Induced allostery in the directed evolution of an enantioselective Baeyer-Villiger monooxygenase

Article abstract of DOI:10.1073/pnas.0911656107

The molecular basis of allosteric effects, known to be caused by an effector docking to an enzyme at a site distal from the binding pocket, has been studied recently by applying directed evolution. Here, we utilize laboratory evolution in a different way, namely to induce allostery by introducing appropriate distal mutations that cause domain movements with concomitant reshaping of the binding pocket in the absence of an effector. To test this concept, the thermostable Baeyer-Villiger monooxygenase, phenylacetone monooxygenase (PAMO), was chosen as the enzyme to be employed in asymmetric Baeyer-Villiger reactions of substrates that are not accepted by the wild type. By using the known X-ray structure of PAMO, a decision was made regarding an appropriate site at which saturation mutagenesis is most likely to generate mutants capable of inducing allostery without any effector compound being present. After screening only 400 transformants, a double mutant was discovered that catalyzes the asymmetric oxidative kinetic resolution of a set of structurally different 2-substituted cyclohexanone derivatives as well as the desymmetrization of three different 4-substituted cyclohexanones, all with high enantioselectivity. Molecular dynamics (MD) simulations and covariance maps unveiled the origin of increased substrate scope as being due to allostery. Large domain movements occur that expose and reshape the binding pocket. This type of focused library production, aimed at inducing significant allosteric effects, is a viable alternative to traditional approaches to designed directed evolution that address the binding site directly.

Full text of DOI:10.1073/pnas.0911656107

Induced allostery in the directed evolution of an
enantioselective Baeyer–Villiger monooxygenase
Sheng Wu, Juan Pablo Acevedo, and Manfred T. Reetz1
Max-Planck-Institut für Kohlenforschung, Kaiser-Wilhelm-Platz 1, 45470 Mülheim an der Ruhr, Germany
Edited by Marc Ostermeier, Johns Hopkins University, Baltimore, MD, and accepted by the Editorial Board January 5, 2010 (received for review October 9, 2009)
The molecular basis of allosteric effects, known to be caused by an
effector docking to an enzyme at a site distal from the binding
pocket, has been studied recently by applying directed evolution.
Here, we utilize laboratory evolution in a different way, namely to
induce allostery by introducing appropriate distal mutations that
cause domain movements with concomitant reshaping of the
binding pocket in the absence of an effector. To test this concept,
the thermostable Baeyer–Villiger monooxygenase, phenylacetone
monooxygenase (PAMO), was chosen as the enzyme to be em-
ployed in asymmetric Baeyer–Villiger reactions of substrates that
are not accepted by the wild type. By using the known X-ray struc-
ture of PAMO, a decision was made regarding an appropriate site
at which saturation mutagenesis is most likely to generate mutants
capable of inducing allostery without any effector compound
being present. After screening only 400 transformants, a double
mutant was discovered that catalyzes the asymmetric oxidative
kinetic resolution of a set of structurally different 2-substituted cy-
clohexanone derivatives as well as the desymmetrization of three
different 4-substituted cyclohexanones, all with high enantioselec-
tivity. Molecular dynamics (MD) simulations and covariance maps
unveiled the origin of increased substrate scope as being due to
allostery. Large domain movements occur that expose and reshape
the binding pocket. This type of focused library production, aimed
at inducing significant allosteric effects, is a viable alternative to
traditional approaches to “designed” directed evolution that
address the binding site directly.
Remote effects in the directed evolution of enantioselective en-
zymes have been noted earlier but these do not involve allostery
(7–11, 16). Here, we describe the experimental implementation
of induced allostery using a Baeyer–Villiger monooxygenase
(BVMO) as the model enzyme (17–21).
Phenylacetone monooxygenase (PAMO), discovered by
Fraaije and Janssen by using genome mining, is the only thermo-
stable BVMO (22) and the only monooxygenase of its kind to be
analyzed by X-ray crystallography as reported by Maleti, Mattevi
and coworkers (23). Due to its robustness, it would appear to
be an ideal candidate for synthetic organic applications, C-C-
activating asymmetric transformations in oxidative kinetic resolu-
tion of racemic ketones, and/or desymmetrization of prochiral
substrates being potentially possible. Unfortunately, the substrate
scope of PAMO is limited to phenylacetone and related linear
phenyl-substituted derivatives (22, 24). To address this problem,
we have previously applied rational design using site-specific mu-
tagenesis (25) as well as directed evolution based on saturation
mutagenesis (26), in both cases the targeted sites being a loop
next to the binding pocket. PAMO is a monomeric 62-kDa
flavin-dependent enzyme that, like other BVMOs, requires
the cofactor NADPH for FAD reduction (22, 23). Reduced
FAD reacts with O2 forming an alkyl-hydroperoxide that, in
the deprotonated form, adds nucleophilically to the carbonyl
function of the substrate with formation of the short-lived
Criegee intermediate in the rate determining step. Subsequent
fragmentation/elimination generates an ester (or lactone) and
water, like other BVMOs (17–21). In the case of PAMO, it
was suggested on the basis of the X-ray structure that Arg337
participates in the stabilization of the Criegee intermediate
(23). This provides a structural basis for a deeper understanding
of the enzyme’s mechanism. The crystal structure was obtained
without the cofactor NADPþ∕NADPH and in the absence of
a substrate/inhibitor. Further information was recently provided
by Lau, Berghuis and coworkers, who reported the X-ray struc-
ture of another BVMO, the cyclohexanone monooxygenase
(CHMO) from a Rhodococcus strain, with bound FAD and
NADPþ in the absence of a substrate/inhibitor (27). Importantly,
two different structures were identified, a closed and an open
form both, presumably, participating at various points of the
catalytic cycle. These structural data illustrate the mechanistic
and structural complexity of BVMOs and, consequently, the chal-
lenge in performing knowledge-driven directed evolution.
