See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/232009962

Computational Enzymology

Article  in  Methods in molecular biology (Clifton, N.J.) · January 2013

DOI: 10.1007/978-1-62703-017-5_4 · Source: PubMed

CITATIONS READS

16 271

2 authors, including:

Alessio Lodola
Università di Parma
164 PUBLICATIONS   3,338 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Melatonin receptor ligands View project

All content following this page was uploaded by Alessio Lodola on 20 May 2014.

The user has requested enhancement of the downloaded file.

 Chapter 4

Computational Enzymology
Alessio Lodola and Adrian J. Mulholland

Abstract
Techniques for modelling enzyme-catalyzed reaction mechanisms are making increasingly important
contributions to biochemistry. They can address fundamental questions in enzyme catalysis and have the
potential to contribute to practical applications such as drug development.

Key words: QM/MM, Enzyme, Catalysis, Protein dynamics, Biomolecular simulation, Quantum
mechanics/molecular mechanics

1. Introduction

Molecular modelling and simulations can explore mechanisms of

biological catalysts (i.e., enzymes) at a level of detail that cannot be
achieved experimentally (1–11). Modelling can unravel the
mechanisms of enzyme-catalyzed reactions, identify the origins of
catalysis, analyze effects of mutations and genetic variations, and
help to develop structure–activity relationships (12–14). Since its
origins (15, 16), computational enzymology has grown enor-
mously, particularly in recent years. There has also been a significant
improvement in the accuracy of computational methods. For exam-
ple, it is now possible to achieve an unprecedented level of accuracy
in calculations on enzyme-catalyzed reactions with combined quan-
tum mechanics/molecular mechanics (QM/MM) methods (17).
In the best cases, calculations can give activation energies that agree
extremely well with experiment. High-level quantum chemical
methods allow calculations of energy barriers, in the best cases,
near “chemical accuracy” (1 kcal/mol) (18). Quantitative predic-
tions at this level were only previously possible for very small
molecules. Carefully parameterized empirical molecular simulation
approaches also give excellent agreement with experiments for
enzyme reactions (19).

Luca Monticelli and Emppu Salonen (eds.), Biomolecular Simulations: Methods and Protocols, Methods in Molecular Biology,
vol. 924, DOI 10.1007/978-1-62703-017-5_4, # Springer Science+Business Media New York 2013

67
68 A. Lodola and A.J. Mulholland

Identifying the chemical mechanisms of enzymes solely from

experiments is often difficult. Many mechanisms in the literature are
probably wrong in important details, e.g., as more recent experi-
ments and simulations have shown for hen egg-white lysozyme (20,
21). The physical origins of enzyme catalysis also continue to be
hotly debated. Recent controversies have centered on “low-bar-
rier” hydrogen bonds (22–25), so-called near-attack conformations
(26, 27) enzyme dynamics (28, 29), quantum tunnelling (30–33),
and entropic effects (34). The applicability of transition state theory
to enzyme reactions has also been questioned (35). Molecular
simulations are proving to be crucial in testing these proposals.
Transition states are central to understanding chemical reactiv-
ity and catalysis, but experiments cannot directly study them in
enzymes because of their extremely short lifetimes, and because of
the large size and complexity of enzymes. Molecular modelling can
analyze transition states directly and identify interactions involved in
catalysis (e.g., a conserved proline residue that specifically stabilizes
the transition state for aromatic hydroxylation in the flavin depen-
dent monooxygenases para-hydroxybenzoate hydroxylase and phe-
nol hydroxylase (36, 37)). Such interactions may not be apparent
(and may not exist) in available experimental structures. This type of
knowledge can assist in ligand design, e.g., as a potential route to
enhanced affinity. Also, in contrast to some experimental
(e.g., structural) studies, which may require mutation of the enzyme
or use of alternative (e.g., inefficient) substrates (e.g., to slow down
the reaction, and prolong the lifetime of intermediates, to allow
their spectroscopic or structural characterization), molecular mod-
elling can study directly the “wild type” reaction (i.e., the reaction as
it occurs in the naturally occurring enzyme). Computational enzy-
mology interacts fruitfully with experiments, which can validate
modelling approaches, which in turn can interpret experimental
findings and suggest new experiments.

2. Materials
and Methods

2.1. Choice of Enzyme An enzyme structure from X-ray crystallography (ideally high reso-
Structure for Modelling lution), is the usual starting point for modelling an enzyme-
catalyzed reaction. A crystal structure of an enzyme alone, with
no ligand bound at the active site, may not be useful, because it is
difficult to predict binding modes and protein conformational
changes associated with binding. NMR structure ensembles can
give useful complementary information on dynamics and interac-
tions, but generally do not define atomic positions precisely enough
for mechanistic modelling. In some cases a homology model may
be sufficiently reliable, but should be treated with great caution: the
positions, relative orientations and packing of side chains may not
 4 Computational Enzymology 69

be modelled sufficiently accurately. An example of the successful use

of homology modelling for mechanistic studies is an investigation
of the substrate binding mode and reaction mechanism of a malarial
protease with a novel active site, by automated docking, and molec-
ular dynamics/reaction free energy profile simulations (38).
The structure used for modelling must accurately represent the
reacting enzyme complex; a crystal structure of an enzyme-
inhibitor complex is often a good choice. The inhibitor should
resemble the substrate, product, transition state or an intermediate,
in its bound conformation. One should remember that there can be
local structural uncertainty due e.g., to protein dynamics, and
conformational variability or disorder, even in high-resolution
structures. It is usually not possible to determine crystal structures
of active enzyme-substrate complexes, unless specialized conditions
or variant substrates or enzymes are used. In some cases, structures
of several complexes along the reaction pathway may be available
(30). Calculations can then model the chemical and structural
changes, unstable intermediates and transition state structures,
providing a picture of the whole reaction in the enzyme. For
many enzymes, the conformational changes that take place during
the reaction are small, and so modelling based on a single structure
of an enzyme complex can give a reliable picture of the reaction. It
may, however, be necessary to consider the effects of conforma-
tional fluctuations in the protein (see below).

2.2. Effects of Protein Proteins have many conformational substates, and a single structure
Dynamics may not be truly representative for modelling a mechanism (39).
Extensive conformational sampling may be needed to generate a
representative ensemble of structures. To calculate free energy
profiles, i.e., potentials of mean force (40), a simulation method
must be capable of calculating trajectories of many picoseconds at
least (or a similarly large number of configurations in a Monte
Carlo simulation). Alternatively, molecular dynamics simulations
can be used to generate multiple structural models for subsequent
mechanistic calculations, to ensure wide sampling of possible
enzyme configurations. If multiple different crystal structures of
the same enzyme are available, these may be suitable as different
starting models, and similarly help to examine the effects of struc-
tural variation on the reaction.
Protein dynamics are believed to be important to their
biological functions in many cases (41). It is well known that
many enzymes undergo large conformational changes during
their reaction cycles (42). The possible relationship of dynamics
to enzyme catalysis is more controversial. It has been proposed that
protein dynamics contribute significantly to enhancing the rates of
reaction in enzymes, but simulations indicate that the effect of
protein dynamics in determining the rates of chemical reactions in
enzymes is relatively small (43). Protein conformational changes
70 A. Lodola and A.J. Mulholland

(e.g., involved with substrate binding or product release) can,

however, be rate-limiting for the overall reaction in many enzymes
(44), and in some cases are coupled to chemical changes
(e.g., facilitating product release). Quantum effects such as nuclear
tunnelling are important in reactions involving hydrogen transfer
(30, 45, 46) and the effects of protein dynamics on reactions
involving quantum tunnelling is an area of particularly active debate
(5, 10, 28, 47).

2.3. Determining Determining the chemical mechanism is an essential first step in

the Mechanism studying an enzyme-catalyzed reaction. This is not trivial: many
“textbook” mechanisms are probably wrong. The first aim is to
establish the identities and functions of catalytic residues; many
mechanisms in the biochemical literature assign functions to resi-
dues which are probably incorrect. Next, any specific interactions
that stabilize transition states or reaction intermediates should be
identified and analyzed. A typical computational approach to mod-
elling reactions is to optimize the structures of key species (such as
transition state structures); preferably entire reaction pathways
should be optimized or simulated.

2.4. Analyzing Catalysis To understand catalysis, i.e., to understand why a reaction in an

enzyme reaction is faster than an uncatalyzed or nonenzymic reac-
tion, the two reactions should be compared (although deciding on
an appropriate “reference” reaction may be difficult (6)). Practical
applications often have simpler aims such as predicting the effects of
a mutation on activity or on the specificity of an enzyme for alter-
native substrates. Overall, understanding enzyme mechanism, spec-
ificity and catalysis presents a range of challenges, and different
types of modelling or simulation methods are needed to investigate
different types of question, as outlined below.

