This is an open access article published under an ACS AuthorChoice License, which permits

copying and redistribution of the article or any adaptations for non-commercial purposes.

Research Article

Cite This: ACS Catal. 2018, 8, 3395−3410 pubs.acs.org/acscatalysis

P450-Catalyzed Regio- and Diastereoselective Steroid Hydroxylation:

Efficient Directed Evolution Enabled by Mutability Landscaping
Carlos G. Acevedo-Rocha,†,‡,¶ Charles G. Gamble,§ Richard Lonsdale,†,‡,∥ Aitao Li,†,‡,⊥ Nathalie Nett,‡
Sabrina Hoebenreich,‡ Julia B. Lingnau,† Cornelia Wirtz,† Christophe Fares,† Heike Hinrichs,†
Alfred Deege,† Adrian J. Mulholland,∥ Yuval Nov,# David Leys,§ Kirsty J. McLean,§
Andrew W. Munro,*,§ and Manfred T. Reetz*,†,‡
†
Max-Planck-Institut für Kohlenforschung, Kaiser-Wilhelm-Platz 1, 45470 Muelheim, Germany
‡
Department of Chemistry, Philipps-University, Hans-Meerwein-Strasse 4, 35032 Marburg, Germany
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

§
Manchester Institute of Biotechnology, School of Chemistry, University of Manchester, Manchester M1 7DN, U.K.
Downloaded via CHINESE UNIV OF HONG KONG on April 22, 2024 at 07:06:40 (UTC).

∥
Centre for Computational Chemistry, School of Chemistry, University of Bristol, Cantock’s Close, Bristol BS8 1TS, U.K.
⊥
Hubei Collaborative Innovation Center for Green Transformation of Bio-resources, Hubei Key Laboratory of Industrial
Biotechnology, College of Life Sciences, Hubei University 368 Youyi Road, Wuchang Wuhan 430062, China
#
Department of Statistics, University of Haifa, Haifa 31905, Israel
*
S Supporting Information

ABSTRACT: Cytochrome P450 monooxygenases play a crucial

role in the biosynthesis of many natural products and in the
human metabolism of numerous pharmaceuticals. This has
inspired synthetic organic and medicinal chemists to exploit
them as catalysts in regio- and stereoselective CH-activating
oxidation of structurally simple and complex organic compounds
such as steroids. However, levels of regio- and stereoselectivity
as well as activity are not routinely high enough for real
applications. Protein engineering using rational design or
directed evolution has helped in many respects, but simulta-
neous engineering of multiple catalytic traits such as activity,
regioselectivity, and stereoselectivity, while overcoming trade-
offs and diminishing returns, remains a challenge. Here we show
that the exploitation of information derived from mutability landscapes and molecular dynamics simulations for rationally
designing iterative saturation mutagenesis constitutes a viable directed evolution strategy. This combined approach is illustrated
by the evolution of P450BM3 mutants which enable nearly perfect regio- and diastereoselective hydroxylation of five different
steroids specifically at the C16-position with unusually high activity, while avoiding activity−selectivity trade-offs as well as
keeping the screening effort relatively low. The C16 alcohols are of practical interest as components of biologically active
glucocorticoids.
KEYWORDS: directed evolution, cytochrome P450 monooxygenase, regioselectivity, stereoselectivity, mutability landscapes,
iterative saturation mutagenesis, steroids, C−H activation

■ INTRODUCTION
Cytochrome P450 (CYP) monooxygenases belong to a large
mutants may not be optimal. The situation in the case of CYP-
catalyzed hydroxylation of steroids or other natural products
protein superfamily (>20 000 members) that is involved in the such as terpenes or alkaloids is even more challenging, because
biosynthesis of complex natural products including steroids, the desired catalytic profile is more complex encompassing
terpenes, alkaloids, flavonoids, and vitamins;1 in the metabolism regioselectivity, diastereoselectivity, and activity.6 Parallel to
of therapeutic drugs;1,2 and in the breakdown of pollutants.2 these efforts, organic chemists have responded to the need for
Since the early 1950s, many different fungal and bacterial late-stage CH-activating hydroxylation of distinct types of
strains containing CYPs have also been used as catalysts in compounds by developing novel reagents and catalysts, but
organic and medicinal chemistry, as first demonstrated in the limitations in terms of regioselectivity persist.7
landmark industrial semisynthetic preparation of cortisol.3
Protein engineering of many different enzyme types based on Received: January 29, 2018
rational design4 or directed evolution5 allows the enhancement Revised: March 5, 2018
and reversal of enantioselectivity, although activity of the best Published: March 8, 2018

© 2018 American Chemical Society 3395 DOI: 10.1021/acscatal.8b00389

ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

When faced with the problem of CYP-based selective rational design based on site-directed mutagenesis, Comman-
hydroxylation of substrates not studied previously, screening deur and co-workers engineered P450BM3 mutants in the
wildtype (WT) or mutants evolved earlier for other purposes hydroxylation of testosterone at positions 16α/β with
may lead fortuitously to a certain regio- and steresoelectivity at selectivities of 60−85%, but with suboptimal substrate
a desired (or undesired) position in the substrate, which, if conversion and 10-fold activity loss.18 Other CYPs have also
necessary, can then be improved by further mutagenesis.6 An been engineered or used to control regio- and stereoselective
impressive recent example concerns the late stage trans- steroid hydroxylation at other positions.6h,19a−d However, they
formation of an allylic CH2−entity into the respective α,ß- generally suffer from the same issue: High selectivity is
unsaturated ketone function in the natural product synthesis of observed only at low substrate conversion, while high substrate
nigelladine A by the groups of Stoltz and Arnold.8 Upon testing conversion compromises selectivity.12 A different approach
CYP mutant libraries previously generated for other sterically without the use of mutagenesis is to use protected or
demanding substrates, a mutant was identified which ensures a functionalized substrates according to the Gringl-concept or
regioselectivity of 3.8:1 in favor of the desired hydroxylation at to employ additives, but this technique has not been applied to
position C7, followed by chemical oxidation to the ketone in steroids and cannot be expected to be general for this class of
the final step.8 In other work, P450BM3, a widely used self- substrates.19d−f Thus, evolving both regio- and stereoselectivity as
sufficient fatty acid CYP from Bacillus megaterium,9 was well as high activity for CYP-catalyzed steroid hydroxylation at
elegantly employed by Fasan and co-workers in the late-stage position C16 (or at other targeted positions) remains a daunting
hydroxylation of the antimalarial drug artemisinin.10a Trans- task.
formations at the chemically easily accessible positions such as In some of our previous “steroid mutant libraries” some
C9 and C10 had been performed earlier using synthetic weakly C16-selective P450BM3 mutants of very low activity were
reagents and catalysts.11 In contrast, it was known that P450- identified.16a Rather than using these mutants as starting
mediated metabolism occurs, inter alia, at the chemically templates for SM and applying the same ISM-based procedure
difficult to access C6 and C7 positions.11 In order to reach this for evolving possible improvements, we decided to explore an
difficult goal by P450BM3 catalysis, a mutant FL#62 with 16 alternative strategy. We hoped that this would be more efficient
point mutations was first tested which had been identified while minimizing the screening effort. The basic idea is the
earlier in a fingerprint process based on saturation mutagenesis utilization of mutability landscapes (MLs)20 as a rational guide
(SM) and color-based screening of 12 500 mutants using bulky for choosing optimal reduced AAAs in ISM experiments, in
surrogate substrates.12 Gratifyingly, this variant showed in the some cases flanked by molecular dynamics (MD) computations
reaction of artemisinin a selectivity of C7(S):C7(R):C6a = for additional assistance. This approach is different from
83:10:7, calling for improvement by directed evolution.10a previous strategies in which (R)- and/or (S)-enantioselectivities
Using the technique of iterative saturation mutagenesis of other enzyme types were evolved by applying SM at
(ISM),13 final selectivities of 100% C7(S), 100% C7(R), and individual residues using, inter alia, NNK codon degeneracy
94% C6a were achieved, although in the latter case with a 3-fold encoding all 20 canonical amino acids.5,13 This enables positive
decrease in total turnover number (TTN) relative to the parent mutations to be identified. The conventional procedure of
enzyme.10a Later, the selective hydroxylation of the anti- combining positive mutations can then be applied, although
leukemic agent parthenolide was also accomplished, although this does not always work. Recently, we have used the
one mutant showed a strong activity trade-off (17-fold loss).10b information from such NNK-scans for further genetic
Trade-off effects have been observed in other directed optimization by employing the respective newly introduced
evolution studies of CYPs,6 as in protein engineering of other amino acids as building blocks in reduced AAAs for SM at
enzyme types.5,14 On the other hand, Wong and co-workers multiresidue sites.21 This technique, in contrast to simply
reported the highly regioselective (96%) hydroxylation of combining positive mutations, focuses on more extensive yet
testosterone at C15 with excellent substrate conversion (83%) rationally chosen protein sequence space and reduces the
and little screening effort using a panel of only ∼100 P450BM3 screening effort drastically. While in several cases successful, it
variants.15 Although C15-selectivity was not planned in terms should be remembered that only positive mutants are
of targeting this particular position, this kind of result is of sequenced, which are then involved in subsequent decision-
significant synthetic value. making. When focusing on enantioselectivity as a typical
Some time ago, we reported the directed evolution of parameter, for example, no information is provided regarding
P450BM3 in the selective hydroxylation of testosterone.16 Since possible mutations that influence other parameters such as
WT does not accept this substrate nor steroids in general, the activity, either in a deleterious or positive manner.
known variant F87A6b was tested, which led to a ∼1:1 mixture In contrast to the traditional procedure based on NNK-
of the respective 2β- and 15β-products. For further improve- encoded SM at individual residues lining the binding pocket,13
ment, 20 positions surrounding the extensive binding pocket MLs generate extensive additional information of significant
were identified, grouped into small multiresidue randomization value, if exploited properly. Massive sequencing data of MLs
sites, and subjected to ISM using relatively large amino acid provide a complete fingerprint of the influence on selectivity as
alphabets (AAAs).16a Mutants with good conversion (ca. 80%) well as activity of all theoretically possible amino acid mutations
and 97% 2β- or 96% 15β-selectivity were found, although at each position, thereby revealing both positive and negative
activity was suboptimal. It was also necessary to assay a total of effects. As already delineated, mutations that increase activity
∼9000 transformants per substrate by automated HPLC may exert an opposing effect on regio- or stereoselectivity, and
analysis, which constitutes a considerable screening effort vice versa, and this fundamental problem also pertains when
(i.e., 18,000 samples for 2 substrates). In contrast to C2- and considering regio- versus stereoselectivity. MLs have been used
C15-selectivity, steroid functionalization at position C16 is of previously in an increasing number of directed evolution studies
particular synthetic value because the respective alcohols are of other proteins.20 In some cases, positive mutations were
components of biologically active glucocorticoids.17 Using combined with generation of multimutational variants.
3396 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

Scheme 1. Model Reaction for Regio- and Diastereoselective Control of Testosterone (1) Hydroxylation To Generate the
Respective 16α (2) or 16β (3) Alcohols

Figure 1. Mutability landscape of P450BM3 mutant F87A toward testosterone (1) hydroxylation. The five target residues are indicated on the left
side, while the traits under investigation are shown in percentage on the right: Substrate conversion (black) as well as selectivity toward 2β- (red),
15β- (blue), or 16β-hydroxytestosterone (green) are shown. The values of the parent enzyme are shown in squares. Other oxidation sites occur
minimally at positions 19, 1β, and at other unknown positions (<5%). The data (Table S1) represent the %HPLC average values based on three
independent experiments (n = 3). Reaction conditions: 1 mM 1 in 600 μL KPi buffer pH 8.0 for 24 h/220 rpm/37 °C. Average OD values: ∼3.0.

