jensen MP, Que L (2004) Chem Rev 104:939
Pestovsky O, Stoian S, Bominaar EL, Shan XP, Mnck E, Que L, Bakac A (2005)
Ghosh A, de Oliveira FT, Yano T, Nishioka T, Beach ES, Kinoshita I, Munck E,
Ryabov AD, Horwitz CP, Collins TJ (2005) J Am Chem Soc 127:2505
Neese F (2006) J Inorg Biochem 100:716
Rohde JU, In JH, Lim MH, Brennessel WW, Bukowski MR, Stubna A, Mnck E,
Nam W, Que L (2003) Science 299:1037
Price JC, Barr EW, Tirupati B, Bollinger JM, Krebs C (2003) Biochemistry 42:7497
Que L, Ho RYN (1996) Chem Rev 96:2607
Krebs C, Price JC, Baldwin J, Saleh L, Green MT, Bollinger JM (2005) Inorg Chem
44:742

84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
94.
95.
83
Schneboom JC, Neese F, Thiel W (2005) J Am Chem Soc 127:5840

Baerends EJ, Ellis DE, Ros P (1973) Chem Phys 2:41
Dunlap BI, Connolly JWD, Sabin JR (1979) J Chem Phys 71:3396
Noodleman L (1981) J Chem Phys 74:5737
Noodleman L, Davidson ER (1986) Chem Phys 109:131
Shaik S, de Visser SP, Kumar D (2004) J Biol Inorg Chem 9:661
Shaik S, de Visser SP, Ogliaro F, Schwarz H, Schrder D (2002) Curr Opin Chem Biol
6:556
Hirao H, Kumar D, Thiel W, Shaik S (2005) J Am Chem Soc 127:13007
Meyer K, Bill E, Mienert B, Weyhermller T, Wieghardt K (1999) J Am Chem Soc
121:4859
Aliaga-Alcalde M, DeBeer George S, Mienert B, Bill E, Wieghardt K, Neese F (2005)
Kababya S, Nelson J, Calle C, Neese F, Goldfarb D (2006) J Am Chem Soc 128:2017
Kostka KL, Fox BG, Hendrich MP, Collins TJ, Rickard CEF, Wright LJ, Mnck E
(1993) J Am Chem Soc 115:6746
Top Curr Chem (2007) 268: 85132

DOI 10.1007/128_2006_082
First-Principles Approach to Vibrational Spectroscopy

of Biomolecules
Carmen Herrmann Markus Reiher (u)
Laboratorium fr Physikalische Chemie, ETH Zrich, Hnggerberg Campus HCI,
Wolfgang-Pauli-Str. 10, CH-8093 Zrich, Switzerland
markus.reiher@phys.chem.ethz.ch
1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
86
2
2.1
2.2
2.3
Calculation of the Vibrational Normal Modes and Frequencies . .

BornOppenheimer Approximation . . . . . . . . . . . . . . . . . .
Harmonic Approximation . . . . . . . . . . . . . . . . . . . . . . . .
Efcient Calculation of Vibrational Normal Modes and Frequencies
3
3.1
3.2
3.3
Calculation of the Electronic Potential Energy Hypersurface

Ground State Hypersurfaces . . . . . . . . . . . . . . . . . .
Excited State Hypersurfaces . . . . . . . . . . . . . . . . . .
Environment Effects and Molecules in Solution . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
4
4.1
4.2
4.3
4.4
Calculation of Intensities in Vibrational Spectroscopy:

General Considerations . . . . . . . . . . . . . . . . . .
Interaction Between Radiation and One Molecule . . .
Orientational Averaging . . . . . . . . . . . . . . . . . .
Consideration of Experimental Conditions . . . . . . .
Efcient Calculation of Intensities for Large Molecules
.
.
.
.
.
.
.
.
.
.
5
5.1
5.2
5.3
5.3.1
5.3.2
5.3.3
5.3.4
5.3.5
5.3.6
5.3.7
5.4
5.4.1
5.4.2
5.4.3
5.5
Selective Vibrational Spectroscopic Techniques . . . . . . . .

Difference-IR and -Raman . . . . . . . . . . . . . . . . . . . .
Two-Dimensional Techniques . . . . . . . . . . . . . . . . . . .
Techniques Sensitive to Chirality: VCD and ROA . . . . . . . .
VCD Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ROA Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Molecular Conformation . . . . . . . . . . . . . . . . . . . . .
Efcient Computational Procedures for Large (Bio-)Molecules
Interpretation of Calculated Spectra for Large (Bio-)Molecules
Solvent Effects . . . . . . . . . . . . . . . . . . . . . . . . . . .
Local Contributions . . . . . . . . . . . . . . . . . . . . . . . .
Raman Techniques in Electronic Resonance . . . . . . . . . . .
Spontaneous Resonance Raman Scattering . . . . . . . . . . .
Resonance CARS . . . . . . . . . . . . . . . . . . . . . . . . .
Resonance ROA . . . . . . . . . . . . . . . . . . . . . . . . . .
NRVS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
91
91
92
93
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
95
96
96
97
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
98
98
102
103
103
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
105
105
107
110
112
113
114
115
116
116
118
118
118
122
124
124
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
125
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
126
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
86
C. Herrmann M. Reiher
Abstract Vibrational spectroscopy of biomolecules like enzymes, nucleic acids, carbohydrates, lipids, and their components, is in most cases the vibrational spectroscopy of
large molecules in aqueous solution or in vivo. Since large molecules in solution are
likely to yield conventional infrared (IR) and Raman spectra with many close-lying peaks,
spectroscopic techniques which lter out information selectively are of special interest
in this eld. Because of the large size of the investigated molecules and the lack of reliable rules of thumb for many special techniques, accurate rst-principles calculations
are an important means of interpreting the resulting spectra. First-principle calculations on biomolecules in solution have to cope with the challenges arising from the size
of the systems under study, which make selective computational techniques an essential tool in order to be able to investigate biomolecular systems of reasonable size with
sufcient computational accuracy. This is why the focus of this work is on the selective rst-principles calculation of vibrational spectra of biomolecules obtained by special
techniques such as difference IR and Raman, Vibrational Circular Dichroism, Raman
Optical Activity, resonance Raman, and also Coherent Anti-Stokes Raman Spectroscopy,
two-dimensional IR and Nuclear Resonance Vibrational Spectroscopy. For each of these
techniques, a short introduction of their relevance for studies on biomolecules is given.
Theoretical as well as practical aspects of calculating the corresponding intensities are
discussed and complemented by references to original work on these topics.
Keywords Biomolecules First-principles methods Vibrational spectroscopy
Abbreviations
CARS coherent anti-Stokes Raman scattering
CD
circular dichroism
CIS
conguration interaction using single excitations
DNA
deoxyribonucleic acid
DFT
density functional theory
HF
HartreeFock theory
IR
infrared
KS
KohnSham
MD
molecular dynamics
PCM
polarizable continuum model
PES
potential energy surface
ROA
(vibrational) Raman optical activity
RPA
random phase approximation
TDA
TDDFT time-dependent density functional theory
TDHF time-dependent HartreeFock
VCD
vibrational circular dichroism
2D-IR two-dimensional infrared (spectroscopy)
1
Introduction
In order to understand important processes of life, such as biological energy
conversion and storage, carbon and nitrogen xation, and the transcription
Vibrational Spectroscopy of Biomolecules
87
of genes, it is necessary to acquire a detailed knowledge of the underlying biochemical processes. Therefore, information is needed on the molecular structures of biologically relevant systems such as enzymes, deoxyribonucleic acid
(DNA), viruses and their components, in particular amino acids and peptides, nucleic acids, sugars, and metalloenzyme active sites. Structural information on biomolecules is in principle available by a number of spectroscopic
techniques such as X-ray crystallography [1, 2], NMR spectroscopy [3, 4], and
vibrational spectroscopy. Using X-ray crystallography, a very detailed picture of the nuclear positions within a biomolecule may be obtained directly
from experimental data, but the applicability of this technique is limited to
molecules which can be prepared in crystalline form (however, see also the recent reports on X-ray diffraction by single molecules [5, 6] and noncrystalline
samples [7]). Furthermore, the resolution may not be sufcient for an unambiguous assignment of the positions of comparatively light nuclei such as
hydrogen or, in some cases, carbon, nitrogen or oxygen [8], which may be key
atoms in biochemical reaction mechanisms. For example, the identity of the
central ligand in the FeMo cofactor of nitrogenase (which is supposed to be
N, but might also be O or C) is still not resolved [9, 10]. In addition, the need
for crystalline samples may restrict X-ray structure investigations to inactive
forms of the system under study. Vibrational spectroscopy, on the other hand,
is complementary to X-ray crystallography in the sense that it can monitor
biomolecules and their reactions (including active species and intermediates
as well as excited states) in their natural environment, which is in most cases
in (aqueous) solution, or even in vivo. However, the information vibrational
spectroscopy provides on molecular structures is not obtained directly from
experiment as in X-ray crystallography. It rather has to be extracted from
a set of vibrational frequencies and corresponding intensities either by wellestablished empirical rules based on experimental experience and symmetry
considerations, or by comparison to computed vibrational spectra.
Most experimental experience on vibrational spectroscopy has so far been
gained with traditional infrared (IR) and Raman vibrational spectroscopy.
As most biomolecules are comparatively large and have low (mostly no)
symmetry, however, the resulting IR and Raman vibrational spectra contain a large number of vibrational bands which may overlap depending on
the experimental resolution. As a consequence, new vibrational spectroscopic
techniques such as resonance Raman, Vibrational Circular Dichroism (VCD),
and Vibrational Raman Optical Activity (ROA) have been developed and
made routinely applicable, which lter out certain vibrational bands based
on special selection criteria. Because of the complicated molecular structure of most biomolecules and since experimental experience with these
techniques is often still limited compared to traditional IR and Raman, theoretical calculations are needed to assign the spectra. In order to be of use
for more than a few special cases, such simulations should be sufciently accurate and applicable to any class of biomolecules and type of spectroscopy.
88
This is provided by a rst-principles approach to the simulation of vibrational spectra of biomolecules, which is the focus of this work. A thematically
related and very comprehensive recent review on theoretical and experimental techniques for the vibrational spectroscopy of biomolecules is provided
in [11]. It is complementary to this work in the sense that it comprises numerous examples to demonstrate the current technical state-of-the-art, whereas
this work concentrates rather on conceptual issues and uses examples to illustrate an underlying idea, which is the targeted ltering of information
from vibrational spectra in experiment and in particular in rst-principles
calculations.
It is possible to circumvent the harmonic approximation in the calculation
of molecular vibrations by rst-principles molecular dynamics (MD) simulations [12] (see also Kirchner and collaborators, 2006, in this volume). In
rst-principles MD simulations, the motion of the nuclei is described by classical mechanicsonly the interaction potential is not. This review, however,
concentrates on static quantum chemistry, where the motion of the nuclei is
described in a statistical manner, that is no explicit reference is made to the
time variable in the equations of motion. A rst-principles approach to vibrational spectroscopy implies that, in contrast to molecular dynamics based
on (classical) molecular mechanics, the elementary particles in the theoretical treatment are the atomic nuclei and the electrons, not the atoms linked by
some (modeled) interaction potentials. These particles are described based
on the laws of quantum mechanics, and consequently rst-principles interaction potentials are free from any empirical parameters other than fundamental natural constants. They are thus physically correct, provided the
electronic structure of the molecules can be described with sufcient accuracy. The quantum mechanical treatment of molecules (see also Sects. 2 and 3
for more details) is usually based on the BornOppenheimer approximation [13], which decouples the description of nuclei and electrons. The nuclei
are moving on the potential energy hypersurface (PES) obtained by solving
the electronic Schrdinger equation. Furthermore, it is common to employ
the harmonic approximation, in which the molecular vibrations are treated as
a set of decoupled harmonic oscillators.
This procedure entails three problems:
1. Relevant biomolecules may easily be larger than 1001000 atoms, which
makes an accurate theoretical description of their electronic structure and
the whole set of vibrational modes a challenging or even impossible task.
This problem is aggravated if solvent effects are considered. Thus, it is
possible that no advantage can be taken from the detailed correlation between spectral information and molecular structure provided by quantum
chemical calculations. In experiment, the size of the systems under study
leads to the reverse problem: The nite resolution may render individual peaks inseparable from one another, so that, although all information
89
is present in principle, it cannot be extracted from the overwhelming

amount of data.
2. The BornOppenheimer approximation may break down at points on the
electronic PES where several electronic states are lying close in energy,
such as points near avoided crossings or conical intersections and on excited state hypersurfaces.
3. The system under study or the way in which it is studied may require
corrections to the harmonic approximation. Without these anharmonic
corrections, neither overtone nor combination bands are included in
the theoretical description. For one-dimensional vibrational spectroscopic
techniques, the neglect of anharmonicity results rst of all in frequency
shifts, whereas in two-dimensional vibrational spectroscopy, without anharmonicity the whole spectrum would vanish since positive and negative
peaks would cancel.
The anharmonicity problem is thus in particular relevant for biomolecules in
the theoretical treatment of two-dimensional vibrational spectroscopic techniques. For one-dimensional techniques, a consideration of anharmonic corrections may be necessary to answer special questions, but in general, a correct description of the intensities associated with certain selected frequencies
is more important for an unambiguous correlation between vibrational peaks
and molecular vibrations.
The possible breakdown of the BornOppenheimer approximation may
be of importance for techniques in electronic resonance, where the properties of excited states are probed. Apart from the special case of VCD and far
from resonance, employing the BornOppenheimer approximation is usually
justied.
As far as the dimensionality problem is concerned, the overcrowding of
vibrational spectra can be avoided in experiment either by reducing the information or by extending the space on which it is presented. The rst approach
results in special techniques such as resonance Raman, ROA or VCD spectroscopy, which amplify vibrational bands with certain properties and thus
lter out information. The second approach is realized in two-dimensional IR
and Raman techniques, where the intensities are plotted versus two frequency
axes instead of one. From a theoretical point-of-view, the dimensionality
problem arising from the size of the system under study can be divided into
two subproblems which are separable in the theoretical treatment as well as
in the sense that they can be carried out one after the other in actual calculations: (1) the calculation of the vibrational normal modes and frequencies
and (2) the calculation of the corresponding intensities. In order not to loose
accuracy in solving these problems by mode-selective techniques, a necessary condition is that no (or only well-controlled) approximations in addition
to those already made in the calculation of full vibrational spectra are introduced (well-controlled means that the approximate size of the introduced
90
error is known). For the normal modes, this can be achieved by using the
mode-tracking protocol [14], which makes the selected calculation of exact (harmonic) normal modes possible based on guessed vibrations that are
easily constructed for a particular scientic problem [1521]. Such iterative
renement schemes have also been applied to large fractions of a vibrational
spectrum of a large biomolecule (usually the upper or lower part of the spectrum is obtained [22, 23]) instead of the selected, narrow sections of the spectrum as in mode-tracking. The corresponding intensities can be calculated
subsequently by numerical direct differentiation of the appropriate molecular property tensors along those selected normal modes. This combination
of techniques makes the calculation of the relevant parts ofin principle
all kinds of vibrational spectra much faster, and extends the range of (bio-)
molecules accessible by accurate rst-principles methods.
To summarize, whereas the anharmonicity and the BornOppenheimer
problems are decisive only for certain techniques (often one may accept the
numerical error introduced), the dimensionality problem is of importance
for the rst-principles treatment of all types of vibrational spectroscopies for
biomolecules. Thus, this work concentrates on the selective calculation of vibrational frequencies and corresponding absorption or scattering intensities
for biomolecules, and covers only experimental techniques which lter out
special information compared to conventional IR and Raman spectroscopy.
Furthermore, the techniques should be able to investigate biomolecules in
their natural environment, that is either in vivo, in aqueous solution, or, for
certain lipophilic molecules, in organic solution. Regarding these requirements, the following techniques are of particular interest: Difference IR and
Raman techniques, which can monitor reactions, two-dimensional IR techniques, which may yield information on vibrational couplings and relaxation
dynamics, chirality-sensitive techniques such as Vibrational Circular Dichroism (VCD) and Raman Optical Activity (ROA), which probe chirality as well
as conformations, and Raman techniques in electronic resonance, which are
sensitive for electronically excited molecular states (including Coherent AntiStokes Raman Spectroscopy (CARS), which may replace resonance Raman
in the case of strongly uorescent samples). Furthermore, due to its special,
nucleus-specic selection criteria, quantum chemical calculations for Nuclear
Resonance Vibrational Spectroscopy (NRVS) are shortly summarized, which
is suited for the investigation of iron-containing systems in crystalline form.
All these techniques have in common that they probe the same set of
molecular vibrations, but differ in which (and how) subsets of these vibrations are ltered out via the corresponding intensities. As normal modes and
intensities can be calculated separately, they are treated here in separate sections: Sects. 2 and 3 summarize the calculation of vibrational normal modes
and frequencies, the latter focusing on the calculation of the electronic PES,
while Sect. 4 outlines the theoretical background of absorption or scattering
intensity calculations in general and Sect. 5 summarizes the rst-principles
91
description of several selective vibrational spectroscopic techniques. Since

the focus in our group is currently on resonance Raman and Vibrational Raman Optical Activity, these two techniques are treated in greater detail than
the remaining ones.
As far as the notation is concerned, vectors are printed in bold-face and
small letters, matrices in bold-face and capitals. As an exception, nuclear coordinates are indicated by bold-face capitals, R referring to cartesian and Q to
normal coordinates.
2
Calculation of the Vibrational Normal Modes and Frequencies
2.1
BornOppenheimer Approximation
The basis of the quantum chemical description of molecular vibrations of
(bio-)molecules with more than a few atoms is the BornOppenheimer approximation [13], which allows for a separate description of the movements
of the atomic nuclei and the electrons. These are coupled in the timeindependent nonrelativistic Schrdinger equation

nuc,e + V e,e I = EI I ,
molI = T K + T e + V nuc,nuc + V
(1)
H
(where I labels the state of the system) by the electron-nucleus-interaction
Hamiltonian V nuc,e , whereas the contributions to the total molecular Hamil mol which describe the kinetic energy of the nuclei (T K ), the kinetic
tonian H
energy of the electrons (T e ), the interaction between the nuclei (V nuc,nuc ),
and the interaction between the electrons (V e,e ) refer to either nuclear or
electronic coordinates only and thus do not prevent a separation of the
Schrdinger equation into a nuclear and an electronic part. The Born
Oppenheimer approximation is based on the adiabatic approximation, in
which the wave function I (R, r) is separated into an electronic and a nuclear part. The nuclear part a (R) depends on the 3M coordinates RA of all
M nuclei collected in the vector R, and the electronic part e,i (r,
R) depends
explicitly on the 3N coordinates r i of all N electrons (summarized as the vector r), and parametrically on the nuclear coordinates1 (which is expressed by
the tilde),

R .
(2)
I R, r/ = a R/ e,i r,
1
Parametrical dependence of the electronic wave function on the nuclear coordinates means that
these coordinates enter the electronic Schrdinger equation as constant values. Different electronic wave functions are thus obtained for different sets of nuclear coordinates, i.e., for different
molecular structures.
92
The indices a and i label nuclear and electronic energy levels, respectively,
and are dened by the composite total index I. By neglecting the action of the
nuclear kinetic energy operator onto the electronic wave function, the adiabatic approximation turns into the BornOppenheimer equation. As a result,
(r,
after inserting Eq. 2 into Eq. 1, multiplying from the left with e,i
R) and
integrating over all electronic coordinates, the nuclear Schrdinger equation

T K + Ee,i (R) a (R) = Etot a (R)
(3)
is obtained, which describes the nuclei as moving on the PES dened by
Ee,i (R) of the electronic state i. The PES may be obtained pointwise by solving
the electronic Schrdinger equation,

e,e e,i r,
T e + V nuc,nuc + V nuc,e + V
R = Ee,i R/ e,i r,
R ,
(4)
for different positions of the nuclei. This provides the potential in which the
nuclei are moving. The solution of the electronic Schrdinger equation in
practice is shortly summarized in Sect. 3. As far as the nuclear Schrdinger
equation is concerned, it is sufcient to know that it is possible to calculate
the electronic PES and its derivatives pointwise.
2.2
Harmonic Approximation
The most simple expression for the nuclear wave function a (R) is a product
of functions which each only depend on one nuclear coordinate. This product will be an exact solution of Eq. 3 if the individual coordinates are not
coupled through the terms in the Hamiltonian operator in Eq. 3. The rst
step to achieve such a decoupling is to express the dependence of the PES on
1/2
mass-weighted cartesian nuclear coordinates R(m)
A = MA RA (where the index
A refers to the x, y, or z coordinate of a specic nucleus and its nuclear mass)
as a Taylor series around the nuclear equilibrium structure R(m) = 0,

3M

E
e,i
(m)
Ee,i R(m) = Ee,i (0) +
RA
(5)
(m)
R
A
A=1
0

3M 3M
2 Ee,i
1
(m) (m)
+
RA RB + ... ,
(m)
(m)
2
RA R
A=1 B=1
and to truncate it after the second-order term (harmonic approximation,

see Fig. 1). Then, considering that the energy gradients are zero at the
equilibrium (= minimum) structure, choosing the equilibrium energy as
zero and transforming it to a set of 3M mass-weighted normal coordinates
(m)
QA which makes all mixed second derivatives of the electronic energy
93
Fig. 1 Qualitative potential energy curves and the rst four vibrational energy levels for
the anharmonic case (right-hand side) and its harmonic approximation (left-hand side)
along a mass-weighted vibrational normal mode QA(m) in the ith electronic
state (which

may be the ground state or an excited state). The total energy Ee,i {QI } is related to the
Q
Q
energy for elongations along one normal mode Ee,iI by Ee,i ({QI }) = I Ee,iI . In the anharmonic case, no vibrational energy levels are found above the dissociation level indicated
by the dotted line
(m)
2 Ee,i /Q(m)
A QB
vanish, a decoupled description is possible. The nu

clear kinetic energy operator T K = A 2 /Q(m)2
A , is also diagonal with respect to these coordinates, so that the product
a
(m)
QI
A =B

(m)
(m)
(m)
= 1 Q1
2 Q2
... M QM
(6)
is an exact solution of the nuclear Schrdinger equation in harmonic approximation, which is now separable into 3M independent equations of harmonic
oscillator form with a mass equal to unity (since mass-weighted coordinates
are employed). The vibrations of the molecule are thus described as a system
of decoupled harmonic oscillators. The eigenfunctions of harmonic oscillator equations are known, and their eigenvalues (the angular frequencies I )
are equal to the square root of the second energy derivatives. Of course, since
= 2 and = /c, vibrational frequencies and wavenumbers can be
easily calculated from the angular frequencies.
2.3
Efficient Calculation of Vibrational Normal Modes and Frequencies
The conventional approach to obtain the transformation from cartesian to
normal coordinates is to calculate all entries of the mass-weighted Hessian

(m)
matrix (the matrix containing the second derivatives 2 Ee,i /R(m)
R
of
A
B
0
94
the electronic energy with respect to nuclear coordinates), and to diagonalize

it,
Q(m)T H (m) Q(m) = H (Q,m) .
(7)
The mass-weighted normal modes are the columns of the transformation matrix Q(m) . For nonlinear molecules, the diagonalization of the Hessian yields
3M 6 vibrational normal modes, as well as three rotational and three translational modes with eigenvalues of zero.
By diagonalizing the Hessian matrix, all vibrational normal modes and frequencies of the system under study are obtained. In the special vibrational
spectroscopic techniques which are discussed in this work, however, the focus
is on ltering out selected vibrations from an overcrowded vibrational spectrum. A calculation of all modes and frequencies would therefore produce
a large amount of rather useless data at high computational cost. This can
be circumvented by the mode-tracking protocol, which allows for the selective calculation of vibrational frequencies and normal modes from eigenpairs
of the Hessian matrix through subspace iteration [14, 17] (for representative
applications, see [1520, 24, 232]). The mode-tracking protocol can be summarized as follows:
(m,0)
1. Guess a vibration Q , which may be chosen based on chemical intuition

or on a precedent vibrational analysis with a lower level of accuracy.
2. Use this guessed vibration as a rst basis vector b1 in a Davidson-like
algorithm [25]:
b1 = Q(m,0)
.
(8)
(i) from the Hessian matrix H (m)

3. Construct the small Davidson matrix H
and the matrix B(i) which contains all basis vectors of iteration i as
columns,
(i) = B(i)T H (m) B(i)
H
(9)
and diagonalize it:

(i) (i)
(i) c(i)
H
= c
(10)
of the Hessian from c(i)

4. Calculate the approximate eigenvectors Q(m,i)
and
(i)
B :
=
Q(m,i)
i

(i)
c,j bj
j=1
5. Check for convergence.
(11)
95
6. If the convergence criteria are not satised, create one new basis vector per
(m)
(i)
tracked normal mode Q from the corresponding residuum vector r ,
r (i)
=
i

(i)
c,j H (m) bj (i)
b
j
(12)
j=1
and go back to step 3 (for details on convergence criteria and creating new
basis vectors using a preconditioner see, e.g., [14]).
In order to calculate the matrix product B(i)T H (m) B(i) without computing the
Hessian matrix H (m) , a matrix = H (m) B(i) with columns j is dened. These
j are evaluated by applying the chain rule for differentiation,

2 Ee,i
2 Ee,i
j,I =
bj,B =
(m)
R(m)
R(m)
A RB
A bj 0
B
0

Ee,i
=
.
(13)
bj R(m)
A
0
The components of j can thus be calculated as numerical derivatives of the

gradient components along the basis vectors bj . Several eigenvectors may
be optimized simultaneously using the DavidsonLiu (also known as blockDavidson) method [26, 27].
In contrast to other techniques such as partial Hessian vibrational analysis [28], where a block corresponding to an interesting subsystem is cut out of
the Hessian matrix and diagonalized separately, mode-tracking will provide
exact normal modes and frequencies if the convergence criteria are chosen
sufciently tight. For systems with a repeated structural motif, such as the
amide bond in peptides, vibrational modes which are dominated by motions
of the atoms belonging to this structural motif may also be calculated selectively in an approximate fashion by transfer of molecular properties from
small model systems to larger molecules of interest. A general algorithm relying on the transfer in Cartesian coordinates instead of internal ones has been
developed by Bour and coworkers [2931], and rened for arbitrary polypeptides in solution by Choi et al. [3234]. This scheme may also be applied
to the property tensors determining the spectral intensities, and has accordingly been applied to IR, Raman, VCD and ROA spectroscopy [29] (see also
Sect. 5.3).
3
Calculation of the Electronic Potential Energy Hypersurface
In the context of vibrational spectroscopy, the aim of calculating the electronic PES by solving the electronic Schrdinger equation is to provide
96
a physically correct potential Ee,i (R) in which the nuclei are moving (Eq. 3).
Furthermore, the shapes of excited state energy surfaces are needed for spectroscopic techniques in or near electronic resonance. The PES can only be
calculated pointwise, but an analytical form describing the dependence of the
total energy on the nuclear coordinates, such as the truncated Taylor series
around the molecular equilibrium structure in the harmonic approximation,
may be approximated based on these energy values at certain points and
derivatives of the energy with respect to nuclear displacements.
3.1
Ground State Hypersurfaces
In order to provide physically reasonable electronic energies and its derivatives, a fast but sufciently accurate electronic structure method is needed.
Currently, KohnSham density functional theory (KS-DFT) [3537], which
in principle fully describes electron correlation, is the most general rstprinciples method which meets these requirements, and in practice the only
one which is applicable to large biomolecules. Its efciency is comparable
to that of HartreeFock calculations and can be further enhanced by using
density-tting techniques [38, 39] provided pure density functionals are employed. Because of the size of the systems under study, relativistic effects
required to be described if heavy elements are present are usually neglected,
with the exception of the scalar-relativistic effects included in effective core
potentials (see, e.g., [40]).
3.2
Excited State Hypersurfaces
Excited state hypersurfaces can be calculated either by explicit methods or by
using linear response theory. The rst class comprises methods such as Conguration Interaction Singles (CIS, compare also the TammDancoff approximation (TDA) [41, 42]) or Complete Active Space theory (CASSCF/CASPT2,
see [43] for a review on applications to various organic photosystems),
while the latter encompasses time-dependent HartreeFock (TDHF, also
known as random phase approximation (RPA) in physics) as well as the
electron-density based time-dependent DFT (TDDFT). Since many important biomolecules are far too large to be treated entirely by multi-reference
methods, it is desirable to be able to calculate accurate excited state energies and properties with rst-principles methods based on a single Slaterdeterminant. An extensive review of such methods is provided by [44]. As
an alternative approach, multireference ab-initio methods may be combined
with a MM force eld within a Quantum Mechanics/Molecular Mechanics
(QM/MM) framework, which has been successfully applied, e.g., to the excited state of the visual pigment rhodopsin [45, 46]. For a comparative review
97
on different quantum chemical methods applied to excited state calculations

on biomolecules see [47].
3.3
Environment Effects and Molecules in Solution
In order to study the properties of (bio-)molecules with rst-principles
methods, it is necessary to separate a suitable subsystem from its environment and treat it as an isolated system [48, 49]. In order to achieve this in the
case of molecules in solution, it is often necessary to consider not only the solute, but also the surrounding solvent. Solvent effects may change the nuclear
arrangement in the molecular equilibrium structures, as well as vibrational
normal modes and frequencies, or electronic properties by changing the wave
function/electron density compared to the free molecule (see [50] for the
quantum chemical description of solvent effects on a biomimetic arginine
fork model). They can be modelled either explicitly in the form of solvent
molecules which are attached to the molecule under study at appropriate positions, or implicitly in the form of dielectric continuum models, or by a combination of both. A perfect description of solvent effects is achieved within
the CarParrinello molecular dynamics framework under periodic boundary
conditions. To bridge the gap to quantum chemical isolated-molecule calculations is not trivial [51]. A very extensive up-to-date review on different
continuum models and their application to molecular property calculations
is provided by [52]. Reasonable positions and orientations of explicit solvent
molecules may be either obtained by intuition or from MD snapshots. For
a review on CarParrinello Molecular Dynamics for biological systems, see
the article by Kirchner et al. in this volume. In the case of large biomolecules
with a localized region of interest, such as metalloproteins, it may be necessary (and sufcient) to cut out this region from the rest of the molecule
and to treat it either separately or to model the remainder in a more approximate fashion, e.g., in the framework of a QM/MM approach [53] (see
also the contribution by Senn and Thiel in this volume). For a recent review on the technical state-of-the-art of the QM/MM-treatment of complex
biological processes, see [54], and for applications of QM/MM to enzyme
catalysis [5557]. Another approach to such problems is frozen-density embedding [58], where the system under study is surrounded by a larger system whose effect is represented by an orbital-free density which enters in
terms of an effective potential in the KohnSham equations for the system
under study. This approach is also applicable to excited state calculations
within TDDFT [59].
98
4
Calculation of Intensities in Vibrational Spectroscopy:
General Considerations
In order to provide a basis for the theoretical treatment of the more special vibrational spectroscopic techniques, the concepts of the calculation of IR
and Raman intensities shall be outlined in the following. For further details
on how to compute IR and Raman intensities, we refer to [60]. The theory
of light absorption or (noncoherent) scattering by molecules in solution can
be divided into three steps, starting from the most simple model case and
approaching experimental conditions stepwise:
1. Description of the interaction between radiation and one molecule.
2. Orientational averaging if the molecules are randomly oriented.
3. Consideration of experimental conditions like environment (usually solvent) effects, temperature, ...
4.1
Interaction Between Radiation and One Molecule
It is in principle possible to treat both the radiation and the molecule either
classically or quantum mechanically. In order to obtain quantitatively correct
absorption and scattering intensities, however, the molecule under study has
to be described by the laws of quantum mechanics, whereas the radiation may
be treated by classical electrodynamics. This combination is called a semiclassical treatment. For simple processes, it yields the same intensity expressions
as a (more tedious) quantum mechanical treatment of the radiation (but fails
in special cases such as spontaneous emission). A molecule in the electromagnetic eld of a light wave is described within the semiclassical approximation
by a Hamiltonian
=H
mol + H
mol-rad ,
H
(14)
mol is the Hamiltonian of the isolated molecule dened in Eq. 1, and

where H
mol-rad a Hamiltonian describing its interaction with the radiation. In a fully
H
rad describing the radiation
quantum mechanical treatment, a third term H
When a molecule is interwould be included in the total Hamiltonian H.
acting with radiation, essentially three things (or a combination of them)
can happen: The molecule can emit light stimulated by the radiation and
change to a lower-energy state, it can absorb light resulting in a transition to
a higher-energy state, and it can scatter light and either keep its initial state
(Rayleigh scattering) or undergo transitions upon scattering (Raman scattering). The goal of the theoretical treatment is to derive expressions for the
emission, absorption, or scattering intensities, respectively, related to each of
these processes. The intensity expressions will depend on the energy levels of
99
the molecule as well as on the energy and thus the frequency of the radiation,
and should be so general that they may be adopted for any relation between
these energies.
mol-rad to the total Hamiltonian is
The contribution of the Hamiltonian H
mol, so that interactions
usually considered much smaller than the one of H
with an oscillating electromagnetic eld
E of a light wave with an angular
frequency L and a (complex) amplitude
E0 ,

E =
E0 eiL t ,
(15)
are usually described by perturbation theory. The perturbation operator

mol-rad in the case of a dynamic electromagnetic eld with electric and magH
netic eld amplitudes E0 and B0 can be written for an uncharged molecule
as [61]
mol-rad =
H
3

i=1
3
3
3
1

i Ei 0
m
i Bi 0 + ... ,
ij i Ej 0
3
i=1 j=1
(16)
i=1
i are the x, y, and z components of the electric dipole,

where
i , ij , and m
the electric quadrupole, and the magnetic dipole operator, respectively, and
(Ei )0 , (i Ej )0 , (Bi )0 etc. are the x, y, and z components of the electric and
magnetic eld and its gradients at the gauge origin, which vary with time, of
mol-rad is therefore time-dependent, so that a timecourse. The Hamiltonian H
dependent version of perturbation theory has to be employed. It has been
found that it is often sufcient in this context to restrict the perturbational
treatment to rst order.
The intensity of the radiation scattered by a molecule can then be obtained
by calculating the (oscillating) electric and magnetic multipole moments induced in the molecule by the light wave and treating them as classical sources
of radiation. For Rayleigh scattering, the induced multipole moments are oscillating with the same phase and frequency as the incident radiation and
can be computed from the expectation values of the molecular multipole operators using the perturbed time-dependent molecular wave functions and
subtracting the corresponding permanent molecular multipole moment from
this expectation value. Raman scattering can be described within a semiclassical formalism by introducing transition multipole moments between an
initial state |i and a nal state |f ,

(1)
(1)
=
f
|
|i
,
(17)
which are oscillating with an angular frequency equal to L (Stokes) or

L + (Anti-Stokes), and are unrelated in phase to the incident radiation
due to the random phase of the molecular vibrations.
For ordinary Raman scattering, only a linear dependence of the induced
electric dipole moment on the electric eld amplitude is considered. The
100
(1)
induced electric dipole transition moment

is thus proportional to the
0
amplitude
E of the oscillating electric eld of the radiation,
E0 .
=
(1)
(18)
Higher-order effects can be included by considering nonlinear dependencies

on the electric eld described in terms of (higher-order) hyperpolarizabilities. For the sake of simplicity of notation, complex quantities are used in
Eq. 18 and in the following, denoted by a tilde, although for Raman scattering, only the real part of the induced dipole moment is needed [62, 63]. The
(electric dipoleelectric dipole) polarizability
can be obtained from timedependent perturbation theory as [64]

|r r|
f |
|r r|
|i
|i
1 f |
+
=
, {x, y, z} . (19)

ri L 12 ir
rf + L + 12 ir
r =f ,i
is a component of the molecular electric dipole operator, and r is a damping factor equal to the inverse of the life time of the molecular state r. The
oscillating induced dipole described by Eq. 18 is treated as a classical source
of radiation, which results in the intensity formula [65]
(1)
I() =
4 | |2 sin2 ()
32 2 c3 0
(20)
for the radiation scattered at an angle relative to the direction of the incident beam. c is the speed of light, 0 the permittivity of the vacuum, and is
the angular frequency of the scattered radiation. In the case of spectroscopic
techniques which probe molecular chirality, i.e., VCD and ROA, restricting
the theoretical treatment to induced dipole moments in the form of Eq. 18
would yield intensity differences of zero for all vibrations. Therefore, contributions from the molecular induced magnetic dipole and electric quadrupole
moments have also to be taken into account, and Eqs. 20 and 18 have to be
adjusted accordingly (see [64]).
In practice, the transition polarizability components ( ) cannot be
evaluated in the form given by Eq. 19, since this would require knowledge
about all excited molecular states |k. It can be simplied, however, by making
assumptions which are justied for special cases, and thus restrict the applicability of the resulting formulae to those special cases. In the case of nonresonant Raman scattering, the frequency of the incident radiation is assumed
to be far below electronic transition frequencies of the molecule (which also
implies that the damping factors can be neglected), and the electronic ground
state to be nondegenerate. As mentioned in Sect. 2, by employing the adiabatic approximation, the molecule can be treated as moving on an electronic
101
energy hypersurface. In analogy to the simplication of the dependence of

the electronic energy on nuclear coordinates in the calculation of vibrational
motions and frequencies, the polarizability is therefore commonly expanded
into a Taylor series, which is truncated after the rst-order term (electric harmonic approximation)2 ,

(Q) = 0 +
Qp + ... .
(21)
Qp
p
0
Then, the real part of Eq. 19 can be simplied to

= 0 f |i +
f |Qp |i ,
Qp
p
(22)
which is known as Placzeks approximation. |i and |f are abbreviations for

the nuclear wave functions of the initial and the nal state. Since |i and |f
are approximated as products

of harmonic oscillator eigenfunctions (Sect. 2),
the matrix elements f |Qp |i can be deduced from the quantum mechanical treatment of the harmonic oscillator. For vibrational transitions involving
only the change of one vibrational normal mode from quantum number n to
n + 1, they are equal to /2p , and zero in all other cases. In the case of resonant Raman scattering, Eq. 19 is simplied by the fact that the summands
containing the electronic state in resonance are much larger than the others.
This is treated in more detail in Sect. 5.4.
For absorption associated with a molecular transition from the initial
state |i to the nal state |f , the transition rate can be calculated via timedependent perturbation theory and is given by

dPf i 2
1
mol-rad |i |2 (E ) = | f |H
mol-rad|i |2 ( ) ,
=
| f |H
(23)
2
dt

which is Fermis Golden Rule. The transition rate is the change of the probability to nd the system (which was originally in the initial state |i) in the
nal state |f w.r.t. time, i.e., the number of transitions |f |i per time.
The prefactor in Eq. 23 depends on whether the density of states per energy
range (E ) is expressed using the energy of the states or the associated frequencies. By considering only the electric-dipole term in the perturbation
operator in Eq. 16 and based on similar considerations as for the transition
polarizability, the transition matrix elements in Fermis Golden Rule given by
Eq. 23 can be calculated from the derivatives of the
dipole moment
molecular

along mass-weighted vibrational normal modes /Qp 0 (see [60] for more
details).
2
In contrast to this, the expansion is truncated after the second-order term in the mechanical harmonic approximation. Since the quantum chemical calculation of Raman intensities involves both
approximations, this is then called the double harmonic approximation.
102
To summarize, the following approximations are made in the established

theoretical double harmonic approximation of IR and Raman spectroscopy:
1. BornOppenheimer approximation.
2. (Mechanical) harmonic approximation.
3. Semi-classical treatment: The contributions from classical electrodynamics are expressions for the dependence of the induced multipole moments
on the incident radiation, as well as for the dependence of the scattered radiation on the induced multipole moments. Quantum mechanics provides
vibrational frequencies and transition matrix elements and thus accounts
for all quantities depending on the molecule.
4. The time-dependent perturbation theory employed for the derivation of
Eqs. 18 and 23 is restricted to rst-order.
5. Only the electric dipole moment induced by the incident radiation is considered, and all other induced electric and magnetic multipole moments
are neglected.
6. The electromagnetic eld of the incident radiation is considered to be constant over the whole molecule at a given time, since the wavelength of the
radiation is usually larger than the molecular region which is affected by
the interaction with it. A typical wavelength employed for Raman measA, whereas typical bond distances
urements would be, for example, 5000
A. In IR spectroscopy, this ratio is more favorable due
are of the order 12
to the larger wavelengths of the incident radiation.
7. Molecular property tensors are assumed to depend linearly on nuclear
distortions close to the equilibrium structure (electrical harmonic approximation).
4.2
Orientational Averaging
In order to obtain IR and Raman intensity expressions for molecular gases,
solutions, and uids, i.e., for a large number of randomly oriented molecules,
classical orientational averaging of the intensity expressions derived for one
molecule may be employed. Since the radiation scattered in a spontaneous
Raman process is incoherent due to the random phase of the molecular vibrations [65], the intensity of the radiation scattered by N molecules is N
times the intensity scattered by one molecule, averaged over all possible
molecular orientations. Orientational averaging can be achieved by replacing space-xed polarizability tensor elements by the classical average over
molecule-xed polarizability elements for all possible molecular orientations.
As a result, intensity formulae depending on the polarization of the incoming
and the detected radiation beam as well as on the angle between both beams
are obtained. For details on this topic, we refer to [60], which also contains
a detailed comparison of the calculation of IR and Raman spectra.
103
4.3
Consideration of Experimental Conditions
The third point, the inclusion of experimental conditions, comprises rst
of all the description of solvent and temperature effects. Solvent effects are
essentially a modication of the electronic potential (and property) energy
hypersurface and have therefore been addressed in Sect. 3.
Quantum chemical calculations of this hypersurface are usually based on
equations valid for a temperature of 0 K, i.e., by excluding all nuclear motion
effects. In many cases, the results obtained are close to those which are measured at higher temperature (e.g., room temperature). As far as temperature
effects are concerned, they can be included into the intensity expressions by
assuming a Boltzmann distribution of the molecules over initial vibrational
energy levels. Furthermore, if a molecule has several conformations which are
so close in energy to the ground state that at experimental temperatures they
are non-negligibly populated in the sample, the inuence of the temperature
onto the vibrational spectra may be incorporated by calculating spectra for
every relevant conformation and superimposing them with weighting factors
calculated by assuming a Boltzmann distribution of the different conformations. The relative energies of the molecular conformations may of course be
changed by solvation.
Another feature which is affected by the environment is the line broadening of spectral peaks. In contrast to MD simulations (see, e.g., [66]), this
information is not obtained from a static quantum chemical calculation. It
may be modelled, however, by plotting Gaussian or Lorentzian functions
centered at the calculated vibrational frequencies and using an appropriate
halfwidth (which may be in the range of 8 to 30 cm1 , for example).
4.4
Efficient Calculation of Intensities for Large Molecules
The main focus of this work is on the efcient calculation of the relevant
parts of vibrational spectra obtained by special techniques which lter out
information compared to conventional IR and Raman spectroscopy. On the
basis of the approximations outlined above, vibrational absorption and scattering intensities are usually calculated from derivatives of components of
molecular property tensors with respect to mass-weighted normal coordinates. This applies to traditional IR and Raman spectroscopy as well as to
many special techniques such as difference IR or ROA. These derivatives
along mass-weighted normal modes are commonly obtained by calculating
all cartesian derivatives of the property of interest and then transforming
them to a normal mode basis. This automatically yields derivatives along all
normal modes of the molecule. One is often interested, however, in intensities
associated with a few characteristic vibrations or within a certain frequency
104
range only. In such cases, direct differentiation along selected mass-weighted

normal modes allows for an efcient calculation of the relevant derivatives
alone [67]. It furthermore allows for restricting the calculation of the normal
modes to the interesting ones, in particular by employing the mode-tracking
protocol described in Sect. 2.
The numerical derivative of any molecular property A (like, e.g., the polarizability or the electric dipole moment) along the mass-weighted normal
mode Q(m)
p is given by [67]

norm A R s Rnorm
A
R
+
s
R
eq
R
eq
R
p
p
A
,
(24)
=
norm
(m)
2sQp |Qp |
Qp
where sR is the step-size in (non-mass-weighted) cartesian coordinates, and
sQp is a dimensionless normal-mode specic step size equal to
3M
1/2

[unit of length]
norm 2
s Qp = s R
(Qpi ) /Mi
,
(25)
[unit of mass]1/2
i=1
and the superscript norm indicates normalized coordinates.

Whereas the calculation of the cartesian property derivatives can be done
either analytically or numerically (both possibilities have been implemented
in various quantum chemical programs), the direct differentiation along
mass-weighted normal modes has in practice to be carried out numerically.
In principle, analytical derivatives can be calculated faster than numerical
ones, and some authors argue that the lacking implementation of analytical
derivatives for certain special techniques such as ROA is a serious drawback for rst-principles calculations of the corresponding intensities (see
e.g., [68]). This may be true for small molecules, but in the case of many
comparatively large biomolecules, the selective numerical differentiation has
the opposite effect, since it enhances the computational efciency and renders larger molecules accessible to calculations. In addition, these numerical
methods are facile to implement and offer easy parallelization and restart
options. These advantages are, in principle, all available in pure analytical
schemes, but their development is rather complicated and tedious as analytic
electronic energy derivatives are to be derived for various electronic structure
methods.
Another possibility of making the prediction of vibrational spectra more
efcient is to employ computational methods of different quality for the
normal modes and the molecular property tensors. Since molecular polarizabilities, especially the generalized ones needed for the calculation of ROA
spectra, require a higher computational effort than the molecular energies, it
may speed up the calculation of Raman and ROA spectra considerably to use
a small basis set for the polarizabilities. It turned out that Raman and ROA
intensities can be calculated with reasonable accuracy within HartreeFock
105
and density functional theory by employing surprisingly small, tailored basis sets for the property tensors in combination with more accurate normal
modes [69, 70]. This is due to error cancellations when calculating the derivatives of the property tensorsalthough the absolute values may be wrong,
their dependence on nuclear coordinates may still be reproduced correctly
when employing numerical differentiation (see [60] for a discussion of this effect in the context of the electric dipoleelectric dipole polarizability tensor).
5
Selective Vibrational Spectroscopic Techniques
In this section, selective vibrational spectroscopic techniques are shortly introduced, and principles and applications of the corresponding quantum
chemical intensity calculations are summarized.
5.1
Difference-IR and -Raman
Difference-IR and -Raman spectra can be obtained by recording a series of
conventional IR or Raman spectra during the course of a reaction and presenting them as intensity differences with respect to the spectrum of the
educts. In contrast to techniques such as resonance Raman or VCD and
ROA, the selection of certain bands from the full vibrational spectrum is
not achieved by ltering out certain scattering or absorption components via
the experimental setup, but by plotting the difference of time-resolved IR or
Raman spectra. Molecular vibrations involving mainly atoms which either
exhibit large displacements along the reaction path, or which are indirectly
affected by the reaction through changes in the electron distribution, will
show up as positive or negative peaks in the difference spectra, while the
bands associated with all other parts of the molecule cancel upon subtraction. Provided the observed bands can be assigned unambiguously to (local)
molecular vibrations, difference spectra can thus answer the question which
parts of a molecule are involved in a reaction. Furthermore, information on
the reaction kinetics and on the change of electron density distribution can
be obtained.
Difference vibrational spectroscopy is a comparatively young technique
(early time-resolved difference IR spectra have been reported in 1980 [71]),
since it requires at the same time a high sensitivity of the experimental setup
and a high stability of the measuring conditions. Difference IR spectroscopy
has been applied successfully to protein reactions (see [72] for an up-to-date
review, [73] for a review of IR spectroscopy of proteins in general including a section on difference IR, and [74] for a very recent application to light
absorption by photoreceptors). Difference Raman spectroscopy, besides its
106
applicability to proteins [75], is particularly suited for the investigation of localized interactions in nucleic acid complexes (for a review, see [76], for an
example of its application to DNA recognition by proteins, [77], and [78] for
a survey of difference Raman spectroscopy of biomolecules in general).
Vibrational bands can be experimentally assigned to molecular regions
by comparison to vibrational spectra of reference systems obtained either by
controlled mutation of one selected amino acid residue in peptides, or by
site-specic isotope labeling. Because of the large effort associated with assuring that the desired reference compound is synthesized, and of recording the
spectra under exactly the same experimental conditions as the original sample, quantum chemical calculations can be valuable for reliable and efcient
band assignments.
Although most reactions take place at a limited region of the protein only,
interactions with the peptide chain and possible conformational changes in
the protein may play a central role during its course. Thus, it may be necessary to include the whole protein structure in a simulation. Difference IR
spectroscopy can be applied to proteins of up to 100 kDa, including proteins
of the cell membrane. Since systems of that size are far from being accessible
by pure rst-principles methods, QM/MM-calculations are of crucial importance. QM/MM simulations of vibrational spectra of biomolecules focussed,
for instance, on bacteriorhodopsin [7981], the GTPase Ras [82, 83], and CO
binding in myoglobin [84].
For biological systems which shall be modelled as close to their natural
state as possible, the choice of a molecular structure as a starting point of the
vibrational analysis is not obvious. In principle, the calculation of vibrational
frequencies and normal modes within the harmonic approximation requires
that the molecule under study is in an energy minimum with respect to all
nuclear degrees of freedom. But a geometry optimization of the whole system
might lead to a structure which is far away from the one adopted in solution
or during the reaction. However, the harmonic approximation can also be
justied for partly optimized structures if only those modes are interpreted
which are centered on the optimized molecular subsystem [21]. Therefore, it
is possible to perform a constrained optimization for only the QM subsystem starting from an X-ray structure (as done in [84]) or from a snapshot
taken from a preceding MD simulation (as in [82]), and to use it for a subsequent vibrational analysis. This vibrational analysis is usually performed
by simply diagonalizing that part of the QM/MM Hessian matrix which contains the QM atoms [66, 82, 84], since the calculation and diagonalization of
the whole QM/MM Hessian matrix would imply an enormous computational
effort producing a lot of useless information. In principle, it is also possible
to calculate selectively the exact normal modes of the QM subsystem only
from the complete QM/MM Hessian matrix by employing the mode-tracking
protocol. This approach will be applied to myoglobin in a separate study to
be published by the authors. It has the advantage of yielding normal modes
107
of the whole QM/MM system, whereas the diagonalization of parts of the

Hessian matrix would entail an uncontrollable loss of accuracy due to the
neglected coupling of the QM system to the MM system. Molecular dipoles
and their derivatives have been implemented into QM/MM schemes [85], so
that IR intensities can be calculated within a QM/MM approach. The simplest
computational approach to vibrational difference spectra is then to calculate
the spectra of educts and products separately and to subtract them. Furthermore, since IR difference spectra can be recorded with a time resolution
down to a few ns or even further if the step scan technique is applied [72], it
is possible to extract pure difference IR spectra of intermediate products on
the reaction path using mathematical techniques such as multivariate curve
resolution [86] or factor analysis combined with singular value decomposition [87], both of which do not rely on special assumptions on the nature of
the reaction. By comparison of these extracted intermediate spectra to simulated ones for assumed intermediate structures, very detailed information on
the reaction path should be obtained.
5.2
Two-Dimensional Techniques
Two-dimensional vibrational spectroscopy is based on the idea of recording
IR (or Raman) spectra of molecules in selectively excited vibrational states.
Because of the anharmonicity of the PES and vibrational couplings, positive
and negative intensity differences arise with respect to the ordinary onedimensional IR (or Raman) spectrum. There are two different experimental
setups, which yield similar two-dimensional spectra [88]: double-resonance
spectroscopy (or dynamic hole burning), and pulsed Fourier transform spectroscopy (or heterodyne-detected photon echo spectroscopy). Both are compared from a theoretical as well as from an experimental point-of-view
in [88]. In the former approach, spectra are obtained by exciting the sample
with a laser beam (pump pulse) with a narrow frequency range and recording a Fourier Transform-IR spectrum with a second laser beam (probe pulse)
covering the whole frequency range of interest. By recording such spectra
for different frequencies of the pump pulse, a two-dimensional IR spectrum
with two frequency axes is obtained. Pulsed Fourier transform spectroscopy
consists of sending three consecutive ultrashort laser pulses with dened
time delays through the sample and Fourier transforming the outgoing signal with respect to the delay times between the rst and the second pulse
and the time after the third pulse. Because of the tunable time-delays with
a time-resolution down to picoseconds, two-dimensional IR (2D-IR) spectroscopy is particularly suited for the investigation of vibrational relaxation
dynamics.
The rst 2D-IR experiment on biomolecules was performed by Hamm
et al. in 1998, employing the double-resonance variant for the investigation
108
of amide I bands in peptides [89]. Among others, the particular value of

two-dimensional IR spectroscopy for peptides is emphasized by a recent experimental study of a model octapeptide, indicating that 2D-IR spectroscopy
allows for a discrimination between -helical and 310 -helical secondary structures [90], which is difcult to achieve by other spectroscopic techniques.
(The introduction of [90] also provides a good overview of 2D-IR spectroscopy of peptides with many literature references. For a review of this
topic, see e.g., [91].) 2D-IR may furthermore give valuable insight into the
tertiary structure of peptides [89, 92]. For an overview of its methodology
and applications, see e.g., [9395], and [96] for a conceptual comparison to
multidimensional NMR spectroscopy.
A second dimension can also be introduced into the IR spectrum by combining it with a UV pump pulse (IR/UV double resonance spectroscopy). [97]
provides an application to the investigation of the secondary structure of various gas-phase isolated peptides. In contrast to 2D-IR, however, this method
is restricted to molecules which contain a chromophore and will not be discussed further for the sake of brevity. Two-dimensional Raman [98] and
resonance Raman [99] spectroscopy is also possible, but has so far been
applied primarily to the investigation of liquids and small molecules, respectively, rather than to large biomolecules in solution.
Two-dimensional IR spectroscopy solves the problem of interpreting IR
spectra with many overlapping bands not by ltering out some of these bands
and thus reducing information, but by spreading out the information onto
a larger space, which is provided by two wavenumber axes instead of one.
In particular, broad amide I bands which appear structureless in the onedimensional IR spectrum, may be split into several cross-peaks in the twodimensional spectrum [100]. In addition, tunable ltering mechanisms are
available in the form of the polarization of the pump and probe laser beams.
By carefully choosing these polarizations, diagonal peaks may be completely
suppressed, so that only off-diagonal coupling remains [101].
2D-IR spectra consist of pairs of positive and negative peaks. Calculations
of their intensities often focus on the contribution of anharmonicity of the
potentials of the vibrational modes, whereas the contributions of electrical anharmonicity and level-dependent dephasing dynamics are neglected. A comparative study of these three contributions is provided by [102]. The diagonal
peaks arise because of the anharmonicity of the vibrational modes (in Fig. 2,
this results in a difference in length of arrows 1 and 2: if transition 1 is excited by the pump pulse, the IR spectrum recorded with the probe pulse will
show a hole (i.e., a negative peak) in comparison to the one-dimensional IR
spectrum at the wavenumber corresponding to transition 1, and an additional
positive peak at the wavenumber corresponding to transition 2, which due to
the anharmonicity is located at slightly lower wavenumbers than the negative
peak). Off-diagonal peaks are due to couplings between different vibrations
(as can be seen from the difference between arrows 3 and 4).
109
Fig. 2 Qualitative energy levels of an anharmonic oscillator coupled to another one
For a more detailed and sophisticated explanation see e.g., [103]. If all
vibrations were strictly harmonic, so that all vibrational levels were evenspaced and no coupling would arise, no peaks would be observed in the 2D-IR
spectrum, since positive and negative contributions would appear at the same
positions and cancel exactly. In order to describe 2D-IR spectra, it is thus
necessary to go beyond the (mechanical) harmonic approximation. When anharmonic contributions are considered, the potential energy surface in terms
of mass-weighted normal coordinates takes the form

3M

1
2 Ee,i
(m)
=
Q(m)2
(26)
Ee,i Q
I
(m)2
2
2
QI
I=1
3M 3M 3M
1
3 Ee,i
(m) (m) (m)
QI QJ QK + ...
+
(m)
(m)
(m)
6
I=1 J=1 K=1 QI QJ QK
0
The coupling between different normalcoordinates is introduced
via the
(m)
(m)
(m)
. Anharcubic (and higher-order) force constants 3 Ee,i /QI QJ QK
0
monic force constants up to sixth order with up to three different normal
mode indices have been included into predictions of 2D-IR spectra of acetic
acid dimers based on DFT calculations [104]. The calculation of fundamental
frequencies and anharmonic vibrational constants may be achieved either by
vibrational self-consistent eld methods or by vibrational perturbation theory (see [105] for a comparison of both). For large molecules, however, the
perturbation theory approach may fail due to the large number of anharmonic resonances. The vibrational self-consistent eld approach, on the other
hand, requires a signicantly larger computational effort, so the problem of
including mechanical anharmonic effects accurately in calculations of large
molecules remains a difcult if even possible task. Electrical anharmonicity,
110
on the other hand, is commonly believed to play a minor role and thus usually
neglected [104].
The goal of a theoretical description of 2D-IR spectra is to predict the
positions, intensities, and shapes of the peaks within a selected wavenumber range [106]. Since 2D-IR is a nonlinear technique, the description of the
whole spectrum based on rst-principles calculations is very involved (for
a detailed account of the theory of nonlinear spectroscopy see [107]). Thus,
for realistic systems, model Hamiltonians are used instead, whose parameters may be adjusted based on rst-principles calculations. Two-dimensional
IR spectra of peptides are usually calculated based on a vibrational-exciton
model [108, 109], which describes selected vibrational states as a set of
coupled oscillators. The associated coupling constants have been estimated
so far either based on semi-empirical models or from the anharmonic force
constants obtained from DFT calculations [100, 104, 110, 111], in some cases
including a continuum model description of the solvent [112]. It is also possible to use molecular dynamics or Monte-Carlo simulations in order to
construct uctuating vibrational-exciton models [113]. Two different variants
of this technique have been compared and applied to a 17-residue peptide
in [114]. Results from MD simulations have also been compared to structural
and dynamical parameters for the subpicosecond conformational dynamics
of small peptides obtained from 2D-IR experiments [115].
Because of the restriction of the exciton model to a certain class of vibrations (such as amide I vibrations in peptides), the calculations are automatically selective for a certain subset of vibrational modes.
5.3
Techniques Sensitive to Chirality: VCD and ROA
Vibrational Circular Dichroism (VCD) and Vibrational Raman Optical Activity (ROA) spectra are in principle recorded in the same way as standard IR
and Raman spectra, respectively, but instead of total intensities, the intensity differences between right and left circularly polarized light are plotted.
These differences are zero for achiral molecules and in general nonzero for
chiral molecules. Thus, information related to the chirality of the molecule
under study is ltered out from the vibrational spectrum. As far as the basic idea is concerned, the two techniques are similar to electronic circular
dichroism (CD), where absorption intensity differences between right and left
circularly polarized light in the frequency region of electronic transitions are
recorded. Advantages compared to CD are, however, that VCD and ROA allow
for a determination of the absolute conguration of molecules which do not
contain a chromophore and are thus inaccessible to CD spectroscopy. Also,
this determination can be more reliable since it is based on a larger number of
peaks. The ability to determine absolute congurations of biomolecules (and
pharmaceuticals) is of crucial importance due to the different interactions
111
of different enantiomers with molecular structures in living organisms. The

large number of signals arising from vibrational motions of all parts of the
molecule instead of only those involved in electronic transitions furthermore
provides more structural information than electronic spectra.
Although the rst ROA experiment was carried out in 1973 [116] and thus
one year earlier than the rst VCD experiment [117], ROA became a routinely applicable technique only recently. This is mainly due to the fact that
the scattering intensity differences are of the order of about 103 times the
total scattering intensity at most, and that total scattering intensities are considerably smaller than usual absorption intensities, so that very sensitive
and stable instruments are needed. Apart from this, ROA outperforms VCD
in two ways: First, it can be recorded in the natural environment of most
biomolecules, namely in aqueous solution, and second, ROA spectra can be
recorded down to lower wavenumbers ( 100 cm1 [118]) than VCD spectra
( 600 cm1 [119]).
VCD has primarily been used for the structure determination of peptides
and proteins, but also for nucleic acids and monosaccharides (for reviews, see
e.g., [120122]). The range of compounds investigated by ROA so far [123, 124]
is similar to the one of VCD, except that it comprises also disaccharides [125]
and even intact viruses [126]. Because of the sensitivity of the intensity differences to conformational changes, both techniques are particularly suited for
the investigation of protein secondary structures. ROA, e.g., allows for a discrimination between helical and -sheet peptide structures based on empirical
correlations between assumed protein structures and characteristic signal patterns [127, 128]). Furthermore, experimental data suggest that ROA might
provide information on tertiary structures [123, 124]. For a comparison of the
congurational and conformational sensitivity of VCD and ROA, see [129] and
for a general quantitative comparison of both techniques [119].
In order to improve and verify empirical correlations between structure
and spectral intensities, and in particular in order to make unambiguous
assignments of absolute congurations possible, quantum chemical calculations are needed for both VCD and ROA. Whereas the development of
a quantum chemical ROA theory, whose basic equations were published in
1971 [130], preceded the consistent formulation of the theory of VCD in
1983/1985 [131, 132], the number of VCD calculations published so far is
much larger due to the faster experimental development of this technique.
An excellent description of the theoretical background of ROA and of the interaction of chiral molecules with radiation in general can be found in [64].
For a short summary of theoretical background, current experimental instrumentation and selected applications of ROA see [133], and on theoretical and
experimental vibrational optical activity in general, emphasizing the need of
rst-principles calculations see [134]. Furthermore, [135] provides a survey
on ROA (and Raman) spectroscopy both from an experimental and a theoretical point-of-view.
112
It is necessary to outline the theoretical background briey in order to provide a basis for the discussion of special problems related to the calculation of
VCD and ROA spectra for biomolecules. For computational details, we refer to
the reviews mentioned above and to the literature cited in the following.
5.3.1
VCD Theory
Absorption intensity differences between right and left circularly polarized
light in an isotropic sample are proportional to the isotropic rotational
strength Rf i , which can be calculated from the Rosenfeld equation [136],

Rf i = Im i||f f |m|i .
(27)
This equation describes vibrational as well as electronic circular dichroism,
depending on whether the initial and the nal state |i and |f differ only in
their vibrational or also in their electronic

state. The evaluation of the electric transition dipole matrix element i||f is known from IR theory, so the
calculation of VCD spectra focusses on the magnetic transition dipole matrix element f |m|i
. For nonmagnetic (closed-shell) molecules, the electronic
contribution to f |m|i is zero for matrix elements between nondegenerate
electronic states within the BornOppenheimer approximation, which was
the reason for the delay in the formulation of the theory of VCD compared
to ROA. By replacing the concept of nuclear-coordinate dependent energy
and property surfaces by the idea of velocity dependent

ones, however, it
is possible to formulate computable expressions for f |m|i based on Born
Oppenheimer wave functions [131, 132],

m

p |i
f |Q
,
(28)
f |m(Q)|i
=
p
Q
p
0

p |i can be calculated within the (mechanical)
where the matrix element f |Q
harmonic approximation from the harmonic oscillator model. The derivative
p of the norof the magnetic dipole moment along the time derivative Q
mal mode Qp can be expressed by using the magnetic eld perturbation
method [132, 137, 138] as

m
0 0
.
(29)
= 2Im
p
B Qp
Q
0
Both wave function derivatives can be calculated either analytically (employing coupled HF (or KS) perturbation theory [139]) or numerically [140].
Equation 29 has been the basis for numerous computational studies on VCD.
A summary of the excellent performance of DFT for that purpose is provided
by [141], and one on the use of theory to deduce absolute congurations from
experiment by [142].
113
5.3.2
ROA Theory
The derivation of the relevant formulas for the ROA intensity differences is
essentially the same as in Raman spectroscopy, except that in addition to
the electronic dipole moment, the contribution of the molecular electronic
quadrupole and magnetic dipole moments interacting with the radiation to
the induced multipole moments have to be taken into account [64, 130, 143,
144]. These three induced multipole moments can be expressed as a function
of the electric eld amplitude of the incident radiation, and of several complex generalized polarizability tensors. The latter are the (ordinary) electric
dipoleelectric dipole polarizability , the electric dipolemagnetic dipole
polarizability G (the prime indicating imaginary parts of complex quantities), and the electric dipoleelectric quadrupole polarizability A. Depending
on whether the right-handed or the left-handed enantiomer is under the inuence of radiation with a certain polarization state, the magnitude of the
induced multipole moments and thus the intensity of the scattered radiation
in a given polarization state will be slightly different.
The next step is to calculate this intensity of the scattered radiation
from the induced multipole moments, which can be done within the semiclassical approach based on equations known from classical electrodynamics.
Whereas in VCD only one experimental setup is of importance, in ROA different arrangements are possible, which are distinguishable by the polarization
of the incident and detected radiation and, analogous to conventional Raman
spectroscopy, by the angle to the incident light beam in which the scattered
radiation is detected [64, 133]. For each setup, different mathematical expressions for the calculation of the corresponding intensity differences are
obtained. Because of the good signal-to-noise ratio and the low susceptibility to measuring artifacts, the backscattering variant of ROA is employed in
modern instruments. Therefore, the corresponding expression for the intensity difference between right and left circularly polarized light relative to the
total scattered intensity, obtained by applying the principles mentioned above
and employing Placzeks approximation, is given as an example,

IR IL
180 = R L
(30)
I +I

1
8 3 mn G
mn G
+ 3 mn A ,
mn
mn
mn

=
,

c 7 mn
+ mn
mn
mn
where summation over repeated indices is implied. The quantities indicated

by the subscript mn are transition matrix elements between vibrational states
|vn and |vm of the generalized polarizability tensors, whose dependence on
114
the normal coordinates is, as in the theory of conventional Raman scattering,

expressed as a Taylor series truncated after the linear term (electric harmonic approximation). The combinations of these transition matrix elements
in Eq. 30 also appear in the intensity expressions corresponding to the other
experimental setups, so they are usually summarized as Raman and ROA invariants (for details, see e.g., [133] or [64]). Within the electric harmonic
approximation, the transition matrix elements in Eq. 30 are proportional to
derivatives of the corresponding molecular generalized polarizability tensor
along mass-weighted normal modes. Since derivatives of the electric dipole
magnetic dipole polarizability G along nuclear coordinates have not been
implemented in an analytical form yet [68], and analytical derivatives of the
electric dipoleelectric quadrupole polarizability A have become available
only very recently [145], all three derivatives are usually calculated numerically at present. The polarizability tensors themselves are usually calculated
employing linear response theory [146]. As in the calculation of VCD spectra, it is necessary to ensure gauge invariance of the intensity differences by
employing London atomic orbitals [146, 147], since both the electric dipole
magnetic dipole and the electric dipoleelectric quadrupole polarizability
depend on the choice of the coordinate origin. This origin dependence is not
reproduced correctly otherwise. As far as selection rules are concerned, all
Raman active modes are in principle also visible in ROA spectra.
ROA calculations on biomolecules have so far essentially been carried out
for peptides and proteins (see e.g., [148], which also gives a survey of the
current state-of-the-art of computational methods for predicting ROA spectra). Furthermore, very recent reviews on the quantum mechanical prediction
of chiroptical vibrational properties [68] and chiroptical properties in general [149] are available.
5.3.3
Molecular Conformation
ROA [118, 133, 150] and VCD [151, 152] calculations have shown that in conformationally exible molecules, intensity differences arising from different
conformations (e.g., of methyl groups or ring structures) can cancel, so that
certain vibrational bands vanish in the ROA spectrum. This can be regarded
as an additional mechanism ltering out information, which distinguishes
ROA from Raman spectroscopy, where contributions from different conformations are always additive. In order to reproduce this behavior accurately
in quantum chemical calculations, it is in principle necessary to calculate the
relative energies of all relevant conformations (which may depend strongly
on the solvent) and to superimpose their spectra multiplied by weighting factors obtained by assuming a Boltzmann distribution of the conformers [118].
For larger biomolecules, where already the simulation of the spectrum of
one conformer is a computational challenge, another (though conceptually
115
less satisfying) approach is to neglect the intensity differences of all vibrational bands which involve mainly the conformationally exible groups (in
case such a partitioning of modes is possible).
5.3.4
Efficient Computational Procedures for Large (Bio-)Molecules
Whereas calculated VCD intensity differences are not considerably more expensive than the corresponding IR intensities, the calculation of ROA intensity differences is much more computer time-consuming than the calculation
of the normal modes, vibrational frequencies and also Raman intensities. Two
strategies may be combined in order to overcome this problem. First, the
molecular property tensors may be computed more efciently. Currently, efcient density-tting techniques like those available for the calculation of the
molecular polarizability [153] are not yet available for generalized property
tensors. Therefore, calculations of the property tensors are accelerated by employing a smaller basis set than the one used for the vibrational modes and
frequenciesfor ROA, a specic basis set has been developed which yields
excellent results despite its limited size. It comprises diffuse p-type functions
on hydrogen atoms, which are not contained even in much larger basis sets,
but which improve the accuracy of the results dramatically [69]. In order to
gain additional efciency, it has been proposed to neglect the contributions
of the electric dipoleelectric quadrupole polarizability [154], which are often small, and reduce to zero for a certain experimental setup [133]. This
reduces the computational efciency considerably, at the cost of a (hardly
controllable) loss of accuracy. It is also possible to replace the linear response
theory by a cheaper, but more approximate sum-over-states formalism [155]
for the property calculations, which has yielded good results for oligopeptides. In addition, calculated ROA intensity differences (for a given set of
normal modes) have been found to be virtually independent of the chosen
density functional, so that calculations of the relevant properties with pure
density functionals and thus with the speed-up provided by the resolutionof-the-identity technique [38, 39] are in principle possible [70]. As a second
way of improving the computational efciency needed for large biomolecules,
the calculation of intensity differences can be restricted to a small set of vibrational modes. For the assignment of absolute congurations as well as for
the investigation of molecular conformations, calculations of intensity differences are not necessarily needed for all vibrational bands, but may be
sufcient for a few characteristic ones. Intensity differences may be computed selectively for these modes from property derivatives (and additionally
wave function derivatives, in the case of VCD) obtained by direct differentiation along normal modes. For ROA, the direct numerical differentiation along
normal modes has been implemented and applied to deca-alanine [232].
Which vibrational normal modes are considered to be characteristic may not
116
be obvious from the beginning, but can be determined by dening a set of

modes assumed to have large associated intensity differences based on chemical intuition. This set can be enlarged stepwise if it turns out to be not
sufcient for an unambiguous assignment of the molecular conguration. In
the case of calculated ROA spectra for peptides, the relevant normal modes
chosen will usually be the amide modes (especially amide III), since the intensity differences of these modes are especially sensitive to the backbone
conformation. An approximate calculation of VCD spectra of large molecules
has also been performed based on the transfer of molecular property tensors
in cartesian coordinates, which relates spectra of large molecules with repeated similar subunits (such as peptides) to molecular properties computed
for the fragments [29, 34].
5.3.5
Interpretation of Calculated Spectra for Large (Bio-)Molecules
When calculating VCD or ROA spectra of large biomolecules, the possibility
that individual peaks may cancel, and the fact that the mechanical harmonic
approximation has to be employed, creates a problem with respect to the
interpretation of the computed spectra. The intensity differences of closelying vibrational peaks may have different signs, and due to the mechanical
harmonic approximation as well as the neglect of solvent effects, the corresponding wavenumbers may easily be shifted by 20 cm1 or more. In contrast
to small molecules, where the individual peaks are sufciently separated from
each other, the high density of peaks in the spectra of large molecules may
lead to cancellations of peaks in the experimental spectrum which can turn
into a sequence of a positive and a negative peak in the calculated spectrum,
and vice versa. Since the harmonic approximation cannot be avoided in calculations of vibrational spectra of large molecules, especially ROA spectra
calculated by rst-principles methods will have difculties to predict the detailed structure of a given spectral region (such as amide III) with certainty,
but only the intensity difference integrated over the whole region is likely to
be characteristic [232]. This may be accounted for by plotting ROA spectra in
a band-averaged way in addition to a Lorentzian (or Gaussian) peak broadening (Fig. 3). How the relevant information can be extracted from calculated
ROA spectra of large molecules is considered in detail by [232].
5.3.6
Solvent Effects
As far as the calculation of VCD and ROA spectra is concerned, solvent effects
can become even more important than for techniques which are not sensitive
to chirality, since not only the relative magnitudes, but also the signs of vibrational bands may change in different solvents [156, 157], either indirectly
117
Fig. 3 ROA spectroscopy of helical all-S deca-alanine. Top left: Experimental ROA spectrum of poly-l-alanine in 30% dichloroacetic acid/70% CHCl3 (reproduced from data
presented in [127]). Top right: Symbolical representation of (all-S)-deca-alanine. Bottom:
Selectively calculated ROA intensity differences of (all-S)-deca-alanine, at an excitation
wavelength of 488.80 nm. Bottom left: Lorentzian intensity distribution, half-width of
peaks 15.0 cm1 . Bottom right: Band-averaged ROA intensity differences. Individual peaks
have been included as a line spectrum scaled by 0.04. Calculated spectra were reproduced from data given in [232]. Intensity differences in
A4 a.m.u.1 ; normal modes and
frequencies: DFT(BP86/RI), TZVP; generalized polarizability tensors: DFT(BLYP), Zuber
and Hugs rDSP:3-21
due to changes in the conformational equilibrium or directly through solventinduced changes in the electronic structure of the solute and its dependence
on the nuclear coordinates. In [156] for example, solvent effects on VCD spectra of proline, modeled by a combination of the integral equation formulation
of the polarizable continuum model (PCM) and explicit water molecules, are
found to inuence the relative energies of different conformations considerably, whereas the structural parameters are quite invariant to solvent effects.
A complementary study has been carried out for ROA [157], employing the
same PCM version but no explicit solvent molecules, which points out the
importance of considering the solvent inuence on both the vibrational normal modes and the property tensors. It has been reported in previous studies
that the explicit consideration of water molecules in the rst solvation shell of
solvated alanine improves ROA results considerably in combination with an
118
Onsager continuum model [158]. Further investigations on the relative performance of continuum and explicit solvent models for the prediction of ROA
spectra are in preparation and will be published by the authors in a separate
study. Furthermore, DFT investigations combined with the PCM on VCD spectra have shown the need for better quantitative accuracy in the predicted IR
and VCD intensities [151, 152], which suggests that solvent effects may have to
be considered more accurately in some cases than is possible with the PCM.
In particular when hydrogen bonds between solvent and solute are important,
explicit solvent molecules will, in general, be necessary. The larger the number of solvent molecules, the more important mode-selective techniques will
become for the feasibility of such studies.
5.3.7
Local Contributions
Additional insight into the generation of VCD and ROA signals may be
gained from assessing and comparing the contributions from different parts
of the molecule under study. Local contributions to VCD intensity differences, which may be visualized as a continuous distribution over the whole
molecule in the form of an electron transition current density, have been derived by Freedman, Nae and coworkers [159162]. A scheme to decompose
ROA and Raman intensities into contributions from individual pairs of atoms
(which may be grouped to pseudo-atomic contributions with a certain arbitrariness comparable to electron population analysis) has been presented
by Hug [163, 164]. It reveals that long-range couplings may play an important role in ROA spectroscopy [165]. The possibility of local decomposition
is a great advantage of rst-principles calculations over data gained from
experimental work, since it provides information on where signals are generated and thus might nally lead to a set of rules for the prediction of VCD and
ROA spectra.
5.4
Raman Techniques in Electronic Resonance
Traditional as well as special Raman scattering techniques have a resonance
variant, where the energy of the incident light is close to an electronic transition
energy of the molecule under study. This usually results in the amplication of
a few peaks and thus to a simplication of the vibrational spectrum.
5.4.1
Spontaneous Resonance Raman Scattering
Whereas Raman intensities far from electronic resonance depend mainly
on molecular ground-state properties, resonance Raman spectra contain in-
119
formation on those electronically excited states whose excitation frequencies are close to the frequency of the incident laser beam (usually only
one or a few; for a qualitative scheme of the former case see Fig. 4). From
a simplied point-of-view, the intensities of those vibrational modes along
which the equilibrium structure is shifted most upon electronic excitation
are largely amplied in comparison to ordinary Raman spectroscopy. Thus,
resonance Raman spectroscopy is of particular interest in order to lter
out those vibrational modes which are located on a chromophore in large
biomolecules [166168]. Its importance for the investigation of biomolecules
is highlighted, e.g., in a special issue on the structure and dynamics of
biomolecules probed by (mainly resonance) Raman spectroscopy in J Raman Spectrosc 2005:36(4). For a review of the applications of resonance
Raman and other Raman techniques to problems relevant for the life sciences see [169]. Resonance Raman spectroscopy has been applied to enzymes
whose active site is a chromophore such as heme proteins (for a review on
structural and functional properties of hemoglobins revealed by resonance
Raman spectroscopy see e.g., [170]), where the vibrational frequencies are
sensitive to the oxidation and the spin state as well as to the coordination
number of the central iron porphyrin group [171175], and to redox-active
tyrosines in various metalloenzymes [176]. Furthermore, it has been used for
the resonance enhancement of vibrations of the peptide backbone [177, 178]
and of aromatic side chains [179] in order to elucidate their conformation and
thus the secondary structure. Resonance Raman spectroscopy can also be em-
Fig. 4 Schematic representation of resonance Raman scattering for one electronically excited state in resonance (IMDHO = independent mode, displaced harmonic oscillator
model)
120
ployed as a time-resolved technique down to subpicosecond resolution, e.g.,

for the investigation of the excited-state conformational dynamics of nickel
porphyrins [180] or the dynamics of nitric oxide reductases and hemecopper
oxidases [181, 182].
The calculation of resonance Raman intensities starts from the same expression for the transition polarizability obtained from time-dependent perturbation theory as the treatment of nonresonant Raman scattering (Eq. 19).
However, since the energy difference between transitions to electronically
excited molecular states and the incident radiation is not large as assumed
in the derivation of the Placzek approximation [62], this drastic simplication has to be replaced by better suited ones. If the frequency difference
between the incident radiation and an electronic transition becomes small,
one or a few denominator(s) of the terms on the left-hand side of Eq. 19 will
approach zero and the corresponding terms will dominate over the others,
which consequently can be neglected. Following the same line of reasoning,
the terms on the right-hand side of Eq. 19 can also be neglected since they
will all be considerably smaller than those dominant terms. The sum therefore reduces to one or a few terms which need to be evaluated. In practice,
mainly two approaches are employed: the transform theory as presented by
Peticolas and Rush [183] based on KramersKronig relations between the optical absorption and the frequency-dependence of the dispersive part of the
polarizability [184], and a time-dependent treatment from which a simple
short-time approximation can be deduced [185187]. The performance of the
two approaches as well as the limitations of the underlying approximations
have been assessed in detail in [188]. Both have been found to yield comparable results for various molecules [188, 189]. On the basis of the assumptions
that (1) the BornOppenheimer and (2) the (mechanical) harmonic approximation are valid, that (3) only one excited electronic state s is in resonance,
that (4) only FranckCondon-type scattering occurs, and that (5) the excited
state normal modes differ from their ground state counterparts only in a shift
of their equilibrium position (independent mode, displaced harmonic oscillator (IMDHO) model), transform theory predicts the resonance Raman
intensity of a normal mode Qp as

2
4 2
IQp = L 3S |el
0s | qp | L L p | .
(31)
L and S refer to the angular frequencies of the incident and the scattered
(Stokes-)laser beam, respectively, p to the vibrational angular frequency
associated with the normal mode Qp , |(L ) (L p )|2 is a Kramers
Kronig scaling factor which may be obtained from the (either calculated or
measured) absorption spectrum, and 2qp denotes the displacement of the
equilibrium structure of the excited state s relative to the ground state along
the dimensionless normal mode qp , which is related to the mass-weighted
121

normal mode Qp as qp = Qp 2cp /. The short-time approximation is
based on considering scattering as connected to ground state vibrational
wave packets with short life-times propagating on the excited state energy
PES [186, 187]. It describes the intensities associated with preresonant Raman
scattering as
IQp = 2p 2qp .
(32)
Within the IMDHO model, and assuming that only short-time dynamics of
the nuclei are important, the excited-state displacements along dimensionless
normal modes may be estimated from the excited state gradients with respect
to the dimensionless normal modes,

Ee,s
= p qp .
(33)
qp
0
The excited state gradients can be efciently calculated within TDDFT [190
194]. It is obvious from both methods that resonance Raman scattering is
largely determined by how much the vibrational normal mode under consideration is shifted in the excited electronic state compared to the ground state.
For theoretical details on both approaches, see [188], and for a recent comparison of experimental resonance Raman spectra of large biomolecules to
calculated ones see a study on Ruthenium complexes to be published by the
authors of this paper and [195, 196]. In addition, a more general version of the
Placzek approximation polarizability has been proposed based on the shorttime approximation [197201], which reduces to the conventional Placzek
approximation in the nonresonant case and which has the advantage of automatically including all possible excited states. It has recently been applied to
uracil as a test molecule and found to compare well with experiment as well
as to the short-time approximation based on excited state gradients for this
system [202].
The computational effort associated with calculations based on transform
theory is in general much higher than for those relying on the short-time approximation. Both require a ground-state structure optimization and a subsequent calculation of the ground-state normal modes and frequencies, but
in the case of transform theory, an additional structure optimization in the
relevant excited state(s) is needed, whereas in the case of the short-time approximation, a comparatively fast computation of the excited state gradients
at the ground-state equilibrium structure is sufcient.
An alternative approach to the calculation of (pre-) resonance Raman
spectra is based on performing CPMD simulations in the ground state as well
as in the excited state of interest, and to deduce vibrational frequencies and
associated preresonance Raman intensities from the obtained data. This technique has been suggested and applied to a model chromophore of the DsRed
protein in [203].
122
5.4.2
Resonance CARS
In contrast to nonresonant Raman scattering, resonance Raman spectra may
suffer from the uorescence of the sample, since the frequency of the scattered Stokes radiation (which is usually detected instead of the Anti-Stokes
radiation, since the latter is much less intense) may be close to the frequency
of the radiation emitted by the uorescent species. This problem may be
overcome, however, by employing resonance Coherent Anti-Stokes Raman
(CARS) scattering (see [204] for a general review). Although CARS can be
principally used in as well as out of electronic resonance, only its resonance
form is treated here, because it is the more widely used version of both for
the investigation of biomolecules. In contrast to spontaneous Raman scattering, where the phase of the molecular vibrations of each molecule is arbitrary,
the radiation scattered by different molecules in CARS has a xed phase relation. This is achieved by using a set of coherent intense laser pulses (which
have to be intense enough to make nonlinear effects observable, and short
enough in order not to decompose the sample despite their high intensity):
The molecules are excited simultaneously by a pump beam (with angular
frequency L , corresponding to a molecular electronic transition for resonance CARS) and a Stokes beam (with angular frequency S < L ). The result
is a set of molecules in one or a few selected vibrational state(s), vibrating with a phase determined by the two exciting beams. These articially
excited molecules are probed in principle like in conventional Raman spectroscopy, using a probe beam with angular frequency P , usually equivalent
to L , and detecting an Anti-Stokes radiation beam with angular frequency
AS = L + P S . When the difference between the angular frequencies
of the two incident beams, L S , approaches the angular frequency of
a molecular vibrational transition M , the intensity of the coherent scattered
radiation will greatly enhance. Thus, by tuning the frequency of the incident
laser beams, one or a few molecular vibrations may be ltered out (provided
the duration of the laser pulses is long enough to allow for a sufcient frequency resolution, which usually is the case when the time-scale is around
a few picoseconds). This tunable ltering mechanism represents an advantage
over the linear selective spectroscopic techniques discussed so far, where the
choice of which modes are ltered out is essentially controlled by the properties of the molecule under study. Since the scattered radiation is coherent, its
intensity is nonzero essentially only in one direction depending on the directions of the three incident laser beams. This directed signal and the fact that
an intense Anti-Stokes signal is recorded allows for a distinction from uorescence, which is detected at lower wavenumbers [205, 206]. CARS has been
applied to various biomolecules, such as porphyrins [207, 208], bacteriorhodopsin [209], and low-frequency modes around 150 cm1 of -carotene which
are assumed to play an important role in the energy transfer in photosyn-
123
thesis [210], including applications as a microscopic imaging technique [211,