2.5. Methods for “Molecular mechanics” (MM) methods can model protein struc-
Modelling Enzymes ture and dynamics well, but standard MM methods cannot be used
and Enzyme-Catalyzed directly to model chemical reactions, because of their simple func-
Reactions tional forms (e.g., harmonic terms for bond stretching, and an
inability to model changes in electronic distribution because of
2.5.1. Molecular Mechanics
the invariant atomic point partial charge model). The simplicity of
MM “force fields” (potential functions) allows long timescale
(e.g., now up to millisecond) simulations of protein dynamics,
and simulations of large proteins. Molecular dynamics simulations
can study conformational changes (which are rate-limiting in many
enzymes under typical conditions): e.g., simulations of the human
scavenger decapping enzyme (DcpS) found a cooperative periodic
opening and closing of the dimer, over tens of nanoseconds (48).
Molecular dynamics simulations can also investigate substrate con-
formational behavior, which can help to develop mechanistic ideas,
 4 Computational Enzymology 71

but simulations of substrate complexes without consideration of

the reaction can sometimes be misleading (49).
Computer programs for biomolecular dynamics simulations
include AMBER (50), CHARMM (51), GROMOS (52), NAMD
(53) and TINKER (54); these should not be confused with force
fields, which may have the same or similar names. A force field
consists of the energy function and the parameters. Current protein
force fields use similar potential energy functions, in which bonds
and valence angles are represented by harmonic terms, electrostatic
interactions are represented by invariant point charges on atoms, a
simple representation that cannot capture the full electrostatic
properties of a molecule. Dispersion and exchange repulsion are
included by a simple Lennard-Jones function (usually of the 12-6
variety). Current widely used all-atom MM force fields for proteins
are OPLS/AA (55, 56), CHARMM22-27 (57), and AMBER
(PARM99 (50, 58–60)). Force fields for other types of biological
macromolecules (e.g., lipids, nucleic acids (61, 62)) consistent with
these protein force fields have also been developed (e.g., the
CHARMM27 (63, 64) and AMBER nucleic acid parameters (65)
and CHARMM parameters for lipids (66)). Most biomolecular
MM force fields have been developed to be compatible with simple
point charge models of water, e.g., the TIP3P water model (67).
Current standard biomolecular MM force fields only include elec-
tronic polarization in an average, invariant way. The next genera-
tion of protein MM force fields will probably include electronic
polarization explicitly (68, 69).
Standard MM potential functions cannot be applied to model
the breaking and making of bonds (and electronic reorganization)
in a chemical reaction. Also, MM force field parameters are devel-
oped based on the properties of stable molecules, and so will usually
not be applicable to transition states and intermediates. MM func-
tions and parameters can be developed specifically for reactions
(e.g., using different functional forms, such as Morse curves to
allow bond breaking), which has been successful for organic reac-
tions in solution (70). Such parameters are, however, applicable
only to a particular reaction, or small class of reactions. Also, the
form of the potential function imposes limitations, such as the
neglect of electronic polarization.

2.5.2. Empirical Valence In the empirical valence bond (EVB) method (6), a few resonance
Bond Methods structures are chosen to represent the reaction. The energy of each
resonance form is given by a simple empirical force field, with the
potential energy given by solving the related secular equation. The
EVB Hamiltonian is calibrated to reproduce experimental data for a
reaction in solution, or ab initio results can be used (71). The
surrounding protein and solvent are modelled by an empirical
force field, with appropriate treatment of long-range electrostatics.
The activation free energy of activation is calculated from free
72 A. Lodola and A.J. Mulholland

energy perturbation simulations (72). The free energy surfaces can

be calibrated by comparison with experimental data for reactions in
solution. The EVB method allows the use of a non-geometrical
reaction coordinate, which allows evaluation of nonequilibrium
solvation effects (6). A mapping procedure gradually moves the
system from the reactants to products. The simplicity of the EVB
potential function allows extensive molecular dynamics simula-
tions, giving good sampling (73). The EVB method is now widely
used for studying reactions in condensed phases, particularly in
enzymes (74–80).

2.5.3. Quantum Chemical In most enzymes, the chemical changes occurring in the reaction
Calculations on Small (Active are confined to a relatively small region, the active site of the
Site) Models enzyme. One approach to the study of enzyme-catalyzed reactions
is to study only the active site, using quantum chemical methods
(this is sometimes called the “supermolecule” or cluster approach).
Such models can represent important features of an enzyme reac-
tion, and can identify likely mechanisms. The active site model
should contain molecules representing the substrate(s) (and any
cofactors) and enzyme residues involved in the chemical reaction or
in binding substrate. Important functional groups (such as catalytic
amino acid side chains) are represented by small molecules, e.g.,
acetate can represent an aspartate side chain, imidazole for histi-
dine, etc.). The initial positions of these groups are usually coordi-
nates taken from a crystal structure, or from a molecular dynamics
simulation of an enzyme complex.
Quantum chemical calculations (i.e., methods that calculate
molecular electronic structure using quantum mechanics, e.g., ab
initio molecular orbital or density functional theory calculations)
can give excellent results for reactions of small molecules. The best
“ab initio” methods (such as CCSD(T)), which include correlation
between electrons, can calculate rate constants for reactions involv-
ing very few atoms (in the gas phase) with small error bars, similar
to experiments on these systems. Such calculations require very
large computational resources, however, severely limiting the size
of the system that can be treated. More approximate methods,
(such as the semiempirical molecular orbital techniques AM1 and
PM3), are computationally much cheaper, and can model larger
systems (containing of the order of hundreds of atoms). Techni-
ques (e.g., “linear-scaling” methods) have been developed that
allow semiempirical electronic structure calculations on whole pro-
teins (81–83). Semiempirical methods are, however, inaccurate for
many applications (e.g., typical errors of over 10 kcal/mol for
barriers and reaction energies, though specifically parameterized
semiempirical methods can give improved accuracy for a particular
reaction (40, 47)). Density functional theory (DFT) methods
(e.g., applying the B3LYP functional) are considerably more accu-
rate, while also allowing calculations on relatively large systems
 4 Computational Enzymology 73

(e.g., active site models of the order of 100 atoms), larger than is
feasible with correlated ab initio calculations. Many DFT methods,
however, lack important physical interactions, such as dispersion,
which are important in the binding of ligands to proteins. Disper-
sion effects can also be important in the calculation of energy
barriers (84). DFT often gives barrier heights that are too low by
several kcal/mol, and it can be difficult to assess the accuracy of
results, because DFT does not offer a route to their systematic
improvement or testing.
Calculations on active site models can provide models of transi-
tion states and intermediates (see below). This has proved particu-
larly useful for studying metalloenzymes, using DFT methods. In
many metalloenzymes, all the important chemical steps take place at
one metal center (or a small number of metal ions bound at one
site). The metal also holds its ligands in place, limiting the require-
ment for restraints to maintain the correct active site structure.
Calculations on small clusters can give useful mechanistic insight
(85, 86): e.g., a mechanism can be ruled out if the calculated barriers
for it are significantly higher than the experimentally derived activa-
tion energy, based on the accuracy of the computational method.
The effects of the environment are usually either omitted, or
included only in an approximate way (e.g., by continuum solvation
methods, which cannot fully represent the heterogeneous electro-
static environment in an enzyme). It is useful to test the sensitivity of
the results to the choice of, e.g., dielectric constant.
To calculate the energy barrier for a reaction in a cluster model,
structures of the reactant, transition state, intermediates and pro-
ducts of the reaction should be optimized. Doing this while main-
taining the correct orientations of the groups in the protein can be
difficult. Small models may also lack some important functional
groups. It is important to consider carefully which groups to
include, striking a balance between computational feasibility and
the desire for a larger, more extensive model. A larger model is not,
however, always a better model: a larger model will be susceptible
to greater conformational complexity: conformational changes,
even outside the active site, may artificially affect relative energies
along the reaction path). Also, charged groups can have unrealisti-
cally large effects on reaction energies. One should test the sensitiv-
ity of the results to the choice of model (and also to factors such as
the choice of density functional).