allosteric effects ∣ enzymes ∣ molecular dynamics simulations ∣ protein
engineering
rotein allostery has been recognized as a positive or negative
P
cooperative event, leading to a structural change at the bind-
ing site as a result of distal docking of a molecule acting as an
effector (1–5). Allosteric effects can be influenced by such factors
as variation in pH, temperature, ionic strength, and covalent
modification as well as mutational changes. A variety of experi-
mental and computational techniques have been utilized to un-
ravel the intricacies of this phenomenon (1–5), but also to achieve
useful applications such as the creation of protein switches (6).
Among the techniques that have been applied to address these
challenges is directed evolution, a method that is generally used
to engineer the catalytic profiles of enzymes or the binding prop-
erties of proteins (7–11). In the endeavor to make directed evo-
lution of enantioselective enzymes more efficient than in the past,
we recently introduced the concept of iterative saturation muta-
genesis according to which sites around the binding pocket are
randomized iteratively, a given site being composed of one or
more amino acid positions in the protein (12–14.) When applying
this technique to enzymes displaying strong allosteric effects or
coupled motions along the reaction coordinate (15), it is neces-
sary to consider such phenomena to make reasonable decisions
regarding the appropriate randomization sites, that is, to identify
those residues that do align the binding pocket when the actual
catalytic transformation occurs. A very different approach would
be the idea of “inducing” allosteric communication at a remote
site that results in altered structure and dynamics at the active
site, thereby influencing enzyme activity and enantioselectivity.
Results and Discussion
Experimental Platform. It is not a trivial task to identify remote sites
in an enzyme at which saturation mutagenesis can be expected
to generate mutations that would trigger allosterically-induced
Author contributions: S.W., J.P.A., and M.T.R. designed research; S.W. and J.P.A. performed
research; S.W., J.P.A., and M.T.R. analyzed data; and S.W., J.P.A. and M.T.R. wrote the paper.
The authors declare no conflict of interest.
This article is
Editorial Board.
a PNAS Direct Submission. M.O. is a guest editor invited by the
1To whom correspondence should be addressed. E-mail: reetz@mpi-muelheim.mpg.de.
This article contains supporting information online at www.pnas.org/cgi/content/full/
0911656107/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.0911656107
PNAS ∣ February 16, 2010 ∣ vol. 107 ∣ no. 7 ∣ 2775–2780

domain movements with the creation of a structurally new
binding pocket in the absence of an effector. Suggestions about
movements of the NADP-binding domain believed to be coupled
to the catalytic mechanism of PAMO (23) led us to consider mu-
tations strategically located in regions distal from the active site
that might induce such structural reorganizations (28–30). Close
inspection of the PAMOX-ray structure shows that the number of
likely possibilities is in fact limited. One option would be to focus
on hinge regions but, in the case of PAMO and other BVMOs,
these are characterized by highly conserved sequences. There-
fore, we devised a different strategy by searching for potential
randomization sites at or near two domain interfaces. For exam-
ple, residues at the interface between the NADP-binding domain
and the so-called helical domain of PAMO could be chosen in
the expectation that appropriate mutations would lead to strong
attractive interactions with concomitant structural changes. A
different possibility appeared more promising, namely, to identify
hot spots in PAMO that would bring the NADP-binding domain
spatially closer to the FAD-binding domain (Fig. 1).
We spotted in the PAMO X-ray structure several already
strongly attractive contacts between the two domains, structurally
not far from the hinge region. They occur between the loop
segment Trp177-Glu180 (NADP-binding domain) and the loop
segment Tyr56–Tyr60 (FAD-binding domain), the π-cation inter-
action between His179 and Arg59 being an example (Figs. 1 and
4). Located directly behind, and in loose contact with the loop
segment Tyr56-Tyr60, is the N-terminal region (Ala91-Glu95)
of an α-helix that, by nature, is a rigid secondary element. We
speculated that appropriate mutations in this part of the helix
could lead to strong attractive interactions with loop segment
Tyr56-Tyr60 that, in turn, would result in the movement of loop
segment Trp177-Glu180. Therefore, Gln93/Pro94 was chosen
as the randomization site, a decision that was also guided by
the realization that position 92 is the last member of a long loop
extending parallel to loop segment Asp66-Tyr64 that is in direct
contact with the flavin ring. Exploratory docking experiments in
which phenylacetone was placed in the binding pocket of
WT PAMO, indicated that the substrate is about 18 Å away from
position 93 (Fig. S1).