Our combined approach in the present study is fundamen- the often observed poor or incorrect stereoselectivity, limited
tally different because we utilized the ML-derived information substrate scope and insufficient stability.22 Directed evolution
in designing SM-based evolutionary pathways for efficient has progressed during the last two decades to a point where
directed evolution. We chose testosterone (1) as the model these problems can be addressed and generally solved.5,6,13 The
substrate in a multiparameter genetic optimization procedure most common gene mutagenesis methods in directed evolution
encompassing activity, regioselectivity and diastereoselectivity are error-prone PCR (epPCR, a shotgun technique), DNA
at position C16 (Scheme 1). Minimization of trade-offs was a shuffling (a recombination-based method), and SM with
central theme. We also aimed for some degree of generality generation of focused mutant libraries.5 Despite impressive
regarding C16-selectivity by investigating four additional advances, reliable prediction of the influence of even a single
steroids without further mutagenesis experiments. Finally, by mutation on protein function remains difficult,23 not to speak
obtaining crystal structures of mutants, we hoped to gain of several in combination. Random methods like epPCR are
insight into the origin of altered catalytic profiles. We note that often employed, especially when structural or functional
simultaneous multiparameter optimization not only constitutes knowledge is not available. However, such approaches usually
a challenge in CYP-catalyzed reactions, but in directed require screening of large protein combinatorial libraries.24 In
evolution of selective enzymes in general.5,6 contrast, SM at sites lining the enzyme binding pocket
Brief Analysis of Current Directed Evolution Strat- (combinatorial active-site saturation test (CAST)) allows the
egies. Before turning to the results, we briefly review the creation of smaller libraries that are more likely to contain
present status of directed evolution techniques, because this improved variants.13 If necessary, ISM can be employed, which
puts our study into the proper framework. Enzymes have been means that the best mutant identified in a library at a given site
used for a long time for enantio- and regioselective trans- is used as the template for SM-based randomization at another
formations in organic chemistry and biotechnology, but this site.13 Indeed, it has been shown that SM is more efficient than
approach has suffered from the traditional limitations regarding epPCR and DNA shuffling for improving selectivity and/or
3397 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

activity.25 In view of Emil Fischer’s lock-and-key hypothesis and further mutagenesis at positions V78, V87, and L437.
Daniel Koshland’s induced-fit model, this is not surprising. Independently, earlier experiments in our laboratory showed
However, when targeting stereo- and regioselectivity, the labor- that introducing mutation A330W into the F87A variant allows
intensive screening step is the bottleneck,5,13 because selection almost complete oxidation of 1 to form the 2β alcohol.16b
systems such as phage display, and related techniques do not Although this mutation may improve activity in M01/M11, we
work in this area of directed evolution.26 Efficient methods for reasoned that this semirational design approach might not be
creating small and smart libraries are therefore needed. the best strategy in the light of previous failed attempts.
Different strategies have been developed to create such focused Furthermore, M01-derived mutants are somewhat unstable due
libraries:5,13,21 (1) Split large randomization sites for SM into to disruption of salt bridges and hydrophobic contacts and are
smaller ones and use large or small AAAs. (2) Identify thus not ideal biocatalysts for industrial applications.27 For
beneficial mutations by acquiring experimentally mutational these reasons, we hypothesized that ISM-based CASTing would
data. (3) Employ computational approaches to minimize library engineer highly active, selective and stable biocatalysts by
size in silico. A more extensive analysis with additional starting with the WT enzyme.
references is provided in the Supporting Information. Library Design and Statistical Analysis. Having analyzed

■ RESULTS AND ANALYSES

Learning from Sequence−Function Relationships. As
sequence-function relationships from 11 residues systematically
(R47, T49, Y51, V78, A82) and partially (S72, F87, M185,
L188, A330 and L437), we considered more extensive SM.
reported earlier,16a we started with mutant F87A, which However, it may not be necessary to mutagenize all residues.
unselectively leads to equimolar amounts of 2β- and 15β- Because R47W and T49E improve activity significantly, but the
hydroxytestosterone. Previously, we used this template to build former mutation exhibits higher selectivity for product 3
three combinatorial libraries for possible ISM experiments: (dubbed 3-selectivity) (Figure 1), position T49 was not
Libraries A (R47/T49/Y51) and C (M185/L188) were considered. Given that L181, L188, and M185 are close to
randomized using NDT codon degeneracy (12 amino acids), each other, we decided to eliminate the latter because L188 is
whereas NNK was used for constructing library B (V78/A82). present in mutants M01/M11/F87A, while keeping L181 due
Residues R47 and Y51 interact with the carboxylate group of to its proximity to V78 and A82 (see below), which may enable
the natural fatty acid substrates near the mouth of the substrate potential synergistic interactions.28 Finally, 10 residues were
entry cavity, whereas residues V78/A82 are located at the active chosen. Before considering residue grouping, AAA size and
site of the heme domain (see below). Because mutations to identity, we applied the Patrick and Firth29 as well as Nov30
small amino acids (Gly/Cys/Ser) in residues M185/L188 metrics in two library designs that were successfully used in
(library C) correlated mostly with high conversion of 1 (50− various enzymes5,13,21 to estimate screening effort at different
80%) but not with selectivity,16 we have now investigated in degrees of library coverage assuming absence of amino acid bias
more detail the role of each amino acid substitution at the five (Table S2).
residues of previous libraries A and B (R47, T49, Y51, V78, Table S2 shows the effort required for screening
A82) by building a ML with mutant F87A as the template. combinatorial libraries of up to 10 residues randomized to 2
In total, 95 mutants were constructed and screened using or 4 amino acids. In the case of the binary AAA (2-AAA),
whole cells as described in the SI (Figure 1). Interestingly, screening can be quite low (767 samples) when coverage is
mutations R47W, T49E, and A82L/F/E/Q enhance substrate 53% but increases by about 4-fold (3076 samples) with 95%
conversion by ∼2−3-fold, whereas polar (T/Y/H/K/R/D/E/ library coverage. Such a library maximizes the chances of
Q/N) substitutions at position 78 diminish significantly or reshaping a binding pocket while decreasing the structural
completely abolish activity regardless of the side-chain length. diversity at each site. On the other hand, an AAA of 4 members
Moreover, regioselectivity for position 2β or 15β is controlled (4-AAA) maximizes a more localized structural diversity, but it
by many diverse substitutions at positions Y51 or V78/A82, increases the screening effort dramatically, irrespective of library
respectively. In contrast, A82W is the only mutation that shifts coverage. To resolve this issue, ISM could be used by screening
regioselectivity significantly (3% → 41%) toward position 16β two or more multisites composed of 4 residues randomized
without compromising activity. These results indicate that only with 4-AAAs, as only 766 samples are necessary for 95%
a minor fraction of the mutations are beneficial, while the coverage. Although this is a practical number, the effort rises
majority of mutations are neutral and many are entirely proportionally to the number of small libraries. In view of
deleterious in terms of activity and regioselectivity. Moreover, previous observations that extensive library coverage is not
the emergence of mutation A82W for shifting regioselectivity necessary for obtaining acceptable mutants,16b we divided the
confirms its importance regardless of genetic background, as 10 target residues in two multisites (A and B), each composed
substantiated by subsequent data (see below). These lessons of 5 residues. Their randomization with a reduced 4-AAA
are important for designing reduced AAAs. requires 767 samples for 53% library coverage, which is a
As stated above, Commandeur et al. reported in a rational practical number given our low-throughput HPLC screening
design approach two mutants dubbed “M01” (R47L/F87V/ method of 8 min per sample (see SI). This decision led to the
L188Q/E267V/G415S) and “M11” (M01 plus mutations question as to which residues should be included in groups A
E64G/F81I/E143G) which are relevant to our study: Mutation and B.
A82W shifts regioselectivity in the hydroxylation of 1 from 15β This grouping was chosen so that the respective residues are
to 16β (3).18d A similar effect was caused by introducing the spatially close to one another, allowing maximal probability of
F87I mutation,18c whereas mutation S72I in mutants M01/ cooperative effects. However, primer costs should also be
A82W and M11 inverted stereoselectivity in favor of 16α- considered if nondegenerate codons are used (especially if two
hydroxytestosterone (2).18e Unfortunately, selectivity did not or more contiguous codons are simultaneously targeted with a
improve beyond ∼80% in any mutant, while enzymatic activity large AAA). Because mutagenesis of F87 toward a small residue
was severely compromised by at least 10-fold, even upon is crucial for accepting substrate 1, this residue was assigned to
3398 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