212] and the investigation of living cells [213].
CARS spectra may be recorded with femtosecond time resolution (see [214]
for a review) which allows for investigating ultrafast molecular dynamics. Ultrashort laser pulses are also used in the context of CARS microscopy [215].
The broad frequency range of the incident laser beams, which is hardly
avoidable when using ultrashort laser pulses, may be overcome by feedbackcontrolled optimization of pulse shapes [216], or by additionally varying the
relative delay between the exciting laser pulses [214], so that mode-selectivity
can be achieved. There are at least two ways of measuring time-dependent
CARS spectra: First, an initial pump laser may be used, which is followed after
a time delay by a CARS experiment where the three CARS pulses are interacting with the sample simultaneously. Second, a time-delay can be introduced
in a normal CARS experiment between the simultaneous pump and Stokes
pulses on the one hand and the probe pulse on the other hand. For an application of the former technique to the ground state repopulation dynamics
via S1 to S0 internal conversion of -carotene see [217], and for a comparison
of this technique to the latter one see [218]. Both techniques are compared
and reviewed from a theoretical point-of-view in [219]. More references may
be found in a special issue on Femtosecond Coherent Raman Spectroscopy
(J Raman Spectrosc 2000:31).
A rigorous rst-principles calculation of a CARS spectrum has to the best
of our knowledge not been reported in the literature so far. Because of its
coherent nonlinear nature, the theoretical description of CARS spectroscopy
is much more involved than the one of spontaneous Raman scattering (for
detailed accounts of the theoretical foundations, see [220222]). The CARS
intensities of a bulk of molecules cannot necessarily be calculated as the
orientational-averaged sum of the scattering intensities of all molecules, since
the scattered radiation is coherent. The derivation of CARS intensities is thus
usually formulated in a more general way than for linear Raman techniques.
The incident radiation introduces a polarization in the sample of molecules
under study, which depends on the third-order susceptibility of the sample.
This third-order susceptibility is not a single-molecule property as in the theoretical treatment of spontaneous Raman scattering, but a bulk property of
the sample under study. If rst-principles calculations are performed in the
context of CARS spectroscopy, they have so far rather been restricted to the
molecular equilibrium structure and vibrational normal modes [207], but do
not include any intensities.
It is possible to predict excitation proles, that is the dependence of CARS
intensities on the frequency of the pump laser, either by employing a sumover-states formalism or transform theory [223]. However, both have not
been employed as true rst-principles methods yet, but in connection with
a set of tted parameters. In addition, the observed proles are not necessarily reproduced for all laser wavelengths and vibrational modes [223]. The
124
challenge of calculating resonance CARS intensities is furthermore illustrated

by a study aiming at predicting CARS signals for a sample of dipicolinic
acid [224]. This study is based on considering equations of motion of the
sample in terms of density matrix elements, coupled to Maxwells equations,
and takes into account the particle density as well as propagation effects of
the laser pulses. However, the complicated system under study is reduced to
a four-level system and needs estimated radiative decay and dephasing times
as an input.
CARS spectra may be converted to the corresponding Raman spectra [206], so in principle, it might seem sufcient to be able to predict resonance Raman spectra. However, accurate rst-principles calculations of CARS
spectra appear to be valuable nonetheless. The CARS signal is often obscured
by interference with the nonresonant background, which may be suppressed
by techniques such as polarization-sensitive CARS experiments [207]. It can
be helpful to be able to predict this background interference with theoretical
methods.
5.4.3
Resonance ROA
Resonance variants have also been introduced for special forms of Raman
spectroscopy. Resonance ROA spectroscopy is still at a very early stage of development. The rst (and, so far, the only) experiment has been published
in 1998 [225], two years after the derivation of the underlying theory [226].
In analogy to the connection between resonance Raman and electronic absorption spectra, the theory of ROA is closely related to the one of electronic
circular dichroism. It predicts (and is conrmed by the experiment) that in
contrast to ROA spectroscopy far from electronic resonance, resonance ROA
signals of chiral molecules all have the same sign, provided that only one
excited electronic state is in resonance. In this case, the resonance ROA spectrum is equal to the resonance Raman spectrum scaled by a factor equal
to one or one-half the magnitude of the electronic CD anisotropy ratio of
the electronic state which is in resonance, depending on the experimental
setup. The resonance ROA spectra of two enantiomers are related by a reection at the wavenumber axis. Altogether, resonance ROA may be regarded as
a means to assess whether only one excited electronic state is in resonance (by
comparison to resonance Raman spectra recorded with unpolarized light),
combined with a detection tool for absolute congurations.
5.5
NRVS
Nuclear Resonance Vibrational Spectroscopy (NRVS) is an exception in the
group of spectroscopic techniques included in this review, since it is the only
125
technique which cannot be applied to biomolecules in solution, but which

is restricted to solid samples, measured at very low temperature (typically
around 30 K). However, it possesses outstanding selective properties and is
not affected by the optical selection rules. NRVS lters out the vibrations
which involve a Mssbauer active nucleus, with intensities proportional to the
displacement of that nucleus in a certain vibrational normal mode. The most
important Mssbauer active isotope from a biochemical point-of-view is 57 Fe,
so that NRVS is especially useful for heme-containing compounds. The idea of
NRVS is to record the vibrational structures of Mssbauer bands. These structures are not observable in conventional Mssbauer experiments, but need
a sophisticated experimental setup based on a tunable X-ray synchrotron
source, currently allowing a resolution down to 78 cm1 . This demanding
experimental setup is one of the main reasons why NRVS experiments have
become feasible only recently. An up-to-date review on details of the experimental methodology and applications to iron species, comparing experimental results to DFT calculations, is given in [227]. Since NRVS intensities are
directly related to the contributions of the iron atom(s) to the vibrational
normal modes and the vibrational frequencies [228, 229], only a diagonalization of the Hessian matrix is necessary to calculate them. Recent applications
of NRVS to biomolecules have been focussing on myoglobin [229, 230], cytochrome f [230], rubredoxin [229], and the low-frequency modes of the
FeMo cofactor in intact nitrogenase [231].
6
Conclusion
This review provides a survey of the quantum chemical foundations and applications of rst-principles calculations of vibrational spectra of biomolecules.
Because of their large size (which can easily be 1001000 atoms or more), the
conventional vibrational spectra of biomolecules tend to contain much more
information on a limited space than can be interpreted. Thus, this work focussed on the theory of techniques which lter out information compared
to conventional IR and Raman spectroscopy, in particular resonance Raman,
difference IR and Raman, two-dimensional techniques, VCD, ROA, CARS,
and NRVS. The complexity of the systems under study and the early stage of
development of many special spectroscopic techniques makes rst-principles
calculations a very valuable tool for the interpretation of vibrational spectra.
For the same reason, i.e., their complexity, many biomolecules are not accessible by conventional quantum chemical methods. This requires either to
cut out a central fragment and analyze it with an accurate (but computationally expensive) method such as (in the context of large molecules) DFT, to
treat the whole molecule with cheaper, but less accurate (e.g., semi-empirical)
methods, or to use a combination of both (like QM/MM). A conceptionally
126
different approach to this problem is to describe the whole biomolecular system with accurate methods, but to restrict the calculation to the relevant
information only. Two such methods, mode-tracking to calculate vibrational
normal modes and direct numerical differentiation of molecular property
tensor along normal modes to obtain the corresponding intensities, were presented and compared to conventional computational schemes.
By selective calculation of the relevant normal modes and intensities only,
large biomolecules are made accessible to accurate quantum chemical calculations in the future, and accordingly, solvent effects, whose treatment by
continuum models is often not sufcient to model all spectral features (especially in the context of VCD and ROA), may be treated explicitly for much
larger systems.
Acknowledgements M.R. is grateful to the Fonds der Chemische Industrie (FCI) for
a Dozentenstipendium. C.H. gratefully acknowledges funding by a ChemiefondsDoktorandenstipendium of the FCI. The authors thank Dr. Johannes Neugebauer for
valuable comments on earlier versions of this manuscript, and Dr. Georg Brehm and
Priv.-Doz. Dr. Michael Schmitt for helpful discussions.
References
1. Ladd MF, Palmer R (2003) Structure determination by X-ray crystallography. Kluwer
Academic/Plenum Publishers, New York
2. Drenth J (1999) Principles of Protein X-Ray Crystallography. Springer, Berlin
3. Wagner G (1997) Nature Struct Biol 4:841844
4. Clore GM, Gronenborn AM (1997) Nature Struct Biol 4:849853
5. Hajdu J (2000) Curr Opin Struc Biol 10:569573
6. Miao J, Hodgson KO, Sayre D (2001) Proc Natl Acad Sci USA 98:66416645
7. Miao J, Chapman HN, Kirz J, Sayre D, Hodgson KO (2004) Annu Rev Biophys Biomol
Struct 33:157176
8. Sommerhalter M, Lieberman RL, Rosenzweig AC (2005) Inorg Chem 44:770778
9. Einsle O, Tezcan FA, Andrade SLA, Schmid B, Yoshida M, Howard JB, Rees DC (2002)
Science 297:16961700
10. Yang T-C, Maeser NK, Laryukhin M, Lee H-I, Dean DR, Seefeldt LC, Hoffman BM
(2005) J Am Chem Soc 127:1280412805
11. Jalkanen KJ, Jrgensen VW, Claussen A, Rahim A, Jensen GM, Wade RC, Nardi F,
Jung C, Degtyarenko IM, Nieminen RM, Herrmann F, Knapp-Mohammady M, Niehaus TA, Frimand K, Suhai S (2005) Int J Quantum Chem 106:11601198
12. Marx D, Hutter J (2000) Ab Initio Molecular Dynamics: Theory and Implementation. In: Grotendorst J (ed) Modern Methods and Algorithms of Quantum Chemistry, Proceedings. NIC Series, Vol 3. John von Neumann Institute for Computing,
Jlich
13. Born M, Oppenheimer R (1927) Ann Phys 84:457
14. Reiher M, Neugebauer J (2003) J Chem Phys 118:16341641
15. Neugebauer J, Reiher M (2004) J Phys Chem A 108:20532061
16. Neugebauer J, Reiher M (2004) J Comput Chem 25:587597
17. Reiher M, Neugebauer J (2004) Phys Chem Chem Phys 6:46214629

18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
127
Kaledin AL (2005) J Chem Phys 122:184106

Reiher M, Neugebauer J (2005) J Chem Phys 123:117101
Kaledin AL, Kaledin M, Bowman JM (2005) J Chem Theory Comput 2:166174
Herrmann C, Reiher M (2006) Surf Sci 9:18911900
Filippone F, Parrinello M (2001) Chem Phys Lett 345:179
Filippone F, Meloni S, Parrinello M (2001) J Chem Phys 115:636
Adler TB, Borho N, Reiher M, Suhm MA (2006) Angew Chem Int Ed 45:34403445
Davidson ER (1975) J Comput Phys 17:8794
Liu B (1978) In: Moler C, Shavitt I (eds) Numerical Algorithms in Chemistry: Algebraic Methods. Lawrence Berkeley Laboratory LBL-8158, Livermore, CA
Murray CW, Racine SC, Davidson ER (1992) J Comput Phys 103:382
Head JD (1997) Int J Quantum Chem 65:827
Bour P, Sopkov J, Bednrov L, Malo
n P, Keiderling TA (1997) J Comput Chem
18:646659
Bour P, Kubelka J, Keiderling TA (2000) Biopolymers 53:380
Bour P, Kubelka J, Keiderling TA (2002) Biopolymers 65:45
Ham S, Cha S, Choi J-H, Cho M (2003) J Chem Phys 119:1451
Choi J-H, Ham S, Cho M (2003) J Phys Chem B 107:9132
Choi J-H, Cho M (2004) J Chem Phys 120:43834392
Hohenberg P, Kohn W (1964) Phys Rev 136:864871
Kohn W, Sham LJ (1965) Phys Rev 140:11331138
Parr RG, Yang W (1989) Density-Functional Theory of Atoms and Molecules. International Series of Monographs on Chemistry, Vol 16. Oxford Science Publications,
New York
Baerends EJ, Ellis DE, Ros P (1973) Chem Phys 2:41
Dunlap BI, Connolly JWD, Sabin JR (1979) J Chem Phys 71:33963402
Andrae D, Hussermann U, Dolg M, Stoll H, Preuss H (1990) Theor Chim Acta
77:123
Tamm I (1945) J Phys (USSR) 9:449
Dancoff SM (1950) Phys Rev 78:382
Garavelli M (2006) Theor Chem Acc 116:87105
Dreuw A, Head-Gordon M (2005) Chem Rev 105:40094037
Andruni
ow T, Ferr N, Olivucci M (2004) Proc Natl Acad Sci USA 101:17908
17913
Sinicropi A, Andruniow T, Ferre N, Basosi R, Olivucci M (2005) J Am Chem Soc
127:1153411535
Helms V (2002) Curr Opin Struc Biol 12:169175
Reiher M (2003) Found Chem 5:2341
Reiher M (2003) Found Chem 5:147163
Kirchner B, Reiher M (2005) J Am Chem Soc 127:87488756
Kirchner B (2005) J Chem Phys 123:204116
Tomasi J, Mennucci B, Cammi R (2005) Chem Rev 105:29993093
Cramer CJ (2004) Essentials of Computational Chemistry: Theories and Models. 2nd
edn. Wiley, New York
Riccardi D, Schaefer P, Yang Y, Yu H, Ghosh N, Prat-Resina X, Knig P, Li G, Xu D,
Guo H, Elstner M, Cui Q (2006) J Phys Chem B 110:64586469
Friesner RA, Guallar V (2005) Annu Rev Phys Chem 56:389427
Shurki A, Warshel A (2003) Adv Prot Chem 66:249313
Gao J, Truhlar DG (2002) Annu Rev Phys Chem 53:467505
Wesolowski TA, Warshel AJ (1993) J Phys Chem 93:80508053
128
59. Neugebauer J, Louwerse MJ, Baerends EJ, Wesolowski TA (2005) J Chem Phys 122:
094115
60. Neugebauer J, Reiher M, Kind C, Hess BA (2002) J Comput Chem 23:895910
61. Barron LD, Gray CG (1973) J Phys A 6:59
62. Placzek G (1934) Rayleigh-Streuung und Raman-Effekt. In: Marx E (ed) Handbuch
der Radiologie, Vol 6. Akademische Verlagsgesellschaft, Leipzig
63. Born M, Kuang K (1954) Dynamical Theory of Crystal Lattices. Clarendon Press,
Oxford
64. Barron LD (2004) Molecular Light Scattering and Raman Optical Activity. 2nd edn.
Cambridge University Press, Cambridge
65. Long DA (2002) The Raman Effect: A Unied Treatment of the Theory of Raman
Scattering by Molecules. Wiley, New York
66. Klhn M, Mathias G, Ktting C, Nonella M, Schlitter J, Gerwert K, Tavan P (2004)
J Phys Chem A 108:61866194
67. Reiher M, Neugebauer J, Hess BA (2003) Z Phys Chem 217:91103
68. Polavarapu PL (2006) Int J Quantum Chem 106:18091814
69. Zuber G, Hug W (2004) J Phys Chem A 108:21082118
70. Reiher M, Liegeois V, Ruud K (2005) J Phys Chem A 109:75677574
71. Siebert F, Mntele W, Kreutz W (1980) Biophys Struct Mech 6:139146
72. Ktting C, Gerwert K (2005) Chem Phys Chem 6:881888
73. Barth A, Zscherp C (2002) Quarterly Reviews of Biophysics 35:369430
74. Kottke T, Batschauer A, Ahmad M, Heberle J (2006) Biochemistry 45:24722479
75. Tuma R (2005) J Raman Spectrosc 36:307319
76. Benevides JM, Overman SA, Thomas GJ Jr (2005) J Raman Spectrosc 36:279299
77. Serban D, Benevides JM, Thomas GJ Jr (2003) Biochemistry 42:73907399
78. Callender R, Deng H (1994) Annu Rev Biophys Biomol Struct 23:215245
79. Hayashi S, Tajkhorshid E, Pebay-Peyroula E, Royant A, Landau EM, Navarro J, Schulten K (2001) J Phys Chem B 105:1012410131
80. Rajamani R, Gao JL (2002) J Comput Chem 23:96105
81. Rousseau R, Kleinschmidt V, Schmitt UW, Marx D (2004) Angew Chem Int Ed
43:48044807
82. Klhn M, Schlitter J, Gerwert K (2005) Biophys J 88:38293844
83. Topol IA, Cachau RE, Nemukhin AV, Grigorenko BL, Burt SK (2004) Biochim Biophys
Acta 1700:125136
84. Freindorf M, Shao Y, Brown ST, Kong J, Furlani TR (2006) Chem Phys Lett 419:563
566
85. Cui Q, Karplus M (2000) J Chem Phys 112:11331149
86. Ruckebusch C, Duponchel L, Sombret B, Huvenne JP, Saurina J (2003) J Chem Inf
Comput Sci 43:19661973
87. Hessling B, Souvignier G, Gerwert K (1993) Biophys J 65:1929
88. Cervetto V, Helbig J, Bredenbeck J, Hamm P (2004) J Chem Phys 121:59355942
89. Hamm P, Lim M, Hochstrasser RM (1998) J Phys Chem B 102:61236138
90. Maekawa H, Toniolo C, Moretto A, Broxterman QB, Ge N-H (2006) J Phys Chem B
110:58345837
91. Ozaki Y, Murayama K, Wu Y, Czarnik-Matusewicz B (2003) Spectroscopy 17:79100
92. Hamm P, Lim M, De Grado WF, Hochstrasser RM (1999) Proc Natl Acad Sci USA
96:20362041
93. Hamm P, Hochstrasser RM (2001) In: Fayer MD (ed) Ultrafast Infrared and Raman
Spectroscopy. Marcel Dekker, New York
94. Jonas DM (2003) Annu Rev Phys Chem 54:425463
129
95. Hochstrasser RM (2006) Dynamical Models for Two-Dimensional Infrared Spectroscopy of Peptides. In: Berry S, Jortner J (eds) Advances in Chemical Physics,
Vol 132. John Wiley, Hoboken, New Jersey
96. Scheurer C, Mukamel S (2002) J Chem Phys 116:68036816
97. Chin W, Piuzzi F, Dimicoli I, Mons M (2006) Phys Chem Chem Phys 8:10331048
98. Tokmakoff A, Lang MJ, Larsen DS, Fleming GR, Chernyak V, Mukamel S (1997) Phys
Rev Lett 79:27022705
99. Chen PC, Joyner CC (2006) J Phys Chem A 110:79897993
100. Woutersen S, Hamm P (2001) J Chem Phys 115:77377743
101. Voronine D, Abramavicius D, Mukamel S (2006) J Chem Phys 124:034104
102. Khalil M, Tokmakoff A (2001) Chem Phys 266:213230
103. Bredenbeck J (2005) Transient 2D-IR Spectroscopy: Towards Ultrafast Structural Dynamics of Peptides and Proteins. Thesis, Universitt Zrich
104. Dreyer J (2005) Int J Quantum Chem 104:782793
105. Neugebauer J, Hess BA (2003) J Chem Phys 118:72157225
106. Golonzka O, Khalil M, Demirdven N, Tokmakoff A (2001) Phys Rev Lett 86:2154
2157
107. Mukamel S (1995) Principles of Nonlinear Optical Spectroscopy. Oxford Series in
Optical and Imaging Sciences, Vol 6. Oxford University Press, Oxford
108. Zhang WM, Chernyak V, Mukamel S (1999) J Chem Phys 110:50115028
109. Wang J, Hochstrasser RM (2004) Chem Phys 297:195
110. Abramavicius D, Zhuang W, Mukamel S (2004) J Phys Chem B 108:1803418045
111. Wang J, Chen J, Hochstrasser RM (2006) J Phys Chem B 110:75457555
112. Gorbunov RD, Kosov DS, Stock G (2005) J Chem Phys 122:224904
113. Scheurer C, Piryatinski A, Mukamel S (2001) J Am Chem Soc 123:31143124
114. Zhuang W, Abramavicius D, Hayashi T, Mukamel S (2006) J Phys Chem B 110:3362
3374
115. Woutersen S, Mu Y, Stock G, Hamm P (2001) Proc Natl Acad Sci USA 98:1125411258
116. Barron LD, Boogard MP, Buckingham AD (1973) J Am Chem Soc 95:603
117. Holzwarth G, Hsu EC, Mosher HS, Faulkner TR, Moscowitz A (1974) J Am Chem Soc
96:251252
118. Kapitn J, Baumruk V, Kopeck
y V, Bour P (2006) J Phys Chem A 110:46894696
119. Qu X, Lee E, Yu G-S, Freedman TB, Nae LA (1996) Appl Spectrosc 50:649657
120. Polavarapu PL, Zhao C (2000) Fresenius J Anal Chem 366:727734
121. Taniguchi T, Miura N, Nishimura S-I, Monde K (2004) Mol Nutr Food Res 48:246254
122. Keiderling TA (2002) Curr Opin Chem Biol 6:682688
123. Barron LD, Hecht L, Bell AF, Wilson G (1996) Appl Spectrosc 50:619629
124. Barron LD, Ford SJ, Bell AF, Wilson G, Hecht L, Cooper A (1994) Faraday Discuss
99:217232
125. Alasdair LH, Bell F, Barron LD (1994) J Am Chem Soc 116:51555161
126. Blanch EW, Hecht L, Syme CD, Volpetti V, Lomonossoff GP, Nielsen K, Barron LD
(2002) J Gen Virol 83:25932600
127. Mc Coll IH, Blanch EW, Hecht L, Barron LD (2004) J Am Chem Soc 126:81818188
128. Mc Coll IH, Blanch EW, Gill AC, Rhie AGO, Ritchie MA, Hecht L, Nielsen K, Barron LD (2003) J Am Chem Soc 125:1001910026
129. Tam CN, Bour P, Keiderling TA (1997) J Am Chem Soc 119:70617064
130. Barron LD, Buckingham AD (1971) Mol Phys 20:1111
131. Nae LA, Freedman TB (1983) J Chem Phys 78:71087116
132. Stephens PJ (1985) J Phys Chem 89:748752
133. Barron LD, Hecht L, Mc Coll IH, Blanch EW (2004) Mol Phys 102:731744
130
134. Nae LA (1997) Annu Rev Phys Chem 48:357386

135. Polavarapu PL (1998) Vibrational Spectra: Principles and Applications with Emphasis in Optical Activity. Studies in Physical and Theoretical Chemistry, Vol 85.
Elsevier, New York
136. Rosenfeld LZ (1928) Z Physik 52:161
137. Galwas PA (1983) Thesis, Cambridge University
138. Buckingham AD, Fowler PW, Galwas PA (1987) Chem Phys 112:114
139. Amos RD, Handy NC, Jalkanen KJ, Stephens PJ (1987) Chem Phys Lett 133:2126
140. Lowe MA, Segal GA, Stephens PJ (1986) J Am Chem Soc 108:248256
141. Stephens PJ, Develin FJ (2000) Chirality 12:172179
142. Freedman TB, Cao X, Dukor RK, Nae LA (2003) Chirality 15:743758
143. Atkins PW, Barron LD (1969) Mol Phys 16:453
144. Buckingham AD, Dunn MB (1971) J Chem Soc A, p 1988
145. Quinet O, Ligeois V, Champagne B (2005) J Chem Theory Comput 1:444452
146. Helgaker T, Ruud K, Bak KL, Jrgensen P, Olsen J (1994) Faraday Discuss 99:165
147. Ruud K, Helgaker T, Bour P (2002) J Phys Chem A 106:74487455
148. Pecul M, Ruud K (2005) Int J Quantum Chem 104:816829
149. Crawford TD (2006) Theor Chem Acc 115:227245
150. Kapitn J, Baumruk V, Bour P (2006) J Am Chem Soc 128:24382443
151. He J, Petrovich A, Polavarapu PL (2004) J Phys Chem A 108:16711680
152. He J, Petrovic AG, Polavarapu PL (2004) J Phys Chem B 108:2045120457
153. van Gisbergen SJA, Snijders JG, Baerends EJ (1999) Comput Phys Comm 118:119138
154. Zuber G, Goldsmith M-R, Beratan DN, Wipf P (2005) Chem Phys Chem 6:595597
155. Bour P (2001) J Comput Chem 22:426435
156. Cappelli C, Monti S, Rizzo A (2005) Int J Quantum Chem 104:744757
157. Pecul M, Lamparska E, Cappelli C, Frediani L, Ruud K (2006) J Phys Chem A
110:28072815
158. Jalkanen KJ, Nieminen RM, Friemand K, Bohr J, Bohr H, Wade RC, Tajkhorshid E,
Suhai S (2001) Chem Phys 265:125151
159. Nae LA (1997) J Phys Chem A 101:78267833
160. Freedman TB, Shih M-L, Lee E, Nae LA (1997) J Am Chem Soc 119:1062010626
161. Freedman TB, Gao X, Shih M-L, Nae LA (1998) J Phys Chem A 102:33523357
162. Freedman TB, Lee E, Nae LA (2000) J Phys Chem A 104:39443951
163. Hug W (2001) Chem Phys 264:5369
164. Hug W, Haesler J (2005) Int J Quantum Chem 104:695715
165. Zuber G, Hug W (2004) Helv Chim Acta 87:22082234
166. Spiro TG, Czernuszewicz RS (2000) In: Que L Jr (ed) Physical Methods in Bioinorganic Chemistry. University Science Books, Sausalito, CA
167. Loehr TM (1998) In: Solomon EI, Hodgson KO (eds) Spectroscopic Methods in
Bioinorganic Chemistry. American Chemical Society, Washington, DC
168. Czernuszewicz RS, Spiro TG (1999) In: Solomon EI, Lever ABP (eds) Inorganic Electronic Structure and Spectroscopy, Vol 1. Wiley, New York
169. Petry R, Schmitt M, Popp J (2003) Chem Phys Chem 4:1430
170. Egawa T, Yeh S-R (2005) J Inorg Biochem 99:7296
171. Kitagawa T, Mizutani Y (1994) Coord Chem Rev 135/l36:685735
172. Spiro TG, Wasbotten IH (2005) J Inorg Biochem 99:3444
173. Hudecek J, Anzenbacherov E, Anzenbacher P, Munro AW, Hildebrandt P (2000)
Arch Biochem Biophys 383:7078
174. Spiro TG (1983) In: Lever ABP, Gray HB (eds) Iron Porphyrins, Part 2. AddisonWesley, Reading, MA
131
175. Spiro TG, Li XY (1988) In: Spiro TG (ed) Resonance Raman Spectra of Heme- and
Metalloproteins. Wiley, New York
176. Berthomieu C, Hienerwadel R (2005) Biochim Biophys Acta 1707:5166
177. Asher SA, Mikhonin AV, Bykov S (2004) J Am Chem Soc 126:84338440
178. Mikhonin AV, Ahmed Z, Ianoul A, Asher SA (2004) J Phys Chem B 108:1902019028
179. Huang C-Y, Balakrishnan G, Spiro TG (2006) J Raman Spectrosc 37:277282
180. Jeong DH, Kim D, Cho DW, Jeoung SC (2001) J Raman Spectrosc 32:487493
181. Pinakoulaki E, Koutsoupakis C, Stavrakis S, Aggelaki M, Papadopoulos G, Daskalakis V, Varotsis C (2005) J Raman Spectrosc 36:337349
182. Negrerie M, Cianetti S, Vos MH, Martin J-L, Kruglik SG (2006) J Phys Chem B
110:1276612781
183. Peticolas WL, Rush T III (1995) J Comput Chem 16:12611270
184. Hassing S, Mortensen OS (1980) J Chem Phys 73:10781083
185. Myers AB (1996) Chem Rev 96:911
186. Lee S-Y, Heller EJ (1979) J Chem Phys 71:4777
187. Heller EJ, Sundberg RL, Tannor D (1982) J Phys Chem 86:1822
188. Neugebauer J, Hess BA (2004) J Chem Phys 120:1156411577
189. Neugebauer J, Baerends EJ, Efremov EV, Ariese F, Gooijer C (2005) J Phys Chem A
109:2100
190. Caillie CV, Amos RD (1999) Chem Phys Lett 308:249255
191. Caillie CV, Amos RD (2000) Chem Phys Lett 317:159164
193. Hutter J (2003) J Chem Phys 118:39283934
194. Rappoport D, Furche F (2005) J Chem Phys 122:064105
195. Jarzecki AA, Spiro TG (2005) J Phys Chem A 109:421430
196. Petrenko T, Ray K, Wieghardt KE, Neese F (2006) J Am Chem Soc 128:44224436
197. Schulman JM, Detrano R, Musher JI (1972) Phys Rev A 5:1125
198. Schulman JM, Detrano R (1974) Phys Rev A 10:1192
199. Schulman JM, Lee S-Y (1981) J Chem Phys 74:4930
200. Lee S-Y (1982) J Chem Phys 76:3064
201. Lee S-Y (1983) J Chem Phys 78:723
202. Jensen L, Zhao LL, Autschbach J, Schatz GC (2005) J Chem Phys 123:174110
203. Tozzini V, Giannozzi P (2005) Chem Phys Chem 6:17861788
204. Zheltikov AM (2000) J Raman Spectrosc 31:653667
205. Begley RF, Harvey AB, Byer RL (1974) Appl Phys Lett 25:387390
206. Rinia HA, Bonn M, Mller MU (2006) J Phys Chem B 110:44724479
207. Schlcker S, Koster J, Nissum M, Popp J, Kiefer W (2001) J Phys Chem A 105:9482
9488
208. Otto C, Voroshilov A, Kruglik SG, Greve J (2001) J Raman Spectrosc 32:495501
209. Popp J, Popp A, Ujj L, Atkinson GH, Sheves M, Ottolenghi M (1996) J Ram Spectrosc
27:8795
210. Wohlleben W, Buckup T, Herek JL, Motzkus M (2005) Chem Phys Chem 6:1
211. Knutsen KP, Messer BM, Onorato RM, Saykally RJ (2006) J Phys Chem B 110:5854
5864
212. Zumbusch A, Holtom G, Sunney Xie X (1999) Phys Rev Lett 82:41424145
213. van Manen H-J, Kraan YM, Roos D, Otto C (2005) Proc Natl Acad Sci USA
102:1015910164
214. Kiefer W, Materny A, Schmitt M (2002) Naturwissenschaften 89:250258
215. Volkmer A, Book LD, Xie XS (2002) Appl Phys Lett 80:15051507
216. Konradi J, Singh AK, Scaria AV, Materny A (2006) J Raman Spectrosc 37:697704
132
217. Siebert T, Maksimenka R, Materny A, Engel V, Kiefer W, Schmitt M (2002) J Raman

Spectrosc 33:844854
218. Siebert T, Schmitt M, Grfe S, Engel V (2006) J Raman Spectrosc 37:397403
219. Stock G, Domcke W (1997) Adv Chem Phys 100:1
220. Eesley GL (1981) Coherent Raman Spectroscopy. Pergamon Press, Oxford
221. Yuratich MA, Hanna DC (1977) Mol Phys 32:671
222. Teets RE (1984) Opt Lett 9:226228
223. Funk J-M, Schmitt U, Kiefer W, Materny A (2000) J Raman Spectrosc 31:743753
224. Sariyanni Z-E, Rostovtsev Y (2004) J Mod Optic 51:26372644
225. Vargek M, Freedman TB, Lee E, Nae LA (1998) Chem Phys Lett 287:359364
226. Nae LA (1996) Chem Phys 205:309322
227. Scheidt WR, Durbin SM, Sage JT (2005) J Inorg Biochem 99:6071
228. Sage JT, Durbin SM, Sturhahn W, Wharton DC, Champion PM, Hession P, Sutter J,
Alp EE (2001) Phys Rev Lett 86:49664969
229. Xiao Y, Wang H, George SJ, Smith MC, Adams MWW, Jenney FE Jr, Sturhahn W,
Alp EE, Zhao J, Yoda Y, Dey A, Solomon EI, Cramer SP (2005) J Am Chem Soc
127:1459614606
230. Adams KL, Tsoi S, Yan J, Durbin SM, Ramdas AK, Cramer WA, Sturhahn W, Alp EE,
Schulz C (2006) J Phys Chem B 110:530536
231. Xiao Y, Fisher K, Smith MC, Newton WE, Cass DA, George SJ, Wang H, Sturhahn W,
Alp EE, Zhao J, Yoda Y, Cramer SP (2006) J Am Chem Soc 128:76087612
232. Herrmann C, Ruud K, Reiher M (2006) Chem Phys Chem 7:21892196
Top Curr Chem (2007) 268: 133171

DOI 10.1007/128_2006_083
CarParrinello Molecular Dynamics Simulations

and Biological Systems
Jens Thar Werner Reckien Barbara Kirchner (u)
Institute for Physical and Theoretical Chemistry, Wegelerstr. 12, 53115 Bonn, Germany
kirchner@thch.uni-bonn.de
1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
134
2
2.1
2.2
Molecular Dynamics Simulations . . . . . . . . . . . . . . . . . . . . . . .

Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Thermostats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
137
137
139
3
3.1
3.2
3.3
Electronic Structure . . . . . . . . . . . . . . . . . . . . . . .
HohenbergKohn Theorems . . . . . . . . . . . . . . . . . .
KohnSham Equations . . . . . . . . . . . . . . . . . . . . .
Exchange Correlation Functional and Dispersion Corrections
.
.
.
.
141
141
142
143
4
4.1
4.2
Basis Sets and Pseudopotentials . . . . . . . . . . . . . . . . . . . . . . . .

Plane Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pseudopotentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
145
145
147
5
5.1
CarParrinello Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . .
Fictitious Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
149
150
QM/MM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
151
7
7.1
7.2
7.3
7.4
7.5
Literature on Biological Systems . . . . . . . . . .

Proteins, Peptides, and Amino Acids . . . . . . . .
Metal-Containing Biomolecules: Metallo-enzymes
DNA and RNA . . . . . . . . . . . . . . . . . . . .
Miscellaneous, Biomimetics . . . . . . . . . . . .
Excited States MD . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
155
155
156
157
158
158
Case Study: Biomimetic Nitrogen Fixation . . . . . . . . . . . . . . . . . .
159
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
165
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
166
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Abstract The basic concepts of CarParrinello molecular dynamics (CPMD) and its application to biological systems are reviewed. In CarParrinello simulations the electronic
structure is calculated on the y, i.e., during the course of simulations the potential is
adjusted according to the chemical events occurring in the system. This allows for more
unbiased simulations of chemical processes, as opposed to classical molecular dynamics,
which utilize predened potentials. In contrast to BornOppenheimer molecular dynamics, where the nuclei are propagated under classical equations of motions and where the
wave function is calculated in every time step, CPMD maps this two-component classi-
134
J. Thar et al.
cal/quantum system onto a two-component purely classical system. A short introduction

into classical mechanics, which provides the equations of motions, is given. DFT will be
briey outlined because it represents the standard method for obtaining the electronic
structure during a CPMD simulation. QM/MM approaches, which combine the CPMD
method for a core region with a classical description of its environment, are also reviewed. These schemes provide a more reliable description of biological systems, whose
properties are strongly inuenced by their surroundings. A brief overview over the application of CarParrinello simulations to biological systems is provided. Finally, a case
study to illustrate some of the basic possibilities of this method is discussed.
Abbreviations
BOMD BornOppenheimer molecular dynamics
CPMD CarParrinello molecular dynamics
DFT
Density functional theory
GGA
Generalized gradient approximation
HF
HartreeFock
LDA
Local density approximation
MD
Molecular dynamics
MM
Molecular mechanics
MP2
Second order MllerPlesset theory
PP
Pseudopotential
PW
Plane wave
QM/MM Quantum mechanics/molecular mechanics
TDDFT Time-dependent density functional theory
WDA
Weighted density approximation
1
Introduction
Chemical and biological systems are far from simple and can encompass
a complicated interplay between many different sorts of interactions of many
different functional groups. Furthermore, these systems are often not stiff,
but exible and therefore of a dynamical nature. In view of this complicated nature of biological and complex chemical systems the right choice of
a method that is able to include all of these effects is necessary. While traditional, static quantum chemical methods are able to resolve, for instance,
the geometrical or electronic behavior of a relatively small amount of atoms
at a high level of accuracy, traditional molecular dynamics simulation techniques are able to include a large amount of atoms and to temporally resolve
the microscopic level. Thus, when interested in the high resolution of a local
area within the biomolecule, quantum chemical methods are certainly the
right choice. On the other hand, one might be interested in solvent effects
(and therefore be in need of a large amount of atoms) or in the dynamical behavior (and be in need of a dynamical description of the nuclei) so
the right choice of methods will be given by traditional molecular dynamics
CarParrinello Molecular Dynamics
135
simulations. Situations might occur where both a quantum chemical treatment of the electronic structure and a dynamical description of the nuclei are
needed. In 1985, Car and Parrinello introduced with their seminal paper [1]
one such method that is able to combine the advantages of both methods.
They used as an elegant trick the extended Lagrangian technique to map
a two-component classical/quantum mechanical treatment (atoms/electrons)
onto a two-component purely classical treatment. As a result the method is
suitable:
To follow the temporal resolution of the atoms, which is not possible with
static quantum chemical methods, see Fig. 1
To describe solvent effects, since an ensemble of molecules with periodic
boundary conditions can be treated, see Fig. 2 (right picture)
To describe spontaneous events, which is impossible for traditional molecular dynamics simulations that rely on predened potentials, see Fig. 2
(left picture)
To include polarization or cooperative effects, because the many-body nature is properly treated within this method, see Fig. 3
Fig. 1 Diffusion of DMSO molecules in water. The water molecules are dispensed with for
clarity. For simulations see [24]
Fig. 2 Left: Cartoon of a spontaneous event (a double hydrogen transfer in this case), see
Sellmann complexes in Sect. 8 and [5]. Right: Illustration of solvent effects at the example
of an articial guanidinium receptor, Kirchner et al. unpublished results
136
J. Thar et al.
Fig. 3 Various water clusters showing cooperative effects. The numbers give the percentage of energy that is covered by assuming pair additivity, see [6]
Since there is no free meal, disadvantages of the CarParrinello method,

like short simulation times and lower quantum chemical accuracy, are
a byproduct of the limited amounts of computer time available today. Those
disadvantages will be discussed in the relevant sections below. CarParrinello
simulations have already been the subject of many review articles. For additional information, the reader is referred to the excellent reviews of Remler
and Madden [7], Marx and Hutter [8], and Tuckerman and Martyna [9, 10].
The next few sections describe the CarParrinello methodology, which is the
central topic of this article. Since the CarParrinello method uses molecular
dynamics simulations, we will start with a very brief outline of the methodology, however, we will not go into detail because this topic is covered by
another chapter (see the review of Schulten et al. in this volume). The molecular dynamics discussion is followed by a section on electronic structure
method, where only density functional theory is introduced. The reason for
this is that up to now the CarParrinello method is mostly implemented on
the basis of density functional theory. This is not due to shortcomings of
the method, but due to the reasonable amount of computer time available.
After the discussion of the electronic structure method we shed some light
onto the basis sets and the pseudopotentials used. Next, we turn to the Car
Parrinello method in detail. This is followed by a short introduction of the
QM/MM methodology, as most often biological systems are studied using the
CarParrinello method in the framework of a QM/MM ansatz. Again we do
not go into much detail, since this topic is covered by Senn and Thiel in this
volume. We continue with an overview of articles that have appeared in the
literature and have applied the CarParrinello method to biological systems;
one selected example will be highlighted. Finally, conclusions and discussions
end this review article.
137
2
Molecular Dynamics Simulations
We will rst briey review the basics of molecular dynamics simulations,
which are required to understand the ideas behind rst principle molecular
dynamics simulations [11]. The aim of molecular dynamics simulations is to
study a system by recreating it on the computer in a manner that is as close to
nature as possible, i.e., by simulating the dynamics of a system over a physical length of time relevant to properties of interest. Therefore, there are two
topics at the heart of molecular dynamics simulations. First, the equations
of motion must be dened and second, the forces by which the particles are
moved must be described.
To this end, classical mechanics is applied to describe the nuclear motion
of the particles, i.e., the equations of motion. This is an approximation. Extensions that include quantum effects [12, 13] have been explored. It turns out
that employing classical mechanics is a very good approximation for molecular systems, as long as the properties studied are not related to vibrations with
a frequency such that h kb T.
2.1
Equations of Motion
The equations of motion may be written in various ways. We assume a system of N particles interacting via a potential U. Each particle I (usually an
atom) is described by its position RI and conjugated momenta PI = MI R I .
Here MI is the mass of the particle. In our simulations we have a set of all
positions {R1 , R2 , ...RN } and of all momenta {P1 , P2 , ...PN }, which are abbreviated RN and PN , respectively. The equations of motion can be derived from
the EulerLagrange equations:
d L L
=0,
dt R R
(1)
where the Lagrangian function L is given in terms of kinetic minus potential

energy:
N) = K U =
L(RN , R
N

1
I=1
MI R 2I U(RN ) .
(2)
The Lagrangian formalism allows for an easy introduction of restraints by

simply adding the constraining condition. Alternatively, the Hamiltonian for-
138
J. Thar et al.
malism can be used:

N) =
H (RN , R
N
N N
N
R N
I PI L(R , R )
I=1
=K +U=
N

P2I
+ U(RN ) .
2MI
(3)
I=1
The force F I (RN ) on particle I is obtained as the change of U(RN ) due to

an innitesimal displacement of I, i.e., the partial derivative of the potential
U(RN ):
F I (RN ) =
U(RN )
.
RI
(4)
In traditional molecular dynamics simulations an analytical function like the

LennardJones potential is used for U(RN ), such that this analytical form is
simple enough to obtain the analytical derivative.
The description of these forces can be very important for the outcome of
the simulations [14, 15]. Usually the forces on the particles are operated pairwise additively, as already explained in the introduction. U(RN ) is then given
as a sum over two-body potentials u(RIJ ):
U(RN ) =
N
N

I
J>I
1
u(RIJ ) .
2 I
N
u(RIJ ) =
(5)
J =I
This is again an approximation, since many-body or cooperative effects are

neglected. As opposed to the quantum nature of the particle (nuclei) dynamics, cooperative effects are obviously more important in complicated liquids.
A better description of many-body effects is provided by polarizable forces
elds that make up for the neglect of such effects. A second way to circumvent
this problem is given by rst-principle simulations, which explicitly include
cooperative effects in the description of the system under consideration (as
we will see in the next sections) because here the complete electronic structure of the system is treated at once.
The equations of motion according to Hamiltons equation read:
R I =
H
PI
=
,
PI MI
P I =
H
U(RN )
=
= F I (RN ) .
RI
RI
(6)
(7)
Newtons second law can be obtained from this equation by using PI = MI R I ,

taking the time derivative and equating it to the second part of the Hamilton
139
equations and inserting it into Eq. 7:

I = F I (RN ) .
MI R
(8)
Having arrived at the equations of motion, numerical integration helps to nd

the solution. One of the most prominent algorithms was invented by Strmer
and Verlet [16, 17].
Before discussing rst-principle simulations another approach to simulations should be mentioned, namely the extended system approach. In this,
additional degrees of freedom that control a quantity under consideration
can be added to the system [1619]. Thereby, thermostats or barostat can be
simulated [2024].
2.2
Thermostats
The most forward way to keep the temperature (i.e., the kinetic energy) constant in classical molecular dynamics simulations is to scale the velocities of
every single particle [16]. However, one can show that the average temperature of a NVT ensemble (i.e., an ensemble where the number of particles
N, the volume V, and the temperature T are kept constant) uctuates, and
that therefore the scaling of every single velocity can lead to an unphysical system [16]. A convenient method to keep the NVT ensemble is the
Nose(Hoover)-thermostat [20, 21, 23, 24], where a heat bath is coupled to the
system and energy can ow freely between these two systems. The heat bath
is represented by an additional degree of freedom S, while the energy ow
is described by the conjugated momentum PS of S. The complete system is
R, S , S) [20, 21],
describe by the extended Lagrangian: LNose (R,
R, S, S)
LNose = LNose (R,
=
N

MI
I=1
I U(RN ) +
S2 R
Q 2
S Lkb TlnS ,
2
(9)
where kb is the Boltzmann constant and L is a chosen parameter to obtain the

canonical average. Q is considered an effective mass, and controls the speed
with which energy is transferred between the system and the reservoir. It
must be chosen with care, because small values for Q lead to a fast oscillation of the extended variable S, and therefore require a very small time step
to accurately describe the extended system. However, this leads to more time
steps needed to equilibrate the system, which is to be kept at constant temperature. On the other hand, large values of Q lead to a very slow response of
S to changes in this system, and here the convergence of the additional degree
of freedom S requires a larger number of time steps [25]. With the conjugated
140
J. Thar et al.
momenta:
PI
L
I,
= MI S2 R
I
R
(10)
PS
= QS ,
S
(11)
one obtains the Hamiltonian HNose for the complete system:
R, S, S)
HNose = HNose (R,
=
N

I=1
P2I
P2
+ U(RN ) + S + Lkb T ln S .
2
2MI S
2Q
(12)
(13)
The equations of motion are therefore given by:

dRI HNose
PI
=
=
,
dt
PI
MI S2
(14)
dPI
HNose
U(RN )
=
=
,
dt
RI
RI
(15)
dS HNose PS
=
,
=
dt
PS
Q
PI2
Lkb T
M I S2
dPS
HNose
I
=
=
.
dt
S
S
(16)
(17)
For this selection, one must choose L = 3N + 1 to obtain a canonical system.

More commonly, one introduces the scaled variables V [22, 23], which are
also called real variables because they resemble physical observables most
closely. These variables are related to the initial variables by the following
equations:
R = R ,
(18)
P = P/S ,
(19)
S = S ,
(20)
PS = PS /S ,
(21)
t = t/S .
(22)
141
Using relations Eqs. 1822 and introducing the thermodynamic friction coefcient = S PS /Q, we can write the Eqs. 1417 in the following form:
PI
PI
dRI
dRI
=
=
=
S
,
dt
dt
MI S MI
(23)
dPI
d(PI /S) dPI 1 dS
=
PI
=S

dt
dt
dt
S dt
=
U(RN ) S PS
U(RN )
P
PI ,
RI
Q I
RI
1 dS S dS S PS
=
= ,
=
S dt S dt
Q
S PS
Q
dt
S dPS
Q dt
PI2
I MI
Lkb T
Q
= .
(24)
(25)
(26)
From these equations of motion one can see how the thermostat works:
The momentum of each particle depends on the friction coefcient through
Eq. 24, i.e., it is lowered if is positive or it rises if is negative. The friction
coefcient is related to the desired temperature T through Eq. 26. If the temperature of the system is too large, rises, which scales down the momenta.
3
Electronic Structure: Density Functional Theory
Because rst-principle simulations usually employ density functional theory
(DFT), we will give a brief outline of DFT. The reason for the extensive use
of DFT is its low cost while giving a comparatively good description of the
electronic structure.
3.1
HohenbergKohn Theorems
DFT is used as the formulation to describe the electronic structure based
on the HohenbergKohn theorems [26]. The rst HohenbergKohn theorem
states that the ground state density (r):

(r) = (r1 ) = N ... | (x1 , x2 , ..., xN )|2 ds1 dx2 ... dxN ,
(27)
142
J. Thar et al.
with the electronic wave function (x1 , x2 , ..., xN ) depending on the spatial
and spin coordinates xi and with N as the number of electrons in the system, determines the external potential V ext . This theorem is only valid for
a non-degenerate ground state. With N given by:

N = (r) dr ,
(28)
the complete Hamiltonian is known, which yields the wave function and
therefore all properties of the system. As a result, we can write the energy as
a functional of the density:
E[] = T[] + Vext [] + Vee []

= T[] + (r)v(r) dr + Vee [] ,
(29)
(30)
where T[] is the kinetic energy and Vee [] is the electronelectron interaction. The second HohenbergKohn theorem allows the use of the variational
principle so that a trial density provides an approximation E[ ] to the exact
ground state energy E0 []:
(31)
|H| = T[] + (r)v(r) dr + Vee [ ] = E[ ] E0 [] .
The remaining problem is to determine the functionals T[] and Vee [].
3.2
KohnSham Equations
The bright idea of Kohn and Sham was to replace the minimization with respect to an unknown ground state density by a minimization with respect to
a set of orthonormal one-particle orbitals [27]. This idea comprises a substitute of non-interacting electrons for the fully interacting N-electron system.
The auxiliary system is described by a single-determinant wave function
consisting of N orbitals i for N-independent electrons. Inserting this wave
function KS into Eq. 27 yields:
(r) =
N
|i (r)|2 ,
(32)
which equals the density of the fully interacting system. The energy of this
auxiliary system is given by:

Eaux [] = Ts [] + vs (r)(r) dr ,
(33)
143
where the kinetic energy term can be easily evaluated as:

Ts [] =

N

1
i 2 i .
2
i
(34)
Turning back to the system with interacting electrons, we can simply add
and subtract the energy expression in Eq. 33 from the energy of the interacting system. Rearranging the terms and introducing the exchangecorrelation
energy functional: Exc [],
Exc [] = T[] Ts [] + Vee [] J[] ,
(35)
where the Coulomb energy J is given by the classical Poisson integral [28]:

1
1
1

(r)(r
drVH (r)(r) ,
)
dr
dr
=
(36)
J[] =
2
|r r |
2
nally yields the KohnSham energy EKS []:

1
drVH (r)(r) + Exc [] ,
EKS [] = Ts [] + drv(r)(r) +
2
(37)
which can be varied with respect to the orbitals, subject to an orthonormality

constraint. This leads to the KohnSham equations:

Exc []
1 2
i (r) =
ij j (r) .
(38)
+ v(r) + VH (r) +
2
n(r)
j
These equations are one-particle equations involving an effective one-particle

Hamiltonian hKS (see terms in braces of Eq. 38). ij are the Lagrange multipliers for the constraint that orbitals are orthonormal. The art of density
functional theory is now to derive an accurate exchangecorrelation functional empirically. This will be discussed in the next section.
3.3
Exchange Correlation Functional and Dispersion Corrections
The rst approximation to the exchange correlation functional is the so called
local density approximation (LDA) [27]. To derive the functional, the Kohn
Sham energy is obtained for a homogeneous electron gas in a large box in
which a positive charge is spread out to neutralize the system. As a result,
the Coulomb interaction of the electrons with the positive charges (nuclei)
cancels exactly with the inter-electron interaction and the inter-nuclei interaction. The energy ELDA [{}] then nally reads:
ELDA [] = Ts [] + Exc [] ,
(39)
144
J. Thar et al.
where Exc [] can be split up into an exchange part Ex [] and a correlation

part Ec []:
E[] = Ts [] + Ex [] + Ec [] .
(40)
From this ansatz, Ex is obtained as:

Ex [] = 0.7386 (r)4/3 dr .
(41)
To get the correlation part, Nusair et al. [29] tted exact Monte Carlo calculations and obtained Ec [] as the difference between the exact energy and
Ts [] + Ex []. The LDA approximation has two main drawbacks: First, it
originates from a homogeneous electron gas, which is normally not found in
real systems. Also, the dependence on r in the limit of r is not correct.
Becke [30] introduced a gradient-corrected exchange functional term EBx of
the form:

x2
dr ,
(42)
EBx = (r) 1/3
(1 + 6x sin h1 x)
with:
x=
||
.
4/3
(43)
The functional is derived by imposing the correct asymptotic behavior as

a restrain. The parameter is adjusted to = 0.0042 so that EBx reproduces
the exchange energy of six noble gases. Therefore, this functional accounts for
both problems associated with the LDA functionals. It is one of the so-called
generalized gradient approximation (GGA) functionals, which are given by:

Exc = F( , , , , )dr ,
(44)
with = . One problem remains with all the above functionals:
dispersion interactions are not described sufciently well.
A wide variety of functionals to address this problem has been developed
so far: the GGA in combination with an (empirical) 1/r6 van der Waals correction term EvdW , the weighted density approximation (WDA), and methods
based on time-dependent density functional theory [31].
GGAs and 1/r6 van der Waals corrections terms were originally applied
to correct the HartreeFock energy [32]. Based on Anderson, Langreth, and
Lundqvist [33] the van der Waals term is corrected for both separated uniform electron gas and separated atoms with an explicit 1/r6 term. Others [34
38] include an empirical 1/r6 term. A cutoff criterion is employed to separate
the regions where the length scale for the change of the local Fermi wavevector is smaller than the electron screening length. Furthermore, because
of the divergence of this EvdW expression for short distances, damping functions were introduced. Becke and coworkers were able to develop a correction
145
scheme based on the idea of parameterizing the 1/r6 contribution generated

by the instantaneous dipole moment of the exchange hole [39].
The weighted density approximation [40] starts from the exact expression
for the exchange correlation functional to obtain a weighted density approximation. An unknown paircorrelation function enters the expression, which
leads to an introduction of a weighted density parameter. Several suitable
functions were proposed [41], among them functions of GunnarsonJones
type, Gaussians, fourth-order Gaussians, and Lorentzians.
The strategy for the seamless approach is to approximate the density distribution n(r) by LDA or GGA and to divide the Coulomb interaction U(r)
into short- and long-range parts. The contribution for the long-range interaction starts as in WDA from the exact expression. Unlike the WDA approach,
a polarization term enters the equation, which is transformed into a time domain and leads to complicated expressions. If pseudopotentials are used, the
long-range parts can also be approximated by a suitably chosen non-local
form [42]. This approach will be briey discussed in the next section.
4
Basis Sets and Pseudopotentials
We want to recall here that the basis sets used in quantum chemical calculations are usually built up by atom-centered functions. Within the category of
atomic orbitals, there are mostly two types of basis functions used: (1) Slatertype orbitals and (2) Gaussian-type orbitals [43, 44]. In principle, Gaussiantype orbitals are inferior to the Slater-type orbitals, but they are easier to
handle, e.g., two-electron integrals can be calculated analytically. Previously,
it was common to develop a basis set according to the special problems at
hand, for an example see [45]. Nowadays, many special Gaussian basis sets
exist and most often only exponents are changed or functions are added instead of developing a completely new basis set. However, the CPMD method
utilizes plane wave basis sets. The reason for this is that simulations are often
used to obtain insight into bulk materials like liquids or solids, which require
periodic boundary conditions to avoid surface effects. While periodic boundary conditions are naturally incorporated into plane waves, this is not the case
for Gaussian-type orbitals, and additional problems arise, which need to be
handled with care [4649].
4.1
Plane Waves
Plane waves [7, 10]:
fGPW (r) = exp[iG r] ,
(45)
146
J. Thar et al.
form a complete and orthonormal basis with an inherent periodicity, where

G is the reciprocal space vector and r is the electronic coordinate. If a simulation takes place in a cubic box, this box is replicated throughout space to form
an innite lattice. Plane waves are not atom-centered and therefore allow for
a steady description of the wave functions at all positions. This means that the
simulation is not biased by an unequally good description of different spatial
regions. A periodic function can be expanded in the basis of the form given
in Eq. 45:

(r) = (r + L) =
(G) exp[iG r] .
(46)
G
The functions (r) and (G) are related by a three-dimensional Fourier

transform. The direct lattice vectors L connect equivalent points in different
cells. The KohnSham orbitals can be written in Bloch form:
i (r) = i (r, k) = exp[ik r]ui (r, k) ,
(47)
where k is a vector in the rst Brillouin zone. The functions ui (r, k) hold the
periodicity of the direct lattice:
ui (r, k) = ui (r + L, k) .
(48)
The index i runs over all one-particle states and these are associated with an
occupation. Expanding the periodic functions ui (r, k) in the plane wave basis
leads to:

ui (r, k) =
ci (G, k) exp[iG r] ,
(49)
G
and the KohnSham orbitals are given as:

ci (G, k) exp[i(G + k) r] .
i (r, k) =
(50)
When the density is used in the calculations, the integration over the Brillouin
zone has to be approximated by a nite sum over special k-points. Furthermore, the innite sum over G vectors and cells (stems from the prefactor,
which is omitted here for brevity) has to be truncated. Since the KohnSham
potential converges rapidly with increasing modulus of G, such G vectors with
a kinetic energy higher than a given maximum cutoff Ecut are neglected:
1 2
|G| Ecut .
(51)
2
The drawback of the uniform description is that regions that are situated far
apart from the nuclei, and therefore do not contribute signicantly to the
wave function of the system, are described by the same high accuracy and the
therewith connected CPU time as the regions close to the nuclei. This should
not be a severe problem as long as the wave function at the nuclei is described
properly with a reasonable small number of plane waves, i.e., with a rather
147
small cutoff. Unfortunately, this is not the case, because the wave function
changes rapidly close to the nuclei, and normally shows several nodes in this
region. This is the reason why the plane wave description is usually accompanied by pseudopotentials [8].
Another approach uses a mixed auxiliary basis set to describe the density
simply by expanding the orbitals in Gaussians [50, 51]. Such approaches are
available in standard program packages such as the CP2K code [52].
4.2
Pseudopotentials
The rapid change of the wave function in the inner core region of each atom
requires a prohibitively large set of plane waves. This region does not contribute much to the chemical binding so an approximate, but much smoother
(and therefore less complicated), description of the wave function in this region is feasible [8, 53]. This can be achieved by altering the potential of the
nuclear charges and the core electrons inside an (arbitrarily) chosen cutoff radius rc , i.e., by replacing the old potential by a pseudopotential. This
pseudopotential should fulll a number of requirements [54]. First it should
ps
generate a new orbital l = l (, r) that reproduces the exact valence shell
eigenvalues as well as the exact all-electron orbital lae = lae (, r) outside rc .
It should be norm conserving, i.e., the number of electrons in the core region
should be identical for both l and lae . The pseudoorbital l should have the
same logarithmic derivative as well as the same rst energy derivative of the
logarithmic derivative as lae outside rc . A major breakthrough was achieved
by Hamann, Schlter, and Chiang [54] who showed that norm conservation
and the energy derivative of the logarithmic derivative (at the distance R) are
related through the identity:

d d
ln l
2 (rl )
d dr
R
l2 (r)r2 dr .
= 4
R
(52)
To prove this identity [55], one must consider the radial equation:

l(l + 1)
1
l +
+ Veff (r) l = 0
2
2r2
for a spherical atom, with l =
d
dr l .
(53)
This relation can be reformulated into:
2
l
l(l + 1)
d l
+ Veff (r) =
=
+ l2 = l + l2 ,
2
2
2r
l dr l
l

(54)
148
J. Thar et al.
dening the function l = l (, r) as:

l

d
ln l = l .
dr
l
(55)
Differentiating Eq. 54 with respect to yields:

+ 2l l .
l

For any arbitrary function f (r) and any l the relation:
2=
(56)
l
1 2

[
f
(r)]
=
f
(r)
+
2
f (r) = f (r) + 2l f (r)
l
l2 r l
(57)
l into Eq. 56, multiplying by

is valid. Inserting this equation for f (r) =
2
l (r) and integrating over the complete space, one nally obtains the above
identity. The pseudopotential is obtained by optimizing lae and tting l with
the new pseudopotential under the above-mentioned constraints to the real
eigenvalue. Because the pseudopotential must be calculated for every angular
momentum, but acts only as a linear operator on r, it is commonly written as
a semi-local operator V SL :

|Ylm Vl (r)Ylm | ,
(58)
V SL =
lm
where Ylm = Ylm (, l ) = Plm (cos())eiml is a spherical harmonic. This approach yields very accurate results [56], but suffers in plane wave calculations
from the high cost of evaluating the matrix elements containing the semilocal potential.
Kleinman and Bylander [57] addressed this remaining problem by separating the semi-local potential in a local l-independent and a new semi-local
part:

V SL = V local (r) +
|Ylm Vl (r)Ylm | ,
(59)
lm
where the semi-local part is then replaced by a new, computationally less

demanding and fully separable non-local term. This term can be chosen arbitrarily, which therefore allows for additional error corrections.
Goedecker et al. [58, 59] nally introduced a new class of analytic separable pseudopotentials, tailor-made for fast integration on the real space grid,
but still norm conserving. The tting procedure was slightly altered by minimizing a penalty function, which takes into account the difference in charge
as well as the eigenvalues between lae and l .
This approach was extended by Lilienfeld et al. [42] who introduced a second penalty function to account for the long range dispersion interaction represented by MP2 calculations. This second penalty function was minimized
149
using one single additional projector, i.e., one additional angular momentum
quantum number, generating one additional, non-local potential.
5
CarParrinello Simulations
The most straightforward way to introduce rst-principle molecular dynamics is to propagate the nuclei classically and solve Schrdingers equation for
every time step on the y, to obtain the new orbitals. This procedure is
called BornOppenheimer molecular dynamics (BOMD) [60]. Compared to
the BornOppenheimer approximation, where the atoms do not move during the wave function optimization, and the approximations inherent in the
electronic structure method, BOMD is exact. Unfortunately, obtaining the
wave functions in every time step results in a prohibitively large effort of
computational resources, which gives an upper limit to the accuracy of the
wave function. On the other hand, poorly converged wave functions lead to
a drift in the total energy [8, 61]. Therefore, energy conservation can be considered as the lower limit to the accuracy of the wave function. In contrast to
BOMD, in CarParrinello molecular dynamics (CPMD) [1] both systems, i.e.,
the electrons and the nuclei, evolve at the same timescale, effectively eliminating the explicit time-dependence of the quantum subsystem. The obvious
advantage of this procedure is that the wave function is calculated only once,
thus saving a lot of computer resources. The classical kinetic energy Eekin of
the electrons was proposed by Car and Parrinello to be:
1
Eekin =
i | i ,
(60)
2
i
where i are the one-electron KohnSham orbitals, i is the time derivate of
i and is an arbitrary parameter, denoted as the ctitious mass of each
electron. To keep the orbitals orthonormal
one must introduce the constraint

E KS [{i }, RN ] = EKS [{i }, RN ] + ij ij (i | j ij ).
N , {i }, { i }] postulated by Car and ParThe Lagrangian LCP LCP [RN , R
rinello [1] nally reads:
N
N , {i }, { i }] =
LCP [R , R
N

1
MI R 2I
2
I=1
1
i | i E KS [{i }, RN ] .
+
2
i
(61)
Note, that the extended KohnSham energy functional E KS is dependent on

the orbitals {i } and RN . The corresponding Newtonian equations of motion are derived from the associated EulerLagrange equations for the nuclear
150
J. Thar et al.
positions like in classical mechanics:

d LCP LCP
=0,
dt R I
RI
and for the orbitals:
d LCP LCP
=0.
dt i | i |
(62)
(63)
This leads to the CarParrinello equations of motion:
EKS
I (t) =
MI R
+
ij
i | j ,
RI
RI
(64)
ij
i (t) =
EKS
+
ij | j .
i |
(65)
Because this Lagrangian does not depend explicitly on time, the (physically
meaningless) total energy HCP :
HCP =
N

1
I=1
MI R 2I +
1
i
i | i + EKS = EIkin + Eekin + EKS
(66)
of the whole system is conserved. To guarantee that sum of the kinetic energy
of the nuclei, EIkin (and the potential energy EKS ) remains nearly constant, the
parameter is chosen small enough that Eekin is several orders of magnitude
smaller than EIkin + EKS . With this choice of , one creates a coupled classical
system of two energy reservoirs. The hot nuclei, and the cold electrons
with an unequal distribution of energy [62]. According to the equipartition theorem, which states that the energy is distributed equally between all
quadratic degrees of freedom in an equilibrated physical system, one would
expect a constant ow of energy from the hot nuclei to the cold electrons during the course of simulation. This would destroy the conservation
of EIkin . Furthermore, with rising temperature the wave function deviates
more and more from a perfectly converged wave function for the given positions of nuclei. In this case, it is commonly said that the electrons leave the
BornOppenheimer surface [8]. Therefore, one must suppress this energy
ow to maintain the adiabatic separation. To do this one can consider adjusting the parameter or the introduction of thermostats for both energy
reservoirs [63], as shown in Sect. 2.2.
5.1
Fictitious Mass
Pastore et al. [62] as well as Tagney [64] pointed out that the adiabatic separation depends crucially on the separation of the spectrum for the ionic
151
vibrations and the ultrahigh-frequency oscillation of the orbitals. A nice review on the coupling between the electronic excitations and the phonons
was written by Chester [65]. Assuming a linear KohnSham Hamiltonian and
equal occupation of all occupied orbitals, the frequency for the electronic
excitation can be approximated as:
ij = (fj (i j )/)1/2 ,
(67)
where fj is the occupation number of the KohnSham orbital j , i is the

eigenvalue of the unoccupied orbital i , and j is the eigenvalue of j . The
lowest electronic excitation frequency min , i.e., the electronic frequency
closest to the frequencies of the nuclei, is therefore determined by the Kohn
Sham gap. Obviously, min can be adjusted by . This gives rise to the idea of
shifting towards zero, giving the best possible adiabatic separation. However, this idea is limited by the fact that the largest possible time step for
the simulation depends on the inverse of the highest electronic frequency
max [8]. Therefore, choosing means compromising between a sufciently
large frequency gap and a reasonable timescale for the simulation. A second
problem arises for systems with a small, or even without any, KohnSham
gap, especially metals. In this case, the adiabatic separation breaks down and
cannot be saved by a nite , because min goes to zero and the frequency
spectra of the ions and the electrons overlap. Here, thermostats for both the
electrons and the ions can be used to maintain the adiabatic separation [66].
Tagney et al. [64, 67] could also show that the deviation of the forces of the
CPMD simulation compared to BOMD is proportional to the ctitious mass
. For purely ionic systems, where the electronic structure of one atom is
not perturbed by the neighboring atoms, this difference reduces to a simple
rescaling of the ionic masses, which can be corrected a priori [67]. For more
delocalized systems, this difference was shown to be smaller, but cannot be
corrected any further.
6
QM/MM
Despite the success of the CarParrinello method, biological systems like proteins or systems where explicit solvent effects are observed are too large to
be described by density functional theory. For such systems, several hybrid
QM/MM schemes have been developed, rooting back to the original idea proposed by Warshel and Levitt [68]. The basic idea of all the hybrid schemes is
that the system is divided into one (chemically active) region described by the
quantum mechanical (QM) method and one surrounding region described by
classical molecular mechanics (MM). The crucial point of the hybrid methods
is the coupling Hamiltonian that describes the interface region between the
QM and the MM part. Referring to classical molecular mechanics, the con-
152
J. Thar et al.
tributions to the coupling Hamiltonian can be divided into bonded and nonbonded interactions. Bonded interactions only arise when the QM/MM partitioning breaks existing bonds. In this case, the related MM atom has to be
replaced by a hydrogen atom or a suitable pseudopotential [6971] to avoid
unsaturated valence orbitals on the related QM atom. Another solution to this
problem is the so-called frozen orbital method [70, 7274], where suitable
orbitals are located on the related MM atom. These orbitals are not evolved
during the simulation, but nevertheless raise the computational cost for plane
wave codes signicantly [75]. Angles and dihedral terms are completely described by the underlying classical force eld [75]. For the discussion of the
non-bonded interaction, we restrain ourselves to hybrid schemes involving
the CPMD method [7578]. Non-bonded interactions in classical molecular
dynamics consist of the van der Waals contribution and the electrostatic interaction. The van der Waals interaction can be expressed by a LennardJones
potential HvdW :
HvdW =

4ij
iMM
jQM
ij
|Ri Rj |
12
ij
|Ri Rj |
6
,
(68)
where QM denotes the set of all atoms in the QM region, MM denotes the
set of atoms of the MM region, and Rk denotes the position of one nucleus
k in the QM region or one atom k in the MM region. The construction of
an appropriate electrostatic interaction is the most challenging point. In general, the electrostatic Hamiltonian Hel , describing the interaction between the
electrons of the QM system with the charges of the MM system, is given by:
Hel =

jMM

qj
dr
(r)
,
|r Rj |
(69)
where qj is the charge of the MM atom at the position Rj and (r) is the charge
density of the QM part at the electron position r.
Therefore, a partially positively charged MM atom represents a trap for the
electron density of the QM part, because the electrostatic interaction goes to
innity if the electron approaches the MM atom. This phenomenon is called
the spill-out effect (Fig. 4) and is attributed to the missing repulsion of the
core electrons on the MM atoms. All schemes rely on modifying the electrostatic potential of the MM atoms in the vicinity of the QM region, i.e., by
altering the short range Coulomb interaction such that Hel remains nite for
|r Rj | 0.
Eichinger et al. [76] replaced the point charges of the MM atoms close to
the QM part by a Gaussian distribution of charges. Thus, their electrostatic
153
Fig. 4 Plot of electron localization function (ELF) for a water dimer as obtained from
a full quantum calculation employing CPMD and a QM/MM calculation employing
GROMACS-CPMD. = 10 is equivalent to a pure 1/r electrostatic potential, = 1 is
equivalent to a modied electrostatic interaction with a nite value for r 0. Reprinted
with permission from [77]. Copyright 2005, American Institute of Physics
Hamiltonian nally reads:

erf(|r Rj |/ )
Hel =
,
qj N dr(r)
|r Rj |
(70)
jMM
where N is the normalization constant for the electronic density and erf is
the error function.
A second approach was undertaken by Laio et al. [75], where Hel was
modied to:

Hel =
qj dr(r)vj (|r Rj |) ,
(71)
jMM
with:
vj (|r Rj |) =
rcjn |r Rj |n
rcjn+1 |r Rj |n+1
(72)
where rcj is the covalent radius of the atom j and n is an empirically determined parameter.
One further proposal was given by Biswas et al. [77], who expanded
the charge qj on the MM atom in terms of hydrogen-like wave functions
154
J. Thar et al.
(r Rj ):
(r Rj ) =
qj 3
1/2
e|r Rj | ,
(73)
where can be freely adjusted. The electrostatic Hamiltonian in this case

reads:

(r)|(r Rj )|2
N dr dr
Hel =
.
(74)
|r r|
jMM
All of these Hamiltonians approach a nite value of Hel for |r Rj | 0 and

show a 1/|r Rj | behavior for long distances. In order to save CPU time, the
explicit evaluation of the Coulomb interaction with every MM atom can be
replaced by a multipole expansion [75] of the electrostatic eld of the QM
region:

3
)

(Rj R
1
(r)
=C
+
dr
D
(75)
3
|r Rj |
|Rj R|
|Rj R|
=1
4

)(R R
3
)
Rq
1 (Rj R
j
+
Q
+O
5
2
|Rj R|
|Rj R|
,=1
Rq
= Hlr (j) + O
|Rj R|
4
,
where r is the geometrical center of the quantum subsystem and therefore the
origin of the multipolar expansion. C is the monopole, D is the dipole, and
Q is the quadrupole. They are given by:

C = dr(r) ,
(76)
D =
Q

=
) ,
dr(r)(r R

)(r R
) |r R|
2 .
dr(r) 3(r R
(77)
(78)
This multipole expansion is only used for the potential seen by atoms outside
an arbitrarily chosen cut-off radius Rc around every quantum atom. All atoms
inside Rc see the exact, non-expanded potential.
155
7
Literature on Biological Systems
In the last decade, CarParrinello molecular dynamics simulations have developed into a powerful tool for the investigation of biomolecules such as
proteins, peptides, enzymes, or even DNA and RNA. We present a brief review
of applications of CPMD in this area.
7.1
Proteins, Peptides, and Amino Acids
One of the strengths of CPMD is the possibility of observing spontaneous
events. This can be utilized for the investigation of systems where proton
transfers or the formation and breakage of hydrogen bonds are expected. Examples of such investigations are: the simulation of the proton transfer in
a polyglycine analog of the gramicidin A ion channel, by Klein and coworkers [79]; the study of the dynamics and energetics of a fully hydrated model
complex of the pSer133 and the corresponding part of the CREP-binding protein (CBP), by Carloni and coworkers [80]; simulations of the deprotonation
of histidine residues in aqueous solution, by Ivanov and Klein [81]; and studies of the role of intramolecular hydrogen bonds for the torsion of arylamide
derivatives, by Doerksen et al. [82].
Further examples of protein CPMD simulations are the examination of the
binding of copper ions to prion proteins by the Rthlisberger group [83, 84],
and a study by Gaigeot and Sprik where they demonstrated the applicability
of CPMD for the computation of infrared absorption spectra in condensed
molecular systems based on a study of aqueous uracil [85].
Retinal derivatives as chromophores in many light-detecting proteins, such
as the visual pigment rhodopsin and the proton pump bacteriorhodopsin,
have been the subject of detailed investigations: First molecular dynamics
studies of the retinylidene chromophore of rhodopsin were performed by
Buda, Bifone, and coworkers [8689]. Rhrig et al. studied the structure and
dynamics of the retinal chromophore of rhodopsin in vacuum, dissolved in
methanol, and in the protein binding pocket [90] and the photoreaction in
rhodopsin [91]. Marx and coworkers investigated the protonated water networks on the extracellular side of bacteriorhodopsin [92, 93].
Liquid N-methylacetamide was studied by a combination of classical and
CarParrinello molecular dynamics by Whiteld et al. [94]. Investigation of
the hypothetical hydroxide and proton migration along the linear water chain
in aquaporins was carried out by Jensen, Rthlisberger, and Rovira [95]. Simulations of the zwitterionic form of the dipeptide glycine-alanine in water
were undertaken by Hugosson [96]. The 2D (, ) free energy surface of
alanine dipeptide in both gas phase and aqueous solution phase were determined [97] by means of the meta-dynamics approach [98, 99].
156
J. Thar et al.
7.2
Metal-Containing Biomolecules: Metallo-enzymes
CarParrinello molecular dynamics simulations are also established as
a widespread tool for the investigation of metal-containing biomolecules such
as metallo-enzymes. Heme models belong to the rst biologically relevant
systems that were studied by CarParrinello simulations. Parrinello, Rovira,
and coworkers gained insight into the binding of O2 , CO, and NO to iron porphyrin [100103]. They extended the investigation of these reactions to the
natural system, the myoglobin protein [104106]. Recently Spiro and coworkers studied the inuence of a distal histidine residue on O2 and CO binding
in myoglobin [107].
Further early applications of the CPMD method on metal-containing systems were the studies on the role of hydrogen bonds for the enzymatic
function of serine protease by De Santis and Carloni [108], and the simulation of model systems for the adduct between the HIV-1 reverse transcriptase and nucleosidic phosphate by Alber and Carloni [109]. HIV-1 protease and its enzymatic reaction are also under intensive study by Piana
et al. [110113].
Molecular dynamics studies of a biomimetic model of galactose oxidase
were performed by Rthlisberger and Carloni [114]. Consecutively, they investigated together with Doclo and Parrinello the catalytic properties of the
galactose oxidase and the synthetic analog of the active site on the basis of
QM/MM CarParrinello calculations [115].
Further examples of such systems for investigation using CarParrinello
simulations are: the enzymatic catalysis of d-alanine oxidation by the
avioenzyme d-amino acid oxidase [116]; the study of the initial and fundamental step in enzymatic GTP hydrolysis, by Cavelli and Carloni [117]; and
the investigation of different protomeric states of methionine aminopeptidase
active site, by Klein et al. [118].
In addition, Magistrato et al. studied the structural and dynamical properties of the dizinc analog of the synthetic diiron protein Due Ferri 1 [119], and
Klein and coworker investigated the Glu4 His2 Zn2 cofactor that is found in the
metallo-protein Duo Ferro 1 [120].
Carloni and coworker carried out QM/MM simulations of the hydrolysis of
cefotaxime by the monozinc -lactamase [121] and of the enzymatic activity
of orotodine monophosphate decarboxylase [122].
More recent applications are: the study of the role that magnesium cations
play in the cleavage reaction of fully hydrated RNA enzymes, by Boero
et al. [123]; simulations to gain insight into the dynamical exibility of the
bridged binuclear structural motif in the active site of arginase, by Ivanov and
Klein [124]; investigation of the initial dinitrogen reduction step in Sellmanntype nitrogenase model complexes [125]; and the study of the binding mode
of anticancer ruthenium(II) arene complexes to double-stranded DNA, by
157
Gossens et al. [126]. Alagano et al. compared the results of quantum mechanics/free energy perturbation (QM/FEP) calculations of the second proton transfer step in triosephosphate isomerase with the results of QM/MM
CarParrinello simulations, in order to estimate the bias of the FEP simulations [127].
Recently, Rovira and coworkers studied the bacillus 1,3-1,4--glucanase [128]. The hydrogen-bonding pattern between phosphinate and phosphonate inhibitor to aspartic protease was investigated by Vidossich and
Carloni [129].
Giannozzi et al. demonstrated, using the example of porphyrine representatives, that their ultrasoft pseudopotential implementation of rst-principle
molecular dynamics is suitable for the investigation of biologically relevant
metallo-organic systems [130].
7.3
DNA and RNA
Up to now there have been only a few applications of the CarParrinello
methods to DNA and RNA. One of the rst applications of the CPMD code in
this context was the calculation of nucleic acids under laboratory-realizable
conditions, performed by Hutter et al. for sodium guanylyl-3 -5 -cytidine
nonahydrate [131]. Molecular dynamics simulations of such systems were
performed by Carloni et al. [132]. They simulated the hydrolysis of cis-platin
as well as the nal step of the binding process to pGpG dinucleotide, as model
for the DNA.
Further studies were concerned with the hydration and dynamics of
arginine-phosphate adducts in proteins, by Frigyes et al. [133]; with the
interaction between DNA and radicals by calculating the reaction mechanisms of the reaction between isolated guanine and an OH radical, by Mundy
et al. [134]; with the QM/MM examination of the second step of the enzymatic
reaction of caspase-3 (namely the hydrolysis of the acyl-enzyme complex), by
Sulpizi et al. [135, 136]; with the investigation of the structure and binding of
cis-platin to DNA oligomers in aqueous solution, by Spiegel et al. [137]; with
the mixed QM/MM study of the DNA inuence on the reactivity of the guanine radical cation by the Parrinello group [138]; and with the simulation of
the OH radical reaction with guanine or thymine in water in order to obtain
a better understanding of the OH radical-induced DNA-base damage, by Car
and coworkers [139].
Recently a fully hydrated polyd(GpCp) DNA strand was investigated by
Gervasio et al. in order to gain a deeper insight into the charge localization
in DNA bers [140]. Spiegel et al. studied the binding of duocarmycins to
DNA [141].
158
J. Thar et al.
7.4
Miscellaneous, Biomimetics
In this section we will briey review some further applications of the CPMD
method on biological relevant molecules. Early examples are: the application of the projector augmented wave (PAW) method of Blchl [142] on the
investigation of penta-O-methyl--d-glycopyranose inversion and penta-Omethyl--d-glycopyranose inversion by Ionescu et al. [143] and the simulation of the anomeric equilibrium of glucose in aqueous solution by Molteni
and Parrinello [144].
Later, Rovira et al. studied the protonation state of dimethylglyoxime ligands and the dynamics of OH H units in a B12 coenzyme biomimetic
system [145]. Schwarz and coworkers could clarify the unusual temperature
effects on the dissociation patterns of valeramide radical cations observed
in mass spectrometric experiments by simulation of neutral and ionized
valeramide [146]. The solvation of 2,2 -dihydroxyl-1,1 -binaphthyl in water
was simulated by Entel, Kreth et al. [147].
Recent applications are: studies of the dynamics and NMR properties of
a vanadateglycylglycine complex in aqueous solution [148]; investigation
of the role of the aqueous solvent for the acid-catalyzed glycosidic bond
formation, by Stubbs and Marx [149]; investigations of the protonationinduced diastereoisomerism in nicotine, which may have decisive inuence
on the dynamical behavior of the molecules, by Hammond et al. [150]; and
simulations of the shape and exibility of arylamide polymers by classical
molecular dynamics and CPMD calculations, which resulted in the development of new torsional parameters for classical MD simulations of these
compounds [151].
7.5
Excited States MD
CarParrinello molecular dynamics in excited states has become possible in
recent years: The Parrinello group introduced a KohnSham-like formalism
for the treatment of excited singlet states [152, 153]. The Rthlisberger group
implemented this approach in their mixed QM/MM CarParrinello code [83].
Bittner and Kosov used the random phase approximation for the description of electronically excited systems with CarParrinello molecular dynamics [154]. Doltsinis and Marx applied a method for an efcient treatment of
electronically non-adiabatic processes with CPMD by using a surface hopping
scheme for the coupling of the S1 restricted open-shell KohnSham excited
state to the S0 ground state [155].
Hutters implementation of time-dependent linear response density functional theory (LR-TDDFT) to plane wave basis set methods allowed the
simulation of photochemical reactions and uorescence spectra [156, 157].
159
Recently, Rthlisberger and coworkers presented implementations of a linear response and time propagation scheme to molecular dynamics simulations [158].
A few examples of applications of these methods to systems of biological interest are: the investigation of rhodopsin by Molteni et al. [159]; hybrid
QM/MM simulations of the photoreaction in rhodopsin, by the Rthlisberger
group [91]; the calculation of optical properties of molecules in solution,
by Sulpizi et al. [160, 161]; simulations of guanine [162, 163] and methylated
guanine [164], by Langer et al.; the study of photoactive proteins, by the Rthlisberger group [165]; and the investigation of the photochemical activation
scheme of dinitrogen by dinuclear RuII and FeII complexes [166].
8
Case Study: Biomimetic Nitrogen Fixation
In this section we briey review an example of a CarParrinello molecular dynamics simulation, namely the initial dinitrogen reduction step in
Sellmann-type nitrogenase model complexes [125]. These complexes are of
special interest because they are supposed to enable access to articial nitrogen xation at ambient conditions. An introduction to biological nitrogen
xation mediated by the enzyme nitrogenase can be found in recent literature, see [167176].
Sellmann-type complexes are based on the so-called open-side FeMo cofactor (FeMoco) systems, and have only recently been substantiated by some
DFT calculations [172, 173]. Sellmann assumed that FeMoco (a transition
metalsulfur cluster, which is the core of the active center of the enzyme nitrogenase) opens so as to bind nitrogen at the iron centers. In fact, the synthesis of [-N2 {Ru(PiPr3 )(N2 Me2 S2 )}2 ] [N2 Me2 S2 2 = 1,2-ethanediamineN,N -dimethyl-N,N -bis(2-benzenethiolate)(2-)], which is up to now the only
known example of a stable dinuclear metalsulfur N2 complex, was accomplished in 2001 [177, 178]. In further experiments, Sellmann and coworkers
tried to establish a reduction pathway to the corresponding diazene complex
[-N2 H2 {Ru(PiPr3 )(N2 Me2 S2 )}2 ], which had been already fully characterized [179].
Instead of the reduction to diazene, various side reactions were observed
during the experiments. This can be explained by the fact that the rst endothermic reduction step is already difcult to achieve [180]. However, the
results indicate that these model complexes are not suitable for the desired
behavior to reduce nitrogen, and not that Sellmann-type complexes are in
principle not able to succeed in the reduction of nitrogen.
CarParrinello molecular dynamics simulations were performed in order
to clarify whether the reduction of nitrogen with Sellmann-type complexes
is possible in principle or not. The advantage of CarParrinello simulations
160
J. Thar et al.
is the detailed time-resolved picture, which is needed for the discussed processes.
Two different complexes were investigated: a small [Ru S2 N2 ]2 L model
complex (1N2 ) and a larger [Fe NH S4 ]2 L complex (2N2 ) depicted in Fig. 5.
These systems were chosen because they are similar to the rst experimentally observed dinitrogen metalsulfur complex [178, 179, 181] and because
an investigation of complexes with chelate ligands of different rigidity allows
elucidation of the general properties of such complexes. The starting points
of our investigation are the doubly protonated complexes 1N2 2+ and 2N2 2+ .
A twofold reduction of these systems leads to complex A1 on the corresponding neutral electronic energy surface, on which the subsequent reaction takes
place and nally stops at the diimine B (see Fig. 6). A2 is obtained if the preparation of B during the course of the reaction is stopped.
The presented products A1 and A2 of the twofold reduction are highenergy structures on the neutral potential energy surface. Hence, it can be
Fig. 5 Sellmann-type Ru S2 N2 model complex 1N2 (left) and the Fe NH S4 complex (2N2 )
(right)
Fig. 6 Schematic drawing of initial reduction pathway
161
expected that spontaneous events take place and it can be observed how A1
and A2 behave under the effects of temperature, forced bending, and forced
decrease of the nitrogen proton distances. This can indicate a potential mechanism and a possible reaction coordinate for this system. It is not possible to
reach extended statistics because of the very expensive CPMD calculations.
A detailed description of the methodology is presented in the original publication [125].
Constrained and free simulations of complex 1N2 and 2N2 were run at
different temperatures. In one experiment both hydrogen MNN angles were
forced to decrease with time in complex 2N2 . This was done because it had
been shown before that the bending of the nitrogen moiety is an important
step for the reaction [166]. The results of these experiments are depicted in
Fig. 7. Because the symmetry of the complex is preserved during the reaction, only one curve is shown. Temperature regulation was used to prevent the
chelate ligands from decomposing. The initiation of the reaction can be recognized at 600 fs. It involves a sequence of complex processes. The increase
of the KohnSham energy and temperature, which can be observed in Fig. 8,
points to a small barrier at an initial time step of 150 fs.
This barrier is accompanied by:
1. A small elongation (5 pm) of the N N distance, see bold black line in the
left panel of Fig. 7,
2. An increase (150 pm) of the hydrogennitrogen atom distance, see black
line in the left panel of Fig. 7,
3. An elongation (15 pm) of the metalnitrogen distance (dashed line in the
left panel of Fig. 7) and an increase of the FeSH angle (20 degrees), see
right panel of Fig. 7.
Fig. 7 Results of the simulation of the large Fe complex 2N2 , where the MNN angles are forced to decrease in time. Left panel: Nitrogennitrogen distance (bold black
line), nitrogenhydrogen distances (black line), sulfurhydrogen distance (gray line), and
N M distances (dashed line). Right panel: MNN angles (dashed line) and MSH angles
(gray line). All data are plotted against the time step in fs. Distances are in pm and angles
in degrees. Picture reproduced from data of [125]
162
J. Thar et al.
Fig. 8 Forcing both metalnitrogennitrogen angles to decrease: KohnSham energy

