Contribution of both catalytic constant and Michaelis constant to CALB enantioselectivity: Use of FEP calculations for prediction studies

Article abstract of DOI:10.1016/j.molcatb.2011.11.020

Candida antarctica lipase B (CALB) is characterized by its stability and ease of production and is widely used in the pharmaceutical industry. Here we report on the enantioselectivity of the enzyme using both experimental and computational methods. The apparent kinetic parameters were first experimentally determined for enantiopure butan-2-ol and pentan-2-ol substrates. We demonstrate that enantiopreference for the R form of butan-2-ol arises mainly from a lower apparent K_M. This corresponds to a major contribution of ΔΔG_ES, the free energy difference between the ES complex formed with the R and S enantiomers, to ΔΔ^G, the free energy difference between both transition states, in comparison with ΔΔG_kcat, the activation free energy difference. In the case of pentan-2-ol, we show that the enantiopreference for the R form comes from both a lower K_M and a higher k_cat. In addition, we used, for the first time, the free energy perturbation method to evaluate the free energy difference between tetrahedral intermediates formed with R and S alcohol enantiomers for a series of secondary alcohols. This is a valid model for ΔΔ^G. Computational results were found to be in qualitative agreement with experimental data, and enable the determination of substrate orientation in the active site with fair confidence.

Full text of DOI:10.1016/j.molcatb.2011.11.020

Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
Contents lists available at SciVerse ScienceDirect
Journal of Molecular Catalysis B: Enzymatic
journal homepage: www.elsevier.com/locate/molcatb
Contribution of both catalytic constant and Michaelis constant to CALB
enantioselectivity: Use of FEP calculations for prediction studies
L. Chaputa,b, Y.-H. Sanejouandb, A. Balloumia, V. Tranb, M. Grabera,∗
a Université La Rochelle, UMR 6250 LIENSs CNRS – ULR, Pôle Sciences et Technologies, Bâtiment Marie Curie, Avenue Michel Crépeau, 17042 La Rochelle, France
b Unité Biotechnologie, Biocatalyse et Biorégulation (U3B), UMR CNRS 6204 – Faculté des Sciences et des Techniques de Nantes, 2, rue de la Houssinière, BP 92208, 44322 Nantes
Cedex 3, France
a r t i c l e i n f o
a b s t r a c t
Article history:
Candida antarctica lipase B (CALB) is characterized by its stability and ease of production and is widely
used in the pharmaceutical industry. Here we report on the enantioselectivity of the enzyme using both
experimental and computational methods. The apparent kinetic parameters were first experimentally
determined for enantiopure butan-2-ol and pentan-2-ol substrates. We demonstrate that enantiopref-
erence for the R form of butan-2-ol arises mainly from a lower apparent KM. This corresponds to a major
contribution of ꢀꢀGES, the free energy difference between the ES complex formed with the R and S enan-
Received 21 July 2011
Received in revised form
29 November 2011
Accepted 30 November 2011
Available online 16 December 2011
tiomers, to ꢀꢀG‡, the free energy difference between both transition states, in comparison with ꢀꢀGkcat
,
Keywords:
the activation free energy difference. In the case of pentan-2-ol, we show that the enantiopreference for
the R form comes from both a lower KM and a higher kcat. In addition, we used, for the first time, the free
energy perturbation method to evaluate the free energy difference between tetrahedral intermediates
formed with R and S alcohol enantiomers for a series of secondary alcohols. This is a valid model for
ꢀꢀG‡. Computational results were found to be in qualitative agreement with experimental data, and
enable the determination of substrate orientation in the active site with fair confidence.
© 2011 Elsevier B.V. All rights reserved.
Enantioselectivity
Enzyme catalysis
Free energy perturbation
Candida antarctica lipase B
Molecular modelling
1. Introduction
alcohol-binding part of the active site, which contains the stere-
ospecificity pocket [10], defined by Thr42, Ser47 and Trp104. More
Lipase B from Candida antarctica, CALB, is ˛/ˇ hydrolase (EC
3.1.1.3) which catalyses in vivo the hydrolysis of triglycerides. It also
has a high activity and specificity for a wide range of synthetic esters
of primary and secondary alcohols [1]. Due to its stability in organic
media and its large-scale availability, it has found widespread
applications in the enantioselective synthesis of molecules of phar-
maceutical interest and in the resolution of racemic mixtures [2].
CALB displays an enantiopreference for the R form, according to
Kazlauskas rules [3], and its enantioselectivity towards secondary
alcohols is a property that has often been explored [4]. Enantiose-
lectivity is considered to be related to the energy barrier difference
between enantiomers, associated with the transition states formed
during the second step of the bi-bi ping pong enzymatic mech-
anism [5]. The energetic determinants of CALB enantioselectivity
is presently incompletely understood, despite many attempts to
rationalize selectivity.
precisely, the model assumes that the slow-reacting S enantiomer
has to orient its large substituent into the stereospecificity pocket
in order to conserve the hydrogen bond between the oxygen of
the alcohol moiety and His224-Nꢁ, which is essential for catal-
ysis. Although the large substituent does not fit easily into the
sterospecificity pocket, this is the only binding mode which allows
all the essential hydrogen bonds to form in the S enantiomer tran-
sition state and which leads to catalysis. The R enantiomer, on the
other hand, positions its large substituent towards the surface of the
protein and this does not lead to steric limitation. As a consequence,
CALB has an enantiopreference for the R form.
In the past, there have been many other attempts to evaluate
enantiomeric ratio by molecular modelling. For instance, enantios-
electivity has been correlated, with more or less success, to the
difference in the potential energy part ꢀU of the free energy differ-
ence ꢀꢀG‡ between the two enantiomer transition states [11–13].
One of the obvious limitations of this kind of calculation is that,
whilst the potential energy of a protein in explicit solvent is
typically of the order of several thousand kilocalories per mole,
the energy difference between R and S tetrahedral intermediates,
which are good models for the transition states [14], is expected to
be <5 kcal mol−1, as the difference in free energy ꢀꢀG‡ is related
to the enantioselectivity, expressed by the enantiomeric ratio E,
An initial explanation of CALB enantioselectivity was provided
in 1998 by the team of Hult [4,6–9], who suggested that the ori-
entation of R and S enantiomers are significantly different in the
∗
Corresponding author. Tel.: +33 5 46 45 86 30; fax: +33 5 46 45 82 65.
