Chapter

6.

Enantioselective enzymes by computational design
and screening
Hein J. Wijma, Robert J. Floor, Sinisa Bjelic, Siewert J. Marrink, David Baker and Dick B.
Janssen


Computational enzyme design can provide new or improved biocatalysts for
synthetic chemistry. However, current computational tools are not sufficiently
accurate to control substrate binding and orientation in catalytic sites with the level
of precision required for rapid and selective catalysis. Here, we report a strategy
that uses computational design and molecular dynamics simulations to produce
divergent enzyme variants that position a meso substrate in predefined orientations
required for enantioselective conversions. We applied this strategy to develop
enantiocomplementary epoxide hydrolases in which the target substrate
cyclopentene oxide is oriented such that ring opening proceeds predominantly by
attack at either the (R) or (S)carbon atom, producing highly enantioenriched (S,S)
diol or (R,R)diol, respectively. Key features of this strategy (CASCO, catalytic
selectivity by computational design and MD screening for optimal substrate
positioning) are the use of RosettaDesign to design mutations that favor binding of
the substrate in a predefined orientation, the introduction of steric hindrance to
prevent unwanted substrate binding modes, and an in silico ranking of initial
designs by highthroughput molecular dynamics simulations. The scoring procedure
uses nearattack conformations to select variants with a high likelihood of having
the required selectivity. After in silico design and ranking of 2,500 initial multisite
mutants, a small library of 37 variants was constructed which harbored nine mutant
epoxide hydrolases with the desired (R,R) or (S,S)product selectivity. The results
show that a large part of the experimental screening that is common in directed
evolution protocols can be replaced by in silico methods.

Part of this chapter has been published:

Angewandte Chemie International Edition 2015
doi 10.1002/anie.201411415

Chapter 6

Introduction
Recently, computational protein design (CPD) has emerged as a promising method
to design enzymes with activities not found in nature and to improve activities of
existing enzymes. Examples of the de novo designed enzymes include esterases[1],
Kemp eliminases[2], retroaldolases[3] and DielsAlderases[4]. Modification of enzyme
selectivity by computational methods was reported for several enzymes including
homing endonucleases[5], deaminases[6], a peptidase[7], a phosphorylase[8],
phosphotriesterase[9] and kinases[10]. The results described in those publications
raise the possibility that computational design may be used for controlling enzyme
enantioselectivity. A possible strategy could be to place a substrate in an orientation
required for the desired enantioselective conversion and redesign the active site to
stabilize that geometry. In case of prochiral and meso substrates this could yield
enzymes that form a single product enantiomer due to regioselective or
facioselective reactivity. A protein sequence forming such redesigned active sites
can be generated by CPD methods such as implemented in RosettaDesign[11] and
described in several reviews[12]. Protein engineering efforts on enzymes catalyzing
enantioselective substitution reactions on prochiral and meso compounds have
been reported, including directed evolution to produce enantiocomplementary
enzyme pairs that allow formation of both product enantiomers[13]. The approach
included rational design and optimized directed evolution strategies, but in most
cases large mutant libraries had to be used[13]. Whereas computational design can
be used to obtain highly efficient mutant libraries for enhancing thermostability[14],
it remains to be explored if similar methods can be applied to produce
complementary enantioselective biocatalysts for desymmetrization reactions.
An important shortcoming of enzymes obtained by de novo computational
design is that their catalytic activity tends to be orders of magnitude lower than that
of natural enzymes[14, 15]. A main cause was revealed by MD simulations, which
indicated that the designed enzymes bind their substrate in multiple orientations,
most of which are unsuitable for catalysis. This differs from the more homogeneous
substrate orientations observed in natural enzymes and in enzymes improved by
directed evolution[15b, 16]. The abundance of undesired binding orientations in
designed enzymes was confirmed by crystal structures, which showed binding of
substrate analogs in orientations that deviate from the original design[17]. Solving
this problem would also make it possible to use CPD for applications where precise
positioning of the substrate is crucial, such as for obtaining enzymes displaying the
146

Enantioselective enzymes by computational design and screening

high regio or enantioselectivity that is often needed for the production of
intermediates for pharmaceuticals or developing inhibitors for drug design.

Scheme 61. Enzymatic regioselective hydrolysis of alicyclic meso epoxides 13,
resulting in enantioselective diols. The hydrolytic reaction proceeds with inversion
of the configuration. Attack at the (R) carbon thus results in an (S,S)diol while
attack at the (S) carbon results in the (R,R)diol.
A possible solution could be to apply a second independent ranking step after
the CPD. In this study we examine the use of MD simulations to predict substrate
positioning for enantioselective catalysis. While MD simulations are capable of
predicting behavior of enzymesubstrate complexes, the computational costs are
usually too high to cover functionally relevant timescales or to rank more than a
dozen of computationally designed variants. We recently described a
computationally inexpensive MD protocol for predicting substrate orientations
through simulating enzymesubstrate interactions using a large number (e.g. 20) of
short (10 100 ps) and differently initialized MD simulations. It turned out that
several of such short independent MD simulations sampled a much wider diversity
of substrate orientations than a single long MD simulation (e.g. 20 ns). This allowed
a more complete sampling of possible substrate orientations, which resulted in
better prediction of the enantioselectivity of existing enzymes[18] than the use of
standard single long MD simulations at a 100fold lower computational cost. We
refer to this protocol as HTMIMD (highthroughput multiple independent MD)[18].
In the current study, we demonstrate that HTMIMD can be used to obtain

147

Chapter 6

computationally designed enzyme variants that have high enantioselectivity in an

asymmetric synthesis reaction (Scheme 61).
To investigate the feasibility of integrating CPD with HTMIMD for the
redesign of an enantioselective enzyme, we selected a challenging test case of
biocatalytic relevance: the design of enantiocomplementary epoxide hydrolases for
the enantioselective transformation of cyclopentene oxide (1a) to yield either (R,R)
or (S,S)cyclopentane diol (1b) (Scheme 61). Since the stereochemical outcome of
the reaction is controlled by regioselectivity of water attack, accurate placement of
the substrate in the active site is essential. This is highly challenging because
epoxide 1a lacks bulky or polar groups that could position the substrate in a defined
way by directional interactions, such as through hydrogen bonds or electrostatic
attraction.

Scheme 62. Geometries for quantifying near attack conformations during MD
simulations of epoxide 1a bound enzymes. For hydrogen bonds h1h6, the distances
between the donor and acceptor during a NAC should be between 0 and 3.5 and
the angles between the OacceptorHdonorR atoms between 120 and 180. Other criteria
for a NAC are listed in the scheme.
To obtain enantioselective variants, we redesigned the limonene epoxide
hydrolase (LEH, PDBID 1NU3) to position substrate 1a such that the substrate is
orientated either in the proRR or proSS binding mode. The wildtype enzyme has
low enantioselectivity since the substrate can adopt diverse orientations, resulting
148

Enantioselective enzymes by computational design and screening

in a racemic product[19]. LEH functions by general acidbase catalysis using an Asp
ArgAsp catalytic triad (Scheme 62)[1920]. The Asp 132Arg 99 pair abstracts a
proton from a water molecule, triggering nucleophilic attack, and the epoxide
oxygen leaving group is protonated by Asp 101. We call this proRR attack (Scheme
61) if nucleophilic displacement occurs on the (S)configured carbon atom of
epoxide 1a, which leads to the (R,R) diol 1b by inversion of stereoconfiguration at
the reacting carbon atom[19a, 20]. Vice versa, proSS attack occurs on the (R)
configured carbon atom and produces (S,S)diol. The orientation of the substrate in
the active site determines at which carbon atom the nucleophilic attack will take
place, and relatively small differences in the positions of the water molecule and the
carbon atoms of the epoxide determine the enantioselectivity of the enzyme[13b, 19b].