(gray line) and temperature (black line) plotted against time steps. Picture reproduced
from data of [125]
On the basis of these results it can be assumed that in the starting conguration A1 the bending of the N2 moiety is structurally hindered. At a time step
of about 300 fs a further small increase of the KohnSham energy of about
0.025 a.u. appears. It can be explained by a slight barrier for the initiation of
the bending of the FeNN angle (see right panel of Fig. 7). The results of these
constraint simulations indicate that bending of the M1 N1 N2 and M2 N2 N1 angles is necessary for a successful reaction. It seems apparent that this bending
of the N2 moiety is prohibited unless the hydrogen atoms leave the vicinity
of the nitrogen atoms. At this point it is important to note that the transfer of the hydrogen is a concerted process. If only one angle is constrained,
a simultaneous reaction takes place nevertheless.
In different experiments, free simulations without any constraints for both
Sellmann models 1N2 and 2N2 were carried out. The results of simulations
at 50 K, where the starting point is a pre-bent structure (A2 of Fig. 6) of the
small complex 1N2 , are presented in Fig. 9. The reaction of the N2 moiety with
one hydrogen atom (H2 N2) begins after about 450 fs. However, in this case
the transfer of the hydrogen atoms is not a simultaneous process.
It can be understood from the left upper panel of Fig. 9 that at about 450 fs
the vibration of the dinitrogen bond is enhanced and the NN bond elongates.
The vibrations with small amplitudes around 121.5 pm belong to the original
triple bond of N2 , whereas the vibrations with a larger amplitude around
an average NN distance of 129.2 pm can be assigned to a NN double bond.
The bottom left panel of Fig. 9 shows that the vibrational motion of both
NH bonds is anticorrelated. The increase of one NH distance (dashed curve)
at the beginning enables the other hydrogen atom to approach its nitrogen
atom. Then, an oscillating motion of the hydrogen atoms can be observed
163
Fig. 9 Free simulation of the bent small Sellmann-type complex 1N2 : Left upper panel:
Nitrogennitrogen distance and sulfurhydrogen distances. Right upper panel: RuNN
and RuSH angles. Bottom left panel: Nitrogenhydrogen distances. Bottom right panel:
NitrogenRu distances. All data are plotted against the time step in fs. Distances are in
pm and angles in degrees. Picture reproduced from data of [125]
until, at approximately 450 fs, the distance between one hydrogen and the corresponding nitrogen atom drops down rapidly to about 100 pm. This shows
that an angle as well as a distance constraint is not an appropriate reaction
coordinate, because the reaction cannot be described by only one simple reaction coordinate. Only 15 fs later the second hydrogen atom is transferred.
After 450 fs, the distances between the S atoms and the corresponding H
atoms, as well as the amplitude of the vibrational motions, increase abruptly.
This also affects the curve of the RuSH angles; they decrease to below 90.
The bottom right panel of Fig. 9 shows that the N Ru distances do not
change much upon reaction but that they vibrate with smaller amplitude after
the hydrogen transfer. Again, the KohnSham energy and the temperature are
very good indicators for the course of the reaction (Fig. 10). The proton transfer at 450 fs is connected with a notable drop of the KohnSham energy and
a heating-up of the system.
In further experiments [125], the behavior of the Fe complex 2 N2 at
a temperature of 100 K without any constraints was investigated. The results
of these simulations in Fig. 11 demonstrate successful events. In this case it
takes about 1050 fs until the rst proton transfer proceeds, and the delay of
the second proton transfer is almost 100 fs. The bottom left in Fig. 11 indicates that the hydrogen atoms move away from the corresponding nitrogen
164
J. Thar et al.
Fig. 10 KohnSham energy (gray line) and temperature (black line) plotted against time
steps. Picture reproduced from data of [125]
atom before the bending of the FeNN angles takes place. The vibrational motion of these angles is smooth until proton transfer begins. The S H bonds
are not elongated until the beginning of the proton transfer. In the rst 800 fs
the N Fe distance increases by 20 pm only at the side of the complex where
the reactions starts (see dashed curve in the upper left panel of Fig. 11).
From the results of the simulations that led to successful events, it can be
concluded [125] that the bending motion of the MSH angle plays a crucial role
in determining whether the reaction takes place. This motion induces an increase in the NH distance, which in turn induces the start of the bending
motion of the MNN angle. Even for a pre-bent structure the hydrogen nitrogen distance has to grow in order to enable the MNN angle bending mode.
A comparison of the results of the free simulations with the constraint simulation shows that we observe for the latter a simultaneous proton transfer
whereas the former is a stepwise process. The constraints lead to a somewhat
unphysical behavior as is evident from the right panel of Fig. 7, where the
curve of the FeNN angles shows no vibrational motion.
The above case study of the initial proton electron transfer step in
Sellmann-type nitrogenase model complexes has shown that CarParrinello
molecular dynamics is a powerful tool for the investigation of complex reactions. On the basis of these simulations, a three-step reaction mechanism
was suggested for both investigated complexes: The NH distance has to rst
increase. Hereby a bending of the MNN angles is allowed, which leads to an
activated diazenoid N2 species. Vibrations of the MSH angle effect vibrations
of the NH distances, which are in opposite phase to simultaneous vibrations
of the MNN angles. This nally leads to the observed proton transfer, because
165
Fig. 11 Free simulation results of the large Fe complex 2N2 at 100 K. Left upper panel:
Nitrogennitrogen distance and nitrogeniron distances. Right upper panel: FeNN angles.
Bottom left panel: Nitrogenhydrogen distances and sulfurhydrogen distances. Bottom
right panel: FeSH angles. All data are plotted against the time step in fs. Distances are
displayed in pm and angles in degrees. Picture reproduced from data of [125]
the nitrogen atoms can get close to the hydrogen atoms. Since both complexes
with chemically different chelate ligands exhibit an equivalent behavior, it can
be concluded that the observed features of the reactions are universal [125].
Particularly important is the fact that the reduction of nitrogen is a spontaneous event, if a two-electron reduction of a Sellmann-type complex leads to
a trans-diprotonated equlibrium structure. On the basis of our simulations we
can also estimate that the time needed for the twofold proton transfer is in the
order of 5001000 fs in this case.
During the simulations a few side reactions are also observed. One of these
side reaction consists of a partial dissociation of the chelate ligands of the
metal atom. In another side reaction, a hydrogen transfer to the metal center is observed. These side reactions could be related [125] to side products
observed in experiment.
9
Conclusion
We have presented an introduction to the CarParrinello simulation technique with a special emphasis on the applications to biomolecular systems.
166
J. Thar et al.
Although many new developments and improvements are still in progress,

this method is now amongst the standard tools of theoretical chemists. This
is mostly due to the excellent program packages that were provided by the
groups of Parrinello, Car, and Hutter, for example. A large step towards the
appropriate description of biological systems in the CarParrinello framework is due to the developments of the Rthlisberger group. These developments include QM/MM coupling schemes and methodological developments
related to this topic. Pioneering work on the eld of CarParrinello simulations of biomolecules was also undertaken by Carloni and coworkers.
Biological systems are complicated in several aspects. For instance, they can
demand a quantum chemical treatment, they are large and extended systems,
and a dynamical description is often necessary. Therefore the continuing
developments contributed by many different groups are important and will
aid the investigation of challenging topics such as those present in biology. We concentrated in this review only on the CarParrinello simulation
technique, but we want to stress here that many important contributions
also come from scientists working with similar techniques (not treated in
this review), like for instance the tight-binding density functional theory
method (TB-DFT) [182] or BornOppenheimer molecular dynamics simulations (BOMD) [142, 183186].
Acknowledgements We gratefully acknowledge the nancial support of the DFG priority
program SPP 1191 Ionic Liquids and the ERA program, which allows productive collaboration under the project A Modular Approach to Multi-responsive Surfactant/Peptide
(SP) and Surfactant/Peptide/Nanoparticle (SPN) Hybrid Materials. We would furthermore like to acknowledge the nancial support from the collaborative research center
SFB 624 Templates at the University of Bonn. Fruitful discussion with Jrg Hutter, Ursula Rthlisberger, and Markus Reiher are also gratefully acknowledged. JT would like to
thank the Fonds der Chemischen Industrie for a Chemiefonds-Stipendium.
References
1.
2.
3.
4.
5.
6.
7.
8.
Car R, Parrinello M (1985) Phys Rev Lett 55:2471

Kirchner B, Reiher M (2002) J Am Chem Soc 124:6206
Kirchner B, Hutter J (2002) Chem Phys Lett 364:497
Kirchner B, Hutter J (2004) J Chem Phys 121:5133
Reiher M, Sellmann D, Hess BA (2001) Theor Chim Acta 106:379
Kirchner B (2005) J Chem Phys 123:204116
Remler DK, Madden PA (1990) Mol Phys 70:921
Marx D, Hutter J (2000) Ab-initio molecular dynamics: Theory and implementation. Modern methods and algorithms in quantum chemistry, Forschungszentrum
Juelich, NIC Series 1:301
9. Tuckerman ME, Martyna GJ (2000) J Phys Chem B 104:159
10. Tuckerman ME (2002) J Phys Condens Matter 14:1297
11. Alder BJ, Wainwright TE (1957) J Chem Phys 27:1208
167
12. Ermakova E, Solca J, Huber H, Marx D (1995) Chem Phys Lett 246:204
13. Kirchner B, Ermakova E, Steinebrunner G, Dyson AJ, Huber H (1998) Mol Phys
94:257
14. Huber H, Dyson A, Kirchner B (1999) Chem Soc Rev 28:121
15. Kirchner B, Ermakova E, Solca J, Huber H (1998) Chem Eur J 4:379
16. Allen MP, Tildesley DJ (1987) Computer simulations of liquids. Clarendon, Oxford
17. Frenkel D, Smit B (2002) Understanding molecular simulations. Academic, San
Diego CA
18. Edelman A, Arias TA, Smith ST (1998) SIAM J Matrix Anal Appl 20:303
19. Andersen HC, Allen MP, Bellemans A, Board J, Clarke JHR, Ferrario M, Haile JM,
Nose S, Opheusden JV, Ryckaert JP (1984) Technical report, Rapport dactivite scientique du CECAM
20. Nose S (1984) J Chem Phys 81:511
21. Nose S (1984) Mol Phys 52:255
22. Hoover WG (1985) Phys Rev A 31:1695
24. Martyna GJ, Klein ML, Tuckerman ME (1992) J Chem Phys 97:2635
25. Cho K, Joannopoulos JD (1992) Phys Rev A 45:7089
26. Hohenberg P, Kohn W (1964) Phys Rev B 136:864
27. Kohn W, Sham LJ (1965) Phys Rev 140:1133
28. Roos BO, Widmark PO (eds) (2000) European summerschool in quantum chemistry.
Lund University
29. Vosko SH, Wilk L, Nusair M (1980) J Phys 58:1200
31. Filatov M, Thiel W (1998) Phys Rev A 57:189
32. Ahlrichs R, Penco R, Scoles G (1977) Chem Phys 19:119
33. Andersson Y, Langreth DC, Lundqvist BI (1996) Phys Rev Lett 76:102
34. Kamiya M, Tsuneda T, Hirao K (2002) J Chem Phys 117:6010
35. Grimme S (2004) J Comput Chem 25:1463
36. Wu Q, Yang W (2002) J Chem Phys 116:515
37. Zimmerli U, Parrinello M, Koumoutsakos P (2004) J Chem Phys 120:2693
38. Seifert G, Kohler T, Tenne R (2002) J Phys Chem B 106:2497
39. Johnson ER, Becke AD (2005) J Chem Phys 123:024101
40. Lundqvist BI, Andersson Y, Shao H, Chan S, Langreth DC (1995) Int J Quantum
Chem 56:247
41. Rushton PP, Tozer DJ, Clark SJ (2002) Phys Rev B 65:23503
42. von Lilienfeld OA, Tavernelli I, Rthlisberger U (2004) Phys Rev Lett 93:153004
43. Szabo A, Ostlund NS (1996) Modern quantum chemistry. Dover, New York, USA
44. Jensen F (2002) Introduction to computational chemistry. Wiley, Chichester
45. Kirchner B, Huber H, Steinebrunner G, Dreizler H, Grabow J-U, Merke I (1997)
Z Naturforsch 52a:297
46. Feibelman PJ (1987) Phys Rev B 35:2626
47. Jaffe JE, Hess AC (1996) J Chem Phys 105:10983
48. Kudin KN, Scuseria GE (2000) Phys Rev B 61:16440
49. Saunders VR, Dovesi R, Roetti C, Orlando R, Zicovich-Wilson CM, Harrison NM,
Doll K, Civalleri B, Bush I, DArco Ph, Llunell M (2003) CRYSTAL2003 users manual.
University of Torino
50. Lippert G, Hutter J, Parrinello M (1997) Mol Phys 92:477
51. Lippert G, Hutter J, Parrinello M (1999) Theor Chem Acc 103:124
52. http://cp2k.berlios.de/
168
J. Thar et al.
53.
54.
55.
56.
57.
58.
59.
60.
Fuchs M, Schefer M (1999) Comput Phys Commun 119:67

Hamann DR, Schlter M, Chiang C (1979) Phys Rev Lett 43:1494
Martin RM (2004) Electronic Structure. Cornell University Press, Cambridge
Bachelet GB, Hamann DR, Schlter M (1982) Phys Rev B 26:4199
Kleinman L, Bylander DM (1982) Phys Rev Lett 48:1425
Goedecker S, Teter M, Hutter J (1996) Phys Rev B 54:1703
Hartwigsen C, Goedecker S, Hutter J (1998) Phys Rev B 58:3641
Payne MC, Joannopoulos JD, Allan DC, Teter MP, Vanderbilt DH (1986) Phys Rev Lett
56:2656
Pulay P, Fogorasi G (2004) Phys Rev Lett 386:272
Pastore G, Smargiassi E, Buda F (1991) Phys Rev A 44:6334
Galli G, Parrinello M (1992) Phys Rev Lett 69:3547
Tagney P (2006) J Chem Phys 124:044111
Chester GV (1961) Adv Phys 10:357
Bloechl PE, Parrinello M (1992) Phys Rev B 45:9413
Tagney P, Scandolo S (2002) J Chem Phys 116:14
Warshel A, Levitt M (1976) J Mol Biol 103:227
Antes I, Thiel W (1999) J Phys Chem A 103:9290
Zhang YK, Lee TS, Yang WT (1999) J Chem Phys 110:46
Reuter N, Dejaegere A, Maigret B, Karplus M (2000) J Phys Chem A 104:1720
Monard G, Loos M, Thery V, Baka K (1996) Int J Quantum Chem 58:153
Kairys V, Jensen JH (2000) J Phys Chem A 104:6656
Murphy RB, Philipp DM, Friesner RB (2000) J Comput Chem 21:1442
Laio A, VandeVondele J, Rthlisberger U (2002) J Chem Phys 116:6941
Eichinger M, Tavan P, Hutter J, Parrinello M (1999) J Chem Phys 110:10452
Biswas PK, Gogonea V (2005) J Chem Phys 123:164114
Woo TK, Cavallo L, Ziegler T (1998) Theor Chem Acc 100:307
Sagnella DE, Laasonen K, Klein ML (1196) Biophys J 71:1172
Dal M Peraro, Alber F, Carloni P (2001) Eur Biophys J 30:75
Ivanov I, Klein ML (2002) J Am Chem Soc 124:13380
Doerksen RJ, Chen B, Klein ML (2003) Chem Phys Lett 380:150
Colombo MC, Guidoni L, Laio A, Magistrato A, Maurer P, Piana S, Rhrig U,
Spiegel K, Sulpizi M, Vande J Vondele, Zumstein M, Rthlisberger U (2002) Chimia
56:13
Carloni P, Rthlisberger U, Parrinello M (2002) Acc Chem Res 35:455
Gaigeot MP, Sprik M (2003) J Phys Chem B 107:10344
Buda F, de Groot HJM, Bifone A (1996) Phys Rev Lett 77:4474
Bifone A, de Groot HJM, Buda F (1997) Pure and Appl Chem 69:2105
La Penna G, Buda F, Bifone A, de Groot HJM (1998) Chem Phys Lett 294:447
Bifone A, de Groot HJM, Buda F (1997) J Phys Chem B 101:2954
Rhrig UF, Guidoni L, Rthlisberger U (2005) Comput Phys Commun 6:1836
Rhrig UF, Guidoni L, Laio A, Frank I, Rthlisberger U (2004) J Am Chem Soc
126:15328
Rousseau R, Kleinschmidt V, Schmitt UW, Marx D (2004) Phys Chem Chem Phys
6:1848
Rousseau R, Kleinschmidt V, Schmitt UW, Marx D (2004) Angew Chem Int Ed Engl
43:4804
Whiteld TW, Martyna GJ, Allison S, Bates SP, Jason C (2005) Chem Phys Lett
414:210
Jensen MO, Rthlisberger U, Rovira C (2005) Biophys J 89:1744
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
94.
95.

96.
97.
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.
133.
134.
135.
136.
137.
138.
169
Hugosson HW, Laio A, Maurer P, Rthlisberger U (2006) J Comput Chem 27:672

Ensing B, De Vivo M, Liu Z, Moore P, Klein ML (2006) Acc Chem Res 39:73
Huber T, Torda AE, van Gunsteren WF (1994) J Comput-Aided Mol Design 8:695
Iannuzzi M, Laio A, Parrinello M (2003) Phys Rev Lett 90:238302
Rovira C, Ballone P, Parrinello M (1997) Chem Phys Lett 271:247
Rovira C, Parrinello M (1998) Int J Quantum Chem 70:387
Rovira C, Kunc K, Hutter J, Ballone P, Parrinello M (1998) Int J Quantum Chem 69:31
Rovira C, Parrinello M (1999) Chem Eur J 5:250
Rovira C, Parrinello M (2000) Int J Quantum Chem 80:1172
Rovira C, Schulze B, Eichinger M, Evanseck JD, Parrinello M (2001) J Biol Chem
81:435
Rovira C (2003) J Phys Condens Matter 15:1809
De Angelis F, Jarzecki AA, Car R, Spiro TG (2005) J Phys Chem B 109:3065
De Santis L, Carloni P (1999) Proteins 37:611
Alber F, Carloni P (2000) Protein Sci 9:2535
Piana S, Carloni P (2000) Proteins 39:26
Piana S, Sebastiani D, Carloni P, Parrinello M (2001) J Am Chem Soc 123:8730
Piana S, Carloni P, Parrinello M (2002) J Mol Biol 319:567
Piana S, Bucher D, Carloni P, Rthlisberger U (2004) J Phys Chem B 108:11139
Rthlisberger U, Carloni P (1999) Int J Quantum Chem 73:209
Rthlisberger U, Carloni P, Doclo K, Parrinello M (2000) J Biol Inorg Chem 5:236
Tilocca A, Gamba A, Vanoni MA, Fois E (2002) Biochemistry 41:14111
Cavalli A, Carloni P (2002) J Am Chem Soc 2002:3763
Klein CDP, Schiffmann R, Folkers G, Piana S, Rthlisberger U (2003) J Biol Chem
278:47862
Magistrato A, DeGrado WF, Laio A, Rthlisberger U, Vande Vondele J, Klein ML
(2003) J Phys Chem B 107:4182
Papoian GA, DeGrado WF, Klein ML (2003) J Am Chem Soc 125:560
Dal Peraro M, Llarrull LL, Rthlisberger U, Vila AJ, Carloni P (2004) J Am Chem Soc
126:12661
Raugei S, Cascella M, Carloni P (2004) J Am Chem Soc 126:15730
Boero M, Tateno M, Tenkura K, Oshiyama A (2005) J Chem Theory Comput 1:925
Ivanov I, Klein ML (2005) J Am Chem Soc 127:4010
Kirchner B, Reiher M, Hille A, Hutter J, Hess BA. (2005) Chem Eur J 11:574
Gossens C, Tavernelli I, Rthlisberger U (2005) Chimia 59:81
Alagona G, Campanile S, Ghio C, Molin D (2005) J Molec Struct, THEOCHEM
729:131
Biarnes X, Nieto J, Planas A, Rovira C (2006) J Biol Chem 281:1432
Vidossich P, Carloni P (2006) J Phys Chem B 110:1437
Giannozzi P, De Angelis F, Car R (2004) J Phys Chem 120:5903
Hutter J, Carloni P, Parrinello M (1996) J Am Chem Soc 118:8710
Carloni P, Sprik M, Andreoni W (2000) J Phys Chem B 104:823
Frigyes D, Alber F, Pomgor S, Carloni P (2001) J Molec Struct, THEOCHEM 574:39
Mundy CJ, Colvin ME, Quong AA (2002) J Phys Chem B 106:10063
Sulpizi M, Folkers G, Rthlisberger U, Carloni P, Scapozza L (2002) Quant Struct Act
Relat 21:173
Sulpizi M, Laio A, Vande Vondele J, Cattaneo A, Rthlisberger U, Carloni P (2003)
Proteins 52:212
Spiegel K, Rthlisberger U, Carloni P (2004) J Phys Chem B 108:2699
Gervasio FL, Laio A, Iannuzi M, Parrinello M (2004) Chem Eur J 10:4846
170
J. Thar et al.
139.
140.
141.
142.
143.
Wu Y, Mundy CJ, Colvin ME, Car R (2004) J Phys Chem A 108:2922

Gervasio FL, Laio A, Parrinello M, Boero M (2005) Phys Rev Lett 94:158103
Spiegel K, Rthlisberger U, Carloni P (2006) J Phys Chem B 110:3647
Blchl PE (1994) Phys Rev B 50:17953
Ionescu AR, Berces A, Zgierski MZ, Whiteld DM, Nukada T (2005) J Phys Chem A
109:8096
Molteni C, Parrinello M (1998) J Am Chem Soc 120:2168
Rovira C, Kunc K, Parrinello M (2002) Inorg Chem 41:4810
Semialjac M, Schrder D, Schwarz H (2003) Chem Eur J 9:4396
Entel P, Kadau K, Adeagbo WA, Sugihara M, Rollmann G, Zayak AT, Kreth M (2004)
Lecture Notes Phys 642:177
Bhl M (2005) Inorg Chem 44:6277
Stubbs JM, Marx D (2005) Chem Eur J 11:2651
Hammond PS, Wu Y, Harris R, Minehardt TJ, Car R, Schmitt JD (2005) J ComputAided Mol Des 19:1
Pophristic V, Veparala S, Ivanov I, Liu Z, Klein ML, De Grado WF (2006) J Phys Chem
B 110:3517
Frank I, Hutter J, Marx D, Parrinello M (1998) J Chem Phys 108:4060
Molteni C, Frank I, Parrinello M (2001) Comput Mater Sci 20:311
Bittner ER, Kosov DS (1999) J Chem Phys 110:6645
Doltsinis NL, Marx D (2002) Phys Rev Lett 88:166402
Hutter J (2003) J Chem Phys 118:3928
Odelius M, Laikov D, Hutter J (2003) J Molec Struct, THEOCHEM, 630:163
Tavernelli I, Rhrig UF, Rthlisberger U (2005) Mol Phys 103:963
Molteni C, Frank I, Parrinello M (1999) J Am Chem Soc 121:12177
Sulpizi M, Carloni P, Hutter J, Rthlisberger U (2003) Phys Chem Chem Phys 5:4798
Sulpizi M, Rhrig UF, Hutter J, Rthlisberger U (2005) Int J Quantum Chem 101:671
Langer H, Doltsinis NL (2003) J Chem Phys 118:5400
Langer H, Doltsinis NL, Marx D (2005) Comput Phys Commun 6:1734
Langer H, Doltsinis NL (2004) Phys Chem Chem Phys 6:2742
Moret ME, Tapavicza E, Guidoni L, Rhrig UF, Sulpizi M, Tavernelli I, Rthlisberger U (2005) Chimia 59:493
Reiher M, Kirchner B, Hutter J, Sellmann D, Hess BA (2004) Chem Eur J 10:4443
Einsle O, Tezcan FA, Andrade SLA, Schmid B, Yoshida M, Howard JB, Rees DC (2002)
Science 297:1696
Leigh GJ (ed) (2002) Nitrogen xation at the millenium. Elsevier, Amsterdam
Hinnemann B, Nrskov JK (2004) J Am Chem Soc 126:3920
Hinnemann B, Nrskov JK (2003) J Am Chem Soc 125:1466
Dance I (2003) Chem Commun 3:324
Schimpl J, Petrilli HM, Blchl PE (2003) J Am Chem Soc 124:15772
Huniar U, Ahlrichs R, Coucouvanis D (2004) J Am Chem Soc 126:2588
Yandulov DV, Schrock RR (2003) Science 301:76
Sellmann D, Utz J, Blum N, Heinemann FW (1999) Coord Chem Rev 190192:607
Sellmann D, Frsattel A, Sutter J (2000) Coord Chem Rev 200202:541
Reiher M, Hess BA (2004) Adv Inorg Chem 56:55
Sellmann D, Hille A, Heinemann FW, Moll M, Rsler A, Sutter J, Brehm G, Reiher M,
Hess BA, Schneider S (2003) Inorg Chim Acta 348:194
Sellmann D, Hille A, Rsler A, Heinemann F, Moll M, Brehm G, Schneider S, Reiher M, Hess BA, Bauer W (2004) Chem Eur J 10:819
Reiher M, Hess BA (2002) Chem Eur J 8:5332
144.
145.
146.
147.
148.
149.
150.
151.
152.
153.
154.
155.
156.
157.
158.
159.
160.
161.
162.
163.
164.
165.
166.
167.
168.
169.
170.
171.
172.
173.
174.
175.
176.
177.
178.
179.
180.
171
181. Sellmann D, Hautsch B, Rsler A, Heinemann FW (2001) Angew Chem Int Ed Engl
40:1505
182. Elstner M, Porezag D, Jungnickel G, Elsner J, Haugk M, Frauenheim T, Suhai S,
Seifert G (1998) Phys Rev B 58:7260
183. Payne MC, Teter MP, Allan DC, Arias TA, Joannopoulos JD (1992) Rev Mod Phys
64:1045
184. Bockstedte M, Kley A, Neugebauer J, Schefer M (1997) Comput Phys Commun
107:187
185. NWChem: developed and distributed by Pacic Northwest National Labaratory, USA
186. Kresse G, Furthmller J (1996) Phys Rev B 54:169
Top Curr Chem (2007) 268: 173290

DOI 10.1007/128_2006_084
Published online: 22 November 2006

Hans Martin Senn (u) Walter Thiel (u)
Max-Planck-Institut fr Kohlenforschung, Kaiser-Wilhelm-Platz 1,
45470 Mlheim an der Ruhr, Germany
senn@mpi-muelheim.mpg.de, thiel@mpi-muehlheim.mpg.de
1
Overview and Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
175
2
2.1
2.2
2.2.1
2.2.2
2.3
2.3.1
2.3.2
2.3.3
2.3.4
2.4
2.4.1
2.4.2
2.5
2.5.1
2.5.2
2.5.3
2.5.4
2.5.5
The QM/MM Method . . . . . . . . . . . . . . . . . . . . . . . . . . .

Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
QM/MM Energy Expressions . . . . . . . . . . . . . . . . . . . . . . .
Subtractive Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Additive Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Electrostatic Interaction Between Inner and Outer Subsystems . . . .
Mechanical Embedding . . . . . . . . . . . . . . . . . . . . . . . . . .
Electrostatic Embedding . . . . . . . . . . . . . . . . . . . . . . . . . .
Polarized Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . .
Long-Range Electrostatic QMMM Interactions . . . . . . . . . . . . .
Other Non-bonded and Bonded Interactions Between the Subsystems
QMMM van der Waals Interactions . . . . . . . . . . . . . . . . . . .
QMMM Bonded Interactions . . . . . . . . . . . . . . . . . . . . . . .
Covalent Bonds Across the QMMM Boundary . . . . . . . . . . . . .
Overview of Boundary Schemes . . . . . . . . . . . . . . . . . . . . . .
Link Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Boundary Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Frozen Localized Orbitals . . . . . . . . . . . . . . . . . . . . . . . . .
Boundary Schemes: Summary . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
178
178
179
179
181
182
182
183
184
186
187
187
188
189
189
191
197
199
201
3
3.1
3.2
3.3
3.3.1
3.3.2
Choice of QM and MM Models, QM/MM Implementations .

Choice of QM Method . . . . . . . . . . . . . . . . . . . . .
Choice of MM Method . . . . . . . . . . . . . . . . . . . . .
QM/MM Implementations . . . . . . . . . . . . . . . . . . .
Program Architecture and QM/MM Packages . . . . . . . .
QM/MM-Related Approaches . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
202
202
204
205
205
207
4
4.1
4.2
4.2.1
4.2.2
4.2.3
4.3
4.3.1
4.3.2
4.3.3
4.3.4
Optimization and Simulation Techniques Used with QM/MM

General Comments . . . . . . . . . . . . . . . . . . . . . . . .
Stationary Points and Reaction Paths . . . . . . . . . . . . . .
General Optimization Techniques for Large Systems . . . . .
Optimization Techniques Specic to QM/MM . . . . . . . . .
Reaction-Path Techniques . . . . . . . . . . . . . . . . . . . .
Molecular Dynamics and Simulation Techniques . . . . . . .
QM/MM Molecular-Dynamics and Monte Carlo Simulations .
QM/MM Free-Energy Perturbation . . . . . . . . . . . . . . .
Thermodynamic Integration . . . . . . . . . . . . . . . . . .
Transition-Path Sampling . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
208
208
209
210
211
213
213
214
215
217
218
174
H.M. Senn W. Thiel
4.3.5
4.3.6
4.3.7
Metadynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Adiabatic Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
QM/MM Reaction-Path Potential . . . . . . . . . . . . . . . . . . . . . . .
219
219
219
Practical Aspects of Biomolecular Reaction Modelling . . . . . . . . . . .
220
Interpreting the Results: Understanding Enzyme Catalysis . . . . . . . .
224
Survey of Biomolecular QM/MM Studies . . . . . . . . . . . . . . . . . .
227
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
250
Abstract Thirty years after the seminal contribution by Warshel and Levitt, we review
the state of the art of combined quantum-mechanics/molecular-mechanics (QM/MM)
methods, with a focus on biomolecular systems. We provide a detailed overview of the
methodology of QM/MM calculations and their use within optimization and simulation schemes. A tabular survey of recent applications, mostly to enzymatic reactions,
is given.
Keywords QM/MM Combined quantum mechanics/molecular mechanics
Optimization Molecular dynamics Molecular simulations Free-energy methods
Enzymatic mechanisms
Abbreviations
ADMP
Atom-centred density-matrix propagation
BFGS
BroydenFletcherGoldfarbShanno (Hessean update algorithm
in minimizations)
CASSCF
Complete active space self-consistent eld
CCSD
Coupled-cluster theory including single and double excitations
COSMO
Conductor-like screening model
CP-MD
CarParrinello molecular dynamics
DFT
Density-functional theory
DO
Drude oscillator
DTSS
Differential transition-state stabilization
EC
Enzyme class
ECP
Effective core potential
EFP
Effective fragment potential
EGP
Effective group potential
ELMO
Extremely localized molecular orbital
ESP
Electrostatic potential
EVB
Empirical valence bond
FEP
Free-energy perturbation
FQ
Fluctuating charge
GHO
Generalized hybrid orbital
GSBP
Generalized solvent boundary potential
HDLC
Hybrid delocalized coordinates
HF
HartreeFock
IMOMM Integrated molecular orbital/molecular mechanics
KIE
Kinetic isotope effect

L-BFGS
LBHB
LSCF
MC
MC-VEEP
MD
MM
MECP
MEP
MP2
NAC
NEB
OECP
ONIOM
PBC
PES
PPD
P-RFO
QCP
QTCP
QM
QM/MM
SCC-DFTB
RFO
SCF
SLBO
SMD
VEP
TDDFT
TDHF
TI
TPS
TS
US
VTST
ZPE
175
Limited-memory BroydenFletcherGoldfarbShanno algorithm

Low-barrier hydrogen bond
Local self-consistent eld
Monte Carlo
Multicentred valence-electron effective potential
Molecular dynamics
Molecular mechanics
Minimum-energy crossing point
Minimum-energy path
Second-order MllerPlesset perturbation theory
Near-attack conguration
Nudged elastic band
Optimized effective core potentials
Our N-layered integrated molecular orbital/molecular mechanics
Periodic boundary conditions
Potential-energy surface
Polarized point dipole
Partial rational-function optimizer
Quantum capping potentials
Quantum-mechanical thermodynamic-cycle perturbation
Quantum mechanics
Combined quantum mechanics/molecular mechanics
Self-consistent-charge density-functional tight-binding
Rational-function optimizer
Self-consistent eld
Strictly localized bond orbital
Steered molecular dynamics
Variational electrostatic projection
Time-dependent density-functional theory
Time-dependent HartreeFock
Thermodynamic integration
Transition-path sampling
Transition state
Umbrella sampling
Variational transition-state theory
Zero-point energy
1
Overview and Scope
Combined quantum-mechanics/molecular-mechanics (QM/MM) approaches
have become the method of choice for the modelling of reactions in biomolecular systems. On the one hand, the size and conformational complexity of
biopolymers, in particular proteins and nucleic acids, call for highly efcient
methods capable of treating up to several 100 000 atoms and allowing for extensive sampling or simulations over time scales of hundreds of nanoseconds.
Molecular-mechanics (MM) force elds, based on classical empirical poten-
176
H.M. Senn W. Thiel
tials, have been proven to provide an effective means for simulating complex
biomolecules (see the contribution by K. Schulten and co-workers in this
volume). On the other hand, the description of chemical reactions (i.e., bondforming and bond-breaking) and other processes that involve changes in the
electronic structure, such as charge transfer or electronic excitation, require
quantum-mechanical (QM) methods. However, their high computational demands still restrict their applicability to systems of several tens up to a few
hundred atoms.
A natural solution to this dilemma is to use a QM method for the chemically active region (e.g., substrates and cofactors in an enzymatic reaction)
and combine it with an MM treatment for the surroundings (e.g., the full
protein and solvent). The resulting schemes are commonly referred to as
combined or hybrid QM/MM methods. They enable the modelling of reactive
biomolecular systems at reasonable computational cost while providing the
necessary accuracy.
The seminal contribution in the eld is due to Warshel and Levitt [1],
who presented in 1976, exactly thirty years ago, a method that took into account essentially all aspects of the QM/MM approach and applied it to an
enzymatic reaction. Based on an earlier formulation [2], which had been
developed to treat conjugated hydrocarbons by the combination of a semiempirical QM method for the -electrons with classical MM terms for the
-framework, their method was characterized by a most remarkable combination of features. The energy expression included the usual MM terms;
a semi-empirical QM Hamiltonian that accounted for the polarization of the
QM density by the MM point charges as well as by induced dipoles placed
on all MM atoms of the protein and by the permanent dipoles of the water
molecules; the interaction of the point charges, induced dipoles, and permanent dipoles with each other; and classical bonded and van der Waals
QMMM coupling terms. Within certain approximations, the elaborate description of the electrostatic interactions was treated self-consistently. Covalent bonds across the QMMM boundary were saturated by single hybrid
orbitals placed on the frontier MM atom. Structure optimizations were possible owing to the availability of analytical derivatives with respect to the
nuclear positions.
Ten years after this pioneering effort, Singh and Kollman [3] took a major step forward by combining an ab initio QM method (HartreeFock) with
a force eld. While a few others had done this before, they were the rst
to report coupled ab initio QM/MM structure optimization. They used link
atoms (see Sect. 2.1) to cap the covalent bonds across the QMMM boundary. Polarization effects were only included as an a posteriori correction at
xed geometry; however, they allowed for mutual polarization of the QM and
MM regions. The contribution by Field, Bash, and Karplus in 1990 [4] described the coupling of a semi-empirical (AM1 or MNDO) QM method with
the CHARMM force eld in considerable detail. Their formulation again used
177
link atoms and accounted for the polarization of the QM density by the MM
point charges. The paper carefully evaluated the accuracy and effectiveness of
the QM/MM treatment against ab initio and experimental data.
Over the last 10 years, numerous reviews [537] have documented the
development of the QM/MM approach as well as its application to biomolecular systems. The use of the QM/MM method as an explicit-solvent approach
to model organic reactions in solution (QM solute in MM solvent, calculation of solvation free energies by a Monte Carlo free-energy perturbation
technique) was thoroughly reviewed by Gao [5, 6] in 1996, who considered
various methodological issues, in particular the treatment of polarization,
in detail. Cunningham and Bash [7] described the development and calibration of semi-empirical QM and MM van der Waals parameters for QM/MM
simulations of an enzymatic reaction. Several articles in the Encyclopedia of
computational chemistry [1013] and an ACS symposium series volume [8]
give a comprehensive account of the state of the art in 1998. A complete
and succinct overview of the QM/MM method was provided by Sherwood in
2000 [21]. Lin and Truhlar [37] have very recently given an astute report of
current methodological aspects.
A number of articles have combined, with varying accents, an overview
of QM/MM and other computational methods for biomolecular systems
with application surveys from this area [14, 15, 1720, 2226, 2836]. Among
these, we highlight the contributions by Field (1999, 2002) [19, 24], Mulholland (2001, 2003) [23, 28], and Friesner (2005) [36].
The current review provides a detailed overview of the QM/MM method
and its use within optimization and simulation schemes, and surveys recent
applications. Throughout, we keep the focus on biomolecular systems, leaving aside QM/MM treatments geared towards inorganic, organometallic, or
solid-state systems, nor covering applications from any of these areas.
We start with an introduction into the formalism of the QM/MM method.
The use of different computational models for different regions of space immediately raises issues about how to dene the regions and how to treat their
mutual interaction, and we will give an overview of different approaches that
address these issues. Although the choice of QM and MM methods being
combined is in principle arbitrary, we comment on some special aspects and
list commonly used combinations, some of which are available in commercial programs. Similarly, we highlight issues pertaining to the use of QM/MM
geometry optimization, molecular dynamics (MD), and free-energy simulation techniques. Moreover, practical aspects concerning the setup of QM/MM
calculations on biomolecular systems are discussed, as well as different ways
of analysing and interpreting the results from such calculations. We conclude
with a tabular survey of biomolecular QM/MM studies that have appeared
since 2000.
178
H.M. Senn W. Thiel
2
The QM/MM Method
2.1
Terminology
A very general sketch of the division of the system into QM and MM parts
is shown in Fig. 1. The entire system (S) is partitioned into the inner region
(I) to be treated quantum-mechanically and the outer region (O) described
by a force eld. Inner and outer regions are therefore also frequently referred to as QM and MM regions, respectively. Each atom of the entire system
is assigned to either of the subsystems. Because the two regions generally
(strongly) interact, it is not possible to write the total energy of the entire system simply as the sum of the energies of the subsystems. As detailed below,
coupling terms have to be considered, and it will be necessary to take precautions at the boundary between the subsystems, especially if it cuts through
covalent bonds. The term boundary region is used here rather loosely to designate the region where the standard QM and MM procedures are modied or
augmented in any way. Depending on the type of QM/MM scheme employed,
the boundary region may contain additional atoms (link atoms) used to cap
the QM subsystem that are not part of the entire system, or it may consist of
atoms with special features that appear both in the QM and the MM calculation. Note that the assignment of each atom to either subsystem is no longer
unique in this latter case.
Anticipating the discussion of boundary schemes (Sect. 2.5), we introduce
here some labelling conventions, illustrated in Fig. 2, that apply to covalent
bonds across the QMMM boundary. The QM and MM atoms directly connected are designated Q1 and M1 , respectively, and are sometimes referred
to as boundary, frontier, or junction atoms. The rst shell of MM atoms (i.e.,
those directly bonded to M1 ) is labelled M2 . The next shell, separated from M1
by two bonds, is labelled M3 ; and so on, following the molecular graph outwards from M1 . The same naming procedure applies to the QM side; atoms Q2
Fig. 1 Partitioning of the entire system S into inner (I) and outer (O) subsystems
179
Fig. 2 Labelling of atoms at the boundary between QM and MM regions
are one bond away from Q1 , Q3 are two bonds away, etc. If a link-atom scheme
is applied, the dangling bond of Q1 is saturated by the link atom L.
As a caveat, we emphasize that the classication of QM/MM schemes and
the denition of terms such as link, capping, boundary, junction, or frontier atom are not unique and their usage varies between authors. Moreover,
a given QM/MM method can incorporate aspects from different schemes,
making its classication ambiguous.
If not stated otherwise, for the remainder of this review the classical
potential-energy function (the force eld) of the MM region is assumed to
include bonded terms (bond stretching, angle bending, torsions, out-of-plane
deformations or improper torsions), a LennardJones-type van der Waals
term, and the Coulomb interaction between rigid point charges. A simple,
prototypical MM energy expression of this type, sometimes called a class I
force eld, reads:

2

2

kb d d0 +
k 0 +
k 1 + cos n +
EMM =
bonds
angles

AB
non-bonded
pairs AB
AB
rAB
12
dihedrals
AB
rAB
1 qA qB
+
40 rAB
(1)
where d, , and designate bond distances, angles, and torsions, respectively;

d0 and 0 are the corresponding equilibrium values; and n and are the torsional multiplicity and phase, respectively. The bonded force constants are kb ,
k , and k . rAB is the non-bonded distance, and AB and AB are the van der
Waals parameters between atoms A and B. qA , qB are atomic partial charges
and 0 is the vacuum permittivity. We refer to the literature [3842] for pertinent details and extensions or variations of this general form.
2.2
QM/MM Energy Expressions
2.2.1
Subtractive Schemes
In subtractive QM/MM schemes, three calculations are performed: (i) an MM
calculation on the entire system, S; (ii) a QM calculation on the inner sub-
180
H.M. Senn W. Thiel
system, I; and (iii) an MM calculation on the inner subsystem. The QM/MM

energy of the entire system is then obtained by summing (i) and (ii) and
subtracting (iii) to correct for double counting:
EQM/MM (S) = EMM (S) + EQM (I + L) EMM (I + L) .
(2)
Here, as in the following, the subscript indicates the level of the calculation
while the system on which it is performed is given in parentheses. As written,
Eq. 2 holds for a link-atom scheme, the calculations on the inner subsystem
being performed not on the bare I but on I capped with link atoms, I + L.
For a scheme with special MM boundary atoms (rather than link atoms) that
carry certain features appearing also in the calculations on the inner subsystem, L is understood to refer to these atoms. If no covalent bond is cut by the
QMMM boundary, I + L reduces simply to I.
Conceptually, the subtractive QM/MM scheme can be seen as an MM
approach in which a certain region has been cut out and replaced by a higherlevel treatment. Its main advantage is simplicity. No explicit coupling terms
are needed, avoiding any modication of the standard QM and MM procedures, and the subtraction implicitly corrects for any artefacts caused by the
link atoms, provided that the MM force terms referring to the link atoms reproduce the QM potential reasonably well. These features make a subtractive
scheme fairly straightforward to implement.
On the downside, a subtractive scheme also requires a complete set of MM
parameters for the inner subsystem. These may be difcult or cumbersome to
obtain. Moreover, and more severely, the coupling between the subsystems is
treated entirely at the MM level. This is particularly problematic for the electrostatic interaction, which is typically described by xed atomic charges at
the MM level. Hence, in a subtractive scheme the electrostatic interaction between the subsystems is treated within a simple point-charge model, which is
often a rather severe approximation: First, the charge distribution in the inner subsystem can change (e.g., during a reaction), which cannot be reected
by rigid point charges. Second, the QM calculation does not incorporate the
charges in the outer region, that is, the QM charge density is not polarized by
the environment. A subtractive scheme is therefore not suitable if the electron
density is signicantly inuenced by electrostatic interactions with the outer
region.
Within the classication of QMMM coupling schemes (Sect. 2.3), a strictly
subtractive QMMM method necessarily implies mechanical embedding (i.e.,
the QM density is not polarized by the environment). However, mixed formulations are conceivable that are in principle subtractive, but treat the
electrostatic interaction separately, allowing for a more elaborate coupling
scheme.
As an example for a subtractive QM/MM scheme, we mention the IMOMM
method (integrated molecular orbital/molecular mechanics) by Morokuma
and co-workers [43]. It has subsequently been extended to enable the combi-
181
nation of two QM methods (IMOMO [44]) and further generalized to N layers

(typically, N = 3), each of which can be treated at an arbitrary QM or MM
level (ONIOM, our N-layered integrated molecular orbital and molecular mechanics [4547]). Recent improvements of the ONIOM approach [4850] that
enable the inclusion of MM charges into the QM Hamiltonian (electrostatic
embedding, see Sect. 2.3.2) take it beyond a purely subtractive scheme.
2.2.2
Additive Schemes
The basic energy expression for an additive QM/MM scheme is:
EQM/MM (S) = EMM (O) + EQM (I + L) + EQM-MM (I, O) .
(3)
In contrast to the subtractive scheme of Eq. 2, the MM calculation is now performed on the outer subsystem only. In addition, there appears an explicit
coupling term, EQM-MM (I, O), which collects the interaction terms between
the two subsystems. The capped inner subsystem, I + L, is treated at the QM
level as before.
Assuming a link-atom-based scheme with mechanical embedding, it
is possible to derive the additive energy expression from the subtractive
one [51]. Using the fact that the MM energy is unambiguously decomposable
into contributions depending on exclusive sets of atoms, we can split the MM
terms of Eq. 2 as:
EMM (S) = EMM (O) + EMM (I) + EMM (I, O) ,
(4)
EMM (I + L) = EMM (I) + EMM (L) + EMM (I, L) .
(5)
Substituting these into Eq. 2, the MM contribution from the inner subsystem,
EMM (I), cancels, and we obtain the full QM/MM energy as:
EQM/MM (S) = EMM (O) + EQM (I + L) + EMM (I, O)
(6)
[EMM (L) + EMM (I, L)] .

For a mechanical-embedding scheme with link atoms, EMM (I, O) can be identied with the QMMM coupling term EQM-MM (I, O) of Eq. 3 as it contains for
this case all the interactions between the subsystems.
The subtractive terms in Eq. 6 are referred to as the link-atom correction:
Elink = [EMM (L) + EMM (I, L)] .
(7)
While the link atoms are not part of the entire (i.e., physical or real) system
being modelled, their interaction with one another and the atoms of the inner
subsystem is contained in the term EQM (I + L). A correction is thus formally
justied. However, Elink is in practice often omitted, which can be motivated
by pragmatic arguments: (i) The accuracy and validity of a correction at the
182
H.M. Senn W. Thiel
MM level for QM interactions is questionable. It is therefore unclear if the correction would actually improve the overall model. (ii) The interaction among
the link atoms, EMM (L) (which for a typical force eld consists of electrostatic
and van der Waals contributions), is expected to be small. (iii) EMM (I, L),
the interaction between the link atoms and the inner subsystem, is not small.
However, in many common link-atom schemes, this term is a constant or depends only weakly on the structure because the position of the link atom, in
particular the distance Q1 L, is constrained.
The working equation adopted in the majority of QM/MM schemes is thus
Eq. 3. The exact form of the QMMM coupling term EQM-MM denes a particular QM/MM method. In accordance with the interactions considered in
the force eld, it includes electrostatic, van der Waals, and bonded interactions between QM and MM atoms:
vdW
b
EQM-MM (I, O) = Eel
QM-MM + EQM-MM + EQM-MM .
(8)
The following sections deal in more detail with the individual contributions
to EQM-MM . The electrostatic coupling term (Sect. 2.3.2) arguably has the
largest impact and is also the most technically involved one. The van der
Waals interaction and the bonded terms are discussed in Sect. 2.4. Section 2.5
presents various ways that have been devised to treat covalent bonds across
the QMMM boundary.
2.3
Electrostatic Interaction Between Inner and Outer Subsystems
The electrostatic coupling between the QM charge density and the charge
model used in the MM region can be handled at different levels of sophistication, characterized essentially by the extent of mutual polarization and
classied [51, 52] accordingly as mechanical embedding (model A), electrostatic embedding (model B), and polarized embedding (models C and D).
2.3.1
Mechanical Embedding
In a mechanical-embedding scheme, the QMMM electrostatic interaction is
treated on the same footing as the MMMM electrostatics, that is, at the MM
level. The charge model of the MM method used typically rigid atomic point
charges, but other approaches, e.g., bond dipoles, are also possible is simply applied to the QM region as well. This is conceptually straightforward and
computationally efcient.
However, there are major disadvantages and limitations: (i) The charges in
the outer region do not interact with the QM density, which is thus not directly inuenced by the electrostatic environment. Hence, the QM density is
183
not polarized. (ii) As the charge distribution in the QM region changes, for
instance during a reaction, the charge model needs to be updated. However,
this is problematic because it leads to discontinuities in the potential-energy
surface. (iii) The derivation of, e.g., MM point charges for the inner region
is often not trivial. The procedures vary widely between force elds and can
require considerable effort. Moreover, they may not be general but geared
towards the class of compounds for which the force eld was developed. In
this case, their applicability to the inner region, which is often treated at
the QM level exactly because it is outside the chemical domain of the force
eld, is questionable. (iv) The MM charge model is dependent on, and interlinked with, the other force-eld parameters. Together with these, it is mainly
intended to yield a balanced description of conformational or structural preferences, rather than to reproduce accurately the true charge distribution. It is
therefore not justiable to use charges for the QM part derived from a model
different to the one applied in the force eld.
2.3.2
Electrostatic Embedding
The major shortcomings of mechanical embedding can be eliminated by
performing the QM calculation in presence of the MM charge model. For
instance, by incorporating the MM point charges as one-electron terms in
the QM Hamiltonian, which is thus augmented by an additional term (using
atomic units):
electrons
MO
el
QM-MM
H
=
qM Z
qM
+
,
|ri RM |
|R RM |
(9)
I+L MO
where qM are the MM point charges and Z the nuclear charge of the QM
atoms; the index i runs over all electrons, M over the point charges, and over
the QM nuclei.
In such a scheme (referred to as electrostatic or electronic embedding)
the electronic structure of the inner region can adapt to changes in the
charge distribution of the environment and is automatically polarized by
it. No charge model needs to be derived for the inner region. The QM
MM electrostatic interaction is treated at the QM level, which obviously
provides a more advanced and more accurate description than a mechanicalembedding scheme. Naturally, electrostatic embedding also increases the
computational cost, especially for the calculation of the electrostatic force due
to the QM density acting on the (many) MM point charges.
Special care is required at the QMMM boundary, where the MM charges
are placed in immediate proximity to the QM electron density, which can lead
to overpolarization. This problem is especially pronounced when the boundary runs through a covalent bond, and is therefore discussed later in Sect. 2.5.
184
H.M. Senn W. Thiel
Note that because the QMMM electrostatic interaction term, Eel

QM-MM , is
now calculated by the QM code, it is sometimes considered a contribution to
EQM and included therein. However, in the present review we will strictly adhere to the energy partitioning given by Eqs. 3 and 8, that is, EQM is the pure
QM energy, while Eel
QM-MM is part of EQM-MM .
There remains the issue that the MM charge model is not necessarily wellsuited to interact with the QM electron density. As mentioned above, the
electrostatic MM parameters are not primarily designed to provide a faithful
representation of the real charge distribution. It is, in principle, not legitimate
to stitch a true charge distribution, as provided by the QM calculation, into
the carefully parameterized MM charge model. Nevertheless, this has become
common practice, and experience shows that it generally yields reasonable
results, at least for the combination of a QM density with one of the widely
used biomolecular force elds. The obvious appeal of this approach is that
the MM atomic partial charges are readily available from the force eld and
their inclusion in the QM Hamiltonian is efcient. Electrostatic embedding is
the most popular embedding scheme in use today, certainly for biomolecular
applications.
2.3.3
Polarized Embedding
As electrostatic embedding accounts for the interaction of the polarizable QM
density with rigid MM charges, the next logical step is to introduce a exible MM charge model that is polarized by the QM charge distribution. One
can further divide these polarized-embedding schemes into approaches that
apply a polarizable-charge model in the MM region, which is polarized by
the QM electric eld but does not itself act on the QM density (model C),
and fully self-consistent formulations that include the dipoles into the QM
Hamiltonian and therefore allow for mutual polarization (model D).
There exist various models used to treat polarization in the MM part,
which can broadly be classied as follows. We provide here only a very brief
overview and refer to [40, 53, 54] for more detailed treatments and pertaining
references:
Polarized point dipoles (PPD): Polarizabilities are assigned to atoms (or
other distinguished sites, e.g., centres of mass), which interact with the
electric eld at that site, thus inducing point dipoles. The sources of the
electric eld are the point charges, the other induced dipoles, and possibly the QM charge distribution. Since the dipoles interact with each
other, an iterative procedure must be applied to generate a self-consistent
polarization. Alternatives are a full-matrix direct solution or extended Lagrangean schemes with the dipoles as ctitious degrees of freedom. The
free parameters of the model are the (atomic) polarizabilities. Sometimes,
185
a dipoledipole interaction model is applied that damps the interaction

between close-lying dipoles. In principle, PPD methods can be extended
to include higher-order multipoles.
Drude oscillators (DO): A mobile point charge of opposite sign is connected to a charge site by a harmonic spring, thus forming a dipole. These
dipoles then interact with the local electric eld, as outlined above. DO
models are usually implemented within an extended Lagrangean scheme.
The t parameters of the model are the magnitude of the mobile charge
and the spring constant. In the context of solid-state simulations, the DO
approach is often referred to as shell model; it is also known as charge-onspring model.
Fluctuating charges (FQ): Based on the principle of electronegativity
equalization, the atomic partial charges are optimized with respect to the
total electrostatic energy. The practical advantage of the FQ as well as the
DO approaches is that the description of polarizability is achieved without explicitly introducing additional (i.e., dipoledipole) interactions. In
FQ methods, even the number of chargecharge interactions is unchanged
from the non-polarizable case.
Although the very rst QM/MM approaches were in fact polarized-embedding
schemes [1, 3], they have remained scarce. The main obstacle is the lack of
well-established polarizable biomolecular MM force elds. A variety of polarizable solvent models is available, most prominently for the simulation
of liquid water (recent examples include [5561]). The development of polarizable protein force elds, however, is very much a work in progress. We
mention contributions from the developers of the CHARMM [6265] and
AMBER [6668] force elds, Friesner and co-workers [6972], and Ren and
Ponder [73], and refer to [40] for a recent review.
Apart from the availability of polarizable force elds, there are also some
computational and technical issues that need to be considered in polarizedembedding schemes. Model D requires the coupling of the self-consistency
cycles for the QM charge density and the MM polarizable-charge model,
which increases the computational effort and may create convergence problems. Additional complications also arise at the boundary between the subsystems, where the QM density and the MM charge model interact in close
proximity, see Sect. 2.5.2.
As far as the use of polarized-embedding schemes in QM/MM calculations is concerned, there is only limited experience. A PPD model D at
the semi-empirical QM level was applied to a fairly large biomolecular system [74], and tests on small organic molecules with a PPD model C at the
semi-empirical, HartreeFock, and DFT QM levels have been reported [51,
52]. Otherwise, polarized-embedding QM/MM calculations were restricted
to explicit-solvation (in particular, hydration) studies, where the solute is
treated at the QM level and the solvent by a polarizable force eld [18, 7577].
186
H.M. Senn W. Thiel
2.3.4
Long-Range Electrostatic QMMM Interactions
An accurate description of the electrostatic forces on the QM subsystem due
to the environment is essential for a reliable modelling of the structure and
function of biomolecules. Including all the electrostatic interactions explicitly
is computationally challenging, and QM/MM electrostatic cutoffs are problematic because of the long-range nature of the Coulomb interaction. Several
recent studies have shown that cutoffs can introduce signicant artefacts [78
80]. While the reliable and efcient treatment of the electrostatic interactions
is a well-established topic in the area of classical MD simulations, it has only
recently found increased attention in the context of QM/MM methods; we
highlight here some recent developments:
Ewald methods: For simulations done under periodic boundary conditions (PBC), Ewald methods provide an accurate treatment of long-range
electrostatics. A linear-scaling particle-mesh Ewald scheme for QM/MM
simulations has recently been presented by York and co-workers [78].
Although accurate, the PBC/Ewald approach generally suffers from high
computational demands because of the large number of explicit solvent
molecules that need to be included. The biomolecule of interest is immersed in a box of explicit solvent, whose size must be chosen large
enough to minimize artefacts caused by the articially imposed periodicity. The large number of degrees of freedom further increases the simulation cost because it prolongs the required equilibration times. Different
approaches have therefore been proposed that include the electrostatic interactions explicitly only from an active region around the QM part.
Charge scaling [81]: Karplus and co-workers have proposed a procedure
for QM/MM free-energy simulations where only a limited number of explicit solvent molecules is considered and the charges are scaled to mimic
the shielding effect of the solvent. The energies obtained are then corrected using continuum-electrostatics (linearized PoissonBoltzmann or
nite-difference Poisson) calculations.
Variational electrostatic projection (VEP) [82, 83]: In the popular stochastic-boundary method [8487] for MD simulations, the spherical active
zone (treated by standard Hamiltonian dynamics) is surrounded by
a stochastic buffer shell governed by Langevin dynamics; the solute
atoms in this buffer are positionally restrained, and the solvent molecules
are subject to a boundary potential. The remaining parts of the system are
held xed and form the external environment.
Proposed by Gregersen and York, the VEP method aims at reducing the
cost of calculating the electrostatic forces due to the external environment on the atoms in the two inner regions. It uses Gaussians to expand the charge distribution of the environment on a discretized spher-
187
ical surface enclosing the moving part of the system. The procedure
is related to the continuum-solvent models of the COSMO (conductorlike screening model) type. An improved variant of the VEP method is
the VEP-RVM (reverse variational mapping) method. A charge-scaling
implementation of the VEP and VEP-RVM approaches has also been
presented [83].
Generalized solvent boundary potential (GSBP) [80]: The spherical solvent boundary potential (SSBP) [88, 89] includes a small number of
solvent molecules explicitly while the surrounding ones are represented by
a (spherical) effective boundary potential. The GSBP method generalized
this scheme to boundaries of arbitrary shape [90]. All atoms in the inner
region are treated by explicit dynamics, while the xed environment is included in terms of a solvent-shielded static eld and a PoissonBoltzmann
reaction eld.
The GSBP approach has recently been extended and adapted for QM/MM
simulations by Cui and co-workers [80]. They successfully validated and
applied it in pKa calculations [79], proton-transfer processes [91], and
other biomolecular simulations [92]. They stress in particular the need to
treat the QMMM and MMMM electrostatics in a balanced manner to
prevent artefacts.
2.4
Other Non-bonded and Bonded Interactions Between the Subsystems
In addition to the electrostatic interaction discussed in the previous section, there are also van der Waals and bonded contributions to the QMMM
coupling term, Eq. 8. Their treatment is considerably simpler than for the
electrostatic coupling as they are handled purely at the MM level, irrespective
of the class (subtractive or additive) of QM/MM scheme.
2.4.1
QMMM van der Waals Interactions
The van der Waals interaction is typically described by a LennardJones
potential, as shown in Eq. 1; alternative functional forms, e.g., with an exponential repulsive term, are sometimes used instead. However, the present
discussion is unaffected by the exact form of this term. The rst issue that
arises here is the same as discussed above in the context of mechanical
embedding: the availability and suitability of MM parameters for the inner
region. It is not uncommon that certain QM atoms are not covered by any
of the atom types and assignment rules of the force eld. Secondly, even
if suitable van der Waals parameters exist for a given conguration, QM
atoms can change their character, e.g., during a reaction. This then raises the
question of whether one should switch the parameter set, say, from a re-
188
H.M. Senn W. Thiel
actant description to a product description; and if so, at which point

along the reaction path. Finally, there is the overarching problem that, strictly
speaking, MM parameters are not separable and transferable, but only valid
within the parameterization for which they were derived, that is, for MMMM
interactions.
In practice, however, all these complications are very much alleviated by
the short-range nature of the van der Waals interaction. While every atom of
the inner region is involved in van der Waals interactions with all the atoms
of the outer region, only the closest-lying ones contribute signicantly. Unoptimized van der Waals parameters therefore affect only the QM atoms close to
MM atoms, that is, those at the boundary. If one is concerned that this might
inuence the result, one solution is to move the QMMM boundary further
away from the incriminated QM atom. Similar considerations apply to the
ambiguity of choosing a xed set of van der Waals parameters, where switching between parameter sets would introduce additional problems rather than
increase the quality of the model. The effect can simply be checked by comparing the results obtained with different parameters (e.g., derived for the
reactants and the products).
Friesner and co-workers [93] in their QM/MM scheme have re-optimized
the QM van der Waals parameters against structures and bonding energies of
hydrogen-bonded pairs of small models for amino acids. The van der Waals
radii thus obtained are 510% larger than those of the force eld used (OPLSAA); the van der Waals well depths were left unchanged. The increased van
der Waals repulsion compensates for the too-strong QMMM electrostatic
attraction caused by the MM point charges overpolarizing the QM density.
Recently, a set of van der Waals parameters optimized for B3LYP/AMBER was
presented by a different group [94]. Cui and co-workers [95] showed that
thermodynamic quantities in the condensed phase (e.g., free energies), calculated from QM/MM simulations, are rather insensitive towards the QMMM
van der Waals parameters. As expected, they do, however, inuence the detailed structure around the QM region.
With respect to the QMMM van der Waals coupling, subtractive and additive schemes are identical. In an additive scheme, the simple rule is that only
pairs consisting of one atom from the inner and one atom from the outer subsystem are considered in EvdW
QM-MM . This yields exactly the same van der Waals
terms as a subtractive scheme, where the QMQM van der Waals pairs are
subtracted out.
2.4.2
QMMM Bonded Interactions
The formal reservations raised above against using MM parameters to describe QMMM interactions also apply of course in the case of the bonded
(bond stretching, angle bending, torsional, etc.) interactions. And again, the
189
solution is entirely pragmatic. One usually retains the standard MM parameter set and complements it as necessary with additional bonded terms not
covered by the default assignment rules of the force eld. As the bonded interactions are by denition strictly localized to the boundary, one can validate
the results by extending the inner region, shifting the boundary and, hence,
potentially less reliable interaction terms away from the chemically active
region.
For the bonded QMMM interaction, there is an operational difference between subtractive and additive schemes with respect to the treatment of link
atoms, which leads to different terms being included in the nal energy expression. A subtractive scheme removes by construction the QMQM bonded
interactions (i.e., those involving atoms from the capped inner region I + L
only) and retains all mixed QMMM bonded terms. It thus implicitly corrects for the link atoms. For instance (see Fig. 2), the stretching terms Q2 Q1
and Q1 L and the bending term Q2 Q1 L are removed, while Q1 M1 and
Q2 Q1 M1 are included.
By contrast, an additive scheme requires an explicit set of rules that govern
which bonded contributions are to be included in EbQM-MM , thereby avoiding double-counting of (possibly implicit) interactions. The general rule is
that every bonded term that depends on atoms from both the inner and the
outer subsystem is included (note that the link atoms do not belong to either region). However, terms like Q2 Q1 M1 or Q3 Q2 Q1 M1 are excluded
to prevent double-counting. For example, when the angle Q2 Q1 M1 is distorted, the link atom placed along Q1 M1 needs to move as well. This leads
to restoring forces on Q1 and M1 , as discussed in Sect. 2.5.2. Hence, the angular distortion is implicitly accounted for, and the bending term Q2 Q1 M1
is omitted. Commonly, only angle terms of the form M1 Q1 M1 and torsion
terms where at least one of the two central atoms is QM are retained [96].
However, the exact rules by which bonded interactions between QM and MM
atoms are included depend on the details of the boundary scheme employed.
2.5
Covalent Bonds Across the QMMM Boundary
2.5.1
Overview of Boundary Schemes
This section is concerned with the various approaches that have been devised to treat covalent bonds cut by the QMMM boundary. The simplest
solution is of course to circumvent the problem by dening the subsystems
such that the boundary does not pass through a covalent bond. This is trivially fullled for explicit-solvation studies, where the solute and maybe the
rst solvation shell are described at the QM level, surrounded by MM solvent molecules. It is sometimes possible also for biomolecular systems; for
190
H.M. Senn W. Thiel
instance, if the reactants of an enzymatic reaction (substrates, cofactors)

are not covalently bound to the enzyme and no protein residue is directly
involved in the chemical transformation. In many cases, however, it is unavoidable that the QMMM boundary cuts through a covalent bond. Such
situations arise when one needs to include certain protein residues in the inner region or would like to treat chemically inactive parts of the substrate or
cofactor at the MM level to reduce the computational cost. One then has to
deal mainly with three issues (see Sect. 2.1 for atom labelling conventions):
(i) The dangling bond of the QM atom Q1 must be capped; simply assuming a truncated QM region (i.e., treating the bond as being homolytically or
heterolytically cleaved) would be entirely unrealistic. (ii) For electrostatic or
polarized embedding, one has to prevent overpolarization of the QM density, in particular, by the partial charge on M1 . This is problematic especially
when link atoms are used. (iii) The bonded MM terms involving atoms from
both subsystems have to be selected such that double-counting of interactions
is avoided (see Sect. 2.4.2). Overall, the boundary scheme should provide
a balanced description of the QMMM interaction at the border between the
two subsystems.
The different boundary schemes can be categorized into three groups, examined in more detail in the following sections:
Link-atom schemes introduce an additional atomic centre L (usually a hydrogen atom) that is not part of the entire, real system. It is covalently
bound to Q1 and saturates its free valency.
In boundary-atom schemes, the MM atom M1 is replaced by a special
Janus boundary atom that appears in both the QM and the MM calculation. On the QM side, it mimics the cut bond and possibly also the
electronic character of the MM moiety attached to Q1 . In the MM calculation, it behaves as a normal MM atom.
Localized-orbital schemes place hybrid orbitals at the boundary and keep
some of them frozen. They serve to cap the QM region, replacing the cut
bond.
Where to Cut
While cutting through covalent bonds can often not be avoided, as discussed
above, one can minimize its ramications by an appropriate choice of the
boundary, that is, of the bonds being cut. Apart from the obvious rule that the
QMMM frontier should be as distant from the chemically active region as
the size of the QM part (i.e., the computational demand) allows, one can give
some additional guidelines. A minimum requirement is that QM atoms participating in bond making or breaking should not be involved in any bonded
coupling term [50]. Since the dihedral terms extend at most two bonds into
the inner region (depending on the details of the boundary scheme in use,
191
Sect. 2.4.2), one is on the safe side if such atoms are at least three bonds away
from the boundary.
The bond being cut should be non-polar and not involved in any conjugative interaction (multiple bonding, hyperconjugation, stereoelectronic
interaction). A good place to cut is thus an aliphatic, innocent CC bond,
whereas cutting through an amide bond, which has partial double-bond character, is less desirable.
Another restriction is introduced by MM charge groups. It is common
practice in biomolecular force elds to collect several connected atoms into
a group with integer, normally zero, charge. This is advantageous since the
electrostatic interaction between neutral charge groups can be neglected to
rst order, thus enabling the construction of a molecule from these neutral units without reparameterization of the partial charges. Cutting through
a charge group is to be avoided because it creates an articial net charge in
the immediate vicinity of the QM density. It may also interfere with certain
algorithms that calculate the MM electrostatic interactions based on charge
groups. Finally, it is desirable, but not compulsory, that the total charge of the
MM atoms being replaced by the QM part is zero. In other words, the hole
created in the MM region that surrounds the QM part (and therefore the QM
part itself) should be neutral, such that the leading electrostatic interaction
between QM and MM subsystems is the dipole contribution.
2.5.2
Link Atoms
The appeal of the link-atom method, adopted already by early QM/MM studies [3, 4], lies in its conceptual simplicity: the free valency at Q1 created by the
QMMM separation is capped by introducing an additional atom that is covalently bonded to Q1 . This link atom L is in most cases a hydrogen atom, but
any monovalent atom or group is in principle conceivable. One thus performs
the QM calculations on an electronically saturated system consisting of the inner subsystem and the link atom(s), I + L. The bond Q1 M1 is described at
the MM level.
Although simple, the introduction of an additional atomic centre, which is
not part of the real system, entails consequences that need to be addressed:
Each link atom introduces three articial (structural) degrees of freedom
not present in the real system. This causes complications during structure
optimizations and raises the question of how the position of the link atom
is to be determined (discussed in more detail below).
The link atom, and with it the QM electron density, is spatially very close
to the MM frontier atom M1 . If M1 bears a partial charge and the QM
density is allowed to be polarized (i.e., electrostatic or polarized embedding), the point charge on M1 will overpolarize the density. Different
192
H.M. Senn W. Thiel
approaches to alleviate or eliminate this unphysical effect are presented

below.
The link atom introduces articial interactions with other link atoms and
the inner region. This has already been treated in Sect. 2.2.
Another issue of the link-atom method is that the link atom is generally
chemically and electronically different from the group it replaces. Attempts to
overcome this have led to the more elaborate boundary schemes discussed in
the following sections below. In the context of link atoms, Morokuma and coworkers suggested shifting the energy level of a selected orbital by means of
an additional one-electron operator in the Hamiltonian [97]. They proposed
mimicking the electronic effect of a substituent by a hydrogen link atom with
appropriately shifted energy levels [98]. A similar idea was explored using
an angular-momentum-dependent localized potential within the projectoraugmented waves method [99].
Despite their shortcomings, link atoms are the most popular and most
widely used boundary method. Correspondingly, a large variety of link-atom
schemes has evolved. One of the rst is due to Singh and Kollman [3], followed by Field, Bash, and Karplus [4]. Modied or extended formulations
were reported by several groups [96, 100104]. Another line of developments
derives from Morokumas IMOMM method [43, 46, 47, 105, 106]. A doublelink-atom method has also been proposed [107], in which a second link atom
is introduced to also saturate the MM region.
2.5.2.1
Placement of the Link Atom
In some schemes [3, 4, 102], the link atoms are treated as independent atomic
centres, thus introducing three additional structural degrees of freedom per
link atom. The link atoms are initially positioned at a certain distance along
the Q1 M1 bond vector, but are completely free during structure optimization. To mitigate the inconsistencies that arise when the articial bond Q1 L
is no longer collinear with the real Q1 M1 bond, it was suggested [96] that
a classical angle term LQ1 M1 be introduced, with an equilibrium value of
0 that keeps the bonds aligned.
To remove the excess degrees of freedom altogether, one can eliminate
them by the use of constraints. This was rst realized by Maseras and Morokuma [43], who constrained the MM frontier atom M1 to lie along Q1 L and
xed both the Q1 L and the Q1 M1 distances using a formulation in internal
coordinates. As this eliminates four, rather than three, degrees of freedom per
link atom, the procedure was subsequently modied [47] by dening the position of the link atom as a function of the positions of Q1 and M1 in Cartesian
coordinates:
RL (RQ1 , RM1 ) = RQ1 + g(RM1 RQ1 ) .
(10)
193
This denition places L along Q1 M1 , and the distance Q1 L is related to the

distance Q1 M1 by the scaling factor g. Exactly three degrees of freedom are
thus removed. Most current link-atom schemes are based on Eq. 10, using
different denitions for g.
Woo et al. [105] eliminate the coordinates of M1 , rather than L, by making
RM1 a function of RQ1 and RL :
RM1 (RQ1 , RL ) = RQ1 + g (RL RQ1 ) .
(11)
This formulation is equivalent to Eq. 10 in that it also removes three degrees

of freedom. However, the link bond Q1 M1 is not described at the MM level
in this case, its length being determined according to Eq. 11 from the Q1 L
distance, which is calculated at the QM level. The Q1 M1 bond is therefore
in principle allowed to break if Q1 L breaks. The factor g is related to g by
simply exchanging M1 and L in the denitions of g below.
If g is chosen to be a constant [47, 105], the distance Q1 L varies with the
length of the Q1 M1 bond (or the other way round in Woos scheme [105]).
A suitable choice for g is the ratio of the equilibrium bond lengths for Q1 L
and Q1 M1 [47]:
g=
d0 (Q1 L)
;
d0 (Q1 M1 )
(12)
where the values of d0 can be taken, e.g., from the force eld.
Alternatively, the link atom can be positioned at a constant distance from
Q1 [100, 103, 104, 106] by dening g as:
g=
d0 (Q1 L)
.
|RM1 RQ1 |
(13)
The Q1 L bond length is determined by the constant d0 (Q1 L), which is typically assigned different values for different types of QMMM bonds. Note
that g now depends on the positions RQ1 and RM1 .
Tavan and co-workers [101] supplement Eq. 13 with a term reecting the
deviation of Q1 M1 from its equilibrium value, making Q1 L again variable.
Their correction also accounts for the different stiffness of the Q1 M1 and
Q1 L bonds.
In principle, Eq. 10 or Eq. 11 can be extended to include other atoms in the
denition of the link-atom position. We are, however, not aware of any linkatom scheme that makes use of this generalization.
2.5.2.2
Link-Atom Forces
If the position of the link atom is not independent but expressed as a function
of other atomic positions, Eqs. 10, 12 and 13, its coordinates are eliminated
194
H.M. Senn W. Thiel
from the set of coordinates used to describe the entire system. In other
words, the link atoms appear only in the internal description of the QM/MM
coupling scheme and are transparent to geometry optimization or molecular dynamics algorithms, which handle the entire coordinate set. However,
as the QM calculation treats the link atoms generally on the same footing as
the atoms of the inner subsystem, there exist forces acting on the link atoms.
These forces are relayed onto the atoms appearing in the denition of the
link atom coordinates. The link atoms are then effectively force-free, and their
coordinates in the next geometry or time step are fully determined by the positioning rule, rather than being propagated according to the forces acting on
them.
The distribution of the forces acting on a link atom onto the atoms used in
its denition is formulated in terms of the chain rule. The dependence of the
total QM/MM energy on the coordinates of the entire system, {RI }, I S, and
the link-atom coordinates, RL , L L, can be expressed as:

EQM/MM {RI } = EQM/MM {RI }, RL {RI } .
(14)
The derivative of the energy with respect to an atomic position RK , K S,
is then obtained from the chain rule (dropping the QM/MM subscript for
clarity):

E
E
E RL
=
+
.
RK RK RL RK
(15)
The link-atom contribution to the force, the second term of Eq. 15, vanishes if
atom K is not involved in the denition of link atom L. There is a corresponding force contribution on K for each link atom in which K is involved.
RL
The notation R
designates the Jacobian matrix JK constructed from the
K
partial derivatives of RL with respect to RK . The elements of this 3 3 matrix
depend on the denition used to determine the position of the link atom. If
Eq. 10 is used together with the constant g of Eq. 12, the Jacobians JQ1 and JM1
take a particularly simple, diagonal form [47]:
JQ1 =
JM1 =
RL
RQ1
RL
RM1
= (1 g) ,
(16a)
= g ,
(16b)
where is the Kronecker symbol and , are Cartesian components.