group A. This means that we planned to start with WT Table 1. Semi-Rational Choice of Reduced Amino Acid
P450BM3 as the parent gene. Residues S72 and A82 were also Alphabets for 10 Target Residues
assigned to this first group because mutations S72I and A82W
group target AAAa rationaleb reference(s)
are required for the formation of products 2 and 3, respectively.
We also included R47 and L437 in this group because the A R47 W enhances 1 conversion to any this work
product by ∼2-fold in F87A
former could act as an activity-enhancer residue, whereas the
L enhances 1 conversion to any this work
latter is close to S72, A82, and F87 (Figure 2), which may also product by ∼1-fold in F87A and18
T neutral effect on 1 conversion to any this work
product in F87A
A S72 V moderate 3-selectivity in mutants 18e
M01/M11
I high 2-selectivity in mutants M01/ 18e
M11
F unknown effect; large hydrophobic -
side chain
A A82 W high 3-selectivity in mutants F87A this work
and M01/M11 and 18d, e
F enhances 1 conversion to 15β this work
hydroxy-1 by ∼0.5-fold in F87A
M enhances 1 conversion to 15β this work
hydroxy-1 by ∼2-fold in F87A
Figure 2. P450BM3 library design. The 10 target residues are divided in A F87 V moderate effect on 3-selectivity in 18c
groups A (green) and B (blue), each composed of 5 members. mutants M01/M11
I moderate effect on 2-selectivity in 18c
mutants M01/M11
result in cooperative (more than additive) mutational effects. M moderate effect on 3-selectivity in 18c
The remaining residues Y51, V78, L181, L188, and A330 were mutants M01/M11
assigned to group B. We then defined the following strategy: A L437 A potentially improves steroid 6h
(1) Screen library A with the goal of finding selective mutants conversion
for 2 and 3, thus assuring at an early stage the emergence of S positive effect on 1 conversion to 2 18e
or 3 in M11
each diastereoselectivity. (2) Next perform ISM by generating
N positive effect on 1 conversion to 2 18e
two B libraries, one for each template, followed by screening. In or 3 in M11
total, three combinatorial libraries were created, each with 1024 B Y51 W diversity; neutral effect on 1 this work
possible variants, but a low library coverage (e.g., 53%) conversion to any product in F87A
requiring the screening of less than 1000 samples per C diversity; negative effect on 1 this work
mutagenesis round (Table S2). conversion to any product in F87A
The question that follows is the selection of optimal AAAs. H diversity; negative effect on 1 this work
conversion to any product in F87A
Although the same alphabet could be used to target all
B V78 M positive effect on 3-selectivity/1 31
multisites,13,21 gathering data from the experimental system conversion in F87A/A82W
enables a more rational approach. ML- and NNK-based studies C positive effect on 3-selectivity/1 31
usually consider only beneficial mutations as randomization conversion in F87A/A82W
building blocks that may induce additive mutational effects. D positive effect on 3-selectivity/1 31
However, nonadditive (more than additive), that is, coopera- conversion in F87A/A82W
tive, effects are becoming increasingly important in protein B L181 G unknown effect; small nonpolar side -
chain
engineering, especially in SM and ISM.28 For this reason, C unknown effect; small polar side -
instead of only considering beneficial mutations in AAAs, in chain
some cases beneficial and neutral or nonbeneficial mutations Q unknown effect; large polar side -
are also included. chain
Table 1 summarizes the rationale behind the formation of B L188 G potential positive effect on 1 16b
conversion to 2β or 15β hydroxy-1
each AAA. For example, at position 47 we include mutation in F87A
R47L as it has been reported to increase oxidation activity C potential positive effect on 1 16b
toward a wide range of natural products,6b and it could be conversion to 2β or 15β hydroxy-1
important for steroid oxidation as well, as exemplified by in F87A
mutants M01/M11.8 However, mutation R47W, which seems Q potential positive effect on 1
conversion to 2 or 3 in M11
18e
to be very important for activity, has not been described. We B A330 W high 1 conversion to 2β hydroxy-1 in 31
also included mutation R47T because it provides a small side- mutant F87A
chain and still shows detectable 3-selectivity and basal activity L low 1 conversion to 2β hydroxy-1 31
(Figure 1). and moderate 3-selectivity in F87A
In other cases, our intention was to maximize chemical V high 1 conversion to 2β hydroxy- 31
1and low 3-selectivity in F87A
diversity by including side-chain substitutions with unpredict- a
able effects allowing for unexpected results. For example, for AAA = Amino acid alphabet. bM01: R47L/F87V/L188Q/E267V/
residue S72F we basically keep a hydrophobic residue like the G415S. M11 is mutant M01 plus E64G/F81I/E143G.
others previously reported V/I, but with increased side-chain
size. In the case of L181 and L188, we simply chose the same
alphabet G/C/Q in both cases, although L188G/C/Q were the many other mutations that are known to increase activity
only reported ones. In the vast P450BM3 literature,6b there are toward a wide range of substrates, a notable example being
3399 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

Table 2. Selected Hits (Improved Mutants) for Diastereoselective Control of 2 and 3

mutationsb
library A library B conv. (%) selectivity (%)
a
ISM round code R47 S72 A82 F87 L437 Y51 V78 L181 L188 A330 1 2 3 othersc
WT inactive - - -
first LIFII L I F I I 71 ± 5 82 ± 6 3±1 15 ± 1
TIMI T I M I 43 ± 4 90 ± 9 1±0 9±1
LIFI L I F I 83 ± 6 83 ± 9 3±1 14 ± 1
WWV W W V 91 ± 4 0 71 ± 3 29 ± 2
WMV W M V 93 ± 5 0 14 ± 0 86 ± 5
WMI W M I 59 ± 8 1±0 55 ± 8 44 ± 5
second LIFI-WQ L I F I W Q 96 ± 2 89 ± 2 4±0 7±1
LIFI-CW L I F I C W 89 ± 1 92 ± 1 4±2 4±1
LIFI-WQM L I F I W Q M 98 ± 1 89 ± 1 2±1 9±1
LIFI-WC L I F I W C 90 ± 4 93 ± 3 4±1 3±1
WWV-HQM W W V H Q M 74 ± 2 1±1 87 ± 1 12 ± 1
WWV-WQW W W V W Q W 63 ± 5 11 ± 0 17 ± 1 72 ± 7
WWV-Q W W V Q 84 ± 3 0 85 ± 2 15 ± 1
WWV-M W W V M 92 ± 2 0 81 ± 2 19 ± 1
third WIFI-WC W I F I W C 95 ± 3 96 ± 2 4±0 0
LIFI-YC L I F I Y C 77 ± 5 92 ± 5 2±0 6±1
LIWI-WC L I W I W C 62 ± 3 95 ± 5 2±0 3±1
LIFV-WC L I F V W C 64 ± 1 64 ± 1 2±0 34 ± 1
LIFI-WQ L I F I W Q 62 ± 6 93 ± 8 3±1 4±1
LWV-Q L W V Q 66 ± 4 0 82 ± 5 18 ± 1
WWV-WQ W W V W Q 67 ± 7 1±1 72 ± 8 27 ± 3
WFV-Q W F V Q 92 ± 3 0 19 ± 0 81 ± 3
WWI-Q W W I Q 36 ± 7 7±1 76 ± 9 17 ± 4
WWV-C W W V C 76 ± 5 0 82 ± 6 18 ± 1
a
Evolutionary step: In the first round, 950 mutants were screened upon CASTing at the five residues of library A using BM3 WT as parent enzyme.
The three most active/selective variants for each product are shown. In the second step, mutants LIFI and WWV were chosen as parents; for each
template, CASTing at the five residues of library B was performed, resulting in the screening of 950 mutants each (1900 in total). The four most
active/selective mutants for each product are shown. In the third round, upon MD simulations, five mutants were created for each template LIFI-WC
and WWV-Q. The screening results for the 10 resulting mutants are shown. bAt each ISM round, the data for the respective 16α and 16β alcohols
are shown above and below, respectively. cOthers refer to 15β- > 2β- > unknown products; 6α-, 15α-, 19, 6β, 7α-, 2α-, 11α-, 1β, 11β, 2β or 4-
hydroxytestosterone were not detected. The values are the mean of three independent experiments displaying the standard error mean (n = 3).
Reaction conditions: 1 mM 1 in 600 μL KPi buffer pH 8.0 for 24 h/220 rpm/37 °C. Average OD values: ∼3.0.

L188Q which is present in mutants M01/M11. We included we first tested the “Gibson method” using gaps between oligos
this substitution as well as L188G/C or A82W and others at (Figure S1). However, all attempts using this method failed due
position V78 that have been shown to improve 1 conversion to frame shifting of the coding region along residues S72 to F87
and 3-selectivity in mutant F87A.16,31 Similarly, mutations (data not shown). This issue could be solved using gap-free
L437S/N could improve the conversion of substrate 1 as oligos in another library design, but we decided to use the
reported by Commandeur and colleagues,18e while L437A original primers by performing a variant of QuikChange and
could potentially improve BM3 activity on steroids as reported overlap extension PCR5 to generate library A. Its quality was
by the group of Arnold.6h These mutations were also included. acceptable (Figure S2), as judged by a quick quality control, a
Overall, of the 30 non-WT amino acids, only 4 lack a cautionary step that sequences the pooled library plasmids to
rationale (indicated by -), which corresponds to about 13% avoid futile screening of nonexisting diversity.16,33 Ten 96-well
(random). The remaining mutations are rationally selected on microtiter plates (total of 950 colonies plus 10 controls) were
the basis of ML data (9 out of 30 makes 30%), our own F87A screened using whole cells, as reported earlier.16 We found at
mutants16b,31 (8 out of 30 equals 27%) as well as those mutants least three very good biocatalysts for the formation of each
reported by the groups of Commandeur18 (8 out of 30 makes alcohol, with values of 82−90% selectivity (71−83% con-
27%) and Arnold6h (1 out of 30 equals 3%). Thus, most of the version) for 2 and 47−71% (59−91%, respectively) for 3
AAAs are rationally based on our own data (57%), followed by (Table 2).
that of Commandeur (27%). We next chose the 2-selective LIFI (R47L/S72I/A82F/F87I)
Library Construction, Quality Control, and Screening. and 3-selective WWV (R47W/A82W/F87V) mutants as
The above statistical analysis shows that each multisite requires templates for ISM. We generated two more B libraries, this
a screening effort of 767 samples for 53% library coverage time using a two-step QuikChange reaction for SM, but the
(Table S2). It assumes that no bias is created during initial quality control revealed suboptimal results, even after
mutagenesis, which however is generally not the case.32 Thus, various optimization attempts (Figure S3). We nevertheless
library bias should be taken into account when library creation screened 950 colonies per case (1900 transformants in total).
is not perfect.30 For library construction in the assembly step, Interestingly, the addition of mutations Y51W/L181C to LIFI
3400 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

(LIFI-WC) enhanced both 2-selectivity (83 → 93%) and

activity (83 → 90%), while the single L181Q mutation in
WWV (WWV-Q) enhanced 3-selectivity (71 → 85%), but
decreased conversion somewhat (91 → 84%) (Table 2).
Although the catalytic profiles are already quite good, further
improvements seemed possible. At this stage, instead of more
extensive screening, we turned to MDs with the aim of
understanding the role of the mutations and with the hope of
gaining computational guides for additional protein engineer-
ing. It should be noted that in simultaneous genetic
multiparameter optimization, it is not optimal to choose a
mutant for further mutagenesis that has the very best score in
one property (e.g., activity). Rather, as practiced here and noted
earlier,33 rational compromises in terms of balanced (minimal)
trade-offs need to be made. For instance, mutants LIFI-WQM
and WWV-M show the highest respective conversion values of
1 toward 2 (98%) and 3 (92%), but the selectivity is ca. 4%
lower than in mutants LIFI-WC and WWV-Q.
Molecular Dynamics Simulations. Traditionally, MD
simulations are performed to investigate the origin of stereo-
and regioselectivity of biocatalysts,5,6 but seldom for guiding the
actual engineering process, with a few recent examples.34 MDs
were performed on the LIFI-WC (R47L/S72I/A82F/S72I−
Y51W/L181C) and WWV-Q (R47W/A82W/F87V-L181Q)
mutants. Preliminary experiments on the apo enzyme models
were carried out to allow relaxation of the protein in the
presence of the introduced mutations. Three 50 ns unrestrained
simulations were performed, followed by clustering of these
geometries. The root-mean-square deviation (RMSD) of the
backbone α carbon atoms from the initial structure was
measured for the MD simulations of both mutants (LIFI-WC
and WWV-Q) in the reaction with 1 (Figures S4−S5). The Figure 3. Substrate docking in models of the engineered P450BM3
values are small and oscillate around a fixed value for most of mutants. Active sites of the (a) LIFI-WC and (b) WWV-Q mutants
the simulations, suggesting stable model systems. The root- are shown with testosterone (green) docking in the poses showing the
mean-square fluctuations (RMSF) of the backbone atoms of highest scores. The mutated residues are shown in space-fill format
the individual amino acids were averaged over the simulation. (magenta). Mutant models were created based on the 1BU7 WT
crystal structure.35 The two models shown display 1 in binding poses
The regions of high RMSF correspond to the loops between
that prove to be consistent with the experimentally observed
the alpha helices (Figures S6−S7). Docking of 1 to the selectivity, with the hydrogen atom to undergo abstraction closest to
representative structures from the clustering calculation was the catalytically active oxidant Compound I.36,37
performed before further MDs were carried out on the
enzyme/substrate complexes. As a control, docking and
simulations were also performed with the WT enzyme; no during the entire MD trajectories (Figures S10−S11). This
productive 1 binding modes were observed (data not shown), suggests that the active site is not entirely optimized for binding
in agreement with the lack of activity for WT P450BM3 with of 1 in positions that would result in exclusive hydroxylation at
testosterone. This is due to steric hindrance around the heme their respective position. This is further supported by the data
caused by the F87 side chain. In contrast, productive binding in Table S5, which show the average distance between the O
orientations were observed for 1 in the LIFI-WC and WWV-Q atom of the catalytically active species heme-Fe IVO
mutants that are consistent with the observed selectivity porphyrin radical cation (Cpd I)1 and the hydrogen atoms at
(Figure 3), i.e., poses were observed in which hydrogen the 16α and 16β positions. Apart from one of the three LIFI-
abstraction leading to formation of 2 and 3 would be favored WC simulations, all simulations have a smaller value for the
(Tables S3−S4). Movies S1 and S2 respectively show that the average O−H16β distance, compared to O−H16α. Remarkably,
binding orientations of 1 are consistent with the experimentally the docking and MD calculations can explain the roles of the
observed hydroxylation selectivity by mutants LIFI-WC and majority (e.g., R47, S72, A82, F87, L181Q) but not of all the
WWV-Q (highlighted in stick representation are the testoster- mutated residues (Y51, L181C). For instance, residues Q73
one molecule, heme modeled as Compound I, and the mutated (caused by mutation S72I) and S72 form a H-bond to the
residues). Moreover, a hydrogen bonding interaction is carbonyl group of 1 to position the substrate in the right
observed between the carbonyl oxygen of 1 and the Q73 side orientation in mutants LIFI-WC and WWV-Q, respectively.
chain in the LIFI-WC variant (Figure S8), whereas in WWV-Q Likewise, mutation L181Q enables formation of a H-bond to
a hydrogen bonding interaction is observed between the residue T436 in mutant WWV-Q (see below), but the role of
carbonyl oxygen atom of 1 and the S72 side chain (Figure S9). mutation L181C in LIFI-WC is not clear. Finally, introducing
Starting from the above docking poses, the hydrogen L/W in position 47 breaks a salt bridge to Q352, whereas
bonding interactions observed between the carbonyl group of substitutions I/V at position F87 remove steric congestion
1 and the side chains of the mutant enzymes are not present around the heme and F/W at position A82 reshape the active
3401 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