In the present study, we opted for positions 93∕94 as a
“combined site” for simultaneous saturation mutagenesis (31)
using NDT codon degeneracy (14) encoding 12 amino acids
(Phe, Leu, Ile, Val, Tyr, His, Asn, Asp, Cys, Arg, Ser, and
Gly). About 400 transformants were screened for activity and
enantioselectivity using gas chromatography (GC), the model
reaction being the oxidative kinetic resolution of 2-ethylcyclo-
hexanone (rac-1d) (Scheme 1) in conjunction with the known
thermostable secondary alcohol dehydrogenase from Thermoa-
naerobacter ethanolicus as the NADPH regeneration enzyme
and isopropanol as the reductant (32). The aim was to perform
laboratory evolution using a single substrate, subsequently, to
isolate the enzyme variants, and to test the hit(s) as catalysts
in the analogous transformation of a set of other compounds
without performing any additional mutagenesis/screening experi-
ments. The model compound 1d as well as most of the other sub-
strates (1b, 1c, and 1e–k) are not accepted by the WT PAMO
(22–24), meaning that under the standard reaction conditions
(Materials and Methods), compounds rac-1b–k do not react to
provide lactones to any detectable extent, not even when increas-
ing the reaction time to 48 h or longer. 2-Phenylcyclohexanone
(1a) is the only compound in the list that shows at least some
activity (7 U∕mg) using WT PAMO, but conversion after a pro-
longed reaction time is less than 10% under standard conditions
and enantioselectivity as measured by the selectivity factor is
poor (E ¼ 1.5).
Identification and Characterization of Hits. The library generated by
saturation mutagenesis at positions 93∕94 by using NDT codon
degeneracy was found to contain two mutants that catalyze mea-
surable conversion of ketone rac-1d within a reaction time of 24 h
as specified in the GC screening process, namely Gln93Val/
Pro94Phe and Gln93Asn/Pro94Asp. Due to amino acid bias
inherent in the nature of the saturation mutagenesis protocol
(14, 31), the possibility that hits were missed cannot be excluded
inspite of the high percentage of coverage. Because the double
mutant Gln93Asn/Pro94Asp proved to be the much more active
of the two identified hits, we subsequently focused all further
efforts on this variant. It leads to a selectivity factor of E ¼ 50
in the model reaction, favoring (S)-2d. Remarkably, all other sub-
strates are likewise accepted by this mutant, and with the excep-
tion of ketone 1b, good to excellent enantioselectivity was
uniformly achieved (Table 1). Although substrate acceptance
has been made possible by our method, activities are modest.
Combining the present approach with such techniques as
DNA shuffling (7–11) or iterative saturation mutagenesis at sites
aligning the reshaped binding pocket (11, 12) may allow for
further improvements.
We also tested the catalytic potential of the best PAMO variant
Gln93Asn/Pro94Asp as a catalyst in the desymmetrization of
4-substituted cyclohexanone derivatives 3a–c. Scheme 2 shows
Fig. 1. Cartoon representations of the different domains in the PAMO struc-
ture. FAD domain is shown in Green, Dark Blue corresponds to the NADP-
binding domain, and Light Blue to the helical domain. The NADP- and
FAD-binding domains are linked by two antiparallel β-strands in a hinge-like
structure configuration (Black). NADP-binding and helical domains can be
considered as separate entities connected by two segments (Red). The ran-
domization site Gln93/Pro94 in the N-terminal region of the α-helix is also
labeled.
Scheme 1. Oxidative kinetic resolution of rac-1 catalyzed by PAMO
mutants.
2776
∣
www.pnas.org/cgi/doi/10.1073/pnas.0911656107
Wu et al.

Table 1. Oxidative kinetic resolution of ketones rac-1a–k using PAMO variant Gln93Asn/Pro94Asp under standard
reaction conditions (Materials and Methods). Details of the procedure for expression, purification, and enzymatic
activity assays are included in SI Material and Methods. Note that the R∕S assignment may change in accordance with
the Cahn-Ingold-Prelog convention; the absolute configuration is as shown in Scheme 1.
Entry
Substrate
Specificactivity (U∕mg)
Conv. (%)
Residual ketone 1
Lactone 2
E-value
ee (%)
38
2.7
Abs. conf.
ee (%)
Abs. conf.
1
2
3
4
5
6
7
8
rac-1a
rac-1b
rac-1c
rac-1d
rac-1e
rac-1f
rac-1g
rac-1h
rac-1i
rac-1j
rac-1k
64
40
35
49
42
51
24
39
48
58
52
45
17
15
21
37
36
23
15
42
43
11
S
S
R
R
R
R
S
S
S
S
S
95
18
91
95
95
92
94
>99
86
R
R
S
S
S
92
1.5
25
50
68
40
42
>200
25
46
25
37
14
19
17
43
58
17
S
R
R
R
R
R
9
10
11
98
96
>200
55
that, in all cases, excellent enantioselectivity resulted. Interest-
ingly, the absolute configuration of the lactone products are
uniformly opposite to what is observed when using CHMO as
the catalyst (17–21). The latter has been reported to deliver
an ee-value of only 52% in the desymmetrization of substrate
4c, whereas the evolved PAMO mutant Gln93Asn/Pro94Asp is
highly enantioselective in the opposite absolute sense (Scheme 2).
which case very low activity and stereorandom behavior (E ≈ 1)
was registered, similar to WT PAMO. This proves that a strong
cooperative effect is operating in the double mutant. The results
also indicate, but do rigorously prove, that our initial choice
regarding simultaneous randomization at two positions was the
better of the two possible strategies for targeting positions 93
and 94. Finally, we tested the activity of the two single mutants
in the reaction of phenylacetone and discovered that they are
similar to WT PAMO: Mutant Gln93Asn (Km ¼ 0.143 mM,
kcat ¼ 3.1 s−1, and kcat∕Km ¼ 21; 000 M−1 s−1), mutant Pro94Asp
(Km ¼ 0.48 mM, kcat ¼ 1.3 s−1, and kcat∕Km ¼ 27; 000 M−1 s−1).