If the bond length Q1 L is kept constant by applying Eq. 10 with the g
of Eq. 13, off-diagonal elements also appear in the Jacobians because of the
dependence of g on the positions of the two frontier atoms [100]. They are,
195
however, still symmetric:
JQ1 = (1 g) + g
RQ1 ,M1
RQ1 ,M1 ,
(17a)
RQ1 ,M1
JM1 = g g
RQ1 ,M1 ,
(17b)
where
RQ1 ,M1 designates the unit vector pointing from Q1 to M1 ,
RQ1 ,M1 =
(RM1 RQ1 )/|RM1 RQ1 |.
The corresponding expressions for the forces in Woos scheme, Eq. 11, are
obtained by replacing g by g and exchanging M1 and L in Eqs. 16 and 17.
2.5.2.3
Electrostatic Interactions at the Boundary
For the embedding schemes that allow the QM electron density to be polarized by the environment (i.e., electrostatic and polarized embedding, see
Sect. 2.3), there exists the problem that the QM density is overpolarized by
the rigid point charges of the MM charge model. While this artefact is always present to some extent when a point charge interacts with a polarizable
charge distribution, it is the more pronounced (i) the closer the point charge
approaches the QM density, and (ii) the more spatially exible the density is.
The problem is therefore especially critical at the QMMM boundary in
the presence of link atoms. If there are no covalent bonds across the boundary, the van der Waals interaction prevents the atoms from approaching each
other too closely. At a link, however, the link atom, which is part of the
QM region, is positioned in immediate proximity to the frontier MM atom,
A. Different approaches to alleviate the
typically at a distance of about 0.5
resulting spurious polarization effect are discussed below. One should also
keep in mind the possibility of other, non-bonded close contacts. In the context of biomolecules, hydrogen bonds across the QMMM boundary can lead
to non-bonded distances between the hydrogen and the acceptor atom of
A.
around 1.61.7
Overpolarization is less severe when small, atom-centred basis sets are
used in the QM calculation, e.g., a semi-empirical method with a minimal
basis. Larger basis sets, which include polarization and diffuse functions,
provide more exibility to place electron density further away from the nuclei and are therefore more prone to overpolarization. Especially affected are
methods using plane waves (see Sect. 3.1).
We continue by describing different approaches that have been put forward
to mitigate overpolarization within link-atom schemes:
Deleting one-electron integral terms: The spurious electrostatic interaction
between the QM density and the MM point charges is dominated on the
QM side by the one-electron terms in the Hamiltonian associated with the
basis functions and the nuclear charge of the link atoms. It has been sug-
196
H.M. Senn W. Thiel
gested that these terms be deleted [4, 51, 52, 104, 108], which effectively removes the interaction of the link atoms with the MM charge environment.
While this may be acceptable for semi-empirical QM methods, it becomes
problematic when used with higher-level ab initio or DFT methods and
larger basis sets because it leads to an unbalanced representation of the
polarization and electrostatic potential of the QM density [52, 108, 109].
A variant of this procedure is to delete only those terms that involve the
link atom and the charge of the MM frontier atom M1 , which was, however, found to yield inconsistent results [52].
Deleting point charges: The complementary approach to excluding the link
atom from interacting with the environment is to delete from the Hamiltonian one or more of the MM point charges at the boundary. These
charges then do not interact any more with the QM density as a whole.
Different implementations of this idea have been proposed: (i) deletion of
only the charge on M1 [96, 110114]; (ii) deletion of the charges on M1
and M2 [114]; (iii) deletion of those on M1 , M2 , and M3 [3, 114]; or (iv)
deletion of those on the atoms belonging to the same charge group as
M1 [52, 96, 102]. With the exception of the latter approach, these schemes
suffer from the creation of a net partial charge near the QM region, which
leads to severe artefacts [114, 115], and they do not conserve the total
charge of the system. Excluding the charge group to which M1 belongs
from interaction with the QM density avoids the most serious problems,
at least if the charge group is neutral. However, all these deletion schemes
distort the electric eld of the environment in the vicinity of the QM region, where it affects the QM density the most, which is not satisfactory.
Shifting point charges: To cure the problems of deleted-charge schemes,
different charge-shifting formulations have been introduced. They share
the common feature of preserving the charge and sometimes also the
dipole in the boundary region, while still removing the overpolarizing
partial charge from M1 . In the charge-shift scheme of Sherwood and coworkers [21, 103, 116, 117], the charge of M1 is distributed evenly over the
M2 atoms. The dipole created by shifting the charges is compensated for
by a pair of point charges placed near each M2 atom, which generate
dipoles of the same magnitude and opposite direction. In a variant of this
approach [115], the charge of M1 is distributed over the other atoms of the
charge group, with (charge shift) or without (divided frontier charge)
dipole correction.
In Lin and Truhlars redistributed-charge scheme [114], the charge of M1
is replaced by charges at the midpoints of the M1 M2 bonds. They also
propose modication of the values of the M2 and the redistributed charges
such that the M1 M2 bond dipoles are conserved (redistributed charges
and dipoles scheme). These formulations can be seen as classical substitutes of the generalized-hybrid-orbital (GHO) scheme (see below).
197
Recent evaluation studies [114, 115, 118] have demonstrated the importance of preserving charges and dipoles in the link region.
Charge smearing: Another alternative is to replace the point charge on M1
(and possibly other MM atoms near the QM region) by a charge distribution, which signicantly reduces the spurious overpolarization. A simple
choice for the form of the charge distribution is a spherical Gaussian,

MM (r) = qMM ( MM )3 exp (|r RMM |/MM )2 , where qMM , MM ,

and RMM are the charge, width, and centre of the charge distribution,
respectively. Such a scheme has been applied by Eichinger et al. [101]
in their QM/MM implementation in the cpmd code [119], which uses
plane waves as the basis set. It was also used within conventional QM/MM
methods based on atom-centred basis functions [107, 118]. The delocalized charge representation is applied based on a distance criterion (e.g., to
A of a QM atom) [101] or according to the connecall MM atoms within 5
tivity at the link (e.g., to M1 or M1 and M2 ) [107, 118]. The value chosen
A.
for the width parameter MM varies between 0.8 and 4
2.5.3
Boundary Atoms
Boundary-atom schemes replace the MM frontier atom M1 by a special
Janus boundary atom that participates as an ordinary MM atom in the
MM calculation but also carries QM features to saturate the free valency of
Q1 . They avoid the complications of link-atom approaches related to the introduction of additional atoms, such as ambiguous placement and articial
interactions, and enable one to mimic the electronic properties of the MM
moiety at the link.
Most of the boundary-atom schemes proposed are based on a type of
monovalent pseudopotential (or effective potential) that is parameterized to
reproduce certain desired properties and that is located at the position of M1 :
Adjusted connection atoms [120]: Dened within semi-empirical QM
methods (MNDO, AM1, PM3), adjusted connection atoms feature one valence electron in an s-orbital. They were tted using a set of 30 molecules
to mimic the structural and electronic (charges, dipole moment, formation energy) features of a methyl group by adjusting the atomic parameters of the respective method. The scheme is intended to saturate a cut
CC single bond.
Pseudobonds [121, 122, 738]: The pseudobond approach, developed for
ab initio and DFT methods, uses a monovalent, uorine-like boundary
atom with seven valence electrons, Z = 7, and an angular-momentumdependent effective core potential (ECP); in the original formulation [121],
it carries a uorine 321G or 631G basis set. With the aim of capping a C(sp3 )C(sp3 ) bond, the six ECP parameters (for 631G ) were
198
H.M. Senn W. Thiel
determined from six structural and electronic properties of ethane, as calculated with B3LYP. A given pseudobond is therefore specic to the bond
type and to the basis set used in the parameterization. It is, however, independent of the MM force eld and only weakly dependent on the QM
method. Although developed within B3LYP, the pseudobond parameters
are transferable to other DFT or HF calculations. The MM point charges
of the charge group to which M1 belongs are deleted.
In a recent modication of the scheme [122], C(sp3 )carbonyl C(sp2 )
and C(sp3 )N(sp3 ) pseudobonds were presented in addition to C(sp3 )
C(sp3 ). A simpler, angular-momentum-independent form of the ECP was
adopted, together with an STO-2G basis set on the boundary atom, which
is determined by six parameters.
Effective group potentials (EGP) [123127]: Aimed at replacing ligands
such as CO, NH3 , CH3 , or cyclopentadienyl in transition-metal complexes,
effective group potentials were also proposed [126] for use as boundary
atoms in QM/MM schemes. We are, however, not aware of any applications
so far.
The EGP can be regarded as a type of generalized ECP, expressed as an expansion over Gaussian projectors that may be located on different centres.
The EGP is determined such that a reduced representation of the system
mimics as closely as possible a suitably chosen reference system. Only the
electrons involved in bonding are described explicitly, e.g., one in the case
of CH3 .
Quantum capping potentials (QCP) [128, 129]: These combine the form of
shape-consistent ECPs with the idea of a one-electron potential. By adding
additional spherical terms to a standard carbon ECP with four valence
electrons and tting to molecular properties of ethane, effective potentials
with one explicit electron were obtained. These quantum capping potentials (QCPs) mimic a methyl group at the QM/MM boundary.
Effective Hamiltonians from a minimum principle [130]: A formal framework was proposed in which an effective QM/MM Hamiltonian is dened
that provides the best approximation in a least-squares sense to the full
QM Hamiltonian under the condition that no basis functions are located
in the outer region. ECPs or point charges, however, may be present. This
effective Hamiltonian is transferable, that is, it is independent of the inner
subsystem.
From this formalism, both a one-electron ECP and a classical potential
term were obtained that describe a methyl group in ethane and were
applied to several small test systems. Moreover, the LSCF (see below),
pseudobond, QCP, and EGP schemes were shown to be derivable from
within the formalism.
Optimized effective core potentials (OECP) [131]: QM methods using plane
waves as basis set are often implemented within the pseudopotential approach to eliminate the core electrons from the calculation. It seems
199
therefore natural to exploit the same formalism to describe boundary

atoms in plane-wave-based QM/MM calculations. Rthlisberger and coworkers [132] have used a one-electron ECP, empirically optimized to
reproduce the CC distance in ethane.
Recently, the same group [131] proposed a scheme that provides a systematic way of deriving optimized effective core potentials for use as
boundary atoms. It is based on a form for the pseudopotential frequently
employed in plane-wave calculations. The parameters are obtained by
minimizing a penalty function depending on the electron density. The
scheme was successfully tested on the methyl group of acetic acid, which
was replaced by a seven-electron OECP.
Multicentred valence-electron effective potentials (MC-VEEP) [133]: These
effective potentials are able to treat both ground and excited electronic
states correctly. They exploit the established quantum-chemical ECP approach and build on ideas similar to those used for the QCP method and
the minimum effective Hamiltonian. At the QM/MM boundary, these oneelectron potentials replace a methyl group.
2.5.4
Frozen Localized Orbitals
The approach of using a frozen hybrid orbital to saturate the dangling bond
at the QMMM boundary dates back to Warshel and Levitt [1]. Different
schemes have been elaborated that share the idea of placing a set of suitably
oriented localized orbitals on one of the frontier atoms and keeping some of
these orbitals frozen, that is, they do not participate in the SCF iterations:
Local self-consistent eld (LSCF) [134138]: In the LSCF method, developed by Rivail and co-workers, one starts out with a QM calculation
on a model system that contains the frontier bond to be described. Applying a localization scheme, one constructs a strictly localized bond orbital
(SLBO) for this bond. The SLBO has contributions from the frontier atoms
only and is assumed to be transferable. In the QM/MM calculation, it is excluded from the SCF optimization and does therefore not mix with other
orbitals. Its orientation is always kept along the Q1 M1 vector. It may be
described as a kind of frozen lone pair on Q1 pointing towards M1 .
To compensate for the additional electron introduced with the doubly occupied SLBO, an extra charge of 1e is placed on M1 , which interacts with
all other MM charges. On the model compound, a special classical bond
potential with ve parameters is tted, which is used in the QM/MM calculation together with the SLBO (parameters for common bonds are listed
in [138]). The MM charges on M1 and M2 are adjusted as necessary to
obtain a balanced description of the frontier bond and the polarization
of the QM region, while maintaining the overall charge [138, 139]. It has
200
H.M. Senn W. Thiel
also been suggested that the total point charge on M1 (i.e., compensation
charge + MM partial charge) be replaced by a Gaussian charge distribution [140].
Very recently, the performance of various localization schemes used in the
construction of the SLBOs was assessed and compared to extremely localized molecular orbitals (ELMOs) [141, 142]. The latter were found to be
superior because of their better transferability. They avoid the somewhat
arbitrary deletion of orbital contributions not localized on the frontier
atoms.
Frozen orbitals [93, 143, 144]: Friesner and co-workers have presented
a formulation of the LSCF procedure that differs in some technical details from the original one; for instance, the compensation charge is
placed at the midpoint of the Q1 M1 bond. Furthermore, there is a major conceptual difference as compared to most other QM/MM schemes
in that the QMMM interactions at the boundary are heavily parameterized: (i) Several electrostatic correction terms are introduced that reduce
the short-range electrostatic interactions at the interface, following the
spirit of 12, 13, and 14 electrostatic exclusion and scaling rules used
in many force elds. These corrections also require the assignment of
point charges to the atoms of the inner subsystem and involve the optimization of the MM, QM, and bond partial charges in the boundary
region. (ii) As mentioned in Sect. 2.4.1, the van der Waals parameters of
the QM atoms are re-optimized. (iii) Certain classes of hydrogen bonds
across the boundary are described by an additional repulsive term. (iv)
The QMMM bonded terms are re-optimized, rather than taken directly
from the force eld.
The goal of this extensive parameterization is to reproduce as closely as
possible the conformational and reaction energetics in the boundary region. A database of parameters has been derived for QMMM bonds in
the backbone and on the side chains of proteins. The parameterization is
specic for the basis set and the QM method.
Generalized hybrid orbitals (GHO) [145150]: The GHO method of Gao
and co-workers is closely related to the LSCF and frozen-orbital approaches in that it constructs localized hybrid orbitals and freezes some
of them. However, it places the set of localized hybrid orbitals on M1 ,
rather than Q1 . M1 thus becomes a boundary atom. (The classication
of boundary methods into boundary-atom and frozen-orbital schemes is
therefore somewhat arbitrary.) The orbital pointing towards Q1 is active
and participates in the SCF iterations, while the remaining auxiliary
hybrids are kept frozen and are not allowed to mix with the other orbitals.
The standard C(sp3 ) boundary atom in a HF or DFT calculation has
four electrons, Z = 4, and a minimal basis set, from which four localized
201
sp3 hybrid orbitals are constructed. The MM point charge of M1 is distributed equally over the three frozen auxiliary hybrids. They thus provide
a type of pseudopotential that mimics the electronic character at the link.
The hybridization is completely determined by the local geometry at the
boundary, that is, by the relative positions of Q1 , M1 , and M2 . In contrast
to the LSCF and frozen-orbital methods, there is thus no need for parameterization calculations on model compounds to derive the localized
hybrids.
However, to improve the structure at the boundary, in particular the
Q1 M1 distance and Q1 M1 M2 angles, some additional parameters are
introduced. Depending on the QM level at which the GHO scheme is implemented, certain classical bonded terms involving M1 are modied or
added and/or certain integrals are scaled. For instance, in the HF and DFT
implementations, the one-electron, one-centre kinetic-energy integrals involving orbitals on M1 or on M1 and Q1 are scaled, which introduces seven
parameters.
2.5.5
Boundary Schemes: Summary
Several studies have evaluated the merits and drawbacks of different boundary methods. As link atoms are the most widely used boundary scheme,
most of these assessments [52, 109, 114, 115, 118, 139, 140, 151] compare linkatom approaches, which differ in particular in the way that the charges at
the boundary are handled. Some have also compared link-atom to localizedorbital schemes [109, 139, 140].
Approaches based on hybrid orbitals are certainly more fundamental from
the theoretical point of view, providing a boundary description essentially
at the QM level. They also avoid some of the complications inherent in the
link-atom method arising from introducing additional atoms. However, they
are technically considerably more complicated, not least because of the orthogonality constraints required to prevent the mixing of frozen and active
orbitals. In addition, the localized orbitals themselves, or specic parameter
sets related to them, must be determined beforehand, involving calculations
on model compounds. These parameters are usually not transferable and
need to be reconsidered upon changing the MM force eld or the QM method
or basis set.
The conclusion from the available evaluations is that localized-orbital approaches can sometimes be tweaked more specically towards a given application, but that the performance of link-atom schemes is generally on
par. They both provide reasonable accuracy when used with care; in particular, one should minimize distortions of the charge distribution at the
boundary.
202
H.M. Senn W. Thiel
3
Choice of QM and MM Models, QM/MM Implementations
3.1
Choice of QM Method
The QM/MM formalism as such is sufciently exible to accommodate almost any QM method. The particular choice thus follows the same criteria
as in pure QM studies and is not further elaborated on here. Especially with
link-atom schemes, only minimal changes to the QM code are required. Essentially, the QM code must be able to perform the SCF treatment in the presence of the external point-charge eld, representing the MM charge model in
the case of electronic or polarized embedding. Other boundary schemes can
require somewhat larger modications.
In practice, many current biomolecular QM/MM applications use DFT as
the QM method due to its favourable cost/accuracy ratio. Traditionally, semiempirical QM methods have been the most popular, and they remain important for QM/MM molecular dynamics, where the aspect of computational
cost is paramount. They are also very useful in method evaluation studies because they can be expected to enlighten the same problems as would occur
with more costly ab initio methods. Especially in the context of biomolecular
QM/MM studies [115, 152, 153, 739], the semi-empirical, DFT-inspired SCCDFTB (self-consistent-charge density-functional tight-binding [154]) method
due to Elstner and co-workers appears promising because it approaches,
within the validity domain of the parameterization, the accuracy of DFT at
the cost of a semi-empirical treatment.
At the other end of the spectrum are the post-HartreeFock ab initio
electron-correlation methods, such as those based on MllerPlesset perturbation theory (e.g., to second order, MP2) or coupled-cluster theory (e.g.,
CCSD including single and double excitations or CCSD(T) adding a perturbative treatment of triple excitations). Recent developments (exemplied by
the work of Schtz, Werner, and collaborators [155162]) have extended the
size of systems that can be treated with these methods by almost an order of
magnitude to several tens of atoms. They take advantage of the short-ranged
nature of electron correlation and are commonly referred to as local methods
(e.g., LMP2, LCCSD); their computational effort scales linearly with system
size. The superior accuracy of high-level ab initio methods can therefore now
also be exploited for biomolecular QM/MM studies [740], certainly at the
level of energy calculations at xed geometries (i.e., single points).
203
Plane-Wave Methods
We highlight here specic issues of QM methods employing plane waves as
the basis set within a QM/MM framework. The application of such methods is
intimately linked to CarParrinello rst-principles molecular dynamics [163,
164], which almost always treats the electronic-structure problem within DFT
in a plane-wave basis. There are two main issues, both concerning the electrostatic QMMM coupling used in electrostatic or polarized embedding.
The rst is a fundamental issue, the severe overpolarization of an electron density expanded in plane waves by a bare point charge, sometimes
referred to as the electron spill-out problem. While any charge distribution is overpolarized to some extent when interacting with a point charge (see
Sect. 2.5.2) the effect is completely deleterious when plane waves are used.
Because they form an origin-less basis set, the density is pulled away from
the nuclei and localizes in the purely attractive potential around the point
charges. The problem has been dealt with in different ways:
By substituting the point charges within a certain distance from the QM
region by Gaussian charge distributions (see Sect. 2.5.2) [101].
By coupling the point charges to a model density of atom-centred Gaussians [165], which reproduces the multipoles of the true density and is
variational with respect to the true density [166].
By smoothly replacing the Coulomb potential at short range by a form
which goes to a constant at zero distance [132].
By representing the point charges as Slater-type s-functions (or, more generally, partial-wave expansions), which also provides a nite potential at
zero distance [167].
The second problem is of a more technical or algorithmic nature. Directly
evaluating the (possibly modied) Coulomb interaction between the MM
point charges and the QM density presents a considerable computational
effort. In a plane-wave scheme, the electrostatic potential due to the total
density is represented on a real-space grid in the unit cell. Hence, computation of the Coulomb energy requires Nq Nr evaluations, where the number
of charges, Nq , is of the order of 104 and the number of grid points, Nr
is of the order of 1003 . Several techniques have been proposed to reduce
the cost:
A hierarchical multipole expansion is used to represent the electrostatic
potential at the grid points due to the point charges [101]. However,
this scheme is not symmetrical (i.e., not Hamiltonian and therefore not
energy-conserving) in that the electrostatic potential acting on the MM
charges due to the density is derived from a point-charge model of the
density.
Beyond a chosen distance, the Coulomb interaction is evaluated from
a multipole expansion of the density [132] and directly otherwise. An
204
H.M. Senn W. Thiel
intermediate layer can be dened, where the density is represented by

variational electrostatic-potential-derived (ESP) charges [168, 169] to calculate the interaction with the MM charges.
The point charges are represented by Gaussian charge distributions and
their potentials by sums of Gaussians with different widths (Gaussian expansion of the electrostatic potential, GEEP) [170]. These Gaussians are
then mapped onto the suitable grid level of a multigrid scheme. The potential on the nest grid, which couples to the density, is obtained by
sequentially interpolating starting from the coarsest grid level. This procedure reduces the computational cost by up to two orders of magnitude
without introducing cutoff parameters that need to be adjusted. It scales
linearly for systems as small as a few hundred atoms. Recently, an extension to periodic boundary conditions has appeared [741].
3.2
Choice of MM Method
The QM/MM formalism is also largely independent of the choice of the MM
method. Subtractive QM/MM schemes are generally more easily compatible
with wider classes of force elds than additive ones. As far as force elds for
(bio)molecular systems are concerned [171175], they are all valence force
elds using point-charge models. As discussed in Sect. 2.3.3, there is no established polarizable biomolecular force eld available as yet. One can further
differentiate between all-atom and united-atom force elds. The latter describe explicitly only selected hydrogen atoms (typically the polar ones) and
replace the remaining ones by a suitably parameterized united atom representing, e.g., a CH2 unit.
We list here the most widely used biomolecular force elds as well as a few
more general ones:
Biomolecular force elds: AMBER [66, 176178], CHARMM [39, 179
182], GROMOS [183, 184], OPLS-AA [185187]
General-purpose force elds: MM3 [188194], MM4 [195205],
MMFF [206213], UFF [214217]
We give only the family names, without detailing specic variants. We emphasize that it is important to distinguish very clearly the force eld proper
from the force-eld engine (i.e., the program in which it is implemented),
especially when they bear the same name, and to specify precisely which parameter set was used (as characterized, e.g., by the exact designation and citation). We also note that biomolecular typically includes proteins [171, 172]
and in many cases also nucleic acids [173, 174], but less frequently carbohydrates [175] or lipids. See [4042] for recent surveys on available force
elds.
205
3.3
QM/MM Implementations
3.3.1
Program Architecture and QM/MM Packages
There are essentially three main approaches in which QM/MM implementations have been realized: (i) by adding QM capabilities to an MM package;
(ii) by adding MM functionality to a QM package; or (iii) in a modular manner by coupling existing QM and MM programs to a central QM/MM engine.
Approaches (i) and (ii) take advantage of the inherent strengths of the respective base program. MM packages are designed to handle large, complex
systems and offer the corresponding simulation and analysis tools, whereas
quantum-chemistry programs traditionally provide, e.g., efcient algorithms
to locate stationary points on the potential-energy surface.
The modular approach (iii) minimizes as far as possible modications of
the standard QM and MM codes and offers more exibility. The external QM
and MM packages are linked via interfaces to a central core that supplies the
QM/MM coupling as well as routines for structure optimization, molecular
dynamics, etc. The system is relatively easily extended by interfacing to additional QM or MM programs. When updated versions of the external programs
become available, they can normally be used immediately or with only minimal changes to the interface routines. The core also provides a common user
interface to the external programs, at least for the most common tasks.
There are also drawbacks to the modular architecture: (i) the increased
complexity of the program because of the need to create a unifying framework that is able to accommodate external programs with possibly very different designs; (ii) the considerable amount of data being transferred between
the core and the external programs and between different core modules,
which is often done by writing and reading les on disk; and (iii) the repeated
start-up, initialization, and close-down of the external programs on each call.
The latter two points have implications for the efciency of the whole program system.
We list here examples for each of the three types of QM/MM implementations and describe in somewhat more detail the modular QM/MM package
ChemShell that is co-developed in our laboratory:
MM packages with QM: amber [218, 702], boss (MC simulations, semiempirical QM only) [219, 703], charmm [180, 704]
QM packages with MM: adf [220, 705], gamess-uk [221, 706], Gaussian [707], NWChem [222, 708], QSite/Jaguar [709]; CarParrinello MD
codes with QM/MM capabilities: cpmd [101, 119, 131, 132, 168, 169, 223,
224], cp-paw [99, 105, 165, 225227, 710].
Modular QM/MM packages: ChemShell [117, 711], qmmm [712]
206
H.M. Senn W. Thiel
The ChemShell package

Following the modular design approach sketched above, ChemShell is
presently interfaced with the following QM packages (see Fig. 3): gamessuk [221, 706], Gaussian03 [707], Molpro [713], mndo99 [228], orca [714],
and turbomole [715]. On the MM side, interfaces are available to charmm
[180, 704], gromos96 [184, 716], and gulp [229, 230, 717]. These external
programs are supplied with input data (coordinates of atoms and point
charges, command le) by the core and return energy and gradient data.
The core modules provide the following key functionalities:
QM/MM coupling: mechanical and electrostatic embedding, shell model
for solid-state embedding; link atoms with charge shifting; double link
atoms with charge smearing
Structure optimization using different types of coordinates (Cartesian, internal, hybrid delocalized coordinates)
Molecular-dynamics driver for NVE, NVT, and NPT simulations with
rigid-body motion (quaternions) or SHAKE constraints
Vibrational frequencies from nite-difference second derivatives
A general force-eld engine compatible with the functional form of various force elds (e.g., AMBER, CHARMM, UFF)
Utility routines for data management, coordinate manipulations, etc.
Fig. 3 Schematic design of the modular ChemShell QM/MM package
207
The MD driver and the force-eld engine are built on routines from
dl_poly [718] that are tightly integrated into the core.
The controlling framework and the user interface are implemented within
the tool command language (Tcl). The user input is in the form of Tcl scripts.
The various functions are implemented either directly in Tcl or as Tcl wrappers to lower-level routines written in C or Fortran.
Current development projects include the incorporation of a polarizable
water model due to van Gunsteren and co-workers [61] within the gromos
interface as well as algorithms for QM/MM calculations on excited states (localization of conical intersections, surface hopping).
3.3.2
QM/MM-Related Approaches
Some QM/MM or QM/MM-like schemes, which have been applied to
biomolecular systems, are sufciently conceptually distinct from the standard QM/MM implementations to be mentioned separately:
Empirical valence-bond (EVB) method [231234]: Within a valence-bond
description of the reactive part of the system, the energies of the interacting diabatic (resonance) states are represented by simple empirical
potential terms. They incorporate the interaction of the charges of the
states with the environment. No QM (electronic-structure) calculations
are involved. Two-state models are the most common. The EVB energies
are calibrated to reproduce experimental or ab initio data. The method has
been applied notably by Warshel and collaborators to model the inuence
of the solvent or protein environment on reactions. It offers the possibility
of driving a reaction from reactants to products by means of a mapping
potential that gradually transforms the diabatic state describing the reactant into the product state, which can be considered a kind of electronic
reaction coordinate.
Effective fragment potentials (EFP) [235238]: From separate calculations
on model systems, one derives a set of one-electron terms (the EFP)
that account for the electrostatic, inductive, and repulsive interactions
of a moiety. For instance, the electrostatic potential is represented by
a distributed multipole expansion. The EFP is then incorporated into the
Hamiltonian of a QM calculation, where it describes the effects of the environment on the QM part. The method was originally designed to model
the solvent environment in QM calculations by means of EFPs. These fragments may be considered the MM part of a QM/MM scheme.
The scope was extended to biomolecular systems, using an LSCF-type procedure (see Sect. 2.5.4) to treat covalent bonds across the boundary [237,
238]. Recent developments have lifted the limitation that the fragments
be xed in space [239241]. By dening relatively small fragments (a few
208
H.M. Senn W. Thiel
atoms) and treating the fragmentfragment interaction by an empirical

potential, the MM subsystem can be represented as a chain of effective
fragments, which are internally rigid but are allowed to move relative to
each other.
QM/MM structure renement [242244]: Ryde and co-workers recently introduced an approach that integrates QM calculations into the renement
of X-ray crystallographic or NMR-derived structural data of biological
macromolecules, in particular proteins.
Conventionally [245], an MM model of the structure is rened against
the experimental electron-density data. Whereas these models have been
carefully tuned for the macromolecular part, they are clearly less reliable
when it comes to substrates, products, or other ligands bound in the active site or to cofactors and prosthetic groups, that is, compounds of other
chemical classes also present in the structure. This poses the dilemma that
the structural information in and around the active site, the focus of most
mechanistic studies, is usually the least dependable [246].
The new approach alleviates this problem by describing, e.g., the nonprotein components in the active site of an enzyme at the QM level, while
the remainder is treated by the force eld. In contrast to standard QM/MM
schemes, but in line with established renement procedures, no electrostatic interactions are considered, neither between QM and MM regions
nor within the MM part. The program, named ComQum-X, is derived
from a QM/MM implementation by Ryde and co-workers. Merz and coworkers have recently presented a similar approach [742].
4
Optimization and Simulation Techniques Used with QM/MM
4.1
General Comments
The QM/MM method as presented in Sect. 2 is a potential-energy scheme
yielding the energy (and the forces) at a given structure. As such it can be
combined with any approach that makes use of this information to update
the structure, that is, with any optimization, molecular-dynamics, or Monte
Carlo technique. Moreover, a single QM/MM energy and gradient evaluation is not signicantly more costly than the corresponding QM calculation
alone.
However, it is an almost intrinsic feature of the QM/MM method that it
is applied to large systems with at least hundreds, but often several 10 000
structural degrees of freedom. Two types of problems hence arise: (i) There
are algorithmic complications due to the mere system size. Algorithms for
209
manipulating the coordinates (e.g., in optimization schemes) whose computational effort per step scales as O(N 2 ) or O(N 3 ), where N is the number of
degrees of freedom, are not computationally viable any more. At the same
time, procedures with more favourable scaling, well-established from pure
MM calculations, are often not suitable either because they need much more
(however less costly) steps, which is not affordable with QM/MM methods.
(ii) Conformational complexity is an issue. The number of steps required to
converge an optimization or to obtain sufcient sampling increases with the
number of degrees of freedom.
The optimization and simulation techniques presented in the following
sections follow two strategies to address these problems: (i) Some are generally suitable for large systems, e.g., by scaling as O(N). (ii) Others take
advantage specically of the partitioning of the system into the QM region,
where evaluating energy and gradients is computationally demanding, and
the MM region, where these calculations are almost for free.
Related to the determination of stationary points (minima, saddle points)
on the potential-energy surface (PES) for biomolecular reactions, a comment
is in order that projects beyond the purely algorithmic issues discussed below.
The vast size of the conguration space available to a reacting biomolecular system as compared to a small-molecule reaction drastically reduces
the signicance of single stationary points. As opposed to typical QM studies
on smaller molecular systems, knowledge of the optimized reactant, transition state, and product is not everything, even if a reaction path connecting
them has been determined and even if we disregard nite-temperature effects
for the moment. As recently re-emphasized by Warshel and co-workers [247],
one should in QM/MM optimization studies consider at least several representative transition states with their corresponding minima. While this does
not fully substitute an extensive congurational sampling, it reects at least
partly the inuence of the conformational diversity of the environment on
the reaction. An expedient way to generate a selection of environmental congurations is, for instance, to take snapshots from a classical MD trajectory
and use them as starting structures in subsequent QM/MM optimizations, as
demonstrated, e.g., in [248].
4.2
Stationary Points and Reaction Paths
For general reviews on algorithms for locating minima and (rst-order) saddle points on the PES, we refer to [27, 249253]. We rst briey consider
optimization methods treating the complete system on equal footing, which
thus are generally applicable to large systems and are not specic to the
QM/MM approach. Secondly, we describe techniques specically designed
for, and taking advantage of, the QMMM partitioning.
210
H.M. Senn W. Thiel
4.2.1
General Optimization Techniques for Large Systems
Among the most efcient algorithms developed to locate stationary points
on the PES are quasi-Newton methods using some form of internal coordinates. Efciency here means that they take a minimal number of steps to
convergence, that is, a minimal number of energy and gradient calculations.
They require an (approximate) Hessean, which is updated, e.g., according
to the BroydenFletcherGoldfarbShanno (BFGS) formula for minima. The
matrices involved in the conversions between Cartesian and internal coordinates and the Hessean itself are N N; their diagonalization is an O(N 3 )
operation, storage scales as O(N 2 ), which becomes impractical for large
systems.
A popular optimization algorithm that scales linearly in both CPU and
memory usage is the limited-memory BFGS (L-BFGS), which stores only the
diagonal of the Hessean and uses information only from a limited number of previous steps. However, it does not resolve the scaling problem of
the coordinate-conversion steps. Optimization methods that reduce the overall scaling to O(N 2 ) or even O(N) by various algorithmic enhancements
have been developed, e.g., by Farkas and Schlegel [254257], Pulay and coworkers [258260], and others [261263].
The scheme by Billeter et al. [264] divides the system into fragments and
performs expensive coordinate manipulations only within these fragments
(divide-and-conquer approach). Each fragment is described by a set of
non-redundant delocalized coordinates [265], built from redundant primitive
internals [266]. However, to preserve the information about the fragments
absolute position in space, its Cartesian coordinates are included together
with the internals in the delocalization procedure, affording hybrid delocalized coordinates (HDLCs). Using them within the L-BFGS algorithm provides
an overall O(N) scheme.
For the optimization of transition states, the idea of micro-iterations [43,
267, 268] has been combined with HDLCs [264]. A core fragment is dened
that contains the atoms immediately involved in the reaction. Within this
core, a partial rational-function optimizer (P-RFO) with Powell updates of
the approximate Hessean is applied; the initial core Hessean is calculated exactly. With the core kept frozen, the environment fragments are fully relaxed
using the L-BFGS algorithm in HDLCs. One P-RFO step is then taken in the
core, and the procedure is iterated to convergence. This scheme thus searches
for a rst-order saddle point, controlled by the (low-dimensional) core Hessean, while minimizing the energy with respect to the environmental degrees
of freedom. The partitioning into core and environment is exact in the limit
of a completely relaxed environment (i.e., exactly vanishing gradients). The
environment then adiabatically follows the core degrees of freedom. This is
approximated by applying a stricter convergence criterion to the environment
211
than to the core. These procedures have been implemented in the optimizer
HDLCOpt [264] and are part of the ChemShell package.
A combination of an RFO with an L-BFGS algorithm for core and environment, respectively, has been described that uses Cartesian coordinates [269].
The core and environment optimizations are alternatingly iterated to convergence.
4.2.2
Optimization Techniques Specific to QM/MM
The principal idea here is to exploit the division of the system into QM and
MM regions in the spirit of the micro-iterative scheme [43, 267, 268]. The
core/environment partitioning thus reects the QM/MM partitioning. Different coordinates and optimization algorithms can be applied in the two
regions; for instance, plain Cartesians with a conjugate-gradient or truncatedNewton algorithm for the MM degrees of freedom (which avoids costly coordinate or Hessean manipulations) and an efcient quasi-Newton algorithm in
some form of internal coordinates for the QM degrees of freedom. If desired,
additional atoms beyond the QM part may also be included in the core region.
There are essentially two possibilities to achieve mutual convergence of
the optimizations in the two parts: (i) the adiabatic approach, in which the
MM system is fully relaxed in each QM step [43, 93, 139, 270, 271]; and (ii)
the alternating scheme, where the QM and MM optimizations are alternatingly iterated to convergence [3, 272, 273]. The QM atoms are always kept
xed during the MM optimization and vice versa. Different options within the
micro-iterative scheme (including the frequency of environment minimizations, the size of the core, and approximations to the electrostatic QMMM
interaction) are assessed in [274] for an enzymatic reaction.
The use of different coordinate systems and optimizers can compromise
the convergence behaviour and lead to numerical instabilities. The reasons
are that the approximate QM Hessean contains no information about the MM
atoms (i.e., the QM and MM degrees of freedom are treated as decoupled)
and that the Hessean-update procedures may become unstable as a result
of spurious force contributions on QM atoms due to the environment [271].
An adiabatic optimization scheme for minima that remedies these complications for mechanical embedding has very recently been presented [271].
It uses Cartesians for the MM atoms and redundant internal coordinates in
an RFO-type algorithm for the QM atoms. QM and MM degrees of freedom
are explicitly coupled in the QM optimization by virtue of MM contributions
to the Hessean, which are analytically evaluated. The scheme takes advantage of O(N) techniques and employs fast-multipole methods to evaluate the
electrostatic interactions.
Separating the QM and MM optimizations is particularly straightforward
for mechanical embedding and has been used, for instance, by Maseras and
212
H.M. Senn W. Thiel
Morokuma [43] with their IMOMM scheme. For electrostatic embedding,

however, the difculty arises that the electrostatic QMMM interaction is
evaluated at the QM level. That is, a QM calculation is required in each MM
step to let the density adapt to the new MM conguration and to obtain the
forces on the MM atoms due to the QM density. To restore the decoupling of
QM and MM calculations, one can represent the QM density in the MM calculations by a point-charge model. Schemes based on atomic charges tted to
the electrostatic potential (ESP) have been proposed, the basic assumptions
being that the QM charge distribution is sufciently accurately described by
the charges and that its relaxation during the MM optimization can be neglected.
Yang and co-workers [273] implemented an alternating scheme. They derive the ESP charges from the optimized QM part, use them to optimize the
MM part, then re-optimize the QM part, and so on. The drawback of this procedure is that the nal structure and energy are not consistent because the
structure is obtained using the ESP charges, whereas the energy is calculated
from the full QM density.
An adiabatic approach has been presented by Friesner and co-workers [93],
who use the electrostatic force calculated with the ESP charges as a correction
during the MM optimization. In each step of the QM optimization, the true
QM
el, ESP
force F el,
QM-MM (0) and its approximation F QM-MM (0) are obtained. Their difference is added to the ESP-derived force in each step k of the subsequent full
MM optimization:
el, QM
el, ESP
el, ESP
F el
QM-MM (k) = F QM-MM (k) + F QM-MM (0) F QM-MM (0) .
(18)
When the MM optimization has converged, the next QM step is taken. Once
mutual convergence has been achieved, no more MM steps are needed at
ESP
el, ESP
a given QM conguration, hence the correction F el,
QM-MM (k) F QM-MM (0) vanishes, yielding a fully consistent structure and energy.
A modication of this procedure has been suggested [270], in which the
density and the MM conguration (i.e., SCF calculations and MM optimizations) are iterated to convergence before the next QM step is taken. This eliminates the approximation that the change in the density due to the changed
MM conguration is negligible and guarantees that the QM optimizer always obtains fully consistent forces. It was also proposed that fast multipole
methods be used instead of the ESP charge model to calculate the QMMM
electrostatic interaction. However, we are not aware of a practical implementation of these suggestions as of this writing.
Mart et al. proposed use of a lower-level QM scheme to calculate the
QMMM electrostatic interaction during the MM optimization, rather than
a point-charge model [275277]. This leads to smoother convergence behaviour because the QM density can adapt to changes in the MM environment, and is not much costlier than a point-charge description.
213
4.2.3
Reaction-Path Techniques
The nudged-elastic-band (NEB) method [278282] represents the minimumenergy path (MEP) by a chain of states in conguration space (also called
replicas) linked by springs. The optimized MEP provides a continuous path
connecting reactant and product states without assuming a specic reaction
coordinate. Yang and co-workers have extended the NEB method for large
systems [283]. In particular, they include in the path denition only selected
degrees of freedom involved in the reaction and make sure that the environment follows the reaction smoothly.
The same group also adapted the path-optimization procedure by Ayala
and Schlegel [284] for the use within a QM/MM approach [285]. They make
use of their alternating QM/MM optimization scheme [273] and again restrict
the degrees of freedom considered in the denition of the distance between
congurations to the relevant coordinates. Recently, they have combined their
NEB and path-optimization schemes into a two-step procedure [286]. The
NEB method is used to obtain an approximate reaction path, which is subsequently rened with the path optimizer.
A chain-of-replicas method has been proposed [287] that replicates only
a selected subsets of atoms and restrains them based on the root-mean-square
rotational and translational distance to the adjacent steps. Implemented in
charmm, it is suitable for use with QM/MM approaches and is efciently
parallelized.
4.3
Molecular Dynamics and Simulation Techniques
As mentioned above, the QM/MM energy and forces can in principle be
used within any molecular dynamics (MD) or Monte Carlo (MC) scheme. In
most cases, the rationale behind such simulations is the sampling of phase
(or rather conguration) space to calculate statistical-thermodynamical ensemble averages. Typical examples include free-energy differences, such as
reaction or activation free energies or free energies of solvation.
As the amount of sampling necessary to obtain converged averages is considerable, especially for complex systems such as biomolecules, the computational demands of these simulations are extremely high, even considering
the fast increase in available computing power. Similarly to the optimization
schemes presented in the previous section, simulation approaches have therefore been developed that reduce the computational cost by taking advantage
of the QMMM partitioning; in particular, one tries to avoid the expensive
direct sampling of the QM contribution.
We highlight in this section a number of simulation techniques that have
been applied in QM/MM calculations on biomolecular systems.
214
H.M. Senn W. Thiel
4.3.1
QM/MM Molecular-Dynamics and Monte Carlo Simulations
Historically, full QM/MM simulations (i.e., with freely moving QM atoms)
were rst applied as explicit-solvent models to calculate solvation free energies or reaction free energies in solution. Both molecular dynamics and
Monte Carlo were used as sampling methods, along with standard free-energy
methods such as free-energy perturbation (FEP), umbrella sampling, and
thermodynamic integration (see [288292] for general overviews on MD,
MC, and free-energy methods). In most cases, a semi-empirical QM method
was used. We note here contributions by Bash, Field, and Karplus [293], Gao
and co-workers [5, 6, 294, 295], and Merz and co-workers [296299]; the latter
group also utilized rst-principles QM methods (DFT, HF).
Probably the rst full QM/MM MD simulation on a biomolecular system
(a metalloenzyme) was reported by Hartsough and Merz in 1995 [300] with
a semi-empirical QM method. Whereas calculations at this level are now well
affordable (e.g., [301]), QM/MM MD with a rst-principles QM method has
remained very computationally demanding also by todays standards and is
not routinely performed.
In the above-mentioned studies, the QM energy and forces are taken from
a converged SCF calculation in each step of the dynamics, that is, they employ a BornOppenheimer MD scheme. An alternative is CarParrinello MD
(CP-MD) [163, 164], where the wave functions are treated as ctitious dynamic variables within a Lagrangean scheme and follow the nuclei on the
y. QM/MM approaches based on CP-MD have been developed by Woo
et al. [105, 165, 225227], Tavan and co-workers [101], and Rthlisberger and
collaborators [131, 132, 168, 169, 223, 224] (see also Sects. 3.1 and 3.3.1). Notably, the latter group has been active in the area of biomolecular simulations.
While the CarParrinello scheme is most commonly implemented with
DFT and plane waves, a very much related extended-Lagrangean formulation using atom-centred Gaussians has been presented, the atom-centred
density-matrix propagation (ADMP) method [302306]. It has recently been
combined with the ONIOM QM/MM scheme [307]. We are, however, not
aware of any biomolecular applications so far.
Also within QM/MM MD, one can apply the adiabatic micro-iterative
scheme by optimizing the MM degrees of freedom in each step of the QM
dynamics. In a sense, this can be considered a limiting case of the multipletime-step technique, where the MM region is oversampled with respect to
the QM region by performing multiple time steps in the MM part per QM
time step. Oversampling the MM environment generates an effective freeenergy surface, on which the QM degrees of freedom move. The adiabatic and
multiple-time-step schemes have been implemented within QM/MM CP-MD
by Woo et al. [226, 227]. They are, however, generally applicable to QM/MM
MD approaches.
215
Because of the considerable computational cost of full QM/MM MD simulations, semi-empirical QM methods are still a popular choice in this context. Several approaches have been suggested to improve the accuracy of
the results. These include tuning the semi-empirical Hamiltonian with specic reaction parameters (SRP) [308311]; adding a valence-bond correction
term [312, 313]; and correcting the potential of mean force obtained at a lower
QM level with an interpolative scheme based on few higher-level calculations [314316].
4.3.2
QM/MM Free-Energy Perturbation
A QM/MM free-energy perturbation (FEP) scheme to calculate free-energy
differences along a predened reaction path has been proposed by Yang
and co-workers [273]; Rod and Ryde recently presented a similar formulation [317, 318], dubbed QTCP (quantum-mechanical thermodynamic-cycle
perturbation). The basic idea is to sample only the MM degrees of freedom, thus signicantly reducing the computational effort compared to full
QM/MM sampling.
One determines rst a series of QM/MM-optimized structures along the
chosen reaction path. The Helmholtz free-energy difference between two
states (i.e., points on the path) a and b is then written as:
ab
ab
Aab
QM/MM = AQM + Aenv .
(19)

The contribution from the QM region Aab
is expressed
QM
differ as the

ZPE
ence of the QM
EQM plus zero-point energy EQM , nite energies

th
temperature UQM , and entropic SQM contributions obtained within
the harmonic approximation [319]:
ZPE
th
Aab
QM = EQM + EQM + UQM TSQM .
(20)
The harmonic frequencies for the QM part are calculated at the end points
of the path (assuming that they are stationary points), in the presence of the
xed MM region. Note that, for N QM degrees of freedom, all N (N 1 in the
case of a transition state) vibrational degrees of freedom are considered in the
calculation of the partition function; there are no rotational and translational
contributions.
The free-energy contribution of the environment, Aab
env , is obtained
from Zwanzigs perturbation formula [320] (which, despite its name, has
nothing to do with a perturbational treatment in the usual sense). It incorporates the free-energy contributions from the MM region and the QMMM
coupling. One samples the MM degrees of freedom in state a and considers
216
H.M. Senn W. Thiel
the perturbation due to the QM region of state b:

a
ab
Aab
,
env = RT ln exp Epert /(kB T)
MM
(21)
where

b
a
R
E
R
=
E
(t);
R
(t);
R
Eab
QM-MM
MM
QM-MM
MM
pert
QM
QM .
(22)
Here, R is the universal gas constant, kB is Boltzmanns constant, and the

QMMM coupling term, EQM-MM , is dened as in Eq. 8. aMM designates the
canonical ensemble average obtained by sampling over only the MM degrees
of freedom in presence of the QM atoms xed in state a; that is, the forces
governing the dynamics are those of state a. In each MD step, that is, for
each conguration of the MM atoms RMM (t), the
coupling energy is

QMMM
b
evaluated once with
the QM atoms in state b RQM and once with the QM
atoms in state a RaQM . Although the free-energy contribution due to the

MM degrees of freedom is incorporated in Aenv , the MM energy, EMM , is not
explicit in Epert ; EMM (RMM ) does not depend on the QM coordinates and
therefore cancels out.
The sampling could also be performed by MC, rather than MD, methods,
which is, however, less common in practice. If one samples from the (isobaricisothermal) NPT, instead of the (canonical) NVT, ensemble, one obtains
a Gibbs, rather than a Helmholtz, free energy. The difference is negligible for
most practical purposes.
To calculate EQM-MM and the forces on the MM atoms, Yang and coworkers proposed use of an ESP charge model of the QM density, as they do
in their QM/MM optimization scheme (see Sect. 4.2.2) [273]. This avoids the
costly evaluation of the electrostatic QMMM interaction at the QM level, but
approximates the continuous density by point charges and implies a frozen
density (i.e., the density cannot adapt to the changing MM environment). We
have recently conrmed that the use of the ESP charge model is a justied
approximation and that the QM/MM FEP approach yields results comparable to thermodynamic integration or umbrella sampling at a fraction of the
cost [321]. The most time-consuming parts of the procedure are the optimizations to dene the reaction path.
The reaction path used in QM/MM FEP has usually been dened in
terms of a preconceived reaction coordinate. However, one could also use
a minimum-energy path (MEP) determined by a reaction-path optimization
method (see Sect. 4.2.3).
QM/MM FEP (sometimes also called QM/MM-FE) is related to older techniques, with which it shares the common feature that sampling of the QM
degrees of freedom is avoided. In the mid-1980s, Jorgensen and co-workers
proposed a scheme for calculating the reaction free energy in solution along
a reaction coordinate dened by a series of structures optimized at the (ab
217
initio) QM level in the gas phase [322, 323]. They then derived classical potential terms describing the solutesolvent interactions at each point of the path
and calculated the free energy along the reaction coordinate by MC umbrella
sampling. Similar techniques were used for reactions in enzymes by Merz
and co-workers (thermodynamic integration with MD sampling) [324] and
Kollman and co-workers (FEP with MD sampling) [325]. In all these QM freeenergy (QM-FE) approaches, the reaction path is determined in the gas phase
and the QM and MM potentials are not coupled. This amounts to assuming
that the inuence of the environment on the reaction path as well as polarization effects are negligible; both approximations are absent in QM/MM
FEP.
To calculate reaction and activation free energies in enzymes, Warshel
and co-workers [326329] developed a scheme in which the sampling is
performed on an EVB reference potential tted to ab initio data. A linearresponse approximation is then applied to evaluate the free energy of transfer
from the EVB to the ab initio surface.
Different formulations have also been proposed for solving the problems
that arise when performing alchemical FEP simulations with a QM/MM
potential [330333].
4.3.3
Thermodynamic Integration
In standard thermodynamic-integration (TI) simulations, the system is taken
along a reaction coordinate , which parameterizes the Hamiltonian, and the
mean generalized force acting on this coordinate is integrated to give the freeenergy difference:
b
ab
=
a
dA()
=
d
d
b
a
H ()
d

,
(23)
where denotes the canonical ensemble average at xed . In the pointwise

TI approach, one constrains to a series of values, runs NVT MD simulations to sample the average force at each location, and integrates these average
forces numerically. (Note that if is coordinate-dependent, i.e., a structural
or geometric constraint, the correct general expression for H ()/ must
correct for the fact that in a constrained simulation the conjugate momentum
of is also constrained [334342].)
Alternatively, one can move the reaction coordinate continuously from a to
b over a time interval :

H ()
ab
A
,
(24)
= dt
218
H.M. Senn W. Thiel
with = /t. If a constraint is used, this is commonly referred to as slowgrowth or adiabatic switching TI. If one uses a guiding potential, that is,
a moving restraint, it is known as steered MD (SMD) [343, 344]. A crucial
assumption underlying all these methods is that the system is always in equilibrium. The switching process from a to b is assumed to be reversible, that is,
innitely slow.
In real simulations, however, is changed at a nite rate. It then follows
from the second law of thermodynamics that the average work W expended
in switching from a to b in a nite time is an upper bound to the free energy,
Aab W ; Aab = W only for .
However, Jarzynski discovered a remarkable equality [345350, 743], valid
under very general conditions, that connects the irreversible work for arbitrary switching times to the equilibrium free energy:

(25)
exp{ A/(kB T)} = exp{ W /(kB T)} 0 ,
where 0 denotes an average of W values over initial states taken from an
equilibrium (canonical) ensemble.
This relation has become the foundation for fast-growth [351] or
multiple-steering [349, 352, 353] simulations, in which irreversible-work
values from deliberately non-adiabatic switching processes are accumulated
and exponentially averaged according to Eq. 25; it has also been applied in the
analysis of single-molecule pulling experiments [354357]. Jarzynskis equality immediately raises questions about the dependence of the free-energy
error on the switching time and about the efciency of performing many fast
switches versus fewer slow ones. These issues have been amply discussed in
the literature, comparing fast-growth to other free-energy simulations [358
368]. Applications of the fast-growth method in the context of enzymatic
QM(DFT)/MM simulations are reported in [369, 744, 755].
4.3.4
Transition-Path Sampling
Developed by Chandler and collaborators [370380], transition-path sampling (TPS) uses Monte Carlo importance sampling in the space of trajectories connecting the reactant with the product basin and yields an ensemble
of reactive trajectories (see the contribution by C. Dellago and P.G. Bolhuis in
this volume). It does not involve any preconceived knowledge about the reaction path or the transition state. All that is required are low-dimensional
order parameters (e.g., combinations of bond lengths) that are able to unambiguously separate reactant from product congurations. TPS has recently
been applied to an enzymatic reaction [381]; semi-empirical QM/MM MD
was used to generate the trajectories.
219
4.3.5
Metadynamics
The metadynamics approach by Laio, Parrinello, and co-workers [382390] is
capable of exploring the free-energy surface without prior information about
the location of minima or transition states. It is related to earlier techniques
such as the local-elevation method [391], conformational ooding [392, 393],
or the at-histogram method for MC [394396]. Metadynamics follows the
dynamics in the space of a set of collective coordinates that characterize the
process of interest, and is coupled to the real dynamics of the system via
a history-dependent bias potential. Controlled by the metadynamics, the bias
potential drives the system away from regions of the free-energy surface it has
already visited by lling them up. The bias potential is constructed from
Gaussians of a specied width that are deposited in the space of the collective
coordinates at a chosen frequency. Once the free-energy landscape has been
lled to a certain level (related to the temperature) with Gaussians, their sum
represents a cast of the free-energy surface, with all minima and transition regions. This technique has recently been applied to an enzyme reaction within
CarParrinello QM/MM MD [397].
4.3.6
Adiabatic Dynamics
The phase-space sampling can be enhanced in constant-temperature MD by
coupling selected degrees of freedom to a separate thermostat and keeping them at a higher temperature than the remainder of the system. To
minimize the heat ow from the hot degrees of freedom into the cooler
surroundings, the masses of the hot atoms are scaled up, thus creating an
adiabatic separation between the hot, slow and the cool, fast degrees of freedom. Such techniques have been presented by Tuckerman and co-workers
(adiabatic free-energy dynamics, AFED) [398, 399] and VandeVondele and
Rthlisberger (canonical adiabatic free-energy sampling, CAFES) [400]. The
dynamics of the slow degrees of freedom is effectively performed on the
free-energy surface generated by the environment. CAFES has been used in
a CarParrinello QM/MM MD simulation [401].
4.3.7
QM/MM Reaction-Path Potential
Lu and Yang [402] extended the idea of the reaction-path Hamiltonian by
Miller et al. [403] to large systems described at the QM/MM level. Using the
energies, vibrational frequencies, and electronic response properties of the
QM region along a QM/MM reaction path (e.g., the minimum-energy path),
they constructed a harmonic reaction-path potential. It provides an analytical
220
H.M. Senn W. Thiel
expression for the QM/MM potential energy along the path, accounting for
the coupled dynamics of QM and MM parts. This potential can subsequently
be used for sampling, for instance, to perform thermodynamic integration
with constrained MD.
5
Practical Aspects of Biomolecular Reaction Modelling
When performing a QM/MM study of a biomolecular reaction, particularly
an enzymatic reaction, one generally has to invest a considerable amount of
work into the setup and preparation of the system prior to the actual QM/MM
calculations. We provide here a brief list of possible issues that need to be
addressed in this process. Many of them are not exclusive to the QM/MM
approach, but arise also in classical MD studies:
Developing MM parameters: As it is useful to be able to perform classical MD simulations during the preparation stage (as described below),
one needs a set of MM parameters for the entire system, including the
prospective QM part. While one can restrain, e.g., substrates or cofactors to their experimental position, thus avoiding the need for accurate
bonded parameters, it is highly desirable to develop at least reasonable
non-bonded parameters, in particular atomic charges. They largely control the arrangement of polar or charged residues and water molecules
around these moieties.
Different biomolecular force elds follow different procedures to derive
partial charges, which may involve QM calculations on the molecule in question or representative fragments thereof. The effort required to obtain or
generate MM parameters can be a decisive factor for the choice of force eld.
Structure validation: The starting point of a biomolecular QM/MM study
is almost always an experimental structure, most commonly determined
by single-crystal X-ray diffraction. To check the structure for inconsistencies [404], it is generally advisable to use one or several of the existing
validation tools, for example what check [405, 719]; a web interface to
several such tools is available [720]. One should also bear in mind that
the structure is usually least reliable in and around the active site, where
non-protein molecules (substrates, cofactors) are present [246].
It is often not possible to distinguish between C, N, and O based on crystallographic data alone. A common error are therefore ipped side-chain
amide groups of asparagine and glutamine as well as ipped histidine
imidazole rings. These are detected by, e.g., what check [405, 719] and
Reduce [406, 407].
Besides the technical correctness of a specic structure, there is the
more fundamental question of the sensitivity of the computational results
221
to the initial crystal structure [408]. Structures of the same enzyme family or of the same enzyme crystallized with different inhibitors or under
different conditions can differ in subtle details, which may, however, impact the results. There is no guarantee that the crystal structure actually
corresponds to the productive form of the enzyme.
Adding hydrogen atoms: The positions of the hydrogen atoms are rarely
resolved in X-ray structures, and generally no hydrogen coordinates are
provided. For the simulations, certainly all hydrogen atoms in the QM
region as well as the polar and aromatic MM hydrogen atoms need to
be added; if an all-atom force eld is used, all hydrogen positions are
required. For aliphatic and aromatic hydrogen atoms, this is a routine
procedure; most biomolecular simulation packages are capable of reliably
placing these hydrogen atoms according to standard bond distances and
angles.
The situation is less clear for polar hydrogen atoms, especially if they cap
a rotatable bond, e.g., the serine side-chain hydroxy group. The orientation of these bonds, and hence the position of the corresponding hydrogen
atoms, is controlled by the local hydrogen-bond network as well as steric
factors. Several programs and algorithms are available (either as part of
a larger modelling package or as stand-alone programs) that attempt to
determine the optimum position for these hydrogen atoms. Examples include charmm [180, 704], what if [409, 721], or Reduce [406, 407]; an
assessment of such procedures has recently appeared [410].
Most difcult are the acidic hydrogen atoms, whose presence or absence
is ambiguous. Depending on the local pKa value, ionizable groups can be
protonated or not. Most prominent are the imidazole side chain of histidine and the side-chain carboxylates of aspartate and glutamate, but it
can also be necessary to consider the phenolic hydroxy group of tyrosine,
the thiol of cysteine, as well as the lysine ammonium and the arginine
guanidino group. The protonation state controls the availability of acidic
or basic sites, and it directly inuences the charge and thus the long-range
electrostatic eld. The decision where to protonate is therefore a rather
delicate one, which may well have mechanistic implications.
The most rigorous solution, calculating the local pKa values using
PoissonBoltzmann or even QM/MM methods [331, 411], is not routinely
used because of the computational effort involved. The other extreme,
namely to choose the protonation states of the free amino acids in water at
pH 7, is not satisfactory. A compromise are methods that assign the protonation state based on the local hydrogen-bonding network and structural
considerations; for instance what if [409, 721] or Reduce [406, 407]. An
alternative is the calculation of local pKa values with a fast increment
method, such as propka [412, 722]. One should bear in mind that these
empirical approaches neglect the non-protein part of the enzyme and are
therefore less reliable in and around the active site.
222
H.M. Senn W. Thiel
The protonation states depend of course on the pH chosen for the simulation. It should be noted that the optimal pH, where the enzyme activity is
highest, is often quite different from the pH at which the structure was determined because a pH close to the isoelectric point tends to be favourable
for crystallization. The pH for kinetic measurements, from which, e.g.,
rate constants or activation barriers are derived, can again differ from the
optimum pH.
Through the hydrogen-bonding network, the assignment of protonation
states is linked to the amide and histidine ips (vide supra) and to the
hydration of the structure (vide infra). These three steps are therefore interdependent and may require an iterative treatment. One should also note
that the results of different empirical algorithms, e.g., to detect ipped
residues or to assign protonation states, may well be at variance. In these
cases, the only practical solution is to visually inspect the structure and to
take a decision based on chemical insight.
Choice of simulation system: Apart from the denition of the QM region,
which should incorporate the electronically active part of the structure,
one has also to decide on how much of the environment is to be included
in the simulation and in which way. The cleanest way is certainly to include the whole protein plus a hydration shell (see below). The next choice
is then the size of the active region, where the atoms are free to move during optimization or MD, while the remaining parts are xed or positionally
restrained. In MD simulations, the number of degrees of freedom is not
critical. Optimizations, however, tend to become problematic if more than
a few thousand (about 2000) atoms are included. Not only can convergence
become an issue, but it is also more difcult to ensure that all the many environment degrees of freedom remain consistent among changing active
site congurations, e.g., along a reaction path. For instance, if a hydrogen
bond in the environment rearranges from one step to the next, the continuity of the energy prole is broken. A reasonable selection for the active
region therefore includes all residues within, e.g., 10
A of the QM part.
Solvation/hydration: Only water molecules occupying a well-dened position are crystallographically resolved; mobile ones are not visible. At least
those parts of the system that are allowed to move during the MD or in
optimizations (plus a buffer region around them) must therefore be hydrated with additional water molecules. There are two main options: one
can either place the whole system in a box of water under periodic boundary conditions; or one superimposes a sufciently sized sphere of water
molecules (cut out from an MD simulation of liquid water) onto the active
region of the structure. In the latter case, a spherical boundary potential
prevents the water molecules from diffusing away and simulates the interface to the bulk. In either procedure, water molecules too close to an
existing atom are deleted, and the remaining ones are subjected to repeated minimization and MD runs.
223
This protocol should ensure that the surface of the protein as well as
cavities and channels connected to the surface are adequately hydrated.
However, there may exist small internal cavities without connection to the
surface that cannot be hydrated in this manner. Apart from inserting water molecules manually, one can use programs like dowser [413, 723] that
identify such cavities and attempt to determine whether water molecules
should be placed there.
Classical MD simulations: Once the system setup has been completed, it
is generally advisable to run classical MD simulations, either of the whole
system (under periodic boundary conditions) or of a suitably chosen active region. The purpose of these MD runs is twofold: (i) By gradually
releasing the positional restraints applied to the protein part during the
hydration phase and performing free MD, one can validate the setup.
Signicant differences between the X-ray structure and the average positions hint at problems with the protonation or hydration procedures.
(ii) Snapshots from the MD simulation can serve as initial structures for
subsequent optimizations.
At this stage, one can introduce or modify moieties not present in the experimental structure. For instance, an inhibitor bound in the active site is
replaced by the substrate; a cofactor is changed into its chemically active
form; or an amino acid is exchanged to generate a mutant. Depending on the
manipulations, it may be appropriate to check the protonation states and/or
rehydrate and to let the system adapt to the changes with another MD run.
Counter ions, charge neutralization: There is no general consensus on
whether it is necessary or desirable to neutralize the total charge of the
system by the addition of counter ions. For charged surface residues (e.g.,
carboxylate or ammonium side chains), there will always be a corresponding counter ion (typically sodium or chloride, respectively) nearby in
solution. One can thus argue that counter ions should also be present in
the simulation to balance and screen these charges. It has also been suggested that they be neutralized by (de)protonation, especially if they are
xed and outside of the hydration sphere.
Even if one chooses to compensate for the surface charges, there normally
remains a net charge due to buried charged groups. To produce an overall
neutral system, one may add additional counter ions in the hydrated part
(raising the question of where exactly to place them) or remove some of
the surface counter ions added previously (raising the question of which
ones to eliminate).
Because of these ambiguities, a common choice is to leave the total charge
as it is after assigning the protonation states according to the chosen pH.
Exceptions are simulations of the relative stability of differently charged
QM regions (e.g., redox processes). In this case, the environment should
be neutral as a non-zero net charge would create an electrostatic eld, thus
articially stabilizing one charge state over the other [330, 414, 415].
224
H.M. Senn W. Thiel
6
Interpreting the Results: Understanding Enzyme Catalysis
The results from QM/MM calculations and simulations (potential and freeenergy proles, structures, dynamics, etc.) are, of course, on the rst level
amenable to all the same interpretations as those obtained with any other
potential-energy method. However, because of the inherent separation between the QM reactive part and the MM environment, and the possibility of easily dissecting the total energy according to various decomposition
schemes (e.g., QM/MM/QMMM, electrostatic/steric, by residue), QM/MM
methods lend themselves to investigations into the sources of the catalytic
power of biocatalysts. Simulations and theoretical tools provide the means for
a detailed understanding of how enzymes achieve their sometimes spectacular rate enhancements, in particular of the factors contributing to the barrier
lowering and the role of enzyme structure and its exibility.
Since the proposal by Linus Pauling [416] that enzymes work by binding
the transition state more strongly than the substrate, thus lowering the activation barrier, a variety of concepts and ideas have been advanced to explain
the origin of enzymatic catalysis; we refer to several recent reviews on this
topic [233, 234, 417421, 745750], including a thematic issue of Chemical Reviews [751]. (Note that these articles usually deal not only with interpretive
concepts but also with computational methods for the investigation of enzyme catalysis, including, but not restricted to, QM/MM approaches. They
thus complement the reviews cited in Sect. 1.)
No single model should be expected to encompass in its entirety a phenomenon as complex as enzyme catalysis. Some proposals are more qualitative notions rather than quantitative models, and many are subject to critical,
sometimes controversial, discussion in the literature. One difculty is the
loose denition of certain terms and concepts. To mention just one basic example, there is no general agreement on the reference system to which the
enzymatic reaction should be compared and, therefore, how precisely the
rate acceleration due to the enzyme (its catalytic effect) is to be dened. As
advocated by Warshel [746], the emost natural choice seems to be the corresponding reaction in water that proceeds by the same mechanism. We give
here a brief, partial list of catalytic models and concepts:
Transition-state stabilization: The environment of the active site is organized such that the TS is preferentially stabilized. In particular, hydrogenbonding interactions and the electrostatic eld in the active site are considered [233, 234, 417, 746].
Substrate destabilization: Under this category, a number of effects related
to the preparation or preorganization of the substrate can be subsumed;
including steric connement [422426], the entatic state [427430],
induced t [431], orbital steering [432, 433], the spatial-temporal hy-
225
pothesis [434], cratic free energy [435], and near-attack conformations

(NAC) [436447].
Desolvation: The substrate is less well solvated in the active site than in
solution [448, 449].
Entropic effects: The motion of the substrate is restricted in the active site
(entropic trap) [325, 435, 450452].
Low-barrier hydrogen bonds (LBHB): Special hydrogen bonds with a signicant covalent contribution stabilize the TS [453458].
Dynamic effects: Specic protein motions or vibrations couple to the substrate to lower the barrier [417, 459463, 748].
Nuclear quantum effects: The enzyme increases the transmission coefcient through vibrationally enhanced tunnelling and other nuclear quantum effects [419, 421, 463467, 745, 750].
Chemical effects: All chemical factors in which the reaction in solution and
in the enzyme differ can be collected under this heading; for instance,
changes in the mechanism or the participation of a generalized acid or
base, of a cofactor, or a metal centre. The covalent hypothesis proposes
that these effects are at work in the most procient enzymes [468]. One
may argue, however, that they are not catalytic effects proper and should
be considered separately.
From the current literature, electrostatic transition-state stabilization appears
to emerge as the dominant factor in the actual chemical step of enzymatically catalysed transformations. The electrostatic basis of enzyme catalysis
has been stressed in a recent review that evaluates a number of concepts and
proposals in great detail [746].
Besides the plethora of catalytic proposals, one can more generally characterize the role of the enzyme and the microscopic steps involved in an
enzymatic reaction [24]:
The enzyme is prepared for the reaction. This can include changes in the
tertiary or quaternary structure, binding a cofactor, and possibly activating the cofactor. (The latter constitutes a, possibly catalytic, reaction in
its own right but we consider it here as a preparatory step preceding the
reaction step under study.)
The substrate binds to the enzyme. While this may be a simple association, it can also entail a more complicated mechanism, involving enzyme
structural changes or guiding of the substrate to the active site by specic
interactions. In this case, the binding step can contribute to the selectivity
of the enzymatic reaction.
The substrate is prepared in the active site, both chemically and structurally. Chemical preparation involves, e.g., (de)protonating the substrate
to activate it towards the reaction. Structural preparation (or preorganization) includes both enthalpic (adopting a specic conformation different
from the one in solution) and entropic (loss of translational and rotational
degrees of freedom) aspects.
226
H.M. Senn W. Thiel
The actual chemical reaction takes place in the active site. The chemical
step is in the focus of most of the models listed above that attempt to
identify and explain the specic catalytic effect of the enzyme.
We also mention some analysis procedures that have proven useful in obtaining insight into enzymatic reactions:
Energy decomposition by residue: To probe the effect of a selected MM
residue on an energy difference (e.g., the activation barrier), certain interactions of this residue are switched off, and the energy difference is
re-evaluated. Typically, this procedure is applied to the electrostatic interaction by deleting the residues point charges. A common way is to remove
the electrostatic interactions sequentially residue-by-residue in order of
decreasing distance from the active site and to plot the energy difference
as a function of the distance. Also known as perturbation analysis, this
approach was established early on [469] and has remained popular since;
see, e.g., [310, 470474].
Decomposition of the electrostatic energy: The electrostatic QMMM interaction energy can be decomposed into the permanent interaction energy
(the interaction between the environment and the unpolarized QM density) and the polarization energy. The latter can be further divided into
the polarization-stabilization and the electronic-distortion terms. Such
a dissection gives insight into the role of electronic polarization; see,
e.g., [5, 294, 475].
Differential transition-state stabilization (DTSS): This analysis focuses on
the specic barrier-lowering effect of the enzyme relative to the gas phase,
considering the substrate(s) in their active-site reactant and TS structures [476, 477]. The DTSS can be further decomposed into long-range
multipole, short-range penetration, exchange, delocalization, and correlation components; the multipole and penetration terms together form the
electrostatic contribution. The procedure can be expanded by performing
the analysis on a per-residue basis.
General energy decompositions: A generic energy decomposition scheme
can be applied to dissect the contributions related to changing the environment from a reference (typically aqueous solution) to the enzyme [301,
420, 478]. For the substrates, the process of transferring them from solution into the enzyme can be subdivided into three steps: (i) The free reactants are individually distorted in solution from their minimum structure
into the conformation that they adopt in the binding site. This preparation
energy is strictly positive. (ii) The distorted reactants, still in solution,
are brought into the relative position and orientation they adopt in the
enzyme; this complexation energy can be either stabilizing or destabilizing. (iii) The distorted and oriented substrates are placed into the active
site. The corresponding energy is the interaction energy of the prepared
substrate complex.
227
The analogous procedure can be applied to the transition state, affording

a decomposition of the activation barrier. The energy difference in solution between the prepared substrate complex and the enzyme-optimized
TS structure is compared to the full QM/MM barrier, providing a measure
for the specic catalytic effect of the enzyme.
An alternative approach follows the inherent partitioning of the QM/MM
energy. Any energy difference, such as the activation barrier, can be dissected into QM, QMMM, and pure MM contributions. The QMMM
component is further split into QMMM electrostatic, QMMM van der
Waals, and QMMM bonded terms. The MM component may be likewise divided into MM electrostatic, MM van der Waals, and MM bonded
terms.
7
Survey of Biomolecular QM/MM Studies
Since the pioneering work of Warshel and Levitt in 1976 [1], QM/MM
methods have become increasingly popular in the area of biomolecular modelling. The large majority of applications have been concerned with enzymatic reactions. Obviously, QM/MM approaches are the method of choice
for the investigation of chemical transformations, while MM methods serve
well in purely structural studies. The focus on enzymes is certainly due to
the ubiquity and importance of these biocatalysts, but also to the relatively
wide availability of experimental, especially structural and kinetic, data and
of specialized force elds (which is, of course, in turn related to the relevance
of enzymes). Other biomolecules (nucleic acids, carbohydrates, lipids) have
found comparably less attention.
We provide here a survey of biomolecular QM/MM studies in tabular form,
considering contributions that have appeared between 2000 and April 2006
(see Tables 19). References to earlier work can be found in the reviews cited
in Sect. 1. The number of publications has been steadily increasing since the
mid-1990s, but especially from 2003 onwards. The list is extensive, but certainly not exhaustive. Apart from inadvertent omissions, a contribution is
likely to be missing if none of the pertinent keywords (QM/MM, combined
quantum mechanics/molecular mechanics, etc.) is present in the title or abstract. The entries are grouped by the type of biomolecule investigated; the
sorting of enzymes follows the IUBMB (International Union of Biochemistry and Molecular Biology) enzyme classication (EC nomenclature [724]).
Note that the reaction investigated does not necessarily correspond to the enzymes main function, which determines its classication. To make the tables
less congested, the entries in the columns biomolecule, process studied,
QM level, and MM level are not repeated in successive rows for the same
Process studied
OH transfer
Cholesterol oxidase
Liver alcohol
dehydrogenase
AM1
AM1, HF
AM1
AMBER
CHARMM
CHARMM
CHARMM
AMBER
CHARMM
CHARMM
GROMOS
HF, DFT, MP2

AM1
AM1
CHARMM
MM level
AM1
QM level
SCC-DFTB,
DFT, HF
Electronic excitations CIS
FAD oxidation
SCC-DFTB
FAD reduction
H+ , H transfer
H transfer
Acyl-CoA dehydrogenase H+ , H transfer
Phenol hydroxylase
Flavin-dependent oxidoreductases
p-Hydroxybenzoate
OH transfer
hydroxylase
Biomolecule
Table 1 Oxidoreductases (EC 1), O2 - and electron-transport proteins
[486]
[487]
[330]
[472]
[484]
[485]
[480]
[301]
[321]
[479]
Refs.
Comparison of free-energy
methods
Inuence of substrate
[481]
protonation state
Inuence of substrate deproto- [482]
nation, substituted substrates
VTST with tunnelling
[483]
Inuence of substituted substrates, avin cofactors
Comments
US(MD), rate
constants, KIE
Rate constants, KIE VTST with tunnelling
Opt, reaction paths, VTST with tunnelling
rate constants, KIE
Opt, US(MD)
Inuence of specic residues,
mutations
Single points
FEP(MD)
Redox potentials
FEP
Redox potential from dual-topology-single-coordinate
FEP
Opt
Opt
TI(MD)
Opt, QM/MM FEP,
US(MD)
Opt, TI(MD)
Opt
Calculation type
228
H.M. Senn W. Thiel
Process studied
QM level
Substrate binding,
deprotonation
Dihydrofolate reductase
H+ transfer
H transfer
Inhibitor binding
H+ , H transfer
Lactate dehydrogenase
MM level
AMBER,
Tripos
AMBER
AMBER
AM1
PM3
AMBER
CHARMM
AM1, PM3
AM1
CHARMM
PM3
AM1
PM3
Opt
Constrained MD
Opt
FEP(MD)
MD
US(MD), rate
constants, KIE
Opt
TI(MD)
Opt
FEP(MD)
Opt
FEP(MD)
[494]
[495]
[493]
[490]
[381]
[491]
[492]
[488]
[489]
Refs.
Inuence of mutations
QM alchemical perturbation
[501]
[502]
[503]
Comparison to semi-empirical all-QM [496]

[497]
Comparison to semi-empirical all-QM [498]
Overlapping mechanical embedding
[499]
Analysis of substrate polarization
[475]
[500]
Inuence of titratable residues

Corrections for AM1, adjusted
vdW parameters
Comparison to ab initio/continuum
solvent; role of LBHB
Calculation type Comments
OPLS-AA US(MD)
CHARMM, Opt, US(MD)
OPLS-AA
AM1
CHARMM Opt
TPS
AM1, MP2
Opt
AM1
Opt
NAD(P)-dependent oxidoreductases
Class 3 aldehyde
H+ transfer
PM3
dehydrogenase
Thioacetal formation
Biomolecule
Table 1 continued

229
Process studied
Non-haem-iron oxidoreductases
Methane
Structure
monooxygenase
O2 binding
O2 activation,
OH transfer
H abstraction
Phe hydroxylase
Hydroxylation
Ribonucleotide reductase Structure
HIF-1 asparaginyl
O2 activation
hydroxylase
Haem-dependent oxidoreductases
Cytochrome c peroxidase Electronic structure
of ferryl intermediate
Acetohydroxy acid
Alkyl migration
isomeroreductase
TTQ-dependent oxidoreductases
Methylamine
H+ transfer
dehydrogenase
Biomolecule
Table 1 continued
CHARMM
DFT
Opt
Opt, KIE
AMBER
Opt
AMBER
Opt
EFP (AMBER) Opt
OPLS-AA
Opt
Opt
Opt
KIE
Opt, rate constants, KIE
Opt, rate constants, KIE
Opt, reaction paths,
rate constants, KIE
US(MD)
Calculation type
DFT
DFT
CASSCF
DFT
AMBER
CHARMM
AMBER
PM3
PM3, HF
PM3
OPLS
AM1
QM level MM level
Electronic structure
Different substrates
Comparison of mechanisms

Comments
[515]
[512]
[513]
[509]
[514]
[509]
[510]
[511]
[505]
[506]
[507]
[508]
[504]
Refs.
230
H.M. Senn W. Thiel
Spin states, structure of haem

Substrate binding
Characterization of ferryl
intermediate
Cytochrome P450
Electronic structure of intermediate,

Mssbauer parameters
Product release
Substrate binding, O2 activation,
hydroxylation
Hydroxylation
Ascorbate peroxidase
Electronic structure of ferryl
intermediate
Horseradish peroxidase Electronic structure of ferryl
intermediate
Haem oxygenase
Active-site structure
EPR, Mssbauer parameters

of ferryl intermediate
Formation of ferryl intermediate
H abstraction, OH transfer
H abstraction
Process studied
Biomolecule
Table 1 continued
Calculation type
CHARMM Opt
AMBER
DFT
DFT
Opt
[526]
[515]
CHARMM Opt
CHARMM Opt
DFT
[528]
[527]
[524]
[525]
OPLS-AA
[752]
[522]
[523]
[520]
[521]
[516]
[517]
[518]
[519]
Refs.
Opt
Opt
Review
Comparison between
human isoforms
Inuence of mutation
Comments
[753]
[415]
Opt
Opt
CHARMM Opt
AMBER
Opt
CHARMM Opt
Opt
MM level
Opt
Opt
CHARMM, Opt
OPLS-AA
CHARMM Opt
CHARMM Opt
DFT
QM level

231
DFT
DFT
DFT
DFT
NO oxidation to NO3
O2 binding
O2 binding
with bound CO
Vibrational analysis
with bound CO
Ligand binding
CO binding
CO dissociation
V-, Cu-dependent oxidoreductases

V haloperoxidase
Active-site protonation DFT
state, formation of hydroperoxo intermediate
V-dependent chloroperoxidase Active-site structure
DFT
Other haem proteins

Truncated haemoglobin N
(trHbN)
Various haem proteins
Haemerythrin
Myoglobin
DFT
HF, DFT
DFT
H+ transfer
Form of NHA intermediate
NO binding
Nitrite reductase
NO synthase
QM level
Process studied
Biomolecule
Table 1 continued
Opt
CP-MD
Opt, MECP
Opt
Opt
AMBER
CHARMM
CHARMM
CP-MD,
metadynamics
Opt
AMBER
OPLS-AA
OPLS-AA
Opt
Opt
CP-MD
Opt
Opt
Opt
Opt
Opt
Calculation type
AMBER
OPLS-AA
CHARMM
AMBER
AMBER
AMBER
MM level
Protonation state of
active site
Non-adiabatic TST
Analysis of protein
structural changes
Inuence of mutations,
haem oxidation state
Comments
[542]
[397]
[541]
[538]
[539]
[540]
[537]
[534]
[511, 535]
[536]
[533]
[532]
[529]
[530]
[531]
Refs.
232
H.M. Senn W. Thiel
Oxidation, protonation,
dehydrogenation
Galactose oxidase
Structure
EPR parameters
Redox potential
Active-site structure,
redox potentials
Plastocyanin
Other oxidoreductases
Mammalian glutathione Active-site structure
peroxidase
Other proteins
Thioredoxin
Disulde reduction
Granulocyte colonyMet oxidation by H2 O2
stimulating factor
Blue copper proteins
Due ferri 1
Other metalloproteins
Rubredoxin
H abstraction
O2 activation
Multicopper oxidase
Dopamine -monooxygenase
Process studied
Biomolecule
Table 1 continued
HF, DFT, MP2 UFF

DFT
CHARMM
Opt
Opt, TI(MD)
Opt
AMBER
HF, DFT
CP-MD
Opt
Opt
Opt
CP-MD
Opt
Refs.
Zn analogue of a
mimetic Fe protein
Inuence of mutations
Comparison of different
Cuoxo species
[553]
[554]
[552]
[549]
[550]
[551]
[548]
[547]
[546]
[545]
Electronic structure of various

[543]
protonation and oxidation states
Comparison to mimetic
[544]
DFT
AMBER
AMBER
AMBER
AMBER
CHARMM
AMBER
MM level
Opt
Single points
UFF, AMBER Opt
DFT
DFT
DFT
DFT
DFT
DFT
QM level

233
DFT, MP2
DFT
Serine phosphorylation
AMBER
CHARMM
DFT
DFT
DFT, MP2
AMBER
OPLS-AA
CHARMM
AMBER
HF, DFT
CHARMM
AM1
CHARMM
Methyl transfer
Histone lysine methyltransferase SET7/9

Human purine nucleoside
phosphorylase
Protein kinase, thymidine
phosphorylase
Pyrimidine nucleoside
phosphorylase
cAMP-dependent protein
kinase
CHARMM
AM1
MM level
AM1
Nucleophilic epoxide
opening
Methyl transfer
Glutathione S-transferase
Catechol O-methyltransferase
QM level
Vibrations in bound substrate, intermediate

Phosphorylation
Process studied
Biomolecule
Table 2 Transferases (EC 2)
Opt
Opt
Opt
Opt
Opt, MD
Opt, US(MD)
Opt, KIE
Opt, US(MD)
Opt, MD
QM/MM FEP
Opt, QM/MM FEP
US(MD)
Calculation type
Effect of enzyme
phosphorylation
Evaluation of
semi-empirical QM
Corrections to PMF
Comments
[474]
[563]
[562]
[561]
[560]
[478]
[556]
[557]
[558]
[317, 318]
[559]
[555]
Refs.
234
H.M. Senn W. Thiel
CHARMM
CHARMM
HF, MP2
INDO/S-CIS
DFT
Fluorinase
CHARMM
OPLS-AA
AM1
AM1, HF, DFT
H+ transfer, phosphorylation
H+ transfer,
phosphoryl transfer
Enolization
Fluorescence quantum yield
SN 2 displacement
HIV-1 reverse transcriptase

Hypoxanthine-guanine-xanthine
phosphoribosyltransferase
Citrate synthase
MM level
QM level
Process studied
Biomolecule
Table 2 continued
Opt
MD
Opt
Opt, US(MD)
Opt, US(MD)
Calculation type
[566]
[567]
[568]
[564]
[565]
Comments Refs.

