VU Research Portal

Application of free energy calculations for drug design

de Beer, S.B.A.

2012

document versionPublisher's PDF, also known as Version of record

Link to publication in VU Research Portal

citation for published version (APA)de Beer, S. B. A. (2012). Application of free energy calculations for drug design.

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ?

Take down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

E-mail address:vuresearchportal.ub@vu.nl

Download date: 15. Aug. 2021

CHAPTER 1 INTRODUCTION

Introduction

10

Chapter 1

11

1.1 Introduction

1.1.1 Computational drug discovery and design

In modern drug discovery, a detailed understanding of interactions between small chemical

compounds and biomolecular macromolecules (e.g. medicinal agents and their targets) is of

crucial importance.1 The search for drug-like compounds that selectively bind to a molecular

target and interfere with its receptor function or enzymatic activity demands a multi- and

interdisciplinary approach. Hereby, computer modeling serves as an important tool to

understand the relevant ligand-receptor or ligand-enzyme interactions. Nowadays, it is difficult

to imagine drug discovery without computation.2

Since a few decades, drug discovery relies greatly on computational efforts, to provide insights at

the atomic level that experiments typically fail to resolve and to make predictions that can aid

and guide experiments. Thus, computational methods and experimental observations usually

complement each other in an interdisciplinary setting. The rationalization of experimental

findings at an atomistic level can guide the design of active compounds. Furthermore,

assessments of the binding free energy (ΔG) offer useful insights in the ligand binding process.3

There is overall consensus in the scientific community that one of the biggest challenges in

computational drug design nowadays lies in the accurate and efficient calculation of binding free

energies.2,4

The binding free energy characterizes the equilibrium between the ligand in solution and bound

to its molecular target, e.g. a protein. Furthermore, the binding free energy depends on different

types of interactions, occurring upon ligand binding. In order to score ligands according to their

binding affinity, three actors should be taken into account: the ligand, the protein, and the

solvent in which both are solvated. Figure 1.1 schematically indicates how each of these three

actors may contribute to the binding free energy, which is the sum of all contributions. From the

perspective of the ligand, complexation with the protein should lead to favorable interactions

such as hydrogen bonds, charge-charge or pi-pi interactions, etc. These interactions will add

favorably to the enthalpy of binding. On the other hand, upon binding, the ligand loses

conformational, translational and rotational freedom, resulting in an unfavorable entropic

contribution to the binding process.

According to the conformational selection model, a protein can adopt an ensemble of different

conformations.5 Low-energy conformations are most likely and will be observed more often,

while higher-energy conformations will occur only rarely. However, ligands often bind to

unfavorable protein conformations. For tightly binding ligands, the favorable protein-ligand

enthalpy is typically sufficient to stabilize the protein when it is in a high-energy conformational

state. From the perspective of the protein, there is an unfavorable enthalpic contribution.

Introduction

12

Because the ligand now locks the protein in a given conformational state, its flexibility is also

reduced, which leads to an additional unfavorable entropic contribution.

Desolvation of the protein and ligand may lead to an unfavorable enthalpic contribution as

solute-solvent interactions are broken, but a large favorable entropic contribution can be

expected, due to the release of ordered water molecules.

Experimental techniques to directly assess the thermodynamics of ligand-receptor or ligand-

enzyme association, such as isothermal titration calorimetry, improved in recent years and can

provide thermodynamic details of the binding process.6 Parallel to the continuous increase in

computational power, several classes of computational methods have been developed, that can

be used to get a more detailed insight into the mode and affinity of (drug) compounds to their

(off-)target protein. These methods are all affiliated with a qualitative and/or quantitative

assessment of binding free energies, and differently trade off speed versus physical accuracy. With

the current wealth of available three-dimensional structures of proteins and their complexes with

ligands, structure-based drug design studies to identify key ligand interactions and free energy

calculations to quantify the thermodynamics of binding are topics of utmost interest.

Figure 1.1. Schematic representation of the contributions of the three main actors in ligand binding to the total

binding free energy. By interacting with the protein, the ligand usually yields a favorable enthalpic contribution, but

loses rotational and conformational freedom, leading to an unfavorable entropic contribution. Upon binding, the

protein may be locked in a high energy conformation leading to unfavorable enthalpic and entropic contributions.

Desolvation leads to an unfavorable enthalpic contribution, but to a favorable entropic contribution.

Chapter 1

13

The ultimate goal of both qualitative and quantitative approaches is to determine or predict the

mode of binding, the selectivity and the binding free energy that is associated with protein-ligand

interactions, using computational methods that efficiently assess the factors determining the

binding process, such as the specific interactions contributing to protein-ligand recognition.

Whether carried out in a qualitative or quantitative manner, binding free energy methods (and

the associated current challenges to their application) form the underlying common motive

throughout this thesis. The current chapter will give an overview of the various methods that

have been used, followed by a detailed discussion of methods to calculate binding free energies.

Finally, the various protein targets studied in this thesis will be introduced, and a short outline of

the thesis will be given.

1.1.2 Computational methods – an introduction

In this thesis a general distinction is made that divides the various computational methods into

two categories.

1) A structure-based, qualitative assessment of protein-ligand interactions governing the

binding process.

2) A quantitative assessment of the binding affinity of ligands for protein targets.

Note that this is a simplification and that overlap may occur between the two categories.

Figure 1.2 shows the contributions of representative computational methods (Docking,

Quantitative Structure-Activity Relationship (QSAR) and Free Energy Calculation) to

publications in the field of computational chemistry (panel A). Free energy methods are further

subdivided in panel B into: Free Energy Perturbation (FEP),7 Linear Interaction Energy theory

(LIE),8 Molecular-Mechanics Poisson Boltzmann/generalized Born solvent accessible surface

area methods (MM-PBSA),9 One-step Perturbation (OSP),10 and Thermodynamic Integration

(TI).11 As Figure 1.2 shows, clearly the computationally inexpensive docking methods prevail,

however, their number is relatively decreasing with respect to the more robust free energy

calculations, when comparing the time frames 1950-2008 to 2008-current (panel C).

QSAR and (automated) docking studies are the most commonly used virtual screening methods

in computational drug design.12 Whereas the former relates physicochemical properties of

compounds to their biological activity for datasets of potential receptor or enzyme agonists or

antagonists, the latter predicts their binding modes and scores their affinities.

Introduction

14

Figure 1.2. Relative numbers of publications that make use of different computational methods, based on the

numbers of citations in Google scholar (July 2012). Total numbers of publications per method (1950-July 2012) are:

Docking (384,000; restricting the search ‘Docking’ to ‘Molecular Docking’ yields 313,000 hits); QSAR (78,700); Free

energy calculations (15,000), depicted in Panel A. FEP (6860); LIE (1050); MM-PBSA (2270); OSP (213); TI (6430),

depicted in Panel B. Panel C represents relative publication numbers for Docking, QSAR and Free energy

calculations for two time frames, namely 1950-2008, and 2008-present.

Molecular Dynamics (MD) simulations provide detailed insight into molecular movements,

accounting for a greater number of relevant molecular conformations by virtue of more extensive

sampling, and, if carried out with an explicit representation of the solvent, allow for a better

description of solvation effects. MD simulations commonly form the basis for methods for free

energy calculation, such as FEP, LIE, OSP or TI, which quantify ligand-protein binding

affinities.

Section 1.1.3 describes the various approaches in more detail together with their implementation

in terms of currently available computational tools.

Figure 1.3 distinguishes five different levels of computational studies that are addressed in this

thesis. Note that the scheme presented in Figure 1.3 shows close resemblance to the well-known

screening funnel that is adopted in drug discovery processes, where, in the search of active lead

compounds, large databases of compounds are screened in an increasingly accurate and

increasingly time-consuming manner. Structure-based methods have been widely used in the

design and discovery of new ligands.13 These methods vary from approximate but higher-

throughput virtual screening methods to more robust and rigorous free energy calculations.12a

Chapter 1

15

Fig

ure

1.3

. Sc

hem

atic

rep

rese

ntat

ion

of f

ive

leve

ls o

f st

ruct

ure-

base

d co

mpu

tatio

nal

appr

oach

es u

sed

in t

his

thes

is t

oget

her

with

som

e of

the

ir l

imita

tions

.

Lis

ted

softw

are

pack

ages

(in

italic

s) a

re n

on-e

xhau

stiv

e ex

ampl

es. S

ee th

e te

xt fo

r re

fere

nces

to th

e va

riou

s m

etho

ds a

nd li

sted

sof

twar

e.

Introduction

16

Depending on the application (e.g., computational prediction of drug binding or rationalization of

substrate selectivity), a suitable method is chosen. As mentioned before, there may be overlap

between the levels depicted in Figure 1.3. A major aim of this thesis is to push several of the

methods forward to overcome the indicated limitations. In Chapter 2, protein flexibility is

accounted for in an (automated) docking procedure using a plasticity model based on MD

simulations. In Chapters 4 and 7, inefficient free energy calculations are carried out more

efficiently by using alternative approaches. In Chapter 6, the aspect of solvation effects is

thoroughly reviewed.

1.1.3 Five levels of computer-aided drug design

Figure 1.3 presents a ranking of structure-based computational approaches according to the

increased quantitative nature of the acquired information. In the following sections, the different

approaches that are used in the current thesis are discussed in turn, starting from the most

qualitative one. Level 1 Visualization of 3D structures

Structure-based drug design can already be carried out at the most simplistic level, using nothing

more than a protein target structure. Due to significant improvements in the fields of protein

crystallization, X-ray diffraction and nuclear magnetic resonance spectroscopy, an exponentially

increasing amount of protein structural data is becoming publicly available. While starting with

only seven structures in 1971, there were 83.106 protein structures available from the protein

databank (PDB)14 in June 2012. By looking at the geometrical features of a protein binding

pocket and examining chemical complementarity with a possible ligand, one can directly gain

structural insights into the target of biomolecular interest. For instance, using computer software

with basic visualization tools and a graphical interface (such as the PyMOL15 and Molecular

Operating Environment16 (MOE) graphic systems), one may obtain useful three-dimensional

insight into the protein active site and can thus suggest critical interactions between the ligand

and the protein. Getting acquainted with the active site topology is a crucial but sometimes

underestimated step, not only towards the design of new lead candidates, but also towards the

design of new enzymes with the aim of biocatalytically and selectively producing specific

metabolites.