E-mail address: mgraber@univ-lr.fr (M. Graber).
1381-1177/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.molcatb.2011.11.020

30
L. Chaput et al. / Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
as follows: ꢀꢀG‡ = − RT ln E [15–17]. In addition, when the calcu-
lations are based on energy-minimized structures, they face the
multi-minima problem, although this may be partly overcome by
simulated annealing protocols [6]. Finally potential energy calcula-
tions assume that the entropy contribution to enantioselectivity
is of minor importance, whereas a pronounced contribution of
entropy to CALB enantioselectivity has been demonstrated in sev-
eral cases [18,19]. Indeed it can represent as much as 50 % of the
differential activation free energy in absolute value [20].
They described the enantioselectivity of subtilisin in the resolution
of a racemic mixture of sec-phenethyl alcohol by a transesterifica-
tion reaction with vinyl acetate acyl donor [34].
In addition to these computational studies, there has been a lot
of experimental work showing enantiopreference of wild type CALB
for the R form of secondary alcohols. The classical method to quan-
tify enantioselectivity is to measure enantiomeric ratio E, by using
a formula with enantiomeric excess of substrates and products
and conversion ratio [35], instead of using the original definition
of E = (kcat/KM)R/(kcat/KM)S, which involves expensive and fastidi-
ous kinetic studies with enantiopure substrates. E values obtained
by these rapid methods are useful for optimizing reaction yield and
product purity, but fail to give information about the reaction step
at which the enantiopreference occurs.
In the present work, we have highlighted the contribution
of both catalytic constant and Michaelis constant to CALB enan-
tioselectivity. Kinetic parameters for CALB-catalysed acylation of
enantiopure butan-2-ol and pentan-2-ol were first experimentally
determined. This allows the determination of the difference in reac-
tion free energy profile for the two enantiomers, which leads to a
better understanding of the origin of enantioselectivity. FEP calcu-
lations were then performed for tetrahedral intermediates formed
with CALB and five different secondary alcohols with relatively
high structural similarity (butan-2-ol, pentan-2-ol, hexan-3-ol, 3-
methylbutan-2-ol, and 4-methylpentan-3-ol). A system in which
the side chains of both R and S alcohols simultaneously exist was
therefore built and the alchemical transformation consisted of
going step by step from an interaction of the enzyme with the R
form to an interaction with the S enantiomer. It is worthwhile stat-
ing that such an approach also provides concrete predictions of the
substrate orientation within the CALB active site.
Other types of computational approaches have also been used
for quantitative prediction of lipase enantioselectivity. For instance,
Braiuca et al. [21] adapted the 3D-QSAR to quantitatively predict
CALB enantioselectivity towards a wide set of substrates with rel-
atively good accuracy. In 2000, Schultz et al. indirectly assessed
the instability of the tetrahedral intermediate formed by the lipase
from Pseudomonas cepacia with the S enantiomer of different
secondary alcohols, compared to the R form, by measuring the
hydrogen bond distance between the oxygen atom of the free alco-
hol in the active site and His224-Nꢁ[22]. It appeared that the S
enantiomer has greater difficulty to form this essential-to-catalysis
hydrogen bond, suggesting that the energy for the reaction to
proceed towards the formation of the tetrahedral intermediate is
higher for the S substrate. Moreover, a correlation was established
between the length of the S enantiomer hydrogen bond and the
enantiomeric ratio [22]. More recently, Juhl et al. [23] used a cus-
tomised docking method, where enzyme induced-fit was included
by minimization of the enzyme–substrate complex to predict CALB
and W104A mutant enantioselectivities for 1-phenylethyl butyrate.
García-Urdiales et al. [24] used a different strategy and measured
the steric constraints due to the nucleophile observed during stan-
dard molecular dynamics simulations. They showed that, in the
case of the S form, higher CALB enantioselectivity correlates with a
higher number of van der Waals unfavourable contacts [24].
There are existing methods to calculate free energy differences.
One possibility is to use the thermodynamic integration method to
quantitatively predict CALB enantioselectivity. Zhou [25] calculated
free energy differences for butan-2-ol and four other sec-alcohols
with either a bulky group or a bromide as substituent. Calcu-
lations were performed by modifying the dihedral angle value
defined around the chiral carbon to transform, in a stepwise man-
ner, from the R enantiomer to the S enantiomer. The free energy
perturbation (FEP) method is an alternative approach. It has been
applied for several decades [26–28] and remains a powerful tool
today. This method is based on measurements of the progressive
transformation of the system from an initial state to a final state,
usually by following a non-physical (often coined alchemical) path.
Throughout this process, free energy variation is measured. As a
consequence, FEP calculations provide both internal energy and
entropy contributions to the free energy. FEP has been extensively
employed to compute free energy differences in numerous appli-
cations [29], including enzymatic stereospecificity [30] and allows
for the calculation of differences as small as 1 kcal mol−1 [31]. In
practice, however, FEP calculations remain a difficult challenge,
with well-known limitations [32]. Firstly, the molecular model and
the force field used to describe the system thermodynamics must
yield realistic probabilities for its most representative conforma-
tions. Secondly, low frequency motions and long time relaxations
of the system must be handled properly to obtain a fair sampling
of the relevant conformational space [33]. This is often done by
restricting the number of degrees of freedom chosen to model the
system but such a choice then affects the sets of sampled confor-
mations. Another difficulty is finding a good convergence for the
calculations [31].
2. Experimental
2.1. Enantioselectivity measurements
Enantiomeric ratio E was calculated from apparent kinetic
parameters obtained with enantiopure butan-2-ol and pentan-2-
ol, according to the equation E = (kcat/KM)R/(kcat/KM)S. Enantiomeric
ratio values for butan-2-ol, pentan-2-ol and hexan-3-ol, reported
in Section 4, were previously obtained in our laboratory, from
enantiomeric excess of substrates and products [35], in a contin-
uous solid–gas reactor with methylpropanoate as acyl donor, and
immobilized CALB, as previously described [20,36]. For branched
substrates, values were picked from the literature [4], which were
obtained at 39 ◦C in hexane, with S-ethyl thiooctanoaote as acyl
donor.
2.2. Enzyme and chemicals for kinetics
Novozym® 435 (immobilized C. antarctica lipase B) was kindly
provided by Novozymes A/S, Bagsvaerd, Denmark. R and S pure
enantiomers (99 %) of butan-2-ol and pentan-2-ol were purchased
from Sigma–Aldrich (St. Louis, USA), whilst methylpropanoate was
from Fluka (St. Quentin-Fallavier, Switzerland).