Another point of interest concerns the question of whether the
mutational change in going from WT PAMO to the double
mutant Gln93Asn/Pro94Asp shifts the substrate acceptance from
phenylacetone to the above compounds, or whether substrate
scope has been widened to include all of the compounds. The
latter pertains as shown by kinetic experiments using phenylace-
Single Site Saturation Mutagenesis Libraries. Although the deconvo-
lution experiments show that the two single point mutants
Gln93Asn and Pro94Asp are inactive, it was not absolutely clear
whether other point mutations at these two positions could lead
to active single mutants. Therefore, two separate single residue
saturation mutagenesis libraries were generated by randomiza-
tion at positions 93 and 94, respectively, using, in this case,
NNK codon degeneracy. In each case 200 transformants were
screened, but no active variants were found. It is, thus, clear once
more that simultaneous randomization at a site comprising more
than one amino acid position opens the door for potentially
achieving pronounced cooperativity between the respective resi-
dues, that is, more than additivity. We do not suggest, however,
that saturation mutagenesis at sites comprising single amino acid
positions is always a futile exercise.
tone as the substrate: WT PAMO (Km ¼ 0.071 mM, kcat
¼
2.3 s−1
,
and kcat∕Km ¼ 32; 000 M−1 s−1), mutant Gln93Asn/
Pro94Asp (Km ¼ 0.054 mM, kcat ¼ 1.5 s−1
27; 000 M−1 s−1).
,
and kcat∕Km
¼
Thermostability Experiments. To check whether substrate accep-
tance and enantioselectivity were evolved at the expense of
the robustness of PAMO, T6500 values were measured, the tempera-
ture at which 50% of the activity has been lost following a heat
treatment for 60 min (33, 34). The results show that thermostabil-
60
ity is essentially maintained: WT PAMO (T ¼ 58.2 °C); variant
50
60
50
Gln93Asn/Pro94Asp (T ¼ 56.2 °C).
Deconvolution Experiments. The immediate question arose as to
whether the double mutant is really necessary for achieving
activity and enantioselectivity, or whether the corresponding
single variants, characterized by point mutations, would likewise
function well. For this purpose, PAMO single mutants Gln93Asn
and Pro94Asp were generated by site-specific mutagenesis and
tested for activity as catalysts in the model reaction involving
rac-1d. Neither of the single mutants showed any activity whatso-
ever. The same was found for all other substrates, except for 1a in
MD Simulations. To gain insight into the source of enhanced activ-
ity and thereby to check our original postulate regarding possible
allosteric communication, we turned to MD simulations of
WT PAMO and of the superior double mutant Gln93Asn/
Pro94Asp. Being guided by the X-ray structure of PAMO (23),
the Schrödinger software package was applied to the two proteins
harboring FAD, initially in the absence of NADPþ and, sub-
sequently, with bound cofactor. The actual MD simulations were
performed with the Desmond program by using an algorithm for
high-speed parallel execution (35). Two different strategies for
applying MD simulations were tested to avoid any possible
misinterpretations due to the uncertainty regarding the catalytic
mechanism of PAMO, that is, the question whether the presence
of bound NADPþ influences the shape of the binding pocket
during the different steps in the catalytic cycle (23). The first
MD simulations were performed with the protein structures
having bound FAD, but excluding the coenzyme NADPþ. The
productive MD for WT PAMO, the evolved mutant Gln93Asn/
Pro94Asp, and the point mutant Pro94Asp were run during
Scheme 2. Desymmetrization of ketones 3a–c using PAMO variant Gln93/
Pro94Asp.
Wu et al.
PNAS
∣
February 16, 2010
∣
vol. 107
∣
no. 7
∣
2777

2 ns. Subsequently, average structures of the WT and mutants
were calculated from the conformers obtained in the second
ns of simulation following equilibration (Fig. 2).
WT PAMO. Interestingly, the conformational differences
between WT PAMO and the evolved double mutant
Gln93Asn/Pro94Asp are similar to the observed structural differ-
ences between the two reported crystalline forms of the CHMO.
The two conformers correspond to a closed and a more open
conformation, interconversion resulting in domain shifts with
concomitant sliding of the cofactor NADPþ. In the open form
of CHMO, the increase in surface accessibility of the binding
pocket relative to the situation in the closed form was associated
with the movement of the NADP-binding domain, much like the
mutation-induced allosteric effect in our system (27). Backbone
movements around the active site in addition to repositioning of
NADPþ were also described in the two forms of the CHMO (27),
very similar to the changes that we observe when going from WT
PAMO to the double mutant. For example, the two crystal struc-
tures reveal the reorientation of two residues (Arg132/Lys328)
much like that of Arg337 and Lys 336 in our system, as well
as similar modifications of the H-bonding around NADPþ. More-
over, reasonable hypotheses regarding the nature of the domain
movement required for catalysis in the Baeyer–Villiger mono-
oxygenase family were suggested (27) that have direct bearing
on our results. The allosterically induced domain movements
in our system, similar in nature, were brought about by strate-
gically located mutations. This resulted in a highly altered cata-
lytic profile with respect to substrate acceptance and enantios-
electivity.