2.5.4. Combined Quantum “Hybrid” methods that combine quantum chemical methods with
Mechanics/Molecular molecular mechanics allow more extensive calculations, on larger
Mechanics Methods models of enzymes, than is possible with purely quantum chemical
for Modelling Enzyme techniques. Such QM/MM methods are very important in compu-
Reactions tational enzymology. The QM/MM approach is simple: a small
part of the system, at the active site, is treated quantum mechani-
cally, i.e., by an electronic structure method of one of the types
74 A. Lodola and A.J. Mulholland

discussed above, which allows modelling of the electronic rearran-

gements involved in the breaking and making of chemical bonds.
The QM region contains the reacting groups of the enzyme, sub-
strate and any cofactors. The rest of the system is treated by MM.
QM/MM calculations can be carried out at ab initio (87) or
semiempirical (88) molecular orbital, density functional (89) or
approximate density functional (e.g., self-consistent charge density
functional tight-binding, SCC-DFTB (90)) levels of QM electronic
structure calculation. Different types of coupling between the QM
and MM regions are possible (see below).
Many different QM/MM implementations are available,
in several widely used programs. Reaction pathways and transition
state structures can be optimized (91, 92). Molecular dynamics
simulations are possible with cheaper QM/MM methods (such
as semiempirical or SCC-DFTB level QM) (93). Free energy differ-
ences, such as activation free energies can be calculated, as can
quantum effects such as tunnelling and zero-point corrections (5,
30, 40). High-level QM/MM calculations (e.g., ab initio or density
functional level QM) are required for some systems and also have an
important role in testing more approximate methods. The compu-
tational demands of high level (e.g., ab initio, (94)) QM/MM
calculations (17) typically limit their application to “single point”
energy calculations on structures optimized at lower levels (95).
DFT/MM methods can be used for energy minimization/geome-
try optimization to generate reaction paths.
QM/MM methods can also be used in free energy perturba-
tion simulations (96), e.g., to calculate relative binding affinities,
and in molecular docking and scoring (97). QM/MM methods
provide several advantages over MM methods in studies of ligands
bound to proteins, including potentially a better physical descrip-
tion of a ligand (e.g., including electronic polarization), and avoid-
ing the need for time-consuming MM parameterization for the
ligand.

Interactions Between One of the main differences between various QM/MM methods
the QM and MM Regions is the type of QM/MM coupling employed i.e., in how the inter-
actions (if any) between the QM and MM systems are treated
(98). The simplest linking of QM and MM methods involves a
straightforward “mechanical” embedding of the QM region in
the MM environment, treating interactions between the QM and
MM regions only by MM (i.e., the QM system is represented by
(MM) point charges in its interaction with the MM environment).
In calculations of this type, the QM/MM energy of the whole
system, ETOTALQM/MM, is calculated in a simple subtractive
scheme. This simple subtractive approach can be applied to all
combinations of theory levels (for example combining different
levels of QM treatment (QM/QM) as opposed to QM with MM)
and forms the basis for the (simplest form of the) multilayer
 4 Computational Enzymology 75

ONIOM (Our own N-layered Integrated molecular Orbital and

molecular Mechanics) method (99). A QM/QM calculation
involves a high and a low level of QM theory, with a small region
treated by a high level and the entire model treated at the low
level; polarization is included at the lower level of QM theory.
More intensive QM/MM calculations include polarization of
the QM region by the MM environment. This is likely to be
important for many enzymes, given their polar nature. QM/MM
methods of this type include electrostatic interactions between the
QM and MM regions in the QM calculation, thus modelling polar-
ization of the QM system by the MM atoms, by directly including
the MM atomic charges of the MM group in the QM calculation.
The electronic structure calculation thus includes the effects of the
MM atoms. A further level of complexity would involve polariza-
tion of the MM region also through the use of a polarizable MM
force field, and potentially self-consistent polarization of the MM
region through an iterative procedure. Models of this sort are vastly
more computationally intensive and may not always yield better
results (100). QM/MM methods that include polarization of the
MM system have been developed for small molecular systems
(101). Current standard MM force fields for biological macromo-
lecules do not model changes in polarization, however.
In typical QM/MM calculations, the energy of the QM atoms,
EQM, is given by a molecular orbital or DFT method, and the energy
of the atoms in the MM region, EMM, is given by MM. A boundary
term, EBoundary, is usually necessary to account for the effects of the
surroundings, e.g., to include the effects of parts of the protein that
are not included in the simulation model. It may also be necessary to
scale/reduce charges at the boundary of the simulation system: this
represents the effects of dielectric screening in a crude sense, to
avoid overestimating the effects of charged groups on the active
site (102). The QM/MM interaction energy, EQM/MM typically
consists of terms due to electrostatic interactions and van der
Waals interactions, and any bonded interaction terms. In many
implementations, MM bonding terms (energies of bond stretching,
angle bending, torsion angle rotation, etc.) are included for all QM/
MM interactions which involve at least one MM atom (88). In an ab
initio QM/MM calculation, the MM atomic charges are generally
included directly through one-electron integrals. The treatment of
QM/MM electrostatic interactions is a little less straightforward
when semiempirical molecular orbital methods such as AM1 and
PM3 are used, because they treat only valence electrons directly,
including the core electrons together with the nucleus as an atomic
“core.” In semiempirical QM/MM methods such as the AM1/
CHARMM method of Field et al., the electrostatic interactions
between QM and MM atoms are calculated by treating the MM
atoms as if they where semiempirical atomic cores (88).
76 A. Lodola and A.J. Mulholland

QM/MM van der Waals interactions (representing dispersion

and exchange repulsion interactions between QM and MM atoms)
are usually calculated by a molecular mechanics procedure
(e.g., through Lennard-Jones terms), exactly as the corresponding
interactions would be calculated between MM atoms not interact-
ing through bonding terms. MM van der Waals parameters must
therefore be chosen for each QM atom: these interactions are
significant at short distances, and are important in determining
QM/MM interaction energies and geometries. The van der Waals
parameters are important in differentiating MM atom types in their
interactions with the QM system, e.g., for MM atoms of the same
charge, which would otherwise be indistinguishable to the QM
system; van der Waals interactions are also important for interac-
tions of the QM system with nearby MM atoms whose charges are
close to zero. Often, standard MM van der Waals (Lennard-Jones)
parameters optimized for similar MM groups are used for QM
atoms in QM/MM calculations. This is convenient, but it is always
important to consider whether the van der Waals parameters pro-
vide a reliable description of QM/MM interactions. Where neces-
sary, the (MM) van der Waals parameters for the QM atoms can be
optimized to reproduce experimental or high-level ab initio results
(e.g., structures and interaction energies) for small molecular com-
plexes (93). One limitation of current QM/MM approaches of this
type is that the same van der Waals parameters are typically used for
the QM atoms throughout a simulation: in modelling a chemical
reaction, the chemical nature of the groups involved (treated by
QM) may change, altering their interactions, and so the use of
unchanging MM parameters may be inappropriate. Riccardi et al.
have investigated the effects of van der Waals parameters in QM/
MM (SCC-DFTB/CHARMM22) simulations (103). Different
parameter sets gave very different results for gas-phase clusters
and solvent structures around the solutes. However, condensed
phase thermodynamic quantities (e.g., the calculated reduction
potential and potential of mean force) were less sensitive to the
van der Waals parameters. These authors concluded that work to
improve the reliability of QM/MM methods for condensed phase
energetic properties should focus on factors other than van der
Waals interactions between QM and MM atoms, such as the treat-
ment of long-range electrostatic interactions.

Treatment of Long-Range To reduce computational requirements, the model may include

Electrostatic Interactions only a part of the whole protein (for example, a rough sphere
in QM/MM Simulations around the active site). In simulating a truncated protein system,
it is necessary to include restraints or constraints in the boundary
region to force the atoms belonging to it to remain close to their
positions in the crystal structure. One common approach is the
stochastic boundary molecular dynamics method (104, 105), in
which the simulation system is divided into a reaction region,
 4 Computational Enzymology 77

a buffer region and a reservoir region. Typically, the whole simula-

tion system may include all residues with an atom within a distance
of e.g., 15–25 Å of an atom in the active site. The buffer region
would contain atoms in the outer layer (e.g., outer 5 Å shell).
Atoms in the reaction region are treated by standard Newtonian
molecular dynamics, and are not subject to positional restraints.
The protein heavy atoms in the buffer region are restrained to
remain close to their (e.g., crystallographically determined) posi-
tions by harmonic forces, while a solvent deformable boundary
potential prevents “evaporation” of water. Atoms in the buffer are
subject to frictional and random forces (hence the term “stochas-
tic”) to represent exchange of energy with the surroundings (reser-
voir region). Atoms in the reservoir region are usually not included
because their presence (as fixed atoms) has been found to cause
excessive rigidity of the protein.
Ideally, long-range electrostatic interactions should be included
explicitly. Schemes for treatment of long-range electrostatic inter-
actions in QM/MM simulations have been developed, to allow
periodic boundary simulations (106). An alternative approach, for
QM/MM calculations under spherical boundary conditions (107)
is the generalized solvent boundary potential (GSBP) method
(108). This retains the practical advantage of treating a truncated
system, avoiding having to include the entire macromolecule in a
periodic simulation. The effects of the bulk solvent and macromol-
ecule atoms outside the simulation system are included at the
Poisson-Boltzmann level. Simulations using the GSBP method
were found to be more consistent with experimental data. Conven-
tional stochastic boundary molecular dynamics simulations pro-
duced artifacts, depending on the treatment of electrostatic
interactions. It was suggested that the commonly used interaction
truncation schemes should not be applied if possible in QM/MM
simulations, in particular for simulations that may involve extensive
conformational sampling.