2.3. Enzymatic reactions with enantiopure butan-2-ol and
pentan-2-ol
Initial rate measurements were performed at 45 ◦C in 2-
methylbutan-2-ol. 4 ml of the reaction mixtures containing various
amounts of butan-2-ol (100–4000 mM for the
S form and
100–1000 mM for the R form) or pentan-2-ol (100–900 for the S
form and 100–1500 mM for the R form) were incubated for 10 min
prior to addition of 10 mg of Novozym® 435 for the acylation of
butan-2-ol or R pentan-2-ol with 430 mM methylpropanoate, or
Nevertheless, the FEP method has been successfully used to
study enzyme enantioselectivity. Colombo et al. [34] obtained free
energy differences in good agreement with experimental results.

L. Chaput et al. / Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
31
rate-limiting energy barrier. From a technical point of view, the
energy restraint was included in the potential energy of the sys-
tem, and was effective all along the simulation. Free energy was
calculated as the summation of energy differences, therefore the
energetic restraint was balanced. The only way, then, for the
restraint to affect the free energy is by modifying the conforma-
tion sampling, and that is exactly the point which was improved by
using H-bond restraint, which enhances the sampling around the
transition state conformation.
Fig. 1. Orientation of the R and S side chain of a pair of enantiomers within the
stereospecificity pocket of CALB, sketched as a truncated circle. L and M are the
large and medium chains of the substrate, respectively.
3.2. Molecular dynamics simulations
20 mg of Novozym® 435 for reaction with S pentan-2-ol. 200 ␮l
samples were taken at intervals and centrifuged at 14, 000 × g. The
supernatant was analysed by gas chromatography (GC), after two
times dilution with 2-methylbutan-2-ol.
CHARMM c35 program [39] and the CHARMM22 all-atom force
field were used. Force field parameters for the tetrahedral interme-
diate were taken from the literature [40]. These parameters were
obtained from ab initio calculations and were specifically developed
for CHARMM22 force field. Other parameters required for mod-
elling the alkyl side chains of alcohols were defined by homology
with available CHARMM22 parameters.
2.4. GC analysis
Quantitative analysis of reaction products were conducted using
a 7890 GC system from Agilent for the analysis of 1-methylpropyl
propanoate (55 ◦C, 20 min) and of 1-methylbutyl propanoate (55 ◦C
15 min, 3 ◦C min−1, 85 ◦C 5 min), at a flow rate of 1.5 ml min−1
with a Chirasil-Dex CB (25 m, 0.25 mm i.d., 0.25 ␮m ˇ-cyclodextrin,
Chrompack, France) column. Products were detected by FID and
quantified using HP Chemstation software. External calibration
was performed with chemically synthesized esters from the cor-
responding alcohol and propanoic anhydride in pyridine at room
temperature.
In order to mimic our experimental conditions, within a
solid–gas reactor, a 7 A water layer was added to the CALB struc-
˚
ture, all crystallographic water molecules were retained for correct
solvation of the active site. Then all water molecules were energy-
minimized, with 10,000 steps of conjugated gradient followed by
2000 steps with the ABNR (Adopted Basis Newton–Raphson) algo-
rithm.
Next, the whole system was energy-minimized again, except
for atoms kept fixed throughout this study, namely, all atoms more
˚
than 24 A away from the oxygen of the alcohol moiety, and all ˛-
˚
carbon atoms more than 22 A away. As a result, 75% of the protein
3. Computational methods
was left totally free to move. Our criterium is expected to main-
tain the enzyme closer to the crystallographic structure, as well
as to improve the convergence of FEP calculations by prevent-
ing the unwanted contribution of remote events, such as drifts
of loops or of residues far from the active site. It is, therefore, a
compromise between having a good convergence whilst leaving
significant possibility for the active site to freely accommodate the
tetrahedral intermediate and the side chains of the substrates. In
addition, the protocol was designed to evaluate small conforma-
tional rearrangements such as those occurring when going from
the R to the S enantiomer. Sampling larger conformation move-
ments, like ˛-helix would require much longer sampling duration
at each ꢂ, which was not the aim of the article. Other authors
introduced similar types of restraints on atoms far from the region
where the study focuses on FEP methods [41,42]. In another exam-
3.1. Enzyme and tetrahedral intermediate structures
˚
The starting CALB enzyme was the R = 1.55 A crystallographic
structure solved by Uppenberg et al. [37] (PDB entry 1TCA). A tran-
sition state analogue crystal structure, obtained with phosphonate
irreversible inhibitor (PDB entry 1LBS) was used to build the tetra-
hedral part of the reaction intermediate. The acyl part is a propanoyl
group, to allow for the correct location of the central part of the
tetrahedral intermediate. The negatively charged oxygen was ori-
ented towards the oxyanion hole to establish hydrogen bonds with
Thr40 and Gln106. The position of the substituents of R and S alco-
hol enantiomers in the active site were orientated according to the
model postulated by Haeffner et al. [4], so that the large chain of
the S alcohol was oriented into the stereospecificity pocket, and
the medium chain towards the active site entry, as shown in Fig. 1.
Inversely, the medium chain of the R alcohol was located in the
stereospecificity pocket, and the large chain was oriented towards
the active site entry. Note that this was done in order to, firstly,
define a starting point for the simulations and, secondly, during the
course of the calculations, the substrates were thus free to reorient
themselves in a different way.
˚
ple, Allouche et al. [43] defined free atoms in a radius of 9 A centred
on the alchemical transformation of one cation to another (Ca2+
towards Mg2+). In our study, however, alchemical transformations
concern more atoms, hence the choice of a much larger radius
for the free part of the system. On the other hand, Trodler and
Pleiss [44] showed by molecular modelling that the structure of
C. antarctica lipase B during molecular dynamics simulations, in
water or in five different organic solvents, exhibits a low deviation
from the crystal structure. Finally, McCabe et al. [45] made circu-
lar dichroism measurements, using synchrotron radiation, to study
conformational changes in water and various organic solvents and
concluded that the secondary structure of CALB in aqueous buffer is
close to its crystal structure. These studies suggest that CALB does
not exhibit significant domain movements and that fixing atoms
far away from the active site may not perturb CALB significantly.