In summary, by visualizing the three-dimensional structure of the target protein-ligand complex,

one can better understand the structure-activity relationship of active compounds from the

relevant interactions between ligand and protein, and suggest new analogs to be synthesized.

Level 2 Modeling of protein structures and mutations

The testing of hypotheses drawn from 3D-visualization requires modeling of (the flexibility of)

key amino acid residues and their interactions with the ligand, which can elucidate and guide

Chapter 1

17

further rationalization studies of the effect of critical mutations on substrate selectivity. When no

structure of the target of interest is available, computational modeling tools like homology

modeling or threading can be used to model a protein structure based on homologous proteins or

known protein folds.17 Furthermore, one can attempt to manually dock a novel ligand in a

protein binding cavity by e.g. superposition onto an already present substrate (e.g. the co-

crystallized substrate in the X-ray structure), followed by energy minimization to reduce steric

strain.

Similarly, when rationalizing the effect of mutations on the (catalytic) behavior of the protein, a

further step is to model these mutations into the active site, and study the effects they may have

on the binding of the substrate or ligand to the active site. However, as these kind of modeling

‘experiments’ usually do not include the appropriate physical interactions or active-site flexibility

associated with substrate recognition, further studies (experimental or computational) are

necessary to confirm the results.

More sophisticated methods include an explicit representation of electronic degrees of freedom,

i.e. rely on quantum-chemical calculations to determine the energy of the system. However these

methods are computationally expensive and limited to small ( <100 atoms) molecules in the gas

phase or in an implicitly-modeled solution environment (see e.g. reference 18).

Level 3 Docking

Using the modeling tools discussed above, a qualitative picture of the mode of ligand/substrate

binding and of the amino acids involved can be obtained. One step towards a more quantitative

assessment of binding affinities is the prediction of protein-ligand complex structures and their

associated affinities, using molecular docking studies, in which binding modes are predicted

together with a rough estimation of affinity (binding score),19 these are the two key features of

docking methods.20 Docking methods rely on known protein and ligand structures and aim to

rapidly generate optimal protein-ligand bound conformations to determine possible binding

modes of the ligand in the protein active site, and to evaluate a score that is related to their

relative ligand binding affinity. Since the amount of computation time spent per compound

typically lies in the orders of seconds or less, these approaches are particularly appealing in

virtual screening schemes employed in drug discovery to search for new lead compounds.21 In

these (pharmaceutical) lead discovery processes, docking is used to screen large libraries of (small)

organic molecules against certain receptor targets in a time-efficient manner. Overall, virtual

screening methods can significantly reduce the number of compounds to be tested in

experimental assays.22

In general, docking methods have been applied with considerable success in correctly identifying

bound conformations of ligand-protein complexes.20 The results of docking studies provide

information on the spatial orientation of a ligand in the protein active site and thus help to

rationalize experimentally observed trends. For example, the site of metabolism (SOM) may be

predicted from the orientation of a substrate in the enzyme active site,23 see Chapters 2 and 3. In

Introduction

18

this manner, the effect of a mutation in the active site can be studied in a more robust way (as

opposed to the previously described molecular modeling techniques) and can lead to the design

of new mutants of an enzyme with renewed and/or enhanced (biocatalytic) features. Ideally, the

different poses obtained from docking are adequately ranked by the implemented scoring

function, which relates and quantifies them in accordance to (experimentally determined)

binding free energies, and hence offers a way to discriminate between binding and nonbinding

ligands. However, despite of the large amount of available literature on docking (Figure 1.2),

studies have rarely been reported in which the scoring function quantitatively ranked ligands

according to their affinity or in which correlation with experimentally determined binding free

energies was observed, in the case of ligands not accounted for by the training set that was used

for parameterization.4b,24

This limitation may not be surprising, considering the fact that empirical scoring functions

implemented in docking tools only partially incorporate factors such as protein flexibility,

(de)solvation effects, long range electrostatic interactions, protonation, and entropic

contributions. The accuracy of the empirical scoring functions is a key to success in docking

studies.24a,25 Although methods are being developed to account for the lack of solvation and the

flexibility in traditional docking studies, these methods perform overall poorly outside their

training sets.26

For this reason, in Chapter 2, a combined MD/docking protocol is applied to a flexible receptor

to account for its protein plasticity, incorporating flexibility in otherwise rigid docking

experiments.

In conclusion, docking methods are trading off physical accuracy for speed. In this thesis,

docking approaches are not necessarily used for a quantitative assessment of protein-ligand

complexes, but rather serve to:

- Rationalize trends in enzyme metabolism observed in experiment (Chapter 2).

- Rationally design novel drug-metabolizing enzymes (Chapter 3).

- Provide initial structures of ligand-protein complexes for use in MD simulations

(Chapters 2 and 4).

Level 4 Molecular dynamics simulations

Even when the structure of a protein-ligand complex has been determined, the inclusion of

protein dynamics can be of vital importance to correctly describe and predict substrate

recognition and binding processes. This usually requires a more rigorous approach to model

prime interactions than the previously described docking protocols which remain essentially rigid

(although various attempts to include protein flexibility are described).27 Docking and scoring are

not sufficiently able to quantify ligand-protein affinities, due to the incomplete implementation of

solvation and protein flexibility. To include these factors, MD simulations are widely used in

biomedical sciences to obtain information on the time evolution and dynamics of proteins and

Chapter 1

19

protein-ligand complexes.28 MD samples conformational ensembles of structures accessible at the

given thermodynamic state point. Besides the inclusion of protein flexibility and a more accurate

description of the interaction energies via a carefully calibrated underlying Hamiltonian (force

field), molecular dynamics simulations also commonly incorporate solvation explicitly. Popular

software packages currently available to conduct MD simulations include AMBER,29

CHARMM,30 GROMACS31, GROMOS32 and NAMD.33 All MD simulations performed in this

thesis have been carried out using the GROMOS0534 and GROMOS1135 biomolecular

simulation packages.

MD can be used in combination with docking to give a better account of protein flexibility. A

recurring limitation of docking studies is that proteins and ligands can adopt multiple

conformations. Situations in which ligand binding leads to a change in protein conformation are

commonly referred to as induced fit.1 As mentioned in the previous section (describing Level 3),

docking algorithms typically treat the receptor as being rigid, considering only one receptor-

ligand complex structure and mobility of selected amino acids.36 However, some proteins are

known to adjust the conformation of the active site in order to accommodate a wide variety of

ligands. This issue is addressed in Chapter 2, in which a protein plasticity model is generated

based on a previously developed combined MD/docking protocol.37 In this protocol, ligands are

docked into multiple receptor conformations, generated using MD simulations.

As demonstrated in Chapters 2-4, the incorporation of protein dynamics can also benefit and

guide the understanding, rationalization and design of site-directed mutation studies, by studying

ligand binding to and behavioral differences between native and mutant protein on a molecular

level.38

To address protein-ligand binding using molecular dynamics simulations in a quantitative way,

Åqvist et al.8 developed the Linear Interaction Energy (LIE) method.8 In LIE, the binding free

energy is predicted in a semi-empirical way, i.e. assuming a linear dependence of the binding free

energy on the polar and nonpolar changes in ligand-surrounding energies from MD averages.

LIE is a so-called ‘end-point’ method, which relies on conformations of only the free and bound

ligand configuration and computes the binding free energy from differences in the corresponding

average ligand-surrounding energies.4b,39

Binding free energy calculation on the basis of LIE theory can be more accurate than when using

scoring functions implemented in docking tools, but it is less computationally expensive than Free

Energy Perturbation (FEP) or Thermodynamic Integration (TI), Section 1.2. Therefore, LIE has

high potential for application in structure-based drug design.24a,40 However, there still are

challenges to overcome, which will not be further discussed in this thesis. Elsewhere the LIE

method and modifications to the original approach, as well as current and future challenges in its

application to drug design have been extensively evaluated.3,4b,24a,39,41

Introduction

20

Level 5 Free energy calculations

Ultimately, in computational drug design one is interested in an accurate, efficient and robust

quantitative assessment of ligand-protein binding affinity.3 As mentioned previously, even though

structure-based design methods such as molecular docking routinely estimate binding affinities by

means of an empirical scoring function, there is overall consensus that they are too inaccurate to

reliably rank the obtained poses of multiple ligands.3,25b

The affinity of a ligand for a given target protein is a thermodynamic quantity and depending on

interaction energies and entropies. Therefore, the strength of MD-based free energy calculations

lies in the thorough inclusion of entropic contributions to the binding affinity, arising from

solvent effects, as well as protein and ligand flexibility. Note that these effects are generally not

accounted for by docking methods (Level 3). Free energy calculations, however, rely on statistical

mechanical principles (i.e. a physically more sound analog of a scoring function) and are correct

in the limit of infinite sampling (see Section 1.2).

In the methods described in Section 1.2, relative binding free energies are used to compute

differences in binding affinity between (similar) ligands towards a common receptor, and to

provide an accurate ranking of ligands.4c,39,42

During the 1980s, first advances were made to perform computer simulations of biomolecular

systems to calculate free energy changes.43 Early applications were mainly devoted to

development and calibration of methods using small data-sets.44 Nowadays, the focus has shifted

to further development of free energy calculations to render them more applicable for industrial

purposes and to guide drug discovery, design and lead optimization.4b,45 In the last decade,

several examples of (successful) cases of lead optimization guided by free energy calculations have

been reported, including the FEP-guided selection of HIV-1 reverse transcriptase inhibitors in

lead optimization.46 These success stories illustrate the current trend of the use of free energy

calculations as powerful research tools in computer-aided drug design and related fields such as

protein engineering.