Scheme 63. CASCO framework for redesign of catalytic selectivity by
computational protocols. In the current work, the adjustment of the CPD protocol
encompassed the introduction of steric hindrance to prevent unwanted substrate
binding modes.
In this study, we describe an approach we term CASCO (Catalytic selectivity
by computational design and MD screening for substrate positioning) for
engineering enzyme stereoselectivity. This strategy (Scheme 63) combines the use
of RosettaDesign for obtaining initial designs that can bind the substrate in the
desired orientation with HTMIMD to predict substrate binding modes and
selectivities. Subsequently, the most promising variants can be produced for
experimental characterization. If insufficient good variants are found, the design
protocol can be modified (Scheme 63). We observed that variants designed using
standard CPD by RosettaDesign did not position substrate 1a such that only (R,R)
149

Chapter 6

1b or (S,S)1b would be formed. Therefore, the CPD protocol was modified to

explicitly prevent the formation of the unwanted product. This was achieved by
designing steric clashes between the protein and the undesired orientation of the
substrate. Variants obtained using this modified procedure were predicted by MD
simulation to have more defined substrate orientations, in which the substrate
selectively bound in either the proSS or proRR nucleophilic attack orientation. The
37 most promising variants were experimentally constructed and the formed
products were analyzed by chiral gas chromatography, which revealed that 33 of
them were catalytically active and nine were highly regioselective, forming highly
enantioenriched products. This showed that the developed CASCO method and the
screening by HTMIMD is a valuable improvement for computational protein design.

Results
Prediction of LEH enantioselectivity by HTMIMD The use of HTMIMD for
predicting enantioselectivity in the conversion of substrate enantiomers by existing
haloalkane dehalogenases was recently explored[18, 21]. To examine if HTMIMD can
also be used to predict the enantioselectivity of cyclopentene oxide conversion by
LEH variants, which is based on regioselectivity of water attack[1920], we performed
MD simulations using data from LEH mutants that were obtained by directed
evolution[13b]. Briefly, during the MD simulations the fraction of the time was
recorded that the enzymesubstrate complex is in a proRR or proSS attack position.
This was quantified using near attack conformations (NACs), which are geometries
that approach the structure of the transition state of the reaction (Scheme 62)[22]. It
can be debated whether the presence of NACs is a requirement, a cause, or an effect
of transition state stabilization, but in each case their presence relates to productive
catalysis and the ratio between NACs for different enantiomers occurring during MD
simulations can be used to predict enzyme enantioselectivity[18, 21]. The NAC
geometries for proRR and proSS attack (Scheme 62) were based on earlier
published quantum mechanical modeling [19a, 23]. Based on the relative frequencies
of proRR and proSS NACs during MD simulation, the enantiomeric excess (ee) of
product enantiomers as given by Eq.1 can be predicted by Eq. 2.
ee

150

(1)

(2)

Enantioselective enzymes by computational design and screening

To investigate of the accuracy of the method, differences in activation
barriers for proSS and proRR epoxide ring opening were calculated from the
enantiomeric excess. This makes it possible to compare the predictions to highlevel
quantum mechanical calculations, which are known to provide high accuracy[19b] but
are too computationally expensive to apply to more than a handful of variants. Good
correlations (Figure 61) with experimental data were obtained both with 20 MD
simulations of 10 ps and 10 longer simulations 100 ps. The accuracy was similar to
those of QM/MM calculations[19b]. The high quality of the HTMIMD predictions is
consistent with the notion that MD simulations can model steric interactions
accurately and that steric factors, rather than electronic, determine which carbon
atom is preferentially attacked in LEH substrates[19a].

Figure 61. Differences in activation barriers for proSS and proRR epoxide ring
opening calculated with HTMIMD and with a quantum mechanical method
(B3LYP/6311G(2d,2p). The differences in activation energy between proRR and
proSS attack were calculated from published ee values of the produced diols via
E=RTln[(R,R)1b/(S,S)1b][13b]. The energy differences calculated from QM/MM
were reported by Himo and coworkers [19b]. HTMIMD predicted values are based on
the average eepred calculated from NAC frequencies obtained from either 20 different
MD simulations of each 10 ps or 10 differently initialized 100 ps MD simulations. Fit
lines are for 20 x 10 ps MD simulations, E,predicted = 0.7E + 1.5 with R2 = 0.92;
for the 10 x 100 ps MD simulations E,predicted = 0.7E + 1.7 with R2 = 0.90, for
QM/MM E,predicted = 1.4E +2.4 with R2 = 0.95.

151

Chapter 6

Computational design of substrate positioning Asymmetric enzymatic

synthesis of 1b requires an active site geometry that selectively positions substrate
1a for either proRR or proSS hydrolysis[13b, 1920]. Therefore the design procedure
aimed to identify a set of mutations that creates an active site cavity in which the
substrate is positioned selectively in one of the two catalytic orientations. To fulfill
this requirement, we used RosettaDesign[11] to design limonene epoxide hydrolase
variants that have shape complementarity to the substrate in the desired
orientation. Specifically, combinations of substitutions at eleven positions around
the active site (M32, L35, L74, M78, I80, V83, L103, L114, I116, F134, F139) to any
of the nine hydrophobic residues (AFGILMPVW) were investigated. These mutations
can result in 911 (= 3 1010) possible enzyme sequences and active site geometries.
The substrate was docked in the crystal structure of limonene epoxide hydrolase
either in a proRR or a proSS attack position, and RosettaDesign[11, 24] searched
within the sequenceconformational space for substratebound activesite
conformations with a low energy, as defined by the RosettaDesign energy function.
The use of search algorithms to find the low energy conformations within the
sequenceconformational space of a protein is a standard approach in CPD. It is
more likely that the designed threedimensional structure is correctly formed when
a design has a lower Rosetta energy. The algorithms that can be used to find these
low energy solutions within the sequenceconformation space of a protein, such as
Monte Carlo, have been reviewed[12e, 2425]. For LEH, such methods as implemented
in RosettaDesign produced 236 proSS designs and 230 proRR designs.
Screening designs by MD with NAC scoring While CPD protocols, which all aim
to lower the energy of the designed structure, have become standard, for enzymes
they often result in variants that bind the substrate in undesired binding modes[15b,
16b, 17, 26]. Therefore, the CASCO approach used HTMIMD to predict which of the

designed variants were best at selectively binding the substrate either in a proRR or
proSS attack orientation. This analysis indicated that the initially designed LEH
variants would indeed catalyze the desired reaction, but without displaying the
exclusive proRR or proSS regioselectivity required for enantioselective product
formation (Figure 62A). While there are a few variants for which a high ee is
predicted (Figure 62A), these variants occur as infrequently as that the non
selective wildtype enzyme is predicted to have high ee. The prediction that the
substrate will be bound aspecifically by the initial designs concurs with studies that
demonstrate that CPD can generate enzymes with multiple substrate binding
152

Enantioselective enzymes by computational design and screening

orientations[15b, 16b, 17, 26]. Inspection of MD simulation trajectories of the designed
LEH variants indicated that there was too much space around the (1R,2S)
cyclopentene oxide, which enabled the substrate to move both into proRR and pro
SS orientations.
Improved CPD procedure by incorporating steric restraints To improve the
CPD procedure, we decided to specifically introduce steric hindrance to prevent the
undesired substratebinding binding modes. Thus, proRR designs should
specifically hinder the proSS substrate orientation and vice versa. In this approach,
we first analyzed the effect of introducing bulky residues (W, F, or also Y if the
original residue was an F) for each of the 11 target residues lining the active site.
This was done by computationally predicting at which positions the rotamers of
these bulky side chains could sterically clash with 1a docked in a proRR or a proSS
attack position. This procedure indicated that the mutations F134W/Y, I116F/W,
L103F/W, and L114F/W could sterically hinder proRR attack, while the mutations
I80F/W, L103F/W, L35F/W, L74F/W, and M78F/W could prevent proSS attack.
The targeted introduction of these mutations was necessary, since these
substitutions only occurred in low frequencies amongst the initial designs, and
increasing the energy of unwanted substrate orientations is not part of the
RosettaDesign algorithm. Algorithms to automatically carry out such tasks are still
under development, and will aid future design projects[27]. After the introduction of
single mutations to enforce steric hindrance, the other ten residues were
computationally redesigned, which resulted in 945 proRR and 931 proSS designs.
The sequence profile of this set of variants differs from the set of initial designs
obtained without steric restraints.
A much larger fraction of the designs with enforced steric hindrance
produced the desired enantioselectivity in HTMIMD simulations (Figure 62B). Of
the designs without any enforced steric hindrance, 1.6% had an eepred of over 90%,
while over 20% of the designs with the enforced mutation L103F/W has a high
eepred for proRR attack. Furthermore, substitutions L35F/W, M78W, and I80F/W
increased the eepred for proSS attack to 10 to 20% of the designs. Only a minority of
the enforced mutations significantly increased the overall folding energy of the
designs suggesting the steric hindrance mutations were compatible with protein
stability and substrate binding in the desired orientation.