In an attempt to unveil the reason(s) for the pronounced do-
main movements caused by the evolved double mutant, a deeper
analysis of the region nearest to the saturation mutagenesis site
93∕94 was necessary, specifically by analyzing the MD simula-
tions trajectories of WT PAMO, of the double mutant
Gln93Asn/Pro94Asp, and also of the inactive single mutant
variant Pro94Asp. Fig. 4 shows the average backbone structure
and side-chain position of WT PAMO and of the double mutant
Gln93Asn/Pro94Asp. New H-bond interactions such as Asp94/
Arg59 and Trp57/Trp177 are visible. The interaction Asp94/
Arg59 implies a strong salt bridge, anchoring the rigid helix struc-
ture to the loops Tyr56-Tyr60 and Trp177-Glu180. In sharp con-
trast, the strong Asp94/Arg59 salt bridge is not found in the single
mutant Pro94Asp, due to the steric clash between Arg59 and
Glu93 that would arise if this interaction were to occur. The
situation is reversed in the double mutant in that Gln is mutated
to Asn at position 93.
It is logical to expect that substrate accessibility of the active
site is determined primarily by the residues in close proximity to
the flavin on its re-side. The flavin ring binds at the bottom of the
cleft between the FAD- and NADP-binding domains. This active
pocket in the WT PAMO appears as an enclosed cavity lacking
good solvent accessibility and capability to bind large substrates.
In sharp contrast, the MD simulations show that the space above
the re-side of the bound flavin ring in the double mutant is much
more solvent accessible, offering sufficient room for substrates to
bind. (Fig. 2). In agreement with the observed lack of activity of
the single mutant Pro94Asp and WT PAMO, the respective aver-
age structures in both simulations do not show significant confor-
mational differences. Inspection of the active site of WT PAMO,
of the single mutant Pro94Asp, and of the active double mutant
Gln93Asn/Pro94Asp reveals distinct conformational changes
resulting in different spatial relationships of residues near
FAD and, thus, a change in the shape of the binding pocket.
Fig. S2 and Table S1 pinpoint the major geometric differences.
Second round MD calculations were performed with WT
PAMO and the double mutant Gln93Asn/Pro94Asp, each with
bound FAD and NADPþ during 6 ns of productive simulation.
The same basic tendency regarding conformational differences
between WT PAMO and mutant Gln93Asn/Pro94Asp was
observed (Figs. S3 and S4 and Table S2).
Superposition of the average structures of WT PAMO and the
double mutant using FAD and the FAD-binding residues as an-
chors points to a structurally crucial backbone rotation of the
NADP-binding domain in a movement that is reminiscent of a
scissors-opening, reaching displacements between 3–5 Å of
the backbone of distal residues, and conformational changes
of the same order of magnitude occurring at some residues align-
ing the active site (Fig. 3). Table S3 summarizes the distance
values of displacement between equivalent residues forming
the active site in the superposed MD simulations of WT PAMO
and mutant Gln93Asn/Pro94Asp. Surprisingly, the NADPþ co-
factor adopts a somewhat different position in its binding domain
(Fig. 3B). Some NADPþ binding residues in WT PAMO no long-
er participate in the double mutant, Ser196, Glu200, and Arg337
being prime examples, whereas previous non-involved residues
form new side-chain and backbone interactions with the cofactor
as, for example, in the case of Phe387. In summary, the MD
simulation results show that the combined action of the two point
mutations not only orchestrates distinct structural changes glob-
ally, but also locally at the binding pocket.
A few more explanatory comments are in order. Our MD
simulations of WT PAMO point to a strong interaction between
Arg59 and His179 that is also revealed by its crystal structure. It
involves a cation-π interaction with histidine serving as the aro-
matic partner. Such interactions have been found to be stronger
than salt bridges in other systems (36, 37). In the double mutant,
this interaction is maintained but now Arg59 also interacts with
mutated Asp94 via a salt bridge. This strengthens the connection
between the rigid α-helix, where Asp94 is located, and the NADP-
binding domain, an effect that causes exposure of the binding
pocket together with repositioning of the backbone and side-
chain structural components of residues surrounding it. Such
reorganization has a direct bearing on activity and enantios-
electivity.