QM/MM Partitioning Most QM/MM studies of enzymes require partitioning of cova-

Methods and Schemes lently bonded molecules into QM and MM regions. Typically,
some amino acid side chains participate in the reaction, and must
therefore be included in the QM region. Other side chains will
play binding roles, and a MM representation might be inade-
quate. Similarly, it may be more practical to treat only the reactive
parts of large cofactors or substrates by quantum chemical meth-
ods. There are two general QM/MM partitioning techniques that
can be employed: firstly special treatment of orbitals to satisfy the
valence shell of the QM atom at the QM/MM junction, for
example the local self-consistent field (LSCF) method (109,
110) or the generalized hybrid orbital (GHO) method (111).
Alternatively a QM atom (or pseudoatom) can be added at the
78 A. Lodola and A.J. Mulholland

QM/MM boundary, e.g., using a “link atom” or connection

atom method.
The local self-consistent field (LSCF) method (112) uses a
strictly localized bond orbital, also often described as a frozen
orbital, for the QM atom at the frontier between QM and MM
regions. The electron density of the orbital is calculated in advance,
using small models, and does not change during the QM/MM
calculation. The orbitals must be parameterized for each system and
QM method. The LSCF method avoids the need for dummy atoms
and provides a reasonable description of the chemical properties of
the frontier bond. It has been applied at semiempirical (113) and ab
initio (112) QM/MM levels.
The generalized hybrid orbital (GHO) method (114) uses
hybrid orbitals as basis functions on the frontier atom of the MM
fragment. It does not require extensive specific parameterization,
unlike the LSCF method. It uses four hybrid orbitals for an sp3
carbon atom, one of which is included in the self-consistent field
optimization of the QM region, while three are treated as auxiliary
orbitals. The parameters for the frontier atom are optimized to
reproduce properties of full QM systems. The localized orbitals
can be transferred, without specific parameterization of the active
orbital for each new system. A similar approach in DFT/MM
calculations is to freeze the electron density at the QM/MM junc-
tion (115). The GHO method has been applied in QM/MM
calculations at ab initio (116), SCC-DFTB (117) and density func-
tional (118) QM levels.
The “dummy junction atom” or link atom method introduces
so-called link atoms to satisfy the valence of the frontier atom in
the QM system (119). The link atom is usually a hydrogen atom
(88), but other atom types have been used, such as a halogen
(120). The link atom approach has been criticized, e.g., because it
introduces additional degrees of freedom associated with the link
atom, and the fact that a C–H bond is clearly not exactly equiva-
lent to a C–C bond. It is, however, simple and is widely used. The
results can be sensitive to the positioning of the link atom, and
also on exactly which MM atoms are excluded from the classical
electrostatic field that interacts with the QM region. Comparison
of the LSCF and link atom approaches for semiempirical QM/
MM calculations, however, showed that the two methods gave
similar results (121). It has been recommended that the link atom
should interact with all MM atoms except for those closest to the
QM atom to which the link atom is bonded. The link atom
method can give good results, with a good choice of the bound-
ary between QM and MM regions, e.g., across a carbon–carbon
single bond, far from chemical changes, and also preferably not
close to highly charged MM atoms.
Another method for dealing with the QM/MM boundary
between covalently bonded atoms is the connection atom method
 4 Computational Enzymology 79

(110, 122), which uses a monovalent pseudoatom instead of a link

atom. The parameters for the connection atom are optimized for
the partitioned covalent bond. The connection atoms interact with
the other QM atoms as a (specifically parameterized) QM atom,
and with the other MM atoms as a standard carbon atom. This
avoids the problem of a supplementary atom in the system, as the
connection atom and the classical frontier atom are unified. How-
ever, the need to reparameterize for each type of covalent bond at a
given level of quantum chemical theory is potentially laborious
(123). The connection atom method has been developed for
AM1 and PM3 (100), and DFT (122) QM/MM calculations.
Tests indicated that it is more accurate than the standard link
atom approach (100).
To overcome some of the problems that can arise with the
single link atom method (e.g., an unphysical dipole), Brooks et al.
have proposed a “double link atom” method (124). Also, their
Gaussian delocalization method for MM atomic charges could
simplify the calculation of energies and forces: e.g., even at short
distances, the delocalized Gaussian MM method does not require
the MM host atom charge to be excluded from the QM calculation,
as would be necessary when treating it as a point charge. The
delocalized Gaussian method can be combined with many QM/
MM partitioning techniques, such as the link atom, frozen orbital,
or pseudopotential methods. Tests of the delocalized Gaussian MM
and double link atom methods on small model systems indicated
that these methods gave better energetic properties than point
atomic MM charge and single link atom methods.
Cui et al. have tested link atom QM/MM partitioning meth-
ods, for the SCC-DFTB QM method (125), including all the
options available in the CHARMM program, which differ in their
treatment of electrostatic interactions with the MM atoms close to
the QM/MM frontier. They also proposed a divided frontier
charge protocol, in which the partial charge associated with the
MM atom bonded to the QM atom is distributed evenly on the
other MM atoms in the same MM group. Tests of these various link
atom schemes showed that QM/MM proton affinities and depro-
tonation energies are highly dependent on the particular link atom
scheme employed: standard single link atom methods gave errors of
up to 15–20 kcal/mol compared to pure QM calculations. Other
schemes were found to give better results. Activation barriers and
reaction energies were found, however, to be fairly insensitive to the
choice of link atom scheme (e.g., within 2–4 kcal/mol) because of
cancellation of errors. This is encouraging: the effect of using
different link atom schemes in QM/MM simulations was found
to be relatively small for chemical reactions in which the total
charge does not change. Other technical details, such as the treat-
ment of long-range electrostatics, are likely to play a more signifi-
cant role in determining energetics generally, and should be treated
carefully for reliable results.
80 A. Lodola and A.J. Mulholland

2.6. Modelling Enzyme With QM or QM/MM methods, potential energy surfaces of

Reactions by enzyme reaction mechanisms can be explored at a level of accuracy
Calculating Potential that can enable discrimination between different mechanisms: e.g.,
Energy Surfaces if the barrier for a proposed mechanism is significantly larger than
that derived from experiment (using transition state theory), within
the limits of accuracy of the computational method and experimen-
tal error, then that mechanism can be considered to be unlikely. A
mechanism with a calculated barrier comparable to the apparent
experimental barrier (for that step, or failing that for the overall
reaction) is more likely. However, to calculate rate constants also
requires reliable methods to calculate enthalpies, energies, and free
energies of reaction and activation, given the potential energy
surface. Traditional approaches to modelling reactions (e.g., in
the gas phase) rely on the identification of stationary points (reac-
tants, products, intermediates, transition states) via geometry opti-
mization, followed by computation of second derivatives to enable
relatively simplistic evaluation of zero-point corrections, thermal
and entropy terms. Algorithms developed for small molecules are
often not suitable for large systems: e.g., direct calculation, storage
and manipulation of Hessian matrices becomes extremely difficult.
A basic means of modelling approximate reaction paths is the
“adiabatic mapping” or “coordinate driving” approach. The energy
of the system is calculated by minimizing the energy at a series of
fixed (or restrained, e.g., by harmonic forces) values of a reaction
coordinate, e.g., the distance between two atoms. This approach
has been applied with success to many enzymes (1, 2), but it is only
valid if one conformation of the protein can represent the state of
the system at a particular value of the reaction coordinate. A single
minimum energy structure of this conformation may adequately
represent the several closely related structures making up the react-
ing conformational state. Minimizing the QM/MM potential
energy of such a representative conformation along the reaction
coordinate should give a reasonable approximation of the enthalpic
component of the potential of mean force (the free energy profile)
for the reaction. In contrast, simple calculations of potential energy
surfaces are likely to be unsuccessful or misleading for enzyme
reactions involving large movements of charge or large changes in
solvation (e.g., particularly for solvent–exposed sites, where rear-
rangement of water molecules might involve an unrealistically large
energy penalty where adiabatic mapping calculations are used,
(126)).
Due to the complexity of protein internal motions, many con-
formational substates exist, and a single structure might not be
truly representative. If this is the case, calculations including exten-
sive sampling of the system to obtain configurationally averaged
free-energy changes are needed, as opposed to energy minimiza-
tions, which do not include entropic effects and are sensitive to
starting geometries (126). A more simple approach to investigating
 4 Computational Enzymology 81

conformational effects is to use molecular dynamics simulations,

and/or to use multiple different crystal structures, to generate
multiple models for mechanism (e.g., adiabatic mapping) calcula-
tions, to ensure wide sampling of possible enzyme configurations
(127), with averaging, or Boltzmann-weighted averaging, of
energy barriers.
Despite the limitations and drawbacks of the adiabatic mapping
approach, it has been applied in many QM/MM applications. It has
the advantage that it is simple to apply, and does not require
intensive calculations, such as second derivative evaluations, or
simultaneous treatment of several points on a pathway. It can be
useful for initial scans of potential energy surfaces, and for generat-
ing approximate models of transition states and intermediates, in
which some allowance is made for structural relaxation to chemical
changes at the active site. It is suitable only for reactions involving
small chemical and structural changes, involving a small number of
groups. For some enzymes, this type of approach has been validated
through a correlation of calculated QM/MM barriers with activa-
tion energies derived from experiment (37).
As mentioned before, approaches based purely on calculations
of a potential energy surface may not account for significant con-
formational fluctuations of the protein. Conformational changes,
even on a small scale, may be coupled to, or significantly affect,
chemical changes. Fluctuations of the active site can greatly affect
the energy barrier. In the case of fatty acid amide hydrolase, con-
formational fluctuations do not affect the general shape of the
potential energy surfaces, but consistency between experimental
and calculated barriers is observed only with a specific (and rarely
occurring) arrangement of the enzyme-substrate complex (49).
These findings indicate that investigation of different protein con-
formations is essential for a meaningful determination of the ener-
getics of enzymic reactions for calculations of potential energy
profiles or surfaces.