An initial heating stage of 50 ps, from 100 to 300 K, was per-
formed before every FEP calculation and then the temperature was
increased by 1 K every 100 steps. For accurate results, the system
must be well equilibrated with long initial equilibration periods,
corresponding to 800 ps (trajectory 1) and 1000 ps (trajectories 2
and 3). During all molecular dynamics simulations, the non-bonded
However, the essential hydrogen bond involved in the transition
state between His224-Nꢁ and the alcohol oxygen of the tetrahedral
intermediate, was prevented from disruption during the molecular
dynamics simulations by a harmonic constraint of 100 kcal mol−1
˚
applied on the distance between heavy atoms exceeding 2.8 A.
Without this constraint, the hydrogen bond would probably have
been disrupted during the dynamics. When hydrogen bond disrup-
tion is observed in the case of the S enantiomer [38], it is usually
considered as a measure of the difficulty this enantiomer expe-
riences before reaching an orientation which allows the reaction
to proceed. In the present work, our choice has been to maintain
the system as close to the transition state as possible, in order
to obtain estimates for free energy differences at the top of the

32
L. Chaput et al. / Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
Table 1
Kinetic constants for
pairs list was updated every 20 steps and the temperature was
checked every 5000 steps.
app.
R and S enantiopure butan-2-ol and pentan-2-ol (V
max
expressed in mmol min−1 mg−1 of immobilized enzyme, and KM expressed in mM).
Butan-2-ol
Pentan-2-ol
3.3. Free energy perturbation protocol
KM
R
S
258
1284
70
>472
A system was built with distinct, and simultaneously existing,
side chains of both R and S secondary alcohols, including the chi-
ral carbon, also called the dual topology approach. In accordance
with the alchemical transformation principle, the oxygen atom is
shared by the two enantiomers. A hybrid potential energy U(r, ꢂ)
is associated with such a system, which is a function of r, the atom
coordinates, and ꢂ, a coefficient scale to quantify the interaction
energy of the R enantiomer with the rest of the enzyme. The corre-
sponding coefficient for the S enantiomer is 1 − ꢂ. The free energy
difference, ꢀF along the path starting at ꢂi = 0 and ending at ꢂi = 1
was calculated using the exponential formula [46]:
app.
max
R
S
V
0.062
0.051
0.048
≈0.001
E
6
>324
perturbation at each step was then calculated using the FREE mod-
ule of CHARMM and the wide sampling method. Moreover, the
trajectory at ꢂi was used to compute incremental ꢀF values from
ꢂi − 0.025 to ꢂi and from ꢂi to ꢂi + 0.025. The molecular dynamics
transformation of R to S, followed by the reverse path, took about
18 h when run on 24 parallelized Bi-Xeon processors.
n−1
ꢀ
− ˇ−1 ln ꢀe−ˇꢀU
i+1 ꢁꢂ
ꢂ →ꢂ
ꢀFR→S = FS − FR
=
i
i
i=0
4. Results and discussion
where FS and FR represent the free energy of the enzyme with,
respectively, the R and S alcohol moiety linked to the tetrahe-
4.1. Experimental determination of apparent kinetic constants for
enantiopure butan-2-ol and pentan-2-ol
dral intermediate. ꢀ ꢁꢂ represents the canonical ensemble average
i
obtained with ꢂi, where ˇ is the inverse of the thermal energy,
namely, kbT, kB being the Boltzman constant and T, the tempera-
ture. The sum of the n successive terms calculated along the R to
S transformation leads to the total free energy difference between
the complex with R and S enantiomers. The value obtained with
the sum computed for the reverse transformation of S to R should
rigorously give the opposite value to the R to S transformation.
Such a comparison is a standard indicator of the quality of a FEP
calculation.
Experimental apparent kinetic parameters obtained for enan-
tiopure butan-2-ol and pentan-2- ol are presented in Table 1. The
KMapp. for the R form of butan-2-ol is 258 mM and for the S form
is 1284 mM. In the case of pentan-2-ol, we obtained a KMapp. of
70 mM for the R form. However, CALB was not saturated when
using S pentan-2-ol concentrations up to 1500 mM, consequently
the KMapp. value was above 472 mM, the alcohol concentration giv-
ing half of the reaction rate obtained at 1500 mM. Concentrations
In practice, ꢂi was linearly incremented by steps of 0.05, along
twenty (n = 20) successive equilibration and productive dynamics,
of 25 and 50 ps each. The decision to divide the transformation
into twenty steps yields a small free energy perturbation at each
step and, as a consequence, better accuracy. Too high a number of
sub-trajectories is not useful because if ꢂi is too close to 0 (or to
1), weak interactions between alcohol side chains and the enzyme
yield unrealistic behaviour characterized by wide movements of
the side chains and, therefore, by a poor sampling of the conforma-
tional space. This is why the ꢂi value was initially set at 0.975 and
0.025 for the interaction between, respectively, the R and S alco-
hol side chain and the enzyme. The transformation of the substrate
from R to S was followed by a 100 ps equilibration step and, then,
by the reverse transformation (S to R), in order to assess the conver-
gence of each FEP calculation. Twenty productive sub-trajectories
were thus obtained for each transformation of an R enantiomer to
an S enantiomer, and the same number obtained for the reverse
path. In addition, each FEP calculation was made twice, starting
from two sets of initial conditions.
The free energy perturbation (FEP) protocol is presumed to be
reliable when the energy computed for the R to S transformation
and for the reverse path give similar values with opposite signs.
This presumption cannot, in fact, be taken for granted. Indeed,
reaching convergence in free energy perturbation calculations is
a challenge that depends on several parameters involved in the
process. The FEP method assumes that the part of the phase space
where the energies of both compared systems are significantly dif-
ferent can be extensively explored. This can only be achieved by
using enough computer time to sample the conformations with
significant energy differences between two neighbouring states
along the path followed during each transformation. In practice,
FEP calculations were performed using the BLOCK command avail-
able in CHARMM [39], which allows one to set the value of ꢂi. The
above 1500 mM of S pentan-2-ol were not used owing to the exor-
app.
bitant price of this enantiopure alcohol. The V
for butan-2-ol is
max
0.062 for the R form and 0.051 for the S form, in mmol min−1 mg−1
of immobilized enzyme. Pentan-2-ol gave values of 0.048 and
around 0.001 mmol min−1, for the R and S forms respectively,
results obtained by non linear regression of the Michaelis–Menten
equation. The data demonstrate that for butan-2-ol, the enantio-
preference for the R substrate can be attributed mainly to the KM
difference, which is lower, by a factor of 5 for the R alcohol, sug-
gesting that the enzyme affinity for the R alcohol is greater than for
the S form. In contrast, the apparent Vmax is only slightly higher for
the R enantiomer. However, in the case of pentan-2-ol, the enantio-
preference arises from both a lower KM and a much higher Vmax for
the R form. Assuming that 1 g of Novozym® 435 contains 1 ␮mol
of active CALB [47], kcat can be calculated from Vmax values and are
equal to 1033 s−1 and 850 s−1 for the R and S form of butan-2-ol,
and to 800 s−1 and 17 s−1 for the R and S form of pentan-2-ol.