153

Chapter 6

Figure 62. HTMIMD based prediction of NACs and enantiomeric excess for proRR
attack. Each symbol represents one variant produced by RosettaDesign. A) variants
for proRR attack obtained by computational design without enforcing selective
steric hindrance; B) variants designed for proRR attack with a Phe or Trp
introduced at position 103 to selectively hinder binding in a proSS attack position.
Experimental characterization of computationally designed enzyme variants
To examine the quality of the design protocol, we expressed the designs for which
the highest enantioselectivities were predicted and measured their properties. For
this, the 10 designs with the highest predicted proRR selectivity and 27 proSS
designs were selected. Both designs that featured clear steric hindrance (proRR4
to proRR10, proSS6 to proSS27,Table 61) and designs that did not feature
clear steric hindrance but for which a high chiral selectivity was predicted (proRR
1 to proRR3, proSS1 to proSS5) were included.
154

Enantioselective enzymes by computational design and screening

Table 6-1. Computational predictions, experimentally observed ee values and thermostabilities of designed LEH variants.
NAC
pro-RR
(%)

NAC
pro-SS
(%)

ee
(%)

TM,app
(C)

M32L_L35M_M78I_F134W
M32L_M78F_I80V_L103W
M32L_M78L_F134W_F139L
M32L_M78G_L103F_I116V_F139L
M32L_M78G_L103F_F139M

54.5
54.3
54.5
3.49
1.43

1.96
0.028
1.55
0.00
0.00

5.6 (R,R)
3.3 (R,R)
9.7 (R,R)
49.1 (S,S)
14.4 (R,R)

68.0
69.5
71.0
67.0
60.0

pro-RR-6
pro-RR-7
pro-RR-8
pro-RR-9

M32A_M78G_L103F_F139L
M32L_L74I_I80V_L103F_F139W
M32L_L74I_I80V_L103F_F139L
M78I_I80V_L103F

0.86
0.65
5.95
0.58

0.00
0.00
0.00
0.01

10.4 (R,R)
54.4 (R,R)
84.5 (R,R)
73.2 (R,R)

59.0
60.5
66.0
57.5

pro-RR-10
pro-SS-1
pro-SS-2
pro-SS-3
pro-SS-4
pro-SS-5
pro-SS-6
pro-SS-7
pro-SS-8

M32L_L74I_L103F_F139W
M32L_M78L_I80F_L103I_I116V_F139L
M32L_L35W_L74F_M78F_I80A_L103I_F139L
M32L_L35M_L74I_M78L_I80F_V83I_L103I_I116V_F139L
M32L_L35M_L74I_M78L_I80F_L103I_I116V_F139L
M32L_L35W_M78L_I80A_I116V_F139W
M78W_F139W
M32L_L35M_L74I_M78I_I80F_V83I_L103V_I116V_F139L
M32L_I80W_L103I_I116V_F139W

4.51
0.43
1.86
1.24
1.35
3.38
0.01
0.01
0.04

0.11
57.3
55.5
51.0
46.0
45.5
18.7
1.20
5.25

47.1 (R,R)
77.6(S,S)
63.8(S,S)
81.4(S,S)
56.8(S,S)
70.1(S,S)
32.1(R,R)
62.8 (S,S)
B
NP

62.0
78.5
63.5
77.0
76.5
C
NO
62.0
75.0
76.5

pro-SS-9
pro-SS-10
pro-SS-11
pro-SS-12
pro-SS-13
pro-SS-14

M32L_L35M_L74F_I80F_V83I_L103I_I116V_F139L
M32L_L35W_L74W_M78A_I80A_L103I_I116V_F139L
M32L_L35M_L74I_M78V_I80F_V83G_L103V_F139L
M32L_M78I_I80F_L103I_I116V_F139L
M32L_L74F_I80F_L103I_I116V_F139L
M32L_L35A_M78L_I80W_L103I_I116V_F139L

0.24
0.17
0.02
0.18
0.04
0.64

33.16
20.30
2.65
16.2
3.42
48.1

78.8 (S,S)
62.4 (S,S)
13.9 (R,R)
74.7 (S,S)
78.3 (S,S)
62.0 (S,S)

71.0
60.0
NO
76.5
72.5
72.0

pro-SS-15
pro-SS-16
pro-SS-17
pro-SS-18
pro-SS-19
pro-SS-20

M32A_L35M_M78L_I80W_L103V_I116V_F139W
M32L_L35W_L74F_M78F_I80A_I116V_F139L
M78L_I80F_I116V_F139W
M32L_M78L_I80F_I116V_F139W
M32A_L35W_M78I_I80A_V83A_I116V
M32L_L35W_M78L_I80V_L103I_F139W

0.42
0.69
0.30
0.30
0.89
0.00

31.1
48.1
14.8
14.6
39.8
4.90

1.5 (R,R)
90.3 (S,S)
60.3 (S,S)
57.3 (S,S)
NP
44.6 (R,R)

61.0
68.0
63.0
70.0
48.5
NO

pro-SS-21
pro-SS-22
pro-SS-23
pro-SS-24
pro-SS-25
pro-SS-26

M32L_M78I_I80W_L103I_I116V_F139L
M32L_L35M_M78L_I80W_V83I_L103I_I116V_F139W
L74I_I80F_I116V_F139W
M32L_L35W_L103I_I116V_F139W
M32L_L35W_M78I_I80A_L103I_I116V_F139L
M32L_L35M_M78L_I80F_L103I_I116V_F139L

0.00
0.20
0.29
0.05
0.22
0.25

0.05
32.6
20.2
2.96
9.70
10.7

NP
NP
43.2 (S,S)
21.2 (S,S)
71.4 (S,S)
64.8 (S,S)

80.0
75.5
62.0
NO
64.5
75.5

pro-SS-27
WT (LEH-P)

M32L_L74F_M78A_I80F_L103I_I116V_F139L
None

0.21

8.94

78.6 (S,S)
23.5 (R,R)

71.0
68.5

Variant name

Computationally designed active site mutations

pro-RR-1
pro-RR-2
pro-RR-3
pro-RR-4
pro-RR-5

The designed variants were constructed in a previously engineered
thermostable variant of LEHP because use of a thermostable template increases the
chance that mutations are tolerated without formation of misfolded protein[28]. This
variant has a 20C higher apparent melting temperature while it has a similar
catalytic activity as the wildtype enzyme[14a]. The crystal structure does not show
significant deviations around the active site (R.J. Floor et al., Chapter 5 of this thesis).

155

Chapter 6

Conversion assays with the purified mutant enzymes with analysis by chiral GC
revealed that 33 of the 37 variants were catalytically active of which 28 (85 %) had
the designed chiral selectivity (proRR or proSS) (Table 61). Of these 28 variants,
nine (33 %) were highly selective, producing the desired diol with an ee of over
75%. The most enantioselective variants were proRR8 which contains five
mutations and produces (R,R)1b with an ee of 85% and proSS16 which has seven
mutations and forms (S,S)1b with 90% ee (Table 62). Variants proRR8 and pro
SS16 each have a bulky residue (L103F and L35W, respectively) that was
introduced to incorporate steric hindrance and prevent a binding mode that would
allow nucleophilic attack by water at the unwanted position ((2S) and (1R)carbon
atoms, respectively (Figure 64), which we consider essential for the high chiral
selectivity of these variants.