MD simulations in combination with free energy calculation methods have been successfully

applied to quantify binding affinities.28c However, also in this case numerous challenges are still

to be overcome, of which several will be addressed in this thesis. In Chapters 4, 5 and 7, efforts

are made to achieve increased accuracy and higher efficiency levels when applying MD to

calculate binding free energies.

Chapter 1

21

1.2 Free energy calculations

1.2.1 State of the art

After establishing that free energy calculations give the most detailed and quantitative insight in

protein-ligand interactions among all five categories of computational tools addressed in Section

1.1, the current status of the usage of free energy calculation methods for affinity prediction will

be evaluated.45c,47 The theoretical background of free energy calculations is outlined in Section

1.2.2. Together with the continuous increase in computer power and advances in related areas of

statistical mechanics and enhanced sampling techniques, binding free energy calculations have

become useful tools in drug design and in the rationalization of biophysical experiments. This is

also reflected in the relative increase in the number of scientific research papers and reviews that

have been published over the last years on this topic, see Figure 1.2.

In structure-based drug design, free energy calculations are often applied in the context of a

thermodynamic cycle approach42a,43a,44c combining so-called alchemical transformations between

structurally related compounds. This has proven to be a successful tool to guide drug

development.4c,46a,48 Since it is virtually infeasible to run a molecular dynamics simulations long

enough to thoroughly capture the ligand-protein association/dissociation equilibrium,

calculation of absolute free energy differences, associated with ligand binding (ΔGbind) mostly

remains outside the range of computational chemistry.39 Alternatively, the absolute binding free

energy may also be calculated from alchemical approaches, disappearing a ligand from the

protein active site and from an aqueous solution.49 Note that the term absolute binding free

energy, commonly used in the field, still refers to free energy differences along the binding process.

However, in drug development the main interest is typically to determine affinities of a series of

potential drug candidates relative to each other. Therefore, the main focus usually lies on the

calculation of relative binding free energies (ΔΔGbind) between (series of) compounds.

The use of thermodynamic cycles involving alchemical transformations43a between two ligands

(L1 and L2) in order to calculate ΔΔGbind for the two ligands and a given protein target (P) in

aqueous solution is illustrated in Figure 1.4. Since the free energy is a thermodynamic state

function, it is a path-independent quantity, which means that the order of the binding event does

not matter: the computed free energy only depends on the representation of the initial (ligand in

solution) and final (ligand bound to protein) state of the binding process.

Introduction

22

Therefore, the free energy change along the cycle in Figure 1.4 amounts to zero, so that ΔΔGbind

can be expressed as:

ΔΔGbind = ΔG2 - ΔG1= ΔG4 - ΔG3 (Eq. 1.1)

Eq. (1.1) relates the free energy difference of the two horizontal branches (ΔG1 and ΔG2), which

indicate the individual affinities of the ligands for the protein, to the free energy difference of the

vertical branches (ΔG3 and ΔG4), which correspond to non-physical alchemical transformation of

L1 in L2, in the bound and free state, respectively. Use of thermodynamic cycles is a standard

approach to calculate relative binding free energies. However, note that the thermodynamic

cycle approach (and calculation of alchemical free energy differences) can also be applied to

calculate free energy changes of different types of (bio)chemical events other than ligand binding,

such as protein folding, solvation or conformational changes.44c In the light of this thesis,

thermodynamic cycles and alchemical transformations have solely been used in order to

calculate and predict relative ligand-protein binding free energies.

Figure 1.4. Standard thermodynamic cycle for relative binding free energy calculations. To compute the relative

binding free energy of two ligands (L1 in L2) to a given protein (P), L1 is alchemically mutated to L2, both in aqueous

solution and in the protein environment. According to Eq. (1.1), ΔΔGbind is derived by relating the difference

between ΔG1 and ΔG2 to the difference between ΔG3 and ΔG4.

Ultimately, the challenge lies in developing more robust and efficient free energy calculations to

reduce computational cost and thus make this approach feasible for large-scale industrial

applications.48c,50 As will be outlined in the following section, alchemical transformations require

extensive MD simulation times, to include all enthalpic and entropic effects in Figure 1.1.

Chapter 1

23

1.2.2 Theoretical background

In this thesis, relative binding free energies are computed on the basis of statistical principles,45b,51

to estimate differences in binding affinities between similar ligands towards a common receptor

(Figure 1.4 and Eq. 1.1).4c,39,42

The statistical mechanical definition of the Gibbs free energy, G, is based on the partition

function, Z, which is the sum of the Boltzmann weights of all energy levels of the system.42a,44c

€

G = −kBT lnZ(N, p,T) (Eq. 1.2)

In Eq. (1.2), kB is the Boltzmann constant, and (N,p,T) denotes a isothermal-isobaric ensemble at

constant number of particles (N), pressure (p) and temperature (T). In this ensemble, the partition

function is defined as:

€

Z(N, p,T) =1

Vh3NN!e−(H (r,p )+ pV ) / kBT∫∫∫ dpdrdV (Eq. 1.3)

where V is the system volume, h is Planck’s constant, H(r,p) is the Hamiltonian representing the

total energy of the system as a function of the coordinates r and momenta p of all particles in the

system (also collectively denoted as the ‘phase space’ of the system). The integral is performed

over all 3N positions r, momenta p and over the volume V. This is quite impractical, since

calculating the absolute free energy according to Eq. (1.2) requires an integration over all 3N

degrees of freedom. Even though the boundaries in computer power are being pushed, this type

of analysis of the partition function still cannot be reached.

However, if one focuses on the thermodynamic cycle depicted in Figure 1.4, the free energy

difference between related states A (L1 in free or bound state) and B (L2 in free or bound state)

corresponds to the alchemical transformations ΔG3 and ΔG4, which can be computed from MD

simulations by the use of the free energy perturbation (FEP) technique by Zwanzig,7

€

ΔGAB =GB −GA = −kBT ln e−(ΔH + pΔV ) / kBT

A (Eq. 1.4)

where ΔH = HB - HA is the difference of the Hamiltonians describing state A and B, and

€

A

denotes ensemble averaging over all the configurations generated in an MD simulation using the

Hamiltonian of state A.

The Zwanzig equation is correct in the limit of infinite sampling, and is only practically

applicable if the configurations sampled for state A are sufficiently relevant for state B. This

restricts its applicability to only small differences between states A and B, or requires introduction

of carefully chosen and possibly unphysical intermediate states, see for example Chapter 4 of this

thesis.

Introduction

24

A particular application of FEP is the one-step perturbation approach (OSP),10,52 applied in

Chapters 4, 5 and 7. OSP makes use of two MD simulations of an (unphysical) reference state, R,

from which free energy differences, ΔGRA, with respect to a set of real compounds can be

deduced, according to:

€

ΔGRA =GA −GR = −kBT ln e−(HA −HR ) / kBT

R (Eq. 1.5)

where HA and HR are the Hamiltonians of the system in state A (representing a real ligand) and

R, respectively. The free energy difference between states A and another real ligand B can then

be derived by using:

€

ΔGAB = ΔGRB − ΔGRA (Eq. 1.6)

Successful application of the OSP method requires the design of a reference molecule able to

sufficiently sample the phase space relevant for the real states, i.e. the Boltzmann weight of the

configurations sampled with Hamiltonian HR should not vanish in terms of the Hamiltonian

pertaining to the real compounds.

A popular and frequently used alternative to FEP is Thermodynamic Integration (TI).11 TI is a

free energy pathway method in which the initial state A is transformed into the final state B using

a coupling parameter λ which is introduced to describe the change of the Hamiltonian H(λ)

pertaining to the initial state A (at λ = 0) to the final state B (at λ = 1). The free energy difference

between the two end states (i.e. ΔG3 and ΔG4 in Figure 1.4) can then be obtained by (numerical)

integration along the λ-dependent path from λ =0 to λ =1 (via alchemical states for 0<λ<1),

over the ensemble average of the partial derivative of the Hamiltonian with respect to λ:

€

ΔGAB =dGdλ

dλ =0

1

∫ ∂H(λ)∂λ λ

dλ0

1

∫ (Eq. 1.7)

1.2.3 Challenges

Despite the sound theoretical basis of Eqs. (1.4-1.7), it occurs that free energies calculated using

the OSP or TI approach are in poor agreement with experimental data. Disagreement with

experiment can be due to inaccurate representation of the Hamiltonian or sampling issues. As a

result, thermodynamic cycles may fail to close, or the difference between forward and backward

calculations of ΔG3 and ΔG4 in Figure 1.4 does not vanish, but amounts to a large number

(hysteresis). The challenges that have to be overcome before the accurate prediction of binding

affinities through free energy calculation methods can efficiently and effectively be applied in

Chapter 1

25

drug discovery and industrial biotechnology processes can be classified into four

categories:42c,48d,53

Challenge 1 Modeling and simulation setup

Accurate and efficient applications of alchemical free energy calculations are most likely

to succeed for a set of substrates (e.g. enzyme inhibitors) that are structurally similar. In

Chapters 4, 5 and 7, special attention is paid to the setup of one-step perturbation

calculations to the choice of the (unphysical) reference state, which is optimized such that

sufficient phase-space overlap with the different compounds of interest is ensured.

Furthermore, accurate force field parameters are to be assigned to all three actors

depicted in Figure 1.1 (ligand, protein and solvent) present in the system, as well as to the

interactions between these actors. Especially the parameterization of the ligand can be a

time-consuming task, and might be one of the reasons why free energy calculations are

not (yet) widely applied in industry.

Challenge 2 Conformational Sampling

Both the protein and the ligand are treated as flexible entities in MD simulations and are

described by many conformational degrees of freedom. Therefore, exploring all

potentially relevant conformations requires a substantial computational effort,54

depending on the timescale of the MD simulations that are conducted. The results of free

energy calculations are particularly sensitive to insufficient conformational or spuriously

biased sampling, as entropic effects have to be accounted for via the integral in Eq. (1.3)

over all sampled conformations.55 Unsampled conformations (however crucial) will not

contribute to the total calculated free energy, which may lead to failure in

thermodynamic cycle closure and to an incorrect prediction of the binding affinity.