site, as reported elsewhere.9f However, in the three latter cases, Q to gain a deeper insight of the reactivity of the substrate in
it is unclear why one specific mutation is required for each the active site. Residue T436 was addressed first (Figure 5a).
mutant. To better understand these “side-chain differences”, we
swapped one amino acid to the corresponding one in each
mutant. In LIFI-WC, L47 was mutated to W and W47 to L in
WWV-Q, and so on for residues 51, 82, 87, and 181. The data
are listed at the bottom of Table 2. Importantly, mutation
L47W increases 2-selectivity (93 → 96%) and conversion (90
→ 95%), but the inverse mutation W47L decreases both 3-
selectivity (85 → 82%) and conversion (84 → 66%), indicating
that Trp at position 47 is crucial to oxidize 1 with great
efficiency regardless of selectivity. The remaining four
mutations in LIFI-WC compromise severely activity without
affecting 2-selectivity, except when I87 was mutated to Val
(LIFV-WC), where selectivity changed to a mix (∼2:1)
between 16β and 15β hydroxy-1. In the WWV-Q-derived
mutants, both enzymatic traits are compromised except in
mutant WVF-Q (W82F) that shows an increase in 1 conversion
by 8%, but a severe decrease of 3-selectivity (85 → 19%),
indicating the presence of strong trade-offs, and that other
mutations elsewhere are needed to overcome these effects. At
this stage, one goal out of two has been reached, with mutant
WIFI-WC (R47W/S72I/A82F/S72I−Y51W/L181C) being
96% selective for 2 while catalyzing almost complete 1
conversion (95%). Mutant WWV-Q, however, still required
improvement.
During the MD simulations of variant WWV-Q, the Figure 5. Mutability landscapes (MLs) of 3-selective mutants. (a)
formation of a hydrogen bond between the amide oxygen Mutant WWV-Q was used as a template to construct all possible single
group of the newly introduced glutamine at position 181 and mutants at position T436. Several mutations cause reduced activity (S/
the hydroxyl group of residue T436 was observed (Figure 4), as I/V/P/G), selectivity (E/Y/F/W) or both (K/L/M), while others
increase activity slightly (R), selectivity (S/I/V/P/G), or both (C/D/
Q/A/H). (b) Mutant WWV-QR was used as template for creating all
variants at position M177. Most mutations to small nonpolar (G) and
polar amino acids (S/D/N/E/K/T/Q) cause enhanced activity and
selectivity regardless of the side-chain size. Nonpolar substitutions of
medium (I/L) and large (W/F) side-chains retain activity but impair
selectivity. The bars show the standard error mean (n = 3). Reaction
conditions: 1 mM 1 in 600 μL KPi buffer pH 8.0 for 24 h/220 rpm/37
°C. Average OD values: ∼3.0.

Highly selective mutants were found (89−93%), but the

activity was decreased significantly (72−45%), indicating the
presence of strong trade-offs. On the other hand, mutation
T436R enhances conversion (84 → 92%) without compromis-
ing 3-selectivity (85%). For this reason, we chose the most
active variant displaying the weakest trade-off effects (i.e.,
WWV-QR (WWV-Q plus T436R)).
Figure 4. MD simulations guide engineering efforts. Hydrogen Next, we constructed another ML, but this time at position
bonding is observed between the side chain oxygen atom of Q181 M177 (Figure 5b). Remarkably, amino acids with small side-
and the hydroxyl hydrogen atom of T436 as well as between the side- chains (S/G) improved 3-selectivity (85 → 92%) and
chain amine hydrogen atom of Q181 and the sulfur atom of M177, conversion (92 → 95%), while large ones (F/W/Y) retained
suggesting that these residues are important for substrate orientation
in the active site and are therefore potential mutagenesis targets. conversion (87−91%) at the expense of 3-selectivity (64−
71%), illustrating the existence of strong diminishing returns.
That is, the closer to the optimum, the smaller the benefit per
supported by their average distance (Figure S12), compared to mutation. At this stage, we had achieved our second goal,
that between the side chain nitrogen atom of Q181 and the namely, engineering a highly active (92% conversion of 1) and
hydroxyl oxygen atom of T436 (Figure S13). In fact, hydrogen 3-selective (92%) enzyme WWV-QRS (R47W/A82W/F87V-
bonding is also observed between the side-chain amine L181Q/T436R/M177S).
hydrogen atom of Q181 and the sulfur atom of M177. Because Biochemical, X-ray, and EPR Structural Character-
these interactions influence substrate orientation above the ization of Best Mutants. Using pure enzymes (Figure S14),
heme-FeIVO, we performed mutagenesis experiments with we first determined the dissociation constant (Kd) for 1 binding
the goal of improving 3-selectivity and activity. by the enzymes in both the full-length and heme domain
NNK-randomization is typically performed in such endeav- versions using UV−vis spectroscopic titrations (Figure S15):
ors, but we constructed MLs20 using the parent enzyme WWV- WIFI-WC has Kd values of 0.4 ± 0.1 and 0.49 ± 0.06 μM,
3402 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

Figure 6. X-ray structure of the WIFI-WC mutant and 1 binding mode comparison with that of N-perfluorododecanoyl-L-tryptophan. (a) Cartoon
representation of the 2-selective mutant-testosterone complex. Electron density for both testosterone ligands is shown as a blue mesh contoured at 1
sigma. The C-alpha atoms of the 6 mutations are shown as magenta spheres. (b) Overlay of the active site regions of the 2-selective mutant-
testosterone complex with the WT P450BM3 N-perfluorododecanoyl-L-tryptophan complex (Watanabe et al., PDB 5B2W, DOI: 10.2210/pdb5b2w/
pdb). The individual mutations (in blue) and corresponding WT residues (in green) are shown in sticks with the corresponding ligands respectively
in yellow (testosterone) and magenta (N-perfluorododecanoyl-L-tryptophan).

respectively; whereas WWV-QRS has Kd values of 0.99 ± 0.11 efficiency of 10% and 42% for WIFI-WC and WWV-QRS,
and 1.12 ± 0.03 μM, respectively, indicating that the CPR respectively. WWV-QRS shows a higher NLR than WIFI-WC,
domain does not affect substrate binding in the heme domain. but in the presence of 1 the coupling efficiency of WWV-QRS
Regardless of the construct, the 2-selective mutant has ca. 2- is ∼4-fold higher, indicating not only that nonspecific NADPH
fold stronger binding affinity for 1 compared with the 3- oxidation is minimized in the 1-bound form, but also that this
selective one, suggesting that the former enzyme may have a enzyme is more efficient than the WIFI-WC variant when
better catalytic profile. Steady-state kinetics were also NADPH is limiting. In view of uncoupling of NADPH
determined for the full-length proteins using a NADPH oxidation from substrate oxidation in the WIFI-WC and
regeneration system (Figure S16). Interestingly, WIFI-WC WWV-QRS mutants, catalytic systems for hydroxylated steroid
gives a typical Michaelis−Menten saturation curve for production are unlikely to function efficiently using simple
formation of 2 with Km, kcat and kcat/Km values of 0.28 ± NADPH addition to drive the reaction. In the standard whole-
0.02 μM, 11.56 ± 0.19 s−1 and 4.13 × 107 M−1 s−1, respectively. cell reactions, a system based on NADP+, glucose, and glucose
However, WWV-QRS shows a different behavior for generation dehydrogenase to regenerate NADPH was used. To determine
of 3 with a relatively high turnover frequency at low (ca. 2 μM) the catalytic profile with this regeneration system, the mutants
substrate concentrations (ca. 13 s−1), but decreased rates at were challenged in vitro in the same conditions as for the
higher 1 concentrations with respective Ki, Km, kcat and kcat/Km whole-cell reactions (1 mM substrate), resulting in TTN of
values of 0.73 ± 0.19 μM, 2.53 ± 0.69 μM, 62.12 ± 14 s−1 and 8660 and 9044 with conversion values of 92 and 97% as well as
2.45 × 107 M−1 s−1, suggesting substrate inhibition (i.e., tight selectivity of 94 and 93% for 2 and 3 by mutants WIFI-WC and
binding of 1 and/or the hydroxylated product(s)),38 a WWV-QRS, respectively. These data show that the NADPH
phenomenon that has been observed by the Sligar group in regeneration system is effective in improving the performance
steroid oxidation by other CYPs.39 of the engineered enzymes.
To establish whether NADPH oxidation is well-coupled to Analysis by differential scanning calorimetry (DSC) indicated
steroid substrate oxidation in these mutants, further studies that the steroid-free WWV-QRS mutant had a Tm of 52.4 ±
were done to quantify hydroxylated product formation in 0.1 °C, rising to 55.4 ± 0.1 °C on binding 1. The steroid-free
relation to NADPH oxidation. Hence, NADPH oxidation in WIFI-WC mutant had a Tm of 55.1 ± 0.1 °C, but a decreased
absence of steroid substrate was measured to determine the Tm (52.3 ± 0.1 °C) on 1 binding, a phenomenon possibly
“leak rate”; that is, how fast oxygen reduction occurs due to related to the presence of two 1 binding sites in this mutant.
electron transfer through P450BM3 flavin/heme cofactors. Rate These values are ∼10 °C lower than WT heme in the presence
constants were then compared for 1-bound enzymes, and of its native substrate,40 but still acceptable for practical
particularly for the characterization of the 3-selective mutant applications in whole cell catalysis. The binding of 1 to both
(Table S16). In the absence of substrate, mutant WWV-QRS enzymes was also investigated by electron paramagnetic
shows a ca. 6-fold enhancement in NADPH leak rate (NLR) resonance (EPR) spectroscopy (Figure S17). The X-band
compared with WIFI-WC (38 vs 6 nmol s−1). However, in the EPR analyses reveal rhombic EPR spectra for both WIFI-WC
presence of 1, the values are quite similar (24 vs 22 nmol s−1). and WWW-QRS mutant in their native forms, and following
The product formation rate of 3, however, is ca. 4.5 times faster the addition of ethanol (2%, as solvent for 1) and then of 1
than that of 2 (9 vs 2 nmol s−1), resulting in higher 1 final itself. In all cases, the spectra are characteristic of correctly
conversion values of 33 vs 8%, respectively, with 1 equiv of folded, low-spin ferric P450 enzymes that have cysteine thiolate
NADPH. Within 2 min, ∼154 nmol NADPH are depleted, but as the proximal ligand to the heme iron. The low-spin spectra
only 16 and 65 nmol of 1 are consumed, indicating a coupling of both native mutants are perturbed most significantly by
3403 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