It is informative to view our findings in the light of the two
recently reported crystal structures of the enzyme cyclohexanone
monooxygenase (CHMO) from an environmental Rhodococcus
strain with bound FAD and NADPþ, fortuitously obtained under
different crystallization conditions (27). This BVMO has a
sequence identity of 43% and high structure similarity with
Finally, to check our conclusions derived from the MD simula-
tions, we considered appropriate covariance maps that are known
to constitute useful indicators of correlated and anti-correlated
motions in enzymes (38–40). Cross-correlation coefficients of
the Cα atoms were first constructed by using a 6 ns simulation
of WT PAMO, including FAD and NADPþ. Analysis con-
firms that strong correlating motions exist between the regions
Trp177-Glu180 (NADP-binding domain), Tyr56-Tyr60 (FAD-
binding domain), Ala91-Glu95, and the residues forming the
active site, particularly Asp66, Leu153, Ser196, Arg337, and
Phe389 (Fig. S5). The analogous procedure was then applied
Fig. 2. Binding pocket surfaces of WT PAMO (Green), mutant Pro94Asp
(Orange), and mutant Gln93Asn/Pro94Asp (Blue). The binding pocket sur-
faces were obtained from the average structure of 1 ns of MD trajectory after
1 ns of equilibration. Productive MD simulations were run during 2 ns includ-
ing coenzyme FAD.
2778
∣
www.pnas.org/cgi/doi/10.1073/pnas.0911656107
Wu et al.

Fig. 3. Cartoon representation of the superimposed WT PAMO (Green) and mutant Gln93Asn/Pro94Asp (Blue) structures. (A) Average structures were
obtained from the conformers of 1 ns of MD trajectory after 5 ns of equilibration. Productive MD simulations were run during 6 ns including coenzyme
FAD and NADPþ. Structural superposition was performed using FAD molecule and FAD-binding residues (residues within 3 Å) as anchor. It is possible to observe
important backbone rotation especially in the NADP domain. (B) Details of the NADPþ binding, comparing WT PAMO (Green) and mutant Gln93Asn/Pro94Asp
(Blue). Arg337 in the mutant shifts away from the flavin ring and does not show the H-bond present in the WT.
to the superior double mutant Gln93Asn/Pro94Asp. The respec-
tive covariance map shows increased correlation between the mu-
tation site and the FAD segment in the interface FAD-NADP
domains that supports our previous conclusion regarding
increased contact as a result of mutagenesis (Fig. S6). Upon
comparing the two covariance maps, important differences are
observed, particularly around the active site residues, hinges,
and the NADP-binding domain (Fig. S7), likewise, in line with
our mechanistic conclusions.
duce domain movements via allostery that lead to a different
shape of the binding pocket, very much like traditional allosteric
communication brought about by effectors docking at remote
sites. Random mutagenesis was focused on a two-residue site dis-
tal from the binding pocket, the choice being based on the anal-
ysis of the PAMO X-ray structure. This structure-based approach
led to the identification of an active double mutant that accepts a
set of otherwise inert 2-substituted cyclohexanone derivatives
with high enantioselectivity in the respective oxidative kinetic
resolutions. MD simulations and covariance maps proved to be
useful both in interpreting the data and in choosing appropriate
mutations sites that can be expected to induce allosteric
effects (39).
Our work bears some relationship to certain diseases caused by
mutations that induce allosteric effects (2, 39). These can shut
down the functionality of allosteric enzymes, or they can cause
conformational changes at the active site that lead to constitutive
activation regardless of whether an effector is bound, as in the
case of the G-protein (2). We believe that knowledge-guided
directed evolution, with the aim to induce allostery of the kind
described herein, may be useful in future directed evolution
studies of enzymes, especially when induced domain movements
Conclusions
Baeyer–Villiger monooxygenases are not considered to be allo-
sterically controlled enzymes (5, 17–21), yet their structures con-
sist of semirigid domains connected via flexible regions or hinges
that are typical features present in allosteric systems. It has been
stated that the involvement of both flexibility and semirigidity in
allosteric enzymes is likely to be a result of evolution, ensuring a
balance between sensitivity or specificity and robustness relating
to functional importance (1). In the case of BVMOs, the move-
ment of domains harboring cofactors is essential for catalysis
(23, 27). In the present study, we have successfully applied
knowledge-based directed evolution to a thermostable Baeyer–
Villiger monooxygenase, PAMO, specifically in the quest to in-
Fig. 4. Comparison of the backbone and side-chain shift between WT PAMO (Green) and mutant Gln93Asn/Pro94Asp (Blue). New H-bonds and a salt bridge
between Arg59 and Asp94 in the mutant variant are illustrated in this figure. It is also possible to observe a shorter distance between the loop Tyr56-Tyr60 and
the NADP-binding domain loop Trp177- Glu180.
Wu et al.
PNAS
∣
February 16, 2010
∣
vol. 107
∣
no. 7
∣
2779

of the indentified hits was sequenced to confirm there were not any other
mutations.
can be predicted with a sufficient degree of certainty. This also
applies to laboratory evolution of binding properties, as in the
case of therapeutic proteins.
Procedure for the Enzymatic Oxidation of Substrates. In a typical protocol, 1 mL
reaction system consists of 2.5 mM substrate (added as a concentrated
solution in acetonitrile), 5 mM isopropanol, 100 μM NADPþ, 1.0 μM purified
PAMO mutant enzyme, and 20ADH (2 U) and 50 mM Tris-HCl buffer (pH 8.0).