Table 6-2. Computationally predicted and experimentally observed enantioselectivities of limonene epoxide hydrolase
variants for the hydrolysis of cyclopentene oxide (1a)
Computational predictions

variant

active site mutations

Experimental analysis

NAC proRR (%)

NAC proSS (%)

major
product

ee
(%)

specific
activity
-

(molmin
1

WT
(LEH-P)

none

-1

mg )

(R,R)-1b

23.9

0.083 0.002

pro-RR-8

M32L/L74I/I80V/L103F/F139L

5.95

0.004

(R,R)-1b

85.5

0.015 0.001

pro-SS-16

M32L/L35W/L74F/M78F/I80A/I116V/F139L

0.69

48.1

(S,S)-1b

90.2

0.010 0.001

All mutations were introduced in the thermostable variant LEH-P.

MD simulations

[14a]

The NAC percentages are those from 10 100 ps

The results suggest that there is a good correlation between the predicted
and the experimental ee (Figure 61). For several variants, an erroneously high
eepred was observed (Table 61); this can occur due to conformational
undersampling of the enzymesubstrate complex during MD simulations[18].
However, experimental characterization of the mutant enzymes revealed that 90%
of the designs were catalytically active. Furthermore, 88% of the active proRR and
83% of the proSS designs had the designed selectivity (Table 61). Nine of these
designs were highly selective with an ee of over 75%. This confirms that the CASCO
156

Enantioselective enzymes by computational design and screening

approach can be used for producing mutant libraries highly enriched in variants
with the desired enantioselectivity.
The two most selective variants (proRR8 and proSS16) for either proRR
or proSS attack were characterized in more detail to compare rates, chiral
selectivity and protein production levels. Both variants were well expressed at 50
mg protein/L culture broth, which is comparable to the yield of variant LEHP[14a].
Using purified enzymes, the chiral selectivity for the formation of diol 1b was
determined. ProRR8 produced (R,R)1b with an ee of 85.5% while proSS16
showed an ee of 90.2% for the production of (S,S)1b (Figure 63). This
demonstrates that, respectively, 92% and 94% of the nucleophilic attack occurred
on the correct carbon atom (see equation 1). The two variants maintained good
catalytic activity for the hydrolysis of 1a, albeit with eightfold lower rates than
displayed by the thermostable LEHP from which they originated (Table 62).
Variants proRR8 and proSS16 had apparent melting temperatures of 69.5C and
66.5C respectively, while the parent LEHP had a melting temperature of 68.0C.
These results indicate that the five or seven introduced mutations do not have a
large effect on the protein stability.

Table 6-3. Catalytic rates and stereoselectivity of LEH variants for larger epoxides.

2a

Substrate
e.e. 2b (%)

.
Specific
activity B

WT (LEH-P)

3.1 (R,R)

2.93 0.21

pro-RR-8

74.7 (R,R)

0.29 0.01

pro-SS-16

93.5 (S,S)

0.18 0.01

3a
e.e. 3b
(%)
24.6 (S,S)
39.1
(R,R)
91.9 (S,S)

Specific
activity B

4a
d.e. 4b

Specific
activity B

0.583 0.004

(%)
97.7 (S,S)

0.030 0.001

97.1 (S,S)

2.6 0.1

0.007 0.001

99.4 (S,S)

62.7 1.4

55.2 2.5

Substrate converted in an enantioconvergent manner; both (1R,2S,4R)-4a and (1S,2R,4R)-4a are

converted to (1S,2S,4R)-4b due to different regioselectivity of nucleophilic attack for the substrate
enantiomers
B

[19a, 20b]
-1

.
-1

Units: mol min mg

The selectivity of the proRR8and proSS16 variants was also determined
for hydrolysis of the larger alicyclic epoxides 2a and 3a (Scheme 61) and the
natural substrate limonene epoxide 4a (Scheme 64). For 2 and 3, proRR8 was

157

Chapter 6

moderately (R,R) selective while proSS16 was highly (S,S)selective (Table 63).
The catalytic activities for hydrolysis of epoxides 2a and 3a were reduced up to
twenty fold compared to the wildtype LEHP. This is a larger decrease than
observed for the intended substrate 1a. The results show that proRR8 and proSS
16 are most improved for the enantioselective conversion of substrate 1a, which is
the substrate that they were designed for. However the variants are able to catalyze
the hydrolysis of larger epoxides, albeit with lower selectivity (Table 63).
At the same time, both designed variants remained highly selective for the
conversion of 4a to the (1S,2S,4R) enantiomer of 4b, which is the reaction catalyzed
with the highest rate by LEH (Scheme 64). Furthermore, proSS16 catalyzes the
hydrolysis of the natural substrate 4a (Table 63) with a similar rate as the wild
type LEHP. The undiminished catalytic activity for the native substrate of LEH by
proSS16 is in agreement with existing ideas that new enzyme activities can evolve
while an enzyme continuous to catalyze its original reaction[29].

Scheme 64. The hydrolysis of limonene epoxide rac4a by LEH.

Figure 63. Chiral gas chromatography elution profiles of 1bdiol enantiomers, as
produced by LEHP, variant proSS16 or variant proRR8, demonstrating high
enantioselectivity of the designed epoxide hydrolases. The plot displays FID
detector signal intensity versus the retention time.
158

Enantioselective enzymes by computational design and screening

Discussion
Comparison of computational design to directed evolution Enantioselective
catalysis is one of the most useful properties of enzymes for chemical synthesis, and
protein engineering methods that introduce or modify enzyme enantioselectivity
are highly relevant[13,

30].

Whereas directed evolution with highthroughput

screening appears highly successful, methods of CPD provide an opportunity to

design enzymes for nonnatural reactions but often lack the level of accurate
substrate positioning required for high selectivity [15b, 1617]. To accurately place the
substrate in a position required for regioselective attack we used RosettaDesign,
with additionally imposed steric constraints and ranking by highthroughput MD
simulations. This CASCO framework enabled the design of highly enantioselective
enzymes without the need for a large screening effort.
The most selective variants obtained, proRR8 and proSS16, produced the
desired diol with an ee of 86% and 90%, respectively (Table 62), while variants
that were discovered by directed evolution[13b], H173 and H178, produced (R,R)
and (S,S)cyclopentanediol with an ee of 80% and 93 %, respectively. Thus, the
enantioselectivities of the best variants obtained by CPD are as good as those
obtained by directed evolution. The variants proRR8 and proSS16 each only have
only one substitution in common with the most selective directed evolution
mutants. These mutations (L74I in proRR8 and H173, I116V in proSS16 and
H178) do not have a large influence on the shape of the active site (see below).
Whereas the outcomes of directed evolution and computational approach are
similar in terms of enzyme performance, they differ in the sequences of the enzyme
that are produced and especially also in the experimental screening effort, which
encompassed 4,700 variants examined by chiral GC for directed evolution[13b] but
only 37 mutants for the CASCO CPD approach described here. Highthroughput
screening of enzymes variants is often laborious, especially if the required protein
expression host has a low transformation efficiency and if the performance assays
are unsuitable for miniaturization. Furthermore, CPD can introduce large jumps in
sequence space in a single round, which is especially important when multiple
mutations acting in synergy provide the desired change in functionality[31].
HTMIMD for scoring designs Within the CASCO framework (Scheme 63), all
variants produced by RosettaDesign are subjected to a computationally inexpensive
HTMIMD protocol that samples the occurrence of nearattack conformations. MD
simulations were done in multiple trajectories with independent initialization to