Table 3. Subset of Mutants Reported To Be 2- and 3-Selective in a Cell-Free System18e

target residues and mutations regio- and stereoselectivity (%)
code R47 E64 S72 F81 A82 F87 E143 L188 E267 G415 1 conv. (%) 2β 15β 16α 16β othersa
WT inactive - - - - -
I I 1±1 0 17 ± 9 0 83 ± 9 0
V V 9±3 14 ± 4 49 ± 8 0 36 ± 6 1±1
LV L V 14 ± 2 16 ± 2 58 ± 8 0 25 ± 3 1±1
LVQ L V Q 39 ± 5 15 ± 1 66 ± 9 0 17 ± 2 2±1
LVQS L V Q S 45 ± 7 17 ± 2 63 ± 10 0 18 ± 3 3±1
M01 L V Q V S 52 ± 5 23 ± 2 35 ± 4 1±1 31 ± 4 10 ± 2
M01-W L W V Q V S 79 ± 8 1±1 15 ± 7 1±1 67 ± 6 16 ± 4
M01-IW L I W V Q V S 37 ± 7 14 ± 4 14 ± 9 49 ± 8 6±1 16 ± 6
M01-VW L V W V Q V S 2±2 0 0 81 ± 9 11 ± 7 8±8
M01-WI L W I Q V S 26 ± 5 2±1 2±1 3±1 82 ± 14 11 ± 2
M11 L G I V G Q V S 77 ± 2 14 ± 1 30 ± 15 2±1 18 ± 1 38 ± 18
M11-I L G I I V G Q V S 28 ± 6 38 ± 8 43 ± 9 5±1 4±1 10 ± 2
M11-II L G I I I G Q V S 13 ± 3 12 ± 3 3±1 44 ± 10 9±2 33 ± 6
a
The major species are unknown followed by 11α-, 2α-, and 19-hydroxytestosterone. The values are the mean of three independent experiments
displaying the standard error (n = 3). Reaction conditions: 1 mM 1 in 600 μL KPi buffer pH 8.0 for 24 h/220 rpm/37 °C. Average OD values: ∼3.0.

ethanol addition, but there is no significant high-spin, ferric were screened using whole cells as described in the
heme iron signal observed for either mutant; a phenomenon experimental section. The most selective and active mutants
likely due to the low temperature required for EPR spectral reported are M01/S72I/A82W (M01-IW) and M01/A82W
analysis of P450s and other heme proteins. (M01-W), which in pure form are ca. 85% 2- and 3-selective,
To obtain further information on the binding modes of 1, respectively, but 2.6- and 8.3-fold less active than M11, the
attempts were made to cocrystallize the 2-selective WIFI-WC most active variant reported.18e In whole cells, mutants M01-
and 3-selective WWV-QRS mutant heme domains in complex IW and M01-W show 49% or 67% 2- or 3-selectivity and 37%
with 1. Crystals of the WWV-QRS variant heme domain bound or 79% 1 conversion, respectively. Clearly, these enzymes show
to 1 could not be obtained, but successful crystallization of the poor selectivity and conversion in whole cells, which is often
WIFI-WC heme domain complex with 1 was achieved (PDB the preferred industrial method for expressing biocatalysts,
5OG9 and Figure 6). The WIFI-WC mutant heme domain because additional and costly steps like cell disruption,
crystals diffracted to a resolution of 2.1 Å (see Table S7 in SI centrifugation and separation are avoided. In fact, the product
for details). These data reveal additional electron density in the formation rates reported for M01-IW and M01-W toward 2 and
2-selective mutant active site for both monomers present in the 3 are 0.048 and 0.11 nmol s−1, respectively.18e Our WIFI-WC
asymmetric unit that clearly corresponds to the presence of two and WWV-QRS mutants show respective values of 2 and 9
1 ligands. One of these is located directly above the heme (1A) nmol s−1 for the formation rates of the 2 and 3 alcohols (i.e.,
with the C16 near the heme iron (Figure 6a). The second one 42- and 80-fold enhancements compared with M01-IW and
(1B) is located adjacent to 1A and occupies the more distant M01-W).
region of the active site near to the R47-Y51 fatty acid Conclusively, our results indicate that whole cell systems can
carboxylate binding motif. The 2-selective mutant-1 complex is significantly improve the performance of suboptimal enzymes.
most like the structure of the WT P450BM3 heme domain in Although ISM as originally developed appears to be more
complex with a “decoy” molecule,41 N-perfluorododecanoyl-L- successful than epPCR and rational design for engineering
tryptophan, and an overlay reveals that 1A occupies the highly active and selective biocatalysts, in the present study the
perfluorododecanoyl binding pocket while 1B is near the L- efficiency had to be improved further for significant results: Our
tryptophan binding region for this decoy molecule (Figure 6b). present strategy required the screening of about 3000 samples
The overlay illustrates how individual mutations each (Figure S19), which is 3-fold lower compared to previous ISM
contribute to 1 binding: S72I, A82F and F87I form the 1A approaches.16 Conversely, using the semi-rationally designed
binding pocket, while R47W, Y51W and S72I contribute to the triple mutant R47L/F87V/L188Q, 1400 colonies were
hydrophobic 1B binding site. There is no direct contact screened after 3 epPCR rounds to obtain the M01 and M11
between the testosterones and L181C. Compared with the WT parent mutants.18 However, these enzymes are somewhat
N-perfluorododecanoyl-L-tryptophan complex, the N-terminal unstable27 and might require extensive screening effort to
region of the heme domain B′-helix has adopted an extended improve not only stability, but also activity and selectivity.
loop structure in the 2-selective mutant-1 complex. This allows Finally, we assessed the potential utility of our enzymes evolved
for expansion of the active site to enable the binding of 1A, and via ISM, MLs and MDs vs epPCR and rational design by
mutation S72I is in this region. Figure S18 shows a stereoview performing upscaling reactions with 30 mg of 1 in 100 mL (ca.
of the binding modes of 1A and 1B in the substrate binding 1 mM). Mutants WIFI-WC or WWV-QRS convert almost all
cavity of the WIFI-WC heme domain. substrate (84−99% conversion) into products 2 or 3 (91−95%
Comparison of ISM and Rational Design. To compare selectivity) within 6 h, whereas mutants M01-IW and M01-
our engineered enzymes with the M01 and M11-derived W18e show low conversion (40−50%) as well as 72 and 82%
mutants reported by Commandeur et al.,18 we created via site- selectivity, respectively, and these values were not improved
directed mutagenesis not only the highly active 2- or 3-selective even after 30 h reaction (Figures S20−S24). Using our best
variants, but also intermediate mutants (Table 3). All variants mutants, we isolated 23.3 and 7.7 mg (yields of 77% and 27%)
3404 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

Scheme 2. Efficient Mutants Found for the C16-Selective Hydroxylation of Other Steroids without Performing Additional
Mutagenesis Experiments: Androstenedione (4), Nandrolone (7), Boldenone (10), and Norethindrone (13) and Their
Respective Alcohols at Position C16a

a
Mutant identities are listed in Table 2. The conversion and selectivity values (% conversion as determined by HPLC analysis) arise from the original
data-sets located in Tables S8−S11 (n = 3). Reaction conditions: 1 mM 4/7/10/13 in 600 μL KPi buffer pH 8.0 for 24 h/220 rpm/37 °C. Average
OD values: ∼3.0.

of 2 and 3, respectively. We faced technical issues during the our knowledge 4 have never been reported: 9, 11, 12 and 14.
product isolation procedure, but these were not optimized These compounds were characterized by NMR and LC-MS to
because the main goal of this experiment was to compare in a determine the stereoconfiguration and molecular mass (see
large scale setup the catalytic profile of our evolved enzymes experimental section in SI). Two ways to induce further fine-
with the previously reported ones.18 tuning would be to either screen all 3000 mutants, or to employ
Exploring Substrate Acceptance by Using Four Other the best mutants as templates for further ISM combined with
Steroids. Although universal catalysts for selective trans- MLs and/or MDs
formations cannot exist due to fundamental reasons, organic
chemists seek catalysts that are efficient not just for a single
substrate. Therefore, we tested four other steroids in the quest
■ DISCUSSION AND OUTLOOK
The goal of this study was to demonstrate how optimally
to achieve regio- and diastereoselectivity at position C16, chosen directed evolution techniques can be combined to
specifically androstenedione (4), nandrolone (7), boldenone develop an efficient strategy that enables the generation of
(10) and norethindrone (13). Instead of screening all 3000 small, high quality libraries of active P450BM3 mutants for
mutants, we focused on the 24 improved mutants that are listed efficient steroid hydroxylation at the specifically targeted C16
in Table 2. In general, many different mutants at each ISM position. The strategy makes use of known mechanistic
round exhibited excellent conversion and selectivity values information such as X-ray structural data, exploratory muta-
(Tables S8−S11). The most efficient mutants are summarized tional experiments based on mutability landscapes (MLs)20 for
in Scheme 2. choosing optimal reduced amino acid alphabets (AAAs) for
In the case of 4, variants WIFI-WC and WWV-HQM display saturation mutagenesis (SM) at residues lining the binding
excellent selectivity (95 and 100%) as well as conversion values pocket (CAST) 13 and iterative saturation mutagenesis
(85 and 93%) toward the 16α (5) and 16β (6) positions, (ISM)13,16,21 in which the choice of the evolutionary pathways
respectively. Substrate 7 was converted to the 16α (8) alcohol was made semi-rationally. At appropriate points, MD
with high diastereoselectivity (98%) and acceptable conversion simulations served as additional guides. A few CAST residues
(71%) using LIFI-CW, whereas WWV-Q produced 16β- had been used earlier in other SM studies, but the specific
hydroxy-nandrolone (9) with good selectivity (90%) and mutations did not serve as a guide in our decision making. As a
conversion (91%). Substrate 10 was also hydroxylated at result, unusually active variants were evolved that catalyze the
positions 16α (11) or 16β (12) with 97% selectivity/70% hydroxylation of steroids regioselectively at the desired C16
conversion or 72% selectivity/71% conversion by mutants position with pronounced α- and β-diastereoselectivity on an
LIWI-CW and WWV-Q, respectively. Although no mutant was optional basis. Thus, the problem of multiparameter
found to yield the 16α alcohol of substrate 13, mutant WMI optimization was solved, certainly in the present study.
yielded the 16β (14) alcohol with 91% selectivity and 59% Along the various evolutionary trajectories, we observed
conversion. In total, 96 samples were screened, providing a pronounced trade-offs and diminishing returns on both
formidable outcome because no additional mutagenesis experi- enzymatic traits, activity and selectivity. These phenomena,
ments were required. Notably, of the 7 products obtained, to which are common in protein engineering and directed
3405 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