Failure in the closure of thermodynamic cycles may indicate inadequate sampling, and

internal consistency of the calculated data forms a crucial prerequisite for the reliability of

free energy calculations. Therefore, insufficient conformational sampling is an ongoing

challenge that is addressed by e.g. the development of enhanced sampling techniques and

faster computing algorithms as well as by continuous efforts to increase computer power.

Challenge 3 Force field accuracy

Molecular recognition involves a combination of inter- and intra-molecular interactions

that need to be modeled in a physically sound way.28c Empirical force fields such as

AMBER,56 CHARMM,57 GROMOS,58 and OPLS-AA59 can be used to calculate the

potential energy, U, by adding up the contributions from bonded and non-bonded terms

of the system. The sum of the potential energy and the kinetic energy, are referred to as

the total energy, or Hamiltonian (H in Eqs. 1.3-1.7) of the system. Its representation

directly influences the outcome of free energy calculations. Since inter-atomic forces are

Introduction

26

strong and short ranged, the accuracy of the employed energy function is strongly

dependent on the detailed representation of these terms. Advances are being made and

more sophisticated force fields are being developed, e.g. presenting more complex atomic

partial charge distributions and/or explicitly accounting for electronic polarization, in

order to achieve higher accuracy.28a In Chapter 5, the parameterization of functional

groups of the ligands is optimized in order to improve the correlation to experimentally

measured data.

In this thesis, the GROMOS force field is used, which is tailored for biomolecular

simulation and is therefore suitable for application on the systems described in Section

1.3.

Challenge 4 Analysis

When computing statistical quantities such as the change in binding free energy, proper

estimation of the statistical error is of vital importance to assess the reliability of the

employed method and accuracy of numerical post-processing. Only given a careful error

assessment, a robust and reliable prediction of binding, or comparison to experimental

data can be postulated. Recently, the efficiencies of various available methods to compute

absolute and relative free energy differences has been analyzed and compared.60

1.3 Protein targets and applications

In this thesis, the computational methods described in Sections 1.1 and 1.2 are applied to various

biomolecular targets. The current section introduces these targets and shortly describes the

addressed research questions.

1.3.1 Drug / substrate metabolism

The fate of a drug entering the body is crucially determined by drug-metabolizing enzymes.

Apart from of drug breakdown and facilitated excretion, metabolic transformations sometimes

give rise to the incidence of toxic effects.61

Cytochrome P450 enzymes (CYP450s) constitute a family of heme containing enzymes. They are

involved in endogenous processes, as well as the biotransformation of xenobiotics, such as

drugs.62 As such, they play an important role in the disposition of drugs and their

pharmacological and toxicological effects. Most members of the CYP450 family carry out mono-

oxygenation (Phase 1 drug metabolism) reactions, in which molecular oxygen is reduced by

NADPH-derived electrons to oxidize a substrate molecule, via insertion of an oxygen atom into a

substrate’s C-H bond while simultaneously forming a water molecule.63

Chapter 1

27

Complementary to experimental studies, many computational efforts have been performed in

order to predict the mode and affinity of drug binding (i.e. activity and substrate selectivity) to

CYP450s, in order to predict possible toxicity effects and to rationalize selective substrate binding

phenomena.23,64 Typically, computational prediction of enzyme activity and selectivity involves

substrate recognition, the site and rate of metabolism, the complete catalytic process, and the

ease of product release. In this thesis, the main focus lies on the prediction of sites of metabolism

(SOMs) only, by studying the orientation of the substrate towards the heme catalytic iron center,

limiting the region accessible to catalytic activity to within a (hypothetical) cut-off distance of 0.6

nm65 to predict whether hydroxylation by means of incorporation of an oxygen atom into a C-H

bond will occur (see Chapters 2 and 3 for more details).

In order to describe the actual metabolism and/or reactivity of the substrate species, one needs to

explicitly model electronic structure effects, e.g. through the use of Quantum Mechanical (QM)

methods.23 The accurate prediction of product formation in reactions catalyzed by CYP450s and

of the corresponding reaction barriers ultimately requires a description of the reactive heme iron-

oxygen species that is responsible for the chemistry of the CYP450 catalytic reaction cycle, the

kinetics of the reaction cycle, and the release of the product.18 The CYP450 catalytic reaction

cycle is schematically depicted in Figure 1.5.

In comparison to the empirical force-field models addressed in this thesis, QM methods are very

time-consuming, and typically only small molecules ( <100 atoms) can be described. Computer

time rapidly increases with an increase in system size, or upon inclusion of many different

conformations. Therefore QM methods are not included in Figure 1.3 amongst the methods

applied in this thesis for binding free energy computation, but they mostly are applied at Level 2

for a limited number of structures, based on advanced energy calculations.

Figure 1.5. Simplified CYP450 catalytic cycle, adapted from reference 64a. The substrate R-H is converted into

R-OH in this cycle.

Introduction

28

In Chapter 2, gas-phase QM calculations were performed to assess the intrinsic reactivity of C-H

bonds of testosterone. More sophisticated methods (compared to gas-phase QM calculations and

the methods presented in Figure 1.3) include reactivity-based approaches and are typically based

on inclusion of the heme moiety into the calculations,66 or on combined QM/MM calculations67

to include the full protein environment in the calculations.

1.3.2 Cytochrome P450 BM3

Due to their high catalytic activity and broad substrate specificity, CYP450s are interesting

targets in biotechnological research. They can serve as biocatalysts for the production of, e.g.,

human metabolites.68 CYP450 enzymes are also used as biocatalysists for industrial purposes, e.g.,

in the synthesis of fine chemicals and commercial products.68a

A recurring target in this thesis is a well-studied bacterial Cytochrome P450 from Bacillus

Megaterium, CYP450 BM3, also known as CYP102A1. Wild-type CYP450 BM3 is a fatty acid

hydroxylase, which shows one of the highest hydroxylation activities ever reported for a

CYP450.69 Although the natural CYP450 BM3 substrates are long-chain fatty acids, its substrate

specificity has been broadened by site-directed and random mutagenesis.70 Moreover, many of

the previously designed and identified biocatalytically active CYP450 BM3 mutants are potent

candidates for use in biotechnology, because they convert a variety of substrates into

therapeutically or diagnostically useful products and display a broad range of substrate specificity

as well as regio- and stereoselectivity.71 By employing genetic engineering techniques, these

enzymes can additionally be modified by random or site-directed mutagenesis to further improve

activity, stability, substrate specificity as well as regio- and stereoselectivity.72 In order to

rationalize results of such mutagenesis studies or to even guide them, in silico modeling has

proven to be a useful and synergetic tool, e.g., to predict and structurally rationalize the effect of

mutations.38,64c,73

Previous combined experimental and computational efforts in our molecular toxicology

laboratory71f have led to the design and elucidation of new drug metabolizing mutants of

BM338,74 (mutants M01 and M11) that can convert a variety of drug-like compounds such as 3,4-

methylenedioxy-methylamphetamine (MDMA) and dextromethorphan. Table 1.1 shows the

mutations with respect to the wild-type enzyme for the BM3 mutants studied in this thesis.

Mutants M01 A82W and M11 L437N were postulated based on computational modeling,75 and

their experimental characterization in the metabolism of testosterone and α-ionones is

rationalized using docking, molecular dynamics and free energy calculation in Chapters 2, 3 and

4.

Figure 1.6 shows the active site of wild-type BM3 as represented by PDB crystal structure

1BU7.77 Amino acid residues mutated in the considered BM3 mutants (Table 1.1) are

highlighted.

Chapter 1

29

Mutant: M0174 M01 A82W75a M01 A82W S72I76 M1174 M11 L437N75b

R47L R47L R47L R47L R47L

E64G E64G

S72I

F81I F81I

A82W A82W

F87V F87V F87V F87V F87V

E143G E143G

L188Q L188Q L188Q L188Q L188Q

E267V E267V E267V E267V E267V

G415Sa G415Sa G415Sa G415Sa G415Sa

L437N Table 1.1. CYP450 BM3 mutants studied in this thesis. aThe mutations have been shown not to contribute to the enhanced activities of the mutants.74 These mutations

were not included in the modeling studies.

Figure 1.6. Ribbon representation of the active site of wild-type CYP450 BM3 (PDB structure 1BU7).77 Mutated

amino acid residues and the catalytically active heme moiety are labeled and depicted in stick format. Residue G415

is not indicated since its location is far from the active site.

Introduction

30

1.3.3 4-Hydroxyphenylpyruvate dioxygenase (HPPD)

HPPD is a Fe(II)-dependent, non-heme oxygenase and catalyzes the conversion of 4-

hydroxyphenylpyruvate to homogentisate, i.e. one of the first steps in the tyrosine catabolic

pathway. This reaction is a chemically complex transformation, with many structural

modifications all occurring in a single catalytic cycle.78 HPPD is a relevant target protein for both

therapeutic and agrochemical research. Because HPPD is involved in tyrosine catabolism,

blocking the formation and accumulation of toxic catabolites by HPPD inhibition has proven a

successful strategy to treat type I tyrosinemia in mammals.78c Interestingly, an effective HPPD

inhibitor used to treat this inherited metabolic disorder in mammals was originally developed to

serve as a herbicidal agent.79

In plants, HPPD is a key enzyme in the pathway producing plasoquinone and tocopherol, which

are both essential cofactors in the photosynthesis cascade from homogentisate. By the

suppression of the formation of important cofactors, the photosynthesis route can be disrupted.

Inhibition of HPPD thus leads to bleaching, ultimately followed by necrosis and death.80 For

many years, HPPD has been a target of interest in the agrochemical industry, and many efforts

have been made in the screening and synthesis of inhibitors, which have led to a large number of

commercially available herbicides.28a,81 In Chapter 5, the computationally efficient one-step

perturbation approach is applied to a series of HPPD inhibitors, and compared to rigorous but

computationally less efficient TI calculations. By slightly adjusting force-field parameters of the

functional groups of the ligands, an accurate OSP model could be generated, based on which

predictions for new inhibitors could be made.