159

Chapter 6

prevent trapping in local energy minima and to ensure effective sampling of

conformational space. This MDbased ranking adds significantly to the scoring and
selection of variants obtained by RosettaDesign[11, 24]. First, MD simulations allow
protein structures to escape from local energy minima and sample different parts of
conformational space. Second, explicit water is present during the MD simulations,
which can compete with internal protein hydrogen bonds and proteinsubstrate
hydrogen bonds and thus will provide a more accurate modeling of solvent
interactions than the implicit solvation energy model used within Rosetta. Third, the
HTMIMD ranking uses a fully flexible proteinbackbone and substrate, whereas
Rosetta backbone sampling allows only restricted backbone motions, which is
considered an important limitation[17a]. Finally, the MD simulations use a different
energy function compared to the Rosetta molecular mechanics function. Thus, the
MD ranking is orthogonal to the Rosetta scoring function and provides an additional
filter before experimental verification is pursued.
Origin of enantioselectivity To discover structural causes of the change in
enantioselectivity, the modeled structures of the two most selective variants were
investigated. For variant proRR8, this analysis showed that major effects are
introduced by the mutations L74I and I80V, which create space for binding of the
substrate in the proRR orientation while binding in the undesired orientation is
prevented by the L103F mutation (Figure 64A). In variant proSS16, the residues
introduced by the mutations L35W and L74F sterically hinder the binding of 1a in
the proRR orientation. The substitution M78F prevents a potential clash with the
introduced sidechain of F74. Concurrently, the smaller side group introduced by
the mutation I116V enlarges the pocket where 1a can bind in the proSS orientation
(Figure 64B). Furthermore, the mutation I80A creates a pocket in the inside of the
enzyme, which makes it possible to introduce a tryptophan at position 35 (by
mutation L35W), whereas in the wildtype I80 would clash with W35. Such coupled
mutations may be required for modifying the selectivity of enzyme active sites and
to accommodate new substrates[31]. The other mutations such as M32L and F139L
likely optimize the packing around the mutated residues. While such dependent
mutations are often required for introducing new enzyme activities[31], their
introduction can be problematic when using enzyme engineering methods such as
rational design or directed evolution, since it is difficult to foresee the effects that
small mutations have on others.

160

Enantioselective enzymes by computational design and screening

Figure 64. Predicted structures of the active sites of limonene epoxide hydrolase
variants A) proRR8 and B) proSS16. The docked substrate 1a is shown in yellow.
The water molecule performing the nucleophilic attack is also indicated. Residues
that were introduced to prevent binding in the opposite pose are labeled in bold.
Conclusions We successfully developed a protocol for redesign of catalytic
selectivity by computation (CASCO) which involves docking of substrate in the
active site in a nearattack conformation required for selective catalysis and
computational design of a large set of multisite mutants that are predicted to
stabilize this reactive pose. Furthermore, the procedure prevents unwanted binding
modes by incorporating steric hindrance and performs in silico ranking of these
primary designs for enantioselectivity by molecular dynamics simulations[18]. The
use of highthroughput molecular dynamics with independent initialization (HTMI
MD) provided, at low computational cost, an independent ranking scheme that was
instrumental both for improving the design methodologies and for selecting the
most promising variants. The use of this combined design and screening protocol
resulted in the rapid discovery of enantiocomplementary epoxide hydrolase
variants. In total only 37 different mutants carrying three to nine substitutions
needed to be screened experimentally to obtain tailored epoxide hydrolases that
produced (R,R) or (S,S)1b with high enantiomeric excess. We anticipate that this
CASCO strategy will facilitate protein engineering efforts aiming at redesigning
product profiles of enzymes relevant for applied biocatalysis.

161

Chapter 6

Materials and methods

CPD: An ensemble of different conformations of cyclopentene oxide was generated using
Openeyes Omega software[32]. Only the chair conformation of cyclopentene oxide was used
for CPD since it is lower in potential energy than the boat conformation[19a]. Furthermore, an
ensemble of different enzyme conformations was used as template for the computational
design. This was implemented by performing the computational design both on the Xray
structure of wild type LEH (pdb code 1NWW) and on structures of 55 snapshots collected
from five independent 1 ns MD simulations, carried out using cyclopentene oxide bound
wildtype LEH. Prior to MD simulation, this substrate was docked using Rosetta enzyme
design[11]. The snapshots from MD simulations were collected every 50 ps between 500 and
1000 ps. During these five MD simulations, the enzyme remained in catalytically active
conformations as verified by NAC analysis. The use of multiple starting structures of the
enzyme increased the number of designs with unique aminoacid sequences.
The ideal geometry for LEH catalysis corresponded to the NAC criteria described in
Scheme 62 and was based on published QM/MM calculations[19a, 33]. Rosetta enzyme design
oriented the substrate optimally for either proSS or proRR attack by applying in silico
forces such that the desired carbon atoms was in NAC for catalytic attack while the other
epoxide carbon atom was not. The distance between water and targeted carbon atom was
constrained to 2.7 0.4 while the distance between water and the other epoxide carbon
atom was maintained at 4.5 0.7 , both with a force constant of 100 Rosetta Energy Units
(REU) 2[11]. For other constraints, 10fold smaller force constants (10 REU 2) were
sufficient to keep the substrate in a NAC. The distance between epoxide oxygen and the
oxygen of Asp101 was maintained at 3.0 0.2 . The Hbond distances between the
nucleophilic water and Tyr53, Asn55, and Asp132 (h2h6) were set to 1.8 0.2 . The
dihedral between the epoxide oxygen, the nonattacked carbon atom, the attacked carbon
atom, and the oxygen atom of the water molecule was 180 20 with a force constant of 10
REU degree2.
Ranking of the designs using HTMIMD: Rosetta enzyme design automatically provides
3D structures of the designs which were used for subsequent ranking. The ranking and
selection within RosettaDesign eliminates all structures that violate the sum of penalty
energies for the above constraints by more than 15 REU, after an energy minimization. The
remaining designs were selected for further analysis by MDsimulations. If more than one
design had the same amino acid sequence, both the design with the least violations of
constraints and the design with the lowest overall energy score were selected for ranking
by HTMIMD.
All MD simulations were carried out with explicit water (TIP3P) and counter ions
(0.5 % NaCl), furthermore longrange electrostatics were enabled. The designed enzymes
were positioned in rectangular simulation cells with at least 7.5 between protein and the

162

Enantioselective enzymes by computational design and screening

periodic boundary of the simulation cell. All original water molecules present in the wild
type LEH structure (pdb: 1NWW) were placed back at the original positions, unless they
clashed with the designed protein structure (as indicated by a distance longer than 1.5 to
any heavy atom of the protein). The salt ions were positioned at electrostatically favorable
positions as calculated by Yasara[34]. To prevent salt ions from interacting with charged
active site residues, no salt ions were positioned in the active site or its immediate vicinity
(criterion: more than 15 away from Asp 101). An energy minimization was carried out
prior to the MD simulation as described previously[35] . All MD simulations were performed
under Yasara using the Yamber3 forcefield[34a], with a leapfrog algorithm using timestep
of 1.33 fs. The nonbonded interaction list was updated every three timesteps and a
Berendsen thermostat was applied to preserve neutral pressure temperature conditions[34a,
36].

LINCS and SETTLE algorithms prevented hydrogen atoms from heating up[37].