evolution efforts, and still constitute widespread bottlenecks,14 showed high stereo and regioselectivity at various positions, of
illustrate the complexity of simultaneously evolving proficient which only three were 2-selective.44 Inspired by this study, two
biocatalysts for more than one catalytic trait. In other ISM new 2-selective CYP154s were recently characterized:45
studies, this issue has been addressed by exploring many CYP154C5 from the bacterium Nocardia farcinica enables the
possible evolutionary pathways,13 but the typical approach formation of 2, yet its TTN of <600 is a 14-fold lower value
requires extensive screening of combinatorial libraries. Such than that achieved by mutant WIFI-WC (8,660) in the present
trading-of f screening does not contribute to “greener” directed study. Our approach shows how P450BM3, the most active CYP
evolution. With reduced screening effort, MLs also enable a monooxygenase,6b can be engineered for optimizing challeng-
deeper understanding of landscape regions that are more prone ing non-native reactions. We focused from the beginning on
to larger trade-offs than others. This is the reason why we chose position C16, allowing access not only to the desired products
mutant WWV-QR (showing diminishing returns on activity 2 or 3 on an optional basis, but also to other steroidal products
without selectivity trade-offs) as template for the next ISM 5 or 6, 8 or 9, 11 or 12 and 14. The potential medicinal utility
cycle. Although MLs are more expensive to construct than of some of these compounds remains to be tested, as in the
NNK-based SM or even epPCR libraries, they provide form of glucocorticoids.17
extensive information which enables an excellent understanding While successful in the selective and rapid oxidative
of sequence−function relationships (as exemplified by mutation hydroxylation of five different steroids at position C16, we do
R47W that is present in two different mutants). A universal not claim that the specific recipe described herein will enable
strategy to minimize total costs cannot be proposed, because any position to be targeted. Rather, the present data indicate
each project is likely to be different, depending upon such that the general approach is promising, but that additional
factors as the screening method, expenditure for primers and randomization sites will probably have to be considered when
specific labor costs in academia versus industry. In fact, if the targeting other positions specifically. Moreover, it would be
screening step is too expensive, MLs can be cost-effective.32a impossible to obtain all target mutants at one specific active site
Moreover, since the cost of DNA synthesis and sequencing is residue using random methods such as epPCR or DNA
decreasing, we expect that the use of MLs as described herein shuffling. This does not mean that epPCR or DNA shuffling
will make this approach cheaper in the future, so that the user should be discarded, because a final round of such techniques
can benefit from the information regarding sequence-function on top of CAST/ISM may lead to further improvements.46
relationships as a basis for rationally designing directed With respect to future applications of CYP directed evolution
evolution experiments.20 Because each unique mutant is in general, we suggest that ISM-based CASTing13,16,21
sequenced and located in a defined well of a microtiter plate, combined with MLs20 and computational techniques47 like
MLs enable fast screening with new substrates and/or to look MDs34 constitutes a viable approach in planned control of
at other traits such as stability with the lowest screening effort. targeted regio- and stereoselective hydroxylation with high
In contrast to our choices in defining hits to be used in activity, while keeping the screening effort as low as possible.
further SM experiments, the most selective mutants exhibited Finally, the combined approach introduced in the present study
strong trade-offs on conversion, suggesting that many more can be expected to be a viable option in future directed
evolution of other enzymes as well.

■
engineering steps would be required for increasing activity.
Again, focusing solely on selectivity does not constitute the
optimal strategy.16,33 Overall, our combined directed evolution ASSOCIATED CONTENT
concept proved to be superior to previous protein engineering *
S Supporting Information
approaches. Most importantly, the invariably occurring trade- The Supporting Information is available free of charge on the
offs between selectivity and activity were avoided. Mutants ACS Publications website at DOI: 10.1021/acscatal.8b00389.
WIFI-WC and WWV-QRS display kcat/Km values of 4.13 × 107 Experimental section; lists of oligonucleotides; data of
and 2.45 × 107 M−1 s−1, respectively, which are close to those mutability landscapes from mutant F87A; QQC of
of WT P450BM3 toward long-chain fatty acids as natural combinatorial libraries; MDs; SDS-PAGE; UV−vis
substrates (5−6 × 107 M−1 s−1).6b,42 Both mutants exhibit spectroscopy; X-ray crystallography; EPR as well as
TTNs close to 9000, suggesting their potential use in industrial steroid screening; and HPLC, LC-MS, and NMR data
applications. (PDF)
The structure of the 1-bound WIFI-WC P450BM3 mutant Movie S1: Testosterone binding in mutant LIFI-WC
heme domain shows that 1 occupies two different positions in (MPG)
the P450, one of which is clearly catalytically relevant and lies Movie S2: Testosterone binding in mutant WWV-Q
relatively coplanar with the P450 heme (1A), positioning its (MPG)

■
C16 close to the heme iron. The second molecule (1B) lies in
the active site entry channel and is orientated orthogonally to AUTHOR INFORMATION
1A. The 1B molecule may stabilize the binding of 1A but is also
Corresponding Authors
likely to impede product dissociation and so lead to slow
*E-mail: [email protected]. Phone: +49 64228
catalysis. In CYP21A2, two progesterone molecules were
25500.
likewise located in the access channel and active site,43
*E-mail: [email protected]. Phone: +44-161-
suggesting that future MDs may require docking of more
3065151.
than one substrate.
Although steroid-specific hydroxylating CYPs of eukaryotic ORCID
and bacterial origin exist, the former frequently show Aitao Li: 0000-0001-6036-4058
recombinant expression difficulties and low catalytic efficiency, Sabrina Hoebenreich: 0000-0002-4715-7123
while the latter have a preferred selectivity for attacking on the Christophe Fares: 0000-0001-6709-5057
β “face”.44 Upon screening 213 bacterial CYPs, only 24 (∼11%) David Leys: 0000-0003-4845-8443
3406 DOI: 10.1021/acscatal.8b00389
ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

Andrew W. Munro: 0000-0002-4642-180X Protocols, 2nd ed.; Gillam, E. M. J., Copp, J. N., Ackerley, D. F., Eds.;
Manfred T. Reetz: 0000-0001-6819-6116 Springer: Berlin, 2014. (g) Goldsmith, M.; Tawfik, D. S. Enzyme
Engineering by Targeted Libraries. Methods Enzymol. 2013, 523, 257−
Present Address 283. (h) Brustad, E. M.; Arnold, F. H. Optimizing Non-natural Protein
¶
C.G.A.-R.: Biosyntia ApS, Copenhagen, 2100, Denmark Function with Directed Evolution. Curr. Opin. Chem. Biol. 2011, 15,
Notes 201−210. (i) Jaeckel, C.; Hilvert, D. Biocatalysts by Evolution. Curr.
The authors declare no competing financial interest. Opin. Biotechnol. 2010, 21, 753−759. (j) Turner, N. J. Directed

■
Evolution Drives the Next Generation of Biocatalysts. Nat. Chem. Biol.
2009, 5, 567−573. (k) Lutz, S.; Bornscheuer, U. T. Protein Engineering
ACKNOWLEDGMENTS
Handbook; Wiley-VCH: Weinheim, 2009. (l) Zeymer, C.; Hilvert, D.
This work was supported by the Max-Planck-Society and the Directed Evolution of Protein Catalysts. Annu. Rev. Biochem. 2018, 87,
LOEWE Research Cluster SynChemBio. We thank Anika DOI: 10.1146/annurev-biochem-062917-012034.
Garczynski for help in determining the yield of upscale (6) Selected key papers and reviews of rational design and directed
reactions. A.W.M. and D.L. acknowledge funding from the evolution of CYPs: (a) O’Reilly, E.; Köhler, V.; Flitsch, S. L.; Turner,
UK Biotechnology and Biological Sciences Research Council N. J. Cytochromes P450 as Useful Biocatalysts: Addressing the
(BBSRC) for grant BB/K001884/1 and for a BBSRC DTP Limitations. Chem. Commun. 2011, 47, 2490−2501. (b) Whitehouse,
C. J. C.; Bell, S. G.; Wong, L.-L. P450 BM3 (CYP102A1): Connecting
PhD studentship award supporting C.G.G. A.J.M. and R.L.
the Dots. Chem. Soc. Rev. 2012, 41, 1218−1260. (c) Dennig, A.;
acknowledge funding from the Engineering and Physical Lülsdorf, N.; Liu, H.; Schwaneberg, U. Regioselective o-Hydroxylation
Sciences Research Council (EPSRC) for grant numbers EP/ of Monosubstituted Benzenes by P450 BM3. Angew. Chem., Int. Ed.
M022609/1, EP/M013219/1, and CCP-BioSim. The research 2013, 52, 8459−8462. (d) Bernhardt, R.; Urlacher, V. B. Cytochromes
of Y.N. was supported by the Israeli Science Foundation, grant P450 as Promising Catalysts for Biotechnological Application:
286/13. Chances and Limitations. Appl. Microbiol. Biotechnol. 2014, 98,