1.3.4 Oligopeptide binding protein A (OppA)

Water molecules can be of considerable importance for the binding and selectivity of a substrate

to its receptor,82 see Chapter 6. Numerous examples of water mediated hydrogen bonds between

protein and ligand are reported in the literature.83 The bacterial oligopeptide binding protein A

of Salmonella typhimurium (OppA) is a well-studied example for which water molecules have a

profound effect on ligand binding. OppA binds small peptides of 3-5 residues, regardless of their

amino acid sequence.84 Whereas other proteins need water molecules to establish high selectivity

in ligand binding, OppA relies on water molecules to accommodate a broad range of ligands

with diverse physico-chemical properties.85 This lack of specificity is due to the majority of

interactions between OppA and the peptide ligands being mediated by water, thus stabilizing e.g.

positive and negative charges or dipole moments of the ligand side chains. For instance, the

crystal structure of the charged tripeptide Lys-Glu-Lys (KEK) in complex with OppA (PDB code

1JEU),86 shows that the ligand is buried in the active site, and that most of the interactions

between KEK and OppA are mediated by nine water molecules. For different tripeptides,

different water configurations have been observed in the active site, as well as different numbers

of water molecules.84,86 The particular challenges associated with the simulation of highly flexible

Chapter 1

31

peptidic ligands, combined with the presence of water molecule networks in the active site pocket

are adressed in Chapter 7, in which thermodynamic cycles were constructed for three different

peptides binding to OppA.

1.4 Objectives and outline

The individual chapters of this thesis all describe different aspects of biomolecular recognition

and binding affinities, at different methodological levels and in the context of different molecular

targets. In every chapter, a specific research question is addressed and progress is made towards

the understanding of the molecular target and/or of the applied methodology. The overall aim

of the thesis can be summarized as demonstrating the breadth of computational methods to

describe molecular interactions of a variety of ligands with relevant and realistic protein targets.

In most cases, the methods are being pushed forward in an attempt to overcome current

methodological limitations described in Figure 1.3.

The thesis roughly follows the methods outlined in Figure 1.3. In Chapter 2, the intrinsic

reactivity of testosterone is evaluated using quantum mechanical calculations (Level 2) and

experimentally observed trends in its metabolism by Cytochrome P450 BM3 mutants are

compared to binding orientations obtained from docking experiments (Level 3), by making use of

a plasticity model based on Molecular Dynamics simulations (Level 4). In Chapter 3, new

CYP450 BM3 mutants are modeled and postulated (Levels 1 and 2) and experimentally observed

differences in product formation are rationalized using docking approaches (Level 3). Even

though these docking studies could offer qualitative explanations for the observed

stereospecificity in product formation, it is found in this chapter that docking methods can not

explain the experimentally observed trends in metabolism in a quantitative manner. Therefore,

the binding affinity of α-ionone substrates and products to two CYP450 BM3 mutants is

quantified in Chapter 4 (Levels 4 and 5), largely explaining the observed stereospecificity in α-

ionone hydroxylation. It is shown that the one-step perturbation approach can be more efficient

and more accurate than the more elaborate thermodynamic integration, through the elegant

design of a suitable reference state.

The use of the one-step perturbation approach (Levels 4 and 5) is explored further for larger

alchemical changes in Chapter 5, in which binding free energies are estimated for a series of

HPPD inhibitors. Again, OSP is compared to TI, allowing for conclusions concerning sampling

efficiency and force-field dependency of the calculated free energies. While Chapters 4 and 5

mostly focus on the effect of sampling to overcome the limitations of Level 3 methods, Chapter 6

turns the attention to another major limitation, namely the explicit inclusion of solvation effects.

In Chapter 6, the role of water molecules in computational drug design is reviewed. One of the

protein targets discussed in this chapter is OppA, which is the target of choice in Chapter 7.

Introduction

32

Lessons learned to enhance sampling in Chapter 4 are applied in TI calculations on relative

binding free energies of various tripeptides, and first attempts are described to capture the effect

of the complex water network on the binding affinity. This thesis is concluded with a summary

and outlook in Chapter 8.

References

(1) Mobley, D. L.; Dill, K. A. Binding of small-molecule ligands to proteins: "what you see" is not always "what you get". Structure 2009, 17, 489-498.

(2) Jorgensen, W. L. The many roles of computation in drug discovery. Science 2004, 303, 1813-1818.

(3) Singh, N.; Warshel, A. Absolute binding free energy calculations: on the accuracy of computational scoring of protein-ligand interactions. Proteins 2010, 78, 1705-1723.

(4) (a) Chipot, C.; Pearlman, D. A. Free energy calculations. The long and winding gilded road. Molecular Simulation 2002, 28, 1-12; (b) Foloppe, N.; Hubbard, R. Towards predictive ligand design with free-energy based computational methods? Curr. Med. Chem. 2006, 13, 3583-3608; (c) Michel, J.; Essex, J. W. Prediction of protein-ligand binding affinity by free energy simulations: assumptions, pitfalls and expectations. J. Comp. Aid. Mol. Des. 2010, 24, 639-658.

(5) Carlson, H. A. Protein flexibility and drug design: how to hit a moving target. Curr. Opin. Chem. Biol. 2002, 6, 447 - 452.

(6) Leavit, S.; Freire, E. Direct measurement of protein binding energetics by isothermal titration calorimetry. Curr. Opin. Stuc. Biol. 2001, 11, 560 - 566.

(7) Zwanzig, R. W. High-temperature equation of state by a perturbation method. I. Nonpolar gases. J. Chem. Phys. 1954, 22, 1420-1426.

(8) Åqvist, J.; Medina, C.; Samuelsson, J. E. A new method for predicting binding affinity in computer-aided drug design. Protein Eng. 1994, 7, 385-391.

(9) Kollman, P. A.; Massova, I.; Reyes, C.; Kuhn, B.; Huo, S. H.; Chong, L.; Lee, M.; Lee, T. S.; Duan, Y.; Wang, W.; Donini, O.; Cieplak, P.; Srinivasan, J.; Case, D. A.; Cheatham III, T. E. Calculating structures and free energies of complex molecules: Combining molecular mechanic and continuum models. Acc. Chem. Res. 2000, 33, 889-897.

(10) Liu, H.; Mark, A. E.; van Gunsteren, W. F. Estimating the relative free energy of different molecular states with respect to a single reference state. J. Phys. Chem. 1996, 100, 9485-9494.

(11) Kirkwood, J. G. Statistical mechanics of fluid mixtures. J. Chem. Phys. 1935, 3, 300-313.

(12) (a) Jorgensen, W. L. Efficient drug lead discovery and optimization. Accounts Chem. Res. 2009, 42, 724 - 733; (b) Moroy, G.; Martiny, V. Y.; Vayer, P.; Villoutreix, B. O.; Miteva, M. A. Toward in silico structure-based ADMET prediction in drug discovery. Drug Discov. Today 2012, 17, 44-55.

(13) Kuntz, I. D. Structure-based strategies for drug design and discovery. Science 1992, 257, 1078-1082.

(14) Berman, H. M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bhat, T. N.; Weissig, H.; Shindyalov, I. N.; Bourne, P. E. The protein data bank. Nucleic Acids Res. 2000, 28, 235-242.

Chapter 1

33

(15) DeLano, W. L. The PyMOL Molecular Graphics System, DeLano Scientific, San Carlos, CA, USA. http://www.pymol.org. 2002.

(16) Molecular Operating Environment (MOE), Montreal, Quebec, Canada: Chemical Computing Group Inc.

(17) Leach, A. R., Molecular Modelling. Principles and applications. 2nd ed.; Prentice Hall: Essex, England, 2001.

(18) Shaik, S.; Cohen, S.; Wang, Y.; Chen, H.; Kumar, D.; Thiel, W. P450 enzymes: their structure, reactivity, and selectivity-modeled by QM/MM calculations. Chem. Rev. 2010, 109, 949-1017.

(19) Huang, S.-H.; Grinter, S. Z.; Zou, X. Scoring functions and their evaluation methods for protein-ligand docking: recent advances and future directions. Phys. Chem. Chem. Phys. 2010, 12, 12899-12908.

(20) Warren, G. L.; Andrews, C. W.; Capelli, A.; Clarke, B.; Lalonde, J.; Lambert, M. H.; Lindvall, M.; Nevins, N.; Semus, S. F.; Senger, S.; Tedesco, G.; Wall, I. D.; Woolven, J. M.; Peishoff, C. E.; Head, M. S. A critical assessment of docking programs and scoring functions. J. Med. Chem. 2006, 49, 5912-5931.

(21) (a) Shoichet, B. K.; McGovern, S. L.; Wei, B.; Irwin, J. J. Lead discovery using molecular docking. Curr. Opin. Chem. Biol. 2002, 6, 439-446; (b) Ferreira, R. S.; Simeonov, A.; Jadhav, A.; Eidam, O.; Mott, B. T.; Keiser, M. J.; McKerrow, J. H.; Maloney, D. J.; Irwin, J. J.; Shoichet, B. K. Complementarity between a docking and a high-throughput screen in discovering new cruzain inhibitors. J. Med. Chem. 2010, 53, 4891-4905.

(22) (a) Stoichet, B. K. Virtual screening of chemical libraries. Nature 2004, 432, 862-865; (b) Klebe, G. Virtual ligand screening: strategies, perspectives and limitations. Drug Discov. Today 2006, 11, 580 - 594; (c) Huang, N.; Jacobson, M. P. Physics-based methods for studying protein-ligand interactions. Curr. Opin. Drug Discov. Devel. 2007, 10, 325-331; (d) Babaoglu, K.; Simeonov, A.; Irwin, J. J.; Nelson, M. E.; Feng, B.; Thomas, C. J.; Cancian, L.; Costi, M. P.; Maltby, D. A.; Jadhav, A.; Inglese, J.; Austin, C. P.; Shoichet, B. K. Comprehensive mechanistic analysis of hits from high-throughput and docking screens against beta-lactamase. J. Med. Chem. 2008, 51, 2502-2511; (e) Cheng, T.; Li, Q.; Zhou, Z.; Wang, Y.; Bryant, S. H. Structure-based virtual screening for drug discovery: a problem-centric review. The AAPS Journal 2012, 14, 133-141.