Furthermore, the charges of the substrate were assigned with the AM1BCC method[38] and
a particle mesh Ewald algorithm was used to calculate the long range (> 7.86 )
electrostatics interactions[39].
For every enzyme variant, multiple independent simulation trajectories were
generated by starting MD simulations with different initial atom velocities[18]. For the ultra
short (10 ps) MD simulations, the temperature was gradually increased from 5 to 298 K in
the first 3 ps of the MD simulation. The next 2 ps was used to equilibrate the temperature,
after which NACs were sampled on the fly every 20 fs for the next 5 ps (only geometric
information was saved, no snapshots were recorded). Earlier, this protocol was found to be
optimal for thoroughly sampling enzymesubstrate conformations at low computational
cost[18]. For longer MD simulations, the temperature was gradually increased from 5 to 298
K in the first 30 ps of the simulation. For the screening, the first 20 MD simulations of 10 ps
were used to predict enantioselectivity using equation 2.
Subsequently, the best variants were selected for a 2nd stage of ranking, which
consisted of 10 longer simulations of 100 ps, of which the last 50 ps were used to sample
NACs. The longer equilibration time used during the 100 ps simulation (50 instead of 5 ps)
increased the chance for the designed structures to escape from a local energy minimum.
Therefore these simulations are complementary to the very short MD simulations. For this
stage, proRR designs were selected which had a NAC during more than 0.5% of the analysis
time and a predicted ee of > 70 % during the 20 MD simulations of 10 ps. For proSS designs
the selection criteria were: NACs during more than 2.5% of the simulation time and a
predicted ee of > 90 % were. These settings were more stringent since more potentially
good designs were found for proSS variants. Variants for which the highest ee was
predicted in round 2 were selected for experimental characterization.
Construction of designed LEH mutants: All epoxide hydrolase mutants predicted by CPD
were constructed in a wildtype background containing the previously described stabilizing

163

Chapter 6

mutations S15P, A19K, E45K, T76K, T85V, N92K, Y96F and E124D[14a]. These mutations
improve the thermostability while their combined introduction did not reduce the catalytic
activity or selectivity for the hydrolysis of epoxide 4[14a]. Mutants were grouped according to
the common ancestor from whom they could be constructed. These common ancestors
were created by singlesite mutagenesis using a QuikChange kit (PfuUltra Hotstart PCR
Master Mix #600630, Agilent, CA, USA) using the LEH gene cloned in a pBAD/MycHisC
expression plasmid[14a]. Subsequently, all variants could be created from these ancestors by
one to three additional rounds of QuikChange mutagenesis. For all reactions, the
QuikChange master mix was used as recommended by the supplier. The sequence of the
resulting plasmids was analyzed by DNA sequencing. Once the correct plasmid was
obtained, it was transformed into E. coli TOP10 cells for expression and protein production.
Protein purification: Overnight E. coli cultures containing plasmids encoding the LEH
variants of interest were grown in deep well plates containing 1 mL of LB, containing 50
g/mL ampicillin. For protein production, 5 mL of TB medium (24 g/L yeast extract, 12 g/L
tryptone, 4 mL/L glycerol, 0.017 M KH2PO4, 0.072 M K2HPO4, 50 g/mL ampicillin, pH 7.0)
was inoculated with 1% of this overnight culture. Subsequently, the cultures were placed in
a 37C incubator, which was shaken at 250 rpm. When the culture reached an OD600 of 0.6,
protein production by the cells was induced by the addition of 0.02% Larabinose. The
culture was incubated for a further 16 h at 30C. Subsequently, cells were harvested by
centrifugation (15 min, 15,000 g, 4C), resuspended in 1 mL buffer (50 mM Hepes, pH 8.0,
500 mM NaCl) and lysed using sonication (1 total time, cycles of 5 on 10 off, 60 watt, 20
kHz, 4C). Finally, cell debris was removed by centrifugation (15 min, 15,000 g rpm, 4C)
and the obtained cell free extract was transferred to clean 1.5 mL Eppendorf tubes.
Protein was purified out of cellfree extract using immobilized metal ion affinity
chromatography in microcentrifuge spin columns (Thermo Fisher Scientific, USA),
containing 200 l NiNTA resin (Qiagen, Germany). The cellfree extract was applied on the
column and the tube containing the column was centrifuged for 2 min at 200 g. The column
was washed three times with wash buffer (50 mM Hepes, pH 8.0, 20 mM imidazole, 500 mM
NaCl) to remove unbound contaminants and the protein was eluted by applying 400 l
elution buffer on the column (50 mM Hepes, pH 8.0, 300 mM imidazole, 500 mM NaCl).
Finally, the protein was concentrated using 10 kDa centricon filters (Merck
Millipore, MA, USA) and desalted by four cycles of diafiltration using 50 mM Hepes, pH 8.0,
buffer. The protein concentration was determined using the Bradford procedure, and purity
was analyzed using SDSPAGE. This procedure typically yielded 250 g of purified protein
per mutant.
For validation of the results obtained during the screening, variants proRR8 and
proSS16 were purified on larger scale using methods reported previously[14a]. Briefly, this
consisted of protein production in E. coli TOP10 cells and subsequent purification of the

164

Enantioselective enzymes by computational design and screening

LEH variants by immobilized metal ion affinity chromatography. Subsequently, the purity of
the resulting protein preparations was analyzed by SDSPAGE (Figure 65). This procedure
yielded 50 mg protein/l culture for both LEHP and the two variants investigated at larger
scale.

Figure 65. SDSpage of purified LEH variants proSS16 and proRR8. The sizes of the
proteins in the molecular weight marker (M, in kDa) are indicated to the left of the gel.
Analysis of catalytic activity: To analyze the catalytic activity of the different variants, a
500 mM stock solution of epoxides 14 in acetonitrile was made. Buffer (50 mM Hepes, pH
8.0) and the substrate solution were mixed resulting in a concentration of 20 mM substrate.
Subsequently 1002000 g enzyme was added (depending on the efficiency of the enzyme
for the substrate) to initiate conversion in a 10 ml volume. Ten aliquots of 1 mL were
transferred to different Eppendorf tubes and incubated at 30C. At regular intervals, a tube
was removed from the incubator and the reaction was quenched by the addition of 800 l
ethylacetate, supplemented with 1 mM dodecane as internal standard. The organic layer
was removed and dried using Na2SO4. Subsequently, epoxide and diol content in the organic
layer was determined using gas chromatography.
To determine epoxide and diol concentrations, 2 l of the ethylacetate extract was
injected into a type 2014 gas chromatograph (Shimadzu, Kyoto, Japan) equipped with a
Heliflex AT5 column (Alltech Associates, Inc., IL, USA) using flame ionization detection. The
oven temperature programs used for the separation of the different compounds were as
follows. For 1 and 2: 4 min at 40C, subsequently heating the column at 10C/min to 90C
and at 20C/min to 250C; for 3: an isothermal program of 20 min at 140C followed by
heating at 20C/min to 250C; for 4: 4 min at 90C followed by heating at 10C/min to
160C and at 20C/min to 250C. The retention times of epoxides and diols were as follows:
epoxide1a 5.7 min, diol1b 12.7 min; 2: epoxide 9.7 min, diol 13.7 min; 3 epoxide 12.5 min,

165

Chapter 6

diol 15.3 min; 4 epoxide 6.0 min, diol 9.5 min. The obtained peak areas were converted into
concentrations using calibration curves constructed by extracting and analyzing known
concentrations of epoxides and diols.
Analysis of enantioselectivity: Chiral GC was used to determine the enantioselectivity of
LEH variants for the hydrolysis of the prochiral acyclic epoxides 14. For initial screening,
200 l of a 1 mg/ml sample of purified protein was mixed with 280 l buffer (50 mM Hepes,
pH 8.0). For confirmation of the obtained enantioselectivity more enzyme was used: 480 l
of a 2.5 mg/ml sample of enzyme. To these reactions, 20 l liquid 1a (0.21 mmol) was
added, or equivalent amounts of 2a, 3a or 4a. These reaction mixtures were incubated for 1
h at 30C. Subsequently, 500 l 4 M NaCl was added and the samples were extracted three
times by the addition of 800 l ethylacetate. The organic layers were removed, combined,
dried by the addition of Na2SO4 and evaporated under vacuum in a speedvec centrifuge. The
obtained residue was resuspended in 100 l ethylacetate supplemented with 1 mm
dodecane as internal standard.
Of the ethylacetate extracts, 2 l was injected on a Hydrodex bTBDAc column
(Aurora Borealis, The Netherlands) in an Agilent HP 6890N GC using FID detection. The
enantiomers of 1b were separated during an isothermal incubation step of 10 min at 150C.
During this procedure, the injector and detector temperatures were kept at 250C. Using
these conditions, the two enantiomers of 1b could be baseline separated, with retention
times of 6.5 and 6.8 min for the (S,S) and (R,R) enantiomers, respectively. The enantiomeric
excess was calculated using equation 1.
The enantiomeric composition of diols 2b and 3b was measured using the same
procedure. The retention times for the diols were: 4.7 and 4.9 min for (S,S) and (R,R)
enantiomers of 2b; 6.2 min and 6.5 min for the (S,S) and (R,R) enantiomers 3b. At least one
of the enantiomers of diols 1b4b was commercially available from Sigma Aldrich (MO,
USA) and was used as reference compound. All assays were performed in duplo. The
diastereomeric excess of the diol obtained after hydrolysis of epoxide4a was determined as
reported previously[14a].
Determination of protein melting temperatures: Apparent protein melting temperatures
were determined using the thermofluor method[40]. This method is based on the increase of
the fluorescence of the dye Sypro Orange, once it binds to the hydrophobic protein interior.
This part becomes exposed due to the protein unfolding. For this, 5 L of 100fold diluted
Sypro Orange (Life Technologies, CA, USA) was mixed with 20 L purified enzyme, with a
protein concentration of 0.52.5 mg/ml. This sample was transferred to a 96well plate (iQ
PCR plate, Biorad, CA, USA) and the plate was sealed with transparent foil (iQ 96well PCR
Plate seal, Biorad). Subsequently, the fluorescence (excitation at 490 nm and emission at
575 nm) was monitored, while the sample was heated from 20 to 99C at 1.1C/min in a