■ REFERENCES
(1) Reviews of P450 enzymes with emphasis on different theoretical,
6185−6203. (e) McIntosh, J. A.; Farwell, C. C.; Arnold, F. H.
Expanding P450 Catalytic Reaction Space Through Evolution and
Engineering. Curr. Opin. Chem. Biol. 2014, 19, 126−134. (f) Roiban,
G.-D.; Reetz, M. T. Expanding the Toolbox of Organic Chemists:
structural, mechanistic and practical aspects: (a) Ortiz de Montellano,
P. R. Hydrocarbon Hydroxylation by Cytochrome P450 Enzymes. Directed Evolution of P450 Monooxygenases as Catalysts in Regio-
Chem. Rev. 2010, 110, 932−948. (b) Guengerich, F. P.; Waterman, M. and Stereoselective Oxidative Hydroxylation. Chem. Commun. 2015,
R.; Egli, M. Recent Structural Insights into Cytochrome P450 51, 2208−2224. (g) Girvan, H. M.; Munro, A. W. Applications of
Function. Trends Pharmacol. Sci. 2016, 37, 625−640. (c) Guengerich, Microbial Cytochrome P450 Enzymes in Biotechnology and Synthetic
F. P.; Munro, A. W. Unusual Cytochrome P450 Enzymes and Biology. Curr. Opin. Chem. Biol. 2016, 31, 136−145. (h) Lewis, J. C.;
Reactions. J. Biol. Chem. 2013, 288, 17065−17073. (d) Shaik, S.; Mantovani, S. M.; Fu, Y.; Snow, C. D.; Komor, R. S.; Wong, C.-H.;
Cohen, S.; Wang, Y.; Chen, H.; Kumar, D.; Thiel, W. P450 Enzymes: Arnold, F. H. Combinatorial Alanine Substitution Enables Rapid
Their Structure, Reactivity, and SelectivityModeled by QM/MM Optimization of Cytochrome P450BM3 for Selective Hydroxylation of
Calculations. Chem. Rev. 2010, 110, 949−1017. (e) Poulos, T. L. Large Substrates. ChemBioChem 2010, 11, 2502−2505. (i) Lewis, J. C.;
Heme Enzyme Structure and Function. Chem. Rev. 2014, 114, 3919− Coelho, P. S.; Arnold, F. H. Enzymatic Functionalization of Carbon−
3962. (f) In Fifty Years of Cytochrome P450 Research; Yamazaki, H., Hydrogen Bonds. Chem. Soc. Rev. 2011, 40, 2003−2021. (j) Fasan, R.
Ed.; Springer: Stuttgart, 2014. (g) Bakkes, P. J.; Riehm, J. L.; Sagadin, Tuning P450 Enzymes as Oxidation Catalysts. ACS Catal. 2012, 2,
T.; Rühlmann, A.; Schubert, P.; Biemann, S.; Girhard, M.; Hutter, M. 647−666. See also: (k) Fasan, R. Enzymatic Catalysis: New
C.; Bernhardt, R.; Urlacher, V. B. Engineering of Versatile Redox Functional Twists for P450s. Nat. Chem. 2017, 9, 609−611.
Partner Fusions that Support Monooxygenase Activity of Functionally (7) Reviews and key studies featuring CH-activating hydroxylation
Diverse Cytochrome P450s. Sci. Rep. 2017, 7, 9570. (h) King-Smith, using human-made reagents or catalysts: (a) Newhouse, T.; Baran, P.
E.; Zwick, C. R., III; Renata, H. Applications of Oxygenases in the S. If C-H Bonds Could Talk: Selective C-H Bond Oxidation. Angew.
Chemoenzymatic Total Synthesis of Complex Natural Products. Chem., Int. Ed. 2011, 50, 3362−3374. (b) White, M. C. Adding
Biochemistry 2018, 57, 403−412. Aliphatic C−H Bond Oxidations to Synthesis. Science 2012, 335, 807−
(2) (a) Urlacher, V. B.; Girhard, M. Cytochrome P450 Mono- 809. (c) Neufeldt, S. R.; Sanford, M. S. Controlling Site Selectivity in
oxygenases: An Update on Perspectives for Synthetic Application. Palladium-Catalyzed C−H Bond Functionalization. Acc. Chem. Res.
Trends Biotechnol. 2012, 30, 26−36. (b) Gillam, E. M. J.; Hayes, M. A. 2012, 45, 936−946. (d) Roduner, E.; Kaim, W.; Sarkar, B.; Urlacher,
The Evolution of Cytochrome P450 Enzymes as Biocatalysts in Drug V. B.; Pleiss, J.; Gläser, R.; Einicke, W. D.; Sprenger, G. A.; Beifuß, U.;
Discovery and Development. Curr. Top. Med. Chem. 2013, 13, 2254− Klemm, E.; Liebner, C.; Hieronymus, H.; Hsu, S. F.; Plietker, B.;
2280. Laschat, S. Selective Catalytic Oxidation of C-H Bonds with Molecular
(3) Hogg, J. A. Steroids, the Steroid Community, and Upjohn in Oxygen. ChemCatChem 2013, 5, 82−112. (e) Nanjo, T.; de Lucca, E.
Perspective: A Profile of Innovation. Steroids 1992, 57, 593−616. C., Jr.; White, M. C. Remote, Late-Stage Oxidation of Aliphatic C−H
(4) Review of protein engineering using rational design: Pleiss, J. In Bonds in Amide-Containing Molecules. J. Am. Chem. Soc. 2017, 139,
Enzyme Catalysis in Organic Synthesis; Drauz, K., Gröger, H., May, O., 14586−14591.
Eds.; Wiley-VCH: Weinheim, 2012, 1, 89. (8) Loskot, S. A.; Romney, D. K.; Arnold, F. H.; Stoltz, B. M.
(5) Recent reviews of enzyme directed evolution: (a) In Directed Enantioselective Total Synthesis of Nigelladine A via Late-Stage C−H
Evolution: Advances and Applications; Alcalde, M., Ed.; Springer: Oxidation Enabled by an Engineered P450 Enzyme. J. Am. Chem. Soc.
Stuttgart, 2017. (b) Reetz, M. T. Directed Evolution of Selective 2017, 139, 10196−10199.
Enzymes: Catalysts for Organic Chemistry and Biotechnology; Wiley- (9) (a) Narhi, L. O.; Fulco, A. J. Characterization of a Catalytically
VCH: Weinheim, 2016. (c) Bommarius, A. S. Biocatalysis: A Status Self-Sufficient 119,000-Dalton cytochrome P-450 Monooxygenase
Report. Annu. Rev. Chem. Biomol. Eng. 2015, 6, 319−345. Induced by Barbiturates in Bacillus Megaterium. J. Biol. Chem. 1986,
(d) Bornscheuer, U. T.; Huisman, G. W.; Kazlauskas, R. J.; Lutz, S.; 261, 7160−7169. (b) Munro, A. W.; Leys, D. J.; McLean, K. J.;
Moore, J. C.; Robins, K. Engineering the Third Wave of Biocatalysis. Marshall, K. R.; Ost, T. W. B.; Daff, S.; Miles, C. S.; Chapman, S. K.;
Nature 2012, 485, 185−194. (e) Denard, C. A.; Ren, H.; Zhao, H. Lysek, D. A.; Moser, C. C.; Page, C. C.; Dutton, P. L. P450 BM3: The
Improving and Repurposing Biocatalysts via Directed Evolution. Curr. Very Model of a Modern Flavocytochrome. Trends Biochem. Sci. 2002,
Opin. Chem. Biol. 2015, 25, 55−64. (f) Gillam, E. M. J.; Copp, J. N.; 27, 250−257. (c) Jovanovic, T.; Farid, R.; Friesner, R. A.; McDermott,
Ackerley, D. F.; Directed Evolution Library Creation: Methods and A. E. Thermal Equilibrium of High- and Low-Spin Forms of