2.7. Calculating Free The rate constant of a reaction is actually related not to the potential
Energy Profiles energy barrier, but to the free energy barrier, according to transition
for Enzyme-Catalyzed state theory. The techniques above calculate potential energy bar-
Reactions riers, for a particular conformation. Techniques that sample config-
urations along a reaction coordinate give a more sophisticated and
extensive description, by taking account of multiple conformations
and estimating entropic effects, and can be essential for modelling
some types of enzyme reactions. Simulations of this type provide
estimates of the free energy profile along a specific (reaction) coor-
dinate, which is often referred to as the potential of mean force.
Molecular dynamics and Monte Carlo methods in principle allow
such sampling, but do not provide good sampling of high energy
regions, such as in the vicinity of transition states. Conformational
82 A. Lodola and A.J. Mulholland

sampling of processes of chemical change therefore requires

specialized techniques, e.g., to bias the simulation to sample the
transition state region. Umbrella sampling is such a method, which
is widely used in molecular dynamics simulations e.g., with QM/
MM techniques, to model enzymic reactions (27). In this tech-
nique, a biasing potential is applied to force the system to remain
close to a specific value of a defined reaction coordinate. Often, an
umbrella sampling simulation will begin with simulation of a transi-
tion state or reactant complex; an umbrella (e.g., harmonic) poten-
tial restrains the reaction coordinate to a value corresponding to
e.g., the reactants. In other, subsequent simulations, the reference
value of the restraint is changed by a small amount to sample other
regions of the reaction coordinate. Often, the reaction coordinate is
defined in terms of bond lengths, in which case a typical difference
between the points would be 0.1–0.2 Å. The neighboring potentials
should give overlapping distributions: this can be achieved by
choosing an appropriate spacing of reaction coordinate values for
different simulations, and an appropriate magnitude of the force
constant of the restraint. The number of simulations is a balance
between accuracy and efficiency. The reaction coordinate values
during the (restrained) simulations are recorded. The effects of the
restraining potentials are removed in the analysis and combined,
typically by the weighted histogram analysis method (WHAM).
This gives the unbiased potential of mean force along the reaction
coordinate. It is important also to test for convergence with respect
to length (and numbers) of simulations.
QM/MM umbrella sampling simulations are possible with low
levels of QM theory, such as semiempirical molecular orbital meth-
ods (e.g., AM1 or PM3). Often, such methods are highly inaccu-
rate for reaction barriers and energies. Their accuracy can be
improved significantly by reparameterization for a specific reaction.
For example, specifically parameterized semiempirical QM/MM
methods have been used to investigate model reactions of glutathi-
one-S-transferase (GST) enzymes. QM/MM umbrella sampling
molecular dynamics simulations of the reaction of phenanthrene
9,10-oxide in a glutathione-S-transferase, identified a single amino
acid as a likely determinant of stereospecificity in the epoxide ring
opening (93). Similarly, specifically parameterized QM/MM meth-
ods have been applied to model the reaction between glutathione
and 1-chloro-2,4-dinitrobenzene. The results of QM/MM
umbrella sampling molecular dynamics simulations of this reaction
in the M1-1 GST isoenzyme, in mutant enzymes, and in solution,
agreed very well with experiment (128). QM/MM molecular
dynamics simulations are much more computationally demanding
than MM simulations, because of the computational expense of the
evaluation of the QM forces. Typical QM/MM umbrella sampling
applications have involved trajectories of picoseconds to nanose-
conds (multiple simulations of 30 ps each, at each value of the
 4 Computational Enzymology 83

reaction coordinate, in the case of Bowman et al. (128)), with

semiempirical QM methods. Approaches based on Monte Carlo
simulations (129, 130) avoid the requirement for force calcula-
tions, and are also promising.

3. Notes

The choice of an appropriate method for the particular enzyme and

questions of interest is vital. Careful testing and validation is impor-
tant. Quantitative predictions of reaction rates or the effects of
mutation remain very challenging, but for many enzymes, with
appropriate methods, useful predictions can be made with some
confidence. It is important to validate predictions from modelling
by comparisons with experimental data. An example is comparison
of calculated barriers for a series of alternative substrates with
activation energies derived from experimental rates: demonstration
of a correlation can validate mechanistic calculations (36, 131).
Some enzymes have become important model systems in the devel-
opment and testing of computational methods and protocols: these
include chorismate mutase (1, 8, 10, 17, 27, 87), citrate synthase
(3, 23, 132), P450cam (1, 2, 7, 11, 84), para-hydroxybenzoate
hydroxylase (2, 17, 36), triosephosphate isomerase (5, 13, 102),
fatty acid amide hydrolase (49, 95, 127, 129), and methylamine
dehydrogenase (5, 7, 46, 47, 133).

Acknowledgments

AJM is an EPSRC Leadership Fellow.

References
1. Lonsdale R, Ranaghan KE, Mulholland AJ 5. Garcia-Viloca M, Gao J, Karplus M, Truhlar
(2010) Computational enzymology. Chem DG (2004) How enzymes work: analysis by
Commun 46:2354–2372 modern rate theory and computer simula-
2. Senn HM, Thiel W (2009) QM/MM meth- tions. Science 303:186–195
ods for biomolecular systems. Angew Chem 6. Warshel A (2003) Computer simulations of
Int Ed Engl 48:1198–1229 enzyme catalysis: methods, progress, and
3. van der Kamp MW, Mulholland AJ (2008) insights. Annu Rev Biophys Biomol Struct
Computational enzymology: insight into 32:425–443
biological catalysts from modelling. Nat 7. Friesner RA, Guallar V (2005) Ab initio quan-
Prod Rep 25:1001–1014 tum chemical and mixed quantum mechan-
4. Mulholland AJ (2005) Modelling enzyme ics/molecular mechanics (QM/MM)
reaction mechanisms, specificity and catalysis. methods for studying enzymatic catalysis.
Drug Discov Today 10:1393–1402 Annu Rev Phys Chem 56:389–427
84 A. Lodola and A.J. Mulholland