235
Process studied
Structure of tetrahedral
intermediate
Peptide hydrolysis
Inhibitor binding
-Lytic protease
Thermolysin
-Trypsin
Subtilisin
-Chymotrypsin
SN 2 attack on reactive
inhibitor
Inhibitor binding,
aziridinium formation
Peptidases and other amidases

Aminopeptidase
Peptide hydrolysis
Leu aminopeptidase
Peptide hydrolysis
Met aminopeptidase
Active-site protonation
states
Carboxypeptidase A
Inhibitor binding
Biomolecule
Table 3 Hydrolases (EC 3)
AM1
AM1
DFT
PM3
DFT
AM1, DFT
AMBER
AMBER,
Tripos
EFP
AMBER
EFP
AMBER,
Tripos
AMBER
AM1
HF
Tripos
Tripos
AMBER
MM level
AM1
AM1
DFT
QM level
Opt, NMR shifts,

H/D fractionation
Opt
Opt
Opt, NMR shifts,

H/D fractionation
Single point
Opt, NMR shifts
Opt
Opt
Opt
Opt
Opt
CP-MD
Calculation type
Structural and spectroscopic

properties of H-bonds

Comments
[577]
[495]
[575]
[576]
[575]
[574]
[573]
[572]
[495]
[569]
[570]
[571]
Refs.
236
H.M. Senn W. Thiel
Caspases
Histone-deacetylase-like
protein
Peptide deformylase
DFT
DFT
DFT
Thioester hydrolysis
Amide hydrolysis
N-Peptidyl formamide
hydrolysis
AM1
Inhibitor binding
OPLS-AA
Opt
CP-MD, TI(MD),
KIE
AMBER, Tripos Opt
CHARMM
MD
GROMOS
CP-MD
AMBER
Opt
AMBER
US(MD)
Structure, protonation
state of bound inhibitors
[591]
[495]
[588]
[589]
[590]
[401]
[582]
[579]
[580]
[581]
[578]
Refs.
[586, 587]
Comments
SCC-DFTB CHARMM
Calculation type
[583]
[584]
[485]
[585]
DFT
Deacylation
Substrate binding
MM level
EFP (OPLS-AA, Opt

AMBER)
AMBER
QM/MM FEP
Opt, QM/MM FEP Inuence of mutations
HF
Opt, QM/MM FEP,
NMR parameters
AM1
CHARMM
Opt
QM/MM-PB/SA;
different substrates
HF
CHARMM
MD
HF, DFT
US(MD)
PM3
AMBER
Opt
PM3, DFT
Opt
HF, MP2
QM level
Inhibitor binding
Inhibitor binding,
peptide hydrolysis
Kumamolisin-As (sedolisin, Peptide hydrolysis
serine-carboxyl peptidase)
HIV-1 protease
Peptide hydrolysis
Elastase
Peptide hydrolysis
Trypsin
Acylation
H+ transfer
Process studied
Biomolecule
Table 3 continued

237
PM3
DFT
DFT
AM1
Deacylation
Active-site structure,
dynamics with bound
substrate
Lys carboxylation
Deacylation
Monozinc -lactamase
Monozinc CphA
-lactamase
Dinuclear zinc
metallo--lactamase
OXA-10 -lactamase
Penicillin-binding proteins
Substrate binding
Protonation states
Metallo--lactamase
Zn--lactamase
DFT
SCC-DFTB
DFT
SCC-DFTB
Deacylation
Deacylation
AM1, DFT
HF, MP2
DFT
AM1, DFT
DFT
AM1
PM3
AM1
DFT
QM level
Class C -lactamase
Class A -lactamase
Deacylation
Acylation
Structure, interaction
with bound inhibitors
Acylation
Matrix metalloproteinases
-Lactamase
Process studied
Biomolecule
Table 3 continued
AMBER
OPLS-AA
AMBER
AMBER
CHARMM
AMBER
CHARMM
OPLS-AA
CHARMM
AMBER
CHARMM
CHARMM
AMBER
CHARMM
OPLS-AA
MM level
Opt
Opt
MD
MD
CP-MD
Opt, MD
Opt
Opt
Opt
Opt
Opt
Opt
Opt
Opt
Opt
Opt
Opt
Opt
Calculation type
Comparison to -lactamase
Bound substrate or
acyl intermediate
Structural inuence of
active-site protonation
Comparison to penicillinbinding protein
Part of a docking approach
Comments
[605]
[599]
[606]
[604]
[602]
[603]
[600]
[601]
[593]
[594]
[595]
[596]
[473]
[597]
[598]
[598]
[599]
[592]
Refs.
238
H.M. Senn W. Thiel
Opt
Opt
DFT
AM1, HF, DFT CHARMM
AM1
Ester hydrolysis
Acylation
Oxocarbenium formation, H2 O attack

Oxocarbenium formation, hydrolysis
Substrate binding
DFT
HF, MP2
DFT
HF, DFT
AM1
Structure of bound
substrate
Phosphate hydrolysis
Myosin
RNase A
SCC-DFTB
8-Oxoguanine DNA
glycosylase I
Bacillus 1,31,4-glucanase
Phosphatases
F1 -ATPase
Human butyrylcholinesterase
Lipase
Glycosylases
Uracil-DNA glycosylase
DFT, MP2
MP2
DFT
DFT, MP2
HF,
HF,
HF,
HF,
CHARMM
CHARMM
AMBER
CHARMM
CHARMM
CHARMM
AMBER
OPLS-AA
AMBER, UFF
AMBER
Opt
Opt
Opt
Opt
CP-MD
US(MD)
Opt
Opt
Opt
Opt
Opt
Opt
Opt
Acylation
AMBER
CHARMM
PM3
PM3, DFT
[614]
[613]
[609]
[248]
[610]
[611, 612]
[607]
[608]
Refs.
[617]
[618]
[619]
[620]
[616]
Mixed implicit/explicit
[81]
solvent with charge scaling
Alchemical transformation [615]
Lactam aminolysis
Acylation
MM level
Human serum albumin

Fatty acid amide
hydrolase
Esterases
Acetylcholinesterase
QM level
Process studied
Biomolecule
Table 3 continued

239
Other hydrolases
Soluble epoxide
hydrolase
Adenosine deaminase
Cytidine deaminase
4-ChlorobenzoylCoA dehalogenase
Haloalkane dehalogenase
HF, MP2
QM level
Ester hydrolysis
AM1
PM3
SN 2 displacement
PM3
OPLS-AA
CHARMM
CHARMM
HF, DFT
AMBER
SCC-DFTB CHARMM
PM3
CHARMM
Nucleophilic attack
Inhibitor (de)hydration
SN Ar displacement
AMBER
CHARMM
DFT
AM1
Calculation type
Comments
Opt
US(MD), reaction VTST
rates, KIE
Opt, US(MD)
US(MD),
Reactive-ux calculation
recrossing factor
Opt
Semi-empirical QM with
valence-bond correction
Opt
US(MD)
US(MD)
CP-MD
Opt
CP-MD
EFP (AMBER) Opt
MM level
HF
Vibrational spectrum,
DFT
structure of bound GTP
GTP hydrolysis
Human H-Ras p21ras

GAP-334
Human H-Ras p21ras
Process studied
Biomolecule
Table 3 continued
[313]
[632]
[633]
[309]
[449]
[625, 626]
[627, 628]
[629631]
[624]
[622]
[623]
[621]
Refs.
240
H.M. Senn W. Thiel
Enolization, dehydration
Phosphate elimination
CC bond formation
Decarboxylation
Enolase
Methylglyoxal synthase
Macrophomate synthase
Ornithine decarboxylase
Diol dehydratase
H+ transfer
H abstraction,
OH transfer
Orotidine 5 -monophosphate H+ transfer,

decarboxylase
decarboxylation
H+ transfer
Carbonic anhydrase II
CO2 binding, hydration
Vibrational frequencies
Long-range H+ transfer
Process studied
Biomolecule
Table 4 Lyases (EC 4)

MM level
Calculation type
PM3
DFT
AMBER
MD
Opt
Opt
Opt, MD
MD
Opt
OPLS-AA FEP(MC)
CHARMM US(MD), rate
constants, KIE
AMBER
CP-MD, fast-growth TI
DFT
OPLS-AA
PM3, DFT CHARMM
PM3
SCC-DFTB
DFT
PM3
AM1
HF, DFT
AMBER
QM/MM-FEP
SCC-DFTB CHARMM Opt
QM level
B12 -dependent
Collective
reaction coordinate
Comparison to triosephosphate isomerase

Alternative mechanisms
Comments
[640]
[641]
[637]
[638]
[639]
[91]
[369]
[635]
[636]
[471]
[634]
Refs.

241
H+ transfer
H+ transfers
H+ transfers,
enolization
Glutamate racemase
Triosephosphate
isomerase
Rate constants, KIE
Opt
Rate constants
DFT
AM1
OPLS-AA
CHARMM
Opt
Opt, US(MD)
Opt, MD
Opt
Opt
Opt
Opt, QM/MM FEP

Opt
Reaction-path opt,
QM/MM FEP
Opt
Opt, MD
Opt
Calculation type
SCC-DFTB
CHARMM
CHARMM
CHARMM
CHARMM
AMBER
AM1, DFT
PM3
H+ transfers,
epimerization
Mandelate racemase
AM1
DFT
AM1, DFT
AM1, DFT,
MP2
AMBER
HF, DFT
H+ transfers
4-Oxalocrotonate
tautomerase
MM level
QM level
Process studied
Biomolecule
Table 5 Isomerases (EC 5)
VTST; comparison to
models; inuence of
specic vibrational modes
VTST; role of tunnelling
Evaluation of QM methods, size of

QM region, size of active region,
boundary schemes
Comparison to methylglyoxal
synthase
Alternative mechanisms
Comparison of mechanisms for

different substrates
Effect of mutations
Effect of mutations
Different substrates
Comments
[651]
[649]
[650]
[634]
[647]
[648]
[310]
[311]
[151]
[754]
[645]
[646]
[642]
[643]
[644]
Refs.
242
H.M. Senn W. Thiel
Chorismate mutase
Cyclophilin A
AMBER
CHARMM
AMBER
OPLS-AA
CHARMM
HF
DFT
AM1
SCC-DFTB
CHARMM
CHARMM
AMBER
AMBER
MM level
DFT
AM1
AM1
PM3, HF,
DFT
PM3
SCC-DFTB
Deamidation
1,2 H-shift
1,2 H-shift, H+ transfer

Peptide cis/trans
isomerization at Pro
Claisen rearrangement
HF
H+ transfer
Triosephosphate
isomerase
Xylose isomerase
QM level
Process studied
Biomolecule
Table 5 continued
US(MD)
Opt, FEP(MC)
Opt
Opt
Opt, US(MD)
Opt
Opt
US(MD), KIE
US(MD)
Opt
Opt, KIE
Opt, recrossing factor
Calculation type
QM/MM implementation in
the SIESTA real-space
DFT program
Inuence of substrate
conformation
Inuence of specic residues
on substrate conformations
VTST; evaluation of
different boundary schemes
Effect of mutations
Reaction-path potential,
reactive-ux calculation
Comments
[660]
[444]
[139]
[656]
[657]
[658]
[659]
[654]
[655]
[653]
[140]
[652]
Refs.

243
MethylmalonylCoA mutase
Claisen rearrangement
Chorismate mutase
Substrate conformation
Substrate conformation
Radical rearrangement
Claisen, Cope
rearrangement
Substrate binding
Process studied
Biomolecule
Table 5 continued
DFT
SCC-DFTB
AM1
AM1, DFT
AM1
AM1, DFT
SCC-DFTB,
DFT
HF
HF, MP2
DFT
AM1
AM1, DFT
AM1
HF
AM1
QM level
CHARMM
CHARMM
AMBER
OPLS-AA
AMBER
MM level
MD
US(MD)
Opt
Opt
Opt, US(MD)
Opt
Opt, MD, KIE

Opt
FEP(MD)
Opt
Opt, KIE, equilibrium
isotope effect
Opt
Opt
Fast-growth TI(MD),
US(MD)
FEP(MC)
Opt, KIE
Opt
Calculation type
Comparison of substrate
conformations
Comparison to FMO
all-QM treatment
Effect of mutation
Dual-level QM approach
DTSS analysis
Comments
[670]
[671]
[672]
[669]
[668]
[667]
[666]
[276]
[442]
[477]
[665]
[755]
[661]
[662]
[445]
[663]
[664]
Refs.
244
H.M. Senn W. Thiel
Process studied
Biomolecule
Threonyl-tRNA synthetase
Table 6 Ligases (EC 6)
AM1, PM3
QM level
CHARMM
MM level
Opt
Calculation type
Comments
[673]
Refs.

245
Bacteriorhodopsin
ENZYMIX
HF, DFT
QCFF/PI
H+ transfer
Photo-isomerization
Structure of bound
SCC-DFTB
intermediates
Vibrational analysis of HF
different states
Structure and dynamics DFT
of solvated proton
CHARMM
HF, DFT; CIS,

TDHF, TDDFT
SCC-DFTB
Opt
Opt
CP-MD
CHARMM
AMBER
GROMOS
Opt
Opt, MD
Quantum dynamics
Opt
Opt
Opt, excitation
energies, IR
CHARMM, MD, MC
OPLS
AMBER
Opt
AMBER
DFT, SCC-DFTB
HF, CASSCF,
MRMP
AM1, CIS
CP-MD
Opt
Calculation type
CHARMM
AMBER
MM level
DFT
AM1
Bacterial
photosynthetic
reaction centre
QM level
Quinone IR spectra
Structure of bacteriochlorophyll dimer
Electron transfer
Electronic excitation,
H+ transfer
Electronic excitation
Process studied
Biomolecule
Table 7 Photo-active proteins
[677]
[272]
[674]
[675, 676]
Refs.
Protein in lipid
bilayer
Semi-classical
trajectories
Coordinate
driving vs. reactionpath opt
Reaction-path opt
[685]
[684]
[683]
[680]
[681]
[682]
[679]
QM/MM MC in solution, [678]

MM MD in enzyme/lipid
[49]
Effect of mutations
Comments
246
H.M. Senn W. Thiel
Process studied
Structure of bound
chromophore
Chromophore structure,
electronic excitation
NMR parameters
Chromophore structure
and dynamics
Phot-LOV1
Photo-excitation
Photo-active yellow Photo-isomerization
protein
Green uorescent
protein
Rhodopsins
Bovine rhodopsin Structure of bound

chromophore
Structure of bound
retinal analogue
Electronic excitation,
structure of photoproduct
Biomolecule
Table 7 continued
CHARMM
SCC-DFTB, CASSCF,
CASPT2
DFT
DFT
HF, DFT
PM3, CASSCF
AMBER
GROMOS
CHARMM
CASSCF, CASPT2
AM1, TDDFT
Various
UFF
HF, SAC-CI
CASSCF, CASPT2
Various
DFT
HF, CASSCF
AMBER
DFT
DFT, TDDFT
AMBER
AMBER
CHARMM
MM level
SCC-DFTB
QM level
Opt
MD
Opt
Opt
Opt, excitation
energies
Opt
Opt
Opt
Opt
Opt
CP-MD
MD
Opt
Opt
Surface hopping
TDDFT on isolated QM part only
Method comparison
Comparison of different
environments
Different retinal
derivatives
Review
[700]
[701]
[699]
[698]
[694]
[695]
[696]
[697]
[693]
[691]
[692]
[690]
[688]
[689]
[687]
MD
Refs.
[686]
Comments
MD
Calculation type

247
CVFF
OPLS-AA
CHARMM
AMBER,
Tripos
AM1
DFT
HF
AM1
Structure of
antigenantibody
complex
Ligand charges for
docking
Inhibitor binding
EFP
HF, MP2
TIP3P
AMBER
MM level
pKa calculation
SCC-DFTB
Structure
Turkey ovomucoid
third domain
Hen egg white lysozyme
Fab HyHEL-5/Fab
D44.1
Various proteinligand
complexes
Thrombin
AM1, DFT
Interaction with water
-Helix antifreeze
protein
Crambin
QM level
Process studied
Biomolecule
Table 8 Other proteins, proteinligand complexes
ESP on vdW
surface
Opt
Single points
Opt
Opt
MD
Energy
Calculation type
All-QM protein in
MM solvent, comparison
to MM MD
Comments
[495]
[730]
[729]
[728]
[727]
[726]
[725]
Refs.
248
H.M. Senn W. Thiel
structure, NMR
parameters
Structure
Structure
Structure
Structure
Structure
SN 2 attack
Proteinnucleotide
association
DNAcisplatin
DNAHoechst 33258
DNAnetropsin
DNAPt complex
DNAPt2 complex
DNARu complex
DNAduocarmycin
U1ARNA
Process studied
Biomolecule
Table 9 Oligonucleotide adducts
AM1
HF
HF, DFT
DFT
DFT
DFT
DFT
DFT
QM level
AMBER
CHARMM
UFF
AMBER
AMBER
AMBER
OPLS-AA
AMBER
MM level
Opt
Opt
Opt
CP-MD
CP-MD
CP-MD
Opt
CP-MD
Calculation type
Effect of substituents
Comments
[732]
[733]
[732]
[734]
[735]
[736]
[737]
[731]
Refs.

249
250
H.M. Senn W. Thiel
molecule if they remain unchanged. Comments, however, apply only to the

specic row.
Acknowledgements H.M.S. thanks Drs. Johannes Kstner and Tell Tuttle for insightful
discussions. This work was supported by the Volkswagenstiftung.
References
1. Warshel A, Levitt M (1976) Theoretical studies of enzymic reactions: dielectric, electrostatic and steric stabilization of the carbonium ion in the reaction of lysozyme.
J Mol Biol 103(2):227249
2. Warshel A, Karplus M (1972) Calculation of ground and excited state potential surfaces of conjugated molecules I: formulation and parametrization. J Am Chem Soc
94(16):56125625
3. Singh UC, Kollman PA (1986) A combined ab initio quantum mechanical and molecular mechanical method for carrying out simulations on complex molecular
systems: applications to the CH3 Cl + Cl exchange reaction and gas phase protonation of polyethers. J Comput Chem 7(6):71830
4. Field MJ, Bash PA, Karplus M (1990) A combined quantum mechanical and molecular mechanical potential for molecular dynamics simulations. J Comput Chem
11(6):70033
5. Gao J (1996) Methods and applications of combined quantum mechanical and
molecular mechanical potentials. In: Lipkowitz KB, Boyd DB (eds) Reviews in computational chemistry, vol 7. VCH, New York, pp 119185
6. Gao J (1996) Hybrid quantum mechanical/molecular mechanical simulations: an alternative avenue to solvent effects in organic chemistry. Acc Chem Res 29(6):298305
7. Cunningham MA, Bash PA (1997) Computational enzymology. Biochimie 79(11):
687689
8. Gao J, Thompson MA (eds) (1998) Combined quantum mechanical and molecular
mechanical methods. ACS symposium series, vol 712. American Chemical Society,
Washington, DC
9. Merz KM Jr (1998) Quantum mechanical-molecular mechanical coupled potentials.
In: Gao J, Thompson MA (eds) Combined quantum mechanical and molecular mechanical methods. ACS symposium series, vol 712. American Chemical Society,
Washington, DC, pp 215
10. Gao J (1998) Hybrid quantum mechanical/molecular mechanical (QM/MM)
methods. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry,
vol 2. Wiley, Chichester, pp 12571263
11. Amara P, Field MJ (1998) Combined quantum mechanical and molecular mechanical
potentials. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry,
vol 1. Wiley, Chichester, pp 431437
12. Ruiz-Lpez MF, Rivail JL (1998) Combined quantum mechanics and molecular mechanics approaches to chemical and biochemical reactivity. In: von Ragu Schleyer P
(ed) Encyclopedia of computational chemistry, vol 1. Wiley, Chichester, pp 437448
13. Merz KM Jr, Stanton RV (1998) Quantum mechanical/molecular mechanical
(QM/MM) coupled potentials. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 4. Wiley, Chichester, pp 23302343
14. Friesner RA, Beachy MD (1998) Quantum mechanical calculations on biological
systems. Curr Opin Struct Biol 8(2):257262
251
15. Beck B, Clark T (1998) Some biological applications of semiempirical MO theory.

Perspect Drug Disc Des 911:131159
16. Mordasini T, Thiel W (1998) Combined quantum mechanical and molecular approaches. Chimia 52(6):288291
17. Monard G, Merz KM Jr (1999) Combined quantum mechanical/molecular mechanical methodologies applied to biomolecular systems. Acc Chem Res 32(10):904911
18. Hillier IH (1999) Chemical reactivity studied by hybrid QM/MM methods.
THEOCHEM 463(12):4552
19. Amara P, Field MJ (1999) Hybrid potentials for large molecular systems. In:
Leszczynski J (ed) Computational molecular biology. Theoretical and computational
chemistry, vol 8. Elsevier, Amsterdam
20. Bruice TC, Kahn K (2000) Computational enzymology. Curr Opin Chem Biol 4:540
544
21. Sherwood P (2000) Hybrid quantum mechanics/molecular mechanics approaches.
In: Grotendorst J (ed) Modern methods and algorithms of quantum chemistry. NIC
Series, vol 1. John von Neumann Institute for Computing, Jlich, pp 257277
22. Lyne PD, Walsh OA (2001) Computer simulation of biochemical reactions with QMMM methods. In: Becker OM, Watanabe M (eds) Computational biochemistry and
biophysics. Dekker, New York, pp 221236
23. Mulholland AJ (2001) The QM/MM approach to enzymatic reactions. In: Eriksson LA (ed) Theoretical biochemistry: processes and properties of biological systems. Theoretical and computational chemistry, vol 9. Elsevier, Amsterdam, pp 597
653
24. Field MJ (2002) Simulating enzyme reactions: Challenges and perspectives. J Comput Chem 23(1):4858
25. Gogonea V (2002) The QM/MM method. An overview. Internet Electron J Mol Des
1(4):173184
26. Castillo R, Oliva M, Mart S, Moliner V, Tun I, and Andrs J (2002) Hybrid
QM/MM studies on chemical reactivity. Recent Res Dev Quantum Chem 3:5174
27. Monard G, Prat-Resina X, Gonzlez-Lafont A, Lluch JM (2003) Determination of enzymatic reaction pathways using QM/MM methods. Int J Quant Chem 93(3):229
244
28. Ridder L, Mulholland AJ (2003) Modeling biotransformation reactions by combined
quantum mechanical/molecular mechanical approaches: from structure to activity.
Curr Top Med Chem 3(11):12411256
29. Nray-Szab G, Berente I (2003) Computer modelling of enzyme reactions.
THEOCHEM 666667:637644
30. Ryde U (2003) Combined quantum and molecular mechanics calculations on metalloproteins. Curr Opin Chem Biol 7(1):136142
31. Perkyl M (2003) Ab initio methods in the study of reaction mechanisms their
role and perspectives in medicinal chemistry. In: Carloni P, Alber F (eds) Quantum
medicinal chemistry. Methods and principles in medicinal chemistry, vol 17. Wiley,
Weinheim, pp 157176
32. Perruccio F, Ridder L, Mulholland AJ (2003) Quantum-mechanical/molecularmechanical methods in medicinal chemistry. In: Carloni P, Alber F (eds) Quantum
medicinal chemistry. methods and principles in medicinal chemistry, vol 17. Wiley,
Weinheim, pp 177198
33. Rivail JL (2004) Hybrid quantum mechanical/molecular mechanical methods. In:
Bultinck P, De Winter H, Langenaeker W, Tollenaere JP (eds) Computational medicinal chemistry for drug discovery. Dekker, New York, pp 119131
252
H.M. Senn W. Thiel
34. Friesner RA (2004) Combined quantum and molecular mechanics (QM/MM). Drug
Discovery Today: Technol 1(3):253260
35. Mulholland AJ (2005) Modelling enzyme reaction mechanisms, specicity and catalysis. Drug Discovery Today 10(20):13931402
36. Friesner RA, Guallar V (2005) Ab initio quantum chemical and mixed quantum mechanics/molecular mechanics (QM/MM) methods for studying enzymatic catalysis.
Annu Rev Phys Chem 56:389427
37. Lin H, Truhlar DG (2006) QM/MM: What have we learned, where are we, and where
do we go from here? Theor Chem Acc, Online rst, 8 July 2006
38. Maple JR (1998) Force Fields: A general discussion. In: von Ragu Schleyer P
(ed) Encyclopedia of computational chemistry, vol 2. Wiley, Chichester, pp 1015
1024
39. MacKerell AD Jr (2001) Atomistic models and force elds. In: Becker OM, MacKerell AD Jr, Roux B, Watanabe M (eds) Computational biochemistry and biophysics. Dekker, New York, pp 738
40. Ponder JW, Case DA (2003) Force elds for protein simulations. In: Daggett V (ed)
Protein simulations. Advances in protein chemistry, vol 66. Academic Press, San
Diego, pp 2785
41. Liljefors T, Gundertofte K, Norrby P-O, Pettersson I (2004) Molecular mechanics
and comparison of force elds. In: Bultinck P, De Winter H, Langenaeker W, Tollenaere JP (eds) Computational medicinal chemistry for drug discovery. Dekker,
New York, p 128
42. MacKerell AD Jr (2004) Empirical force elds for biological macromolecules:
Overview and issues. J Comput Chem 25(13):15841604
43. Maseras F, Morokuma K (1995) IMOMM: A new integrated ab initio + molecular
mechanics geometry optimization scheme of equilibrium structures and transition
states. J Comput Chem 16(9):11701179
44. Humbel S, Sieber S, Morokuma K (1996) The IMOMO method: Integration of different levels of molecular orbital approximations for geometry optimization of large
systems: Test for n-butane conformation and SN 2 reaction: RCl + Cl . J Chem Phys
105(5):19591967
45. Svensson M, Humbel S, Froese RDJ, Matsubara T, Sieber S, Morokuma K (1996)
ONIOM: A multilayered integrated MO + MM method for geometry optimizations
and single point energy predictions. A test for DielsAlder reactions and Pt(P(tBu)3 )2 + H2 oxidative addition. J Phys Chem 100(50):1935719363
46. Froese RDJ, Morokuma K (1998) Hybrid methods. In: von Ragu Schleyer P (ed)
Encyclopedia of computational chemistry, vol 2. Wiley, Chichester, pp 12441257
47. Dapprich S, Komromi I, Byun KS, Morokuma K, Frisch MJ (1999) A new ONIOM
implementation in Gaussian98. Part I. The calculation of energies, gradients, vibrational frequencies and electric eld derivatives. THEOCHEM 461462:121
48. Komromi I, Muszbek L (2002) Application of the IMOMM (integrated molecular
orbital molecular mechanics) method for biopolymers. In: Buzaneva E, Scharff P
(eds) Frontiers of multifunctional nanosystems. NATO science series II: Mathematics, physics and chemistry, vol 57. Kluwer, Dordrecht, pp 1728
49. Vreven T, Morokuma K (2003) Investigation of the S0 S1 excitation in bacteriorhodopsin with the ONIOM(MO:MM) hybrid method. Theor Chem Acc 109(3):125
132
50. Vreven T, Byun KS, Komromi I, Dapprich S, Montgomery JA Jr, Morokuma K,
Frisch MJ (2006) Combining quantum mechanics methods with molecular mechanics methods in ONIOM. J Chem Theory Comput 2(3):815826
253
51. Bakowies D, Thiel W (1996) Hybrid models for combined quantum mechanical and
molecular mechanical approaches. J Phys Chem 100(25):1058010594
52. Antes I, Thiel W (1998) On the treatment of link atoms in hybrid methods. In: Gao J,
Thompson MA (eds) Combined quantum mechanical and molecular mechanical
methods. ACS Symposium Series, vol 712. American Chemical Society, Washington,
DC, pp 5065
53. Rick SW, Stuart SJ (2002) Potentials and algorithms for incorporating polarizability
in computer simulations. In: Lipkowitz KB, Boyd DB (eds) Reviews in computational
chemistry, vol 18. Wiley, Hoboken, NJ, pp 89146
54. Yu H, van Gunsteren WF (2005) Accounting for polarization in molecular simulation. Comput Phys Commun 172(2):6985
55. Stern HA, Rittner F, Berne BJ, Friesner RA (2001) Combined uctuating charge and
polarizable dipole models: Application to a ve-site water potential function. J Chem
Phys 115(5):22372251
56. Ren P, Ponder JW (2003) Polarizable atomic multipole water model for molecular
mechanics simulation. J Phys Chem B 107(24):59335947
57. Grosseld A, Ren P, Ponder JW (2003) Ion solvation thermodynamics from simulation with a polarizable force eld. J Am Chem Soc 125(50):1567115682
58. Lamoureux G, Roux B (2003) Modeling induced polarization with classical Drude
oscillators: Theory and molecular dynamics simulation algorithm. J Chem Phys
119(6):30253039
59. Lamoureux G, MacKerell AD Jr, Roux B (2003) A simple polarizable model of water
based on classical Drude oscillators. J Chem Phys 119(10):51855197
60. Yu H, Hansson T, van Gunsteren WF (2003) Development of a simple, self-consistent
polarizable model for liquid water. J Chem Phys 118(1):221234
61. Yu H, van Gunsteren WF (2004) Charge-on-spring polarizable water models revisited: from water clusters to liquid water to ice. J Chem Phys 121(19):95499564
62. Patel S, Brooks CL III (2004) CHARMM uctuating charge force eld for proteins:
I Parameterization and application to bulk organic liquid simulations. J Comput
Chem 25(1):116
63. Patel S, MacKerell AD Jr, Brooks CL III (2004) CHARMM uctuating charge force
eld for proteins: II Protein/solvent properties from molecular dynamics simulations
using a nonadditive electrostatic model. J Comput Chem 25(12):15041514
64. Vorobyov IV, Anisimov VM, MacKerell AD Jr (2005) Polarizable empirical force
eld for alkanes based on the classical Drude oscillator model. J Phys Chem B
109(40):1898818999
65. Anisimov VM, Lamoureux G, Vorobyov IV, Huang N, Roux B, MacKerell AD Jr
(2005) Determination of electrostatic parameters for a polarizable force eld based
on the classical Drude oscillator. J Chem Theory Comput 1(1):153168
66. Wang J, Cieplak P, Kollman PA (2000) How well does a restrained electrostatic potential (RESP) model perform in calculating conformational energies of organic and
biological molecules? J Comput Chem 21(12):10491074
67. Cieplak P, Caldwell J, Kollman PA (2001) Molecular mechanical models for organic
and biological systems going beyond the atom centered two body additive approximation: aqueous solution free energies of methanol and N-methyl acetamide,
nucleic acid base, and amide hydrogen bonding and chloroform/water partition
coefcients of the nucleic acid bases. J Comput Chem 22(10):10481057
68. Wang Z-X, Zhang W, Wu C, Lei H, Cieplak P, Duan Y (2006) Strike a balance: optimization of backbone torsion parameters of AMBER polarizable force eld for
simulations of proteins and peptides. J Comput Chem 27(6):781790
254
H.M. Senn W. Thiel
69. Banks JL, Kaminski GA, Zhou R, Mainz DT, Berne BJ, Friesner RA (1999)
Parametrizing a polarizable force eld from ab initio data. I. The uctuating point
charge model. J Chem Phys 110:741754
70. Stern HA, Kaminski GA, Banks JL, Zhou R, Berne BJ, Friesner RA (1999) Fluctuating
charge, polarizable dipole, and combined models: parameterization from ab initio
quantum chemistry. J Phys Chem B 103(22):47304737
71. Kaminski GA, Stern HA, Berne BJ, Friesner RA, Cao YX, Murphy RB, Zhou R, Halgren TA (2002) Development of a polarizable force eld for proteins via ab initio
quantum chemistry: First generation model and gas phase tests. J Comput Chem
23(16):15151531
72. Kaminski GA, Stern HA, Berne BJ, Friesner RA (2004) Development of an accurate and robust polarizable molecular mechanics force eld from ab initio quantum
chemistry. J Phys Chem A 108(4):621627
73. Ren P, Ponder JW (2002) Consistent treatment of inter- and intramolecular polarization in molecular mechanics calculations. J Comput Chem 23(16):14971506
74. Thompson MA, Schenter GK (1995) Excited states of the bacteriochlorophyll b dimer
of Rhodopseudomonas viridis: a QM/MM study of the photosynthetic reaction center
that includes MM polarization. J Phys Chem 99(17):637486
75. Gao J (1997) Energy components of aqueous solution: insight from hybrid QM/MM
simulations using a polarizable solvent model. J Comput Chem 18(8):10611071
76. Bryce RA, Vincent MV, Malcolm NOJ, Hillier IH, Burton NA (1998) Cooperative
effects in the structuring of uoride water clusters: ab initio hybrid quantum mechanical/molecular mechanical model incorporating polarizable uctuating charge
solvent. J Chem Phys 109:30773085
77. Jensen L, van Duijnen PT, Snijders JG (2003) A discrete solvent reaction eld model
within density functional theory. J Chem Phys 118(2):514521
78. Nam K, Gao J, York DM (2005) An efcient linear-scaling Ewald method for longrange electrostatic interactions in combined QM/MM calculations. J Chem Theory
Comput 1(1):213
79. Riccardi D, Schaefer P, Cui Q (2005) pKa calculations in solution and proteins with
QM/MM free energy perturbation simulations: a quantitative test of QM/MM protocols. J Phys Chem B 109(37):1771517733
80. Schaefer P, Riccardi D, Cui Q (2005) Reliable treatment of electrostatics in combined
QM/MM simulation of macromolecules. J Chem Phys 123(1):014905/114
81. Dinner AR, Lopez X, Karplus M (2003) A charge-scaling method to treat solvent in
QM/MM simulations. Theor Chem Acc 109(3):118124
82. Gregersen BA, York DM (2005) Variational electrostatic projection (VEP) methods
for efcient modeling of the macromolecular electrostatic and solvation environment in activated dynamics simulations. J Phys Chem B 109(1):536556
83. Gregersen BA, York DM (2006) A charge-scaling implementation of the variational
electrostatic projection method. J Comput Chem 27(1):103115
84. Berkowitz M, McCammon JA (1982) Molecular dynamics with stochastic boundary
conditions. Chem Phys Lett 90:215217
85. Brnger A, Brooks CL III, Karplus M (1984) Stochastic boundary conditions for
molecular dynamics simulations of ST2 water. Chem Phys Lett 105(5):495500
86. Brooks CL III, Brnger A, Karplus M (1985) Active site dynamics in protein molecules:
a stochastic boundary molecular-dynamics approach. Biopolymers 24:843865
87. Brooks CL III, Karplus M (1989) Solvent effects on protein motion and protein effects on solvent motion: dynamics of the active site region of lysozyme. J Mol Biol
208:159181
255
88. King G, Warshel A (1989) A surface constrained all-atom solvent model for effective
simulations of polar solutions. J Chem Phys 91(6):36473661
89. Beglov D, Roux B (1994) Finite representation of an innite bulk system: solvent
boundary potential for computer simulations. J Chem Phys 100:90509063
90. Im W, Bernche S, Roux B (2001) Generalized solvent boundary potential for computer simulations. J Chem Phys 114(7):29242937
91. Knig PH, Ghosh N, Hoffmann M, Elstner M, Tajkhorshid E, Frauenheim T,
Cui Q (2006) Toward theoretical analysis of long-range proton transfer kinetics in
biomolecular pumps. J Phys Chem A 110(2):548563
92. Riccardi D, Schaefer P, Yang Y, Yu H, Ghosh N, Prat-Resina X, Knig P, Li G,
Xu D, Guo H, Elstner M, Cui Q (2006) Development of effective quantum mechanical/molecular mechanical (QM/MM) methods for complex biological processes.
J Phys Chem B 110(13):64586469
93. Murphy RB, Philipp DM, Friesner RA (2000) A mixed quantum mechanics/molecular mechanics (QM/MM) method for large-scale modeling of chemistry
in protein environments. J Comput Chem 21(16):14421457
94. Freindorf M, Shao Y, Furlani TR, Kong J (2005) LennardJones parameters for the
combined QM/MM method using the B3LYP/6-31+G /AMBER potential. J Comput
Chem 26(12):12701278
95. Riccardi D, Li G, Cui Q (2004) Importance of van der Waals interactions in QM/MM
simulations. J Phys Chem B 108(20):64676478
96. Eurenius KP, Chateld DC, Brooks BR, Hodoscek M (1996) Enzyme mechanisms
with hybrid quantum and molecular mechanical potentials. I. Theoretical considerations. Int J Quant Chem 60(6):11891200
97. Koga N, Morokuma K (1990) A simple scheme of estimating substitution or substituent effects in the ab initio MO method based on the shift operator. Chem Phys
Lett 172(34):243248
98. Matsubara T, Maseras F, Koga N, Morokuma K (1996) Application of the new integrated MO + MM (IMOMM) method to the organometallic reaction Pt(PR3 )2 + H2
(R = H, Me, t-Bu, and Ph). J Phys Chem 100(7):25732580
99. Senn HM (2001) Transition-metal-catalysed hydroamination of alkenes: ab initio
molecular-dynamics studies. PhD Thesis, Diss ETH no 13972. ETH Zrich, Zrich
100. Eichler U, Klmel CM, Sauer J (1997) Combining ab initio techniques with analytical potential functions for structure predictions of large systems: method and
application to crystalline silica polymorphs. J Comput Chem 18(4):463477
101. Eichinger M, Tavan P, Hutter J, Parrinello M (1999) A hybrid method for solutes in
complex solvents: density functional theory combined with empirical force elds.
J Chem Phys 110(21):1045210467
102. Lyne PD, Hodoscek M, Karplus M (1999) A hybrid QMMM potential employing
HartreeFock or density functional methods in the quantum region. J Phys Chem
A 103(18):34623471
103. de Vries AH, Sherwood P, Collins SJ, Rigby AM, Rigutto M, Kramer GJ (1999) Zeolite structure and reactivity by combined quantum-chemicalclassical calculations.
J Phys Chem B 103(29):61336141
104. Field MJ, Albe M, Bret C, Proust-De Martin F, Thomas A (2000) The Dynamo library for molecular simulations using hybrid quantum mechanical and molecular
mechanical potentials. J Comput Chem 21(12):10881100
105. Woo TK, Cavallo L, Ziegler T (1998) Implementation of the IMOMM methodology
for performing combined QM/MM molecular dynamic simulations and frequency
calculations. Theor Chem Acc 100(56):307313
256
H.M. Senn W. Thiel
106. Swart M (2003) AddRemove: a new link model for use in QM/MM studies. Int
J Quant Chem 91(2):177183
107. Das D, Eurenius KP, Billings EM, Sherwood P, Chateld DC, Hodoscek M, Brooks BR
(2002) Optimization of quantum mechanical molecular mechanical partitioning
schemes: Gaussian delocalization of molecular mechanical charges and the double
link atom method. J Chem Phys 117(23):1053410547
108. Ferr N, Olivucci M (2003) The amide bond: pitfalls and drawbacks of the link atom
scheme. THEOCHEM 632:7182
109. Reuter N, Dejaegere A, Maigret B, Karplus M (2000) Frontier bonds in QM/MM
methods: a comparison of different approaches. J Phys Chem A 104(8):17201735
110. Waszkowycz B, Hillier IH, Gensmantel N, Payling DW (1990) A theoretical study of
the hydrolysis by phospholipase A2 : the catalytic role of the active site and substrate
specicity. J Chem Soc, Perkin Trans 2 (7):12591268
111. Waszkowycz B, Hillier IH, Gensmantel N, Payling DW (1991) A combined quantum mechanical/molecular mechanical model of the potential energy surface of ester
hydrolysis by the enzyme phospholipase A2 . J Chem Soc, Perkin Trans 2 (2):225
231
112. Waszkowycz B, Hillier IH, Gensmantel N, Payling DW (1991) A quantum mechanical/molecular mechanical model of the inhibition of the enzyme phospholipase A2 .
J Chem Soc, Perkin Trans 2 (11):18191832
113. Vasilyevy VV (1994) Tetrahedral intermediate formation in the acylation step of
acetylcholinesterases. A combined quantum chemical and molecular mechanical
model. THEOCHEM 110(2):12941
114. Lin H, Truhlar DG (2005) Redistributed charge and dipole schemes for combined quantum mechanical and molecular mechanical calculations. J Phys Chem A
109(17):39914004
115. Knig PH, Hoffmann M, Frauenheim T, Cui Q (2005) A critical evaluation of different QM/MM frontier treatments with SCC-DFTB as the QM method. J Phys Chem B
109(18):90829095
116. Sherwood P, de Vries AH, Collins SJ, Greatbanks SP, Burton NA, Vincent MV,
Hillier IH (1997) Computer simulation of zeolite structure and reactivity using embedded cluster methods. Faraday Discuss 106:7992
117. Sherwood P, de Vries AH, Guest MF, Schreckenbach G, Catlow CRA, French SA,
Sokol AA, Bromley ST, Thiel W, Turner AJ, Billeter S, Terstegen F, Thiel S, Kendrick J,
Rogers SC, Casci J, Watson M, King F, Karlsen E, Sjvoll M, Fahmi A, Schfer A,
Lennartz C (2003) QUASI: a general purpose implementation of the QM/MM approach and its application to problems in catalysis. THEOCHEM 632(13):128
118. Amara P, Field MJ (2003) Evaluation of an ab initio quantum mechanical/molecular
mechanical hybrid-potential link-atom method. Theor Chem Acc 109(1):4352
119. CPMD, v. 3.9, IBM Corp., 19902001, MPI fr Festkrperforschung, Stuttgart, 1997
2004. http://www.cpmd.org/
120. Antes I, Thiel W (1999) Adjusted connection atoms for combined quantum mechanical and molecular mechanical methods. J Phys Chem A 103:92909295
121. Zhang Y, Lee T-S, Yang W (1999) A pseudobond approach to combining quantum
mechanical and molecular mechanical methods. J Chem Phys 110(1):4654
122. Zhang Y (2005) Improved pseudobonds for combined ab initio quantum mechanical/molecular mechanical methods. J Chem Phys 122(2):024114/17
123. Alary F, Poteau R, Heully J-L, Barthelat J-L, Daudey J-P (2000) A new method for
modelling spectator chemical groups in ab initio calculations: effective group potentials. Theor Chem Acc 104(34):174178
257
124. Poteau R, Ortega I, Alary F, Solis AR, Barthelat J-C, Daudey J-P (2001) Effective
group potentials. 1. Method. J Phys Chem A 105(1):198205
125. Poteau R, Alary F, El Makarim HA, Heully J-L, Barthelat J-C, Daudey J-P (2001)
Effective group potentials. 2. Extraction and transferability for chemical groups involved in covalent or donoracceptor bonds. J Phys Chem A 105(1):206214
126. Bessac F, Alary F, Carissan Y, Heully J-L, Daudey J-P, Poteau R (2003) Effective group
potentials: a powerful tool for hybrid QM/MM methods? THEOCHEM 632(13):
4359
127. Carissan Y, Bessac F, Alary F, Heully J-L, Poteau R (2006) What can we do with an
effective group potential? Int J Quant Chem 106(3):727733
128. DiLabio GA, Hurley MM, Christiansen PA (2002) Simple one-electron quantum capping potentials for use in hybrid QM/MM studies of biological molecules. J Chem
Phys 116(22):95789584
129. DiLabio GA, Wolkow RA, Johnson ER (2005) Efcient silicon surface and cluster
modeling using quantum capping potentials. J Chem Phys 122(4):044708/15
130. Yasuda K, Yamaki D (2004) Simple minimum principle to derive a quantummechanical/molecular-mechanical method. J Chem Phys 121(9):39643972
131. Lilienfeld OA, Tavernelli I, Rothlisberger U, Sebastiani D (2005) Variational optimization of effective atom centered potentials for molecular properties. J Chem Phys
122(1):014113/16
132. Laio A, VandeVondele J, Rothlisberger U (2002) A Hamiltonian electrostatic coupling scheme for hybrid Car-Parrinello molecular dynamics simulations. J Chem Phys
116(16):69416947
133. Slavcek P, Martnez TJ (2006) Multicentered valence electron effective potentials:
A solution to the link atom problem for ground and excited electronic states. J Chem
Phys 124(8):084107/110
134. Thry V, Rinaldi D, Rivail J-L, Maigret B, Ferenczy GG (1994) Quantum mechanical
computations on very large molecular systems: the local self-consistent eld method.
J Comput Chem 15(3):269282
135. Monard G, Loos M, Thry V, Baka K, Rivail J-L (1996) Hybrid classical quantum
force eld for modeling very large molecules. Int J Quant Chem 58(2):153159
136. Assfeld X, Rivail J-L (1996) Quantum chemical computations on parts of large
molecules: the ab initio local self consistent eld method. Chem Phys Lett 263:100
106
137. Assfeld X, Ferr N, Rivail JL (1998) The local self-consistent eld: principles and
applications to combined quantum mechanicalmolecular mechanical computations
on biomacromolecular systems. In: Gao J, Thompson MA (eds) Combined quantum
mechanical and molecular mechanical methods. ACS symposium series, vol 712.
American Chemical Society, Washington, DC, pp 234249
138. Ferr N, Assfeld X, Rivail J-L (2002) Specic force eld parameters determination for
the hybrid ab initio QM/MM LSCF method. J Comput Chem 23(6):610624
139. Hall RJ, Hindle SA, Burton NA, Hillier IH (2000) Aspects of hybrid QM/MM calculations: the treatment of the QM/MM interface region and geometry optimization with
an application to chorismate mutase. J Comput Chem 21(16):14331441
140. Nicoll RM, Hindle SA, MacKenzie G, Hillier IH, Burton NA (2001) Quantum mechanical/molecular mechanical methods and the study of kinetic isotope effects:
modelling the covalent junction region and application to the enzyme xylose isomerase. Theor Chem Acc 106(12):105112
141. Fornili A, Sironi M, Raimondi M (2003) Determination of extremely localized
molecular orbitals and their application to quantum mechanics/molecular mechan-
258
142.
143.
144.
145.
146.
147.
148.
149.
150.
151.
152.
153.
154.
155.
156.
157.
158.
159.
H.M. Senn W. Thiel

ics methods and to the study of intramolecular hydrogen bonding. THEOCHEM
632(13):157172
Fornili A, Moreau Y, Sironi M, Assfeld X (2006) On the suitability of strictly localized
orbitals for hybrid QM/MM calculations. J Comput Chem 27(4):515523
Philipp DM, Friesner RA (1999) Mixed ab initio QM/MM modeling using frozen
orbitals and tests with alanine dipeptide and tetrapeptide. J Comput Chem
20(14):14681494
Murphy RB, Philipp DM, Friesner RA (2000) Frozen orbital QM/MM methods for
density functional theory. Chem Phys Lett 321(12):113120
Gao J, Amara P, Alhambra C, Field MJ (1998) A generalized hybrid orbital (GHO)
method for the treatment of boundary atoms in combined QM/MM calculations.
J Phys Chem A 102(24):47144721
Amara P, Field MJ, Alhambra C, Gao J (2000) The generalized hybrid orbital method
for combined quantum mechanical/molecular mechanical calculations: formulation
and tests of the analytical derivatives. Theor Chem Acc 104(5):336343
Garcia-Viloca M, Gao J (2004) Generalized hybrid orbital for the treatment of
boundary atoms in combined quantum mechanical and molecular mechanical calculations using the semiempirical parameterized model 3 method. Theor Chem Acc
111(26):280286
Pu J, Gao J, Truhlar DG (2004) Generalized hybrid orbital (GHO) method for combining ab initio HartreeFock wave functions with molecular mechanics. J Phys
Chem A 108(4):632650
Pu J, Gao J, Truhlar DG (2004) Combining self-consistent-charge density-functional
tight-binding (SCC-DFTB) with molecular mechanics by the generalized hybrid orbital (GHO) method. J Phys Chem A 108(25):54545463
Pu J, Gao J, Truhlar DG (2005) Generalized hybrid-orbital method for combining
density functional theory with molecular mechanicals. Chem Phys Chem 6(9):1853
1865
Lennartz C, Schfer A, Terstegen F, Thiel W (2002) Enzymatic reactions of triosephosphate isomerase: a theoretical calibration study. J Phys Chem B 106(7):1758
1767
Cui Q, Elstner M, Kaxiras E, Frauenheim T, Karplus M (2001) A QM/MM implementation of the self-consistent charge density functional tight binding (SCC-DFTB)
method. J Phys Chem B 105(2):569585
Elstner M, Frauenheim T, Suhai S (2003) An approximate DFT method for QM/MM
simulations of biological structures and processes. THEOCHEM 632(13):2941
Elstner M, Porezag D, Jungnickel G, Elsner J, Haugk M, Frauenheim T, Suhai S,
Seifert G (1998) Self-consistent-charge density-functional tight-binding method for
simulations of complex materials properties. Phys Rev B 58(11):72607268
Schtz M, Hetzer G, Werner H-J (1999) Low-order scaling local electron correlation
methods. I. Linear scaling local MP2. J Chem Phys 111(13):56915705
Schtz M, Werner H-J (2000) Local perturbative triples correction (T) with linear
cost scaling. Chem Phys Lett 318(45):370378
Schtz M (2000) Low-order scaling local electron correlation methods. III. Linear
scaling local perturbative triples correction (T). J Chem Phys 113(22):998610001
Schtz M, Werner H-J (2001) Low-order scaling local electron correlation methods.
IV. Linear scaling local coupled-cluster (LCCSD). J Chem Phys 114(2):661681
Schtz M (2002) Low-order scaling local electron correlation methods. V. Connected
triples beyond (T): Linear scaling local CCSDT-1b. J Chem Phys 116(20):8772
8785
259
160. Schtz M, Manby FR (2003) Linear scaling local coupled cluster theory with density
tting. Part I: 4-external integrals. Phys Chem Chem Phys 5(16):33493358
161. Werner H-J, Manby FR, Knowles PJ (2003) Fast linear scaling second-order MllerPlesset perturbation theory (MP2) using local and density tting approximations.
J Chem Phys 118(18):81498160
162. Werner H-J, Manby FR (2006) Explicitly correlated second-order perturbation theory using density tting and local approximations. J Chem Phys 124(5):054114/112
163. Car R, Parrinello M (1985) Unied approach for molecular dynamics and densityfunctional theory. Phys Rev Lett 55:24712474
164. Marx D, Hutter J (2000) Ab initio molecular dynamics: Theory and implementation. In: Grotendorst J (ed) Modern methods and algorithms of quantum chemistry.
NIC Series, vol 1. John von Neumann Institute of Computing, Jlich, pp 301449,
http://www.fz-juelich.de/nic-series/Volume3/marx.pdf
165. Woo TK, Blchl PE, Ziegler T (2000) Towards solvation simulations with a combined
ab initio molecular dynamics and molecular mechanics approach. THEOCHEM
506(13):313334
166. Blchl PE (1995) Electrostatic decoupling of periodic images of plane-waveexpanded densities and derived atomic point charges. J Chem Phys 103:7422
7428
167. Biswas PK, Gogonea V (2005) A regularized and renormalized electrostatic coupling
Hamiltonian for hybrid quantum-mechanical-molecular-mechanical calculations.
J Chem Phys 123:164114/19
168. Laio A, VandeVondele J, Rothlisberger U (2002) D-RESP: Dynamically generated
electrostatic potential derived charges from quantum mechanics/molecular mechanics simulations. J Phys Chem B 106(29):73007307
169. Laio A, Gervasio FL, VandeVondele J, Sulpizi M, Rothlisberger U (2004) A variational
denition of electrostatic potential derived charges. J Phys Chem B 108(23):7963
7968
170. Laino T, Mohamed F, Laio A, Parrinello M (2005) An efcient real space multigrid
QM/MM electrostatic coupling. J Chem Theory Comput 1(6):11761184
171. MacKerell AD Jr (1998) Protein Force Fields. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 3. Wiley, Chichester, pp 21912200
172. MacKerell AD Jr (2005) Empirical force elds for proteins: current status and future
directions. In: Spellmeyer DC (ed) Annual reports in computational chemistry, vol 1.
Elsevier, Amsterdam, pp 91102
173. Cieplak P (1998) Nucleic acid force elds. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 3. Wiley, Chichester, pp 19221930
174. Cheatham TE III (2005) Molecular modeling and atomistic simulation of nucleic
acids. In: Spellmeyer DC (ed) Annual reports in computational chemistry, vol 1.
Elsevier, Amsterdam, pp 7589
175. Woods RJ (1998) Carbohydrate force elds. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 1. Wiley, Chichester, pp 220233
176. Cornell WD, Cieplak P, Bayly CI, Gould IR, Merz KM Jr, Ferguson DM, Spellmeyer DC, Fox T, Caldwell JW, Kollman PA (1995) A second generation force eld
for the simulation of proteins, nucleic acids, organic molecules. J Am Chem Soc
117:51795197
177. Kollman P, Caldwell JW, Ross WS, Pearlman DA, Case DA, DeBolt S, Cheatham TE III,
Ferguson D, Seibel G (1998) AMBER: a program for simulation of biological and
organic molecules. In: von Ragu Schleyer P (ed) Encyclopedia of computational
chemistry, vol 1. Wiley, Chichester, pp 1113
260
H.M. Senn W. Thiel
178. Duan Y, Wu C, Chowdhury S, Lee MC, Xiong G, Zhang W, Yang R, Cieplak P, Luo R,
Lee T, Caldwell J, Wang J, Kollman P (2003) A point-charge force eld for molecular
mechanics simulations of proteins based on condensed-phase quantum mechanical
calculations. J Comput Chem 24(16):19992012
179. MacKerell AD Jr, Bashford D, Bellott M, Dunbrack RL, Evanseck JD, Field MJ, Fischer S, Gao J, Guo H, Ha S, Joseph-McCarthy D, Kuchnir L, Kuczera K, Lau FTK,
Mattos C, Michnick S, Ngo T, Nguyen DT, Prodhom B, Reiher WE, Roux B,
Schlenkrich M, Smith JC, Stote R, Straub J, Watanabe M, Wiorkiewicz-Kuczera J,
Yin D, Karplus M (1998) All-atom empirical potential for molecular modeling and
dynamics studies of proteins. J Phys Chem B 102(18):35863616
180. MacKerell AD Jr, Brooks B, Brooks CL III, Nilsson L, Roux B, Won Y, Karplus M
(1998) CHARMM: The energy function and its parameterization with an overview
of the program. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 1. Wiley, Chichester, pp 271277
181. Foloppe N, MacKerell AD Jr (2000) All-atom empirical force eld for nucleic acids:
I. Parameter optimization based on small molecule and condensed phase macromolecular target data. J Comput Chem 21(2):86104
182. MacKerell AD Jr, Banavali NK (2000) All-atom empirical force eld for nucleic acids:
II. Application to molecular dynamics simulations of DNA and RNA in solution.
183. van Gunsteren WF, Daura X, Mark AE (1998) GROMOS force eld. In: von Ragu
Schleyer P (ed) Encyclopedia of computational chemistry, vol 2. Wiley, Chichester,
pp 12111216
184. Scott WRP, Hnenberger PH, Tironi IG, Mark AE, Billeter SR, Fennen J, Torda AE,
Huber T, Krger P, van Gunsteren WF (1999) The GROMOS biomolecular simulation
program package. J Phys Chem A 103(19):35963607
185. Jorgensen WL, Maxwell DS, Tirado-Rives J (1996) Development and testing of the
OPLS all-atom force eld on conformational energetics and properties of organic
liquids. J Am Chem Soc 118(45):1122511236
186. Jorgensen WL (1998) OPLS force elds. In: von Ragu Schleyer P (ed) Encyclopedia
of computational chemistry, vol 3. Wiley, Chichester, pp 19861989
187. Kaminski GA, Friesner RA, Tirado-Rives J, Jorgensen WL (2001) Evaluation and
reparametrization of the OPLS-AA force eld for proteins via comparison with accurate quantum chemical calculations on peptides. J Phys Chem B 105(28):64746487
188. Allinger NL, Yuh YH, Lii JH (1989) Molecular mechanics: The MM3 force-eld for
hydrocarbons. 1. J Am Chem Soc 111(23):85518566
189. Lii JH, Allinger NL (1989) Molecular mechanics. The MM3 force-eld for hydrocarbons. 2. Vibrational frequencies and thermodynamics. J Am Chem Soc
111(23):85668575
190. Lii JH, Allinger NL (1989) Molecular mechanics. The MM3 force-eld for hydrocarbons. 3. The van der Waals potentials and crystal data for aliphatic and aromatic
hydrocarbons. J Am Chem Soc 111(23):85768582
191. Allinger NL, Li FB, Yan LQ (1990) Molecular mechanics: The MM3 force-eld for
alkenes. J Comput Chem 11(7):848867
192. Allinger NL, Rahman M, Lii JH (1990) A molecular mechanics force-eld (MM3) for
alcohols and ethers. J Am Chem Soc 112:82938307
193. Lii JH, Allinger NL (1991) The MM3 force-eld for amides, polypeptides and proteins. J Comput Chem 12(2):186199
194. Allinger NL (1998) Force Fields: MM3. In: von Ragu Schleyer P (ed) Encyclopedia
of computational chemistry, vol 2. Wiley, Chichester, pp 10281033
261
195. Allinger NL, Chen KS, Lii JH (1996) An improved force eld (MM4) for saturated
hydrocarbons. J Comput Chem 17(56):642668
196. Langley CH, Lii JH, Allinger NL (2001) Molecular mechanics (MM4) calculations on
carbonyl compounds. I. Aldehydes. J Comput Chem 22(13):13961425
197. Langley CH, Lii JH, Allinger NL (2001) Molecular mechanics calculations on carbonyl compounds. II. Open-chain ketones. J Comput Chem 22(13):14261450
198. Langley CH, Lii JH, Allinger NL (2001) Molecular mechanics calculations on carbonyl compounds. III. Cycloketones. J Comput Chem 22(13):14511475
199. Langley CH, Lii JH, Allinger NL (2001) Molecular mechanics calculations on carbonyl compounds. IV. Heats of formation. J Comput Chem 22(13):14761483
200. Langley CH, Lii JH, Allinger N (2003) Molecular mechanics (MM4) calculations on
carbonyl compounds. I-IV (Eratum). J Comput Chem 24(10):12831286
201. Langley CH, Allinger NL (2002) Molecular mechanics (MM4) calculations on
amides. J Phys Chem A 106(23):56385652
202. Allinger NL, Chen KH, Lii JH, Durkin KA (2003) Alcohols, ethers, carbohydrates,
and related compounds. I. The MM4 force eld for simple compounds. J Comput
Chem 24(12):14471472
203. Lii JH, Chen KH, Durkin KA, Allinger NL (2003) Alcohols, ethers, carbohydrates,
and related compounds. II. The anomeric effect. J Comput Chem 24(12):1473
1489
204. Lii JH, Chen KH, Grindley TB, Allinger NL (2003) Alcohols, ethers, carbohydrates,
and related compounds. III. The 1,2-dimethoxyethane system. J Comput Chem
24(12):14901503
205. Lii JH, Chen KH, Allinger NL (2003) Alcohols, ethers, carbohydrates, and related
compounds. IV. Carbohydrates. J Comput Chem 24(12):15041513
206. Halgren TA (1996) Merck molecular force eld. 1. Basis, form, scope, parameterization, and performance of MMFF94. J Comput Chem 17(56):490519
207. Halgren TA (1996) Merck molecular force eld. 2. MMFF94 van der Waals and
electrostatic parameters for intermolecular interactions. J Comput Chem 17(56):
520552
208. Halgren TA (1996) Merck molecular force eld. 3. Molecular geometries and vibrational frequencies for MMFF94. J Comput Chem 17(56):553586
209. Halgren TA, Nachbar RB (1996) Merck molecular force eld. 4. Conformational energies and geometries for MMFF94. J Comput Chem 17(56):587615
210. Halgren TA (1996) Merck molecular force eld. 5. Extension of MMFF94 using
experimental data, additional computational data, and empirical rules. J Comput
Chem 17(56):616641
211. Halgren TA (1998) Force Fields: MMFF94. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 2. Wiley, Chichester, pp 10331035
212. Halgren TA (1999) MMFF VI. MMFF94s option for energy minimization studies.
213. Halgren TA (1999) MMFF VII. Characterization of MMFF94, MMFF94s, and other
widely available force elds for conformational energies and for intermolecular interaction energies and geometries. J Comput Chem 20(7):730748
214. Rapp AK, Casewit CJ, Colwell KS, Goddard WA III, Skiff WM (1992) UFF, a full
periodic table force eld for molecular mechanics and molecular dynamics simulations. J Am Chem Soc 114:1002410035
215. Casewit CJ, Colwell KS, Rapp AK (1992) Application of a universal force eld to
organic molecules. J Am Chem Soc 114:1003510046
262
H.M. Senn W. Thiel
216. Casewit CJ, Colwell KS, Rapp AK (1992) Application of a universal force eld to
main group compounds. J Am Chem Soc 114:1004610053
217. Rapp AK, Colwell KS, Casewit CJ (1993) Application of a universal force eld to
metal complexes. Inorg Chem 32:34383450
218. Case DA, Cheatham TE, Darden T, Gohlke H, Luo R, Merz KM Jr, Onufriev A,
Simmerling C, Wang B, Woods RJ (2005) The AMBER biomolecular simulation programs. J Comput Chem 26(16):16681688
219. Jorgensen WL, Tirado-Rives J (2005) Molecular modeling of organic and biomolecular systems using BOSS and MCPRO. J Comput Chem 26(16):16891700
220. te Velde GT, Bickelhaupt FM, Baerends EJ, Guerra CF, van Gisbergen SJA, Snijders GJ, Ziegler T (2001) Chemistry with ADF. J Comput Chem 22(9):931967
221. Guest MF, Bush IJ, van Dam HJJ, Sherwood P, Thomas JMH, van Lenthe JH,
Havenith RWA, Kendrick J (2005) The GAMESS-UK electronic structure package:
algorithms, developments and applications. Mol Phys 103(68):719747
222. Kendall RA, Apra E, Bernholdt DE, Bylaska EJ, Dupuis M, Fann GI, Harrison RJ,
Ju JL, Nichols JA, Nieplocha J, Straatsma TP, Windus TL, Wong AT (2000) High performance computational chemistry: an overview of NWChem a distributed parallel
application. Comput Phys Commun 128(12):260283
223. Colombo MC, Guidoni L, Laio A, Magistrato A, Maurer P, Piana S, Rhrig U,
Spiegel K, Sulpizi M, VandeVondele J, Zumstein M, Rthlisberger U (2002) Hybrid
QM/MM Car-Parrinello simulations of catalytic and enzymatic reactions. Chimia
56(12):1319 2002.
224. Sebastiani D, Rthlisberger U (2003) Advances in density-functional-based modeling techniques recent extensions of the Car-Parrinello approach. In: Carloni P,
Alber F (eds) Quantum medicinal chemistry. Methods and principles in medicinal
chemistry, vol 17. Wiley, Weinheim, pp 539
225. Woo TK, Margl PM, Blchl PE, Ziegler T (1997) A combined CarParrinello QM/MM
implementation for ab initio molecular dynamics simulations of extended systems:
application to transition metal catalysis. J Phys Chem B 101:78777880
226. Woo TK, Margl PM, Deng L, Ziegler T (1998) A combined Car-Parrinello quantum
mechanicalmolecular mechanical implementation for ab initio molecular dynamics simulations of extended systems. In: Gao J, Thompson MA (eds) Combined
quantum mechanical and molecular mechanical methods. ACS symposium series,
vol 712. American Chemical Society, Washington, DC, pp 128147
227. Woo TK, Margl PM, Deng L, Cavallo L, Ziegler T (1999) Towards more realistic computational modeling of homogenous catalysis by density functional theory:
Combined QM/MM and ab initio molecular dynamics. Catal Today 50:479500
228. Thiel W (2004) MNDO99, V. 6.1. Max-Planck-Institut fr Kohlenforschung, Mlheim
an der Ruhr, Germany
229. Gale JD, Rohl AL (2003) The general utility lattice program (GULP). Mol Simul
29(5):291341
230. Gale JD (2005) GULP: Capabilities and prospects. Z Kristallogr 220(56):552554
231. Warshel A (1991) Computer modeling of chemical reactions in enzymes and solutions. Wiley, New York
232. qvist J, Warshel A (1993) Simulation of enzyme reactions using valence bond force
elds and other hybrid quantum/classical approaches. Chem Rev 93:25232544
233. Shurki A, Warshel A (2003) Structure/function correlations of proteins using MM,
QM/MM, and related approaches: Methods, concepts, pitfalls, and current progress.
In: Dagett V (ed) Protein simulations. Advances in protein chemistry, vol 66. Academic Press, San Diego, pp 249313
263
234. Warshel A (2003) Computer simulations of enzyme catalysis: Methods, progress, and
insights. Annu Rev Biophys Biomol Struct 32:425443
235. Jensen JH, Day PN, Gordon MS, Basch H, Cohen D, Garmer DR, Krauss M,
Stevens WJ (1994) Effective fragment method for modeling intermolecular
hydrogen-bonding effects on quantum mechanical calculations. In: Smith DA (ed)
Modeling the hydrogen bond. ACS symposium series, vol 569. American Chemical
Society, Washington, DC, pp 139151
236. Day PN, Jensen JH, Gordon MS, Webb SP, Stevens WJ, Krauss M, Garmer D, Basch H,
Cohen D (1996) An effective fragment method for modeling solvent effects in quantum mechanical calculations. J Chem Phys 105(5):19681986
237. Kairys V, Jensen JH (2000) QM/MM boundaries across covalent bonds: a frozen
localized molecular orbital-based approach for the effective fragment potential
method. J Phys Chem A 104(28):66566665
238. Gordon MS, Freitag MA, Bandyopadhyay P, Jensen JH, Kairys V, Stevens WJ (2001)
The effective fragment potential method: a QM-based MM approach to modeling
environmental effects in chemistry. J Phys Chem A 105(2):293307
239. Grigorenko BL, Nemukhin AV, Topol IA, Burt SK (2002) Modeling of biomolecular systems with the quantum mechanical and molecular mechanical method based
on the effective fragment potential technique: proposal of exible fragments. J Phys
Chem A 106(44):1066310672
240. Nemukhin AV, Grigorenko BL, Bochenkova AV, Topol IA, Burt SK (2002) A QM/MM
approach with effective fragment potentials applied to the dipeptide-water structures. THEOCHEM 581(13):167175
241. Nemukhin AV, Grigorenko BL, Topol IA, Burt SK (2003) Flexible effective fragment QM/MM method: validation through the challenging tests. J Comput Chem
24(12):14101420
242. Ryde U, Olsen L, Nilsson K (2002) Quantum chemical geometry optimizations in
proteins using crystallographic raw data. J Comput Chem 23(11):10581070
243. Ryde U, Nilsson K (2003) Quantum chemistry can locally improve protein crystal
structures. J Am Chem Soc 125(47):1423214233
244. Ryde U, Nilsson K (2003) Quantum renementa combination of quantum chemistry and protein crystallography. THEOCHEM 632(13):259275
245. Tronrud DE (2004) Introduction to macromolecular renement. Acta Crystallogr D:
Biol Crystallogr 60:21562168
246. Kleywegt GJ, Henrick K, Dodson EJ, van Aalten DMF (2003) Pound-wise but pennyfoolish: How well do micromolecules fare in macromolecular renement? Structure
11(9):10511059
247. Klhn M, Braun-Sand S, Rosta E, Warshel A (2005) On possible pitfalls in ab initio quantum mechanics/molecular mechanics minimization approaches for studies
of enzymatic reactions. J Phys Chem B 109(32):1564515650
248. Zhang Y, Kua J, McCammon JA (2003) Inuence of structural uctuation on enzyme
reaction energy barriers in combined quantum mechanical/molecular mechanical
studies. J Phys Chem B 107(18):44594463
249. Schlegel HB (1995) Geometry optimization on potential energy surfaces. In:
Yarkony DR (ed) Modern electronic structure theory. Advanced series in physical
chemistry, vol 2/I. World Scientic, Singapore, pp 459500
250. Schlegel HB (1998) Geometry optimization I. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 2. Wiley, Chichester, pp 11361142
251. Schlick T (1998) Geometry optimization II. In: von Ragu Schleyer P (ed) Encyclopedia of computational chemistry, vol 2. Wiley, Chichester, pp 11421157
264
H.M. Senn W. Thiel
252. Jensen F (1998) Transition structure optimization techniques. In: von Ragu
Schleyer P (ed) Encyclopedia of computational chemistry, vol 5. Wiley, Chichester,
pp 31142123
253. Schlegel HB (2003) Exploring potential energy surfaces for chemical reactions: an
overview of some practical methods. J Comput Chem 24(12):15141527
254. Farkas , Schlegel HB (1998) Methods for geometry optimization of large molecules.
I. An O(N 2 ) algorithm for solving systems of linear equations for the transformation
of coordinates and forces. J Chem Phys 109(17):71007104
255. Farkas , Schlegel HB (1999) Methods for optimizing large molecules. Part II.
Quadratic search. J Chem Phys 111(24):1080610814
256. Farkas , Schlegel HB (2002) Methods for optimizing large molecules. Part III. An
improved algorithm for geometry optimization using direct inversion in the iterative
subspace (GDIIS). Phys Chem Chem Phys 4(1):1115
257. Farkas , Schlegel HB (2003) Geometry optimization methods for modeling large
molecules. THEOCHEM 666:3139
258. Paizs B, Fogarasi G, Pulay P (1998) An efcient direct method for geometry optimization of large molecules in internal coordinates. J Chem Phys 109(16):65716576
259. Baker J, Kinghorn D, Pulay P (1999) Geometry optimization in delocalized internal
coordinates: an efcient quadratically scaling algorithm for large molecules. J Chem
Phys 110:49864991
260. Paizs B, Baker J, Suhai S, Pulay P (2000) Geometry optimization of large
biomolecules in redundant internal coordinates. J Chem Phys 113(16):65666572
261. Nmeth K, Coulaud O, Monard G, ngyn JG (2000) Linear scaling algorithm for
the coordinate transformation problem of molecular geometry optimization. J Chem
Phys 113(14):55985603
262. Nmeth K, Coulaud O, Monard G, ngyn JG (2001) An efcient method for the coordinate transformation problem of massively three-dimensional networks. J Chem
Phys 114(22):97479753
263. Prat-Resina X, Garcia-Viloca M, Monard G, Gonzlez-Lafont A, Lluch JM, Boll JM,
Anglada JM (2002) The search for stationary points on a quantum mechanical/molecular mechanical potential-energy surface. Theor Chem Acc 107(3):147153
264. Billeter SR, Turner AJ, Thiel W (2000) Linear scaling geometry optimization and
transition state search in hybrid delocalized internal coordinates. Phys Chem Chem
Phys 2:21772186
265. Baker J, Kessi A, Delley B (1996) The generation and use of delocalized internal
coordinates in geometry optimization. J Chem Phys 105:192212
266. Pulay P, Fogarasi G (1992) Geometry optimization in redundant internal coordinates. J Chem Phys 96(4):28562860
267. Moliner V, Turner AJ, Williams IH (1997) Transition-state structural renement with
GRACE and CHARMM: realistic modeling of lactate dehydrogenase using a combined quantum/classical method. Chem Commun (14):12711272
268. Turner AJ, Moliner V, Williams IH (1999) Transition-state structural renement with
GRACE and CHARMM: exible QM/MM modeling for lactate dehydrogenase. Phys
Chem Chem Phys 1(6):13231331
269. Prat-Resina X, Gonzlez-Lafont , Lluch JM (2003) How important is the renement
of transition state structures in enzymatic reactions? THEOCHEM 632(13):297
307
270. Vreven T, Morokuma K, Farkas , Schlegel HB, Frisch MJ (2003) Geometry optimization with QM/MM, ONIOM, and other combined methods. I: Microiterations
and constraints. J Comput Chem 24(6):760769
265
271. Vreven T, Frisch MJ, Kudin KN, Schlegel HB, Morokuma K (2006) Geometry optimization with QM/MM methods. II: Explicit quadratic coupling. Mol Phys 104(57):
701714
272. Hayashi S, Ohmine I (2000) Proton transfer in bacteriorhodopsin: structure, excitation, IR spectra, and potential energy surface analyses by an ab initio QM/MM
method. J Phys Chem B 104(45):1067810691
273. Zhang Y, Liu H, Yang W (2000) Free energy calculation on enzyme reactions with an
efcient iterative procedure to determine minimum energy paths on a combined ab
initio QM/MM potential energy surface. J Chem Phys 112(8):34833492
274. Prat-Resina X, Boll JM, Gonzlez-Lafont , Lluch JM (2004) Geometry optimization and transition state search in enzymes: different options in the microiterative
method. Int J Quant Chem 98(4):367377
275. Mart S, Moliner V, Tun I, Williams IH (2005) Computing kinetic isotope effects
for chorismate mutase with high accuracy: a new DFT/MM strategy. J Phys Chem B
109(9):37073710
276. Mart S, Moliner V, Tun I (2005) Improving the QM/MM description of chemical
processes: a dual level strategy to explore the potential energy surface in very large
systems. J Chem Theory Comput 1(5):10081016
277. Mart S, Moliner V, Tun I (2006) Improving the QM/MM description of chemical
processes: a dual level strategy to explore the potential energy surface in very large
systems (Erratum). J Chem Theory Comput 2(1):216
278. Mills G, Jnsson H (1994) Quantum and thermal effects in H2 dissociative adsorption: evaluation of free-energy barriers in multidimensional quantum systems. Phys
Rev Lett 72(7):11241127
279. Jnsson H, Mills G, Jacobsen KW (1998) Nudged elastic band method for nding minimum energy paths of transitions. In: Berne BJ, Ciccotti G, Coker DF (eds)
Classical and quantum dynamics in condensed phase simulations. World Scientic,
Singapore, pp 387404
280. Henkelman G, Uberuaga BP, Jnsson H (2000) A climbing image nudged elastic
band method for nding saddle points and minimum energy paths. J Chem Phys
113(22):99019904
281. Henkelman G, Jnsson H (2000) Improved tangent estimate in the nudged elastic
band method for nding minimum energy paths and saddle points. J Chem Phys
113(22):99789985
282. Henkelman G, Jhannesson G, Jnsson H (2002) Methods for nding saddle points
and minimum energy paths. In: Schwartz SD (ed) Theoretical methods in condensed
phase chemistry. Progress in theoretical chemistry and physics, vol 5. Kluwer, Dordrecht, pp 269300
283. Xie L, Liu H, Yang W (2004) Adapting the nudged elastic band method for determining minimum-energy paths of chemical reactions in enzymes. J Chem Phys
120(17):80398052
284. Ayala PY, Schlegel HB (1997) A combined method for determining reaction paths,
minima, and transition state geometries. J Chem Phys 107:375384
285. Liu H, Lu Z, Cisneros GA, Yang W (2004) Parallel iterative reaction path optimization in ab initio quantum mechanical/molecular mechanical modeling of enzyme
reactions. J Chem Phys 121(2):697706
286. Cisneros GA, Liu H, Lu Z, Yang W (2005) Reaction path determination for quantum mechanical/molecular mechanical modeling of enzyme reactions by combining rst order and second order chain-of-replicas methods. J Chem Phys
122(11):114502/17
266
H.M. Senn W. Thiel
287. Woodcock HL, Hodoscek M, Sherwood P, Lee YS, Schaefer HF III, Brooks BR (2003)
Exploring the quantum mechanical/molecular mechanical replica path method:
a pathway optimization of the chorismate to prephenate Claisen rearrangement catalyzed by chorismate mutase. Theor Chem Acc 109(3):140148
288. Allen MP, Tildesley DJ (1987) Computer simulation of liquids. Oxford University
Press, Oxford
289. Beveridge DL, DiCapua FM (1989) Free energy via molecular simulation: application
to chemical and biomolecular systems. Annu Rev Biophys Biophys Chem 18:431492
290. Straatsma TP, McCammon JA (1992) Computational alchemy. Annu Rev Phys Chem
43:407435
291. Frenkel D, Smit B (2002) Understanding molecular simulation, 2nd ed. Computational science series, vol 1. Academic Press, San Diego
292. Adcock SA, McCammon JA (2006) Molecular dynamics: survey of methods for simulating the activity of proteins. Chem Rev 106:15891615
293. Bash PA, Field MJ, Karplus M (1987) Free energy perturbation method for chemical reactions in the condensed phase: a dynamic approach based on a combined
quantum and molecular mechanics potential. J Am Chem Soc 109:80928094
294. Gao J, Xia X (1992) A priori evaluation of aqueous polarization effects through
Monte Carlo QM-MM simulations. Science 258(5082):631635
295. Gao J (1992) Absolute free-energy of solvation from Monte-Carlo simulations using
combined quantum and molecular mechanical potentials. J Phys Chem 96(2):537
540
296. Stanton RV, Hartsough DS, Merz KM Jr (1993) Calculation of solvation free energies using a density functional/molecular dynamics coupled potential. J Phys Chem
97(46):1186811870
297. Hartsough DS, Merz KM Jr (1995) Potential of mean force calculations on the SN 1
fragmentation of tert-butyl chloride. J Phys Chem 99(1):384390
298. Stanton RV, Little LR, Merz KM Jr (1995) An examination of a HartreeFock/
molecular mechanical coupled potential. J Phys Chem 99(48):1734417348
299. Stanton RV, Hartsough DS, Merz KM Jr (1995) An examination of a densityfunctional/molecular mechanical coupled potential. J Comput Chem 16(1):113128
300. Hartsough DS, Merz KM Jr (1995) Dynamic force eld models: molecular dynamics
simulations of human carbonic anhydrase II using a quantum mechanical/molecular
mechanical coupled potential. J Phys Chem 99(28):1126611275
301. Senn HM, Thiel S, Thiel W (2005) Enzymatic hydroxylation in p-hydroxybenzoate
hydroxylase: a case study for QM/MM molecular dynamics. J Chem Theory Comput
1(3):494505
302. Schlegel HB, Millam JM, Iyengar SS, Voth GA, Daniels AD, Scuseria GE, Frisch MJ
(2001) Ab initio molecular dynamics: propagating the density matrix with Gaussian
orbitals. J Chem Phys 114(22):97589763
303. Iyengar SS, Schlegel HB, Millam JM, Voth GA, Scuseria GE, Frisch MJ (2001) Ab
initio molecular dynamics: Propagating the density matrix with Gaussian orbitals.
II. Generalizations based on mass-weighting, idempotency, energy conservation and
choice of initial conditions. J Chem Phys 115(22):1029110302
304. Schlegel HB, Iyengar SS, Li XS, Millam JM, Voth GA, Scuseria GE, Frisch MJ
(2002) Ab initio molecular dynamics: propagating the density matrix with Gaussian orbitals. III. Comparison with BornOppenheimer dynamics. J Chem Phys
117(19):86948704
305. Iyengar SS, Schlegel HB, Voth GA, Millam JM, Scuseria GE, Frisch MJ (2002) Ab
initio molecular dynamics: propagating the density matrix with Gaussian orbitals.
306.
307.
308.
309.
310.
311.
312.
313.
314.
315.
316.
317.
318.
319.
320.
321.
322.
323.
267
IV. Formal analysis of the deviations from BornOppenheimer dynamics. Isr J Chem
42(23):191202
Iyengar SS, Schlegel HB, Voth GA (2003) Atom-centered density matrix propagation
(ADMP): generalizations using Bohmian mechanics. J Phys Chem A 107(37):7269
7277
Rega N, Iyengar SS, Voth GA, Schlegel HB, Vreven T, Frisch MJ (2004) Hybrid
ab-initio/empirical molecular dynamics: combining the ONIOM scheme with the
atom-centered density matrix propagation (ADMP) approach. J Phys Chem B
108(13):42104220
Gonzalez-Lafont A, Truong TN, Truhlar DG (1991) Direct dynamics calculations
with neglect of diatomic differential overlap molecular orbital theory with specic
reaction parameters. J Phys Chem 95(12):46184627
Lau EY, Kahn K, Bash PA, Bruice TC (2000) The importance of reactant positioning in enzyme catalysis: a hybrid quantum mechanics/molecular mechanics study of
a haloalkane dehalogenase. Proc Natl Acad Sci USA 97(18):99379942
Cui Q, Karplus M (2001) Triosephosphate isomerase: a theoretical comparison of
alternative pathways. J Am Chem Soc 123(10):22842290
Cui Q, Karplus M (2002) Quantum mechanical/molecular mechanical studies of the
triosephosphate isomerase-catalyzed reaction: verication of methodology and analysis of reaction mechanisms. J Phys Chem B 106(7):17681798
Alhambra C, Corchado JC, Snchez ML, Gao JL, Truhlar DG (2000) Quantum dynamics of hydride transfer in enzyme catalysis. J Am Chem Soc 122:81978203
Devi-Kesavan LS, Garcia-Viloca M, Gao J (2003) Semiempirical QM/MM potential
with simple valence bond (SVB) for enzyme reactions. Application to the nucleophilic addition reaction in haloalkane dehalogenase. Theor Chem Acc 109(3):133
139
Corchado JC, Coitio EL, Chuang YY, Fast PL, Truhlar DG (1998) Interpolated variational transition-state theory by mapping. J Phys Chem A 102(14):24242438
Chuang YY, Corchado JC, Truhlar DG (1999) Mapped interpolation scheme for
single-point energy corrections in reaction rate calculations and a critical evaluation
of dual-level reaction path dynamics methods. J Phys Chem A 103(8):11401149
Ruiz-Perna JJ, Silla E, Tun I, Mart S, Moliner V (2004) Hybrid QM/MM potentials
of mean force with interpolated corrections. J Phys Chem B 108(24):84278433
Rod TH, Ryde U (2005) Quantum mechanical free energy barrier for an enzymatic
reaction. Phys Rev Lett 94(13):138302/14
Rod TH, Ryde U (2005) Accurate QM/MM free energy calculations of enzyme
reactions: methylation by catechol O-methyltransferase. J Chem Theory Comput
1(6):12401251
McQuarrie DA (2000) Statistical mechanics. University Science Books, Sausalito, CA
Zwanzig RW (1954) High-temperature equation of state by a perturbation method.
I. Nonpolar gases. J Chem Phys 22(8):14201426
Kstner J, Senn HM, Thiel S, Otte N, Thiel W (2006) QM/MM free-energy perturbation compared to thermodynamic integration and umbrella sampling: application to
an enzymatic reaction. J Chem Theory Comput 2(2):452461
Chandrasekhar J, Smith SF, Jorgensen WL (1984) SN 2 Reaction proles in the gas
phase and aqueous solution. J Am Chem Soc 106(10):30493050
Chandrasekhar J, Smith SF, Jorgensen WL (1985) Theoretical examination of the SN 2
reaction involving chloride ion and methyl chloride in the gas phase and aqueous
solution. J Am Chem Soc 107(1):154163
268
H.M. Senn W. Thiel
324. Zheng YJ, Merz KM Jr (1992) Mechanism of the human carbonic anhydrase II catalyzed hydration of carbon dioxide. J Am Chem Soc 114(26):1049810507
325. Stanton RV, Perkyl M, Bakowies D, Kollman PA (1998) Combined ab initio and
free energy calculations to study reactions in enzymes and solution: amide hydrolysis in trypsin and aqueous solution. J Am Chem Soc 120(14):34483457
326. Bentzien J, Florin J, Glennon TM, Warshel A (1998) Quantum mechanicalmolecular mechanical approaches for studying chemical reactions in proteins and
solution. In: Gao J, Thompson MA (eds) Combined quantum mechanical and molecular mechanical methods. ACS symposium series, vol 712. American Chemical
Society, Washington, DC, pp 1634
327. Bentzien J, Muller RP, Florin J, Warshel A (1998) Hybrid ab initio quantum mechanics/molecular mechanics calculations of free energy surfaces for enzymic reactions: the nucleophilic attack in subtilisin. J Phys Chem B 102:22932301
328.
Strajbl M, Hong G, Warshel A (2002) Ab initio QM/MM simulation with proper sampling: rst principle calculations of the free energy of the autodissociation of water
in aqueous solution. J Phys Chem B 106:1333313343
329. Rosta E, Klhn M, Warshel A (2006) Towards accurate ab initio QM/MM calculations
of free-energy proles of enzymatic reactions. J Phys Chem B 110:29342941
330. Li G, Zhang X, Cui Q (2003) Free energy perturbation calculations with combined
QM/MM potentials complications, simplications, and applications to redox potential calculations. J Phys Chem B 107(33):86438653
331. Li G, Cui Q (2003) pKa Calculations with QM/MM free energy perturbations. J Phys
Chem B 107(51):1452114528
332. Yang W, Bitetti-Putzer R, Karplus M (2004) Chaperoned alchemical free energy simulations: a general method for QM, MM, and QM/MM potentials. J Chem Phys
120(20):94509453
333. Hu H, Yang W (2005) Dual-topology/dual-coordinate free-energy simulation using
QM/MM force eld. J Chem Phys 123(4):041102/14
334. Fixman M (1974) Classical statistical mechanics of constraints: a theorem and application to polymers. Proc Natl Acad Sci USA 71:30503053
335. Carter EA, Ciccotti G, Hynes JT, Kapral R (1989) Constrained reaction coordinate
dynamics for the simulation of rare events. Chem Phys Lett 156:472477
336. Sprik M, Ciccotti G (1998) Free energy from constrained molecular dynamics.
J Chem Phys 109:77377744
337. den Otter WK, Briels WJ (1998) The calculation of free-energy differences by constrained molecular-dynamics simulations. J Chem Phys 109:41394146
338. den Otter WK (2000) Thermodynamic integration of the free energy along a reaction coordinate in Cartesian coordinates. J Chem Phys 112:72837292
339. den Otter WK, Briels WJ (2000) Free energy from molecular dynamics with multiple
constraints. Mol Phys 98(12):773781
340. Darve E, Pohorille A (2001) Calculating free energies using average force. J Chem
Phys 115:91699183
341. Schlitter J, Klhn M (2003) A new concise expression for the free energy of a reaction
coordinate. J Chem Phys 118(5):20572060
342. Schlitter J, Klhn M (2003) The free energy of a reaction coordinate at multiple
constraints: a concise formulation. Mol Phys 101(2324):34393443
343. Isralewitz B, Baudry J, Gullingsrud J, Kosztin D, Schulten K (2001) Steered molecular
dynamics investigations of protein function. J Mol Graphics Model 19(1):1325
344. Isralewitz B, Gao M, Schulten K (2001) Steered molecular dynamics and mechanical
functions of proteins. Curr Opin Struct Biol 11(2):224230
269
345. Jarzynski C (1997) Nonequilibrium equality for free energy differences. Phys Rev
Lett 78:26902693
346. Jarzynski C (1997) Equilibrium free-energy differences from nonequilibrium measurements: a master-equation approach. Phys Rev E 56:50185035
347. Crooks GE (1998) Nonequilibrium measurements of free energy differences for microscopically reversible Markovian systems. J Stat Phys 90:14811487
348. Crooks GE (1999) Entropy production uctuation theorem and the nonequilibrium
work relation for free energy differences. Phys Rev E 60:27212726
349. Park S, Schulten K (2004) Calculating potentials of mean force from steered molecular dynamics simulations. J Chem Phys 120(13):59465961
350. Cuendet MA (2006) Statistical mechanical derivation of Jarzynskis identity for thermostated non-Hamiltonian dynamics. Phys Rev Lett 96(12):120602/14
351. Hendrix DA, Jarzynski C (2001) A fast-growth method of computing free energy
differences. J Chem Phys 114:59745981
352. Jensen M, Park S, Tajkhorshid E, Schulten K (2002) Energetics of glycerol conduction
through aquaglyceroporin GlpF. Proc Natl Acad Sci USA 99:67316736
353. Park S, Khalili-Araghi F, Tajkhorshid E, Schulten K (2003) Free energy calculation
from steered molecular dynamics simulations using Jarzynskis equality. J Chem
Phys 119:35593566
354. Hummer G, Szabo A (2001) Free energy reconstruction from nonequilibrium singlemolecule pulling experiments. Proc Natl Acad Sci USA 98:36583661
355. Jarzynski C (2001) How does a system respond when driven away from thermal
equilibrium? Proc Natl Acad Sci USA 98:36363638
356. Liphardt J, Dumont S, Smith SB, Tinoco I Jr, Bustamante C (2002) Equilibrium information from nonequilibrium measurements in an experimental test of Jarzynskis
equality. Science 296(5574):18321835
357. Egolf DA (2002) Statistical mechanics: far from equilibrium. Science 296(5574):1813
1815
358. Hummer G (2001) Fast-growth thermodynamic integration: error and efciency analysis. J Chem Phys 114:73307337
359. Hu H, Yun RH, Hermans J (2002) Reversibility of free energy simulations: slow
growth may have a unique advantage. (With a note on use of Ewald summation).
Mol Simul 28(12):6780
360. Zuckerman DM, Woolf TB (2002) Theory of a systematic computational error in free
energy differences. Phys Rev Lett 89(18):180602/14
361. Zuckerman DM, Woolf TB (2002) Overcoming nite-sampling errors in fastswitching free-energy estimates: extrapolative analysis of a molecular system. Chem
Phys Lett 351(5-6):445453
362. Zuckerman DM, Woolf TB (2004) Systematic nite-sampling inaccuracy in free energy differences and other nonlinear quantities. J Stat Phys 114(56):13031323
363. Ytreberg FM, Zuckerman DM (2004) Efcient use of nonequilibrium measurement to estimate free energy differences for molecular systems. J Comput Chem
25(14):17491759
364. Gore J, Ritort F, Bustamante C (2003) Bias and error in estimates of equilibrium
free-energy differences from nonequilibrium measurements. Proc Natl Acad Sci USA
100(22):1256412569
365. Fox RF (2003) Using nonequilibrium measurements to determine macromolecule
free-energy differences. Proc Natl Acad Sci USA 100(22):1253712538
366. Rodriguez-Gomez D, Darve E, Pohorille A (2004) Assessing the efciency of free
energy calculation methods. J Chem Phys 120(8):35633578
270
H.M. Senn W. Thiel
367. Lua RC, Grosberg AY (2005) Practical applicability of the Jarzynski relation in statistical mechanics: a pedagogical example. J Phys Chem B 109(14):68056811
368. Lechner W, Oberhofer H, Dellago C, Geissler PL (2006) Equilibrium free energies
from fast-switching trajectories with large time steps. J Chem Phys 124(4):044113/
112
369. Raugei S, Cascella M, Carloni P (2004) A procient enzyme: insights on the mechanism of orotidine monophosphate decarboxylase from computer simulations. J Am
Chem Soc 126(48):1573015737
370. Chandler D (1998) Finding transition pathways: throwing ropes over rough mountain passes, in the dark. In: Berne BJ, Ciccotti G, Coker DF (eds) Classical and
quantum dynamics in condensed phase simulations. World Scientic, Singapore,
pp 5166
371. Dellago C, Bolhuis PG, Csajka FS, Chandler D (1998) Transition path sampling and
the calculation of rate constants. J Chem Phys 108(5):19641977
372. Dellago C, Bolhuis PG, Chandler D (1998) Efcient transition path sampling: application to LennardJones cluster rearrangements. J Chem Phys 108(22):9236
9245
373. Bolhuis PG, Dellago C, Chandler D (1998) Sampling ensembles of deterministic transition pathways. Faraday Discuss 110:421436
374. Csajka FS, Chandler D (1998) Transition pathways in a many-body system: application to hydrogen-bond breaking in water. J Chem Phys 109(3):11251133
375. Dellago C, Bolhuis PG, Chandler D (1999) On the calculation of reaction rate constants in the transition path ensemble. J Chem Phys 110(14):66176625
376. Bolhuis PG, Dellago C, Geissler PL, Chandler D (2000) Transition path sampling:
throwing ropes over mountains in the dark. J Phys Condens Matter 12:A147A152
377. Geissler PL, Dellago C, Chandler D, Hutter J, Parrinello M (2001) Autoionization in
liquid water. Science 291(5511):21212124
378. Bolhuis PG, Chandler D, Dellago C, Geissler PL (2002) Transition path sampling:
throwing ropes over rough mountain passes, in the dark. Annu Rev Phys Chem
53:291318
379. Dellago C, Bolhuis PG, Geissler PL (2002) Transition path sampling. Adv Chem Phys
123:178
380. Zahn D (2005) Unprejudiced identication of reaction mechanisms from biased
transition path sampling. J Chem Phys 123(4):044104/17
381. Basner JE, Schwartz SD (2005) How enzyme dynamics helps catalyze a reaction in
atomic detail: a transition path sampling study. J Am Chem Soc 127:1382213831
382. Laio A, Parrinello M (2002) Escaping free-energy minima. Proc Natl Acad Sci USA
99(20):1256212566
383. Iannuzzi M, Laio A, Parrinello M (2003) Efcient exploration of reactive potential energy surfaces using Car-Parrinello molecular dynamics. Phys Rev Lett
90(23):238302/14
384. Micheletti C, Laio A, Parrinello M (2004) Reconstructing the density of states by
history-dependent metadynamics. Phys Rev Lett 92(17):170601/14
385. Wu Y, Schmitt JD, Car R (2004) Mapping potential energy surfaces. J Chem Phys
121(3):11931200
386. Laio A, Rodriguez-Fortea A, Gervasio FL, Ceccarelli M, Parrinello M (2005) Assessing the accuracy of metadynamics. J Phys Chem B 109(14):67146721
387. Ensing B, Laio A, Parrinello M, Klein ML (2005) A recipe for the computation of the
free energy barrier and the lowest free energy path of concerted reactions. J Phys
Chem B 109(14):66766687
271
388. Raiteri P, Laio A, Gervasio FL, Micheletti C, Parrinello M (2006) Efcient reconstruction of complex free energy landscapes by multiple walkers metadynamics. J Phys
Chem B 110(8):35333539
389. Bussi G, Laio A, Parrinello M (2006) Equilibrium free energies from nonequilibrium
metadynamics. Phys Rev Lett 96:090601/14
390. Ensing B, De Vivo M, Liu ZW, Moore P, Klein ML (2006) Metadynamics as a tool for
exploring free energy landscapes of chemical reactions. Acc Chem Res 39(2):7381
391. Huber T, Torda AE, Gunsteren WF (1994) Local elevation: a method for improving
the searching properties of molecular-dynamics simulation. J Comput-Aided Mol
Des 8(6):695708
392. Grubmller H (1995) Predicting slow structural transitions in macromolecular systems: conformational ooding. Phys Rev E 52(3):28932906
393. Mller EM, de Meijere A, Grubmller H (2002) Predicting unimolecular chemical
reactions: chemical ooding. J Chem Phys 116(3):897905
394. Wang F, Landau DP (2001) Efcient, multiple-range random walk algorithm to calculate the density of states. Phys Rev Lett 86(10):20502053
395. Wang F, Landau DP (2001) Determining the density of states for classical statistical models: a random walk algorithm to produce a at histogram. Phys Rev E
64(5):056101/116
396. Landau DP, Wang F (2002) Determining the density of states for classical statistical
models by a at-histogram random walk. Comput Phys Commun 147(12):674677
397. Raugei S, Carloni P (2006) Structure and function of vanadium haloperoxidases.
J Phys Chem B 110(8):37473758
398. Rosso L, Tuckerman ME (2002) An adiabatic molecular dynamics method for the
calculation of free energy proles. Mol Simul 28:91112
399. Rosso L, Minry P, Zhu Z, Tuckerman ME (2002) On the use of the adiabatic molecular dynamics technique in the calculation of free energy proles. J Chem Phys
116:43894402
400. VandeVondele J, Rothlisberger U (2002) Canonical adiabatic free energy sampling
(CAFES): a novel method for the exploration of free energy surfaces. J Phys Chem
B 106:203208
401. Piana S, Bucher D, Carloni P, Rothlisberger U (2004) Reaction mechanism of
HIV-1 protease by hybrid CarParrinello/classical MD simulations. J Phys Chem B
108(30):1113911149
402. Lu Z, Yang W (2004) Reaction path potential for complex systems derived from combined ab initio quantum mechanical and molecular mechanical calculations. J Chem
Phys 121(1):89100
403. Miller WH, Handy NC, Adams JE (1980) Reaction path Hamiltonian for polyatomic
molecules. J Chem Phys 72(1):99112
404. Nabuurs SB, Spronk CAEM, Krieger E, Vriend G, Hooft RWW (2004) Protein structures: what good is beauty if it cannot be seen? In: Bultinck P, Tollenaere JP,
De Winter H, Langenaeker W (eds) (2004) Computational medicinal chemistry for
drug discovery. Dekker, New York, pp 387403
405. Hooft RWW, Vriend G, Sander C, Abola EE (1996) Errors in protein structures.
Nature 381(6580):272
406. Word JM (2003) Reduce, V. 2.21. Biochemistry Department, Duke University,
Durham, NC, http://www.kinemage.biochem.duke.edu/software/reduce.php
407. Word JM, Lovell SC, Richardson JS, Richardson DC (1999) Asparagine and glutamine: using hydrogen atom contacts in the choice of side-chain amide orientation.
J Mol Biol 285(4):17351747
272
H.M. Senn W. Thiel
408. Garcia-Viloca M, Poulsen TD, Truhlar DG, Gao JL (2004) Sensitivity of molecular dynamics simulations to the choice of the X-ray structure used to model an enzymatic
reaction. Protein Sci 13(9):23412354
409. Vriend G (1990) WHAT IF: a molecular modeling and drug design program. J Mol
Graph 8(1):5256
410. Forrest LR, Honig B (2005) An assessment of the accuracy of methods for predicting hydrogen positions in protein structures. Proteins Struct Funct Bioinf 61(2):296
309
411. Jensen JH, Li H, Robertson AD, Molina PA (2005) Prediction and rationalization of
protein pKa values using QM and QM/MM methods. J Phys Chem A 109(30):6634
6643
412. Li H, Robertson AD, Jensen JH (2005) Very fast empirical prediction and rationalization of protein pKa values. Proteins Struct Funct Bioinf 61(4):704721
413. Zhang L, Hermans J (1996) Hydrophilicity of cavities in proteins. Proteins Struct
Funct Genet 24(4):433438
414. Olsson MHM, Hong GY, Warshel A (2003) Frozen density functional free energy
simulations of redox proteins: computational studies of the reduction potential of
plastocyanin and rusticyanin. J Am Chem Soc 125(17):50255039
415. Altun A, Thiel W (2005) Combined quantum mechanical/molecular mechanical
study on the pentacoordinated ferric and ferrous cytochrome P450cam complexes.
J Phys Chem B 109:12681280
416. Pauling L (1946) Molecular architecture and biological reactions. Chem Eng News
24:13751377
417. Vill J, Warshel A (2001) Energetics and dynamics of enzymatic reactions. J Phys
Chem B 105(33):78877907
418. Kollman PA, Kuhn B, Perkyl M (2002) Computational studies of enzyme-catalyzed
reactions: Where are we in predicting mechanisms and in understanding the nature
of enzyme catalysis? J Phys Chem B 106(7):15371542
419. Gao JL, Truhlar DG (2002) Quantum mechanical methods for enzyme kinetics. Annu
Rev Phys Chem 53:467505
420. Mart S, Roca M, Andrs J, Moliner V, Silla E, Tun I, Bertrn J (2004) Theoretical
insights in enzyme catalysis. Chem Soc Rev 33(2):98107
421. Garcia-Viloca M, Gao J, Karplus M, Truhlar DG (2004) How enzymes work: analysis
by modern rate theory and computer simulations. Science 303(5655):186195
422. Blake CCF, Mair GA, Northy ACT, Phillips DC, Sarma VR (1967) On the conformation of the hen egg-white lysozyme molecule. Proc R Soc London B 167:365377
423. Blake CCF, Johnson LN, Mair GA, North ACT, Phillips DC, Sarma VR (1967) Crystallographic studies of the activity of hen egg-white lysozyme. Proc R Soc London B
167:378388
424. Ford LO, Machin PA, Phillips DC, Tjian R, Johnson LN (1974) Crystal structure of
a lysozyme-tetrasaccharide lactone complex. J Mol Biol 88(2):349360
425. Tapia O, Andrs J, Safont VS (1994) Enzyme catalysis and transition structures
in vacuo. Transition structures for the enolization, carboxylation and oxygenation
reactions in ribulose-1,5-bisphosphate carboxylase/oxygenase enzyme (Rubisco).
J Chem Soc, Faraday Trans 90(16):23652374
426. Khanjin NA, Snyder JP, Menger FM (1999) Mechanism of chorismate mutase: contribution of conformational restriction to catalysis in the Claisen rearrangement. J Am
Chem Soc 121(50):1183111846
427. Vallee BL, Williams RJ (1968) Metalloenzymes: the entatic nature of their active sites.
Proc Natl Acad Sci USA 59(2):498505
273
428. Williams RJ (1971) The entatic state. Cold Spring Harb Symp Quant Biol 36:5362
429. Williams RJP (1985) Metalloenzyme catalysis: the entatic state. J Mol Catal 30(12):
126
430. Williams RJP (1995) Energized (entatic) states of groups and of secondary structures
in proteins and metalloproteins. Eur J Biochem 234(2):363381
431. Koshland DE (1958) Application of a theory of enzyme specicity to protein synthesis. Proc Natl Acad Sci USA 44(2):98104
432. Storm DR, Koshland DE (1970) A source for the special catalytic power of enzymes:
orbital steering. Proc Natl Acad Sci USA 66(2):445452
433. Mesecar AD, Stoddard BL, Koshland DE (1997) Orbital steering in the catalytic
power of enzymes: small structural changes with large catalytic consequences. Science 277(5323):202206
434. Menger FM (1993) Enzyme reactivity from an organic perspective. Acc Chem Res
26(4):206212
435. Kollman PA, Kuhn B, Donini O, Perakyla M, Stanton R, Bakowies D (2001) Elucidating the nature of enzyme catalysis utilizing a new twist on an old methodology:
quantum mechanical-free energy calculations on chemical reactions in enzymes and
in aqueous solution. Acc Chem Res 34(1):7279
436. Bruice TC, Benkovic SJ (2000) Chemical basis for enzyme catalysis. Biochemistry
39:62676274
437. Bruice TC (2002) A view at the millennium: the efciency of enzymatic catalysis. Acc
Chem Res 35:139148
438. Hur S, Bruice TC (2003) The near attack conformation approach to the study of the
chorismate to prephenate reaction. Proc Natl Acad Sci USA 100(21):1201512020
439. Hur S, Bruice TC (2003) Enzymes do what is expected (chalcone isomerase versus
chorismate mutase). J Am Chem Soc 125(6):14721473
440. Hur S, Bruice TC (2003) Comparison of formation of reactive conformers (NACs) for
the Claisen rearrangement of chorismate to prephenate in water and in the E. coli
mutase: the efciency of the enzyme catalysis. J Am Chem Soc 125(19):59645972
441. Hur S, Bruice TC (2003) Just a near attack conformer for catalysis (chorismate to
prephenate rearrangements in water, antibody, enzymes, and their mutants). J Am
Chem Soc 125(35):1054010542
442. Zhang X, Zhang X, Bruice TC (2005) A denitive mechanism for chorismate mutase.
Biochemistry 44(31):1044310448
443. Repasky MP, Guimares CRW, Chandrasekhar J, Tirado-Rives J, Jorgensen WL
(2003) Investigation of solvent effects for the Claisen rearrangement of chorismate
to prephenate: mechanistic interpretation via near attack conformations. J Am Chem
Soc 125(23):66636672
444. Guimares CRW, Repasky MP, Chandrasekhar J, Tirado-Rives J, Jorgensen WL
(2003) Contributions of conformational compression and preferential transition
state stabilization to the rate enhancement by chorismate mutase. J Am Chem Soc
125(23):68926899
445. Ranaghan KE, Mulholland AJ (2004) Conformational effects in enzyme catalysis:
QM/MM free energy calculation of the NAC contribution in chorismate mutase.
Chem Commun (10):12381239
446. Shurki A,
Strajbl M, Vill J, Warshel A (2002) How much do enzymes really gain by
restraining their reacting fragments? J Am Chem Soc 124:40974107
447.
Strajbl M, Shurki A, Kato M, Warshel A (2003) Apparent NAC effect in chorismate
mutase reects electrostatic transition state stabilization. J Am Chem Soc 125:10228
10237
274
H.M. Senn W. Thiel
448. Lee JK, Houk KN (1997) A procient enzyme revisited: the predicted mechanism for
orotidine monophosphate decarboxylase. Science 276(5314):942945
449. Devi-Kesavan LS, Gao J (2003) Combined QM/MM study of the mechanism and
kinetic isotope effect of the nucleophilic substitution reaction in haloalkane dehalogenase. J Am Chem Soc 125(6):15321540
450. Page MI, Jencks WP (1971) Entropic contributions to rate accelerations in enzymic and intramolecular reactions and the chelate effect. Proc Natl Acad Sci USA
68(8):16781683
451. Blow D (2000) So do we understand how enzymes work? Structure 8:R77R81
452. Vill J, Strajbl M, Glennon TM, Sham YY, Chu ZT, Warshel A (2000) How important
are entropic contributions to enzyme catalysis? Proc Natl Acad Sci USA 97:11899
11904
453. Cleland WW, Kreevoy MM (1994) Low-barrier hydrogen-bonds and enzymatic catalysis. Science 264(5167):18871890
454. Frey PA, Whitt SA, Tobin JB (1994) A low-barrier hydrogen-bond in the catalytic
triad of serine proteases. Science 264(5167):19271930
455. Cleland WW, Frey PA, Gerlt JA (1998) The low barrier hydrogen bond in enzymatic
catalysis. J Biol Chem 273(40):2552925532
456. Pan YP, McAllister MA (1997) Characterization of low-barrier hydrogen bonds.
1. Microsolvation effects. An ab initio and DFT investigation. J Am Chem Soc
119(32):75617566
5. Microsolvation of enol-enolate. An ab initio and DFT investigation. J Org Chem
62(23):81718176
6. Cavity polarity effects on the formic acid formate anion model system. An ab
initio and DFT investigation. J Am Chem Soc 120(1):166169
459. Careri G, Fasella P, Gratton E (1979) Enzyme dynamics: the statistical physics approach. Annu Rev Biophys Bioeng 8:6997
460. Karplus M, McCammon JA (1983) Dynamics of proteins: elements and function.
Annu Rev Biochem 52:263300
461. Berendsen HJC, Hayward S (2000) Collective protein dynamics in relation to function. Curr Opin Struct Biol 10:165169
462. Eisenmesser EZ, Bosco DA, Akke M, Kern D (2002) Enzyme dynamics during catalysis. Science 295(5559):15201523
463. Olsson MHM, Parson WW, Warshel A (2006) Dynamical contributions to enzyme
catalysis: critical tests of a popular hypothesis. Chem Rev 106:17371756
464. Hwang JK, Chu ZT, Yadav A, Warshel A (1991) Simulations of quantum-mechanical
corrections for rate constants of hydride-transfer reactions in enzymes and solutions. J Phys Chem 95(22):84458448
465. Kohen A, Klinman JP (1999) Hydrogen tunneling in biology. Chem Biol 6(7):R191
R198
466. Sutcliffe MJ, Scrutton NS (2000) Enzyme catalysis: over-the-barrier or through-thebarrier? Trends Biochem Sci 25(9):405408
467. Truhlar DG, Gao J, Garcia-Viloca M, Alhambra C, Corchado J, Sanchez ML,
Poulsen TD (2004) Ensemble-averaged variational transition state theory with optimized multidimensional tunneling for enzyme kinetics and other condensed-phase
reactions. Int J Quant Chem 100(6):11361152
468. Zhang X, Houk KN (2005) Why enzymes are procient catalysts: beyond the Pauling
paradigm. Acc Chem Res 38(5):379385
275
469. Bash PA, Field MJ, Davenport RC, Petsko GA, Ringe D, and Karplus M (1991)
Computer-simulation and analysis of the reaction pathway of triosephosphate isomerase. Biochemistry 30:58265832
470. Mulholland AJ, Richards WG (1997) Acetyl-CoA enolization in citrate synthase:
a quantum mechanical/molecular mechanical (QM/MM) study. Proteins Struct
Funct Genet 27(1):925
471. Liu H, Zhang Y, Yang W (2000) How is the active site of enolase organized to catalyze
two different reaction steps? J Am Chem Soc 122(28):65606570
472. Cui Q, Elstner M, Karplus M (2002) A theoretical analysis of the proton and hydride transfer in liver alcohol dehydrogenase (LADH). J Phys Chem B 106(10):2721
2740
473. Hermann JC, Hensen C, Ridder L, Mulholland AJ, Hltje HD (2005) Mechanisms
of antibiotic resistance: QM/MM modeling of the acylation reaction of a class A
-lactamase with benzylpenicillin. J Am Chem Soc 127(12):44544465
474. Cheng Y, Zhang Y, McCammon JA (2005) How does the cAMP-dependent protein kinase catalyze the phosphorylation reaction: an ab initio QM/MM study. J Am Chem
Soc 127(5):15531562
475. Garcia-Viloca M, Truhlar DG, Gao J (2003) Importance of substrate and cofactor
polarization in the active site of dihydrofolate reductase. J Mol Biol 327(2):549560
476. Sokalski WA (1985) The physical nature of catalytic activity due to the molecular
environment in terms of intermolecular interaction theory: derivation of simplied
models. J Mol Catal 30(3):395410
477. Szefczyk B, Mulholland AJ, Ranaghan KE, Sokalski WA (2004) Differential
transition-state stabilization in enzyme catalysis: quantum chemical analysis of interactions in the chorismate mutase reaction and prediction of the optimal catalytic
eld. J Am Chem Soc 126(49):1614816159
478. Roca M, Mart S, Andrs J, Moliner V, Tun I, Bertrn J, Williams IH (2003) Theoretical modeling of enzyme catalytic power: analysis of cratic and electrostatic
factors in catechol O-methyltransferase. J Am Chem Soc 125(25):77267737
479. Ridder L, Palfey BA, Vervoort JIM, Rietjens CM (2000) Modeling avin and substrate substituent effects on the activation barrier and rate of oxygen transfer by
p-hydroxybenzoate hydroxylase. FEBS Lett 478(12):197201
480. Ridder L, Harvey JN, Rietjens IMCM, Vervoort J, Mulholland AJ (2003) Ab initio QM/MM modeling of the hydroxylation step in p-hydroxybenzoate hydroxylase.
J Phys Chem B 107(9):21182126
481. Billeter SR, Hanser CFW, Mordasini TZ, Scholten M, Thiel W, van Gunsteren WF
(2001) Molecular dynamics study of oxygenation reactions catalysed by the enzyme
p-hydroxybenzoate hydroxylase. Phys Chem Chem Phys 3:688695
482. Ridder L, Mulholland AJ, Rietjens IMCM, Vervoort J (2000) A quantum mechanical/molecular mechanical study of the hydroxylation of phenol and halogenated
derivatives by phenol hydroxylase. J Am Chem Soc 122(36):87288738
483. Poulsen TD, Garcia-Viloca M, Gao J, Truhlar DG (2003) Free energy surface, reaction paths, and kinetic isotope effect of short-chain acyl-CoA dehydrogenase. J Phys
Chem B 107(35):95679578
484. Alhambra C, Corchado J, Snchez ML, Garcia-Viloca M, Gao J, Truhlar DG (2001)
Canonical variational theory for enzyme kinetics with the protein mean force and
multidimensional quantum mechanical tunneling dynamics. Theory and application
to liver alcohol dehydrogenase. J Phys Chem B 105:1132611340
485. Tresadern G, Faulder PF, Gleeson MP, Tai Z, MacKenzie G, Burton NA, Hillier IH
(2003) Recent advances in quantum mechanical/molecular mechanical calculations
276
486.
487.
488.
489.
490.
491.
492.
493.
494.
495.
496.
497.
498.
499.
500.
H.M. Senn W. Thiel