The mixture in 8 ml glass tube with a sealed cap (to avoid evaporation of
2-alkyl substituted substrates) was shaken at 200 rpm and 30 °C for times
established (between 5–16 h) to control the conversion rate less than 50%
for kinetic resolution. For desymmetrization of rac-3a–c, the reaction time
is 24 h. The reaction was stopped, worked up by extraction with ethyl acetate
(containing 200 mg L−1 internal standard as nonane, dodecane, hexadecane,
and octadecane). The sample was analyzed by achiral and chiral GC to deter-
mine the conversion of and the enantiomeric excess of the residual ketones
and produced lactones. Control experiments were performed for all the
tested substrates with purified WT PAMO enzymes (about 5 μM). Further
details can be found in SI Text.
Materials and Methods
Library Screening. Individual colonies were placed into 2.2-mL 96-deep-well
plates containing 800 μL of LB media with 100 μg mL−1 carbenicilline by a
colony picker QPIX (Genetix). After cell growth at 37 °C overnight with
shaking at 800 rpm, 10 μL of each preculture was transferred into a new plate
containing 800 μL of TB media supplemented with 0.1% L-arabinose as
inducer and 100 μg mL−1 carbenicilline. The duplicate plates were grown
for an additional 24 h to induce PAMO expression. The cultures were
centrifuged at 3200 × g and 4 °C for 6 min and the supernatants were dis-
carded. The original plates were stored. Each cell pellet was resuspended
in 600 μL of 50 mM Tris-HCl (pH 8.0) containing 1 mg mL−1 lysozyme and four
units of DNase I. Lysis were performed at 37 °C and shaken at 800 rpm for 3 h.
Cell debris was precipitated by centrifugation at 3200 × g and 4 °C for 30 min
and 50 μL of each cleared supernatant transferred to a 1.1 mL 96-deep-well
plate. Then in each well, 50 μL of the secondary alcohol dehydrogenase
(20ADH) crude extract (about 10 U) (37), 10 μL of 1 mM NADPþ, and 20 μL
of 100 mM rac-1d in acetonitrile and 370 μL of 50 mM Tris-HCl (pH 8.0) con-
taining 5 mM isopropanol were added. The reaction plates were incubated at
37 °C and shaken at 800 rpm for 24 h. Four Hundred μL of ethyl acetate was
then added to extract the substrate and product from the solution. After
centrifugation, 200 μL of organic layer in each well was transferred into a
new glass-made 96-deep-well plate, and subjected to GC analysis for screen-
ing. Active clones were collected and the results reproduced. The entire gene
MD Simulations. Using the published X-ray structural data of 1W4X, the in
silico structure determinations of the wild type and mutants were performed
with the Schrödinger software package. MD simulations were performed
with the Desmond program by using an algorithm for high-speed parallel
execution (35). Details can be found in SI Materials and Methods.
ACKNOWLEDGMENTS. We thank the Fonds der Chemischen Industrie and the
Deutsche Forschungsgemeinschaft (Schwerpunkt 1170) for generous
support.
1. Cui Q, Karplus M (2008) Allostery and cooperativity revisited. Protein Sci 17:1295–1307.
22. Fraaije MW, et al. (2005) Discovery of a thermostable Baeyer–Villiger monooxygenase
by genome mining. Appl Microbiol Biotechnol 66:393–400.
2. Tsai C-J, del Sol A, Nussinov
R (2009) Protein allostery, signal transmission and
23. Malito E, Alfieri A, Fraaije MW, Mattevi A (2004) Crystal structure of a Baeyer–Villiger
monooxygenase. P Natl Acad Sci USA 101:13157–13162.
dynamics: A classification scheme of allosteric mechanisms. Mol Biosyst 5:207–216.
3. Goodey NM, Benkovic SJ (2008) Allosteric regulation and catalysis emerge via
common route. Nat Chem Biol 4:474–482.
4. Ozkan E, Yu H, Deisenhofer J (2005) Mechanistic insight into the allosteric activation of
a ubiquitin-conjugating enzyme by RING-type ubiquitin ligases. P Natl Acad Sci USA
102:18890–18895.
5. Clarkson MW, Gilmore SA, Edgell MH, Lee AL (2006) Dynamic coupling and allosteric
behavior in a nonallosteric protein. Biochemistry 45:7693–7699.
6. Guntas G, Mansell TJ, Kim JR, Ostermeier M (2005) Directed evolution of protein
switches and their application to the creation of ligand-binding proteins. P Natl Acad
Sci USA 102:11224–11229.
7. Lutz S, Bornscheuer UT (2009) Protein Engineering Handbook (Wiley-VCH, Weinheim).
8. Turner NJ (2009) Directed evolution drives the next generation of biocatalysts. Nat
Chem Biol 5:567–573.
9. Jäckel C, Kast P, Hilvert D (2008) Protein design by directed evolution. Annu Rev
Biophys Biomol Struct 37:153–173.
a
24. Zambianchi F, et al. (2007) Titration and assignment of residues that regulate the
enantioselectivity of phenylacetone monooxygenase. Adv Synth Catal 349:1327–1331.
25. Bocola M, et al. (2005) Converting phenylacetone monooxygenase into phenylcyclo-
hexanone monooxygenase by rational design: Towards practical Baeyer–Villiger
monooxygenases. Adv Synth Catal 347:979–986.
26. Reetz MT, Wu S (2008) Greatly reduced amino acid alphabets in directed evolution:
Making the right choice for saturation mutagenesis at homologous enzyme positions.
Chem Commun pp 5499–5501.