8. Martı́ S, Roca M, Andrés J, Moliner V, Silla E, 22. Cleland WW, Frey PA, Gerlt JA (1998) The
Tuñón I, Bertrán J (2004) Theoretical low barrier hydrogen bond in enzymatic catal-
insights in enzyme catalysis. Chem Soc Rev ysis. J Biol Chem 273:25529–25532
33:98–107 23. Mulholland AJ, Lyne PD, Karplus M (2000)
9. Himo F (2006) Quantum chemical modeling Ab initio QM/MM study of the citrate
of enzyme active sites and reaction mechan- synthase mechanism: a low-barrier hydrogen
isms. Theor Chem Acc 116:232–240 bond is not involved. J Am Chem Soc
10. Warshel A, Sharma PK, Kato M, Xiang Y, Liu 122:534–535
HB, Olsson MHM (2006) Electrostatic basis 24. Schutz CN, Warshel A (2004) The low barrier
for enzyme catalysis. Chem Rev hydrogen bond (LBHB) proposal revisited:
106:3210–3235 the case of the Asp · His pair in serine pro-
11. Senn HM, Thiel W (2007) QM/MM studies teases. Proteins 55:711–723
of enzymes. Curr Opin Chem Biol 25. Molina PA, Jensen JH (2003) A predictive
11:182–187 model of strong hydrogen bonding in pro-
12. Mulholland AJ, Grant GH, Richards WG teins: the Nd1–H–Od1 hydrogen bond in
(1993) Computer modelling of enzyme cata- low-pH a-chymotrypsin and a-lytic protease.
lysed reaction mechanisms. Protein Eng J Phys Chem B 107:6226–6233
6:133–147 26. Hur S, Bruice TC (2003) Just a near attack
13. Mulholland AJ, Karplus M (1996) Simula- conformer for catalysis (chorismate to prephe-
tions of enzymic reactions. Biochem Soc nate rearrangements in water, antibody,
Trans 24:247–254 enzymes, and their mutants). J Am Chem
14. Åqvist J, Warshel A (1993) Simulation of Soc 125:10540–10542
enzyme reactions using valence bond force 27. Ranaghan KE, Mulholland AJ (2004) Con-
fields and other hybrid quantum/classical formational effects in enzyme catalysis: QM/
approaches. Chem Rev 93:2523–2544 MM free energy calculation of the ‘NAC’
15. Warshel A, Levitt M (1976) Theoretical stud- contribution in chorismate mutase. Chem
ies of enzymic reactions: dielectric, electro- Commun 10:1238–1239
static and steric stabilization of the 28. Olsson MH, Parson WW, Warshel A (2006)
carbonium ion in the reaction of lysozyme. Dynamical contributions to enzyme catalysis:
J Mol Biol 103:227–249 critical tests of a popular hypothesis. Chem
16. Scheiner S, Lipscomb WN (1976) Catalytic Rev 106:1737–1756
mechanism of serine proteinases. Proc Natl 29. Hammes-Schiffer S, Watney JB (2006)
Acad Sci USA 73:432–436 Hydride transfer catalyzed by Escherichia
17. Claeyssens F, Harvey JN, Manby FR, Mata coli and Bacillus subtilis dihydrofolate reduc-
RA, Mulholland AJ, Ranaghan KE, Schultz tase: Coupled motions and distal mutations.
M, Thiel S, Thiel W, Werner H-J (2006) Phil Trans Roy Soc B 361:1365–1373
High accuracy computation of reaction bar- 30. Masgrau L, Roujeinikova A, Johannissen LO,
riers in enzymes. Angew Chem Int Ed Hothi P, Basran J, Ranaghan KE, Mulholland
45:6856–6859 AJ, Sutcliffe MJ, Scrutton NS, Leys D (2006)
18. Mulholland AJ (2007) Chemical accuracy in Atomic description of an enzyme reaction
QM/MM calculations on enzyme-catalysed dominated by proton tunneling. Science
reactions. Chem Cent J 1:19 312:237–241
19. Braun-Sand S, Olsson MHM, Warshel A 31. Hatcher E, Soudackov AV, Hammes-Schiffer
(2005) Computer modeling of enzyme catal- S (2007) Proton-coupled electron transfer
ysis and its relationship to concepts in physical in soybean lipoxygenase: dynamical behavior
organic chemistry. Adv Phys Org Chem and temperature dependence of kinetic
40:201–245 isotope effects. J Am Chem Soc
129:187–196
20. Vocadlo DJ, Davies GJ, Laine R, Withers SG
(2001) Catalysis by hen egg white lysozyme 32. Limbach HH, Lopez JM, Kohen A (2006)
proceeds via a covalent intermediate. Nature Arrhenius curves of hydrogen transfers: tun-
412:3835–3838 nel effects, isotope effects and effects of pre-
equilibria. Phil Trans Roy Soc B
21. Bowman AL, Grant IM, Mulholland AJ 361:1399–1415
(2008) QM/MM simulations predict a cova-
lent intermediate in the hen egg white lyso- 33. Nagel ZD, Klinman JP (2006) Tunneling and
zyme reaction with its natural substrate. dynamics in enzymatic hydride transfer. Chem
Chem Commun 7:4425–4427 Rev 106:3095–3118
 4 Computational Enzymology 85

34. Villa J, Strajbl M, Glennon TM, Sham YY, noprotein systems. Phil Trans Roy Soc B
Chu ZT, Warshel A (2000) How important 361:1375–1386
are entropic contributions to enzyme cataly- 47. Ranaghan KE, Mulholland AJ (2010) Com-
sis? Proc Natl Acad Sci USA 97:11899–11904 puter simulations of quantum tunnelling in
35. Lonsdale R, Harvey JN, Manby FR, Mulhol- enzyme-catalysed hydrogen transfer reac-
land AJ (2011) Comment on “A stationary- tions. Interdiscip Sci 2:78–97
wave model of enzyme catalysis” by Carlo 48. Pentik€ainen U, Pentik€ainen OT, Mulholland
Canepa. J Comput Chem 32:368–369 AJ (2008) Cooperative symmetric to asym-
36. Ridder L, Harvey JN, Rietjens IMCM, Ver- metric conformational transition of the apo-
voort J, Mulholland AJ (2003) Ab initio form of scavenger decapping enzyme revealed
QM/MM modeling of the hydroxylation by simulations. Proteins 70:498–508
step in p-hydroxybenzoate hydroxylase. 49. Lodola A, Mor M, Zurek J, Tarzia G, Piomelli
J Phys Chem B 107:2118–2126 D, Harvey JN, Mulholland AJ (2007)
37. Ridder L, Mulholland AJ, Rietjens IMCM, Conformational effects in enzyme catalysis:
Vervoort J (2000) A quantum mechanical/ reaction via a high energy conformation in
molecular mechanical study of the hydroxyl- fatty acid amide hydrolase. Biophys J 92:
ation of phenol and halogenated derivatives L20–L22
by phenol hydroxylase. J Am Chem Soc 50. Case DA, Cheatham TE III, Darden T,
122:8728–8738 Gohlke H, Luo R, Merz KM Jr, Onufriev A,
38. Bjelic S, Åqvist J (2004) Prediction of struc- Simmerling C, Wang B, Woods R (2005) The
ture, substrate binding mode, mechanism, AMBER biomolecular simulation programs.
and rate for a malaria protease with a novel J Comput Chem 26:1668–1688 (see http://
type of active site. Biochemistry amber.scripps.edu/)
43:14521–14528 51. Brooks BR, Bruccoleri RE, Olafson BD,
39. Zhang YK, Kua J, McCammon JA (2003) States DJ, Swaminathan S, Karplus M
Influence of structural fluctuation on enzyme (1983) CHARMM: a program for macromo-
reaction energy barriers in combined quan- lecular energy, minimization, and dynamics
tum mechanical/molecular mechanical stud- calculations. J Comput Chem 4:187–217,
ies. J Phys Chem B 107:4459–4463 (See also http://www.charmm.org)
40. Gao JL, Truhlar DG (2002) Quantum 52. Scott WRP, Hunenberger PH, Tironi IG,
mechanical methods for enzyme kinetics. Mark AE, Billeter SR, Fennen J, Torda AE,
Annu Rev Phys Chem 53:467–505 Huber T, Kruger P, van Gunsteren WF
41. Karplus M, Gao YQ, Ma JP, van der Vaart A, (1999) The GROMOS biomolecular simula-
Yang W (2005) Protein structural transitions tion program package. J Phys Chem A
and their functional role. Phil Trans R Soc A 103:3596–3607, (see http://www.igc.ethz.
363:331–355 ch/gromos/gromos.html)
42. Fersht A (1999) Structure and mechanism in 53. Phillips JC, Braun R, Wang W, Gumbart J,
protein science. A guide to enzyme catalysis Tajkhorshid E, Villa E, Chipot C, Skeel RD,
and protein folding. Freeman, New York Kale L, Schulten K (2005) Scalable molecular
43. Olsson MHM, Warshel A (2004) Solute sol- dynamics with NAMD. J Comput Chem
vent dynamics and energetics in enzyme catal- 26:1781–1802, (see http://www.ks.uiuc.
ysis: the S(N)2 reaction of dehalogenase as a edu/Research/namd/)
general benchmark. J Am Chem Soc 54. Ponder JW, Richards FM (1987) An efficient
126:15167–15179 newton-likemethod for molecular mechanics
44. Wolf-Watz M, Thai V, Henzler-Wildman K, energy minimization of large molecules.
Hadjipaylou G, Eisenmesser EZ, Kern D J Comput Chem 8:1016–1024, (see http://
(2004) Linkage between dynamics and catal- dasher.wustl.edu/tinker/)
ysis in a thermophilic-mesophilic enzyme pair. 55. Jorgensen WL, Maxwell DS, Tirado-Rives J
Nat Struct Mol Biol 11:945–949 (1996) Development and testing of the OPLS
45. Kohen A, Cannio R, Bartolucci S, Klinman JP all-atom force field on conformational ener-
(1999) Enzyme dynamics and hydrogen tun- getics and properties of organic liquids. J Am
neling in a thermophilic alcohol dehydroge- Chem Soc 118:11225–11236
nase. Nature 399:496–499 56. Kaminski GA, Friesner RA, Tirado-Rives J,
46. Sutcliffe MJ, Masgrau L, Roujeinikova A, Johan- Jorgensen WL (2001) Evaluation and repara-
nissen LO, Hothi P, Basran J, Ranaghan KE, metrization of the OPLSAA force field for
Mulholland AJ, Leys D, Scrutton NS (2006) proteins via comparison with accurate quan-
Hydrogen tunnelling in enzyme-catalysed tum chemical calculations on peptides. J Phys
H-transfer reactions: flavoprotein and qui- Chem B 105:6474–6487
86 A. Lodola and A.J. Mulholland