Hypothetical energy profile diagrams for the CALB-catalysed
acylation reactions acting on R and S enantiomers for both butan-
2-ol and pentan-2-ol are shown in Fig. 2.
Apparent KM can be used as an apparent dissociation constant,
Ks, under conditions of steady state assumption. The lower KM for
the R enantiomer corresponds to a ground state stabilization of the
ES-complex, which has a lower energy compared to S enantiomer.
kcatR and kcatS correspond to the difference between ES and ES‡ for
R and S enantiomers. In the case of both pentan-2-ol and butan-
2-ol, kcatR is higher than kcatS. The second order rate constant for
the enzymatic reaction, starting with free enzyme and substrate
is given by kcat/KM and corresponds to the difference between E + S
and ES‡. The enantiopreference for the R form is linked to the energy
barrier difference between enantiomers, associated with the tran-
sition states (ES‡S–ESR‡ ). We demonstrate here that this difference
is primarily due to the KM difference between the R and S form,

L. Chaput et al. / Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
33
4.2. Comparison of FEP results from this study with experimental
data
Table 2 represents the values obtained by FEP methods from
calculations of ꢀFR→S and ꢀFS→R, the free energy difference going
from the R form of the enzyme to the S form of the enzyme and the
reverse path. We obtained positive values for ꢀF when going from
the R to the S enantiomer, except for one trajectory, and obtained
consistently negative values along the reverse path. This corre-
sponds to a lower free energy with the R form than with the S
form, as expected from experimental data. Thus from a qualitative
point of view, FEP results correctly predict the enantiopreference of
CALB for the R form. From a quantitative point a view and for linear
secondary alcohols, average ꢀF values are equal to 1.38 kcal mol−1
for butan-2-ol, 3.62 kcal mol−1 for pentan-2-ol and 2.58 kcal mol−1
for the first trajectory of hexan-3-ol (the exclusion of trajectories
2 and 3 is explained below). These values are in good agreement
with experimental ꢀꢀG‡. For ramified alcohols, the average ꢀF
value is equal to 2.72 for 3-methylbutan-2-ol and the unique ꢀF
value obtained with correct signs for 4-methylpentan-3-ol is equal
to 1.97 kcal mol−1. FEP calculations correctly rank these two sub-
strates, but fail to rank linear and ramified alcohols. The ꢀF values
of branched alcohols are lower than experimental ꢀꢀG‡, and the
standard deviation obtained for 3-methylbutan-2-ol is higher than
for linear alcohols. Calculations with 4-methylpentan-3-ol gave the
weakest convergence, perhaps due to the large bulky substituent
of this alcohol, which may require a pronounced rearrangement of
the enzyme structure in order to accommodate both enantiomers.
It is worth noticing that substrates with very different enan-
tiomeric ratios are expected to have transition states with small
free energy differences between both enantiomers, an important
point when comparing experimental and calculated free energy
differences, ꢀꢀGe‡xp.. Indeed, for experimental E values ranging
between 6.0 and 705 (Table 2), calculated free energy differences
are expected to range between 1 and 4 kcal mol−1, which is far
less than the energy of a typical hydrogen bond, which is about
Fig. 2. Energy profile diagrams for the CALB-catalysed acylation reaction acting on
R (black line) and S (grey line), with E the enzyme, SS and SR the S and R enantiop-
ure substrates, ESS and ESR, the enzyme–enantiopure substrate complexes, ES‡ and
S
ES‡ , the enzyme–enantiopure substrate transition state complexes, PS and PR, the
R
S and R enantiopure products. ꢀꢀGES represents the energy difference between
bound enantiopure substrates, ꢀGkcatS and ꢀGkcatR represent the energy activation
for each enantiomer, ꢀꢀGE‡S represents the free energy difference between R and S
enzyme–enantiopure substrate transition state complexes.
rather than to the kcat in the case of butan-2-ol. In contrast, the
energy barrier difference for pentan-2-ol is associated with both
KM and kcat
.
These results are significantly different from previous experi-
mental results obtained by Magnusson et al. [48] (2005) for CALB
catalysed acylation of pure enantiomers of 1-phenylethanol, a very
bulky secondary alcohol. The large enantiomeric ratio (E = 1, 300,
000) was almost entirely due to the difference in apparent kcat
between the enantiomers, 570 s−1 and 0.00053 s−1 for the R and
S forms, respectively. We find that the respective contributions of
KM and kcat to enantioselectivity are highly variable and probably
depend on substrate structure. In the particular case of a secondary
alcohol with a very bulky large substituent, like 1-phenylethanol,
Magnusson et al. obtained an enantioselectivity almost entirely
due to the difference in kcat. The inverse trend was obtained in the
present work, with another particular case: butan-2-ol which has
only an ethyl as “large” substituent and for which enantioselectivity
is almost entirely due to the difference in KM. In case of pentan-2-ol
an intermediate result was obtained: enantioselectivity arises from
both a lower KM and a much higher kcat for the R form.
4.8 kcal mol−1
.
Hexan-3-ol is a special case in point, in the case of trajectory 1,
FEP results are close to the experimental data (2.61 kcal mol−1), but
trajectory 2 had significantly higher ꢀF values, therefore a third tra-
jectory was performed. Trajectories 2 and 3 both yield an average
ꢀF value equal to 4.21 kcal mol−1, instead of 2.58 kcal mol−1 found
from trajectory 1. Comprehensive analysis of the global structure
of CALB during the three independent trajectories provides a good
indication of the putative implication of secondary structure ele-
ments surrounding the active site. During the first trajectory, which
displays better ꢀF, a small shift of the ˛-helix 10 occurred, com-
pared to the crystallographic structure. This shift was not observed
during the two other trajectories. This ˛-helix shift may be involved
in a better fit between the enzyme and the substrate, explaining
the smaller free energy change during the first trajectory. Inter-
estingly, several groups previously observed the movement of this
˛-helix. Recently, Trodler and Pleiss [44] observed the flexibility of
this helix during molecular dynamics simulations. Skjøt et al. [49]
also demonstrated that during a 10ns dynamic trajectory in a water
box with periodic conditions, ˛-helix 5 and 10 of CALB displayed
significant mobilities. ˛-helix 5 is far from the active site, therefore,
in the case of the substrates described here, it can be assumed that
this helix has a negligible effect on substrate conformation, contrary
to ˛-helix 10, which is directly in contact with the side chain of the
alcohol, particularly through residues Ile189, Ala278 and Ala282.