of enzyme catalysis: hydrogen tunnelling in liver alcohol dehydrogenase and inhibition of elastase by -ketoheterocycles. Theor Chem Acc 109(3):108117
Walker RC, de Souza MM, Mercer IP, Gould IR, Klug DR (2002) Large and fast relaxations inside a protein: calculation and measurement of reorganization energies
in alcohol dehydrogenase. J Phys Chem B 106(44):1165811665
Formaneck MS, Li G, Zhang X, Cui Q (2002) Calculating accurate redox potentials
in enzymes with a combined QM/MM free energy perturbation approach. J Theor
Comput Chem 1(1):5367
Wymore T, Hempel J, Cho SS, MacKerell AD Jr, Nicholas HB Jr, Deereld DW II
(2004) Molecular recognition of aldehydes by aldehyde dehydrogenase and mechanism of nucleophile activation. Proteins Struct Funct Bioinf 57(4):758771
Wymore T, Deereld DW II, Field MJ, Hempel J, Nicholas HB Jr (2003) Initial
catalytic events in class 3 aldehyde dehydrogenase: MM and QM/MM simulations.
Chem-Biol Interact 143144:7584
Moliner V, Williams IH (2000) Flexible QM/MM modelling embraces alternative
mechanisms for lactate dehydrogenase. Chem Commun (19):18431844
Ferrer S, Silla E, Tun I, Oliva M, Moliner V, Williams IH (2005) Dependence of
enzyme reaction mechanism on protonation state of titratable residues and QM level
description: lactate dehydrogenase. Chem Commun (47):58735875
Ferrer S, Ruiz-Perna JJ, Tun I, Moliner V, Garcia-Viloca M, Gonzlez-Lafont A,
Lluch JM (2005) A QM/MM exploration of the potential energy surface of pyruvate
to lactate transformation catalyzed by LDH. Improving the accuracy of semiempirical descriptions. J Chem Theory Comput 1(4):750761
Cummins PL, Gready JE (2000) QM/MM and SCRF studies of the ionization state
of 8-methylpterin substrate bound to dihydrofolate reductase: existence of a lowbarrier hydrogen bond. J Mol Graphics Model 18(1):4249
Cummins PL, Gready JE (2000) Combined quantum and molecular mechanics
(QM/MM) study of the ionization state of 8-methylpterin substrate bound to dihydrofolate reductase. J Phys Chem B 104(18):45034510
Beierlein F, Lanig H, Schrer G, Horn AHC, Clark T (2003) Quantum mechanical/molecular mechanical (QM/MM) docking: an evaluation for known test systems.
Mol Phys 101(15):24692480
Titmuss SJ, Cummins PL, Bliznyuk AA, Rendell AP, Gready JE (2000) Comparison of linear-scaling semiempirical methods and combined quantum mechanical/molecular mechanical methods applied to enzyme reactions. Chem Phys Lett
320(12):169176
Cummins PL, Greatbanks SP, Rendell AP, Gready JE (2002) Computational methods
for the study of enzymic reaction mechanisms. 1. Application to the hydride transfer step in the catalysis of dihydrofolate reductase. J Phys Chem B 106(38):9934
9944
Titmuss SJ, Cummins PL, Rendell AP, Bliznyuk AA, Gready JE (2002) Comparison of linear-scaling semiempirical methods and combined quantum mechanical/molecular mechanical methods for enzymic reactions. II. An energy decomposition analysis. J Comput Chem 23(14):13141322
Cummins PL, Gready JE (2003) Computational methods for the study of enzymic reaction mechanisms. II. An overlapping mechanically embedded method for hybrid
semi-empirical-QM/MM calculations. THEOCHEM 632(13):247257
Garcia-Viloca M, Truhlar DG, Gao J (2003) Reaction-path energetics and kinetics
of the hydride transfer reaction catalyzed by dihydrofolate reductase. Biochemistry
42(46):1355813575
277
501. Thorpe IF, Brooks CL III (2003) Barriers to hydride transfer in wild type and mutant
dihydrofolate reductase from E. coli. J Phys Chem B 107(50):1404214051
502. Cummins PL, Gready JE (2005) Computational methods for the study of enzymic
reaction mechanisms III: a perturbation plus QM/MM approach for calculating relative free energies of protonation. J Comput Chem 26(6):561568
503. Ferrer S, Silla E, Tun I, Mart S, Moliner V (2003) Catalytic mechanism of dihydrofolate reductase enzyme. A combined quantum-mechanical/molecular-mechanical
characterization of the N5 protonation step. J Phys Chem B 107(50):1403614041
504. Proust-De Martin F, Dumas R, Field MJ (2000) A hybrid-potential free-energy study
of the isomerization step of the acetohydroxy acid isomeroreductase reaction. J Am
Chem Soc 122(32):76887697
505. Alhambra C, Snchez ML, Corchado J, Gao J, Truhlar DG (2001) Quantum mechanical tunneling in methylamine dehydrogenase. Chem Phys Lett 347:512518
506. Faulder PF, Tresadern G, Chohan KK, Scrutton NS, Sutcliffe MJ, Hillier IH, Burton NA (2001) QM/MM studies show substantial tunneling for the hydrogen-transfer
reaction in methylamine dehydrogenase. J Am Chem Soc 123(35):86048605
507. Tresadern G, Nunez S, Faulder PF, Wang H, Hillier IH, Burton NA (2003) Direct dynamics calculations of reaction rate and kinetic isotope effects in enzyme catalysed
reactions. Faraday Discuss 122:223242
508. Tresadern G, Wang H, Faulder PF, Burton NA, Hillier IH (2003) Extreme tunnelling
in methylamine dehydrogenase revealed by hybrid QM/MM calculations: potential
energy surface prole for methylamine and ethanolamine substrates and kinetic
isotope effect values. Mol Phys 101(17):27752784
509. Torrent M, Vreven T, Musaev DG, Morokuma K, Farkas , Schlegel HB (2002) Effects of the protein environment on the structure and energetics of active sites of
metalloenzymes. ONIOM study of methane monooxygenase and ribonucleotide reductase. J Am Chem Soc 124(2):192193
510. Hoffmann M, Khavrutskii IV, Musaev DG, Morokuma K (2004) Protein effects on the
O2 binding to the active site of the methane monooxygenase: ONIOM studies. Int
J Quant Chem 99(6):972980
511. Friesner RA, Baik M-H, Gherman BF, Guallar V, Wirstam M, Murphy RB, Lippard SJ (2003) How iron-containing proteins control dioxygen chemistry: a detailed
atomic level description via accurate quantum chemical and mixed quantum mechanics/molecular mechanics calculations. Coord Chem Rev 238239:267290
512. Gherman BF, Lippard SJ, Friesner RA (2005) Substrate hydroxylation in methane
monooxygenase: quantitative modeling via mixed quantum mechanics/molecular
mechanics techniques. J Am Chem Soc 127:10251037
513. Shiota Y, Yoshizawa K (2004) QM/MM study of the mononuclear non-heme iron
active site of phenylalanine hydroxylase. J Phys Chem B 108(44):1722617237
514. Nemukhin AV, Grigorenko BL, Topol IA, Burt SK (2006) Modeling dioxygen binding
to the non-heme iron-containing enzymes. Int J Quant Chem 106(10):21842190
515. Bathelt CM, Mulholland AJ, Harvey JN (2005) QM/MM studies of the electronic
structure of the compound I intermediate in cytochrome c peroxidase and ascorbate
peroxidase. Dalton Trans (21):34703476
516. Schneboom JC, Thiel W (2004) The resting state of P450cam : a QM/MM study.
J Phys Chem B 108(22):74687478
517. Swart M, Groenhof AR, Ehlers AW, Lammertsma K (2005) Substrate binding in the
active site of cytochrome P450CAM. Chem Phys Lett 403(13):3541
518. Schneboom JC, Lin H, Reuter N, Thiel W, Cohen S, Ogliaro F, Shaik S (2002) The
elusive oxidant species of cytochrome P450 enzymes: characterization by combined
278
519.
520.
521.
522.
523.
524.
525.
526.
527.
528.
529.
530.
531.
532.
533.
534.
535.
H.M. Senn W. Thiel

quantum mechanical/molecular mechanical (QM/MM) calculations. J Am Chem Soc
124(27):81428151
Bathelt CM, Zurek J, Mulholland AJ, Harvey JN (2005) Electronic structure of compound I in human isoforms of cytochrome P450 from QM/MM modeling. J Am
Chem Soc 127:1290012908
Cohen S, Kumar D, Shaik S (2006) In silico design of a mutant of cytochrome P450
containing selenocysteine. J Am Chem Soc 128:26492653
Schneboom JC, Neese F, Thiel W (2005) Toward identication of the compound I
reactive intermediate in cytochrome P450 chemistry: a QM/MM study of its EPR and
Mssbauer parameters. J Am Chem Soc 127(16):58405853
Schneboom JC, Cohen S, Lin H, Shaik S, Thiel W (2004) Quantum mechanical/molecular mechanical investigation of the mechanism of CH hydroxylation of
camphor by cytochrome P450cam : theory supports a two-state rebound mechanism.
J Am Chem Soc 126(12):40174034
Altun A, Guallar V, Friesner RA, Shaik S, Thiel W (2006) The effect of heme environment on the hydrogen abstraction reaction of camphor in P450cam catalysis:
a QM/MM study. J Am Chem Soc 128:39243925
Lin H, Schneboom JC, Cohen S, Shaik S, Thiel W (2004) QM/MM study of
the product-enzyme complex in P450cam catalysis. J Phys Chem B 108(28):10083
10088
Guallar V, Friesner RA (2004) Cytochrome P450CAM enzymatic catalysis cycle:
a quantum mechanics/molecular mechanics study. J Am Chem Soc 126(27):8501
8508
Shaik S, Kumar D, de Visser SP, Altun A, Thiel W (2005) Theoretical perspective on
the structure and mechanism of cytochrome P450 enzymes. Chem Rev 105(6):2279
2328
Derat E, Cohen S, Shaik S, Altun A, Thiel W (2005) Principal active species of
horseradish peroxidase, compound I: a hybrid quantum mechanical/molecular mechanical study. J Am Chem Soc 127(39):1361113621
Kamachi T, Yoshizawa K (2005) Water-assisted oxo mechanism for heme metabolism.
J Am Chem Soc 127(30):1068610692
Mart MA, Crespo A, Bari SE, Doctorovich FA, Estrin DA (2004) QMMM study of
nitrite reduction by nitrite reductase of Pseudomonas aeruginosa. J Phys Chem B
108(46):1807318080
Kllrot N, Nilsson K, Rasmussen T, Ryde U (2005) Theoretical study of structure of
catalytic copper site in nitrite reductase. Int J Quant Chem 102:520541
Morao I, Tai Z, Hillier IH, Burton NA (2003) What form of Ng -hydroxy-L-arginine is
the intermediate in the mechanism of NO synthase? QM and QM/MM calculations of
substrate-active site interactions. THEOCHEM 632(13):277285
Fernndez ML, Mart MA, Crespo A, Estrin DA (2005) Proximal effects in the modulation of nitric oxide synthase reactivity: a QM-MM study. J Biol Inorg Chem
10:595604
Crespo A, Mart MA, Kalko SG, Morreale A, Orozco M, Gelpi JL, Luque FJ, Estrin DA
(2005) Theoretical study of the truncated hemoglobin HbN: exploring the molecular
basis of the NO detoxication mechanism. J Am Chem Soc 127(12):44334444
Marti MA, Crespo A, Capece L, Boechi L, Bikiel DE, Scherlis DA, Estrin DA (2006)
Dioxygen afnity in heme proteins investigated by computer simulation. J Inorg
Biochem 100(4):761770
Wirstam M, Lippard SJ, Friesner RA (2003) Reversible dioxygen binding to hemerythrin. J Am Chem Soc 125(13):39803987
279
536. Rovira C, Schulze B, Eichinger M, Evanseck JD, Parrinello M (2001) Inuence of the
heme pocket conformation on the structure and vibrations of the Fe-CO bond in
myoglobin: a QM/MM density functional study. Biophys J 81(1):435445
537. Freindorf M, Shao Y, Brown ST, Kong J, Furlani TR (2006) A combined density
functional theory and molecular mechanics (QM/MM) study of FeCO vibrations in
carbonmonoxy myoglobin. Chem Phys Lett 419(46):563566
538. Rovira C (2003) Applications of Car-Parrinello molecular dynamics in biochemistry
binding of ligands in myoglobin. In: Carloni P, Alber F (eds) Quantum medicinal
chemistry. Methods and principles in medicinal chemistry, vol 17. Wiley, Weinheim,
pp 73112
539. Harvey JN (2004) Spin-forbidden CO ligand recombination in myoglobin. Faraday
Discuss 127:165177
540. Guallar V, Jarzecki AA, Friesner RA, Spiro TG (2006) Modeling of ligation-induced
helix/loop displacements in myoglobin: toward an understanding of hemoglobin
allostery. J Am Chem Soc 128(16):54275435
541. Strickland N, Mulholland AJ, Harvey JN (2006) The FeCO bond energy in myoglobin: a QM/MM study of the effect of tertiary structure. Biophys J 90(4):L27L29
542. Kravitz JY, Pecoraro VL, Carlson HA (2005) Quantum mechanics/molecular mechanics calculations of the vanadium dependent chloroperoxidase. J Chem Theory
Comput 1(6):12651274
543. Rulsek L, Solomon EI, Ryde U (2005) A combined quantum and molecular mechanical study of the O2 reductive cleavage in the catalytic cycle of multicopper oxidases.
Inorg Chem 44(16):56125628
544. Rothlisberger U, Carloni P, Doclo K, Parrinello M (2000) A comparative study of
galactose oxidase and active site analogs based on QM/MM Car-Parrinello simulations. J Biol Inorg Chem 5(2):236250
545. Kamachi T, Kihara N, Shiota Y, Yoshizawa K (2005) Computational exploration of the
catalytic mechanism of dopamine -monooxygenase: modeling of its mononuclear
copper active sites. Inorg Chem 44(12):42264236
546. Yoshizawa K, Kihara N, Kamachi T, Shiota Y (2006) Catalytic mechanism of
dopamine -monooxygenase mediated by Cu(III)Oxo. Inorg Chem 45(7):3034
3041
547. Sundararajan M, Hillier IH, Burton NA (2006) Structure and redox properties of
the protein, rubredoxin, and its ligand and metal mutants studied by electronic
structure calculation. J Phys Chem A 110(2):785790
548. Magistrato A, DeGrado WF, Laio A, Rothlisberger U, VandeVondele J, Klein ML
(2003) Characterization of the dizinc analogue of the synthetic diiron protein
DF1 using ab initio and hybrid quantum/classical molecular dynamics simulations.
J Phys Chem B 107(17):41824188
549. Ryde U, Olsson MHM (2001) Structure, strain, and reorganization energy of blue
copper models in the protein. Int J Quant Chem 81(5):335347
550. Moon S, Patchkovskii S, Salahub DR (2003) QM/MM calculations of EPR hyperne
coupling constants in blue copper proteins. THEOCHEM 632:287295
551. Datta SN, Sudhamsu J, Pandey A (2004) Theoretical determination of the standard
reduction potential of plastocyanin in vitro. J Phys Chem B 108(23):80078016
552. Prabhakar R, Musaev DG, Khavrutskii IV, Morokuma K (2004) Does the active site
of mammalian glutathione peroxidase (GPx) contain water molecules? An ONIOM
study. J Phys Chem B 108(34):1264312645
553. Bergs J, Rickards G, Rauk A, Houe-Levin C (2006) QM/MM study of electron
addition on protein disulde bonds. Chem Phys Lett 421:6367
280
H.M. Senn W. Thiel
554. Chu J-W, Brooks BR, Trout BL (2004) Oxidation of methionine residues in aqueous
solutions: free methionine and methionine in granulocyte colony-stimulating factor.
J Am Chem Soc 126(50):1660116607
555. Ridder L, Rietjens IMCM, Vervoort J, Mulholland AJ (2002) Quantum mechanical/molecular mechanical free energy simulations of the glutathione S-transferase
(M1-1) reaction with phenanthrene 9,10-oxide. J Am Chem Soc 124(33):99269936
556. Ruggiero GD, Williams IH, Roca M, Moliner V, Tun I (2004) QM/MM determination of kinetic isotope effects for COMT-catalyzed methyl transfer does not support
compression hypothesis. J Am Chem Soc 126(28):86348635
557. Roca M, Moliner V, Ruiz-Perna JJ, Silla E, Tun I (2006) Activation free energy
of catechol O-methyltransferase. Corrections to the potential of mean force. J Phys
Chem A 110(2):503509
558. Roca M, Andrs J, Moliner V, Tun I, Bertrn J (2005) On the nature of the
transition state in catechol O-methyltransferase. A complementary study based on
molecular dynamics and potential energy surface explorations. J Am Chem Soc
127(30):1064810655
559. Hu P, Zhang Y (2006) Catalytic mechanism and product specity of the histone
lysin methyltransferase SET7/9: an ab initio QM/MM-FE study with multiple initial
structures. J Am Chem Soc 128(4):12721278
560. Nez S, Antoniou D, Schramm VL, Schwartz SD (2004) Promoting vibrations in human purine nucleoside phosphorylase. A molecular dynamics and hybrid quantum
mechanical/molecular mechanical study. J Am Chem Soc 126(48):1572015729
561. Sheppard DW, Burton NA, Hillier IH (2000) Ab initio hybrid quantum mechanical/molecular mechanical studies of the mechanisms of the enzymes protein kinase
and thymidine phosphorylase. THEOCHEM 506:3544
562. Gao X-F, Huang X-R, Sun C-C (2006) Role of each residue in catalysis in the active
site of pyrimidine nucleoside phosphorylase from Bacillus subtilis: a hybrid QM/MM
study. J Struct Biol 154(1):2026
563. Cheng Y, Zhang Y, McCammon JA (2006) How does activation loop phosphorylation modulate catalytic activity in the cAMP-dependent protein kinase: a theoretical
study. Protein Sci 15(4):672683
564. Rungrotmongkol T, Hannongbua S, Mulholland A (2004) Mechanistic study of
HIV-1 reverse transcriptase at the active site based on QM/MM method. J Theor
Comput Chem 3(4):491500
565. Thomas A, Field MJ (2002) Reaction mechanism of the HGXPRTase from Plasmodium falciparum: a hybrid potential quantum mechanical/molecular mechanical
study. J Am Chem Soc 124(42):1243212438,
566. Mulholland AJ, Lyne PD, Karplus M (2000) Ab initio QM/MM study of the citrate
synthase mechanism. A low-barrier hydrogen bond is not involved. J Am Chem Soc
122(3):534535
567. Kurz LC, Fite B, Jean J, Park J, Erpelding T, Callis P (2005). Photophysics of tryptophan uorescence: link with the catalytic strategy of the citrate synthase from
Thermoplasma acidophilum. Biochemistry 44(5):13941413
568. Senn HM, OHagan D, Thiel W (2005) Insight into enzymatic CF bond formation
from QM and QM/MM calculations. J Am Chem Soc 127:1364313655
569. Schrer G, Lanig H, Clark T (2004) Aeromonas proteolytica aminopeptidase: an
investigation of the mode of action using a quantum mechanical/molecular mechanical approach. Biochemistry 43(18):54145427
570. Schrer G, Horn AHC, Gedeck P, Clark T (2002) The reaction mechanism of bovine
lens leucine aminopeptidase. J Phys Chem B 106(34):88158830
281
571. Klein CDP, Schiffmann R, Folkers G, Piana S, Rthlisberger U (2003) Protonation

states of methionine aminopeptidase and their relevance for inhibitor binding and
catalytic activity. J Biol Chem 278(48):4786247867
572. Cross JB, Vreven T, Meroueh SO, Mobashery S, Schlegel HB (2005) Computational
investigation of irreversible inactivation of the zinc-dependent protease carboxypeptidase A. J Phys Chem B 109(10):47614769
573. Phoon L, Burton NA (2005) Assessment of a mechanism for reactive inhibition of
carboxypeptidase A with QM/MM methods. J Mol Graphics Model 24(2):94101
574. Molina PA, Sikorski RS, Jensen JH (2003) NMR chemical shifts in the low-pH form of
-chymotrypsin. A QM/MM and ONIOM-NMR study. Theor Chem Acc 109(3):100
107
575. Molina PA, Jensen JH (2003) A predictive model of strong hydrogen bonding in proteins: the N1 -H-O1 hydrogen bond in low-pH -chymotrypsin and -lytic protease.
J Phys Chem B 107(25):62266233
576. Pieraccini S, Sironi M, Colombo G (2006) Modeling enzymatic processes: a molecular simulation analysis of the origins of regioselectivity. Chem Phys Lett 418:373
376
577. Antonczak S, Monard G, Ruiz-Lpez M, Rivail JL (2000) Insights in the peptide
hydrolysis mechanism by thermolysin: a theoretical QM/MM study. J Mol Model
6:527538
578. Nemukhin AV, Grigorenko BL, Rogov AV, Topol IA, Burt SK (2004) Modeling
of serine protease prototype reactions with the exible effective fragment potential quantum mechanical/molecular mechanical method. Theor Chem Acc 111(1):
3648
579. Ishida T, Kato S (2003) Theoretical perspectives on the reaction mechanism of serine
proteases: the reaction free energy proles of the acylation process. J Am Chem Soc
125(39):1203512048
580. Ishida T, Kato S (2004) Role of Asp102 in the catalytic relay system of serine proteases: a theoretical study. J Am Chem Soc 126(22):71117118
581. Ishida T (2006) Low-barrier hydrogen bond hypothesis in the catalytic triad residue
of serine proteases: correlation between structural rearrangement and chemical
shifts in the acylation process. Biochemistry 45(17):54135420
582. Grter F, Schwarzl SM, Dejaegere A, Fischer S, Smith JC (2005) Protein/ligand binding free energies calculated with quantum mechanics/molecular mechanics. J Phys
Chem B 109:1047410483
583. Topf M, Vrnai P, Richards WG (2002) Ab initio QM/MM dynamics simulation of the
tetrahedral intermediate of serine proteases: insights into the active site hydrogenbonding network. J Am Chem Soc 124(49):1478014788
584. Topf M, Richards WG (2004) Theoretical studies on the deacylation step of serine
protease catalysis in the gas phase, in solution, and in elastase. J Am Chem Soc
126(44):1463114641
585. Gleeson MP, Hillier IH, Burton NA (2004) Theoretical analysis of peptidyl -ketoheterocyclic inhibitors of human neutrophil elastase: insight into the mechanism
of inhibition and the application of QM/MM calculations in structure-based drug
design. Org Biomol Chem 2(16):22752280
586. Guo H, Wlodawer A, Guo H (2005) A general acid-base mechanism for the stabilization of a tetrahedral adduct in a serine-carboxyl peptidase: a computational study.
J Am Chem Soc 127:1566215663
587. Xu Q, Guo H, Wlodawer A, Guo H (2006) The importance of dynamics in substrateassisted catalysis and specicity. J Am Chem Soc 128(18):59945995
282
H.M. Senn W. Thiel
588. Hensen C, Hermann JC, Nam K, Ma S, Gao J, Hltje H-D (2004) A combined
QM/MM approach to protein-ligand interactions: polarization effects of the HIV-1
protease on selected high afnity inhibitors. J Med Chem 47(27):66736680
589. Sulpizi M, Laio A, VandeVondele J, Cattaneo A, Rothlisberger U, Carloni P (2003)
Reaction mechanism of caspases: insights from QM/MM CarParrinello simulations.
Proteins Struct Funct Genet 52(2):212224
590. Corminboeuf C, Hu P, Tuckerman ME, Zhang Y (2006) Unexpected deacetylation
mechanism suggested by a density functional theory QM/MM study of histonedeacetylase-like protein. J Am Chem Soc 128(14):45304531
591. Madison V, Duca J, Bennett F, Bohanon S, Cooper A, Chu M, Desai J, Girijavallabhan V, Hare R, Hruza A, Hendrata S, Huang Y, Kravec C, Malcolm B, McCormick J,
Miesel L, Ramanathan L, Reichert P, Saksena A, Wang J, Weber PC, Zhu H, Fischmann T (2002) Binding afnities and geometries of various metal ligands in
peptide deformylase inhibitors. Biophys Chem 101102:239247
592. Khandelwal A, Lukacova V, Comez D, Kroll DM, Raha S, Balaz S (2005) A combination of docking, QM/MM methods, and MD simulation for binding afnity
estimation of metalloprotein ligands. J Med Chem 48(17):54375447
593. Pitarch J, Pascual-Ahuir J-L, Silla E, Tun I (2000) A quantum mechanics/molecular
mechanics study of the acylation reaction of TEM1 -lactamase and penicillanate.
J Chem Soc, Perkin Trans 2 (4):761767
594. Daz N, Surez D, Sordo TL, Merz KM Jr (2001) Acylation of class A -lactamases
by penicillins: a theoretical examination of the role of serine 130 and the -lactam
carboxylate group. J Phys Chem B 105(45):1130211313
595. Hermann JC, Ridder L, Mulholland AJ, Hltje H-D (2003) Identication of Glu166 as
the general base in the acylation reaction of class A -lactamases through QM/MM
modeling. J Am Chem Soc 125(32):95909591
596. Castillo R, Silla E, Tun I (2002) Role of protein exibility in enzymatic catalysis: quantum mechanical-molecular mechanical study of the deacylation reaction in
class A -lactamases. J Am Chem Soc 124(8):18091816
597. Meroueh SO, Fisher JF, Schlegel HB, Mobashery S (2005) Ab initio QM/MM study of
class A -lactamase acylation: dual participation of Glu166 and Lys73 in a concerted
base promotion of Ser70. J Am Chem Soc 127:1539715407
598. Hermann JC, Ridder L, Hltje H-D, Mulholland AJ (2006) Molecular mechanisms of
antibiotic resistance: QM/MM modelling of deacylation in a class A -lactamase. Org
Biomol Chem 4:206210
599. Gherman BF, Goldberg SD, Cornish VW, Friesner RA (2004) Mixed quantum
mechanical/molecular mechanical (QM/MM) study of the deacylation reaction in
a penicillin binding protein (PBP) versus in a class C -lactamase. J Am Chem Soc
126(24):76527664
600. Olsen L, Rasmussen T, Hemmingsen L, Ryde U (2004) Binding of benzylpenicillin to
metallo--lactamase: a QM/MM study. J Phys Chem B 108(45):1763917648
601. Gu W, Zhu J, Liu H (2002) Different protonation states of the Bacillus cereus binuclear zinc metallo--lactamase active site studied by combined quantum mechanical
and molecular mechanical simulations. J Theor Comput Chem 1(1):6980
602. Dal Peraro M, Llarrull LI, Rothlisberger U, Vila AJ, Carloni P (2004) Water-assisted
reaction mechanism of monozinc -lactamases. J Am Chem Soc 126(39):12661
12668
603. Xu D, Zhou Y, Xie D, Guo H (2005) Antibiotic binding to monozinc CphA -lactamase from Aeromonas hydropila: quantum mechanical/molecular mechanical and
density functional theory studies. J Med Chem 48(21):66796689
283
604. Park H, Brothers EN, Merz KM Jr (2005) Hybrid QM/MM and DFT investigations
of the catalytic mechanism and inhibition of the dinuclear zinc metallo--lactamase
CcrA from Bacteroides fragilis. J Am Chem Soc 127(12):42324241
605. Li J, Cross JB, Vreven T, Meroueh SO, Mobashery S, Schlegel HB (2005) Lysine carboxylation in proteins: OXA-10 -lactamase. Proteins Struct Funct Bioinf 61(2):246
257
606. Oliva M, Dideberg O, Field MJ (2003) Understanding the acylation mechanisms of
active-site serine penicillin-recognizing proteins: a molecular dynamics simulation
study. Proteins Struct Funct Bioinf 53(1):88100
607. Daz N, Surez D, Sordo TL, Merz KM Jr (2001) A theoretical study of the aminolysis
reaction of lysine 199 of human serum albumin with benzylpenicillin: consequences
for immunochemistry of penicillins. J Am Chem Soc 123(31):75747583
608. Lodola A, Mor M, Hermann JC, Tarzia G, Piomelli D, Mulholland AJ (2005) QM/MM
modelling of oleamide hydrolysis in fatty acid amide hydrolase (FAAH) reveals
a new mechanism of nucleophile activation. Chem Commun (35):43994401
609. Zhang Y, Kua J, McCammon JA (2002) Role of the catalytic triad and oxyanion
hole in acetylcholinesterase catalysis: an ab initio QM/MM study. J Am Chem Soc
124(35):1057210577
610. Hurley MM, Wright JB, Lushington GH, White WE (2003) Quantum mechanics and
mixed quantum mechanics/molecular mechanics simulations of model nerve agents
with acetylcholinesterase. Theor Chem Acc 109(3):160168
611. Gao D, Zhan C-G (2005) Modeling evolution of hydrogen bonding and stabilization of transition states in the process of cocaine hydrolysis catalyzed by human
butyrylcholinesterase. Proteins Struct Funct Bioinf 62:99110
612. Zhan C-G, Gao D (2005) Catalytic mechanism and energy barriers for butyrylcholinesterase-catalyzed hydrolysis of cocaine. Biophys J 89:38633872
613. Funke SA, Otte N, Eggert T, Bocola M, Jaeger K-E, Thiel W (2005) Combination
of computational prescreening and experimental library construction can accelerate
enzyme optimization by directed evolution. Protein Eng Des Sel 18(11):509514
614. Dinner AR, Blackburn GM, Karplus M (2001) Uracil-DNA glycosylase acts by substrate autocatalysis. Nature 413(6857):752755
615. Banerjee A, Yang W, Karplus M, Verdine GL (2005) Structure of a repair enzyme
interrogating undamaged DNA elucidates recognition of damaged DNA. Nature
434:612618
616. Biarns X, Nieto J, Planas A, Rovira C (2006) Substrate distortion in the Michaelis
complex of bacillus 1,31,4--glucanase: insight from rst principles molecular dynamics simulations. J Biol Chem 281:14321441
617. Dittrich M, Hayashi S, Schulten K (2003) On the mechanism of ATP hydrolysis in
F1 -ATPase. Biophys J 85(4):22532266
618. Dittrich M, Hayashi S, Schulten K (2004) ATP hydrolysis in the TP and DP catalytic
sites of F1 -ATPase. Biophys J 87(5):29542967
619. Li G, Cui Q (2004) Mechanochemical coupling in myosin: a theoretical analysis with
molecular dynamics and combined QM/MM reaction path calculations. J Phys Chem
B 108(10):33423357
620. Lopez X, York DM, Dejaegere A, Karplus M (2002) Theoretical studies on the hydrolysis of phosphate diesters in the gas phase, solution, and RNase A. Int J Quant Chem
86(1):1026
621. Topol IA, Cachau RE, Nemukhin AV, Grigorenko BL, Burt SK (2004) Quantum chemical modeling of the GTP hydrolysis by the RASGAP protein complex. Biochim
Biophys Acta 1700(1):125136
284
H.M. Senn W. Thiel
622. Grigorenko BL, Nemukhin AV, Topol IA, Cachau RE, Burt SK (2005) QM/MM modeling the RasGAP catalyzed hydrolysis of guanosine triphosphate. Proteins Struct
Funct Bioinf 60(3):495503
623. Klhn M, Schlitter J, Gerwert K (2005) Theoretical IR spectroscopy based on QM/MM
calculations provides changes in charge distribution, bond lengths, and bond angles of
the GTP ligand induced by the Ras-protein. Biophys J 88(6):38293844
624. De Vivo M, Ensing B, Klein ML (2005) Computational study of phosphatase activity
in soluble epoxide hydrolase: high efciency through a water bridge mediated proton
shuttle. J Am Chem Soc 127(32):1122611227
625. Gleeson MP, Burton NA, Hillier IH (2003) Prediction of the potency of inhibitors of
adenosine deaminase by QM/MM calculations. Chem Commun (17):21802181
626. Gleeson MP, Burton NA, Hillier IH (2003) The mechanism of adenosine deaminase
catalysis studied by QM/MM calculations: the role of histidine 238 and the activity
of the alanine 238 mutant. Phys Chem Chem Phys 5(19):42724278
627. Xu Q, Guo H (2004) Quantum mechanical/molecular mechanical molecular dynamics simulations of cytidine deaminase: from stabilization of transition state
analogues to catalytic mechanisms. J Phys Chem B 108(7):24772483
628. Guo H, Rao N, Xu Q, Guo H (2005) Origin of tight binding of a near-perfect
transition-state analogue by cytidine deaminase: implications for enzyme catalysis.
J Am Chem Soc 127(9):31913197
629. Xu D, Guo H, Gao J, Cui Q (2004) A QM/MM study of a nucleophilic aromatic substitution reaction catalyzed by 4-chlorobenzoyl-CoA dehalogenase. Chem Commun
(7):892893
630. Xu D, Wei Y, Wu J, Dunaway-Mariano D, Guo H, Cui Q, Gao J (2004) QM/MM studies
of the enzyme-catalyzed dechlorination of 4-chlorobenzoyl-CoA provide insight into
reaction energetics. J Am Chem Soc 126(42):1364913658
631. Wu J, Xu D, Lu X, Wang C, Guo H, Dunaway-Mariano D (2006) Contributions of
long-range electrostatic interactions to 4-chlorobenzoyl-CoA dehalogenase catalysis:
a combined theoretical and experimental study. Biochemistry 45:102112
632. Soriano A, Silla E, Tun I, Mart S, Moliner V, Bertrn J (2004) Electrostatic effects in enzyme catalysis: a quantum mechanics/molecular mechanics study of the
nucleophilic substitution reaction in haloalkane dehalogenase. Theor Chem Acc
112(4):327334
633. Nam K, Prat-Resina X, Garcia-Viloca M, Devi-Kesavan LS, Gao J (2004) Dynamics
of an enzymatic substitution reaction in haloalkane dehalogenase. J Am Chem Soc
126(5):13691376
634. Zhang X, Harrison DHT, Cui Q (2002) Functional specicities of methylglyoxal synthase and triosephosphate isomerase: a combined QM/MM analysis. J Am Chem Soc
124(50):1487114878
635. Guimares CRW, Udier-Blagovic M, Jorgensen WL (2005) Macrophomate synthase:
QM/MM simulations address the DielsAlder versus Michaelaldol reaction mechanism. J Am Chem Soc 127(10):35773588
636. Sicinska D, Truhlar DG, Paneth P (2005) Dependence of transition state structure
on substrate: the intrinsic C-13 kinetic isotope effect is different for physiological
and slow substrates of the ornithine decarboxylase reaction because of different
hydrogen bonding structures. J Am Chem Soc 127(15):54145422
637. Lundberg M, Blomberg MRA, Siegbahn PEM (2004) Developing active site models
of ODCase from large quantum models to a QM/MM approach. In: Lee JK (ed)
Orotidine monophosphate decarboxylase: a mechanistic dialogue. Topics in current
chemistry, vol 238. Springer, Berlin Heidelberg New York, pp 79112
285
638. Loferer MJ, Tautermann CS, Loefer HH, Liedl KR (2003) Inuence of backbone
conformations of human carbonic anhydrase II on carbon dioxide hydration: hydration pathways and binding of bicarbonate. J Am Chem Soc 125(29):89218927
639. Garcia-Viloca M, Nam K, Alhambra C, Gao J (2004) Solvent and Protein effects on
the vibrational frequency shift and energy relaxation of the azide ligand in carbonic
anhydrase. J Phys Chem B 108(35):1350113512
640. Tautermann CS, Loferer MJ, Voegele AF, Liedl KR (2003) About the kinetic feasibility of the Lipscomb mechanism in human carbonic anhydrase II. J Phys Chem B
107(43):1201312020
641. Kamachi T, Toraya T, Yoshizawa K (2004) Catalytic roles of active-site amino acid
residues of coenzyme B12 -dependent diol dehydratase: protonation state of histidine
and pull effect of glutamate. J Am Chem Soc 126(49):1620716216
642. Cisneros GA, Liu H, Zhang Y, Yang W (2003) Ab initio QM/MM study shows there
is no general acid in the reaction catalyzed by 4-oxalocrotonate tautomerase. J Am
Chem Soc 125(34):1038410393
643. Cisneros GA, Wang M, Silinski P, Fitzgerald MC, Yang W (2004) The protein backbone makes important contributions to 4-oxalocrotonate tautomerase enzyme catalysis: understanding from theory and experiment. Biochemistry 43(22):68856892
644. Cisneros GA, Wang M, Silinski P, Fitzgerald MC, Yang W (2006) Theoretical and
experimental determination on two substrates turned over by 4-oxalocrotonate tautomerase. J Phys Chem A 110(2):700708
645. Garcia-Viloca M, Gonzlez-Lafont , Lluch JM (2001) A QM/MM study of the
racemization of vinylglycolate catalyzed by mandelate racemase enzyme. J Am Chem
Soc 123(4):709721
646. Prat-Resina X, Garcia-Viloca M, Gonzlez-Lafont A, Lluch JM (2002) On the modulation of the substrate activity for the racemization catalyzed by mandelate racemase
enzyme: a QM/MM study. Phys Chem Chem Phys 4(21):53655371
647. Prat-Resina X, Gonzlez-Lafont , Lluch JM (2005) Reaction mechanism of the mandelate anion racemization catalyzed by mandelate racemase enzyme: a QM/MM
molecular dynamics free energy study. J Phys Chem B 109:2108921101
648. Puig E, Garcia-Viloca M, Gonzlez-Lafont , Lluch JM (2006) On the ionization state
of the substrate in the active site of glutamate racemase. A QM/MM study about the
importance of being zwitterionic. J Phys Chem A 110:717725
649. Guallar V, Jacobson M, McDermott A, Friesner RA (2004) Computational modeling
of the catalytic reaction in triosephosphate isomerase. J Mol Biol 337(1):227239
650. Cui Q, Karplus M (2002) Promoting modes and demoting modes in enzymecatalyzed proton transfer reactions: a study of models and realistic systems. J Phys
Chem B 106(32):79277947
651. Cui Q, Karplus M (2002) Quantum mechanics/molecular mechanics studies of
triosephosphate isomerase-catalyzed reactions: effect of geometry and tunneling on
proton-transfer rate constants. J Am Chem Soc 124:30933124
652. Wang M, Lu Z, Yang W (2002) Transmission coefcient calculation for proton transfer in triosephosphate isomerase based on the reaction path potential method.
J Chem Phys 121(1):101107
653. Konuklar FAS, Aviyente V, Monard G, Lopez RMF (2004) Theoretical approach to
the wear and tear mechanism in triosephoshate isomerase: a QM/MM study. J Phys
Chem B 108(12):39253934
654. Garcia-Viloca M, Alhambra C, Truhlar DG, Gao J (2003) Hydride transfer catalyzed
by xylose isomerase: mechanism and quantum effects. J Comput Chem 24(2):177
190
286
H.M. Senn W. Thiel
655. Li G, Cui Q (2003) What is so special about Arg 55 in the catalysis of cyclophilin A?
Insights from hybrid QM/MM simulations. J Am Chem Soc 125(49):1502815038
656. Mart S, Andrs J, Moliner V, Silla E, Tun I, Bertrn J (2001) Transition structure
selectivity in enzyme catalysis: a QM/MM study of chorismate mutase. Theor Chem
Acc 105(3):207212
657. Mart S, Andrs J, Moliner V, Silla E, Tun I, Bertrn J, Field MJ (2001) A hybrid potential reaction path and free energy study of the chorismate mutase reaction. J Am
Chem Soc 123(8):17091712
658. Lee YS, Worthington SE, Krauss M, Brooks BR (2002) Reaction mechanism of chorismate mutase studied by the combined potentials of quantum mechanics and
molecular mechanics. J Phys Chem B 106(46):1205912065
659. Crespo A, Scherlis DA, Mart MA, Ordejn P, Roitberg AE, Estrin DA (2003) A DFTbased QMMM approach designed for the treatment of large molecular systems:
application to chorismate mutase. J Phys Chem B 107(49):1372813736
660. Guo H, Cui Q, Lipscomb WN, Karplus M (2003) Understanding the role of activesite residues in chorismate mutase catalysis from molecular-dynamics simulations.
Angew Chem Int Ed 42(13):15081511
661. Mart S, Moliner V, Tun I, Williams IH (2003) QM/MM calculations of kinetic
isotope effects in the chorismate mutase active site. Org Biomol Chem 1(3):483487
662. Ranaghan KE, Ridder L, Szefczyk B, Sokalski WA, Hermann JC, Mulholland AJ
(2003) Insights into enzyme catalysis from QM/MM modelling: transition state stabilization in chorismate mutase. Mol Phys 101(17):26952714
663. Ranaghan KE, Ridder L, Szefczyk B, Sokalski WA, Hermann JC, Mulholland AJ
(2004) Transition state stabilization and substrate strain in enzyme catalysis: ab
initio QM/MM modelling of the chorismate mutase reaction. Org Biomol Chem
2(7):96880
664. Ruggiero GD, Guy SJ, Mart S, Moliner V, Williams IH (2004) Vibrational analysis of
the chorismate rearrangement: relaxed force constants, isotope effects and activation
entropies calculated for reaction in vacuum, water and the active site of chorismate
mutase. J Phys Org Chem 17(67):592601
665. Claeyssens F, Ranaghan KE, Manby FR, Harvey JN, Mulholland AJ (2005) Multiple high-level QM/MM reaction paths demonstrate transition-state stabilization in
chorismate mutase: correlation of barrier height with transition-state stabilization.
Chem Commun (40):50685070
666. Guimares CRW, Udier-Blagovic M, Tubert-Brohman I, Jorgensen WL (2005) Effects
of Arg90 neutralization on the enzyme-catalyzed rearrangement of chorismate to
prephenate. J Chem Theory Comput 1(4):617625
667. Ishida T, Fedorov DG, Kitaura K (2006) All electron quantum chemical calculation
of the entire enzyme system conrms a collective catalytic device in the chorismate
mutase reaction. J Phys Chem B 110:14571463
668. Mart S, Andrs J, Moliner V, Silla E, Tun I, Bertrn J (2004) A comparative study
of Claisen and Cope rearrangements catalyzed by chorismate mutase. An insight
into enzymatic efciency: transition state stabilization or substrate preorganization?
J Am Chem Soc 126(1):311319
669. Mart S, Andrs J, Moliner V, Silla E, Tun I, Bertrn J (2000) A QM/MM study of
the conformational equilibria in the chorismate mutase active site. The role of the
enzymatic deformation energy contribution. J Phys Chem B 104(47):1130811315
670. Guo H, Cui Q, Lipscomb WN, Karplus M (2001) Substrate conformational transitions in the active site of chorismate mutase: their role in the catalytic mechanism.
Proc Natl Acad Sci USA 98(16):90329037
287
671. Guo H, Lipscomb WN, Cui Q, Karplus M (2005) Molecular dynamics simulations
of yeast chorismate mutase: substrate conformational transitions and enzymatic
catalysis. In: Puglisi JD (ed) Structure, dynamics and function of biological macromolecules and assemblies. NATO Science Series I: Life and behavioural sciences,
vol 364. IOS Press, Amsterdam, pp 6374
672. Loferer MJ, Webb BM, Grant GH, Liedl KR (2003) Energetic and stereochemical effects of the protein environment on substrate: a theoretical study of methylmalonylCoA mutase. J Am Chem Soc 125(4):10721078
673. Zurek J, Bowman AL, Sokalski WA, Mulholland AJ (2004) MM and QM/MM modeling of threonyl-tRNA synthetase: model testing and simulations. Struct Chem
15(5):405414
674. Nonella M, Mathias G, Eichinger M, Tavan P (2003) Structures and vibrational
frequencies of the quinones in Rb. sphaeroides derived by a combined density functional/molecular mechanics approach. J Phys Chem B 107(1):316322
675. Hughes JM, Hutter MC, Reimers JR, Hush NS (2001) Modeling the bacterial photosynthetic reaction center. 4. The structural, electrochemical, and hydrogen-bonding
properties of 22 mutants of Rhodobacter sphaeroides. J Am Chem Soc 123(35):8550
8563
676. Reimers JR, Hutter MC, Hughes JM, Hush NS (2000) Nature of the special-pair
radical cation in bacterial photosynthesis. Int J Quant Chem 80(6):12241243
677. Hasegawa J, Ishida M, Nakatsuji H, Lu Z, Liu H, Yang W (2003) Energetics of
the electron transfer from bacteriopheophytin to ubiquinone in the photosynthetic
reaction center of Rhodopseudomonas viridis: theoretical study. J Phys Chem B
107(3):838847
678. Rajamani R, Gao J (2002) Combined QM/MM study of the opsin shift in bacteriorhodopsin. J Comput Chem 23(1):96105
679. Bondar N, Elstner M, Fischer S, Smith JC, Suhai S (2004) Can coordinate driving
describe proton transfer coupled to complex protein motions? Phase Transitions
77:4752
680. Bondar AN, Elstner M, Suhai S, Fischer S, Smith JC (2005) Direct proton transfer in a putative L-state intermediate of the bacteriorhodopsin photocycle. Phase
Transitions 78:59
681. Lee Y-S, Krauss M (2004) Dynamics of proton transfer in bacteriorhodopsin. J Am
Chem Soc 126(7):22252230
682. Warshel A, Chu ZT (2001) Nature of the surface crossing process in bacteriorhodopsin: computer simulations of the quantum dynamics of the primary photochemical
event. J Phys Chem B 105(40):98579871
683. Bondar AN, Smith JC, Fischer SK (2005) QM/MM investigation of the hydrogenbonding interactions in putative K and early-M intermediates of the bacteriorhodopsin photocycle. Phase Transitions 78:671675
684. Hayashi S, Tajkhorshid E, Kandori H, Schulten K (2004) Role of hydrogen-bond network in energy storage of bacteriorhodopsins light-driven proton pump revealed by
ab initio normal-mode analysis. J Am Chem Soc 126(34):1051610517
685. Rousseau R, Kleinschmidt V, Schmitt UW, Marx D (2004) Modeling protonated water
networks in bacteriorhodopsin. Phys Chem Chem Phys 6(8):18481859
686. Sugihara M, Buss V, Entel P, Hafner J, Bondar AN, Elstner M, Frauenheim T (2004)
Ab initio, tight-binding and QM/MM calculations of the rhodopsin chromophore in
its binding pocket. Phase Transitions 77(12):3145
687. Sugihara M, Buss V (2005) Retinal versus 13-demethyl-retinal inside the rhodopsin
binding pocketa QM/MM study. Phase Transitions 78:1115
288
H.M. Senn W. Thiel
688. Gascn JA, Sproviero EM, Batista VS (2006) Computational studies of the primary
phototransduction event in visual rhodopsin. Acc Chem Res 39(3):184193
689. Gascon JA, Batista VS (2004) QM/MM study of energy storage and molecular rearrangements due to the primary event in vision. Biophys J 87(5):29312941
690. Sugihara M, Hufen J, Buss V (2006) Origin and consequences of steric strain in the
rhodopsin binding pocket. Biochemistry 45(3):801810
691. Gascn JA, Sproviero EM, Batista VS (2005) QM/MM study of the NMR spectroscopy
of the retinyl chromophore in visual rhodopsin. J Chem Theory Comput 1(4):674
685
692. Rhrig UF, Guidoni L, Rothlisberger U (2005) Solvent and protein effects on the
structure and dynamics of the rhodopsin chromophore. ChemPhysChem 6(9):1836
1847
693. Hayashi S, Tajkhorshid E, Pebay-Peyroula E, Royant A, Landau EM, Navarro J,
Schulten K (2001) Structural determinants of spectral tuning in retinal proteins:
bacteriorhodopsin vs. sensory rhodopsin II. J Phys Chem B 105(41):1012410131
694. Fujimoto K, Hasegawa J, Hayashi S, Kato S, Nakatsuji H (2005) Mechanism of color
tuning in retinal protein: SAC-CI and QM/MM study. Chem Phys Lett 414(13):239
242
695. Ferr N, Olivucci M (2003) Probing the rhodopsin cavity with reduced retinal models at the CASPT2//CASSCF/AMBER level of theory. J Am Chem Soc
125(23):68686869
696. Wanko M, Hoffmann M, Strodel P, Koslowski A, Thiel W, Neese F, Frauenheim T,
Elstner M (2005) Calculating absorption shifts for retinal proteins: computational
challenges. J Phys Chem B 109(8):36063615
697. Patnaik SS, Trohalaki S, Pachter R (2004) Molecular modeling of green uorescent
protein: structural effects of chromophore deprotonation. Biopolymers 75(6):441
452
698. Sinicropi A, Andruniow T, Ferr N, Basosi R, Olivucci M (2005) Properties of the emitting state of the green uorescent protein resolved at the
CASPT2//CASSCF/CHARMM level. J Am Chem Soc 127(33):1153411535
699. Marques MAL, Lpez X, Varsano D, Castro A, Rubio A (2003) Time-dependent
density-functional approach for biological chromophores: the case of the green
uorescent protein. Phys Rev Lett 90(25):258101/14
700. Dittrich M, Freddolino PL, Schulten K (2005) When light falls in LOV: a quantum mechanical/molecular mechanical study of photoexcitation in Phot-LOV1 of
Chlamydomonas reinhardtii. J Phys Chem B 109(26):1300613013
701. Groenhof G, Bouxin-Cademartory M, Hess B, de Visser SP, Berendsen HJC,
Olivucci M, Mark AE, Robb MA (2004) Photoactivation of the photoactive yellow protein: why photon absorption triggers a trans-to-cis isomerization of the
chromophore in the protein. J Am Chem Soc 126(13):42284233
702. http://amber.scripps.edu/
703. http://www.cemcomco.com/
704. http://www.charmm.org/
705. http://www.scm.com/
706. http://www.cse.clrc.ac.uk/qcg/gamess-uk/
707. http://www.gaussian.com/
708. http://www.emsl.pnl.gov/docs/nwchem/
709. http://www.schroedinger.com/
710. http://www.pt.tu-clausthal.de/atp/
711. http://www.chemshell.org/

712.
713.
714.
715.
716.
717.
718.
719.
720.
721.
722.
723.
724.
725.
726.
727.
728.
729.
730.
731.
732.
733.
734.
735.
736.
737.
289
http://comp.chem.umn.edu/qmmm/
http://www.molpro.com/
http://www.thch.uni-bonn.de/tc/orca/
http://www.turbomole.com/
http://www.igc.ethz.ch/gromos/
http://www.ivec.org/GULP/
http://www.cse.clrc.ac.uk/msi/software/DL_POLY/
http://swift.cmbi.ru.nl/gv/whatcheck/
http://biotech.ebi.ac.uk:8400/
http://swift.cmbi.ru.nl/whatif/
http://propka.chem.uiowa.edu/
http://hekto.med.unc.edu:8080/HERMANS/software/DOWSER/Dowser.htm
http://www.chem.qmul.ac.uk/iubmb/enzyme/
Yang Z, Zhou Y, Liu K, Cheng Y, Liu R, Chen G, Jia Z (2003) Computational study on
the function of water within a -helix antifreeze protein dimer and in the process of
ice-protein binding. Biophys J 85(4):25992605
Liu H, Elstner M, Kaxiras E, Frauenheim T, Hermans J, Yang W (2001) Quantum
mechanics simulation of protein dynamics on long timescale. Proteins Struct Funct
Genet 44(4):484489
Li H, Hains AW, Everts JE, Robertson AD, Jensen JH (2002) The prediction of protein
pKa s using QM/MM: the pKa of lysine 55 in turkey ovomucoid third domain. J Phys
Chem B 106(13):34863494
Magalhes AL, Maigret B, Gomes JANF, Ramos MJ (2000) Hybrid QM/MM study of
a fundamental biological interaction surrounded by its protein environment. Internet J Chem 3:11
Cho AE, Guallar V, Berne BJ, Friesner R (2005) Importance of accurate charges
in molecular docking: quantum mechanical/molecular mechanical (QM/MM) approach. J Comput Chem 26(9):915931
Mlinsek G, Novic M, Hodoscek M, Solmajer T (2001) Prediction of enzyme binding:
human thrombin inhibition study by quantum chemical and articial intelligence
methods based on X-ray structures. J Chem Inf Comput Sci 41(5):12861294
Spiegel K, Rothlisberger U, Carloni P (2004) Cisplatin binding to DNA oligomers
from hybrid CarParrinello/molecular dynamics simulations. J Phys Chem B
108(8):26992707
Jia M, Qu W, Yang Z, Chen G (2005) Theoretical study of the factors that affect
the structure and stability of the adduct of a new platinum anticancer drug with
a duplex DNA. Int J Mod Phys B 19(1517/2):29392949
Dolenc J, Borstnik U, Hodoscek M, Koller J, Janezic D (2005) An ab initio QM/MM
study of the conformational stability of complexes formed by netropsin and DNA.
The importance of van der Waals interactions and hydrogen bonding. THEOCHEM
718(13):7785
Magistrato A, Ruggerone P, Spiegel K, Carloni P, Reedijk J (2006) Binding of novel
azole-bridged dinuclear platinum(II) anticancer drugs to DNA: insights from hybrid
QM/MM molecular dynamics simulations. J Phys Chem B 110(8):36043613
Gossens C, Tavernelli I, Rothlisberger U (2005) Rational design of organo-ruthenium
anticancer compounds. Chimia 59(3):8184
Spiegel K, Rothlisberger U, Carloni P (2006) Duocarmycins binding to DNA investigated by molecular simulation. J Phys Chem B 110(8):36473660
Guallar V, Borrelli KW (2005) A binding mechanism in proteinnucleotide interactions: implication for U1A RNA binding. Proc Natl Acad Sci USA 102(11):39543959
290
H.M. Senn W. Thiel
738. Zhang Y (2006) Pseudobond ab initio QM/MM approach and its applications to
enzyme reactions. Theor Chem Acc 116:4350
739. Elstner M (2006) The SCC-DFTB method and its application to biological systems.
Theor Chem Acc 116(1):316325
740. Claeyssens F, Harvey JN, Manby FR, Mata RA, Mulholland AJ, Ranaghan KE,
Schtz M, Thiel S, Thiel W, Werner H-J (2006) High-accuracy computation of reaction barriers in enzymes. Angew Chem Int Ed 45(41):68566859
741. Laino T, Mohamed F, Laio A, Parrinello M (2006) An efcient linear-scaling electrostatic coupling for treating periodic boundary conditions in QM/MM simulations.
J Chem Theor Comput 2:13701378
742. Yu N, Hayik SA, Wang B, Liao N, Reynolds CH, Merz KM Jr (2006) Assigning the
protonation states of the key aspartates in -secretase using QM/MM X-ray structure
renement. J Chem Theor Comput 2:10571069
743. Schll-Paschinger E, Dellago C (2006) A proof of Jarzynskis nonequilibrium work
theorem for dynamical systems that conserve the canonical distribution. J Chem
Phys 125(5):054105/15
744. Xiong H, Crespo A, Marti M, Estrin D, Roitberg AE (2006) Free energy calculations
with non-equilibrium methods: applications of the Jarzynski relationship. Theor
Chem Acc 116(1):338346
745. Olsson MHM, Mavri J, Warshel A (2006) Transition state theory can be used in studies of enzyme catalysis: lessons from simulations of tunneling and dynamical effects
in lipoxygenase and other systems. Phil Trans R Soc B 361:14171432
746. Warshel A, Sharma PK, Kato M, Xiang Y, Liu H, Olsson MHM (2006) Electrostatic
basis for enzyme catalysis. Chem Rev 106(8):32103235
747. Bruice TC (2006) Computational approaches: reaction trajectories, structures, and
atomic motions. Enzyme reactions and prociency. Chem Rev 106(8):31703187
748. Antoniou D, Basner J, Nez S, Schwartz SD (2006) Computational and theoretical
methods to explore the relation between enzyme dynamics and catalysis. Chem Rev
106(8):31703187
749. Gao J, Ma S, Major DT, Nam K, Pu J, Truhlar DG (2006) Mechanisms and free
energies of enzymatic reactions. Chem Rev 106(8):31883209
750. Pu J, Gao J, Truhlar DG (2006) Multidimensional tunneling, recrossing, and the
transmission coefcient for enzymatic reactions. Chem Rev 106(8):31403169
751. Schramm VL (ed) (2006) Chem Rev 106(8)
752. Zheng J, Wang D, Thiel W, Shaik S (2006) QM/MM study of mechanisms for compound I formation in the catalytic cycle of cytochrome P450cam. J Am Chem Soc
128(40):1320413215
753. Altun A, Shaik S, Thiel W (2006) Systematic QM/MM investigation of factors that
affect the cytochrome P450-catalyzed hydrogen abstraction of camphor. J Comput
Chem 27:13241337
754. Tuttle T, Keinan E, Thiel W (2006) Understanding the enzymatic activity of
4-oxalocrotonate tautomerase and its mutant analogues: a computational study.
J Phys Chem B 110:1968519695
755. Crespo A, Mart MA, Estrin DA, Roitberg AE (2005) Multiple-steering QMMM
calculation of the free energy prole in chorismate mutase. J Am Chem Soc
127(19):69406941
Top Curr Chem (2007) 268: 291317

DOI 10.1007/128_085
Transition Path Sampling Simulations

of Biological Systems
Christoph Dellago1 (u) Peter G. Bolhuis2
1 Faculty
of Physics, University of Vienna, Boltzmanngasse 5, 1090 Vienna, Austria

Christoph.Dellago@univie.ac.at
2 vant Hoff Institute for Molecular Sciences, University of Amsterdam,
Nieuwe Achtergracht 166, 1018 WV Amsterdam, The Netherlands
1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
292
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
Transition Path Sampling Techniques . . . . .