27. Mirza IA, et al. (2009) Crystal structures of cyclohexanone monooxygenase reveal com-
plex domain movements and a sliding cofactor. J Am Chem Soc 131:8848–8854.
28. Qu C, Akanuma S, Moriyama H, Tanaka N, Oshima T (1997) A mutation at the interface
between domains causes rearrangement of domains in 3-isopropylmalate dehydro-
genase. Protein Eng 10:45–52.
29. Oue S, Okamoto A, Yano T, Kagamiyama H (1999) Redesigning the substrate specificity
of an enzyme by cumulative effects of the mutations of non-active site residues. J Biol
Chem 274:2344–2349.
30. Atkin KE, et al. (2008) The structure of monoamine oxidase from Aspergillus niger
provides a molecular context for improvements in activity obtained by directed evolu-
tion. J Mol Biol 384:1218–1231.
31. Hogrefe HH, Cline J, Youngblood GL, Allen RM (2002) Creating randomized amino acid
libraries with the QuikChange® multi site-directed mutagenesis kit. BioTechniques
33:1158–1165.
32. Zheng C, Pham VT, Philips RS (1992) Asymmetric reduction of ketoesters with alcohol
dehydrogenase from Thermoanaerobacter ethanolicus. Bioorg Med Chem Lett
2:619–622.
33. Eijsink VGH, Gåseidnes S, Borchert TV, van den Burg B (2005) Directed evolution of
enzyme stability. Biomol Eng 22:21–30.
10. Bershtein S, Tawfik DS (2008) Advances in laboratory evolution of enzymes. Curr Opin
Chem Biol 12:151–158.
11. Reetz MT (2008) Directed evolution as a means to engineer enantioselective enzymes.
Asymmetric Organic Synthesis with Enzymes, ed Gotor V (Wiley-VCH, Weinheim).
12. Reetz MT, Wang L-W, Bocola M (2006) Directed evolution of enantioselective enzymes:
Iterative cycles of CASTing for probing protein-sequence space. Angew Chem
118:1258–1263 Erratum 2556; Angew Chem Int Ed 45:1236–1241; Erratum 2494.
13. Reetz MT, Carballeira JD, Vogel A (2006) Iterative saturation mutagenesis on the basis
of B-factors as
a strategy for increasing protein thermostability. Angew Chem
118:7909–7915 Angew Chem Int Ed 45:7745–7751.
14. Reetz MT, Carballeira JD (2007) Iterative saturation mutagenesis (ISM) for rapid
directed evolution of functional enzymes. Nat Protoc 2:891–903.
15. Wong KF, Selzer T, Benkovic SJ, Hammes-Schiffer S (2005) Impact of distal mutations on
the network of coupled motions correlated to hydride transfer in dihydrofolate reduc-
tase. P Natl Acad Sci USA 102:6807–6812.
16. Reetz MT, et al. (2007) Learning from directed evolution: Further lessons from theo-
retical investigations into cooperative mutations in lipase enantioselectivity.
ChemBioChem 8:106–112.
17. Kayser MM (2009) ‘Designer reagents’ recombinant microorganisms: New and power-
ful tools for organic synthesis. Tetrahedron 65:947–974.
34. Ó’Fágáin C (2003) Enzyme stabilization—recent experimental progress. Enzyme Mi-
crob Tech 33:137–149.
35. Bowers KJ, et al. (2006) Scalable algorithms for molecular dynamics simulations on
commodity clusters. Proceedings of the ACM/IEEE Conference on Supercomputing
(SC06), Tampa, Florida.
36. Reddy AS, Sastry GM, Sastry GN (2007) Cation-aromatic database. Proteins
67:1179–1184.
37. Crowley PB, Golovin A (2005) Cation-pi interactions in protein-protein interfaces.
Proteins 59:231–239.
18. Stewart JD (1998) Cyclohexanone monooxygenase: A useful reagent for asymmetric
Baeyer–Villiger reactions. Curr Org Chem 2:195–216.
19. Mihovilovic MD (2006) Enzyme mediated Baeyer–Villiger oxidations. Curr Org Chem
10:1265–1287.
38. Ichiye T, Karplus M (1994) Collective motions in proteins: A covariance analysis of
atomic fluctuations in molecular dynamics and normal mode simulations. Proteins
11:205–217.
20. Walsh CT, Chen Y-CJ (1988) Enzymic Baeyer–Villiger oxidations by flavin-dependent
monooxygenases. Angew Chem 100:342–352 Angew Chem, Int Ed Engl 27:333–343.
21. Hilker I, Wohlgemuth R, Alphand V, Furstoss R (2005) Microbial transformations 59:
First kilogram scale asymmetric microbial Baeyer–Villiger oxidation with optimized
productivity using a resin-based in situ SFPR strategy. Biotechnol Bioeng 92:702–710.
39. Liu J, Nussinov R (2008) Allosteric effects in the marginally stable von Hippel-Lindau
tumor suppressor protein and allostery-based rescue mutant design. P Natl Acad Sci
USA 105:901–906.
40. Hammes-Schiffer S, Benkovic SJ (2006) Relating protein motion to catalysis. Annu Rev
Biochem 75:519–541.
2780
∣
www.pnas.org/cgi/doi/10.1073/pnas.0911656107
Wu et al.