Unfortunately, the FEP/MD protocol was not designed to handle
events like ˛ helix movement, as observed by Zhao et al. [50].
These authors assessed the free energies for antidepressant bind-
ing to protein receptor by FEP/MD and mentioned that the receptor
was sampled in only one conformation, due to the fact that large
E
values (Table 1) calculated using the formula
E = (kcat/KM)R/(kcat/KM give a value of 6 for butan-2-ol and a
)
S
value above 324 for pentan-2-ol. These values are similar to those
obtained previously in our laboratory at the same temperature, in
a continuous solid–gas reactor with the same acyl donor at 20 %
relative humidity, 6 for butan-2-ol and 330 for pentan-2-ol [20],
using enantiomeric excess to calculate E. Due to the similarity of
the results obtained using different approaches, we chose to add
hexan-3-ol as an additional substrate for our computer simulations,
a substrate for which we had previously determined an E value
of 80 under the same conditions (Table 2). Two supplementary
substrates containing branched substituents (3-methylbutan-2-ol
and 4-methylpentan-3-ol) were also added as extra challenges for
our computational modelling. For these last substrates, E values
were taken from the literature [4]. They were obtained at the same
temperature but with a different acyl donor: S-ethyl thiooctanoate
(Table 2).

34
L. Chaput et al. / Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
Table 2
Calculated ꢀFR→S and ꢀFS→R compared to the experimentally determined ꢀꢀG‡
as derived from the experimentally determined enantioselectivity E. All free energy
R→S exp.
differences are expressed in kcal mol−1
.
‡exp.
Substrate
E
ꢀꢀG
Traj.
ꢀFR→S
ꢀFS→R
|ꢀFavg.
|
sda
R→S
Butan-2-ol
6.0
1.07
3.46
2.61
1
2
1
2
1
+1.10
+0.78
+4.04
+3.57
+2.06
−1.33
−2.30
−3.94
−2.94
−3.09
1.38 0.66
3.62 0.5
Pentan-2-ol
Hexan-3-ol
330
80
2.58
2
3
+4.90
+4.29
−4.84
−4.79
4.21 0.28
3-Methylbutan-2-ol
4-Methylpentan-3-ol
705
109
3.91
2.80
1
2
1
2
+3.50
+3.32
+2.86
−0.07
−2.03
−2.02.
−1.07
+1.60
2.72 0.8
1.97
ncb
a
Standard deviation.
No convergence.
b
movements occurred during a larger time scale, exceeding the time
frame used in the protocol.
experimental data display slightly higher apparent kcat for the R
enantiomer in the case of butan-2-ol and high apparent kcat in the
case of pentan-2-ol, and show that enantioselectivity is also linked
to the difference between apparent KM. The data can be under-
stood by detailing the energy profile diagram of the reaction (Fig. 2)
by using Eyring’s transition state theory, which links the reaction
rate to the activation energy. The Eyring law defines the rate k of a
reaction as a fu‡nction of temperature and of the activation energy:
Our results are in concordance with the alternative protocol of
thermodynamic integration used by Zhou [25] for his free energy
calculations for two bulky chiral alcohols as well as for butan-2-ol.
He failed, however, to predict enantioselectivity for chiral alco-
hols containing bromide and the reproducibility of his results was
not addressed, as only a single trajectory per substrate was per-
formed [25].
k = kBT/h e−ꢀG /RT
.
This energy corresponds to the free energy difference between
a ground state and the top of the energy barrier. In the present
case, the top of this barrier has been modelled by the tetrahedral
intermediate ES‡. The ground state can be defined as the complex
enzyme–substrate ES. Activation energy is equal to ꢀGkcat, corre-
sponding to kcat (see Fig. 2), and for given concentrations of E and
S, the rate constant for the enzymatic reaction, starting with free
enzyme and substrate, is given by kcat/KM and corresponds to the
energy ꢀG‡. In the Michaelis–Menten model, the reaction proceeds
through a classical ligand–enzyme interaction model, and the KM
can be assimilated with the inverse of the affinity constant, under
conditions of steady state. A small KM corresponds to a good affinity
of the substrate for the enzyme, characterized by a low free energy
ꢀGES stabilizing the complex ES. Therefore, as shown in Fig. 2, the
KM is linked to the energy difference ꢀGES between free substrate
and bound substrate.
4.3. Substrate orientations
In our study, the starting orientation of the alcohol substrates
was defined according to the model suggested by Haeffner et al. [6].
For all substrates, except pentan-2-ol, this orientation was well
conserved all along the dynamic trajectory for both S and R alcohol
side chains. So we think that initial orientations we used for the
substrates are acceptable, because we did not observe any rear-
rangement for them, even when substrates get the opportunity to
reorient themselves, when they are weakly interacting with the
enzyme (ꢂ near 0 and 1 for S and R enantiomers, respectively). As
seen in Fig. 3A, in the case of butan-2-ol, the large chain (ethyl
group) of the S alcohol is located in the stereospecificity pocket,
close to amino acids Gly39, Thr40 and Thr42. The large chain of
the R alcohol is oriented towards the entry of the active site. In
the case of pentan-2-ol, a re-orientation of the large substituent
of the S enantiomer alcohol occurs for both trajectories 1 and 2. It
can be observed in Fig. 3B that the large chain bends towards the
active site entry. This is probably due to steric constraints within the
stereospecificity pocket. To our knowledge, this orientation was not
observed previously. Kwon et al. [11] did demonstrate, however,
that for some alcohols, side chains may adopt other orientations
than those described by the Haeffner model.