The Path Ensemble . . . . . . . . . . . . . . .
Dening the Stable States . . . . . . . . . . . .
Sampling the Path Ensemble with Monte Carlo
The Initial Path . . . . . . . . . . . . . . . . .
Computational Issues . . . . . . . . . . . . . .
Analysis of the Reaction Mechanism . . . . . .
Calculation of Rate Constants . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
296
296
298
299
300
301
302
304
3
3.1
3.2
3.3
3.4
3.5
3.6
Applications . . . . . . . . . . . . . . . . .
Biomolecular Isomerization . . . . . . . .
Binding and Unbinding of DNA Base Pairs
Protein Folding . . . . . . . . . . . . . . .
Enzyme Catalysis . . . . . . . . . . . . . .
Lipid Bilayers . . . . . . . . . . . . . . . .
Switches in Biochemical Networks . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
306
306
307
308
311
312
312
Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
313
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
314
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Abstract Transition path sampling is a computational method for the simulation of rare
but important events occurring in complex systems. Based on a statistical mechanics in
trajectory space, transition path sampling can be applied to identify mechanisms and determine rate constants. Here we review the basic ideas and algorithms of transition path
sampling and discuss some recent applications of this methodology to problems in molecular biology including protein folding, enzyme catalysis, and processes occurring in
lipid bilayers.
Keywords Rare events Computer simulation Transition path sampling
Abbreviations
DPS Discrete path sampling
FFS Forward ux sampling
MC Monte Carlo
292
MD
TIS
TPS
TS
TSE
TST
C. Dellago P.G. Bolhuis

Molecular dynamics
Transition interface sampling
Transition path sampling
Transition state
Transition state ensemble
Transition state theory
1
Introduction
Molecular biology strives to describe the structures and processes occurring in the living cell on the molecular level [1]. Molecular processes can
involve proteins, nucleic acids, lipid molecules, or other molecules. Proteinspecic processes include enzyme catalyzed reactions, protein folding, binding and regulation, signaling pathways, and transport by molecular motors
(trafcking, muscles). Nucleic acids are involved in RNA folding, ribozymatic reactions, DNA replication, transcription, chromatin formation, nucleosomes transitions, and translocation. Lipid molecules are of course crucial
for understanding membranes, and are required in the description of protein/nucleic acid translocation, the functioning of ion and water channels,
membrane ion permeation, and membrane protein association. To understand such processes it is of crucial importance to obtain detailed knowledge
of the involved mechanisms on a molecular level.
Computer simulations have been very successful in providing insight into
the structure and dynamics of biomolecules. In particular, the advent of molecular dynamics simulation and the development of appropriate interaction
potentials, a.k.a. force elds, for biomolecules has had an enormous impact
(see, for instance [2]; see also the paper by Schulten et al. in this volume).
However, one of the crucial problems one encounters in the simulation of
many of the above-mentioned processes is that they are rare and occur on
timescales not accessible to molecular dynamics simulations even on todays
fastest computers. For instance, times required by proteins to assemble into
their native structure typically range from milliseconds to even hours. On
the other hand, the timescale of molecular vibrations sets the basic molecular dynamics time step to about one femtosecond. This large separation of
timescales results in a huge number of molecular dynamics steps required for
the observation of just one folding event. Investigating such rare processes
by means of straightforward molecular dynamics simulation is therefore impractical on current computers.
Rare events are related to the existence of high energy barriers or entropic
bottlenecks partitioning the conguration space of the system in different
metastable basins (for instance the folded and unfolded state of a protein). In
equilibrium, the system uctuates in these long-lived states most of the time
293
and the barriers separating them are crossed only rarely. While the time intervals between such transitions may be very long, often the barrier-crossing
event is swift when it occurs. An early approach to solve this timescale problem dates back to the 1930s when Eyring [3] and Wigner [4] introduced
transition state theory (TST). TST is based on the concept of a transition state,
a saddle point in the potential energy surface that the system must cross on
its way from reactants to products. Under the assumption that trajectories
that have crossed the transition state coming from the product region never
recross the transition state, TST yields the reaction rate constant.
Algorithms for the calculations of dynamical corrections to TST were later
introduced by Keck, Bennett, and Chandler [57] providing an, in principle
exact, route to the calculation of reaction rate constants. In the Bennett
Chandler approach, rate constants are computed in a two-step procedure.
First, the free energy prole along a postulated reaction coordinate is calculated and used to determine the transition state theory approximation of the
reaction rate constant. As the transition state region corresponding to the free
energy maximum involves unlikely congurations that are rarely sampled in
equilibrium, advanced methods such as umbrella sampling [8] or thermodynamic integration [9, 10] are required in such a free energy calculation. The
TST rate constants corresponds to the instantaneous ux through a dividing
surface dened by the value of the reaction coordinate at the top of the free
energy barrier. In a second step, a dynamical correction factor taking into
account correlated recrossings of the dividing surface, the so called transmission coefcient, is determined from molecular dynamics trajectories initiated
at the dividing surface. These trajectories can also be analyzed to study the
details of the transition mechanism. In the context of biomolecular simulations, transition state theory, with and without dynamical corrections, was
mainly used to calculate reaction rate constants of enzymatic reactions [11]
but also of conformational changes [12] and trans-membrane transport in
pores [13], to name just a few examples.
The applicability of TST and its more sophisticated variants such as variational TST [14] or the above-mentioned BennettChandler approach, relies
on the knowledge of an appropriate reaction coordinate. A reaction coordinate is a function of the congurational degrees of freedom of the system
that should be capable of characterizing the progress of a transition through
the dynamical bottleneck region. If the reaction coordinates are chosen in
a way that does not capture the essential mechanisms involved in the transition, the methods discussed in the previous section fail. However, identifying
the right coordinates can be exceedingly difcult in a complex phase space.
The generic reaction coordinates for protein folding, for instance, are still
unknown.
In some cases, relevant transition states may be located by searching for
saddle points in the potential energy surface [1517]. The direction of the unstable mode then serves as a reaction coordinate permitting a transition state
294
theory calculation of the reaction rate constant. From databases of stationary

points in the potential energy surface it is then possible to obtain the kinetics
of processes involving transitions between many stable states [18, 19]. While
this approach can yield valuable insights into the properties of biomolecular systems [20, 21], it is limited in its applicability due to the exponential
growth of the number of stationary points with system size. Stochastic transitions between metastable states consisting of almost invariant subsets of
conformations, rather than single potential energy minima, are considered in
Markovian state models [22, 23].
A similar viewpoint, in which the dynamics is regarded as consisting of jumps between long-lived states, is adopted in the parallel replica
method [24, 25], which exploits the availability of highly parallel computer
systems. In this method, independent simulations initiated with different initial conditions are performed in parallel on many processors. Each replica is
periodically tested for transitions from a stable state to a new one. As soon
as a transition is detected on one processor, all simulations are stopped and
then restarted in the new state. Provided that transitions are uncorrelated
with each other, the correct distribution of escape times can be reconstructed
and in the optimum case a speed-up proportional to the number of processors can be achieved. However, the number of needed replicas required to
observe a transition event in a given time increases exponentially with the
barrier height. In the realm of biomolecular simulation, this approach has
been mainly used to determine the folding pathways of small proteins on
thousands of computers running the simulation in parallel [26].
Other rare event approaches that have been applied in biomolecular systems, such as the hyperdynamics method [27, 28] or the ooding method [29,
30] (see also the local elevation method [31]), increase the likelihood of
visiting rare transition state conformations by modication of the potential energy surfaces. The modied energy landscape destabilizes the longlived (meta)stable states and thus accelerates transitions between them. Although the articial potential energy surface alters the dynamics of the
system, certain aspects of the time evolution are conserved on a coarse
grained level.
Modications of the potential energy surface favoring the sampling of rare
transition state congurations can also be implemented by introducing timedependent perturbations that drive the system away from those regions of
conguration space that have been already visited. This idea is the basis of
the metadynamics method that can be used to map out the free energy of
the system as a function of a small set of preselected collective variables [32].
Although, due to the time-dependent bias, the dynamics differs from the
true dynamics of the system, metadynamics is a powerful tool for the identication of possible transition mechanisms. The method, hinging on an
appropriate denition of the collective variables, has been applied to study
various processes ranging from chemical reactions in solution [33] to trans-
295
membrane transport [34]. Such a reduced description is also the basis of

the coarse grained molecular dynamics method, in which short trajectories
in the full phase space are used to estimate effective forces acting in a lowdimensional space spanned by some preselected coarse variables [35]. As
illustrated by simulating conformational changes of alanine dipeptide in water, this method can considerably accelerate the simulation, albeit at the cost
of a detailed description of the dynamics.
In many processes involving rare transitions between long-lived stable
states, not only the initial but also the nal state is known. For instance,
in the case of membrane translocation, enzymatic reactions, or isomerizations detailed information is available both for reactants and products.
This additional knowledge is exploited in several approaches based on pathways that start and end in particular regions of conguration space. These
methods differ in the particular type of dynamics considered. In early work,
Pratt [36] suggested using a Monte Carlo approach to sample the probability distribution of stochastic pathways connecting reactants with products.
Other methods are based on a minimization of the classical mechanical
action [37, 38]. The advantage of these latter methods is that they can be
used with very large time-steps, thus speeding up the simulation considerably. However, the dynamical meaning of such long-timestep trajectories
remains unclear and results obtained with these methods should be veried with other approaches. Paths connecting reactant with products regions
were also considered by Elber and Karplus, who devised a method to determine transition pathways by minimizing an articial path energy, or
action [39]. Other methods follow a similar approach but differ in the particular form of action that is minimized [4042]. Within this class of algorithms the nudged elastic band method [43] permits a particularly efcient
localization of the transition state, which can subsequently be used to estimate reaction rate constants from transition state theory. In this method,
forces perpendicular and tangential to the pathway nudge the trajectory
towards a minimum energy pathway passing through the transition state.
In a biological context the nudged elastic band method has, for instance,
been used to determine pathways for conformational changes in RNA [44].
The same forces used in the nudged elastic band method are also applied
in the so-called string method and its extension, the nite temperature
string method [45, 46]. Applicable to stochastic dynamics, the latter algorithm is based on a constrained sampling of the equilibrium distribution
in a sequence of hyperplanes along a string that connects reactants with
products. These hyperplanes are adapted iteratively such that they are tangential to the isoprobability surfaces consisting of all points that have the
same probability to relax into one stable state rather than the other. As
a result of a string method calculation one obtains a transition tube, in
which the transition occurs with high probability, and also the corresponding free energy prole and transition rate constant. The applicability of the
296
nite temperature string method to biomolecular systems has been demonstrated in an analysis of the isomerization of alanine dipeptide in vacuum and
explicit solvent [47].
A more widely applicable approach, the transition path sampling method
(TPS) [48, 49], consists in rst dening the ensemble of all pathways connecting reactants with products and then sampling this ensemble of dynamical
pathways with Monte Carlo techniques. The TPS method is valid for a wide
range of dynamics ranging from Monte Carlo dynamics to Langevin dynamics and Newtonian dynamics. The method permits the identication of
reaction mechanisms and the calculation of reaction rate constants in complex systems with rough free energy barriers. We will focus on TPS and its
application to biomolecular problems in the remaining part of this article,
which is structured as follows. In Sect. 2 we give a brief overview the basic techniques involved in path sampling, followed by a description of how
to calculate the rate constants. In Sect. 3 we review several papers applying TPS to biologically relevant systems. We end with an outlook on future
possibilities.
2
Transition Path Sampling Techniques
2.1
The Path Ensemble
We are interested in rare event processes that occur in complex systems between two (meta)stable long-lived states, denoted A and B, separated by an
unknown and, in general, rough and high free energy barrier. The system
is (meta)stable in states A and B in the sense that if the system is put there
it will remain there for a long time, and only rarely undergoes transitions
between the stable states A and B. We assume that no long-lived metastable
states exist between states A and B. Further, we assume that states A and B can
be characterized using a small set of order parameters.
The goal of a TPS simulation is to collect all likely transition pathways between these states A and B. The basic idea behind TPS is a generalization of
the standard Monte Carlo importance sampling procedures [5052] to trajectory space. When we consider, for instance, the ensemble of all trajectories
with a given time length that is small compared to the reaction time, most of
these will be localized in or near either stable state A or B. Rare transitions
between A and B will comprise only a small subset of these short trajectories.
For example, if the process of interest occurs roughly once every millisecond,
then only one out of a million 1 ns trajectories will exemplify that process.
TPS provides an efcient means to sample such sub-ensembles of rare trajectories. Subsequent analysis of the path ensemble enables the elucidation of the
297
transition mechanism, i.e., the degrees of freedom that capture the physics of
the transition and how these change during the transition. Since the pathways
collected with TPS are fully dynamical trajectories rather than articial pathways, such as minimum energy pathways or elastic bands, it is also possible
to extract kinetic information from TPS simulations.
In this section we describe very briey the basic concepts of TPS in the
context of biomolecular simulations. For a more detailed description of this
methodology we refer the reader to recent review articles [49, 5356]. To dene the path ensemble we discretize the continuous time evolution of the
system as an ordered sequence of L time slices:
X = {x0 , x1 , x2 , ., xi, ..., xL } .
(1)
Consecutive time slices are separated by a small time increment t and so

xi denotes the complete system at time it. For a Hamiltonian system, for
instance, x = {r, p} includes the positions r and momenta p of all particles.
The probability (density) of observing a given path X depends on the probability of its initial condition x0 and the specic dynamics of the system. For
Markovian processes, the total path probability is:
P [X] = (x0 )
L1

p xi xi+1 ,
(2)
i=0
where (x0 ) is the probability distribution of the initial conditions. For instance, if the system obeys a canonical Boltzmann distribution, (x0 ) = exp
( H(x0 ))/Z,where H is the Hamiltonian, = 1/kB T the inverse temperature, and Z = dx exp[ H(x)] the canonical partition function.
The particular form of the transition probability p(xi xi+1 ) depends
on the exact type of deterministic and stochastic dynamics. For deterministic dynamics (for stochastic transition probabilities see [49, 57]), the state
of the system x at time t = it is completely determined by the state x0 at
time t = 0 : xi = t (x0 ). The time-dependent function t (x0 ) is the propagator of the system. For such deterministic dynamics, the short time transition
probability can be written in terms of a Dirac delta function: p(xi xi+1 ) =
[xi+1 t (xi )]. Accordingly, the path probability for a Newtonian trajectory
is:
L1

P [X] = x0
xi+1 t xi .
(3)
i=0
Other examples of deterministic dynamics include the extended Lagrangian

dynamics of Car and Parrinello [59] and various dynamics with thermostats
such as the NoseHoover thermostat [6062] or the Gaussian isokinetic thermostat [63, 64]. The above denition of path ensembles also applies for these
types of dynamics.
298
To constrain the path ensemble to those rare paths that connect stable
state A with state B, we multiply the path probability P[X] with characteristic
functions:

PAB [X] = hA x0 P [X] hB xL /ZAB ,
(4)
where hA (x) and hA (x) are the characteristic functions of region A and B,
respectively, hA,B (x) = 1 if x A, B and zero otherwise, and ZAB is a normalization factor corresponding to a partition function in trajectory space. Thus,
a path that does not start in A and end in B has a zero weight whereas a path
connecting A and B, may have a non-zero weight depending on the unrestricted path probability P [X]. The transition path ensemble PAB [X] selects
only the reactive trajectories from the ensemble of all possible pathways while
leaving the relative probabilities of the reactive trajectories among each other
unchanged.
2.2
Defining the Stable States
The main advantage of TPS is that it does not require precise knowledge
of the reaction coordinate. Instead, a careful characterization of the initial
region A and the nal region B is important. Often, such characterization
involves nding one or more convenient low dimensional order parameters
q, which have to meet several criteria. First and most importantly, the order
parameters should discriminate A from B, i.e., region A should not overlap
with the basin of attraction of B and vice versa1 . In other words, region A
spanned by hA (x) and region B spanned by hB (x) should be located entirely
within the corresponding basins of attraction. If this criterion is not met, the
TPS algorithm will eventually collect mostly non-reactive trajectories. Correct denitions of regions A and B can be based on straightforward MD
trajectories in both states, but in our experience, identifying discriminating
order parameters can involve considerable trial and error experimentation.
Fortunately, it is easy to detect problems associated with an inappropriate
choice of order parameter. When problems are caused by non-discriminating
regions A and B, the order parameter has to be rened until correct sampling
can be achieved.
A second criterion for the order parameters is that the regions should be
representative: A and B have to be large enough to accommodate typical equilibrium uctuations. This is sometimes difcult to achieve, and therefore this
criterion can be relaxed and the stable states only have to be visited frequently
enough within the path length. If this criterion is not met, important transition pathways might be missing from the transition path ensemble.
1
The basin of attraction of a specic stable state consists of all congurations from which trajectories relax into that stable state.
299
2.3
Sampling the Path Ensemble with Monte Carlo
As mentioned above, the goal of a TPS simulation is to collect reactive trajectories according to their weight in the transition path ensemble. This can be
achieved with a Monte Carlo importance sampling procedure similar to a molecular simulation [51]. A molecular MC algorithm samples congurational
space by generating a new trial conguration from an old one by randomly
selecting a particle and moving it by a random distance. This trial conguration is accepted or rejected according to an acceptance probability satisfying
the detailed balance condition, usually, the Metropolis rule. This basic step is
then iterated. Provided it is ergodic, this algorithm guarantees the sampling
of the desired distribution of congurations. Since this procedure generates
a random walk that visits congurations according to their statistical weight,
it is also called importance sampling. Monte Carlo procedures can sample
arbitrary probability distributions including the path ensemble (Eq. 4), in
which case the random walk is in the space of trajectories. The basic step of
this MC procedure consists of generating a new path, X(n) , from an old one,
X(o) . The Metropolis rule for accepting such a new path is given by:

(n)
(n)

(o)

X
P
X
X
P
AB
gen

,
Pacc X(o) X(n) = min 1,
(5)
PAB X(o) Pgen X(o) X(n)
where Pgen [X(o) X(n) ] is the probability of creating a new path X(n) from
an existing path X(o) . A variety of schemes to generate new trial paths
exist [49], but so far the shooting algorithm has proved to be the most efcient one. (See [49] for other path-generating procedures.) The shooting
algorithm generates new pathways by (a) randomly selecting a time slice x(o)
i
(n)
along the old path, (b) perturbing it to yield xi (this step is necessary for deterministic dynamics), and (c) generating from this time slice two new path
segments forward and backward in time according to the propagation rules
of the underlying dynamics. Together, these two new segments form the new
pathway. Using this path generation scheme and assuming microscopic reversibility holds for the underlying dynamics, the acceptance rule simplies
enormously [49]:

x(n)
i
(n)
(n)
(o)
(n)
Pacc X X
= hA x0 hB xL min 1, (o) .
(6)
xi
Hence, any new trial path connecting the stable states A and B is accepted
with a probability depending only on the shooting points of the old and the
new pathway. This simple acceptance rule also suggests the following algo(n)
rithm. First, select a shooting point x(o)
i at random and modify it to xi . Then,
accept this shooting point when a random number selected from a uniform
300

(n)
(o)
distribution in the interval [0, 1] is smaller than [xi ]/[xi ]. If the shooting point is accepted, grow either one of the path segments. If this segment
reaches the appropriate stable region, grow the other segment. The whole
path is nally accepted if the boundary condition for that segment is also
satised. Rejection of any of these steps results in rejecting the entire trial
path. This procedure is then repeated to harvest an ensemble of transition
paths. The acceptance rule (Eq. 6) is very general and is valid for any procedure that is microscopically reversible. Cases in which it is applicable include
Langevin dynamics, Monte Carlo dynamics, and Newtonian dynamics. We refer the reader to [56] for variants of the shooting algorithm specic to the
various types of dynamics.
For biomolecular processes the pathways can become very long. From
a computational viewpoint it might therefore be useful to minimize the path
length to its absolute minimum: the transition time. This is the part of the
trajectory from exiting A until entering B, thus leaving out the uninteresting
parts of the trajectories that are inside the stables states. This is allowed, as
long as the probability of leaving stable state B again, after coming directly
from A, is as likely as an independent transition. In other words, once B (or
A) has been reached, no recrossings are allowed. It can be shown that the
shooting procedure allows for such a variable path length [65]. The shooting
algorithm remains the same, but the integration of the equations of motion from the shooting point continues only until one of the stable states
is reached. As the path length varies, the probability 1/L to pick a random
shooting point is not symmetric anymore with respect to the reverse move.
The acceptance rule thus now becomes,

L(o) x(n)
i
(n)
(n)
(o)
(n)
Pacc X X
(7)
= hA x0 hB xL min 1,
(o) ,
(n)
L xi
where L(o) and L(n) denote the length of the old and the new path, respectively. This procedure has the advantage that the path length is restricted to
the minimum number of time steps needed to cross the barrier and is hence
automatically always optimal.
2.4
The Initial Path
Initiating a TPS simulation with the algorithms described above requires
an existing pathway connecting A with B. Hence, the generation of an initial transition pathway is an essential step in the application of TPS. The
simplest way to obtain an initial trajectory connecting A and B is by running a long molecular dynamics (or stochastic dynamics) simulation. For
most applications, however, the rarity of the process of interest renders this
straightforward approach impractical. More efcient ways to create an ini-
301
tial trajectory exist, but, in general, produce an atypical trajectory with a very
low weight in the transition path ensemble PAB [X]. Therefore, the rst part
of a TPS simulation starting from such articial trajectory consists of equilibrating the pathway towards the more important regions of trajectory space.
This situation is analogous to that encountered in a conventional MC simulation of a molecular uid. The initial conguration is often a regular lattice,
which subsequently relaxes to the more probable uid regions of conguration space. Similarly, a TPS simulation can start from an articial pathway
which does not even need to be a true dynamical trajectory. Then, repeated
application of the MC steps described above move the pathways towards more
typical regions of pathways and the actual TPS simulation can begin.
Now, how does one create this articial initial pathway? In some situations, high-temperature pathways can be used to initiate a TPS simulation.
Consider, for example, the folding/unfolding of a protein. At physiological temperatures, a protein in its native states unfolds only very rarely on
a molecular timescale. At higher temperatures, however, unfolding occurs
sufciently quickly so that it can be simulated with ordinary molecular dynamics simulations. Such a high temperature trajectory acts then as an initial
pathway to start a TPS simulation at the temperature of interest. If high temperature transition pathways are qualitatively very different from those at
lower temperatures it might be necessary to carry out an annealing procedure, in which the ensemble of pathways if brought to a lower temperature in
small steps. In other cases, one may have some, possibly incomplete, notion
of a reaction coordinate. Controlling this presumed reaction coordinate one
might be able to drive (bias) the system from A to B obtaining a chain of states
from which shooting and shifting moves can be initiated.
A more systematic way to create a new transition pathway is to change the
ensemble PAB [X] gradually from one which includes all trajectories starting
in A (without restrictions on the end point) to one which consists only of trajectories starting in A and ending in B. Conversion of one path ensemble into
another is computationally demanding and in most cases more efcient ways
to generate an initial trajectory exist.
In our experience, no general recipe is available for the generation of an
initial trajectory. Rather, specic procedures have to be developed for this
purpose for each application of the TPS method to a new problem.
2.5
Computational Issues
The computational effort in carrying out a TPS calculation scales linearly
with the number of trajectories harvested. In particular, to obtain N statistically independent trajectories of length t requires the same order of effort
as that required to perform N trajectories. In practice, shooting moves are
accepted with probabilities between 0.1 and 0.7, with an optimal acceptance
302
ratio of about 40% [49]. Further, the correlations in that random walk persist
typically for only two or three accepted moves. For the stochastic algorithms
this number can be much larger. Thus, for instance, 1000 statistically independent 1 ps trajectories are obtained with roughly the same computational
resources required for a single straightforward trajectory of length 10100 ns.
If the rare event occurs with a rate constant of 1 s1 , a straightforward trajectory will probably not show a single transition, while the 1000 transition path
trajectories each will exhibit an independent example of the event.
As explained in the previous section the efciency can be improved by
roughly a factor of two using a exible path length, in order to reduce the
computational effort to a minimum.
Rare events in biomolecular systems often require pathways considerably
longer than a few picoseconds. For instance, a folding protein can take several
nanoseconds to diffuse over the folding free energy barrier, even when there
is no intermediate long-lived (meta)stable state. The deterministic shooting
algorithm runs into problems for such long diffusive folding trajectories.
While a randomly chosen shooting point might seem to lie in the barrier
region (i.e., outside of the stable state denitions) it can in fact already be
completely committed to one of the stable states. In that case, the acceptance ratio will be extremely low. Only when shooting from points around
the true transition state region, can we expect a reasonable acceptance. Unfortunately there is no way to recognize these points a priori. To alleviate
this problem one can employ the stochastic shooting algorithm [66], allowing shooting in one direction, either forward or backwards. To still be able
to use deterministic MD in combination with the stochastic algorithm requires the introduction of a small amount of stochasticity in the trajectories,
for instance by application of the Andersen thermostat. The Andersen coupling constant can be made sufciently small so that there is no noticeable
difference from completely deterministic dynamics [51]. In this case, the acceptance ratio will improve dramatically, in fact to about 50%. The downside
is that the correlation between paths increases. Several tools have been developed to keep track of this correlation [73].
2.6
Analysis of the Reaction Mechanism
As mentioned before, the main advantage of TPS is that it does not require
a priori knowledge of the reaction coordinate. Instead one can analyze the
path ensemble to obtain transition state ensembles and test proposed reaction
coordinates by computing committor distributions.
In a typical complex biomolecular case the order parameters used to dene the stable states are bad reaction coordinates. Driving the system from
A to B along such wrong coordinates neglects important barriers that exist
in orthogonal directions and leads to large hysteresis effects. A good reaction
303
coordinate should capture the essence of the dynamics and should allow the
prediction of where a trajectory initiated from a given conguration will most
likely head. The commitment probability pB or committor enables precisely
that, as it is dened as the probability to relax into stable state B. The committor is a statistical measure for how committed a given conguration is to the
product state. A value of pB = 0 indicates no commitment at all while a value
of pB = 1 indicates that the conguration is fully committed to B. Fractional
values like 0.5 denote that the conguration is equally committed to A and B.
The committor (a.k.a p-fold, or splitting probability [6770]) is a direct statistical indicator of the progress of the reaction. In this sense it is an ideal
reaction coordinate. Thus, if a good reaction coordinate is available, the committor can be parametrized in terms of this coordinate [46]. Parameterizing
the committor in terms of a small number of atomic variables to gain insight
into the reaction mechanism, however, is a highly non-trivial task.
A practical way to compute the committor for an arbitrary conguration
r is to start N independent trajectories and compute the fraction of paths pB
that reach stable state B after time t. For deterministic dynamics an average
fraction is obtained by drawing the initial momenta for r from a Maxwell
Boltzmann distribution. For stochastic dynamics, the average extends also
over noise histories. Assuming that the N trajectories are statistically independent,
the standard deviation in pB calculated from N trajectories is

= pB (1 pB )/N. This expression can be used to terminate a committor
calculation after a certain desired accuracy has been reached. As pB (1 pB )
has its maximum at pB = 1/2 the largest number of trajectories is needed for
congurations with a committor pB = 0.5. Note that the committor depends
on the time-length of the trajectories used. This time should be larger than
the molecular timescale needed to commit to one of the stable states. A timeindependent form of the committor can be dened by counting all trajectories
originating from a given conguration r that reach state B before they reach
state A.
A conguration r with a committor value pB = pA = 0.5 marks a transition
state. Both stable states A and B are equally accessible from that conguration. Transition states dened in this statistical way often do not coincide
with particular features of the potential energy surface, because in general
entropic contributions play an important role in determining the location of
statistically dened transition states in conguration space. The high dimensional surface with committor value 0.5 divides or separates the basins of
attraction of the stable states. Hence, this dividing surface is also called the
separatrix [68].
In principle, it is possible to nd this separatrix by calculating the committor values for all possible congurations, but this is rather impractical.
Moreover, a pB = 0.5 conguration does not hold information on what the
probability of the TS itself is [70, 71]. The path ensemble, however, provides
a weighted ensemble of transition paths, all by necessity crossing the sep-
304
aratrix. Locating congurations with pA = pB on the ensemble of transition

pathways results in the transition state ensemble(TSE). All congurations on
transition pathways that have a committor of pB = 0.5 are members of the
TSE. Each transition path contributes at least one conguration, but possibly
more, to the TSE. Subsequent analysis of the TSE can yield important information about the reaction mechanism.
One can also use the committors to test a proposed parameter q as a reaction coordinate. When q is a good reaction coordinate it should be able
to capture the separatrix, and provide a description of the TSE. Calculating
distributions of committors for constraint ensembles can be used to test the
correctness of the proposed reaction coordinate. This diagnostic tool is not
restricted to TPS but can be applied to any case requiring a test of whether or
not the proposed order parameter describes the transition, or is only partially
correlated (if at all) with the reaction coordinate.
The rst step in this procedure is the calculation of the free energy prole as a function of a variable q. Most likely this prole will show a barrier
with a maximum at q = q . Next, congurations with q = q are generated, for
instance by imposing constraints in a molecular dynamics simulation. Subsequently, commitment probabilities are computed for a representative number
of congurations in this ensemble. If each of these congurations is truly in
the transition state region, all committors will be around pB = 0.5. In this case,
q is a good reaction coordinate. If, in contrast, they are far from one half, say 0
or 1, the congurations are clearly not part of the separatrix, but already committed to A or B. In that case, q is not a good reaction coordinate. Analysis
of the distributions of committors in this way allows location of the correct
reaction coordinates, at least in the neighborhood of the separatrix. The committor distribution can thus be used to estimate how far a postulated reaction
coordinate is removed from the correct reaction coordinate.
2.7
Calculation of Rate Constants
The pathways collected with TPS are true dynamical trajectories and hence
can capture the kinetics of the reaction. If the barrier between A and B is
high enough to cause a separation of timescales, the phenomenological rate
constant kAB for the unimolecular (rst order) reaction A B can be related
to a microscopic time correlation function C(t) of the characteristic functions hA and hB that is linear for times mol < t rxn (mol is the molecular
timescale and rxn the reaction time):

hA x0 hB xt/t
(8)
C(t) =
kAB t .
hA
Here, the brackets denote an ensemble average over the initial condition x0 . In
words, C(t) is the conditional probability that if the system is in A at time 0 it
305
will be in B at time t. To evaluate the time correlation function C(t) in the TPS
framework we can relate it to an order parameter (x) that distinguishes between the reactant region A and product region B and for which the range of
values takes when the system is in A (Amin < < Amax ) does not overlap with
the range of values takes in B (Bmin < < Bmax ). The normalized probability density for nding a particular value of the parameter at the endpoints
of pathways starting in A is given by:

PA , L = (xL ) A ,
(9)
where the average is taken over all pathways starting from A. The function
C(t) and hence the rate constant follows from the integrated probability to
reach B:
Bmax
C(t) =

dPA , L .
(10)
Bmin
This probability is extremely low for high barriers, and the calculation has to
be done using biasing techniques such as umbrella sampling (US), in which
C(t) is obtained by matching the results of TPS simulations carried out for
different -windows (see [49] for a detailed description of this procedure).
Transition Interface Sampling. The efciency of the rate constant procedure
can be improved using a slightly different path sampling approach named
transition interface sampling (TIS) [65]. The TIS method has in common with
the umbrella sampling method that it uses an order parameter to bridge the
barrier region between A and B. However, instead of windows, TIS maps the
conguration space onto a foliation of interfaces characterized by (x). The
TIS method comprises the computation of the ux to leave a stable initial
state and, subsequently, the conditional probability for trajectories leaving the
initial state to reach the nal state The forward rate constant kAB then follows
from:

(11)
kAB = fA P B |A ,
where the rst factor is the ux fA to leave region A (dened by (x) < A )
and the second factor is the conditional probability P(B |A ) to reach region
B (dened by order parameter (x) > B ) once the surface A dening region
A is passed. The ux factor can be measured by starting an MD simulation in
stable state A. The interface A will be crossed often, resulting in a statistically
accurate value for fA . In contrast, the value of P(B |A ) is very low for a high
barrier and cannot be measured directly. The statistics can again be considerably improved by employing a biased path sampling scheme. By sampling
paths with the constraint that they come from region A, cross an interface
i , and then either go to B or return to A, we can measure the probability
306
P(|A ) to reach values of larger than i . These probabilities are binned in

a histogram, and after having performed several simulations for different interfaces i , these histograms are joined into a master curve, from which the
value of P(B |A ) can be extracted. A more detailed description of TIS can
be found in [65]. TIS allows the variable path length description, thereby limiting the required time steps to a minimum. The TIS rate equation is based
on an effective positive ux formalism and is less sensitive to recrossings.
The shifting moves used in TPS to enhance statistics are unnecessary in the
TIS algorithm. Also, multidimensional or discrete order parameters can easily be implemented in TIS. Up to now TIS has only been applied to relatively
small biomolecular systems [72, 73]. The forward ux sampling method (FFS)
put forward by ten Wolde and collaborators [74] also makes use of the same
formalism but differs in the particular way of generating trajectories for the
calculation of the probabilities. For a comparison of TIS and FFS see [80].
Partial Path TIS. Some recently developed methods make use of the diffusive
nature of paths and the corresponding loss of correlation. For instance, the
partial path TIS technique (PPTIS) takes advantage of this loss of correlation
by describing the rare event as a Markovian hopping process between the interfaces, with corresponding hopping probabilities. If the loss of correlation
between three consecutive interfaces is justied, this method exactly gives the
kinetics [81]. The milestoning approach of Elber and collaborators is based
on similar ideas [82].
Discrete Path Sampling. An entirely different approach for the calculation
of rate constants is followed in the discrete path sampling scheme (DPS) of
Wales [18]. Here, pathways between long-lived stable states are viewed as
consisting of many transitions between local energy minima. Overall rate
constants are then computed from the rate constants for these transitions
between minima.
3
Applications
3.1
Biomolecular Isomerization
Conformational changes of biomolecules play a crucial role in biological
function, for instance in protein folding, but also in signal transduction,
and the mechanics of molecular motors. The complexity of such processes
and the timescales involved render a straightforward approach impractical.
Moreover, the dynamical variables for biomolecular isomerization processes
are not known a priori and are most likely multidimensional. In principle,
307
these variables can be revealed using TPS methodology. In one of the earliest attempts to apply TPS on a biomolecular system [83] we tried to nd
the reaction coordinates involved in isomerization of the simplest of peptides:
alanine dipeptide. The common peptide dihedral angles and are satisfactory order parameters for describing the stables states. However, this does
not imply that they also provide an accurate description for the dynamics of
the transition. We established, by analysis of the transition path ensembles
for isomerization of alanine dipeptide in both water and vacuum, that other
variables in addition to and are important. In these simulations we employed the AMBER 94 force eld, with TIP3P water as a solvent. For alanine
dipeptide isomerization in vacuum (from C7eq to Cax ) we found by thorough
committor distribution analysis that there is at least a third angle involved
in the dynamical transition. We identied this angle to be the O C N C
torsion angle. When this angle was constrained, the committor distribution
turned out to be peaked at around pB = 0.5. More recent work by Maragliano
et al. [84] using the nite temperature string method suggests that including
the C C N H torsion angle, besides the above three, is required to make
the committor peaked.
For alanine dipeptide isomerization in TIP3P water (from Ceq to R ) it
was found that the solvent degrees of freedom play a role and should be
incorporated in a dynamical treatment of the process [83]. Ma and Dinner
performed an exhausting search of many possible reaction coordinates and
concluded that a complex coordinate involving the electrostatic torque on
a particular hydrogen atom arising from the solvent is the best candidate
for a description of the reaction coordinate for alanine dipeptide isomerization [85]. In particular, solvent uctuations in the second solvation shell
contributed to this torque. It is therefore not surprising that previous attempts to nd these contributions by investigating solvent ux correlations
did not succeed [86].
A general conclusion following from the above work is that the inuence
of the solvent on the reaction coordinate can be subtle. Furthermore, commitment times might be long on a molecular scale. Such issues will render
the study of isomerization difcult with deterministic TPS, in particular for
larger molecules. A solution to this problem could be to abandon a deterministic approach and describe the transition as a stochastic process, as is done in
PPTIS [81], milestoning [82], and the nite temperature string method [84].
3.2
Binding and Unbinding of DNA Base Pairs
Hagan et al. [87] investigated the room temperature binding and unbinding
pathways of a base pair in a CGC DNA oligomer. The aim was to understand
the microscopic details of DNA hybridization and melting to validate existing coarse-grained models of DNA. Employing TPS with the CHARMM force
308
eld and TIP3P water they found several possible pathways for the unbinding process. The rst path involved hydrogen bond breaking via twisting of
the base pairs followed by unstacking of the C-base. In the second path the Cbase simultaneously breaks the stacking interaction and hydrogen bonds with
the G-base. The authors found, by combining free energy calculations with
path sampling, that four parameters/coordinates are important for the unbinding transition. Two monitor the interaction energy between the ipping
base and its nearest neighbors in the native structure. The other two describe
the distance between the centers of mass of the base and its neighbors. The
rst pair or coordinates describes the rst part of the transition well, where
the twisting changes the interaction energies signicantly. The second pair of
parameters describes the nal region well, where the bases are far apart.
The transition states were located in a at region in the free energy landscape as a function of the distances. The TSE is not determined by the
breaking of DNADNA hydrogen bonds or subsequent formation of DNA
water hydrogen bonds. It is also not the penetration or disappearance of water
between stacking bases, as would be the case for hydrophobic interaction.
Rather the unbinding transition requires the base to diffuse away so far that
re-formation of hydrogen bonds is unlikely. The unbound state is hence stabilized by entropic forces. However, this entropic stabilization is not as strong
as in the case of protein unfolding.
3.3
Protein Folding
Folding lies at the heart of the stability and functioning of each protein in the
living cell. The theoretical understanding of protein folding, although much
improved over the past decades, is still far from complete. What exactly determines how a protein folds into a single native state? What is the ensemble
of reactive pathways? Can we predict the nal structure? These are questions
of crucial impact for molecular biology. In this section, we summarize briey
the current understanding of two-state folding, which most single domain
proteins obey [8895], and describe some recent TPS work applied to this
problem [96].
Two-state proteins have only two stable states, native and unfolded, and
therefore show exponential relaxation behavior. The folding process is conceptually explained in terms of energy funnel landscapes. The unfolded state
has abundant conformational entropy but a high potential energy as well. The
hydrophobic effect drives the collapse of the protein into a molten globule,
a compact solvated state, where the protein still can explore many conformations. Before the transition to the native state can occur via formation of
all native contacts in combination with water expulsion, the protein rst has
to nd a transition state. This is a state topologically similar to the native
state in which a nucleus of tertiary contacts is formed. Some proteins form
309
the secondary structure elements rst, before they form the tertiary structure
(the so-called diffusion collision mechanism). Others can only stabilize their
secondary structures after the nucleation event, (the so-called nucleationcondensation mechanism).
Simulation has contributed much to the understanding of folding but it
is notoriously hard to simulate folding events using accurate atomistic force
elds, rst of all because of the long time scales involved. A protein folds
in microseconds or more, practically prohibiting the use of straightforward
implementation of simulating techniques. Second, the reaction coordinate is
very complex. This prevents the use of biased sampling (see Sect. 1).
Many simulation methods have been developed since the 1980s to circumvent this problem. One way to speed up the simulations is by coarse
graining, e.g., using implicit solvents [75, 76], Go-like models [7779], and
lattice models [92]. An alternative approach is path sampling, for which the
initial and nal states have to be known. Several groups have used two-ended
path sampling methods to study folding events [97, 98]. These studies revealed details about the specic folding transitions but, because of the large
time step between the time slices, do not represent the kinetics accurately.
Here, we will focus only on applications employing transition path sampling
or TPS related methods.
TPS of hydrophobic collapse. The rst step in protein folding is the hydrophobic collapse. Ten Wolde and Chandler [99] studied the collapse of
a hydrophobic homopolymer in solvent. The polymer was described with
a bead-string model, while the solvent was modeled on an Ising-like lattice,
capable of showing a liquidgas phase transition. Using path sampling they
observed a signicant cooperative drying effect around the polymer due to
hydrophobic interactions. One of the main conclusions following from this
work was that the collapse pathways did not follow the free energy landscape saddle point because the solvent coordinate changed on a much faster
timescale than the polymer was able to diffuse.
TPS of GB1 beta hairpin. The 16 residue C-terminal hairpin of protein G-B1 is
a model system to investigate sheet formation. Straightforward MD shows
that at 300 K there is a large barrier to unfolding, preventing the system to
unfold or fold spontaneously in an accessible simulation timescale [100103].
Previous work established that the folding occurs via an intermediate compact state in which the hydrophobic core is formed but no backbone hydrogen
bonds yet. The formation of these hydrogen bonds towards the folded state
seems to be the rate-determining step in the folding process. Application of
TPS techniques to this transition, using the CHARMM22 force eld with an
explicit TIP3P solvent at room temperature with an aggregate simulation time
of several s, elucidated the precise kinetic pathways and obtained the folding
rates [72, 73].
310
The general picture of this process is that folding starts from a state in
which the hydrophobic core holds the two strands together but no hydrogen
bonds are present. The hairpin then has to nd the transition state conformation in which specic backbone hydrogen bonds are bridged by water.
They form a lubricating strip of water between the strands that help to form
the nal hydrogen bonds. The conformational entropy of the backbone is
reduced in this transition state, as is the entropy of the water. When this transition state has been reached the folding can complete by expulsion of this
water.
TIS enabled the calculation of the rate constant, which turned out to be in
agreement with experiments (0.17 s1 ). The TIS results indicate that there
is a folding free energy barrier of more than 10 kB T. However, the computed
free energy barrier is only approximately 34 kB T. The large discrepancy between the rate constant and FE results is caused by an overlap between the
folded and unfolded states, thus apparently lowering the barriers in the projection on the used representations. It follows that free energy landscapes
cannot always be trusted to give correct barrier heights. The low barrier also
shows that TST-based methods [104, 105] cannot be used to calculate rates, as
they will result in extremely low transmission coefcients.
DPS of GB1 beta hairpin. Evans and Wales investigated the same GB1 hairpin with a discrete path sampling method [106]. Inspired by TPS, the DPS
method samples kinetically important pathways from a previously assembled database of metastable states and transition rates between these states
obtained via harmonic TST. The authors employed the CHARMM19 forceeld with the EEF1 implicit solvent. The resulting folding rate was predicted
only an order of magnitude lower than the experimental value. However,
there were some discrepancies with previous work, in particular on the stable states. Whether the discrepancies were due to differences in force elds,
undersampling of conguration space or to difculties in comparing the
structures is yet to be determined.
TPS of Trp-cage. Juraszek and Bolhuis [107] studied the Trp-cage mini-protein
in explicit solvent with TPS. They employed the OPLSAA force eld, with SPC
water, and obtained the path ensemble in an aggregate simulation time of several s. They observed very different folding pathways; one folding according
to the the diffusion-collision mechanism and one following the nucleationcondensation mechanism. The average path length was about 3 ns, indicating
that the transition was extremely diffuse. A committor analysis showed that
the water dynamics was not part of the reaction coordinate, something that
was already suggested for other small proteins by Pande et al. [108]. This
study shows that TPS is indeed capable of nding entire different pathways in
a complex system.
311
3.4
Enzyme Catalysis
Enzymatic reactions are extremely important in molecular biology. The difculty in studying them by computer simulation lies, besides in the quantum
nature of the reaction, in describing the reaction coordinate. In a few studies so far, TPS has enable researchers to test proposed reaction coordinates,
which in most cases led to new insights.
TPS of DNA polymerase . Radhankrishnan and Schlick [109111] performed
path sampling simulations to investigate the mechanism and delity of the
DNA repair process by polymerase. To that end they compared the kinetic
pathways of the incorporation of a correct nucleotide versus that of a incorrect base pair. Because this process consists of many sequential reactions
they split the problem into a series of transitions, a reaction network. Employing QM/MM (Gamess-UK with B3LYP in combination with CHARMM)
simulation for the trajectory generation they obtained path ensembles for
each of the transitions, requiring an aggregate amount of about 100 ns of
QM/MM simulation. The conclusion is that the mismatched base pair in the
polymerase indeed leads to a higher barrier, and hence enables a selection
criterion. (The method used in this work to estimate the free energy barrier
is in our view slightly biased. See, e.g., [112] for an correct treatment).
TPS of Lactate Dehydrogenase. Basner and Schwartz [113] have applied TPS
to the enzymatic reaction of lactate dehydrogenase (LDH). LDH catalyzes the
interconversion of the hydrogen acid acetate and the keto-acid pyruvate with
the coenzyme NADH (nicotinamide adenine dinucleotide). The reaction involves a proton transfer between the histidine of the active site and the oxygen
of the lactate, in addition to a hydride transfer between the NADH coenzyme.
Previous work revealed that the residues close to the active site have inuence on the coenzyme binding, substrate binding and orientation, and on the
reaction event. The order of the proton and hydride transfer was still controversial.
Using the QM/MM methodology (AM1 with CHARMM27) Basner and
Schwartz performed TPS of the proton transfer and hydride transfer. Due
to the demanding calculation, the aggregate simulation time was only a few
ps. The authors argue that the mechanism for the LDH-catalyzed reaction
from pyruvate to lactate involves both sequential and concerted proton and
hydride transfers. They base this conclusion on the fact that both possibilities are found in the transition path ensemble. Large scale residue motions
are paramount to the success of the reaction. The important residues behind
the hydride donor, the coenzyme NADH, move toward the active site, while
the residues located behind the substrate move away from it. Examination
of unreactive trajectories helped in revealing this mechanism. Fluctuations
312
and compression of the residues in the binding pocket turned out to be important. An enzyme-wide collective motion was observed as a contributing
factor of the enzyme rate acceleration. Whereas previous studies always had
to assume the reaction coordinate in advance, this work is the rst to successfully employ TPS for an enzymatic reaction, for which this assumption is not
required.
3.5
Lipid Bilayers
Living cells and their intracellular organelles are protected from the environment by membranes: lipid bilayers. These bilayers serve as compartment
boundaries, provide mechanical strength, host a wide variety of functional
proteins, and act as a capacitor to maintain ionic gradients between the inside
and outside of the cell. The computational study of lipid bilayers has taken an
enormous ight [114117]. Many studies have been carried out using atomistic force elds, but to incorporate the bilayers exibility, the system sizes
need to be large. Most simulation approaches therefore use simplied coarse
grained potentials.
Flip-op of lipids in membranes. Marti and Csajka [118, 119] investigated the
so-called ip-op transition where a lipid molecule with its head-group on
one side of the membrane migrates to the other side. Flip-op has been observed in naturally occurring as well in synthetic membranes. The difculty
of studying such ip-ops lies in the fact that they are very rare. The hydrophilic head-group must go through the hydrophobic interior of the membrane, something that is very unfavorable. The transition therefore happens
only on the order of seconds in a simulation with a reasonable system size.
Marti and Csajka employed a simplied lipid model, consisting of a bead
spring model with LJ interaction and soft core repulsions, for hydrophobic
interactions. They obtained an ensemble of 200 pathways of ip-op by TPS
at different temperatures. The average transition path was found to take less
than a picosecond. This separation of timescales is caused by the height of the
barrier, which was shown to be of the order of 10 kJ/mol at ambient temperature. The transition state ensemble was also obtained and showed very clearly
that a ipping lipid molecule adopts a rolled up conguration in the middle
of the membrane (i.e., parallel to the membrane interface). Also the amount
of disorder/uctuation was found to be higher in the transition state.
3.6
Switches in Biochemical Networks
Biochemical switches are networks of chemical reactions with more than one
steady state. The switches between those states can occur due to uctuations
313
in the system. Such networks are ubiquitous in living cells, e.g., the lac repressor in E. Coli and cell cycle regulation. Simulating such networks has relied
heavily on stochastic simulation methods. However, the stability of the steady
states in these networks make straightforward simulations exceedingly long.
Transition path sampling would be able alleviate this problem, if it were not
for the problem that most biochemical networks are out of equilibrium and
hence lack detailed balance. Allen et al. introduced in [74] a novel method,
based on TIS, that allows one to evaluate the kinetics of such biochemical
switches. Named forward ux sampling (FFS), the method generates the trajectories in a ratchet-like manner. It rst requires the denition of a series of
interfaces in the same way as in TIS. Starting from state A, trajectories are
generated using stochastic dynamics, and the probability of reaching the rst
interface is computed. The crossing points of the rst interface are recorded
and used for the generation of the new trajectories. This procedure is iterated until the last interface and the nal state is reached. The difference from
TIS is that only forward trajectories are generated, hence allowing for timeirreversible dynamics.
Allen et al. applied this method to a model genetic switch and found, although the switch is symmetric on exchanging the nal and initial states,
that the path ensemble (A B) differs from (B A). This implies that the
distribution of transition paths does not follow the steady state phase space
density.
4
Outlook
In this review article we have given an overview of the TPS methods in relation to biologically inspired problems. In biology most process are complex,
and mechanistic molecular insight is sometimes hard to achieve. Molecular
simulation techniques have advanced to the point that complex biological systems can be modeled in atomistic detail. This advancement complements the
experimental trend towards techniques resolving these systems at the molecular level. Characteristic of these biologically important systems are the
large differences in timescales that are relevant for functional activity. TPS
can bridge this difference in timescales.
In the past few years several groups have applied TPS to widely different biologically relevant processes. While the rst applications were on rather modest systems, in past few years it has become clear that even complex processes
such as protein folding can be treated with TPS. As available computer power
increases continuously, we can expect TPS studies on even more complex systems in the near future. In particular, the eld of enzymatic reactions might
show a growth in TPS applications, as these reactions are notoriously hard
to study with straightforward MD (QM/MM). We also might expect applica-
314
tion to other problems such as translocation, DNA condensation, transport

through ion channels, proteinDNA complexes, protein binding, and other
conformational changes. The work on biological networks emphasizes that
the TPS methods are not limited to molecular dynamics. It is even conceivable that they can be used for outstanding problems in population biology
and evolution.
From the methodology point of view there is room for many improvements. The problem of intermediate metastable states has not been solved
satisfactorily. The straightforward solution to this problem is to cut the overall reaction into several subprocess, as was for instance done in [109]. However, this approach requires a lot of knowledge about the process. The use of
Markovian state models or equilibrium kinetic networks might help in this
respect [23]. Also, the difculty of sampling multiple pathways still exists. The
work on protein folding showed that it is sometimes possible to nd qualitatively different pathways. Nevertheless it is very well possible that in other
cases nding different mechanisms requires alternative approaches such as
parallel tempering [49]. Besides the increased computational efciency offered by path sampling, much can be gained by the use of more accurate
coarse-grained force elds. Improvement in this area most certainly will enable the treatment of larger and more complex biological systems.
To summarize, we have shown that TPS has the potential to become a standard simulation tool in theoretical molecular biology studies. As rare events
play an important role in the life sciences it is very well possible that TPS
and its related techniques can make the connection from the single molecule
level all the way up to system biology, population biology, and maybe even
evolutionary biology.
References
1. Alberts B, Johnson A, Lewis J, Raff M, Roberts K, Walter P (2002) Molecular biology
of the cell, 4th edn. Garland Science, New York
2. Karplus M (ed) (2002) Special issue on molecular dynamics simulations of
biomolecules. Acc Chem Res 35:321
3. Eyring H (1935) J Chem Phys 3:107
4. Wigner E (1938) Trans Faraday Soc 34:29
5. Keck JC (1962) Discuss Faraday Soc 33:173
6. Bennett CH (1977) In: Christofferson R (ed) Algorithms for chemical computations,
ACS Symposium series no. 46. Am Chem Soc, Washington DC
7. Chandler D (1978) J Chem Phys 68:2959
8. Torrie GM, Valleau JP (1977) J Comput Phys 23:187
9. Kirkwood J (1935) J Chem Phys 3:300
10. Carter EA, Ciccotti G, Hynes JT, Kapral R (1989) Chem Phys Lett 156:472
11. Garcia-Viloca M, Gao J, Karplus M, Truhlar DG (2004) Science 303:186
12. Lazaridis T, Paulaitis ME (1994) J Am Chem Soc 116:1546
13. Roux B, Karplus M (1991) J Phys Chem 95:4856

14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
315
Truhlar DG, Garrett BC (1984) Ann Rev Phys Chem 35:159

Cerjan CJ, Miller WH (1981) J Chem Phys 75:2800
Doye JPK, Wales DJ (1997) Z Phys D 40:194
Munro LJ, Wales DJ (1999) Phys Rev B 59:3969
Wales DJ (2002) Mol Phys 100:3285
Wales DJ (1005) Phys Biol 2:S86
Evans DA, Wales DJ (2003) J Chem Phys 119:9947
Deuhard P, Huisinga W, Fischer A, Schtte Ch (2000) Lin Alg Appl 315:39
Elmer SP, Park S, Pande VS (2005) J Chem Phys 123:114902
Voter AF (1998) Phys Rev B 57:R13985
Shirts MR, Pande VS (2001) Phys Rev Lett 86:4983
Pande VS, Baker I, Chapman J, Elmer SP, Khaliq S, Larson SM, Rhee YM, Shirts MR,
Snow CD, Sorin EJ, Zagrovic B (2003) Biopolymers 68:91
Voter AF (1997) Phys Rev Lett 78:3907
Hamelberg D, McCammon JA (2005) J Am Chem Soc 127:13778
Grubmller H (1995) Phys Rev E 52:2893
Schulze BG, Grubmller H, Evanseck JD (2000) J Am Chem Soc 122:8700
Huber T, Torda AE, van Gunsteren WF (1994) J Comput-Aided Mol Design 8:695
Laio A, Parrinello M (2002) Proc Natl Acad Sci USA 99:12562
Ensing B, de Vivo M, Liu Z, Moore P, Klein ML (2006) Acc Chem Res 39:73
Ceccarelli M, Danelon C, Laio A, Parrinello M (2004) Biophys J 87:58
Hummer G, Kevrekidis IG (2003) J Chem Phys 118:10762
Pratt LR (1986) J Chem Phys 85:5045
Elber R, Ghosh A, Cardenas A, Stern H (2004) Adv Chem Phys 126:123 (2004)
Passerone D, Parrinello M (2001) Phys Rev Lett 87:108302
Elber R, Karplus M (1987) Chem Phys Lett 139:375
Sevick EM, Bell AT, Theodorou DN (1993) J Chem Phys 98:3196
Voter AF (1997) Phys Rev Lett 78:3908
Gillilan RE, Wilson KR (1996) J Chem Phys 105:9299
Henkelman G, Johannesson G, Jonsson H (2000) In: Schwartz SD (ed) Progress on
theoretical chemistry and physics. Kluwer Academic, Dordrecht, The Netherlands
Mathews DH, Case DA (2006) J Mol Biol 357:1683
E W, Ren W, Vanden-Eijnden E (2002) Phys Rev B 66:052301
E W, Ren W, Vanden-Eijnden E (2005) J Phys Chem 109:6688
Ren W, Vanden-Eijnden E, Maragakis P, E W (2005) J Chem Phys 123:134109
Dellago C, Bolhuis PG, Csajka FS, Chandler D (1998) J Chem Phys 108:1964
Dellago C, Bolhuis PG, Geissler PL (2002) Adv Chem Phys 123:1
Chandler D (1987) Introduction to modern statistical mechanics. Oxford University
Press, New York
Frenkel D, Smit B (2002) Understanding molecular simulation, 2nd edn. Academic,
San Diego, CA
Binder K, Heermann D (2002) Monte Carlo simulation in statistical physics.
Springer, Berlin Heidelberg New York
Bolhuis PG, Chandler D, Dellago C, Geissler PL (2002) Ann Rev Phys Chem 53:291
Dellago C, Chandler D (2002) Bridging the time scale gap with transition path sampling. In: Nielaba P, Mareschal M, Ciccotti G (eds) Molecular simulation for the next
decade. Springer, Berlin Heidelberg New York, p 321
Dellago C (2005) Transition Path Sampling. In: Yip S (ed) Handbook of materials
modeling. Springer, Berlin Heidelberg New York, p 1585
316
56. Dellago C (2007) Transition path sampling and the calculation of free energies.
In: Chipot C, Pohorille A (eds) Free energy calculations: Theory and applications in
chemistry and biology. Springer series in chemical physics, vol 86. Springer, Berlin
Heidelberg New York (in press)
57. Dellago C, Bolhuis PG, Chandler D (1998) J Chem Phys 108:9263
58. Bolhuis PG, Dellago C, Chandler D (1998) Faraday Discuss 110:421
59. Car R, Parrinello M (1985) Phys Rev Lett 55:2471
60. Nos S (1984) J Chem Phys 81:511
61. Nos S (1984) Mol Phys 52:255
63. Hoover WG, Ladd AJC, Moran B (1982) Phys Rev Lett 48:1818
64. Evans DJ (1983) J Chem Phys 78:3297
65. van Erp TS, Moroni D, Bolhuis PG (2003) J Chem Phys 118:7762
66. Bolhuis PG (2003) J Phys Cond Matter 15:S113
67. Onsager L (1938) Phys Rev 54:554
68. Klosek MM, Matkowsky BJ, Schuss Z (1991) Ber Bunsen Phys Chem 95:331
69. Pande V, Grosberg AY, Tanaka T, Shakhnovich EI (1998) J Chem Phys 108:334
70. Hummer G (2004) J Chem Phys 120:516
71. Best RB, Hummer G (2005) Proc Natl Acad Sci USA 102:6732
72. Bolhuis PG (2003) Proc Natl Acad Sci USA 100:12129
73. Bolhuis PG (2005) Biophys J 88:50
74. Allen R, Warren PB, ten Wolde P (2005) Phys Rev Lett 94:018104
75. Qiu D, Shenkin PS, Hollinger FP, Still EC (1997) J Phys Chem A 101:3005
76. Lazaridis T, Karplus M (1999) Prot Struct Func Gen 35:133
77. Taketomi H, Ueda Y, Go N (1975) Int J Pept Protein Res 7:445
78. Clementi C, Garcia AE, Onuchic JN (2003) J Mol Biol 326:933
79. Liwo A, Khalili M, Scheraga HA (2005) Proc Natl Acad Sci USA 102:2362
80. van Erp TS (2006) Condens Mat, abstract cond-mat/0605260
81. Moroni D, Bolhuis PG, van Erp TS (2004) J Chem Phys 120:4055
82. Faradjian AK, Elber R (2004) J Chem Phys 120:10880
83. Bolhuis PG, Dellago C, Chandler D (2000) Proc Natl Acad Sci USA 97:5877
84. Maragliano L, Fischer A, Vanden-Eijnden E, Ciccotti G (2006) J Chem Phys
125:024106
85. Ma A, Dinner AR (2005) J Phys Chem B 109:6769
86. McCormick TA, Chandler D (2003) J Phys Chem B 107:2796
87. Hagan MF, Dinner AR, Chandler D, Chakraborty AK (2003) Proc Natl Acad Sci USA
100:13922
88. Dobson CM (2004) Semin Cell Dev Biol 15:3
89. Fersht A (1999) Structure and mechanism in protein science. Freeman, New York
90. Cheung MS, Chavez LL, Onuchic JN (2004) Polymer 45:547
91. Kubelka J, Hofrichter J, Eaton WA (2004) Curr Opin Struc Biol 14:76
92. Mirny L, Shakhnovich E (2001) Annu Rev Bioph Biom 30:361
93. Shea JE, Brooks CL (2001) Annu Rev Phys Chem 52:499
94. Gnanakaran S et al. (2003) Curr Opin Struc Biol 13:168
95. Snow CD, Sorin EJ, Rhee YM, Pande V (2005) Annu Rev Biophys Biomol Struct 34:43
96. Bolhuis PG (2006) Sampling kinetic protein folding pathways using all atom models.
In: Landau DP, Lewis SP, Schttler H-B (eds) Computer simulations in condensed matter: from materials to chemical biology. Springer proceedings in physics.
Springer, Berlin Heidelberg New York (in press)
97. Eastman P, Gronbech-Jensen N, Doniach S (2001) J Chem Phys 114:3823

98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
317
Elber R, Ghosh A, Cardenas A, Stern H (2003) Adv Chem Phys 126:93

ten Wolde PR, Chandler D (2002) Proc Natl Acad Sci USA 99:6539
Pande VS, Rokhsar DS (1999) Proc Natl Acad Sci USA 96:9062
Roccatano D, Amadei A, Di Nola A, Berendsen HJC (1999) Protein Sci 8:2130
Garcia AE, Sanbonmatsu KY (2001) Proteins 42:345
Zhou RH, Berne BJ, Germain R (2001) Proc Natl Acad Sci USA 98:14931
Chandler D (1978) J Chem Phys 68:2959
Bennett CH (1977) In: Christofferson R (ed) Algorithms for chemical computations:
ACS Symposium series No 46. Am Chem Soc, Washington DC
Juraszek J, Bolhuis PG (2006) Proc Natl Acad Sci USA (in press)
Rhee YM et al. (2004) Proc Natl Acad Sci USA 101:6456
Radhakrishnan R, Schlick T (2004) Proc Natl Acad Sci USA 101:5970
Radhakrishnan R, Yang LJ, Arora K, Schlick T (2004) Biophys J 86:34A
Radhakrishnan R, Schlick T (2005) J Am Chem Soc 127:13245
Moroni D, van Erp TS, Bolhuis PG (2005) Phys Rev E 71:056709
Basner JE, Schwartz SD (2005) J Am Chem Soc 127:13822
Goetz R, Lipowsky R (1998) J Chem Phys 108:7397
Smondyrev AM, Berkowitz ML (1999) J Comput Chem 20:531
Marrink SJ, Jhnig F, Berendsen HJC (1996) Biophys J 71:632
Bemporad D, Essex JW, Luttmann C (2004) J Phys Chem B 108:4875
Marti J, Csajka FS (2004) Phys Rev E 69:061918
Marti J (2004) J Phys: Condens Mat 16:5669
Top Curr Chem (2007) 268: 319347

DOI 10.1007/128_2006_086
PcrA Helicase, a Molecular Motor Studied

from the Electronic to the Functional Level
Markus Dittrich Jin Yu Klaus Schulten (u)
Beckman Institute, University of Illinois at Urbana-Champaign, Urbana,
Illinois, IL 61801, USA
kschulte@ks.uiuc.edu
1
General Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
320
PcrA Helicase, a Prototype Molecular Motor . . . . . . . . . . . . . . . . .
322
Zooming in on ATP Hydrolysis in the Catalytic Site . . . . . . . . . . . . .
323
Molecular Dynamics Simulations of PcrA Function . . . . . . . . . . . . .
332
Langevin Simulations of PcrA Function . . . . . . . . . . . . . . . . . . .
339
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
344
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
346
Abstract Molecular motors are adenosine tri-phosphate (ATP) hydrolysis-driven, cellular proteins responsible for a wide variety of different tasks, such as transport, energy
metabolism, and DNA processing. Their operation cycle spans a wide range of length and
time scales, from the localized and fast chemical reaction in the catalytic site(s) to the
large scale and much slower conformational motions involved in the motors physiological function. From a computational point of view, this means that currently there exists
no single approach capable of capturing the whole spectrum of events during molecular
motor function. In the present review, we show for PcrA helicase, a molecular motor involved in the unwinding of double-stranded DNA, how a combination of computational
approaches can be used to examine PcrAs function in its entirety as well as in detail.
Combined quantum mechanical/molecular mechanical simulations are used to study the
catalytic ATP hydrolysis event and its coupling to protein conformational changes. Molecular dynamics simulations then provide a means of studying overall PcrA function
on a nanosecond time scale. Finally, to reach physiologically relevant time scales, i.e.,
milliseconds, stochastic simulations are employed. We show that by combining the three
stated approaches one can obtain insight into PcrA helicase function.
Keywords ATP hydrolysis DNA helicase Molecular dynamics simulations
Molecular motor QM/MM PcrA Stochastic modeling
Abbreviations
ADP
adenosine di-phosphate
ATP
adenosine tri-phosphate
dsDNA double-stranded DNA
320
DNA
MD
MM
nt
Pi
QM
QM/MM
RESP
RNA
SMD
ssDNA
vdW
M. Dittrich et al.
deoxyribonucleic acid
molecular dynamics
molecular mechanics
nucleotide
phosphate
quantum mechanics
quantum mechanical/molecular mechanical
restrained electrostatic potential
ribonucleic acid
steered molecular dynamics
single-stranded DNA
van der Waals
1
General Introduction
Molecular motors are proteins essential for cellular transport, metabolism,
and signaling. They are powered by the energy stored in nucleotide triphosphates, mostly adenosine tri-phosphate (ATP). At the heart of molecular
motor function is the conversion of the chemical energy stored in ATP into
a mechanical force which drives the physiological function of the system. The
chemo-mechanical coupling underlying the conversion of ATPs energy is still
poorly understood. Indeed, even though several molecular motor proteins are
known at a structural level, e.g., myosin, kinesin, F1 -ATPase, or helicase, and
even though for each of them a vast amount of biochemical data is available,
decades of study have not revealed how the chemo-mechanical coupling is
being achieved in microscopic detail by any of these systems.
However, the availability of structural information allows computational
studies to be conducted and a signicant number have been reported over
the past decade for such diverse systems as myosin [15], kinesin [3, 6],
F1 -ATPase [712], and helicase [1317]. For F1 -ATPase, e.g., molecular dynamics (MD) simulations provided insight into the inner workings of the
rotation of the central stalk inside the hexameric head-unit [8]; combined
quantum mechanical/molecular mechanical (QM/MM) simulations of the
ATP hydrolysis reaction in the catalytic sites [1012] on the other hand revealed a mechanism for efcient ATP catalysis and were able to identify
several residues involved in the coupling of the chemical reaction to larger
scale conformational changes of the protein. On the basis of these available
structures and biochemical data, Oster and coworkers [7] have developed
stochastic and kinetic models for F1 -ATPase function.
Molecular motor proteins present a number of formidable challenges to
the computational researcher. First, their operation involves time scales covering several orders of magnitude, from the fast chemical reaction step which
occurs over the course of a few femtoseconds all the way to the mechani-
PcrA Helicase
321
cal motions responsible for the systems physiological function taking place
on a microsecond up to a millisecond time scale. Second, the events during
motor function span a wide range of spatial scales, from the localized and
electronic level changes accompanying the bond-breaking event in the cataA) to the large scale structural
lytic binding site(s) during ATP catalysis ( 1
A to 10 nm).
changes involved in physiological function (10
From a methodological point of view this means that there is presently no
single computational approach which is able to model the whole spatial and
temporal scale of events underlying molecular motor function. At least three
levels of computational methodology are necessary to capture the relevant degrees of freedom. A QM or QM/MM method is required to investigate the
electronic events underlying the ATP hydrolysis reaction and its interaction
with the protein environment. Typically, however, only localized structural
changes and/or relatively short time scales can be accessed using QM/MM
simulations and MD is needed to study the properties of the full protein
in a solvent environment on a multi nanosecond time scale at the expense
of neglecting electronic level processes such bond-breaking. In conjunction
with methods such as steered MD (SMD) [48, 49] this allows one to study the
larger scale motions underlying motor function, e.g., conformational changes
coupled to binding of reactant ATP or unbinding of hydrolysis products ADP
and Pi as well as protein movements relayed to the motors binding partners
such as actin in case of myosin. However, using even the most powerful computational resources currently available, the time scales accessible to all-atom
MD simulations are still orders of magnitude away from the true physiological ones. Hence, in order to reach the latter regime one has to resort to an
approach which focuses on a small number of essential degrees of freedom.
One might, for example, model the motor function via the center of mass motion of the relevant protein domains on a low-dimensional potential energy
surface governed by a Langevin-type stochastic equation.
The challenge from a modeling point of view, then, is to connect these
approaches to obtain a coherent picture of the overall process. Ideally, the insight gained on each level is used to design and improve the methodology
of the other levels, possibly in an iterative fashion. As an example, the essential degrees of freedom and the potentials governing the stochastic equations
modeling the long time behavior of the system could be extracted from appropriately designed MD simulations. Conversely, insight gained from the
stochastic modeling might then catalyze an improved setup of the MD simulations.
Here, we summarize progress on how such a comprehensive computational approach can be used to gain insight into the function of one particular molecular motor, the DNA helicase PcrA. This motor protein will
be introduced in the next section. We will then lay out the computational
approaches used and nally summarize and interpret recent computational
ndings.
322
M. Dittrich et al.
2
PcrA Helicase, a Prototype Molecular Motor
DNA helicases are important participants in DNA metabolism, which involves
processes such as replication, transcription, and repair. They catalyze the separation of double-stranded DNA (dsDNA) into its single-stranded DNA (ssDNA) components. Helicases are typically found to exist in a monomeric [18,
19], dimeric [20], or hexameric form [21] and unwind DNA either in a 3 5
or 5 3 direction.
PcrA is a monomeric 3 5 helicase, and due to its relatively small size
(80 kDa) constitutes an ideal candidate for computational study. Indeed, PcrA
is one of the smallest molecular motors known to date. Several atomic resolution X-ray structures of PcrA have been reported [18, 19]. The present discussion will focus on two PcrA structures from Bacillus stearothermophillus
that were determined by Velankar et al. [19] in a substrate (with bound ATP
A
analogue AMP-PNP) and product (without ATP/ADP bound) state at 3.3
and 2.9 A resolution, respectively. Figure 1 shows a molecular view of PcrA

which consists of four major domains called 1A, 1B, 2A, and 2B. Also visible
is a strand of ssDNA threading through the protein at the interface formed
by the A and B domains and which is connected to a short piece of dsDNA
bound to the 2B domain. ATP binds to the catalytic binding pocket located at
the interface between the 1A and 2A domains.
Fig. 1 Molecular view of PcrA helicase. a Depicted is the protein in cartoon representation. Also shown is a strand of ssDNA in licorice representation threading through the
center of PcrA at the interface of the 1A (green) and 2A domains (red) and which is attached to a short piece of dsDNA adjacent to the 2B domain (blue); also shown is the 1B
domain (yellow) and a single molecule of ATP (orange) bound to the catalytic site at the
interface between the 1A and 2A domains (depicted in vdW representation). b Shown is
a schematic view of PcrA translocating along ssDNA toward the dsDNA fork; domains 1A,
2A, 1B, and 2B are shown in green, red, yellow, and blue, respectively
PcrA Helicase
323
In the following, we will focus on the mechanism of PcrA movement along

ssDNA. On the basis of their crystal structure data and on the intuitive idea
of alternating afnities, Velankar et al. [19] proposed an inchworm model
for the translocation of PcrA along ssDNA that involves concerted movements
of ssDNA bases across the interface with PcrA. This model does not, however,
address how ATP hydrolysis is coupled to domain motion, what residues are
most crucial during this process, how the proposed alternating afnities can
be justied microscopically and nally, how uni-directionality is achieved.
These are the questions that will be discussed in the following sections. We
will start by looking at the ATP hydrolysis reaction in the catalytic site and
then work our way up toward longer time and larger length scales.
3
Zooming in on ATP Hydrolysis in the Catalytic Site
The catalysis of ATP lies at the heart of PcrA function and takes place in
a single catalytic site at the interface between the 1A and 2A domain (Fig. 1).
Several important questions arise when considering the role of ATP catalysis during the overall translocation process of PcrA along ssDNA. First and
from a purely enzymatic point of view, one would like to know how efcient ATP catalysis is being achieved by the protein. Second, one would like to
investigate how the actual catalytic step is coupled to larger scale protein conformational changes eventually leading to PcrA translocation along ssDNA.
An investigation of both questions requires a proper treatment of the
electronic degrees of freedom during the catalytic reaction in the biological context provided by the protein environment. QM/MM methods present
a powerful and versatile way of addressing these issues, since they allow one
to conduct high level electronic structure calculations of a particular region of
interest in the presence of a classically treated protein environment. The following paragraph provides a brief outline of the principles behind QM/MM
calculations, before we describe some of the insights into PcrA function that
have emerged using this approach.
QM/MM Calculations
The underlying theme of QM/MM simulations [2229] is a quantum mechanical treatment of a certain molecular core region that is undergoing electronic
level changes in the presence of a (large) biomolecular environment given
by, for example, protein, solvent, or DNA/RNA. The environment itself is assumed to be electronically inert and is, therefore, treated classically, typically
with a molecular mechanics method. The main motivation for this separation into a quantum mechanically and a classically treated region is the fact
that quantum mechanical calculations are computationally very expensive
324
M. Dittrich et al.
and can only be applied to relatively small systems of at most a few hundred
atoms. Hence, QM/MM simulations provide a compromise between the need
to treat certain parts of the system at an electronic level and the requirement
for the presence of the sizeable host biomolecule or biomolecular complex.
Depending on the particular QM/MM method chosen, it is sometimes necessary to limit the number of classically treated MM atoms. In the case of
PcrA which contains more than 110 000 atoms in its solvated state (Sect. 4),
the conducted QM/MM calculations described below were restricted to a reduced sub-system containing only 20 000 atoms that was centered around the
nucleotide ATP bound to the catalytic site. This is depicted schematically in
Fig. 2, which shows the full QM/MM system indicating the quantum mechanically (QM) and classically treated regions (MM), respectively.
The remainder of this section will provide a brief overview of QM/MM
simulations in general and the implementation used in the studies summarized in the present manuscript in particular. In all of the following it is
assumed that the BornOppenheimer approximation is valid, i.e., electronic
and nuclear degrees of freedom can be separated. In this case, the QM/MM
Hamiltonian describing the quantum mechanically modeled core region and
the classically treated protein environment can formally be expressed as
H = HQM + HMM + HQM/MM ,
(1)
Fig. 2 QM/MM simulation system for PcrA. Shown is the complete model which is centered around the catalytic binding site located at the interface between the 1A and 2A
domains and contains about 20 000 atoms. The protein is shown in cartoon representation, a piece of bound ssDNA is depicted in licorice representation, and bound ATP
in vdW representation. The classically (MM) and quantum mechanically (QM) treated
regions are indicated by circles and labeled accordingly
PcrA Helicase
325
where HQM can be identied with the Hamiltonian of the time-independent

Schrdinger equation
HQM |({r}) = E |({r})
(2)
and the classical contribution HMM is given by Eq. 14 below.

The main challenge for achieving a proper QM/MM description lies in
the treatment of the interface between the quantum and classical regions,
described by HQM/MM . QM/MM studies of chemical reactions in solution allow for a straightforward separation of both regions, e.g., the solvent being
treated using a classical description and the chemically active solute being
modeled by a QM method. QM/MM simulations of biomolecular systems like,
for example, enzymes, on the other hand, often lead to interfaces separating
the quantum and classical regions that cut through covalent bonds, thereby,
complicating the description signicantly.
Broadly, one can distinguish three types of QM/MM coupling schemes.
These are listed below in the order of increasing complexity:
Mechanical Embedding. This represents the crudest level of description.
The QM and MM regions are not connected by any covalent bonds and interact only via steric constraints, typically implemented in terms of a van
der Waals (vdW) type interaction:

ai
HQM/MM =
HvdW
.
(3)
aMM iQM
The two sums extend over all nuclei in the classically (MM) and quantum mechanically (QM) treated region, respectively. The choice of vdW
parameters is not straightforward. Often, the parameters provided by the
classical forceeld are employed, but QM/MM studies with specically parameterized vdW interactions have also been reported.
Electrostatic Embedding. At this level of QM/MM coupling, in addition to
steric constraints between the QM and MM regions, one also takes into account their electrostatic interaction, leading to a QM/MM Hamiltonian of
the form:

ai
ai
HvdW
+ Hele
.
(4)
HQM/MM =
aMM iQM
Here, Hele describes the electrostatic interaction between the QM and MM

regions. As will be discussed in more detail below, this term can be evaluated in several different ways.
Covalent Embedding. This most sophisticated level of QM/MM scheme involves, in addition to steric and electrostatic interactions, covalent bonds
between QM and MM atoms. Typically, this is the description necessary
for simulations of biomolecular systems, and was used in our QM/MM
approach.
326
M. Dittrich et al.
The next section is devoted to a more in depth exposition of the particular

covalent embedding scheme used in our calculations.
RESP Charge-Based Covalent Embedding
A number of different covalent embedding schemes have been proposed since
the early QM/MM studies of Warshel [30]. Here, rather than trying to be
comprehensive, we focus on the particular method employed in the QM/MM
implementation on which the reported results are based and refer to the literature for additional information. See also the review on QM/MM methods by
Senn and Thiel in this volume.
When describing quantum mechanical events inside biomolecules one frequently encounters covalent bonds across the QM/MM boundary connecting
classically and quantum mechanically treated atoms. This is problematic for
several reasons. First and foremost, cutting these bonds creates an open
valence in the QM region. Several schemes have been devised to deal with
this problem. The probably most straightforward method is the so-called link
atom approach [23] that is also used in our QM/MM method. Here, the dangling bond is simply terminated by a dummy hydrogen atom. More sophisticated schemes have been proposed, for example, the local self-consistent
eld method by Thry et al. [31] in which a frozen hybrid orbital with a predetermined density is used to satisfy the valence at the boundary. However,
even though the introduction of an additional (hydrogen) atom in the link
atom approach can in principle cause articial behavior at the boundary, it
is currently not clear that alternative, more sophisticated methods provide
a much improved description.
A second problematic issue arising from a QM/MM boundary that dissects
covalent bonds is the difculty of achieving a proper separation of the QM
and MM interaction terms in order to avoid the double counting of certain
interactions. This will be discussed in more detail below.
The QM/MM interface Hamiltonian of the covalent embedding scheme describing the quantum-classical interaction can be written as
HQM/MM =

QM/MM
ai
ai
HvdW
+ Hele
+ Hcovalent
.
(5)
aMM iQM
QM/MM
The Hcovalent
contribution maintains the proper conformation across the
covalent quantum-classical boundary and is provided by the molecular mechanics force eld. Quantum mechanically treated atoms in the boundary
region experience all classical bonded force contributions that involve at least
one classical atom.
In most conventional QM/MM approaches the electronic contribution
ai in Eq. 5 is given via the one electron
to the electrostatic interaction Hele
PcrA Helicase
327
operator
ai
Hele,e
=
qa
,
|Ra ri |
(6)
where i indexes the electron coordinates and qa is the partial charge on

the classical atom a at position Ra . Unfortunately, as indicated above, this
approach leads to problems at the covalent QM/MM boundary where electrostatic interactions between quantum and classical atoms are already implicitly
accounted for via the classical bond, angle, and dihedral energies. Subtracting these interactions from the expression Eq. 6 is difcult since the
electronic charge is de-localized over the whole QM segment and, therefore, can not be easily partitioned among atoms. In other words, there is
no quantum mechanical charge operator that would allow the assignment of
partial atomic charges. To address this shortcoming, our QM/MM approach
introduces an effective charge operator p for each atomic site in the QM
segment, that is evaluated via the restrained electrostatic potential (RESP)
charge method [32]. This leads to the following expression for the electrostatic potential
Hele = Hele,e + Hele,n
=

aMM
(7)

Zp qa
p qa
+
,
ap
ap
Ra Rp
Ra Rp
pQM
aMM pQM
where Hele,e and Hele,n are the electronic and nuclear contribution, respectively. Here qa and Zp are the partial charges on the classical atoms and
nuclear charges of the quantum mechanically modeled atoms, respectively.
The ap mediate the proper exclusion of the classical non-bonded interactions
at the QM/MM boundary, and is a scaling factor to ensure the proper total
charge of the QM/MM system.
Using the Hamiltonian Eq. 1 together with the QM/MM interaction energy
Eq. 5, one obtains the following expression for the total energy E
E = |H|
= EQM (P) +
(8)

aMM
p (P)qa
+ EMM .
ap
Ra Rp
pQM
Here, | is the electronic wave function and P the corresponding electronic

density matrix. The term EQM is the expression for the gas phase energy of
the QM part, and EMM includes all remaining non-electronic energy contributions. In the second term, p (P) is the effective charge for atoms in the QM
segment, which is evaluated via the RESP method by minimizing the follow-
328
M. Dittrich et al.
ing expression
p (P)

V (P) + 2e

p (P) Ne
I=
R Rp
pQM
pQM

2
gp p (P) + Zp .
+
(9)
pQM
The rst term on the right-hand side attempts to nd effective charges p (P)
that best reproduce the exact electronic electrostatic potential V (P) subject
to the constraints expressed in the second and third term. Here, sums over
the points of a grid located at positions R surrounding the QM segment
at which the electrostatic potential is evaluated. The are weight factors
whose value depends on the particular choice for the computational grid. The
second term enforces charge conservation and constrains the overall effective charge to be equal to the total number of electrons Ne via a Lagrange
multiplier e . Both V (P) and Ne have to take into account the presence of
the dummy hydrogen atoms. Finally, the third term represents a harmonic
penalty function that can be adjusted by the parameter gp to prevent the appearance of unreasonably large effective charges [32].
The minimization of the expression given by Eq. 9 can be cast into a matrix
equation for p (P) which can then be used to formulate a new Fock operator
F QM/MM = F + F solv ,
(10)
where F is the gas-phase Fock operator and F solv the so-called solvated
Fock operator mediating the electrostatic interaction between the QM and
MM segment of the QM/MM system [33]. The self-consistent solution of the
Roothaan-Hall equations with the Fock operator given by Eq. 10 is the mathematical basis of the QM/MM approach used in the studies described in the
remainder of this section.
Finally, we would like to point out that due to the computational cost involved, the energetics derived from ab initio QM/MM calculations typically
are enthalphies rather than free energies. The latter are required, to, for example, derive true reaction free energies or equilibrium constants. Hence, for
each particular system it is necessary to take into account possible entropic
effects when arguing based on calculated enthalpies. There exist, however,
semi-empirical QM/MM approaches that are capable of computing true free
energy surfaces, at the possible expense of reduced accuracy and transferability [34, 35].
QM/MM Study of the Chemo-Mechanical Coupling in PcrA Helicase
The QM/MM approach just described was employed to investigate the ATP
hydrolysis reaction in the catalytic site of PcrA helicase and its coupling to
PcrA Helicase
329
protein conformational changes, i.e., the chemo-mechanical coupling [36].