(23) Kirchmair, J.; Williamson, M. J.; Tyzack, J. D.; Tan, L.; Bond, P. J.; Bender, A.; Glen, R. C. Computational prediction of metabolism: sites, products, SAR, P450 enzyme dynamics, and mechanisms. J. Chem. Inf. Model. 2012, 52, 617-648.

(24) (a) Stjernschantz, E.; Marelius, J.; Medina, C.; Jacobsson, M.; Vermeulen, N. P. E.; Oostenbrink, C. Are automated molecular dynamics simulations and binding free energy calculations realistic tools in lead optimization? An evaluation of the Linear Interaction Energy (LIE) method. J. Chem. Inf. Model. 2006, 46, 1972-1983; (b) Sousa, S. F.; Fernandes, P. A.; Ramos, M. J. Protein-ligand docking: Current status and future challenges. Proteins 2006, 65, 15-26.

(25) (a) Gohlke, H.; Klebe, G. Statistical potentials and scoring functions applied to protein-ligand binding. Curr. Opin. Struct. Biol. 2001, 11, 231-235; (b) Leach, A. R.; Shoichet, B. K.; Peishoff, C. E. Predicition of protein-ligand interactions. Docking and scoring: successes and gaps. J. Med. Chem. 2006, 49 (5851-5855).

(26) (a) Graves, A. P.; Shivakumar, D. M.; Boyce, S. E.; Jacobson, M. P.; Case, D. A.; Shoichet, B. K. Rescoring docking hit lists for model cavity sites: predictions and experimental testing. J. Mol. Biol. 2008, 377, 914-934; (b) Gohlke, H.; Klebe, G. Approaches to the description and prediction of the binding affinity of small-molecule ligands to macromolecular receptors. Angew. Chem. Int. Ed. 2002, 41, 2644-2676.

(27) (a) Durrant, J. D.; McCammon, J. A. Computer-aided drug-discovery techniques that account for receptor flexibility. Curr. Opin. Pharmacol. 2010, 10, 770-774; (b) Lill, M. A. Efficient incorporation of protein flexibility and dynamics into molecular docking simulations. Biochemistry 2011, 50, 6157-6169.

(28) (a) Adcock, S. A.; McCammon, J. A. Molecular dynamics: survey of methods for simulating the activity of proteins. Chem. Rev. 2006, 106, 1589-1615; (b) Oostenbrink, C.; van Lipzig, M. M. H.; van Gunsteren, W. F., Applications of molecular dynamics simulations in drug design. In Comprehensive Medicinal Chemistry II, Taylor, J. B.;

Introduction

34

Triggle, D. J., Eds. Elsevier: Amsterdam, 2007; Vol. 4, pp 651-668; (c) Durrant, J. D.; McCammon, J. A. Molecular dynamics simulations and drug discovery. BMC Biol. 2011, 9, doi:10.1186/1741-7007-9-71.

(29) Case, D. A.; Cheatham III, T. E.; Darden, T.; Gohlke, H.; Luo, R.; Mertz, K. M. J.; Onufriev, A.; Simmerling, C.; Wang, B.; Woods, R. J. The Amber biomolecular simulation programs. J. Comput. Chem. 2005, 16, 1668-1688.

(30) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.; Swaminathan, S.; Karplus, M. CHARMM: A program for macromolecular energy, minimization, and dynamics calculations. J. Comp. Chem. 1983, 4, 187-217.

(31) (a) Berendsen, H. J. C.; van der Spoel, D.; van Drunen, R. GROMACS: A message-passing parallel molecular dynamics implementation. Comp. Phys. Commun. 1995, 91, 43-56; (b) van der Spoel, D.; Lindahl, E.; Hess, B.; van Buuren, A. R.; Apol, E.; Meulenhoff, P. J.; Tieleman, D. P.; Sijbers, A. L. T. M.; Feenstra, K. A.; van Drunen, R.; Berendsen, H. J. C. Gromacs User Manual version 4.5.4, www.gromacs.org. 2012.

(32) van Gunsteren, W. F.; Billeter, S. R.; Eising, A. A.; Hünenberger, P. H.; Krüger, P.; Mark, A. E.; Scott, W. R. P.; Tironi, I. G., Biomolecular simulation: The GROMOS96 manual and user guide. Vdf Hochschulverlag AG an der ETH Zürich: Zürich, 1996.

(33) Phillips, J. C.; Braun, R.; Wang, W.; Gumbart, J.; Tajkhorshid, E.; Villa, E.; Chipot, C.; Skeel, R. D.; Kalé, L.; Schulten, K. Scalable molecular dynamics with NAMD. J. Comp. Chem. 2005, 26, 1781-1802.

(34) Christen, M.; Hünenberger, P. H.; Bakowies, D.; Baron, R.; Bürgi, R.; Geerke, D.; Heinz, T. N.; Kastenholz, M. A.; Kräutler, V.; Oostenbrink, C.; Peter, C.; Trzesniak, D.; van Gunsteren, W. F. The GROMOS software for biomolecular simulation: GROMOS05. J. Comp. Chem. 2005, 26, 1719 - 1751.

(35) (a) Riniker, S.; Christ, C. D.; Hansen, H. S.; Huenenberger, P. H.; Oostenbrink, C.; Steiner, D.; van Gunsteren, W. F. Calculation of relative free energies for ligand-protein binding, solvation, and conformational transitions using the GROMOS software. J. Phys. Chem. B 2011, 115, 13570-13577; (b) Schmid, N.; Christ, C. D.; Christen, M.; Eichenberger, A. P.; van Gunsteren, W. F. Architecture, implementation and parallelisation of the GROMOS software for biomolecular simulation. Comp. Phys. Commun. 2012, 183, 890-903.

(36) Sotriffer, C. A. Accounting for induced-fit effects in docking: what is possible and what is not? Curr. Top. Med. Chem. 2011, 11, 179-191.

(37) (a) Wong, C. F.; Kua, J.; Zhang, Y.; Straatsma, T. P.; McCammon, J. A. Molecular docking of balanol to dynamics snapshots of protein kinase A. Proteins 2005, 61, 850-858; (b) Hritz, J.; de Ruiter, A.; Oostenbrink, C. Impact of plasticity and flexibility on docking results for Cytochrome P450 2D6: a combined approach of molecular dynamics and ligand docking. J. Med. Chem. 2008, 51, 7469–7477.

(38) Stjernschantz, E.; van Vugt-Lussenburg, B. M. A.; Bonifacio, A.; de Beer, S. B. A.; van der Zwan, G.; Gooijer, C.; Commandeur, J. N. M.; Vermeulen, N. P. E.; Oostenbrink, C. Structural rationalization of novel drug metabolizing mutants of cytochrome P450 BM3. Proteins 2008, 71 (1), 336-352.

(39) de Ruiter, A.; Oostenbrink, C. Free energy calculations of protein-ligand interactions. Curr. Opin. Chem. Biol. 2011, 15, 547-552.

(40) Brandsdal, B. O.; Österberg, F.; Almlöf, M.; Feierberg, I.; Luzhkov, V. B.; Åqvist, J. Free energy calculations and ligand binding. Adv. Protein Chem. 2003, 66, 123 - 158.

(41) Stjernschantz, E.; Oostenbrink, C. Improved ligand-protein binding affinity predicitons using multiple binding modes. Biophys. J. 2010, 98, 2682-2691.

(42) (a) Beveridge, D. L.; DiCapua, F. M. Free energy via molecular simulation: Applications to chemical and biomolecular systems. Annu. Rev. Biophys. Bio. 1989, 18, 431-492; (b) Gilson, M. K.; Zhou, H. X. Calculation of protein-ligand binding affinities. Annu. Rev. Biophys. Biomol. Struct. 2007, 36, 21-42; (c) Christ, C. D.; Mark, A. E.; van Gunsteren, W. F. Basic ingredients of free energy calculations: a review. J. Comp. Chem. 2010, 31, 1569-1582.

Chapter 1

35

(43) (a) Tembe, B. L.; McCammon, J. A. Ligand-receptor interactions. Computers & Chemistry 1984, 8, 281-283; (b) Wong, C. F.; McCammon, J. A. Dynamics and design of enzymes and inhibitors. J. Am. Chem. Soc. 1986, 108, 3830-3832; (c) Warshel, A.; Sussman, F.; King, G. Free energy changes in solvated proteins: microscopic calculations using a reversible charging process. Biochemistry 1986, 25, 8368-8372.

(44) (a) Gao, J.; Kuczera, K.; Tidor, B.; Karplus, M. Hidden thermodynamics of mutant proteins: a molecular dynamics analysis. Science 1989, 244, 1069-1072; (b) McCammon, J. A. Free energy from simulations. Curr. Opin. Struct. Biol. 1991, 1, 196-200; (c) Kollman, P. A. Free energy calculations: applications to chemical and biochemical phenomena. Chem. Rev. 1993, 93, 2395-2417.

(45) (a) Ajay, A.; Murcko, M. A. Computational methods to predict binding free energy in ligand-receptor complexes. J. Med. Chem. 1995, 38, 4953-4967; (b) Gilson, M. K.; Given, J. A.; Bush, B. L.; McCammon, J. A. The statistical-thermodynamic basis for computation of binding affinities: a critical review. Biophys. J. 1997, 72, 1047-1069; (c) Michel, J.; Foloppe, N.; Essex, J. W. Rigorous free energy calculations in structure-based drug design. Mol. Inf. 2010, 29, 570-578.