166

Enantioselective enzymes by computational design and screening

CFX96 QPCR machine (Biorad). The apparent melting temperature (TM,app) was defined as
the maximum of the first derivative of the fluorescence with respect to the temperature.

Acknowledgements
This work was supported by the European Union 7th framework projects
Metaexplore (KBBE20073305, 222625) and Kyrobio (KBBE20115, 289646), by
BEBasic and by NWO (Netherlands Organization for Scientific Research) through an
ECHO grant.
Author contributions: HJW wrote and adapted the algorithms and performed
the computational screening, partially based on methods initially developed by SB
and DB. RJF constructed and characterized the enzyme variants and designed the
laboratory experiments together with DBJ and HJW. HJW, DB, SB, SJW, and DBJ
designed the in silico approach. HJW, RJF and DBJ wrote the manuscript and SB, SJM
and DB corrected it.

References
[1]
[2]

[3]

[4]

[5]

[6]

[7]
[8]
[9]
[10]
[11]

D. N. Bolon, S. L. Mayo, Proc. Natl. Acad. Sci. USA 2001, 98, 1427414279.
D. Rothlisberger, O. Khersonsky, A. M. Wollacott, L. Jiang, J. DeChancie, J. Betker, J. L.
Gallaher, E. A. Althoff, A. Zanghellini, O. Dym, S. Albeck, K. N. Houk, D. S. Tawfik, D.
Baker, Nature 2008, 453, 190195.
L. Jiang, E. A. Althoff, F. R. Clemente, L. Doyle, D. Rthlisberger, A. Zanghellini, J. L.
Gallaher, J. L. Betker, F. Tanaka, C. F. Barbas, D. Hilvert, K. N. Houk, B. L. Stoddard, D.
Baker, Science 2008, 319, 13871391.
J. B. Siegel, A. Zanghellini, H. M. Lovick, G. Kiss, A. R. Lambert, J. L. S. Clair, J. L.
Gallaher, D. Hilvert, M. H. Gelb, B. L. Stoddard, K. N. Houk, F. E. Michael, D. Baker,
Science 2010, 329, 309313.
a) J. Ashworth, J. J. Havranek, C. M. Duarte, D. Sussman, R. J. Monnat, B. L. Stoddard,
D. Baker, Nature 2006, 441, 656659; b) J. Ashworth, G. K. Taylor, J. J. Havranek, S. A.
Quadri, B. L. Stoddard, D. Baker, Nucleic Acids Res. 2010, 38, 56015608; c) S. B.
Thyme, J. Jarjour, R. Takeuchi, J. J. Havranek, J. Ashworth, A. M. Scharenberg, B. L.
Stoddard, D. Baker, Nature 2009, 461, 13001304; d) E. FajardoSanchez, F. Stricher,
F. Pques, M. Isalan, L. Serrano, Nucleic Acids Res. 2008, 36, 21632173.
a) P. M. Murphy, J. M. Bolduc, J. L. Gallaher, B. L. Stoddard, D. Baker, Proc. Natl. Acad.
Sci. USA 2009, 106, 92159220; b) S. D. Khare, Y. Kipnis, P. Greisen, Jr., R. Takeuchi,
Y. Ashani, M. Goldsmith, Y. Song, J. L. Gallaher, I. Silman, H. Leader, J. L. Sussman, B. L.
Stoddard, D. S. Tawfik, D. Baker, Nat. Chem. Biol. 2012, 8, 294300.
S. R. Gordon, E. J. Stanley, S. Wolf, A. Toland, S. J. Wu, D. Hadidi, J. H. Mills, D. Baker, I.
S. Pultz, J. B. Siegel, J. Am. Chem. Soc. 2012, 134, 2051320520.
D. P. Nannemann, K. W. Kaufmann, J. Meiler, B. O. Bachmann, Protein Eng. Des. Sel.
2010, 23, 607616.
M. J. Grisewood, N. P. Gifford, R. J. Pantazes, Y. Li, P. C. Cirino, M. J. Janik, C. D.
Maranas, PLoS ONE 2013, 8, e75358.
L. Liu, P. Murphy, D. Baker, S. Lutz, Chem. Commun. 2010, 46, 88038805.
F. Richter, A. LeaverFay, S. D. Khare, S. Bjelic, D. Baker, PloS one 2011, 6, e19230.

167

Chapter 6
[12]

[13]

[14]

[15]

[16]

[17]

[18]
[19]
[20]

[21]