Here, it was experimentally observed that the KM is much
lower for the R enantiomer. This means that the energy of
the ES complex is lower for the R enantiomer, and the differ-
ence in affinity for the enzyme between the two enantiomers is
characterized by the free energy difference ꢀꢀGES. In addition,
experimental data demonstrate that the apparent Vmax is higher
for the R enantiomer, indicating that the free energy, ꢀGkcat, is
smaller for the R enantiomer. Thus, the difference between tran-
sition state energy ꢀꢀG‡ depends on both ꢀꢀGES and ꢀꢀGkcat
,
with ꢀꢀGkcat = ꢀGkcatS − ꢀGkcatR. Consequently, enantioselectiv-
ity is characterized by the free energy difference ꢀꢀG‡ that
can be expressed as ꢀꢀG‡ = ꢀꢀGES + ꢀꢀGkcat. The resolution of
butan-2-ol by CALB-catalysed acylation demonstrates that the
contribution of the first energetic term ꢀGES is predominant com-
pared to ꢀꢀGkcat, whereas for pentan-2-ol, the two factors are
important.
4.4. Relation between free energy profile of enzymatic catalysis
and experimental and computational results
The free energy differences calculated here, using the FEP
method, are expected to match those for the transition states
ES‡ of the reaction, as catalytic rates depend on the energy bar-
rier associated with the formation of the transition state. The

L. Chaput et al. / Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
35
Fig. 3. Tetrahedral intermediate (TI) and side chains orientation of butan-2-ol and pentan-2-ol enantiomers. S enantiomer in green and R enantiomer in orange. The white
arrow points out the stereospecificity pocket.
4.5. Conclusions
enantioselectivity through differences in energetic pathways
between enantiomers. This is a first attempt to quantify free energy
difference using the FEP method to study CALB enantioselectiv-
ity. In the future this could become a very interesting tool for the
pharmaceutical industry.
Experimental data obtained with enantiopure butan-2-ol shows
that enantiopreference of CALB for the R form arises mainly from
a lower apparent KM and, to a much lesser extent, from a higher
kcat for this enantiomer. With pentan-2-ol, enantiopreference arises
from both a lower KM and a much higher kcat for the R enantiomer,
suggesting that no general rule can be defined for all substrates,
as far as the contribution of the various kinetic parameters to
enantioselectivity is concerned. FEP calculations presented in this
study successfully provided qualitative prediction of the enantio-
preference of CALB for R enantiomers in the case of four of the five
substrates tested. However, the quantitative prediction of the enan-
tiomeric ratio itself proved challenging. In the best cases, namely,
butan-2-ol and pentan-2-ol, the corresponding free energy dif-
ference was overestimated by 0.2–0.3 kcal mol−1, on average. For
hexan-3-ol, only one trajectory amongst the three which were per-
formed, gave almost the same result as the experimental result.
This corresponds also to the only case where a small shift of
the ˛-helix 10 occurs, and this is certainly a case where a bet-
ter accommodation of the enzyme occurred. On the whole, FEP
calculations provide a much more efficient evaluation of energy
difference between enantiomers than potential energy evaluation.
Differences in absolute values between calculated and experimen-
tal ꢀꢀG‡ may be attributed to the limitations of our approach for
modelling global enzyme accommodation, including, for example,
the possible movement of the ˛-helix 10 which may allow CALB to
adapt the shape of its active site to large substrates like hexan-3-ol.
We are fully aware of the fact that the protocol can be improved,
in order to give a better reproducibility and enabling at the same
time, a global enzyme accommodation during simulation. Another
possible source of error is the approximate description of the tran-
sition state of the reaction as a tetrahedral intermediate. In this
respect, QM/MM calculations may prove useful, to better define
both the geometry of the transition state and the distribution of
charges around the chiral centre and also perhaps including a few
key neighbouring amino acid residues [51–53]. Notwithstanding,
FEP calculations can provide results for novel substrates, without
the need for a significant number of experimental data to adjust
the model, as is the case, for instance, with 3D-QSAR methods.
Moreover, FEP calculations can provide clues about tetrahedral
intermediate geometries. Indeed, the analysis of the trajectories
performed during this study strongly supports Haeffners model of
the orientation of the S substrates within the active site of CALB:
the S orientation was preserved in all cases, except that of pentan-
2-ol. The data presented here help in understanding the origin of
Acknowledgements
This study was supported by the French ANR (National Research
Agency) through the EXPENANTIO project (program CP2D). CINES
and IDRIS from GENCI (Grand Equipement National de Calcul
Intensif) are acknowledged for giving access to super-computing
facilities. The manuscript was corrected by a native English scien-
tific translator (http://traduction.lefevere-laoide.net).
References
[1] R.D. Schmid, R. Verger, Angew. Chem. Int. Ed. 37 (1998) 1608–1633.
[2] M.T. Reetz, Curr. Opin. Chem. Biol. 6 (2002) 145–150.
[3] R.J. Kazlauskas, A.N.E. Weissfloch, A.T. Rappaport, L.A. Cuccia, J. Org. Chem. 56
(1991) 2656–2665.
[4] D. Rotticci, F. Haeffner, C. Orrenius, T. Norin, K. Hult, J. Mol. Catal. B: Enzym. 5
(1998) 267–272.
[5] M.-P. Bousquet-Dubouch, M. Graber, N. Sousa, S. Lamare, M.-D. Legoy, Biochim.
Biophys. Acta 26 (2001) 90–99.
[6] F. Haeffner, T. Norin, K. Hult, Biophys. J. 74 (1998) 1251–1262.
[7] C. Orrenius, F. Haeffner, D. Rotticci, N. Öhrner, T. Norin, K. Hult, Biocatal. Bio-
transform. 16 (1998) 1–15.
[8] F. Haeffner, T. Norin, Chem. Pharm. Bull. 47 (1999) 591–600.
[9] D. Rotticci, J.C. Rotticci-Mulder, S. Denman, T. Norin, K. Hult, ChemBioChem 2
(2001) 766–770.
[10] J. Uppenberg, N. Öhrner, M. Norin, K. Hult, G.J. Kleywegt, S. Patkar, V. Waagen,
T. Anthonsen, J.T. Alwyn, Biochemistry 34 (1995) 16581–16838.
[11] C.H. Kwon, D.Y. Shin, J.H. Lee, K.S. Wook, K.J. Won, J. Microbiol. Biotechnol. 17
(2007) 1098–1105.
[12] C. Orrenius, C. van Heusden, J. van Ruiten, P. Overbeeke, H. Kierkels, J.A. Duine,
A.J. Jongejan, Protein Eng. 11 (1998) 1147–1153.