(46) (a) Kim, J. T.; Hamilton, A. D.; Bailey, C. M.; Domoal, R. A.; Wang, L.; Anderson, K. S.; Jorgensen, W. L. FEP-guided selection of bicyclic heterocycles in lead optimization for non-nucleoside inhibitors of HIV-1 reverse transcriptase. J. Am. Chem. Soc. 2006, 128, 15372-15373; (b) Jorgensen, W. L.; Ruiz-Caro, J.; Tirado-Rives, J.; Basavapathruni, A.; Anderson, K. S.; Hamilton, A. D. Computer-aided design of non-nucleoside inhibitors of HIV-1 reverse transcriptase. Bioorg. Med. Chem. Lett. 2006, 16, 663-667; (c) Zeevaart, J. G.; Wang, L.; Thakur, V. V.; Leung, C. S.; Tirado-Rives, J.; Bailey, C. M.; Domoal, R. A.; Anderson, K. S.; Jorgensen, W. L. Optimization of azoles as anti-human immunodeficiency virus agents guided by free-energy calculations. J. Am. Chem. Soc. 2008, 130, 9492-9499.

(47) Simonson, T.; Archontis, G.; Karplus, M. Free energy simulations come of age: protein-ligand recognition. Acc. Chem. Res. 2002, 35, 430-437.

(48) (a) Rami Reddy, M.; Erion, M. D. Calculation of relative binding free energy differences for fructose 1,6-bisphosphatase inhibitors using the thermodynamic cycle approach. J. Am. Chem. Soc. 2001, 123, 6246-6252; (b) Chipot, C.; Rozanska, X.; Dixit, S. B. Can free energy calculations be fast and accurate at the same time? Binding of low-affinity, non-peptide inhibitors to the SH2 domain of the src protein. J. Comput. Aid. Mol. Des. 2005, 19, 765-770; (c) Shirts, M. R.; Mobley, D. L.; Chodera, J. D. Alchemical Free Energy Computations: Ready for Prime Time? Annu. Rep. Comp. Chem. 2007, 3, 41-59; (d) Chodera, J. D.; Mobley, D. L.; Shirts, M. R.; Dixon, R. W.; Branson, K.; Pande, V. S. Alchemical free energy methods for drug discovery: progress and challenges. Curr. Opin. Struct. Biol. 2011, 21, 150-160; (e) Steinbrecher, T.; Case, D. A.; Labahn, A. Free energy calculations on the binding of novel thiolactomycin derivatives to E. coli fatty acid synthase I. Bioorg. Med. Chem. 2012, 20, 3446-3453.

(49) (a) Boresch, S.; Tettinger, F.; Leitgeb, M.; Karplus, M. Absolute binding free energies: a quantitative approach for their calculations. J. Phys. Chem. B 2003, 107, 9535-9551; (b) Woo, H.-J.; Roux, B. Calculation of absolute protein-ligand binding free energy from computer simulations. PNAS 2005, 102, 6825-6830.

(50) Schneider, G. Virtual screening: an endless staircase? Nat. Rev. Drug Discov. 2010, 9, 273-276.

(51) McCammon, J. A. Theory of biomolecular recognition. Curr. Opin. Struct. Biol. 1998, 8, 245-249.

(52) Schäfer, H.; van Gunsteren, W. F.; Mark, A. E. Estimating relative free energies from a single ensemble: Hydration free energies. J. Comp. Chem. 1999, 20, 1604-1617.

(53) Pohorille, A.; Jarzynski, C.; Chipot, C. Good practices in free-energy calculations. J. Phys. Chem. B 2010, 114, 10235-10253.

(54) Christen, M.; van Gunsteren, W. F. On searching in, sampling of, and dynamically moving through conformational space of biomolecular systems: A review. J. Comput. Chem. 2008, 29, 157-166.

(55) (a) Mobley, D. L. Binding mode sampling and its effect on relative free energy calculations. 2012; (b) Mobley, D. L. Let's get honest about sampling. J. Comp. Aid. Mol. Des. 2012, 26, 93-95.

Introduction

36

(56) Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Mertz, K. M.; Ferguson, D. M.; Spellmeyer, D. C.; Fox, T.; Caldwell, J. W.; Kollman, P. A. A second generation force field for the simulation of proteins, nucleic acids and organic molecules. J. Am. Chem. Soc. 1995, 117, 5179-5197.

(57) MacKerell Jr., A. D.; Bashford, D.; Bellot, M.; Dunbrack Jr., R. L.; Evanseck, J. D.; Field, M. J.; Fischer, S.; Gao, J.; Guo, H.; Ha, S.; Joseph-McCarthy, D.; Kuchnir, L.; Kuczera, K.; Lau, F. T. K.; Mattos, C.; Michnick, S.; Ngo, T.; Nguyen, D. T.; Prodhom, B.; Reiher III, W. E.; Roux, B.; Schlenkrich, M.; Smith, J. C., Stote, R.; Straub, J.; Watanabe, M.; Wiórkiewicz-Kuczera, J.; Yin, D.; Karplus, M. All-atom empirical potential for molecular modeling and dynamics studies of proteins. J. Phys. Chem. B. 1998, 102, 3586-3616.

(58) (a) Soares, T. A.; Huenenberger, P. H.; Kastenholz, M. A.; Kraütler, V.; Lenz, T.; Lins, R. D.; Oostenbrink, C.; van Gunsteren, W. F. An improved nucleic acid parameter set for the GROMOS force field. J. Comp. Chem. 2005, 26, 725-737; (b) Lins, R. D.; Hünenberger, P. H. A new GROMOS force field for hexopyranose-based carbohydrates. J. Comp. Chem. 2005, 26, 1400-1412.

(59) Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. Development and Testing of the OPLS All-Atom Force Field on Conformational Energetics and Properties of Organic Liquids. J. Am. Chem. Soc. 1996, 118, 11225-11236.

(60) (a) Bruckner, S.; Boresch, S. Efficiency of alchemical free energy simulations. I. A practical comparison of the exponential formula, thermodynamic integration, and Bennett's acceptance ratio method. J. Comp. Chem. 2011, 32, 1303-1319; (b) Bruckner, S.; Boresch, S. Efficiency of alchemical free energy simulations. II. Improvements for thermodynamic integration. J. Comp. Chem. 2011, 32, 1320-1333.

(61) Sheweita, S. A. Drug-metabolizing enzymes: mechanisms and functions. Curr. Drug Metab. 2000, 1, 107-132.

(62) Walsky, R. L.; Boldt, S. E. In vitro cytochrome P450 inhibition and induction. Curr. Drug Metab. 2008, 9, 928-939.

(63) Guengerich, F. P. Common and uncommon cytochrome P450 reactions related to metabolism and chemical toxicity. Chem. Res. Toxicol. 2001, 14, 611-650.

(64) (a) Stjernschantz, E.; Vermeulen, N. P. E.; Oostenbrink, C. Computational prediction of drug binding and rationalisation of selectivity towards cytochromes P450 Exp. Opin. Drug Metab. Toxicol. 2008, 4, 513-527; (b) Kumar Mishra, N. Computational modeling of P450s for toxicity prediction. Expert Opin. Drug Metab. Toxicol. 2011, 7, 1211-1231; (c) de Graaf, C.; Vermeulen, N. P. E.; Feenstra, K. A. Cytochrome P450 in silico: An integrative modeling approach. J. Med. Chem. 2005, 48, 2725-2755.

(65) de Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; van der Wijst, T.; Jongejan, A.; Vermeulen, N. P. E. Catalytic site prediction and virtual screening of Cytochrome P450 2D6 substrates by consideration of water and rescoring in automated docking. J. Med. Chem. 2006, 49, 2417-2430.

(66) Rydberg, P.; Gloriam, D. E.; Olsen, L. The SMARTCyp cytochrome P450 metabolism prediction server. Bioinformatics 2010, 26, 2988-2989.

(67) Oláh, J.; Mulholland, A. J.; Harvey, J. N. Understanding the determinants of selectivity in drug metabolism through modeling of dextromethorphan oxidation by cytochrome P450. Proc. Natl. Acad. Sci. 2011, 108, 6050-6055.

(68) (a) Guengerich, F. P. Cytochrome P450 enzymes in the generation of commercial products. Nat. Rev. Drug Discov. 2002, 1, 359-366; (b) Urlacher, V. B.; Lutz-Wahl, S.; Schmid, R. D. Microbial P450 enzymes in biotechnology. Appl. Microbiol. Biotechnol. 2004, 64, 317-325.

(69) Warman, A. J.; Roitel, O.; Neeli, R.; Girvan, H. M.; Seward, H. E.; Murray, S. A.; McLean, K. J.; Joyce, M. G.; Toogood, H.; Holt, R. A.; Leys, D.; Scrutton, N. S.; Munro, A. W. Flavocytochrome P450 BM3: an update on structure and mechanism of a biotechnologically important enzyme. Biochem. Soc. Trans. 2005, 33, 747-753.

(70) (a) Glieder, A.; Farinas, E. T.; Arnold, F. H. Laboratory evolution of a soluble, self-sufficient, highly active alkane hydroxylase. Nat. Biotechnol. 2002, 20, 1135-1139; (b) van Vugt-Lussenburg, B. M. A.; Babel, L. C.; Vermeulen, N. P. E.; Commandeur, J. N. M. Evaluation of alkoxyresorufins as fluorescent substrates for

Chapter 1

37

Cytochrome P450 BM3 and site-directed mutants. Anal. Biochem. 2005, 341, 148-155; (c) Whitehouse, C. J. C.; Bell, S. G.; Tufton, H. G.; Kenny, R. J. P.; Ogilvie, L. C. I.; Wong, L.-L. Evolved CYP102A1 (P450BM3) variants oxidise a range of non-natural substrates and offer new selectivity options. Chem. Commun. 2008, 966-968; (d) Weber, E.; Seifert, A.; Antonovici, M.; Geinitz, C.; Pleiss, J.; Urlacher, V. B. Screening of a minimal enriched P450 BM3 mutant library for hydroxylation of cyclic and acyclic alkanes. Chem. Commun. 2011, 47, 944-946.