57. MacKerell AD Jr, Bashford D, Bellott M, the solvation of bovine pancreatic trypsin
Dunbrack RL Jr, Evanseck JD, Field MJ, inhibitor in explicit water: a comparative
Fischer S, Gao J, Guo H, Ha S, Joseph- study of the effects of solvent and protein
McCarthy D, Kuchnir L, Kuczera K, Lau polarizability. J Phys Chem B
FTK, Mattos C, Michnick S, Ngo T, Nguyen 109:16529–16538
DT, Prodhom B, Reiher WE III, Roux B, 69. Shaw KE, Woods CJ, Mulholland AJ (2010)
Schlenkrich M, Smith JC, Stote R, Straub J, Compatibility of quantum chemical methods
Watanabe M, Wiórkiewicz-Kuczera J, Yin D, and empirical (MM) water models in quan-
Karplus M (1998) All-atom empirical poten- tum mechanics/molecular mechanics liquid
tial for molecular modeling and dynamics water simulations. J Phys Chem Lett
studies of proteins. J Phys Chem B 1:219–223
102:3586–3616 70. Lim D, Jenson J, Repasky MP, Jorgensen WL
58. Cornell WD, Cieplak P, Bayly CI, Gould IR, (1999) Solvent as catalyst: computational
Merz KM, Ferguson DM, Spellmeyer DC, studies of organic reactions in solution. In:
Fox T, Caldwell JW, Kollman PA (1995) A Truhlar DG, Morokuma K (eds) Transition
2nd generation force-field for the simulation state modeling for catalysis. American Chemi-
of proteins, nucleic-acids, and organic mole- cal Society, Washington DC
cules. J Am Chem Soc 117:5179–5197 71. Bentzien J, Muller RP, Florian J, Warshel A
59. Mackerell AD (2005) Empirical force fields (1998) Hybrid ab initio quantum mechanics
for proteins: current status and future direc- molecular mechanics calculations of free
tions. Ann Rep Comp Chem 1:91–111 energy surfaces for enzymatic reactions: the
60. Ponder JW, Case DA (2003) Force fields for nucleophilic attack in subtilisin. J Phys Chem
protein simulations. Adv Protein Chem B 102:2293–2301
66:27–75 72. Warshel A (1997) Computer modeling of
61. Cheatham TE III (2005) Molecular modeling chemical reactions in enzymes and solutions.
and atomistic simulation of nucleic acids. Ann John Wiley & Sons, New York
Rep Comp Chem 1:75–90 73. Villa J, Warshel A (2001) Energetics and
62. Cheatham TE III (2004) Simulation and dynamics of enzymatic reactions. J Phys
modeling of nucleic acid structure, dynamics Chem B 105:7887–7907
and interactions. Curr Opin Struct Biol 74. Warshel A, Sharma PK, Chu ZT, Åqvist J
14:360–367 (2007) Electrostatic contributions to binding
63. Foloppe N, MacKerell AD (2000) All-atom of transition state analogues can be different
empirical force field for nucleic acids: I param- from the corresponding contributions to
eter optimization based on small molecule catalysis: phenolates binding to the oxyanion
and condensed phase macromolecular target hole of ketosteroid isomerase. Biochemistry
data. J Comput Chem 21:86–104 46:1466–1476
64. MacKerell AD, Banavali NK (2000) All-atom 75. Bjelic S, Åqvist J (2006) Catalysis and linear
empirical force field for nucleic acids: II appli- free energy relationships in aspartic proteases.
cation to molecular dynamics simulations of Biochemistry 45:7709–7723
DNA and RNA in solution. J Comput Chem 76. Trobro S, Åqvist J (2005) Mechanism of pep-
21:105–120 tide bond synthesis on the ribosome. Proc
65. Cheatham TE, Cieplak P, Kollman PA (1999) Natl Acad Sci USA 102:12395–12400
A modified version of the Cornell et al. force 77. Sharma PK, Xiang Y, Katom M, Warshel A
field with improved sugar pucker phases and (2005) What are the roles of substrate-assisted
helical repeat. J Biomol Struct Dyn catalysis and proximity effects in peptide bond
16:845–862 formation by the ribosome? Biochemistry
66. Feller SE, Yin DX, Pastor RW, MacKerell AD 44:11307–11314
(1997) Molecular dynamics simulation of 78. Hammes-Schiffer S (2004) Quantum-
unsaturated lipid bilayers at low hydration: classical simulation methods for hydrogen
parameterization and comparison with dif- transfer in enzymes: a case study of dihydro-
fraction studies. Biophys J 73:2269–2279 folate reductase. Curr Opin Struct Biol
67. Jorgensen WL, Tirado-Rives J (2005) Potential 14:192–201
energy functions for atomic-level simulations of 79. Liu HB, Warshel A (2007) Origin of the
water and organic and biomolecular systems. temperature dependence of isotope effects in
Proc Natl Acad Sci USA 102:6665–6670 enzymatic reactions: the case of dihydrofolate
68. Kim BC, Young T, Harder E, Friesner RA, reductase. J Phys Chem B 111:7852–7861
Berne BJ (2005) Structure and dynamics of
 4 Computational Enzymology 87

80. Sharma PK, Chu ZT, Olsson MHM, Warshel 92. Prat-Resina X, Bofill JM, Gonzalez-Lafont A,
A (2007) A new paradigm for electrostatic Lluch JM (2004) Geometry optimization and
catalysis of radical reactions in vitamin B12 transition state search in enzymes: different
enzymes. Proc Natl Acad Sci USA 2007 options in the microiterative method. Int J
(104):9661–9666 Quantum Chem 98:367–377
81. Van derVaart A, Gogonea V, Dixon SL, Merz 93. Ridder L, Rietjens IMCM, Vervoort J, Mul-
KM (2000) Linear scaling molecular orbital holland AJ (2002) Quantum mechanical/
calculations of biological systems using the molecular mechanical free energy simulations
semiempirical divide and conquer method. of the glutathione S-transferase (M1-1) reac-
J Comput Chem 21:1494–1504 tion with phenanthrene 9,10-oxide. J Am
82. Khandogin J, York DM (2004) Quantum Chem Soc 124:9926–9936
descriptors for biological macromolecules 94. Ranaghan KE, Ridder L, Szefczyk B, Sokalski
from linear-scaling electronic structure meth- WA, Hermann JC, Mulholland AJ (2004)
ods. Proteins 56:724–737 Transition state stabilization and substrate
83. Khandogin J, Musier-Forsyth K, York DM strain in enzyme catalysis: ab initio QM/
(2003) Insights into the regioselectivity and MM modelling of the chorismate mutase
RNA-binding affinity of HIV-1 nucleocapsid reaction. Org Biomol Chem 2:968–980
protein from linear-scaling quantum meth- 95. Lodola A, Mor M, Hermann JC, Tarzia G,
ods. J Mol Biol 330:993–1004 Piomelli D, Mulholland AJ (2005) QM/MM
84. Lonsdale R, Harvey JN, Mulholland AJ modelling of oleamide hydrolysis in fatty acid
(2010) Inclusion of dispersion effects signifi- amide hydrolase (FAAH) reveals a new mech-
cantly improves accuracy of calculated reac- anism of nucleophile activation. Chem Com-
tion barriers for cytochrome p450 catalyzed mun 35:4399–4401
reactions. J Phys Chem Lett 1:3232–3237 96. Riccardi D, Schaefer P, Cui Q (2005) pKa
85. Himo F, Siegbahn PE (2003) Quantum calculations in solution and proteins with
chemical studies of radical-containing QM/MM free energy perturbation simula-
enzymes. Chem Rev 103:2421–2456 tions: a quantitative test of QM/MM proto-
86. Siegbahn PE, Himo F (2009) Recent devel- cols. J Phys Chem B 109:17715–17733
opments of the quantum chemical cluster 97. Raha K, Merz KM (2004) A quantum
approach for modeling enzyme reactions. mechanics-based scoring function: study of
J Biol Inorg Chem 14:643–651 zinc ion-mediated ligand binding. J Am
87. Woodcock HL, Hodoscek M, Sherwood P, Chem Soc 126:1020–1021
Lee YS, Schaefer HF, Brooks BR (2003) 98. Bakowies D, Thiel W (1996) Hybrid models
Exploring the quantum mechanical/molecu- for combined quantum mechanical and
lar mechanical replica path method: a pathway molecular mechanical approaches. J Phys
optimization of the chorismate to prephenate Chem 100:10580–10594
Claisen rearrangement catalyzed by choris- 99. Svensson M, Humbel S, Froese RDJ, Matsu-
mate mutase. Theor Chem Acc 109:140–148 bara T, Sieber S, Morokuma K (1996)
88. Field MJ, Bash PA, Karplus M (1990) A com- ONIOM: A multilayered Integrated MO +
bined quantum-mechanical and molecular MM Method for geometry optimizations
mechanical potential for molecular-dynamics and single point energy predictions. A test
simulations. J Comput Chem 11:700–733 for Diels Alder reactions and Pt(P(t-
89. Lyne PD, Hodoscek M, Karplus M (1999) A Bu)3)2 + H2 oxidative addition. J Phys
hybrid QM–MM potential employing Har- Chem 100:19357–19363
tree–Fock or density functional methods in 100. Antes I, Thiel W (1999) Adjusted connection
the quantum region. J Phys Chem A atoms for combined quantum mechanical and
103:3462–3471 molecular mechanical methods. J Phys Chem
90. Cui Q, Elstner M, Kaxiras E, Frauenheim T, A 103:9290–9295
Karplus M (2001) A QM/MM implementa- 101. Jensen L, van Duijnen PT (2005) The first
tion of the self-consistent charge density func- hyperpolarizability of p-nitroaniline in 1,4-
tional tight binding (SCC-DFTB) method. dioxane: a quantum mechanical/molecular
J Phys Chem B 105:569–585 mechanics study. J Chem Phys 123:Art. No.
91. Marti S, Moliner V (2005) Improving the 074307
QM/MM description of chemical processes: 102. Cui Q, Karplus M (2002) Quantum mechan-
a dual level strategy to explore the potential ics/molecular mechanics studies of triosepho-
energy surface in very large systems. J Chem sphate isomerase-catalyzed reactions: effect of
Theory Comput 1:1008–1016 geometry and tunneling on proton-transfer
88 A. Lodola and A.J. Mulholland