[13] S. Raza, L. Fransson, K. Hult, Protein Sci. 10 (2001) 329–338.
[14] A. Warshel, G. Naray-Szabo, F. Sussman, J.K. Hwang, Biochemistry 28 (1989)
3629–3637.
[15] H. Eyring, M. Polanyi, Z. Phys. Chem. 12 (1931) 279.
[16] R.S. Philipps, Enzyme Microb. Technol. 14 (1992) 417–419.
[17] R.S. Philipps, Trends Biotechnol. 14 (1996) 13–16.
[18] A.P.L. Overbeeke, C. Orrenius, J.A. Jongejan, J.A. Duine, Chem. Phys. Lipids 93
(1998) 81–93.
[19] J. Ottosson, L. Fransson, K. Hult, Protein Sci. 11 (2002) 1462–1471.
[20] V. Leonard-Nevers, Z. Marton, S. Lamare, K. Hult, M. Graber, J. Mol. Catal. B:
Enzym. 59 (2009) 90–95.
[21] P. Braiuca, K. Lorena, V. Ferrario, C. Ebert, L. Gardossi, Adv. Synth. Catal. 351
(2009) 1293–1302.
[22] T. Schulz, J. Pleiss, R.D. Schmid, Protein Sci. 9 (2000) 1053–1062.
[23] P.B. Juhl, P. Trodler, S. Tyagi, J. Pleiss, BMC Struct. Biol. 9 (2009) 39–55.
[24] E. García-Urdiales, N. Ríos-Lombardía, J. Mangas-Sánchez, V. Gotor-Fernández,
V. Gotor, ChemBioChem 10 (2009) 1830–1838.

36
L. Chaput et al. / Journal of Molecular Catalysis B: Enzymatic 76 (2012) 29–36
[25] Y. Zhou, Ph.D. thesis, Computational Study of Enzyme Enantioselectivity, Uni-
versity of Delft, NL, 2006.
[26] W.L. Jorgensen, Acc. Chem. Res. 22 (1989) 184–189.
M. Feig, S. Fischer, J. Gao, M. Hodoscek, W. Im, K. Kuczera, T. Lazaridis, J. MA, V.
Ovchinnikov, E. Paci, R.W. Pastor, C.B. Post, J.Z. Pu, M. Schaeffer, B. Tidor, R.M.
Venable, H.L. Woodcock, X. Wu, W. Yang, D.M. York, M. Karplus, J. Comput.
Chem. 30 (2009) 1545–1614.
[27] J. Aqvist, J. Phys. Chem. 94 (1990) 8021–8024.
[28] T. Simonson, G. Archontis, M. Karplus, Acc. Chem. Res. 35 (2002) 430–437.
[29] D.Q. McDonald, W.C. Still, J. Comput. Chem. 118 (1996) 2073–2077.
[30] C. de Graaf, C. Oostenbrink, P.H.J. Keizers, B.M.A. van Vugt-Lussenburg, J.N.M.
Commandeur, N.P.E. Vermeulen, Eur. Biophys. J. 36 (2007) 589–599.
[31] A. Aleksandrov, D. Thompson, T. Simonson, J. Mol. Recognit. 23 (2010) 117–127.
[32] C.D. Christ, A.E. Mark, W.F. Van Gunsteren, J. Comput. Chem. 31 (2009)
1569–1582.
[33] A. Di Nola, A.T. Brünger, J. Comput. Chem. 19 (1998) 1229–1240.
[34] G. Colombo, S. Toba, J.K.M. Merz, J. Am. Chem. Soc. 121 (1999) 3486–3493.
[35] C.-S. Chen, Y. Fujimoto, G. Girdaukas, C.J. Sih, J. Am. Chem. Soc. 104 (1982)
7294–7299.
[40] N. Otte, M. Bocola, W. Thiel, J. Comput. Chem. 30 (2009) 154–162.
[41] R. Mutyala, R.N. Reddy, M. Sumakanth, P. Reddanna, M.R. Reddy, J. Comput.
Chem. 28 (2007) 932–937.
[42] J. Wang, Y. Deng, B. Roux, Biophys. J. 91 (2006) 2798–2814.
[43] D. Allouche, J. Parello, Y. Sanejouand, J. Mol. Biol. 285 (1999) 857–873.
[44] P. Trodler, J. Pleiss, BMC Struct. Biol. 8 (2008) 1–10.
[45] R. McCabe, A. Rodger, A. Taylor, Enzyme Microb. Technol. 36 (2005) 70–74.
[46] R.W. Zwanzig, J. Chem. Phys. 22 (1954) 1420–1426.
[47] C. Hedfors, K. Hult, M. Martinelle, J. Mol. Catal. B: Enzym. 66 (2010) 120–123.
[48] A.O. Magnusson, M. Takwa, A. Hamberg, K. Hult, Angew. Chem. Int. Ed. 44 (2005)
4582–4585.
[36] V. Leonard, S. Lamare, M.-D. Legoy, M. Graber, J. Mol. Catal. B: Enzym. 32 (2004)
53–59.
[49] M. Skjøt, L. De Maria, R. Chatterjee, A. Svendsen, S.A. Patkar, P.R. Østergaard, J.
Brask, ChemBioChem 10 (2009) 520–527.
[37] J. Uppenberg, M.T. Hansen, S. Patkar, T.A. Jones, Structure 2 (1994) 293–308.
[38] J. Nyhln, B. Martn-Matute, A.G. Sandstrm, M. Bocola, J.-E. Bckvall, ChemBioChem
9 (2008) 1968–1974.
[39] B.R. Brooks, C.L. Brooks III, A.D. Mackerell, L. Nilsson Jr., R.J. Petrella, B. Roux,
Y. Won, G. Archontis, B. C, S. Boresch, A. Caflisch, L. Caves, Q. Cui, A.R. Dinner,
[50] C. Zhao, D.A. Caplan, S.Y. Noskov, J. Chem. Theory Comput. 6 (2010) 1900–1914.
[51] R. Murphy, D. Philipp, R. Friesner, J. Comput. Chem. 21 (16) (2000) 1442–1457.
[52] Y. Zhang, J. Kua, J. McCammon, J. Am. Chem. Soc. 124 (35) (2002) 10572–10577.
[53] C. Hensen, J. Hermann, K. Nam, S. Ma, J. Gao, H. Holtje, J. Med. Chem. 47 (27)
(2004) 6673–6680.