(71) (a) Otey, C. R.; Bandara, G.; Lalonde, J.; Takahashi, K.; Arnold, F. H. Preparation of human metabolites of propanolol using laboratory-evolved bacterial Cytochrome P450. Biotechnol. Bioeng. 2006, 93, 494-499; (b) Damsten, M. C.; van Vugt-Lussenburg, B. M. A.; Zeldenthuis, T.; de Vlieger, J. S. B.; Commandeur, J. N. M.; Vermeulen, N. P. E. Application of drug metabolising mutants of cytochrome P450BM3 (CYP102A1) as biocatalysts for the generation of reactive metabolites. Chem.-Biol. Interact. 2008, 171, 96-107; (c) Sawayama, A. M.; Chen, M. M. Y.; Kulanthaivel, P.; Kuo, M.-S.; Hemmerle, H.; Arnold, F. H. A panel of Cytochrome P450 variants to produce drug metabolites and diversify lead compounds. Chem. Eur. J. 2009, 15, 11723-11729; (d) Oliver, C. F.; Modi, S.; Sutcliffe, M. J.; Primrose, W. U.; Lian, L.-Y.; Roberts, G. C. K. A single mutation in Cytochrome P450 BM3 changes substrate orientation in a catalytic intermediate and the regiospecificity of hydroxylation. Biochemistry 1997, 36, 1567-1572; (e) Ost, T. W. B.; Miles, C. S.; Murdoch, J.; Cheung, Y.-F.; Reid, G. A.; Chapman, S. K.; Munro, A. W. Rational re-design of the substrate binding site of flavocytochrome P450 BM3. FEBS Letters 2000, 486, 173-177; (f) van Vugt-Lussenburg, B. M. A.; Damsten, M. C.; Maasdijk, D. M.; Vermeulen, N. P. E.; Commandeur, J. N. M. Heterotropic and homotropic cooperativity by a drug-metabolising mutant of cytochrome P450 BM3. Biochem. Biophys. Res. Commun. 2006, 346, 810-818.

(72) (a) Bernhardt, R. Cytochromes P450 as versatile biocatalysts. J. Biotechnol. 2006, 124, 128-145; (b) Seifert, A.; Vomund, S.; Grohmann, K.; Kriening, S.; Urlacher, V. B.; Laschat, S.; Pleiss, J. Rational design of a minimal and highly enriched CYP102A1 mutant library with improved regio-, stereo- and chemoselectivity. ChemBioChem 2009, 10, 853-861.

(73) (a) Clark, J. P.; Miles, C. S.; Mowat, C. G.; Walkinshwaw, M. D.; Reid, G. A.; Daff, S. N.; Chapman, S. K. The role of Thr268 and Phe393 in cytochrome P450 BM3. J. Inorg. Biochem. 2006, 100, 1075-1090; (b) Branco, R. J. F.; Seifert, A.; Budde, M.; Urlacher, V. B.; Ramos, M. J.; Pleiss, J. Anchoring effects in a wide binding pocket: The molecular basis of regioselectivity in engineered cytochrome P450 monooxygenase from B. megaterium. Proteins 2008, 73, 597-607; (c) Kim, K.-H.; Kang, J.-Y.; Kim, D.-H.; Park, S.-H.; Park, S. H.; Kim, D.; Park, K. D.; Lee, Y. J.; Jung, H.-C.; Pan, J.-G.; Ahn, T.; Yun, C.-H. Generation of human chiral metabolites of simvastatin and lovastatin by bacterial CYP102A1 mutants. Drug Metab. Dispos. 2011, 39, 140-150; (d) Seifert, A.; Antonovici, M.; Hauer, B.; Pleiss, J. An efficient route to selective bio-oxidation catalysts: an iterative approach comprising modeling, diversification, and screening, based on CYP102A1. ChemBioChem 2011, 12, 1346-1351.

(74) van Vugt-Lussenburg, B. M. A.; Stjernschantz, E.; Lastdrager, J.; Oostenbrink, C.; Vermeulen, N. P. E.; Commandeur, J. N. M. Identification of critical residues in novel drug metabolising mutants of Cytochrome P450 BM3 using random mutagenesis. J. Med. Chem. 2007, 50, 455-461.

(75) (a) Rea, V.; Kolkman, A. J.; Vottero, E.; Stronks, E. J.; Ampt, K. A. M.; Honing, M.; Vermeulen, N. P. E.; Wijmenga, S. S.; Commandeur, J. N. M. Active site substitution A82W improves the regioselectivity of steroid hydroxylation by cytochrome P450 BM3 mutants as rationalized by spin relaxation nuclear magnetic resonance studies. Biochemistry 2012, 51, 750-760; (b) Venkataraman, H.; de Beer, S. B. A.; Geerke, D. P.; Vermeulen, N. P. E.; Commandeur, J. N. M. Regio- and stereoselective hydroxylation of optically active α-ionone enantiomers by engineered cytochrome P450 BM3 mutants. Adv. Synth. Catal. 2012, 354, 2172-2184.

(76) Venkataraman, H.; de Beer, S. B. A.; van Bergen, L. A. H.; van Essen, N.; Geerke, D. P.; Vermeulen, N. P. E.; Commandeur, J. N. M. A single active site mutation inverts stereoselectivity of 16-hydroxylation of testosterone catalyzed by engineered Cytochrome P450 BM3. Chembiochem 2012, 13, 520-523.

(77) Sevrioukova, I. F.; Li, H.; Zhang, H.; Peterson, J. A.; Poulos, T. L. Structure of a cytochrome P450-redox partner electron-transfer complex Proc. Natl. Acad. Sci. 1999, 96, 1863-1868.

(78) (a) Ryle, M. J.; Hausinger, R. P. Non-heme iron oxygenases. Curr. Opin. Chem. Biol. 2002, 6, 193-201; (b) Hawker, T. R., Hydroxyphenylpyruvate Dioxygenase (HPPD) - the Herbicide Target. In Modern Crop Protection Compounds, Krämer, W.; Schirmer, U., Eds. Wiley-VCH: 2007; Vol. 1; (c) Moran, G. R. 4-Hydroxyphenylpyruvate dioxygenase. Arch. Biochem. Biophys. 2005, 433, 117-128.

Introduction

38

(79) (a) Lindstedt, S.; Holme, E.; Lock, E. A.; Hjalmarson, O.; Strandvik, B. Treatment of hereditary tyrosinaemia type I by inhibition of 4-hydroxyphenylpyruvate dioxygenase. Lancet 1992, 340, 813-817; (b) Lock, E. A.; Ellis, M. K.; Gaskin, P.; Robinson, M.; Auton, T. R.; Provan, W. M.; Smith, L. L.; Prisbylla, M. P.; Mutter, L. C.; Lee, D. L. From toxicological problem to therapeutic use: The discovery of the mode of action of 2-(2-nitro-4-trifluoromethylbenzenoyl)-1,3-cyclohexanedione (NTBC), its toxicology and development as a drug. J. Inher. Metab. Dis. 1998, 21, 498-506.

(80) (a) Prisbylla, M. P.; Onisko, B. C.; Shribbs, J. M.; Adams, D. O.; Liu, Y.; Ellis, M. K.; Hawkes, T. R.; Mutter, L. C. The novel mechanism of action of the herbicidal triketones. Brighton Crop Prot. Conf. - Weeds 1993, 2, 731-738; (b) Schultz, A.; Ort, O.; Beyer, P.; Kleinig, H. SC-0051, a 2-benzoyl-cyclohexane-1,3-dione bleaching herbicide, is a potent inhibitor of the enzyme p-hydroxyphenylpyruvate dioxygenase. FEBS Letters 1993, 318, 162-166.

(81) (a) López-Ramos, M.; Perruccio, F. HPPD: Ligand- and target-based virtual screening on a herbicide target. J. Chem. Inf. Model. 2010, 50, 801-814; (b) Witschel, M. Design, synthesis and herbicidal activity of new iron chelating motifs for HPPD-inhibitors. Bioorgan. Med. Chem. 2009, 17, 4221-4229.

(82) Ladbury, J. E. Just add water! The effect of water on the specificity of protein-ligand binding sites and its potential application to drug design. Chem. Biol. 1996, 3, 973-980.

(83) (a) Meritt, E. A.; Sixma, T. K.; Kalk, K. H.; van Zanten, B. A. M.; Hol, W. G. J. Galactose-binding site in Escherichia coli heat-labile enterotoxin (LT) and cholera toxin (CT). Mol. Microbiol. 1994, 13, 745 - 753; (b) Rao, M. S.; Olson, A. J. Modelling of Factos Xa-inhibitor complexes: A computational flexible docking approach. Proteins 1999, 34, 173-183; (c) Yee, Y. K.; Tebbe, A. L.; Linebarger, J. H.; Beight, D. W.; Craft, T. J.; Gifford-Moore, D.; Goodson, T. J.; Herron, D. K.; Klimkowski, V. J.; Kyle, J. A.; Sawyer, J. S.; Smith, G. F.; Tinsley, J. M.; Towner, R. D.; Weir, L.; Wiley, M. R. N(2)-Aroylanthranilamide inhibitors of human factor Xa. J. Med. Chem. 2000, 43, 873-882; (d) Ball, P. Water as an active constituent in cell biology. Chem. Rev. 2008, 108, 74-108.

(84) Tame, J. R. H.; Murshudov, G. N.; Dodson, E. J.; Neil, T. K.; Dodson, G. G.; Higgins, C. F.; Wilkinson, A. J. The structural basis of sequence-independent peptide binding by OppA. Protein Sci. 1994, 264, 1578-1581.

(85) Sleigh, S. H.; Seavers, P. R.; Wilkinson, A. J.; Ladbury, J. E.; Tame, J. R. H. Crystallographic and calorimetric analysis of peptide binding to OppA protein. J. Mol. Biol. 1999, 291, 393-415.

(86) Tame, J. R. H.; Sleigh, S. H.; Wilkinson, A. J.; Ladbury, J. E. The role of water in sequence-independent ligand binding by an oligopeptide transporter protein. Nat. Struct. Biol. 1996, 3, 998-1001.