rate constants. J Am Chem Soc culations of solvated molecules. J Phys Chem

124:3093–3124 97:8050–8053
103. Riccardi D, Li GH, Cui Q (2004) Importance 116. Pu JZ, Gao JL, Truhlar DG (2004)
of van der Waals interactions in QM/MM Generalized hybrid orbital (GHO) method
simulations. J Phys Chem B 108:6467–6478 for combining ab initio hartree—fock wave
104. Poulsen TD, Garcia-Viloca M, Gao JL, Truh- functions with molecular mechanics. J Phys
lar DG (2003) Free energy surface, reaction Chem A 108:632–650
paths, and kinetic isotope effect of short- 117. Pu JZ, Gao JL, Truhlar DG (2004) Combin-
chain acyl-coa dehydrogenase. J Phys Chem ing self-consistent-charge density-functional
B 107:9567–9578 tight-binding (SCC-DFTB) with molecular
105. Brooks CL III, Karplus M, Pettitt BM (1988) mechanics by the generalized hybrid orbital
Proteins, a theoretical perspective of dynam- (GHO) Method. J Phys Chem A
ics, structure and thermodynamics. Wiley, 108:5454–5463
New York 118. Pu JZ, Gao JL, Truhlar DG (2005)
106. Nam K, Gao JL, York DM (2005) An efficient Generalized hybrid-orbital method for
linear-scaling ewald method for long-range combining density functional theory with
electrostatic interactions in combined QM/ molecular mechanicals. Chemphyschem
MM calculations. J Chem Theory Comput 6:1853–1865
1:2–13 119. Amara P, Field MJ (2003) Evaluation of an
107. Schaefer P, Riccardi D, Cui Q (2005) Reliable ab-initio quantum mechanical molecular
treatment of electrostatics in combined QM/ mechanical hybrid-potential link-atom
MM simulation of macromolecules. J Chem method. Theor Chem Acc 109:43–52
Phys 123:Art. No. 014905 120. HyperChem Users Manual (2002) Hyper-
108. Im W, Berneche S, Roux B (2001) Cube, Inc: Waterloo, Ontario, Canada
Generalized solvent boundary potentials for 121. Reuter N, Dejaegere A, Maigret B, Karplus M
computer simulations. J Chem Phys (2000) Frontier bonds in QM/MM meth-
114:2924–2937 ods: a comparison of different approaches.
109. Monard G, Loos M, Théry V, Baka K, Rivail J Phys Chem A 104:1720–1735
J-L (1996) Hybrid classical quantum force 122. Zhang Y, Lee T-S, Yang W (1999) A pseudo-
field for modelling very large molecules. Int bond approach to combining quantum
J Quantum Chem 58:153–159 mechanical and molecular mechanical meth-
110. Assfeld X, Rivail J-L (1996) Quantum chemi- ods. J Chem Phys 110:46–54
cal computations on parts of large molecules: 123. Monard G, Prat-Resina X, Gonzalez-Lafont
The ab initio local self consistent field A, Lluch JM (2003) Determination of
method. Chem Phys Lett 263:100–106 enzymatic reaction pathways using QM/
111. Gao J, Amara P, Alhambra C, Field MJ MM methods. Int J Quant Chem
(1998) Method for the treatment of bound- 93:229–244
ary atoms in combined QM/MM calcula- 124. Das D, Eurenius KP, Billings EM, Sherwood
tions. J Phys Chem A 102:4714–4721 P, Chatfield DC, Hodoscek M, Brooks BR
112. Ferre N, Assfeld X, Rivail J-L (2002) Specific (2002) Optimization of quantum mechanical
force field parameters determination for the molecular mechanical partitioning schemes:
hybrid ab initio QM/MM LSCF method. Gaussian delocalization of molecular mechan-
J Comput Chem 23:610–624 ical charges and the double link atom method.
113. Antonczak S, Monard G, Ruiz-Lopez MF, J Chem Phys 117:10534–10547
Rivail J-L (1998) Modeling of Peptide 125. Konig PH, Hoffmann M, Frauenheim T, Cui
Hydrolysis by Thermolysin. A Semiempirical Q (2005) A critical evaluation of different
and QM/MM Study. J Am Chem Soc QM/MM frontier treatments with SCC-
120:8825–8833 DFTB as the QM method. J Phys Chem B
114. Garcia-Viloca M, Gao JL (2004) Generalized 109:9082–9095
hybrid orbital for the treatment of boundary 126. Klahn M, Braun-Sand S, Rosta E, Warshel A
atoms in combined quantum mechanical and (2005) On possible pitfalls in ab initio quan-
molecular mechanical calculations using the tum mechanics/molecular mechanics mini-
semiempirical parameterized model 3 mization approaches for studies of enzymatic
method. Theor Chem Acc 111:280–286 reactions. J Phys Chem B 109:15645–15650
115. Wesolowski TA, Warshel A (1993) Frozen 127. Lodola A, Sirirak J, Fey N, Rivara S, Mor M,
density functional approach for ab-initio cal- Mulholland AJ (2010) Structural fluctuations
in enzyme-catalyzed reactions: determinants
 4 Computational Enzymology 89

of reactivity in fatty acid amide hydrolase from free energies. J Chem Phys 128:Art. No.
multivariate statistical analysis of quantum 014109
mechanics/molecular mechanics paths. 131. Ridder L, Mulholland AJ (2003) Modeling
J Chem Theor Comput 6:2948–2960 biotransformation reactions by combined
128. Bowman AL, Ridder L, Rietjens IMCM, Ver- quantum mechanical/molecular mechanical
voort J, Mulholland AJ (2007) Molecular approaches: from structure to activity. Curr
determinants of xenobiotic metabolism: Top Med Chem 3:1241–1256
QM/MM simulation of the conversion of 1- 132. van der Kamp MW, Zurek J, Manby FR, Har-
chloro-2,4-dinitrobenzene catalyzed by M1- vey JN, Mulholland AJ (2010) Testing high-
1 glutathione S-transferase. Biochemistry level QM/MM methods for modeling
46:6353–6363 enzyme reactions: acetyl-CoA deprotonation
129. Acevedo O, Jorgensen WL (2010) Advances in citrate synthase. J Phys Chem B
in quantum and molecular mechanical (QM/ 114:11303–11314
MM) simulations for organic and enzymatic 133. Ranaghan KE, Masgrau L, Scrutton NS,
reactions. Acc Chem Res 43:142–151 Sutcliffe MJ, Mulholland AJ (2007) Analysis
130. Woods CJ, Manby FR, Mulholland AJ (2008) of classical and quantum paths for deprotona-
An efficient method for the calculation of tion of methylamine by methylamine dehy-
quantum mechanics/molecular mechanics drogenase. Chemphyschem 8:1816–1835
View publication stats