168

a) W. S. Mak, J. B. Siegel, Curr. Opin. Struct. Biol. 2014, 27, 8794; b) D. Hilvert, Annu.
Rev. Biochem 2013, 82, 447470; c) A. Zanghellini, Curr. Opin. Biotechnol. 2014, 29,
132138; d) G. Kiss, N. elebilm, R. Moretti, D. Baker, K. N. Houk, Angew. Chem.
Int. Ed. 2013, 52, 57005725; e) H. J. Wijma, D. B. Janssen, FEBS J. 2013, 280, 2948
2960.
a) J. G. E. van Leeuwen, H. J. Wijma, R. J. Floor, J. M. van der Laan, D. B. Janssen,
ChemBioChem 2012, 13, 137148; b) H. Zheng, M. T. Reetz, J. Am. Chem. Soc. 2010,
132, 1574415751; c) G. J. Williams, T. Woodhall, L. M. Farnsworth, A. Nelson, A.
Berry, J. Am. Chem. Soc. 2006, 128, 1623816247; d) G. P. Horsman, A. M. F. Liu, E.
Henke, U. T. Bornscheuer, R. J. Kazlauskas, Chem. Eur. J. 2003, 9, 19331939; e) M.
Alexeeva, A. Enright, M. J. Dawson, M. Mahmoudian, N. J. Turner, Angew. Chem. Int.
Ed. 2002, 41, 31773180.
a) H. J. Wijma, R. J. Floor, P. A. Jekel, D. Baker, S. J. Marrink, D. B. Janssen, Protein Eng.
Des. Sel. 2014, 27, 4958; b) J. E. Diaz, C. S. Lin, K. Kunishiro, B. K. Feld, S. K.
Avrantinis, J. Bronson, J. Greaves, J. G. Saven, G. A. Weiss, Protein Sci. 2011, 20, 1597
1606; c) B. Borgo, J. J. Havranek, Proc. Natl. Acad. Sci. USA 2012, 109, 14941499; d)
H. J. Wijma, R. J. Floor, D. B. Janssen, Curr. Opin. Struct. Biol. 2013, 23, 588594; e) X.
Song, Y. Wang, Z. Shu, J. Hong, T. Li, L. Yao, PLoS Comput. Biol. 2013, 9, e1003129; f)
B. Liu, J. Zhang, Z. Fang, L. Gu, X. Liao, G. Du, J. Chen, J. Ind. Microbiol. Biotechnol.
2013, 40, 697704; g) R. J. Floor, H. J. Wijma, D. I. Colpa, A. RamosSilva, P. A. Jekel,
W. Szymaski, B. L. Feringa, S. J. Marrink, D. B. Janssen, ChemBioChem 2014, 15,
16601672.
a) I. V. Korendovych, D. W. Kulp, Y. Wu, H. Cheng, H. Roder, W. F. DeGrado, Proc. Natl.
Acad. Sci. USA 2011, 108, 68236827; b) H. K. Privett, G. Kiss, T. M. Lee, R. Blomberg,
R. A. Chica, L. M. Thomas, D. Hilvert, K. N. Houk, S. L. Mayo, Proc. Natl. Acad. Sci. USA
2012, 109, 37903795; c) F. Richter, R. Blomberg, S. D. Khare, G. Kiss, A. P. Kuzin, A. J.
T. Smith, J. Gallaher, Z. Pianowski, R. C. Helgeson, A. Grjasnow, R. Xiao, J.
Seetharaman, M. Su, S. Vorobiev, S. Lew, F. Forouhar, G. J. Kornhaber, J. F. Hunt, G. T.
Montelione, L. Tong, K. N. Houk, D. Hilvert, D. Baker, J. Am. Chem. Soc. 2012, 134,
1619716206; d) S. Bjelic, L. G. Nivn, N. elebilm, G. Kiss, C. F. Rosewall, H. M.
Lovick, E. L. Ingalls, J. L. Gallaher, J. Seetharaman, S. Lew, G. T. Montelione, J. F. Hunt,
F. E. Michael, K. N. Houk, D. Baker, ACS Chemical Biology 2013, 8, 749757.
a) G. Kiss, D. Rthlisberger, D. Baker, K. N. Houk, Protein Sci. 2010, 19, 17601773; b)
J. Z. Ruscio, J. E. Kohn, K. A. Ball, T. HeadGordon, J. Am. Chem. Soc. 2009, 131, 14111
14115; c) M. P. Frushicheva, J. Cao, Z. T. Chu, A. Warshel, Proc. Natl. Acad. Sci. USA
2010, 107, 1686916874.
a) R. Blomberg, H. Kries, D. M. Pinkas, P. R. E. Mittl, M. G. Grutter, H. K. Privett, S. L.
Mayo, D. Hilvert, Nature 2013, 503, 418421; b) L. Wang, E. A. Althoff, J. Bolduc, L.
Jiang, J. Moody, J. K. Lassila, L. Giger, D. Hilvert, B. Stoddard, D. Baker, J. Mol. Biol.
2012, 415, 615625.
H. J. Wijma, S. J. Marrink, D. B. Janssen, J. Chem. Inf. Model. 2014, 54, 20792092.
a) K. H. Hopmann, B. M. Hallberg, F. Himo, J. Am. Chem. Soc. 2005, 127, 1433914347;
b) M. E. S. Lind, F. Himo, Angew. Chem. Int. Ed. 2013, 52, 45634567.
a) M. Arand, B. M. Hallberg, J. Y. Zou, T. Bergfors, F. Oesch, M. J. Van der Werf, J. A. M.
de Bont, T. A. Jones, S. L. Mowbray, EMBO J. 2003, 22, 25832592; b) M. J. Van der
Werf, R. V. A. Orru, K. M. Overkamp, H. J. Swarts, I. Osprian, A. Steinreiber, J. A. M. de
Bont, K. Faber, Appl. Microbiol. Biotechnol. 1999, 52, 380385.
A. Westerbeek, W. Szymaski, H. J. Wijma, S. J. Marrink, B. L. Feringa, D. B. Janssen,
Adv. Synth. Catal. 2011, 353, 931944.

Enantioselective enzymes by computational design and screening

[22]
[23]
[24]

[25]
[26]
[27]
[28]
[29]

[30]

[31]
[32]
[33]
[34]

[35]
[36]
[37]
[38]
[39]
[40]

S. Hur, T. C. Bruice, Proc. Natl. Acad. Sci. USA 2003, 100, 1201512020.
B. Schitt, T. C. Bruice, J. Am. Chem. Soc. 2002, 124, 1455814570.
A. LeaverFay, M. J. O'Meara, M. Tyka, R. Jacak, Y. Song, E. H. Kellogg, J. Thompson, I.
W. Davis, R. A. Pache, S. Lyskov, J. J. Gray, T. Kortemme, J. S. Richardson, J. J.
Havranek, J. Snoeyink, D. Baker, B. Kuhlman, Methods Enzymol. 2013, 523, 109143.
C. A. Voigt, D. B. Gordon, S. L. Mayo, J. Mol. Biol. 2000, 299, 789803.
O. Khersonsky, D. Roethlisberger, A. M. Wollacott, P. Murphy, O. Dym, S. Albeck, G.
Kiss, K. N. Houk, D. Baker, D. S. Tawfik, J. Mol. Biol. 2011, 407, 391412.
A. LeaverFay, R. Jacak, P. B. Stranges, B. Kuhlman, Plos One 2011, 6, e20937
e20937.
N. Tokuriki, F. Stricher, L. Serrano, D. S. Tawfik, PLoS Comput. Biol. 2008, 4,
e1000002.
a) D. S. Tawfik, O. Khersonsky, Annu. Rev. Biochem 2010, 79, 471505; b) R. J.
Kazlauskas, U. T. Bornscheuer, in Comprehensive Chirality (Eds.: E. M. Carreira, H.
Yamamoto), Elsevier, Amsterdam, 2012, pp. 465480.
a) U. T. Bornscheuer, G. W. Huisman, R. J. Kazlauskas, S. Lutz, J. C. Moore, K. Robins,
Nature 2012, 485, 185194; b) M. T. Reetz, Angew. Chem. Int. Ed. 2011, 50, 138174;
c) L. G. Otten, F. Hollmann, I. W. Arends, Trends Biotechnol. 2010, 28, 4654.
M. T. Reetz, Angew. Chem. Int. Ed. 2013, 52, 26582666.
P. C. D. Hawkins, A. G. Skillman, G. L. Warren, B. A. Ellingson, M. T. Stahl, J. Chem. Inf.
Model. 2010, 50, 572584.
B. Schitt, T. C. Bruice, J. Am. Chem. Soc. 2002, 124, 1455814570.
a) E. Krieger, T. Darden, S. B. Nabuurs, A. Finkelstein, G. Vriend, Proteins 2004, 57,
678683; b) E. Krieger, J. E. Nielsen, C. A. Spronk, G. Vriend, Journal of molecular
graphics & modelling 2006, 25, 481486.
A. Westerbeek, W. Szymanski, H. J. Wijma, S. J. Marrink, B. L. Feringa, D. B. Janssen,
Adv. Synth. jCatal. 2011, 353, 931944.
H. J. C. Berendsen, J. P. M. Postma, W. F. Van Gunsteren, A. Dinola, J. R. Haak, J. Chem.
Phys. 1984, 81, 36843690.
a) B. Hess, H. Bekker, H. J. C. Berendsen, J. Fraaije, J. Comput. Chem. 1997, 18, 1463
1472; b) S. Miyamoto, P. A. Kollman, J. Comput. Chem. 1992, 13, 952962.
A. Jakalian, D. B. Jack, C. I. Bayly, J. Comput. Chem. 2002, 23, 16231641.
U. Essmann, L. Perera, M. L. Berkowitz, T. Darden, H. Lee, L. G. Pedersen, J. Chem.
Phys. 1995, 103, 85778593.
U. B. Ericsson, B. M. Hallberg, G. T. DeTitta, N. Dekker, P. Nordlund, Anal. Biochem.
2006, 357, 289298.

169

Chapter 6

170