The reactant state conformation of the quantum mechanically treated region
is shown in the upper left-hand side panel of Fig. 3 which also depicts important distances between atoms.
Figure 3 reveals that all binding pocket residues immediately adjacent to
the tri-phosphate part of ATP are either charged or polar, with several water
molecules surrounding the -phosphate group. A magnesium ion is located
between two oxygen atoms of the and -phosphate groups and is hydrogenbonded to additional water molecules and binding pocket residues. Most of
the residues forming the catalytic binding sites are located in the 1A domain
of PcrA with the exception of two arginine residues, R287 and R610, which are
contributed by the neighboring 2A domain (c.f. bottom panel of Fig. 3).
Using QM/MM calculations, the ATP hydrolysis reaction pathway can be
calculated [36] and the conformations of the reactant, transition, and product
state are shown in the upper panel of Fig. 3 with the corresponding energies
depicted in Fig. 4.
The center column of Fig. 3 shows the transition state conformation during
ATP hydrolysis. It features a hydronium ion that evolved via proton transfer
from the nucleophilic water, W1, toward W2. During the conversion from the
Fig. 3 Conformation of the quantum mechanically treated core region in the catalytic site
of PcrA. Shown in the top panel are molecular views of the structures of the reactant (top
left), transition (top middle), and nal (top right) state along the ATP hydrolysis reaction pathway. Important distances between atoms are indicated and given in units of
A.
The bottom panel depicts schematic views of the catalytic site in each of the three states
and shows which of the residues belong to domain 1A or 2A. The bold arrows sketch the
proton relay mechanism discussed in the text. The color coding is as follows: white - hydrogen; light blue carbon; dark blue nitrogen; red oxygen; gold phosphorus; green
magnesium ion (adapted from [36])
330
M. Dittrich et al.
Fig. 4 Energetics of ATP hydrolysis in PcrA. Shown are the energies in kcal/mol of the
reactant, transition, and product state during ATP hydrolysis in the catalytic sites of PcrA.
For the product state, the energies for the wild type and two mutants, Q254N and Q254G,
are provided (adapted from [36])
transition state toward the product conformation, the hydronium ion transfers a second proton to the -phosphate group to yield products ADP and Pi .
This mechanism of proton transfer from the nucleophilic water toward the
-phosphate group of ATP is sketched by the bold arrows in the bottom panel
of Fig. 3 and was termed a proton relay mechanism [10], since it involves two
water molecules, W1 and W2. Until recently, it was thought that direct proton
transfer from the nucleophilic water, W1, toward one of the -phosphate oxygen atoms of ATP is the physiologically relevant pathway. However, QM/MM
studies have shown that the proton relay mechanism is energetically more
favorable by up to 20 kcal/mol [36]. Such a mechanism had also been proposed based on theoretical studies of other ATP-driven molecular motors,
such as F1 -ATPase [10, 12] and myosin [1], and had been proposed also based
on structural and biochemical evidence for ras [37] and F1 -ATPase [38]. All
combined, this evidence suggests that such a proton relay mechanism is responsible for efcient hydrolysis in a wider class of ATP-driven molecular
motors. Since the proton relay mechanism relies on the proper arrangement
of water molecules in the binding pocket, which is in turn facilitated by the
protein environment, it represents a genuinely enzymatic pathway.
Figure 4 gives the energies of important states along the ATP hydrolysis pathway in PcrA. Several features are noteworthy: First, the QM/MM
calculations [36] reveal the presence of a modest transition state barrier
of 20 kcal/mol in good agreement with experimentally measured rates
( 102 s1 ). This moderate barrier is due to the proton relay mechanism as
shown by comparison with calculations of direct proton transfer pathways. In
addition, there are several binding pocket residues that contribute to a lowering of the transition state barrier. This is shown in Fig. 5 which depicts
the change in electrostatic interaction between ATP and the protein environment upon going from the reactant to the transition state. Clearly, K37,
PcrA Helicase
331
Fig. 5 Electrostatic interactions between catalytic core and protein environment. Shown
are the changes in electrostatic interaction between the nucleotide and the protein environment upon going from the reactant to the transition state (adapted from [36])
E224, and R287 stabilize the transition state, whereas K568 leads to a slight
de-stabilization. Somewhat surprisingly, even though R287 provides a large
contribution, its immediate structural neighbor, R610, does not.
The second important feature to notice in Fig. 4 is the endothermicity of
the reaction energy prole in the wild type system, implying that the catalytic
site in the conformation studied has the ability to tightly bind ATP, but not
yet to hydrolyze it. To investigate this in more detail and to identify means by
which the protein could achieve a lowering of the product state energy, in silico mutation studies can be used [36]. Here, selected residues are mutated in
the model system and the reaction energy prole is recalculated either fully
or for selected points along the reaction pathway. In the present case, such
an analysis shows, for example, that the mutations Q254N and Q254G exert
a signicant inuence on the product state energy. This, in turn, allows one
to hypothesize that motion of Q254 with respect to P leads to the lowering
of the product state energy to enable efcient ATP hydrolysis. Q254 links to
a protein pocket known to be involved in ssDNA base ipping [19] during
translocation via the inchworm mechanism. This suggests a direct connection
between translocation along ssDNA and the actual catalytic step in the ATP
binding pocket of PcrA via Q254.
Finally, based on the ATP hydrolysis reaction energy prole in PcrA shown
in Fig. 4, the in silico mutation studies, and the ndings from QM/MM simulations of ATP hydrolysis in F1 -ATPase [10, 12], the overall ATP catalysis reaction energy prole can be assumed to have an equilibrium constant of K 1,
corresponding to an iso-energetic reaction energy prole [36]. This has important implications, since it means that the actual chemical bond-breaking
step in the catalytic sites of PcrA does not release any net free energy and is
332
M. Dittrich et al.
hence not coupled to force generation. Similar to F1 -ATPase [7, 12, 39], it is
likely that the binding of reactant ATP to the initially open catalytic cleft at
the interface between the 1A and 2A domains leads to a closing motion which
provides the actual power stroke propagating PcrA along ssDNA. This picture
ts nicely with the above observation that insertion of R287 from domain 2A
into the catalytic site located in the 1A domain is required for efcient ATP
hydrolysis to take place. Such a coupling of large-scale protein conformation
and the chemical reaction assures that ATP hydrolysis does not occur prematurely, which could destroy the directed motion of PcrA along ssDNA [36].
However, due to the static nature of the employed QM/MM method and the
limited spatial extent of the QM/MM system, one is limited in ones ability to
test the domain motion hypothesis and the coupling of Q254 to ssDNA base
ipping on realistic time and length scales. This is where all-atom MD simulations and coarser-grained descriptions come into play. These methods will
be the topic of the following sections.
4
Molecular Dynamics Simulations of PcrA Function
Continuing our study of molecular motor function in PcrA, we turn to MD
simulations to investigate PcrAs larger scale conformational motions leading
to translocation along ssDNA and to study the coupling of ATP hydrolysis
to unidirectional translocation. Figure 6 shows a molecular view of an allatom simulation system of PcrA, containing the protein, DNA, TIP3P water
molecules, and a number of ions adding up to a total of more than 110 000
atoms. Before discussing some of the ndings for PcrA function obtained via
MD simulations we will give a brief exposition of MD simulation methods.
MD SimulationsA Brief Introduction
Classical simulations are based on the solution of the differential equation
posed by Newtons second law for a system of N interacting particles
x i =
1
Fi
mi
(i = 1, ..., N) .
(11)
Here x i is the second time derivative of the position of particle i, mi its mass,
and F i the total force acting on it. F i depends on the positions of all other
particles. The calculation of the forces F i is the step requiring the most computational effort.
A number of algorithms have been developed to numerically integrate
Eq. 11 and, thereby, obtain the particles trajectories xi (t). Many integration
schemes use the Leap frog algorithm for updating coordinates and velocities
PcrA Helicase
333
Fig. 6 PcrA all-atom simulation system. Shown is the protein in a blue cartoon representation with bound dsDNA and ssDNA shown in red vdW representation. The nucleotide
ATP bound to the catalytic site at the interface between the 1A and 2A domains is shown
as gold-colored vdW spheres. The protein-DNA complex is immersed in a large box
of TIP3P water molecules containing Cl (cyan), Na+ (yellow), and Mg2+ (green) ions
(adapted from [51])
of particle i:

(12)
ri t + t = r i (t) + t vi t + t/2

1
vi t + t/2 = vi t t/2 + t F i ,
(13)
mi
with t being the value of the discrete time-step. In biomolecular simulations, the value of t has to be a fraction of the fastest vibrational period in
the system and, therefore, is in the 12 fs range.
All classical simulations are based on a classical approximation to the
Hamilton operator in the Schrdinger Eq. 2. With this classical Hamiltonian
one can then proceed to calculate energies, forces, and higher derivatives and
use them to, for example, compute the minimum energy conformation or
simulate the dynamical behavior via solving Eq. 11.
In the following we will focus on the classical Hamiltonians that are commonly used for simulations of biomolecular systems. They are referred to
as force elds and are known under names such as CHARMM [40], GROMACS [41], and AMBER [42]. They all have in common a relatively simple
mathematical structure to allow for a computationally efcient force evalu-
334
M. Dittrich et al.
ation. Furthermore, these force elds are empirical, i.e., they contain tting
parameters, that are either determined by comparison to experiments, via
quantum mechanical calculations, or both.
A typical force eld is described by the following Hamiltonian
HMM = Hbnd + Hang + Hdih + HvdW + Hele .
(14)
Here, the rst three terms represent the bonded contributions, namely the
bond
Hbnd =
bonds

a
2
1 bnd
ka xa x0a ,
2
(15)
and angle terms

angles
Hang =
1 ang
2
ka a 0a ,
2
a
(16)
that are both simple harmonic functions of the bond lengths xa and angles
a , respectively. The parameters x0a and 0a denote the corresponding equibnd/ang
librium values and the ka
are the interaction strengths.
The contributions of the dihedral angles a are periodic, typically of small
magnitude, and are given by
Hdih =
dihed

kdih
1 + cos (na 0 ) .
a
(17)
Again, 0 is the equilibrium value of the dihedral angle, kdih

a the interaction
strength, and n determines the periodicity. The last two terms in Eq. 14 are
the non-bonded contributions to the force eld. The short range part is represented by a Lennard-Jones potential

N

ab 12
ab 6
HvdW =
,
(18)
4ab
rab
rab
a =b
where rab is the distance between particles a and b, and the quantities ab and
ab correspond to the depth of the Lennard-Jones potential and to the interparticle distance where the potential is equal to zero, respectively.
Finally, the last term represents the Coulomb interaction between pairs of
atoms with partial charges qa and qb
Hele =
N

qa qb
a =b
rab
(19)
Because of the 1/r dependence, the Coulomb interaction is long-range and,

therefore, a straightforward calculation requires a computational effort of
PcrA Helicase
335
order O(N 2 ) in the number of particles N. Computationally, it is, therefore,

the most expensive part in evaluating Eq. 14 and several numerical approximations, for example the Particle Mesh Ewald method, have been devised to
reduce the computational cost of this term to O(N log N).
Using efcient parallelization schemes, such as that employed by the molecular dynamics program NAMD [43], it is presently possible to calculate
the time evolution of a biomolecular system using Newtons equation of motion combined with an expression for the force eld similar to the one given
in Eq. 14 for a million atoms over a time of tens of ns [44]. MD calculations were traditionally performed in the micro-canonical (NVE) ensemble; currently, most MD simulations are conducted in the canonical (NVT)
and grand-canonical (NPT) ensemble, i.e., under conditions of constant volume/pressure. The temperature of the system can be kept constant by coupling the system to an external heat bath at the desired temperature [45]; the
pressure can be controlled by using, for example, the Nose-Hoover Langevin
piston method [46, 47]. A rather comprehensive introduction into concepts
and algorithms underlying classical MD simulations can be found in [43].
Molecular Dynamics Simulations of PcrA
Having said this, it becomes apparent that the main difculty in applying
MD methods to the study of molecular motors is the vast gap between the
time-scales available to equilibrium MD simulations ( 100 ns) and the ones
relevant physiologically (s to ms). Since the translocation of PcrA by a single nucleotide takes about 20 ms, the accompanying protein motions, among
them, for example, the closing of the catalytic cleft between domains 1A and
2A or the motion of bound ssDNA bases, take place on a similar length timescale. Hence, investigating PcrA function with equilibrium MD simulations is
hampered by this gap in time-scale and different strategies are needed.
One possible approach to elucidate the coupling of ATP binding and hydrolysis to changes in PcrA conformation consists of the analysis of equilibrium
MD trajectories of PcrA in the ATP and ADP+Pi bound states. Figure 7 shows
the per residue RMSD averaged over several ns for both states, clearly illustrating that at least on a ns time scale, there is no signicant difference
between the two, rendering this approach futile.
The time-scale gap can be partially overcome by applying an external
biasing potential, enforcing a certain change in conformation, via, for example steered molecular dynamics simulations (SMD) [48, 49] or umbrella
sampling [50]. The application of these approaches to molecular motors,
however, suffers from several limitations. First, it requires a certain insight
into what the physiologically relevant motions are, in order to bias the simulations properly. Second, biasing can be difcult for all but the simplest
motions. Finally, enforcing ms time-scale protein motions on a ns computational time-scale can lead to artifacts or, in the worst case, destroy the protein.
336
M. Dittrich et al.
Fig. 7 Trajectory averaged per-residue RMSD in PcrA. Shown are the RMSD values based
on 3.5 ns of MD simulations for the ATP (a) and ADP/Pi (b) bound PcrA systems. The
color scale ranges from blue to red and indicates increasing RMSD values. The most mobile part of the structure is a exible loop connecting the 2A and 2B domains colored in
red
Nevertheless, despite all these drawbacks, signicant insight into the system
can be gained if the approach is applied properly and can be used, for example, to calculate potentials of mean force or force-extension curves.
In the case of PcrA, such a biasing method can be employed to investigate the mechanism behind the directionality of ssDNA motion in the 3 5
direction. As proposed in the inchworm translocation model [19] derived
from structural data, the 1A and 2A domains each have different afnities
for ssDNA, and both alternate their afnities between the substrate (with
ATP bound, s) and product (without ATP/ADP bound, p) state. The model,
however, does not provide any microscopic evidence to support such a mechanism. To test this idea further, SMD simulations have been used by attaching
A2 ) to ten phosphorous
ten harmonic springs (force constants of 2 kcal/mol
atoms of the ssDNA bound to PcrA [51]. The ssDNA was then pulled one half
nucleotide (nt) distance forward and backward across the protein-ssDNA interface. The direction of pulling in the s and p states was chosen according
to the anticipated domain motion in each state corresponding to separation
and closure of domains 1A and 2A, respectively. The required pulling forces
were monitored and are reproduced in Fig. 8. These results show that in the
s state, the average force needed to move nucleotides 1518 to the right (corresponding to forward motion of 2A to the left) is smaller than that needed to
move nucleotides 1821 to the left (corresponding to backward motion of 1A
to the right). Similarly, in the p state, the average force needed to move nucleotides 1518 to the left (corresponding to backward motion of 2A to the
right) is larger than the force to move nucleotides 1720 to the right (corresponding to forward motion of 1A to the left). This directly supports the idea
of alternating afnities between the protein domains and ssDNA in different
PcrA states and can explain the directionality of PcrA translocation.
PcrA Helicase
337
Fig. 8 Comparisons of SMD forces arising during ssDNA pulling simulations in PcrA.
Shown are the results with ATP bound (s state) (a) and without ATP bound (p state) (b).
The green/red curve represents the average force needed to move relevant nucleotides
in a direction opposite to the corresponding movement of domain 1A/2A (indicated by
green/red arrows). Numbers are used to label nucleotides on ssDNA. The thin curves were
measured directly from simulations, while the thick curves were smoothed over every ten
data points. The results show that in a, the average force needed to shift the relevant nucleotides, corresponding to the domain movement of 2A, is smaller than the average force
needed to shift the relevant nucleotides corresponding to the domain movement of 1A;
the opposite is true in b (adapted from [51])
In order to further investigate the microscopic mechanism underlying the

alternating afnities between the 1A and 2A domains and the bound ssDNA,
all-atom ns time-scale MD simulations were used to derive an effective potential, Ui (xi ), governing individual domain motion of PcrA. Here, is a state
index describing either s or p; i = 1,2 refers to domain 1A and 2A, respectively, with xi being the position of each domain along the bound ssDNA.
The effective potential Ui (xi ) is then used to formulate a coarse-grained
Langevin dynamics description of PcrA that will be described in the next section. This approach [51] allows one to describe the ATP hydrolysis coupled
uni-directional ssDNA translocation of PcrA on the physiologically relevant
ms time-scale.
Since the translocation of helicase along ssDNA arises through the sequential binding and unbinding of ssDNA units (nucleotides) to the PcrA domain
surfaces, one possible strategy for determining Ui is to calculate the binding free energies of individual ssDNA units with the protein, and then use an
appropriate combination of these energies to obtain Ui . The calculation of
the binding free energy from MD simulations is not straightforward, but can
be accomplished by a semi-empirical method [52]. Here, the binding energy
Eb is approximated as a weighted sum of electrostatic and vdW interactions
between the protein and each individual nucleotide as extracted from equilibrium MD simulations. In the present case, only relative but not absolute
energies are needed. Hence, the solvent (water and ion) contribution to the
energy can be neglected by assuming a homogeneous solvent-nucleotide interaction along the bound ssDNA region (consisting of poly-thymine in the
338
M. Dittrich et al.
simulation model) [51]. From the discrete set of nucleotide-protein interaction energies derived from MD, a continuous site-energy function Eb (x)
for each state (s and p) was constructed by polynomial interpolation [51].
Figure 9 shows the continuous site energies Eb (x) as well as the resulting potential energy functions Ui in both the s and p states.
Now we give an example of how to derive Ui from the binding energies
Eb (x) [51]. Starting from the s state, as domain 2A moves forward (to the left)
by a distance x (x 0.5, in unit of a single nt distance), nucleotides 15
to 18 move backward (to the right) by x, while the other nucleotides remain stationary. This scenario can be inferred from the available X-ray crystal
structures. Hence, the potential U2s , for example, can be expressed as
U2s
18

x =
Eb xj + x Eb xj
x [0, 0.5] .
(20)
j=15
The other Ui can be computed in a similar fashion. In the derivation, we assume a symmetric form of Ui with a barrier at the center at x = 0.5, where
Ai = Ui (0.5) is dened as the barrier height. According to Eq. 20, Ai in-
Fig. 9 Site energies of ssDNA nucleotides and individual domain potentials for PcrA.
Shown are the result for the substrate (a) and the product state (b). Filled dark diamonds
represent the relative binding free energies of nucleotide i at position xi , i.e., the weighted
sum of electrostatic and vdW energies between protein and individual nucleotides. The
separate electrostatic and vdW contributions are indicated through non-lled triangles
and non-lled pentagons, respectively. A smooth site energy function Eb (x) is drawn
through the lled dark diamonds using a third-order polynomial interpolation. The inset shows the potential Ui (x) as derived from Eb (x) that is experienced by domain 2A
(red solid curve) and 1A (green solid curve) as the domains move along ssDNA; the length
scale is in units of one nt distance (6.5
A) (adapted from [51])
PcrA Helicase
339
cludes a sum of site energy terms Eb (xj + 0.5) Eb (xj ). Following [51] one can
write this as
Ai = i + ,
(21)
where i are sums of the differences of binding energies Eb (xj ). is a single tunable parameter that accounts for the effect of additional friction due
to ssDNA-protein interactions and is assumed to be independent of i and .
The value of is adjusted so that the translocation speed of PcrA is consistent
with that measured from experiment, i.e., about one nt distance every 20 ms.
It is important to point out that the difference between amplitudes, |A1
A2 |, is independent of . The calculations show that A1s A2s 9 kcal/mol
in the s state and A1p A2p -12 kcal/mol in the p state [51]. This result, i.e.,
A1s > A2s and A1p < A2p , derived solely on the basis of ns time-scale MD simulations, conrms the idea of alternating afnities between the domains and
ssDNA in different states, rst proposed by Velankar et al. [19]: in the ATPbound (s) state domain 2A moves more easily than 1A while in the empty
binding site case (state p) the opposite is true. As a result, when ATP binds
and the domains 1A and 2A are being pulled together, only 1A moves. When
the domains separate again, domain 2A moves. This scenario of alternating
1A and 2A mobilities along DNA results in a unidirectional translocation as
one can readily realize by grasping with ones hands (representing domains
1A and 2A) a rope (representing ssDNA) and alternatingly making a tight and
a loose st with the left and right hand while pulling the rope with the tight
st (c.f. Fig. 6b in [19]). The derived potential energy functions Ui (Fig. 9)
will be utilized in a Langevin dynamics description presented below, demonstrating indeed uni-directional translocation along ssDNA.
5
Langevin Simulations of PcrA Function
From experimental studies it is known that PcrA translocates along ssDNA
uni-directionally at a speed of about 50 nt/s, presumably consuming one ATP
per single nucleotide step which, therefore, takes about 20 ms [53]. To describe PcrA function on this time-scale, a computational modeling approach
different from QM/MM or MD is required for two major reasons. First, due
to the explicit treatment of electronic and/or atomistic details, the time-scales
accessible to QM/MM and MD simulations do not reach the long time-scales
needed to model physiological PcrA function. Second, on larger length- and
longer time-scales, the essential degrees of freedom governing the dynamics of the system are likely quite different and, therefore, largely inaccessible
from simulations of smaller scale models. Therefore, in order to properly investigate the relevant mechanism on this functional level one needs to adopt
a different approach.
340
M. Dittrich et al.
One possible description of the PcrA ssDNA stepping motor replaces the
two translocating domains 1A and 2A by two featureless beads [51], which
move along ssDNA during each ATP hydrolysis cycle (Fig. 1). This simple
view is based on the two basic conformations of the PcrA-DNA complex
found in the available X-ray crystal structures [19].
A straightforward approach to characterize the bead movement along ssDNA is by means of a Langevin equation describing Brownian motion in
a potential. Let us rst suppose that the interaction between the bead and ssDNA is purely associative, i.e., the attachment of the bead to the DNA strand
does not interfere with its movement along ssDNA. Under these conditions,
the bead motion can be described by a 1-D Langevin equation,
mx = x + f(t) .
(22)
Here is the friction coefcient. The term f (t) represents white noise and
obeys the uctuationdissipation theorem [54]
f(t) = 0 ; f(t) f (t ) = 2 kB T(t t ) .
(23)
Since a particle the size of a protein in solution has a very low mass to viscosity ratio (the Reynolds number) one can work in the strong friction limit in
which the acceleration term mx in Eq. 22 vanishes.
Next, we take into account the interactions between the protein and ssDNA
which prevent the beads from sliding freely along ssDNA. The interactions
can be modeled by additional potentials in Eq. 22 which govern the movement of individual beads. These additional potentials are just the Ui in Eq. 20
that were derived from MD simulation and originate from the imbalance in
binding energy of individual ssDNA nucleotides to PcrA [51]. Since the two
beads, corresponding to domain 1A and 2A, move in a concerted fashion, it
may be advantageous to introduce an extra interaction term acting between
the two domains facilitating the movement of domains. This interaction potential, when non-vanishing, is modeled by a harmonic spring
1
(24)
V = k(x1 x2 l )2 .
2
Here, (s or p) is a state index varying independently from the in Ui .
A2 , a value emThe force constant of the spring, k, adopts a value of 1kB T/
pirically estimated from MD simulations [51]. l is the equilibrium length of
the spring, with lp > ls , since the two domains 1A and 2A are separated farther in the p state than in the s state. For the sake of simplicity, we use lp = l0
and ls = l0 /3 with l0 corresponding to the nucleotide separation in ssDNA
A).
( 6.5
Putting all of this together, the Langevin equation in the strong friction
limit reads
W (x1 , x2 )
+ f (t) ,
(25)
xi =
xi
PcrA Helicase
341
where W (x1 , x2 ) = U1 (x1 ) + U2 (x2 ) + V (x1 , x2 ), with xi , i = 1,2, being

the position of the beads tracking domains 1A and 2A along ssDNA. Equation 25 can be solved numerically assuming discrete time steps of size t,
1 W (x1 , x2 )
xi (t + t) = xi (t)
t + 2D tZ .
(26)
xi
Here, Z is a standard normal random variable (with mean 0 and variance 1)
and D is the diffusion coefcient, according to the uctuation-dissipation
theorem related to the friction coefcient through D = kB T/ . D is choA2 /s, typical for a 3 nm-radius protein in
sen to assume a value of 104
solution [55]. It is important to keep in mind, however, that D can vary signicantly in different physiological environments.
Note that Eq. 25 holds in a particular state , denoted by s or p when V
is absent. Transitions between states are either triggered once some geometrical criterion is satised, or are simulated through a Poisson process with
a specic rate constant. In the latter scenario one generates uniformly distributed random numbers Y (with Y [0, 1]) and transitions are accepted if
Y t (t
1), where is the rate constant for the transition and t is
the discrete time step in Eq. 26. The rate constants are estimated [51] based on
the rate limiting steps assumed in each scenario and the translocation speed
observed experimentally [53].
In the following, we will assume two limiting scenarios [51], the weak
and strong coupling scenario, expecting that the most realistic model will
lie somewhere between the two. We envision that the translocation of PcrA
along ssDNA comes about through an inchworm motion involving separate,
but coupled translocations of its 1A and 2A domains (described by beads)
that are governed by three factors: (1) There exist geometrical constraints that
prohibit the domains to pass each other as well as to separate too far. (2) Binding of ATP favors a narrower separation between domains 1A and 2A while
unbinding of ADP and Pi favors a wider separation between the domains.
(3) Depending on the state (s or p) of PcrA, the domains experience different
effective potentials such that in the ATP bound state (s), 2A can move easily
(low energy barrier A2s ) and 1A can hardly move (high energy barrier A1s )
and in the absence of ATP/ADP (p), 1A can move easily (low energy barrier
A1p ) and 2A can hardly move (high energy barrier A2p ). Below we investigate PcrA translocation in the two scenarios, the weak and strong coupling
scenario, and demonstrate how (1)(3) can endow PcrA with unidirectional
motion in the 3 5 direction.
Weak Coupling Scenario
Here, domain 2A and 1A move with vanishing mutual interaction potential,
i.e. V (x1 , x2 ) = 0. However, the two domains are still coupled by geometrical
constrains. We furthermore assume that the rate limiting steps are indi-
342
M. Dittrich et al.
vidual domain movements, i.e., thermally activated barrier-crossing events

governed by potentials Ui . The barrier-crossing time can be estimated by the
mean rst-passage time [55, 56], which in turn is mostly determined by Ai
A/20 ms
given in Eq. 21. In order to obtain a translocation speed of about 6.5
a value for the parameter of 2.5 kcal/mol was used. The corresponding results of Langevin dynamics simulations are shown in Fig. 10a. The left
panel depicts the stochastic trajectories describing the motion of 1A and 2A
along ssDNA over a period of 10 hydrolysis cycles. The right-side panel shows
Fig. 10 Langevin simulations of ssDNA translocation in PcrA. Depicted are the trajectories for the weak (a) and the strong coupling scenario (b). Left Shown are trajectories of
the two translocating domains, 1A (green) and 2A (red), moving along ssDNA in each scenario; the time is given in units of ATP hydrolysis cycles (one cycle lasts about 20 ms).
Right Illustrated are the individual potentials Ui experienced by domain 1A (green)
and domain 2A (red) moving along ssDNA in different states (p, s in the weak coupling
scenario or pp, ps, ss, and sp in the strong coupling scenario, dened in [51]) in each
scenario. In the weak coupling scenario, two domains are shown as being connected by
a rod, corresponding to the geometric constraint; in the strong coupling scenario (b), the
domains are shown as being connected by an elastic spring with variable equilibrium
lengths, corresponding to the non-vanishing interaction potential V (adapted from [51])
PcrA Helicase
343
schematically the sequence of congurations corresponding to the motion. In

the p state, the two domains separate rst; domain 1A experiences a low barrier (A1p ) and can move readily, while 2A experiences a high barrier (A2p ) and
is essentially stuck. When 1A has moved forward close to 2A, ATP binds
into the cleft between 1A and 2A and the system transits to the s state. In the
s state, the potentials Uis differ qualitatively from Uip , in that now 2A is ready
to move (low barrier A2s ) and 1A becomes stuck (high barrier A1s ). When
2A moves forward far enough, ADP and Pi quickly dissociate and the system
transits back to the p state. Overall, after one cycle, PcrA has shifted one nt
distance compared to the starting state on the ssDNA.
Strong Coupling Scenario
In this scenario, domain 1A and 2A move under the inuence of a nonvanishing interaction potential V (x1 , x2 ), modeled by a harmonic spring.
Hence, the barrier-crossing events of each domain happen under the combined potential of Ui and V . The states in the strong coupling scenario,
accordingly, can be dened through two labels, and . This leads to four
possible states, namely, ( ) = (ss), (sp), (ps), and (pp) as shown in Fig. 10b.
Since A1p < A2p as well as A1s > A2s still holds, the barrier experienced by
1A/2A is lower than that experienced by 2A/1A during ATP binding/ADP +
Pi dissociation. Therefore, the unidirectional translocation of PcrA is maintained, as demonstrated in Fig. 10b. The domain movement itself, however,
can happen fast in this scenario and we assume that the rate limiting steps are
waiting for ATP arrival and ATP hydrolysis; the rate constants of these events
A/20 ms.
can also be tted to yield a translocation speed of about 6.5
From our analysis, it is not clear which of the two limiting scenarios is the
physiologically relevant one for PcrA. Experimental evidence suggests that
PcrA may actually work in a mixed scenario, in which ATP binding serves
as a power-stroke [57, 58], facilitating the forward movement of domain 1A
and assisting PcrA conformational change so that 1A can not move backward, while ATP hydrolysis or ADP + Pi dissociation triggers the thermally
activated forward movement of domain 2A (Myong et al., 2006, personal
communication), assisting PcrA conformational change to prevent 2A from
moving backward.
No matter what the exact scenario is, one can recognize that the unidirectional translocation of PcrA stems from the fact that in the state without
ATP bound, domain 1A experiences a lower energy barrier than 2A; while in
the ATP-bound state, domain 2A experiences a lower energy barrier than 1A,
that is A1p < A2p and A1s > A2s . One can imagine that the opposite relationship A1p > A2p and A1s < A2s will lead to helicase translocation in the reversed
direction.
Since the above Langevin dynamics simulations of PcrA translocation were
conducted with the potentials Ui derived from atomic-scale simulations, and
344
M. Dittrich et al.
recognizing that it is the difference in barrier height |A1 A2 | that directs

the unidirectional movement, it is interesting to evaluate the contributions of
individual residues to this difference term. This reveals that two key amino
acids, R260 and K385, contribute prominently to A1s A2s and A2p A1p , respectively [51], and are therefore proposed to be most important in affecting
the unidirectional translocation of PcrA.
6
Summary
Combining QM/MM calculation, nanosecond MD simulation, and millisecond stochastic modeling, the ATP hydrolysis coupled ssDNA translocation of
a monomeric PcrA helicase has been investigated.
Figure 11 summarizes the combined results and depicts a single PcrA
translocation cycle involving ve steps. In conguration 1, domains 1A and
2A are separated without ATP or ADP bound. The arrangement of the bases
of the ssDNA are shown in the insert to the left, revealing a single base intercalated between side groups of Y257 and F64. Initially, domain 1A can
move more easily along ssDNA than domain 2A. At this point ATP arrives at
the binding site and initiates binding. The binding of ATP causes an attractive force between the domains which consequently start approaching each
other through motion of domain 1A, leading to conguration 3. The close approach of domains 1A and 2A pushes Y257 and F64 together and, thereby,
squeezes the ssDNA base out, it moving from its former position to a new
position as indicated by the arrows in the inserted gures. The close approach
of 2A and 1A brings R610 and R287 into contact with the -phosphate of
ATP; the movement of Y257 is transmitted to Q254. QM/MM calculations
have shown that the three mentioned side groups, R610, R287, and Q254 control the hydrolysis of ATP in PcrA [36]. In conguration 3 the juxtaposition of
the three side groups is optimal for hydrolysis of ATP to ADP + Pi , leading to
conguration 4. At this point the mobility of domains 2A and 1A has reversed,
2A now being able to move more easily. Thermal uctuations lead to the separation of 2A from 1A giving rise to conguration 5, and permitting release of
ADP and Pi , reaching conguration 1 . PcrA has advanced by one step and is
ready for the next cycle.
In summary, we have presented a comprehensive study of PcrA helicase
function covering length scales ranging from the electronic domain using
QM/MM simulations to the molecular level via MD and long time stochastic simulations. The time scales covered span several orders of magnitude
from fast femtosecond time scale bond-breaking events to physiological protein motions on the millisecond time scale. The combination of methods
employed in our computational modeling revealed a detailed picture of PcrA
function, for example, a proton relay mechanism responsible for efcient ATP
PcrA Helicase
345
Fig. 11 Five-step PcrA translocation cycle. The gure shows schematically a translocation cycle for a mixed (see text) scenario involving ve congurations 1, 2, 3, 4, 5 as well
as conguration 1 that is equivalent to conguration 1, except that the system has advanced by one base. The mixed scenario involves both a loaded spring (non-zero potential
V (x1 , x2 ), conguration 2 3) and a step with a random thermal motion (vanishing
potential V (x1 , x2 ), conguration 4 5). In congurations 1, 2, 1 domain 1A (green)
moves more readily than domain 2A (red), while it is the opposite for congurations 3,
4, 5. The insert gures show how the domain and ssDNA base motions are coupled to
the chemistry at the ATP binding site: upon the approach of domains 2A and 1A, R287
and R610 move close to the -phosphate of ATP; Q254 is linked closely to Y257 that
forms a key binding pocket for an ssDNA base, which is squeezed out when the domains approach each other in binding ATP; Q254 was identied as a key participant in
ATP hydrolysis along with the mentioned arginines [36]. The suggested mechanism involves the binding of ATP that pulls domain 1A towards domain 2A; insertion of R287 and
R610 into an optimal (for hydrolysis) position in the ATP binding pocket along with Q254
linked to a key ssDNA interaction site; rapid hydrolysis of ATP that initiates separation of
domains 2A and 1A through movement of 2A alone (adapted from [51])
hydrolysis in the catalytic site and the means by which uni-directional motion along ssDNA is achieved, namely by a ratchet mechanism. Employing
a similar strategy to investigate other molecular motors can be expected to
reveal more about the mechanism by which these remarkable nano-machines
perform their physiological function.
Acknowledgements The molecular images in this paper were created with the molecular
graphics program VMD [59]. This work is supported by grants from the National Institutes of Health PHS-5-P41-RR05969 and the National Science Foundation MCB02-34938.
The authors gladly acknowledge supercomputer time provided by Pittsburgh Supercomputer Center and the National Center for Supercomputing Applications via National
Resources Allocation Committee grant MCA93S028.
346
M. Dittrich et al.
References
1. Li G, Cui Q (2004) J Phys Chem B 108:33423357
2. Minehard TJ, Marzari N, Cooke R, Pate E, Kollman PA, Car R (2002) Biophys J
82:660675
3. Zheng W, Brooks BR (2005) Biophys J 89:167178
4. Fischer S, Windshgel B, Horak D, Holmes KC, Smith JC (2005) Proc Natl Acad Sci
USA 102:68736878
5. Kawakubo T, Okada O, Minami T (2005) Biophys Chem, 115:7785
6. Wriggers W, Schulten K (1999) Proteins: Struct Func Gen 35:262273
7. Oster G, Wang H (2000) Biochim Biophys Acta 1458:482510
8. Bckmann RA, Grubmller H (2002) Nature Struct Biol 9:198202
9. Ma J, Flynn TC, Cui Q, Leslie AGW, Walker JE, Karplus M (2002) Structure 10:921
931
10. Dittrich M, Hayashi S, Schulten K (2003) Biophys J 85:22532266
11. Strajbl M, Shurki A, Warshel A (2003) Proc Natl Acad Sci USA 100:1483414839
12. Dittrich M, Hayashi S, Schulten K (2004) Biophys J 87:29542967
13. Cox K, Watson T, Soultanas P, Hirst J (2003) Proteins: Struct Func Gen 52:254262
14. Betterton M, Julicher F (2003) Phys Rev Lett 91:258103
15. Betterton M, Julicher F (2005) Phys Rev E 71:011904
16. Bhattacharjee S, Seno F (2003) J Phys A: Math Gen 36:L181-L187
17. Bhattacharjee S (2004) Europhys Lett 65:574580
18. Subramanya H, Bird L, Brannigan J, Wigley D (1996) Nature 384:379383
19. Velankar S, Soultanas P, Dillingham M, Subramanya H, Wigley D (1999) Cell 97:7584
20. Wong I, Lohman TM (1992) Science 256:250256
21. Stasiak A, Tsaneva IR, West SC, Benson CJ, Yu X, Egelman EH (1994) Proc Natl Acad
Sci USA 91:76187622
22. Warshel A (1976) Nature 260:679683
23. Singh UC, Kollman PA (1986) J Comp Chem 7:718730
24. Field MJ, Bash PA, Karplus M (1990) J Comp Chem 11:700733
25. Stanley RJ, Boxer SG (1995) J Phys Chem 99:859863
26. Maseras F, Morokuma K (1995) J Comp Chem 16:11701179
27. Bakowies D, Thiel W (1996) J Phys Chem 100:1058010594
28. Lyne PD, Hodoscek M, Karplus M (1999) J Phys Chem A 103:34623471
29. Lyne D, Walsh O (2001) Computer Simulation of Biochemical Reactions with
QM/MM Methods. In: Becker O, Mac Kerell A Jr, Roux B, Watanabe M (eds) Computational Biochemistry and Biophysics. Marcel Dekker Inc., New York
30. Warshel A, Levitt M (1976) J Mol Biol 103:227249
31. Thry V, Rinaldi D, Rivail J-L, Maigret B, Ferenczy GG (1994) J Comp Chem 15:269
282
32. Bayly C, Cieplak P, Cornell W, Kollman P (1993) J Phys Chem 100:1026910280
33. Hayashi S, Ohmine I (2000) J Phys Chem B 104:1067810691
34. Warshel A, Weiss RM (1980) J Am Chem Soc 102:6218
35. Warshel A (2003) Ann Rev Biophys Biomol Struct 32:425443
36. Dittrich M, Schulten K (2006) Structure 14:13451353
37. Scheidig AJ, Burmester C, Goody RS (1999) Structure 7:13111324
38. Ahmad Z, Senior AE (2004) J Biol Chem 44:4605746064
39. Dittrich M, Schulten K (2005) J Bioener Biomemb 37:441444
40. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan S, Karplus M (1983)
J Comp Chem 4:187217
PcrA Helicase
347
41. Lindahl E, Hess B, van der Spoel D (2001) J Mol Mod 7:306317
42. Pearlman DA, Case DA, Caldwell JW, Ross WS, Cheatham T, Debolt S, Ferguson D,
Seibel G, Kollman P (1995) Comput Phys Commun 91:141
43. Phillips JC, Braun R, Wang W, Gumbart J, Tajkhorshid E, Villa E, Chipot C, Skeel RD,
Kale L, Schulten K (2005) J Comp Chem 26:17811802
44. Freddolino PL, Arkhipov AS, Larson SB, Mc Pherson A, Schulten K (2006) Structure
14:437449
45. Berendsen HJC, Postma JPM, van Gunsteren WF, Di Nola A, Haak JR (1984) J Chem
Phys 81:36843690
46. Martyna GJ, Tobias DJ, Klein ML (1994) J Chem Phys 101:41774189
47. Feller SE, Zhang YH, Pastor RW, Brooks BR (1995) J Chem Phys 103:46134621
48. Izrailev S, Stepaniants S, Isralewitz B, Kosztin D, Lu H, Molnar F, Wriggers W, Schulten K (1998) Steered Molecular Dynamics. In: Deuhard P, Hermans J, Leimkuhler B,
Mark AE, Reich S, Skeel RD (eds) Computational Molecular Dynamics: Challenges,
Methods, Ideas. Vol 4. Springer, Berlin Heidelberg New York
49. Isralewitz B, Gao M, Schulten K (2001) Curr Op Struct Biol 11:224230
50. Torrie, Valleau (1977) J Comp Phys 23:187199
51. Yu J, Ha T, Schulten K (2006) Biophys J 91:20972114
52.
Aqvist J (1996) J Comp Chem 17:15871597
53. Dillingham M, Wigley D, Webb M (2000) Biochemistry 39:205212
54. Kubo R (1966) Rep Progr Phys 29:255284
55. Howard J (2001) Mechanics of Motor Proteins and the Cytoskeleton. Sinauer Associates, Inc., Sunderland, MA
56. Szabo A, Schulten K, Schulten Z (1980) J Chem Phys 72:43504357
57. Nishizaka T, Oiwa K, Noji H, Kimura S, Muneyuki E, Yoshida M, Kinosita K Jr (2004)
Nat Struct Mol Biol 11:142148
58. Shimabukuro K, Yasuda R, Muneyuki E, Hara KY, Kinosita K Jr, Yoshida M (2003)
Proc Natl Acad Sci USA 100:1473114736
59. Humphrey W, Dalke A, Schulten K (1996) J Mol Graphics 14:3338
Author Index Volumes 251271

Author Index Vols. 2650 see Vol. 50
The volume numbers are printed in italics

Ajayaghosh A, George SJ, Schenning APHJ (2005) Hydrogen-Bonded Assemblies of Dyes
and Extended -Conjugated Systems. 258: 83118
Albert M, Fensterbank L, Lacte E, Malacria M (2006) Tandem Radical Reactions. 264: 162
Alberto R (2005) New Organometallic Technetium Complexes for Radiopharmaceutical
Imaging. 252: 144
Alegret S, see Pividori MI (2005) 260: 136
Amabilino DB, Veciana J (2006) Supramolecular Chiral Functional Materials. 265: 253302
Anderson CJ, see Li WP (2005) 252: 179192
Anslyn EV, see Houk RJT (2005) 255: 199229
Appukkuttan P, Van der Eycken E (2006) Microwave-Assisted Natural Product Chemistry.
266: 147
Araki K, Yoshikawa I (2005) Nucleobase-Containing Gelators. 256: 133165
Armitage BA (2005) Cyanine DyeDNA Interactions: Intercalation, Groove Binding and
Aggregation. 253: 5576
Arya DP (2005) AminoglycosideNucleic Acid Interactions: The Case for Neomycin. 253:
149178
Bailly C, see Dias N (2005) 253: 89108
Balaban TS, Tamiaki H, Holzwarth AR (2005) Chlorins Programmed for Self-Assembly. 258:
138
Balzani V, Credi A, Ferrer B, Silvi S, Venturi M (2005) Articial Molecular Motors and
Machines: Design Principles and Prototype Systems. 262: 127
Barbieri CM, see Pilch DS (2005) 253: 179204
Barchuk A, see Daasbjerg K (2006) 263: 3970
Bayly SR, see Beer PD (2005) 255: 125162
Beer PD, Bayly SR (2005) Anion Sensing by Metal-Based Receptors. 255: 125162
Bertini L, Bruschi M, de Gioia L, Fantucci P, Greco C, Zampella G (2007) Quantum Chemical
Investigations of Reaction Paths of Metalloenzymes and Biomimetic Models The
Hydrogenase Example. 268: 146
Bier FF, see Heise C (2005) 261: 125
Blum LJ, see Marquette CA (2005) 261: 113129
Boiteau L, see Pascal R (2005) 259: 69122
Bolhuis PG, see Dellago C (2007) 268: 291317
Borovkov VV, Inoue Y (2006) Supramolecular Chirogenesis in HostGuest Systems Containing Porphyrinoids. 265: 89146
350
Boschi A, Duatti A, Uccelli L (2005) Development of Technetium-99m and Rhenium-188 Radiopharmaceuticals Containing a Terminal MetalNitrido Multiple Bond for Diagnosis
and Therapy. 252: 85115
Braga D, DAddario D, Giaffreda SL, Maini L, Polito M, Grepioni F (2005) Intra-Solid and
Inter-Solid Reactions of Molecular Crystals: a Green Route to Crystal Engineering. 254:
7194
Brebion F, see Crich D (2006) 263: 138
Brizard A, Oda R, Huc I (2005) Chirality Effects in Self-assembled Fibrillar Networks. 256:
167218
Bruce IJ, see del Campo A (2005) 260: 77111
Bruschi M, see Bertini L (2007) 268: 146
del Campo A, Bruce IJ (2005) Substrate Patterning and Activation Strategies for DNA Chip
Fabrication. 260: 77111
Carney CK, Harry SR, Sewell SL, Wright DW (2007) Detoxication Biominerals. 270: 155185
Chaires JB (2005) Structural Selectivity of Drug-Nucleic Acid Interactions Probed by Competition Dialysis. 253: 3353
Chiorboli C, Indelli MT, Scandola F (2005) Photoinduced Electron/Energy Transfer Across
Molecular Bridges in Binuclear Metal Complexes. 257: 63102
Clfen H (2007) Bio-inspired Mineralization Using Hydrophilic Polymers. 271: 177
Collin J-P, Heitz V, Sauvage J-P (2005) Transition-Metal-Complexed Catenanes and Rotaxanes in Motion: Towards Molecular Machines. 262: 2962
Collyer SD, see Davis F (2005) 255: 97124
Commeyras A, see Pascal R (2005) 259: 69122
Coquerel G (2007) Preferential Crystallization. 269
Correia JDG, see Santos I (2005) 252: 4584
Costanzo G, see Saladino R (2005) 259: 2968
Credi A, see Balzani V (2005) 262: 127
Crestini C, see Saladino R (2005) 259: 2968
Crich D, Brebion F, Suk D-H (2006) Generation of Alkene Radical Cations by Heterolysis
of -Substituted Radicals: Mechanism, Stereochemistry, and Applications in Synthesis.
263: 138
Cuerva JM, Justicia J, Oller-Lpez JL, Oltra JE (2006) Cp2 TiCl in Natural Product Synthesis.
264: 6392
Daasbjerg K, Svith H, Grimme S, Gerenkamp M, Mck-Lichtenfeld C, Gansuer A, Barchuk
A (2006) The Mechanism of Epoxide Opening through Electron Transfer: Experiment
and Theory in Concert. 263: 3970
DAddario D, see Braga D (2005) 254: 7194
Danishefsky SJ, see Warren JD (2007) 267
Darmency V, Renaud P (2006) Tin-Free Radical Reactions Mediated by Organoboron Compounds. 263: 71106
Davis F, Collyer SD, Higson SPJ (2005) The Construction and Operation of Anion Sensors:
Current Status and Future Perspectives. 255: 97124
Deamer DW, Dworkin JP (2005) Chemistry and Physics of Primitive Membranes. 259: 127
Dellago C, Bolhuis PG (2007) Transition Path Sampling Simulations of Biological Systems.
268: 291317
Deng J-Y, see Zhang X-E (2005) 261: 169190
Dervan PB, Poulin-Kerstien AT, Fechter EJ, Edelson BS (2005) Regulation of Gene Expression
by Synthetic DNA-Binding Ligands. 253: 131
351
Dias N, Vezin H, Lansiaux A, Bailly C (2005) Topoisomerase Inhibitors of Marine Origin

and Their Potential Use as Anticancer Agents. 253: 89108
DiMauro E, see Saladino R (2005) 259: 2968
Dittrich M, Yu J, Schulten K (2007) PcrA Helicase, a Molecular Motor Studied from the
Electronic to the Functional Level. 268: 319347
Dobrawa R, see You C-C (2005) 258: 3982
Du Q, Larsson O, Swerdlow H, Liang Z (2005) DNA Immobilization: Silanized Nucleic Acids
and Nanoprinting. 261: 4561
Duatti A, see Boschi A (2005) 252: 85115
Dworkin JP, see Deamer DW (2005) 259: 127
Edelson BS, see Dervan PB (2005) 253: 131
Edwards DS, see Liu S (2005) 252: 193216
Ernst K-H (2006) Supramolecular Surface Chirality. 265: 209252
Ersmark K, see Wannberg J (2006) 266: 167197
Escud C, Sun J-S (2005) DNA Major Groove Binders: Triple Helix-Forming Oligonucleotides, Triple Helix-Specic DNA Ligands and Cleaving Agents. 253: 109148
Van der Eycken E, see Appukkuttan P (2006) 266: 147
inic M (2005) Systematic Design of Amide- and Urea-Type Gelators with
Fages F, Vgtle F, Z
Tailored Properties. 256: 77131
Fages F, see inic M (2005) 256: 3976
Faigl F, Schindler J, Fogassy E (2007) Advantages of Structural Similarities of the Reactants
in Optical Resolution Processes. 269
Fantucci P, see Bertini L (2007) 268: 146
Fechter EJ, see Dervan PB (2005) 253: 131
Fensterbank L, see Albert M (2006) 264: 162
Fernndez JM, see Moonen NNP (2005) 262: 99132
Fernando C, see Szathmry E (2005) 259: 167211
Ferrer B, see Balzani V (2005) 262: 127
De Feyter S, De Schryver F (2005) Two-Dimensional Dye Assemblies on Surfaces Studied
by Scanning Tunneling Microscopy. 258: 205255
Flood AH, see Moonen NNP (2005) 262: 99132
Fogassy E, see Faigl F (2007) 269
Fricke M, Volkmer D (2007) Crystallization of Calcium Carbonate Beneath Insoluble Monolayers: Suitable Models of MineralMatrix Interactions in Biomineralization? 270: 141
Fujimoto D, see Tamura R (2007) 269
Fujiwara S-i, Kambe N (2005) Thio-, Seleno-, and Telluro-Carboxylic Acid Esters. 251: 87
140
Gansuer A, see Daasbjerg K (2006) 263: 3970
Garcia-Garibay MA, see Karlen SD (2005) 262: 179227
Gelinck GH, see Grozema FC (2005) 257: 135164
Geng X, see Warren JD (2007) 267
George SJ, see Ajayaghosh A (2005) 258: 83118
Gerenkamp M, see Daasbjerg K (2006) 263: 3970
Giaffreda SL, see Braga D (2005) 254: 7194
de Gioia L, see Bertini L (2007) 268: 146
Greco C, see Bertini L (2007) 268: 146
Grepioni F, see Braga D (2005) 254: 7194
352
Grimme S, see Daasbjerg K (2006) 263: 3970

Grozema FC, Siebbeles LDA, Gelinck GH, Warman JM (2005) The Opto-Electronic Properties of Isolated Phenylenevinylene Molecular Wires. 257: 135164
Guiseppi-Elie A, Lingerfelt L (2005) Impedimetric Detection of DNA Hybridization: Towards
Near-Patient DNA Diagnostics. 260: 161186
Di Giusto DA, King GC (2005) Special-Purpose Modications and Immobilized Functional
Nucleic Acids for Biomolecular Interactions. 261: 131168
Haase C, Seitz O (2007) Chemical Synthesis of Glycopeptides. 267
Hansen SG, Skrydstrup T (2006) Modication of Amino Acids, Peptides, and Carbohydrates
through Radical Chemistry. 264: 135162
Harry SR, see Carney CK (2007) 270: 155185
Heise C, Bier FF (2005) Immobilization of DNA on Microarrays. 261: 125
Heitz V, see Collin J-P (2005) 262: 2962
Herrmann C, Reiher M (2007) First-Principles Approach to Vibrational Spectroscopy of
Biomolecules. 268: 85132
Higson SPJ, see Davis F (2005) 255: 97124
Hirayama N, see Sakai K (2007) 269
Hirst AR, Smith DK (2005) Dendritic Gelators. 256: 237273
Holzwarth AR, see Balaban TS (2005) 258: 138
Houk RJT, Tobey SL, Anslyn EV (2005) Abiotic Guanidinium Receptors for Anion Molecular
Recognition and Sensing. 255: 199229
Huc I, see Brizard A (2005) 256: 167218
Ihmels H, Otto D (2005) Intercalation of Organic Dye Molecules into Double-Stranded DNA
General Principles and Recent Developments. 258: 161204
Imai H (2007) Self-Organized Formation of Hierarchical Structures. 270: 4372
Indelli MT, see Chiorboli C (2005) 257: 63102
Inoue Y, see Borovkov VV (2006) 265: 89146
Ishii A, Nakayama J (2005) Carbodithioic Acid Esters. 251: 181225
Ishii A, Nakayama J (2005) Carboselenothioic and Carbodiselenoic Acid Derivatives and
Related Compounds. 251: 227246
Ishi-i T, Shinkai S (2005) Dye-Based Organogels: Stimuli-Responsive Soft Materials Based
on One-Dimensional Self-Assembling Aromatic Dyes. 258: 119160
James DK, Tour JM (2005) Molecular Wires. 257: 3362
Jones W, see Trask AV (2005) 254: 4170
Justicia J, see Cuerva JM (2006) 264: 6392
Kambe N, see Fujiwara S-i (2005) 251: 87140
Kano N, Kawashima T (2005) Dithiocarboxylic Acid Salts of Group 117 Elements (Except
for Carbon). 251: 141180
Kappe CO, see Kremsner JM (2006) 266: 233278
Kaptein B, see Kellogg RM (2007) 269
Karlen SD, Garcia-Garibay MA (2005) Amphidynamic Crystals: Structural Blueprints for
Molecular Machines. 262: 179227
Kato S, Niyomura O (2005) Group 117 Element (Except Carbon) Derivatives of Thio-,
Seleno- and Telluro-Carboxylic Acids. 251: 1985
Kato S, see Niyomura O (2005) 251: 112
353
Kato T, Mizoshita N, Moriyama M, Kitamura T (2005) Gelation of Liquid Crystals with

Self-Assembled Fibers. 256: 219236
Kaul M, see Pilch DS (2005) 253: 179204
Kaupp G (2005) Organic Solid-State Reactions with 100% Yield. 254: 95183
Kawasaki T, see Okahata Y (2005) 260: 5775
Kawashima T, see Kano N (2005) 251: 141180
Kay ER, Leigh DA (2005) Hydrogen Bond-Assembled Synthetic Molecular Motors and
Machines. 262: 133177
Kellogg RM, Kaptein B, Vries TR (2007) Dutch Resolution of Racemates and the Roles of
Solid Solution Formation and Nucleation Inhibition. 269
King GC, see Di Giusto DA (2005) 261: 131168
Kirchner B, see Thar J (2007) 268: 133171
Kitamura T, see Kato T (2005) 256: 219236
Kniep R, Simon P (2007) Fluorapatite-Gelatine-Nanocomposites: Self-Organized Morphogenesis, Real Structure and Relations to Natural Hard Materials. 270: 73125
Komatsu K (2005) The Mechanochemical Solid-State Reaction of Fullerenes. 254: 185206
Kremsner JM, Stadler A, Kappe CO (2006) The Scale-Up of Microwave-Assisted Organic
Synthesis. 266: 233278
Kriegisch V, Lambert C (2005) Self-Assembled Monolayers of Chromophores on Gold Surfaces. 258: 257313
Lacte E, see Albert M (2006) 264: 162
Lahav M, see Weissbuch I (2005) 259: 123165
Lambert C, see Kriegisch V (2005) 258: 257313
Lansiaux A, see Dias N (2005) 253: 89108
Larhed M, see Nilsson P (2006) 266: 103144
Larhed M, see Wannberg J (2006) 266: 167197
Larsson O, see Du Q (2005) 261: 4561
Leigh DA, Prez EM (2006) Dynamic Chirality: Molecular Shuttles and Motors. 265: 185208
Leigh DA, see Kay ER (2005) 262: 133177
Leiserowitz L, see Weissbuch I (2005) 259: 123165
Lhotk P (2005) Anion Receptors Based on Calixarenes. 255: 6595
Li WP, Meyer LA, Anderson CJ (2005) Radiopharmaceuticals for Positron Emission Tomography Imaging of Somatostatin Receptor Positive Tumors. 252: 179192
Liang Z, see Du Q (2005) 261: 4561
Lingerfelt L, see Guiseppi-Elie A (2005) 260: 161186
Liu S (2005) 6-Hydrazinonicotinamide Derivatives as Bifunctional Coupling Agents for
99m Tc-Labeling of Small Biomolecules. 252: 117153
Liu S, Robinson SP, Edwards DS (2005) Radiolabeled Integrin v 3 Antagonists as Radiopharmaceuticals for Tumor Radiotherapy. 252: 193216
Liu XY (2005) Gelation with Small Molecules: from Formation Mechanism to Nanostructure
Architecture. 256: 137
Luderer F, Walschus U (2005) Immobilization of Oligonucleotides for Biochemical Sensing
by Self-Assembled Monolayers: Thiol-Organic Bonding on Gold and Silanization on
Silica Surfaces. 260: 3756
Maeda K, Yashima E (2006) Dynamic Helical Structures: Detection and Amplication of
Chirality. 265: 4788
Magnera TF, Michl J (2005) Altitudinal Surface-Mounted Molecular Rotors. 262: 6397
Maini L, see Braga D (2005) 254: 7194
354
Malacria M, see Albert M (2006) 264: 162

Marquette CA, Blum LJ (2005) Beads Arraying and Beads Used in DNA Chips. 261: 113129
Mascini M, see Palchetti I (2005) 261: 2743
Matsumoto A (2005) Reactions of 1,3-Diene Compounds in the Crystalline State. 254: 263
305
McGhee AM, Procter DJ (2006) Radical Chemistry on Solid Support. 264: 93134
Meyer B, Mller H (2007) Conformation of Glycopeptides and Glycoproteins. 267
Meyer LA, see Li WP (2005) 252: 179192
Michl J, see Magnera TF (2005) 262: 6397
Milea JS, see Smith CL (2005) 261: 6390
Mizoshita N, see Kato T (2005) 256: 219236
Mller H, see Meyer B (2007) 267
Moonen NNP, Flood AH, Fernndez JM, Stoddart JF (2005) Towards a Rational Design of
Molecular Switches and Sensors from their Basic Building Blocks. 262: 99132
Moriyama M, see Kato T (2005) 256: 219236
Murai T (2005) Thio-, Seleno-, Telluro-Amides. 251: 247272
Murakami H (2007) From Racemates to Single Enantiomers Chiral Synthetic Drugs over
the last 20 Years. 269
Mutule I, see Suna E (2006) 266: 49101
Naka K (2007) Delayed Action of Synthetic Polymers for Controlled Mineralization of
Calcium Carbonate. 271: 119154
Nakayama J, see Ishii A (2005) 251: 181225
Nakayama J, see Ishii A (2005) 251: 227246
Neese F, see Sinnecker S (2007) 268: 4783
Nguyen GH, see Smith CL (2005) 261: 6390
Nicolau DV, Sawant PD (2005) Scanning Probe Microscopy Studies of Surface-Immobilised
DNA/Oligonucleotide Molecules. 260: 113160
Nilsson P, Olofsson K, Larhed M (2006) Microwave-Assisted and Metal-Catalyzed Coupling
Reactions. 266: 103144
Niyomura O, Kato S (2005) Chalcogenocarboxylic Acids. 251: 112
Niyomura O, see Kato S (2005) 251: 1985
Nohira H, see Sakai K (2007) 269
Oda R, see Brizard A (2005) 256: 167218
Okahata Y, Kawasaki T (2005) Preparation and Electron Conductivity of DNA-Aligned Cast
and LB Films from DNA-Lipid Complexes. 260: 5775
Okamura T, see Ueyama N (2007) 271: 155193
Oller-Lpez JL, see Cuerva JM (2006) 264: 6392
Olofsson K, see Nilsson P (2006) 266: 103144
Oltra JE, see Cuerva JM (2006) 264: 6392
Onoda A, see Ueyama N (2007) 271: 155193
Otto D, see Ihmels H (2005) 258: 161204
Palchetti I, Mascini M (2005) Electrochemical Adsorption Technique for Immobilization of
Single-Stranded Oligonucleotides onto Carbon Screen-Printed Electrodes. 261: 2743
Pascal R, Boiteau L, Commeyras A (2005) From the Prebiotic Synthesis of -Amino Acids
Towards a Primitive Translation Apparatus for the Synthesis of Peptides. 259: 69122
Paulo A, see Santos I (2005) 252: 4584
Prez EM, see Leigh DA (2006) 265: 185208
355
Pilch DS, Kaul M, Barbieri CM (2005) Ribosomal RNA Recognition by Aminoglycoside

Antibiotics. 253: 179204
Pividori MI, Alegret S (2005) DNA Adsorption on Carbonaceous Materials. 260: 136
Piwnica-Worms D, see Sharma V (2005) 252: 155178
Polito M, see Braga D (2005) 254: 7194
Poulin-Kerstien AT, see Dervan PB (2005) 253: 131
Procter DJ, see McGhee AM (2006) 264: 93134
Quiclet-Sire B, Zard SZ (2006) The Degenerative Radical Transfer of Xanthates and Related
Derivatives: An Unusually Powerful Tool for the Creation of CarbonCarbon Bonds. 264:
201236
Ratner MA, see Weiss EA (2005) 257: 103133
Raymond KN, see Seeber G (2006) 265: 147184
Rebek Jr J, see Scarso A (2006) 265: 146
Reckien W, see Thar J (2007) 268: 133171
Reiher M, see Herrmann C (2007) 268: 85132
Renaud P, see Darmency V (2006) 263: 71106
Robinson SP, see Liu S (2005) 252: 193216
Saha-Mller CR, see You C-C (2005) 258: 3982
Sakai K, Sakurai R, Hirayama N (2007) Molecular Mechanisms of Dielectrically Controlled
Resolution (DCR). 269
Sakai K, Sakurai R, Nohira H (2007) New Resolution Technologies Controlled by Chiral
Discrimination Mechanisms. 269
Sakamoto M (2005) Photochemical Aspects of Thiocarbonyl Compounds in the Solid-State.
254: 207232
Sakurai R, see Sakai K (2007) 269
Sakurai R, see Sakai K (2007) 269
Saladino R, Crestini C, Costanzo G, DiMauro E (2005) On the Prebiotic Synthesis of Nucleobases, Nucleotides, Oligonucleotides, Pre-RNA and Pre-DNA Molecules. 259: 2968
Santos I, Paulo A, Correia JDG (2005) Rhenium and Technetium Complexes Anchored by
Phosphines and Scorpionates for Radiopharmaceutical Applications. 252: 4584
Santos M, see Szathmry E (2005) 259: 167211
Sato K (2007) Inorganic-Organic Interfacial Interactions in Hydroxyapatite Mineralization
Processes. 270: 127153
Sauvage J-P, see Collin J-P (2005) 262: 2962
Sawant PD, see Nicolau DV (2005) 260: 113160
Scandola F, see Chiorboli C (2005) 257: 63102
Scarso A, Rebek Jr J (2006) Chiral Spaces in Supramolecular Assemblies. 265: 146
Scheffer JR, Xia W (2005) Asymmetric Induction in Organic Photochemistry via the SolidState Ionic Chiral Auxiliary Approach. 254: 233262
Schenning APHJ, see Ajayaghosh A (2005) 258: 83118
Schmidtchen FP (2005) Articial Host Molecules for the Sensing of Anions. 255: 129 Author
Index Volumes 251255
Schindler J, see Faigl F (2007) 269
Schoof S, see Wolter F (2007) 267
De Schryver F, see De Feyter S (2005) 258: 205255
Schulten K, see Dittrich M (2007) 268: 319347
356
Seeber G, Tiedemann BEF, Raymond KN (2006) Supramolecular Chirality in Coordination

Chemistry. 265: 147184
Seitz O, see Haase C (2007) 267
Senn HM, Thiel W (2007) QM/MM Methods for Biological Systems. 268: 173289
Sewell SL, see Carney CK (2007) 270: 155185
Sharma V, Piwnica-Worms D (2005) Monitoring Multidrug Resistance P-Glycoprotein Drug
Transport Activity with Single-Photon-Emission Computed Tomography and Positron
Emission Tomography Radiopharmaceuticals. 252: 155178
Shinkai S, see Ishi-i T (2005) 258: 119160
Sibi MP, see Zimmerman J (2006) 263: 107162
Siebbeles LDA, see Grozema FC (2005) 257: 135164
Silvi S, see Balzani V (2005) 262: 127
Simon P, see Kniep R (2007) 270: 73125
Sinnecker S, Neese F (2007) Theoretical Bioinorganic Spectroscopy. 268: 4783
Skrydstrup T, see Hansen SG (2006) 264: 135162
Smith CL, Milea JS, Nguyen GH (2005) Immobilization of Nucleic Acids Using BiotinStrept(avidin) Systems. 261: 6390
Smith DK, see Hirst AR (2005) 256: 237273
Specker D, Wittmann V (2007) Synthesis and Application of Glycopeptide and Glycoprotein
Mimetics. 267
Stadler A, see Kremsner JM (2006) 266: 233278
Stibor I, Zlatuskov P (2005) Chiral Recognition of Anions. 255: 3163
Stoddart JF, see Moonen NNP (2005) 262: 99132
Strauss CR, Varma RS (2006) Microwaves in Green and Sustainable Chemistry. 266: 199231
Suk D-H, see Crich D (2006) 263: 138
Suksai C, Tuntulani T (2005) Chromogenetic Anion Sensors. 255: 163198
Sun J-S, see Escud C (2005) 253: 109148
Suna E, Mutule I (2006) Microwave-assisted Heterocyclic Chemistry. 266: 49101
Sssmuth RD, see Wolter F (2007) 267
Svith H, see Daasbjerg K (2006) 263: 3970
Swerdlow H, see Du Q (2005) 261: 4561
Szathmry E, Santos M, Fernando C (2005) Evolutionary Potential and Requirements for
Minimal Protocells. 259: 167211
Taira S, see Yokoyama K (2005) 261: 91112
Takahashi H, see Tamura R (2007) 269
Takahashi K, see Ueyama N (2007) 271: 155193
Tamiaki H, see Balaban TS (2005) 258: 138
Tamura R, Takahashi H, Fujimoto D, Ushio T (2007) Mechanism and Scope of Preferential
Enrichment, a Symmetry-Breaking Enantiomeric Resolution Phenomenon. 269
Thar J, Reckien W, Kirchner B (2007) CarParrinello Molecular Dynamics Simulations and
Biological Systems. 268: 133171
Thayer DA, Wong C-H (2007) Enzymatic Synthesis of Glycopeptides and Glycoproteins. 267
Thiel W, see Senn HM (2007) 268: 173289
Tiedemann BEF, see Seeber G (2006) 265: 147184
Tobey SL, see Houk RJT (2005) 255: 199229
Toda F (2005) Thermal and Photochemical Reactions in the Solid-State. 254: 140
Tour JM, see James DK (2005) 257: 3362
357
Trask AV, Jones W (2005) Crystal Engineering of Organic Cocrystals by the Solid-State
Grinding Approach. 254: 4170
Tuntulani T, see Suksai C (2005) 255: 163198
Uccelli L, see Boschi A (2005) 252: 85115
Ueyama N, Takahashi K, Onoda A, Okamura T, Yamamoto H (2007) InorganicOrganic
Calcium Carbonate Composite of Synthetic Polymer Ligands with an Intramolecular
NH O Hydrogen Bond. 271: 155193
Ushio T, see Tamura R (2007) 269
Varma RS, see Strauss CR (2006) 266: 199231
Veciana J, see Amabilino DB (2006) 265: 253302
Venturi M, see Balzani V (2005) 262: 127
Vezin H, see Dias N (2005) 253: 89108
Vgtle F, see Fages F (2005) 256: 77131
Vgtle M, see inic M (2005) 256: 3976
Volkmer D, see Fricke M (2007) 270: 141
Vries TR, see Kellogg RM (2007) 269
Walschus U, see Luderer F (2005) 260: 3756
Walton JC (2006) Unusual Radical Cyclisations. 264: 163200
Wannberg J, Ersmark K, Larhed M (2006) Microwave-Accelerated Synthesis of Protease
Inhibitors. 266: 167197
Warman JM, see Grozema FC (2005) 257: 135164
Warren JD, Geng X, Danishefsky SJ (2007) Synthetic Glycopeptide-Based Vaccines. 267
Wasielewski MR, see Weiss EA (2005) 257: 103133
Weiss EA, Wasielewski MR, Ratner MA (2005) Molecules as Wires: Molecule-Assisted Movement of Charge and Energy. 257: 103133
Weissbuch I, Leiserowitz L, Lahav M (2005) Stochastic Mirror Symmetry Breaking via SelfAssembly, Reactivity and Amplication of Chirality: Relevance to Abiotic Conditions.
259: 123165
Williams LD (2005) Between Objectivity and Whim: Nucleic Acid Structural Biology. 253:
7788
Wittmann V, see Specker D (2007) 267
Wright DW, see Carney CK (2007) 270: 155185
Wolter F, Schoof S, Sssmuth RD (2007) Synopsis of Structural, Biosynthetic, and Chemical
Aspects of Glycopeptide Antibiotics. 267
Wong C-H, see Thayer DA (2007) 267
Wong KM-C, see Yam VW-W (2005) 257: 132
Wrthner F, see You C-C (2005) 258: 3982
Xia W, see Scheffer JR (2005) 254: 233262
Yam VW-W, Wong KM-C (2005) Luminescent Molecular Rods Transition-Metal Alkynyl
Complexes. 257: 132
Yamamoto H, see Ueyama N (2007) 271: 155193
Yashima E, see Maeda K (2006) 265: 4788
Yokoyama K, Taira S (2005) Self-Assembly DNA-Conjugated Polymer for DNA Immobilization on Chip. 261: 91112
358
Yoshikawa I, see Araki K (2005) 256: 133165

Yoshioka R (2007) Racemization, Optical Resolution and Crystallization-Induced Asymmetric Transformation of Amino Acids and Pharmaceutical Intermediates. 269
You C-C, Dobrawa R, Saha-Mller CR, Wrthner F (2005) Metallosupramolecular Dye
Assemblies. 258: 3982
Yu J, see Dittrich M (2007) 268: 319347
Yu S-H (2007) Bio-inspired Crystal Growth by Synthetic Templates. 271: 79118
Zampella G, see Bertini L (2007) 268: 146
Zard SZ, see Quiclet-Sire B (2006) 264: 201236
Zhang W (2006) Microwave-Enhanced High-Speed Fluorous Synthesis. 266: 145166
Zhang X-E, Deng J-Y (2005) Detection of Mutations in Rifampin-Resistant Mycobacterium
Tuberculosis by Short Oligonucleotide Ligation Assay on DNA Chips (SOLAC). 261:
169190
Zimmerman J, Sibi MP (2006) Enantioselective Radical Reactions. 263: 107162
inic M, see Fages F (2005) 256: 77131
Z
inic M, Vgtle F, Fages F (2005) Cholesterol-Based Gelators. 256: 3976
Zipse H (2006) Radical StabilityA Theoretical Perspective. 263: 163190
Zlatuskov P, see Stibor I (2005) 255: 3163
Subject Index
Active site models 10

Additive schemes 181
Adiabatic dynamics 219
Amino acids 155
Antiferromagnetic coupling, hydrogenases
2
Aquaporins 155
ATP hydrolysis 319, 320, 323
Density functional theory (DFT) 1, 33, 48

Difference-IR/-Raman 105
Dispersion corrections 143
DMSO, diffusionin water 135
DNA 157
, base pairs, binding/unbinding 307
DNA helicase 319
DNA replication 292
Bacteriorhodopsin 155
Basis sets 145
BennettChandler 293
Bioinorganic spectroscopy 50
Biomimetic models 7
Biomimetics 158
, nitrogen xation 159
Biomolecular isomerization 306
Biomolecular reaction modelling 220
Biomolecules, large 115
Blue copper proteins, spectroscopy 70
BornOppenheimer approximation 91
Boundary atoms 197
Boundary schemes 189
Broken symmetry approach 17
Electron localization function (ELF), water

dimer 153
Electronic potential energy hypersurface
95
Electronic resonance, Raman 118
Electronic structure 141
Embedding 182, 325
ENDOR spectra, DFT 78
Environment effects, molecules in solution
97
Enzymatic mechanisms 173
Enzyme catalysis 15, 224, 311
, cycle 15
Exchange correlation functional 143
Excited state hypersurfaces 96
Excited states MD 158
Carbon monoxide, [Fe] hydrogenase

28
CarParrinello simulations 133, 149
Caspase-3 157
Chirality, VCD/ROA 110
Chromatin formation 292
CO photolysis, TDDFT 28
Coordination compounds 1
, bio-inspired 12
Copper complexes, binuclear mixed
valence 78
CREP-binding protein (CBP) 155
Cysteine radical, g-tensor anisotropy
67
F1-ATPase 320
Fe(V) 76
Fictitious mass 150
First-principles methods 86
Fock operator, solvated 328
Free-energy methods 173
Free-energy perturbation 215
Frozen localized orbitals 199
Galactose oxidase 156
Generalized gradient approximation (GGA)
144
Glucanase 157
360
Subject Index
Ground state hypersurfaces 96

GTP hydrolysis 156
Guanidinium receptor 135
H-cluster, modeling 17
Harmonic approximation 92
Helicases 319
HIV-1 protease 156
HohenbergKohn theorems 141
Hydrogenases 1, 5
, [Fe] hydrogenases 2, 5
, [NiFe] hydrogenases 5
Hyperplanes 295
Hypersurfaces 96
HYSCORE spectra simulations, DFT
78
Importance sampling 299
IR 105
Iron complexes, mononuclear 71
Kinesin 320
KohnSham equations
142
Langevin simulations, PcrA function

339
Leap frog algorithm 332
Linear response theory, ground state 53
Link atoms 191, 326
Lipid bilayers 312
Lipids 292
Local contributions 118
Local density approximation (LDA) 143
Magnetic spectroscopy, ground state 50
Metadynamics 219
Metalloenzymes 1, 156
Molecular conformation 114
Molecular dynamics 173
, simulations 137
, , PcrA 319
Molecular motor 319, 322
Molecular simulations 173
Molecular structures, optimization 9
Monte Carlo simulations 214
Mssbauer spectroscopy 60
Motion, equations 137
Myoglobin 156
Myosin 320
Nitrogen xation 159

Non-equilibrium thermodynamic
integration 217
NoseHoover thermostat 139, 297
NRVS 124
Nucleosomes transitions 292
Optical spectroscopy 62
Optimization 173
, large systems 210
Organic radicals, high precision studies
64
Orientational averaging 102
Path ensemble 296
, Monte Carlo 299
PcrA helicase 319
Peptides 155
Phenoxyl radicals, molecular g-tensors
65
Photochemical reactions, [Fe]
hydrogenases 2
, modeling 26
Plane waves 145
Protein folding 308
Proteins 155
Pseudopotentials 147
QM/MM 151, 319
, biomolecular 227
, bonded interactions 188
, boundary, covalent bonds 189
, combined 173
, energy expressions 179
, free-energy perturbation 215
, implementations 205
, interactions, long-range electrostatic
186
, method 178
, molecular-dynamics 214
, optimization techniques 211
, packages 205
, reaction-path potential 219
, studies, biomolecular 227
, van der Waals interactions 187
QM/MM-related approaches 207
Quantum chemistry 1
Quinone radical anionsolvent interactions
66
Subject Index
361
Raman 105
Reaction-path techniques
213
Reactivity, modeling 9
Resonance CARS 122
Resonance ROA 124
Restrained electrostatic potential (RESP)
charge method 326
Rhodopsin 155, 159
Ribozymatic reactions 292
RNA 157
, folding 292
ROA theory 113
Subtractive schemes 179

Switches, biochemical networks 312
Serine protease 156

Solvated Fock operator 328
Solvent effects 116
Spontaneous resonance Raman scattering
118
Stable states 298
Stationary points 209
Stochastic modeling, PcrA 319
VCD theory 112

Vibrational normal modes 91
Vibrational spectroscopy 86
, intensities 98
TDDFT modeling 26
Thermostats 139, 297
Time-dependent density functional theory
(TDDFT) 26, 38, 158
Transcription 292
Transition state theory (TST) 293
Transition-path sampling 218, 296
Translocation 292
Trans-membrane transport 293
Triosephosphate isomerase 157
Water clusters, cooperative effects 136

Weighted density approximation (WDA)
144

Reiher M. Atomistic Approaches in Modern Biology (2007)

Uploaded by

Document Informationclick to expand document informationbiologia

Document Informationclick to expand document information

Copyright:

Available Formats

Reiher M. Atomistic Approaches in Modern Biology (2007)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Reiher M. Atomistic Approaches in Modern Biology (2007)

Uploaded by

Copyright:

Available Formats

268

Topics in Current Chemistry

Topics in Current Chemistry

Glycopeptides and Glycoproteins

Microwave Methods in Organic Synthesis

Low Molecular Mass Gelators

Atomistic Approaches in Modern Biology

Prof. Jean-Marie Lehn

Dipartimento di Chimica G. Ciamician

Prof. Dr. Armin de Meijere

Prof. Steven V. Ley

Institut fr Organische Chemie

University Chemical Laboratory

Prof. Dr. Kendall N. Houk

Prof. Stuart L. Schreiber

Prof. Dr. Joachim Thiem

Institut fr Organische Chemie

Prof. Barry M. Trost

Prof. Dr. Hisashi Yamamoto

Prof. Dr. F. Vgtle

Topics in Current Chemistry

at springer.com using the search function.

This volume of Topics in Current Chemistry presents an overview of atomistic

The second one by Herrmann and Reiher discusses techniques of vibrational

Quantum Chemical Investigations of Reaction Paths

Theoretical Bioinorganic Spectroscopy

First-Principles Approach to Vibrational Spectroscopy of Biomolecules

CarParrinello Molecular Dynamics Simulations

Top Curr Chem (2007) 268: 146

Quantum Chemical Investigations of Reaction Paths

of Biotechnology and Biosciences,

Investigation of Metalloenzymes with Quantum Chemical Methods . . . .

The Hydrogenase Example . . . . . . . . . . . . . . . . . . . . . . . . . . .

Modeling the Full H-Cluster . . . . . . . . . . . . . . . . . . . . . . . . . .

Modeling Photochemical Reaction Paths . . . . . . . . . . . . . . . . . . .

Conclusions and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . .

Outline of Density Functional Theory . . . . . . . . . . . . . . . . . . . . .

Outline of Time-Dependent Density Functional Theory . . . . . . . . . . .

the investigation of very complex molecular systems, such as biological macromolecules.

Quantum Chemical Investigations of Metalloenzymes

The different description of correlation effects is one of the fundamental

Quantum Chemical Investigations of Metalloenzymes

ition, efcient approaches developed to calculate the Coulomb energy, such

The structural organization of [NiFe] and [Fe] hydrogenases, as well as

Quantum Chemical Investigations of Metalloenzymes

Fig. 2 Structure of the H-cluster found in [Fe] hydrogenases

Fig. 3 Structure of the simplest synthetic model of the [2Fe]H cluster

Fig. 4 Structures of recently reported synthetic models of the [2Fe]H cluster

Quantum Chemical Investigations of Metalloenzymes

Quantum Chemical Investigations of Metalloenzymes

In the eld of coordination compounds related to [Fe] hydrogenases, the

experimentally) can allow one to highlight relationships that otherwise could

Quantum Chemical Investigations of Metalloenzymes

electronic factors at the basis of peculiar reactivity properties, as well as to

Quantum Chemical Investigations of Metalloenzymes

Theoretical investigations based on quantum chemical approaches have

Quantum Chemical Investigations of Metalloenzymes

where B is the interlayer resonance delocalization parameter and S is the total

a-CO S1 1A a S1 1A + CO

a-CO S3 1A a S2 1A + CO