3407 DOI: 10.1021/acscatal.8b00389

ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

Cytochrome P450 BM-3: Repositioning of the Substrate? J. Am. Chem. (18) (a) van Vugt-Lussenburg, B. M. A.; Damsten, M. C.; Maasdijk,
Soc. 2005, 127, 13548−13552. (d) Clodfelter, K. H.; Waxman, D. J.; D. M.; Vermeulen, N. P. E.; Commandeur, J. N. M. Heterotropic and
Vajda, S. Computational Solvent Mapping Reveals the Importance of Homotropic Cooperativity by a Drug-Metabolising Mutant of
Local Conformational Changes for Broad Substrate Specificity in Cytochrome P450 BM3. Biochem. Biophys. Res. Commun. 2006, 346,
Mammalian Cytochromes P450. Biochemistry 2006, 45, 9393−9407. 810−818. (b) van Vugt-Lussenburg, B. M. A.; Stjernschantz, E.;
(e) Schwaneberg, U.; Sprauer, A.; Schmidt-Dannert, C.; Schmid, R. D. Lastdrager, J.; Oostenbrink, C.; Vermeulen, P. E.; Commandeur, J. N.
P450 Monooxygenase in Biotechnology: I. Single-Step, Large-Scale M. Identification of Critical Residues in Novel Drug Metabolizing
Purification Method for Cytochrome P450 BM-3 by Anion-Exchange Mutants of Cytcochrome P450 Using Random Mutagenesis. J. Med.
Chromatography. J. Chromatogr. A 1999, 848, 149−159. (f) Huang, Chem. 2007, 50, 455−461. (c) Vottero, E.; Rea, V.; Lastdrager, J.;
W.-C.; Westlake, A. C. G.; Maréchal, J.-D.; Joyce, M. G.; Moody, P. C. Honing, M.; Vermeulen, N. P. E.; Commandeur, J. N. M. Role of
E.; Roberts, G. C. K. Filling a Hole in Cytochrome P450 BM3 Residue 87 in Substrate Selectivity and Regioselectivity of Drug-
Improves Substrate Binding and Catalytic Efficiency. J. Mol. Biol. 2007, Metabolizing Cytochrome P450 CYP102A1M11. JBIC, J. Biol. Inorg.
373, 633−651. Chem. 2011, 16, 899−912. (d) Rea, V.; Kolkman, A. J.; Vottero, E.;
(10) (a) Zhang, K.; Shafer, B. M.; Demars, M. D.; Stern, H. A.; Fasan, Stronks, E. J.; Ampt, K. A. M.; Honing, M.; Vermeulen, N. P. E.;
R. Controlled Oxidation of Remote sp3 C−H Bonds in Artemisinin Wijmenga, S. S.; Commandeur, J. N. M. Active Site Substitution
via P450 Catalysts with Fine-Tuned Regio- and Stereoselectivity. J. A82W Improves the Regioselectivity of Steroid Hydroxylation by
Am. Chem. Soc. 2012, 134, 18695−18704. (b) Kolev, J. N.; O’Dwyer, Cytochrome P450 BM3Mutants as Rationalized by Spin Relaxation
K. M.; Jordan, C. T.; Fasan, R. Discovery of Potent Parthenolide-Based Nuclear Magnetic Resonance Studies. Biochemistry 2012, 51, 750−760.
Antileukemic Agents Enabled by Late-Stage P450-Mediated CH (e) Venkataraman, H.; de Beer, S. B. A.; van Bergen, L. A. H.; van
Functionalization. ACS Chem. Biol. 2014, 9, 164−173. (c) Kolev, J. N.; Essen, N.; Geerke, D. P.; Vermeulen, N. P. E; Commandeur, J. N. M.
Zaengle, J. M.; Ravikumar, R.; Fasan, R. Enhancing the Efficiency and A Single Active Site Mutation Inverts Stereoselectivity of 16-
Regioselectivity of P450 Oxidation Catalysts by Unnatural Amino Acid Hydroxylation of Testosterone Catalyzed by Engineered Cytochrome
Mutagenesis. ChemBioChem 2014, 15, 1001−1010. P450 BM3. ChemBioChem 2012, 13, 520−523.
(11) (a) Luo, X.-D.; Shen, C.-C. The Chemistry, Pharmacology, and (19) (a) Schmitz, D.; Zapp, J.; Bernhardt, R. Steroid conversion with
Clinical Applications of Qinghaosu (Artemisinin) and its Derivatives. CYP106A2 - Production of Pharmaceutically Interesting DHEA
Med. Res. Rev. 1987, 7, 29−52. (b) Navaratnam, V.; Mansor, S. M.; Sit, Metabolites. Microb. Cell Fact. 2014, 13, 81. (b) Józw ́ ik, I. K.; Kiss,
N. W.; Grace, J.; Li, Q.; Olliaro, P. Pharmacokinetics of Artemisinin- F. M.; Gricman, L.; Abdulmughni, A.; Brill, E.; Zapp, J.; Pleiss, J.;
type Compounds. Clin. Pharmacokinet. 2000, 39, 255−270. (c) Meyer, Bernhardt, R.; Thunnissen, A.-M. W. H. Structural Basis of Steroid
Binding and Oxidation by the Cytochrome P450 CYP109E1 from
D.; Smeilus, T.; Pliatsika, D.; Mousavizadeh, F.; Giannis, A. Synthesis
Bacillus Megaterium. FEBS J. 2016, 283, 4128−4148. (c) Nikolaus, J.;
of Novel C-9 Carbon Substituted Derivatives of Artemisinin. Bioorg.
Nguyen, K. T.; Virus, C.; Riehm, J. L.; Hutter, M.; Bernhardt, R.
Med. Chem. 2017, 25, 6098−6101.
Engineering of CYP106A2 for Steroid 9α- and 6β-Hydroxylation.
(12) Zhang, K.; El Damaty, S.; Fasan, R. P450 Fingerprinting Method
Steroids 2017, 120, 41−48. (d) Larsen, A. T.; May, E. M.; Auclair, K.
for Rapid Discovery of Terpene Hydroxylating P450 Catalysts with
Predictable Stereoselective and Chemoselective Hydroxylations and
Diversified Regioselectivity. J. Am. Chem. Soc. 2011, 133, 3242−3245.
Epoxidations with P450 3A4. J. Am. Chem. Soc. 2011, 133, 7853−7858.
(13) Reviews of directed evolution with emphasis on CAST/ISM:
(e) Polic, V.; Auclair, K. Controlling Substrate Specificity and Product
(a) Reetz, M. T. Laboratory Evolution of Stereoselective Enzymes: A
Regio- and Stereo-Selectivities of P450 Enzymes Without Muta-
Prolific Source of Catalysts for Asymmetric Reactions. Angew. Chem.,
genesis. Bioorg. Med. Chem. 2014, 22, 5547−5554. (f) Hall, E. A.;
Int. Ed. 2011, 50, 138−174. (b) Acevedo-Rocha, C. G.; Hoebenreich, Sarkar, M. R.; Lee, J. H. Z.; Munday, S. D.; Bell, S. G. Improving the
S.; Reetz, M. T. Iterative Saturation Mutagenesis: A Powerful Monooxygenase Activity and the Regio- and Stereoselectivity of
Approach to Engineer Proteins by Systematically Simulating Darwin- Terpenoid Hydroxylation Using Ester Directing Groups. ACS Catal.
ian Evolution. Methods Mol. Biol. 2014, 1179, 103−128. (c) Reetz, M. 2016, 6, 6306−6317.
T.; Carballeira, J. D. Iterative Saturation Mutagenesis (ISM) for Rapid (20) (a) van der Meer, J.-Y.; Biewenga, L.; Poelarends, G. J. The
Directed Evolution of Functional Enzymes. Nat. Protoc. 2007, 2, 891− Generation and Exploitation of Protein Mutability Landscapes for
903. Protein Engineering. ChemBioChem 2016, 17, 1792−1799. (b) van der
(14) (a) Tokuriki, N.; Jackson, C. J.; Afriat-Jurnou, L.; Wyganowski, Meer, J.-Y.; Poddar, H.; Baas, B.-J.; Miao, Y.; Rahimi, M.; Kunzendorf,
K. T.; Tang, R.; Tawfik, D. S. Diminishing Returns and Tradeoffs A.; van Merkerk, R.; Tepper, P. G.; Geertsema, E. M.; Thunnissen, A.-
Constrain the Laboratory Optimization of an Enzyme. Nat. Commun. M. W. H.; Quax, W. J.; Poelarends, G. J. Using Mutability Landscapes
2012, 3, 1257. (b) Tawfik, D. S. Accuracy-rate Tradeoffs: How do of a Promiscuous Tautomerase to Guide the Engineering of
Enzymes Meet Demands of Selectivity and Aatalytic Efficiency? Curr. Enantioselective Michaelases. Nat. Commun. 2016, 7, 10911.
Opin. Chem. Biol. 2014, 21, 73−80. (c) Miton, C. M.; Tokuriki, N. (c) Hecht, M.; Bromberg, Y.; Rost, B. News from the Protein
How Mutational Epistasis Impairs Predictability in Protein Evolution Mutability Landscape. J. Mol. Biol. 2013, 425, 3937−3948.
and Design. Protein Sci. 2016, 25, 1260−1272. (d) Frauenkron-Machedjou, V. J.; Fulton, A.; Zhu, L.; Anker, C.;
(15) Ren, X.; Yorke, J. A.; Taylor, E.; Zhang, T.; Zhou, W.; Wong, L. Bocola, M.; Jaeger, K.-E.; Schwaneberg, U. Towards Understanding
L. Drug Oxidation by Cytochrome P450BM3: Metabolite Synthesis Directed Evolution: More than Half of All Amino Acid Positions
and Discovering New P450 Reaction Types. Chem. - Eur. J. 2015, 21, Contribute to Ionic Liquid Resistance of Bacillus Subtilis Lipase A.
15039−15047. ChemBioChem 2015, 16, 937−945. (e) Mosquna, A.; Peterson, F. C.;
(16) (a) Kille, S.; Zilly, F. E.; Acevedo, J. P.; Reetz, M. T. Regio- and Park, S.-Y.; Lozano-Juste, J.; Volkman, B. F.; Cutler, S. R. Potent and
Stereoselectivity of P450-Catalysed Hydroxylation of Steroids Selective Activation of Abscisic Acid Receptors in vivo by Mutational
Controlled by Laboratory Evolution. Nat. Chem. 2011, 3, 738−743. Stabilization of their Agonist-Bound Conformation. Proc. Natl. Acad.
(b) Hoebenreich, S.; Zilly, F. E.; Acevedo-Rocha, C. G.; Zilly, M.; Sci. U. S. A. 2011, 108, 20838−20843. (f) Park, S.-Y.; Peterson, F. C.;
Reetz, M. T. Speeding up Directed Evolution: Combining the Mosquna, A.; Yao, J.; Volkman, B. F.; Cutler, S. R. Agrochemical
Advantages of Solid-Phase Combinatorial Gene Synthesis with Control of Plant Water use Using Engineered Abscisic Acid Receptors.
Statistically Guided Reduction of Screening Effort. ACS Synth. Biol. Nature 2015, 520, 545−548. (g) Acevedo-Rocha, C. G.; Ferla, M.;
2015, 4, 317−331. Reetz, M. T. Directed Evolution of Proteins based on Mutational
(17) Bureik, M.; Bernhardt, R. In Steroid Hydroxylation: Microbial Scanning. Methods Mol. Biol. 2018, 1685, 87−128.
Steroid Biotransformations Using Cytochrome P450 Enzymes. In Modern (21) (a) Sun, Z.; Lonsdale, R.; Ilie, A.; Li, G.; Zhou, J.; Reetz, M. T.
Biooxidation; Schmid, R. D., Urlacher, V. B., Eds.; Wiley-VCH: Catalytic Asymmetric Reduction of Difficult-to-Reduce Ketones:
Weinheim, Germany, 2007; pp 155−176. Triple-Code Saturation Mutagenesis of An Alcohol Dehydrogenase.

3408 DOI: 10.1021/acscatal.8b00389

ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

ACS Catal. 2016, 6, 1598−1605. (b) Sun, Z.; Wikmark, Y.; Bäckvall, J.- Screening for Mutational Hotspots Using Biased Molecular Dynamics
E.; Reetz, M. T. New Concepts for Increasing the Efficiency in Simulations. ACS Catal. 2017, 7, 6786−6797.
Directed Evolution of Stereoselective Enzymes. Chem. - Eur. J. 2016, (35) Sevrioukova, I. F.; Li, H.; Zhang, H.; Peterson, J. A.; Poulos, T.
22, 5046−5054. L. Structure of a Cytochrome P450−Redox Partner Electron-Transfer
(22) (a) Enzyme Catalysis in Organic Synthesis, 3rd ed.; Drauz, K., Complex. Proc. Natl. Acad. Sci. U. S. A. 1999, 96, 1863−1868.
Gröger, H., May, O., Eds.; Wiley-VCH: Weinheim, 2012. (b) Faber, K. (36) Lonsdale, R.; Harvey, J. N.; Mulholland, A. J. Compound I
Biotransformations in Organic Chemistry, 6th ed.; Springer: Heidelberg, Reactivity Defines Alkene Oxidation Selectivity in Cytochrome
2011. (c) Industrial Biotransformations; Liese, A., Seelbach, K., P450cam. J. Phys. Chem. B 2010, 114, 1156−1162.
Wandrey, C., Eds.; Wiley-VCH: Weinheim, 2006. (37) Lonsdale, R.; Houghton, K. T.; Ż urek, J.; Bathelt, C. M.;
(23) Currin, A.; Swainston, N.; Day, P. J.; Kell, D. B. Synthetic Foloppe, N.; de Groot, M. J.; Harvey, J. N.; Mulholland, A. J. Quantum
Biology for the Directed Evolution of Protein Biocatalysts: Navigating Mechanics/Molecular Mechanics Modeling of Regioselectivity of Drug
Sequence Space Intelligently. Chem. Soc. Rev. 2015, 44, 1172−1239. Metabolism in Cytochrome P450 2C9. J. Am. Chem. Soc. 2013, 135,
(24) Lutz, S. Beyond Directed EvolutionSemi-Rational Protein 8001−8015.
Engineering and Design. Curr. Opin. Biotechnol. 2010, 21, 734−743. (38) Atkins, W. M. Non-Michaelis-Menten Kinetics in Cytochrome
(25) (a) Parikh, M. R.; Matsumura, I. Site-Saturation Mutagenesis is P450-Catalyzed Reactions. Annu. Rev. Pharmacol. Toxicol. 2005, 45,
More Efficient than DNA Shuffling for the Directed Evolution of β- 291−310.
Fucosidase from β-Galactosidase. J. Mol. Biol. 2005, 352, 621−628. (39) Denisov, I. G.; Sligar, S. G. A Novel Type of Allosteric
(b) Paramesvaran, J.; Hibbert, E. G.; Russell, A. J.; Dalby, P. A. Regulation: Functional Cooperativity in Monomeric Proteins. Arch.
Distributions of Enzyme Residues Yielding Mutants with Improved Biochem. Biophys. 2012, 519, 91−102.
Substrate Specificities from Two Different Directed Evolution (40) Butler, C. F.; Peet, C.; Mason, A. E.; Voice, M. W.; Leys, D.;
Strategies. Protein Eng., Des. Sel. 2009, 22, 401−411. (c) Reetz, M. Munro, A. W. Key Mutations Alter the Cytochrome P450 BM3
T.; Prasad, S.; Carballeira, J. D.; Gumulya, Y.; Bocola, M. Iterative Conformational Landscape and Remove Inherent Substrate Bias. J.
Saturation Mutagenesis Accelerates Laboratory Evolution of Enzyme Biol. Chem. 2013, 288, 25387−25399.
Stereoselectivity: Rigorous Comparison with Traditional Methods. J. (41) Shoji, O.; Watanabe, Y. Bringing out the Potential of Wild-type
Am. Chem. Soc. 2010, 132, 9144−9152. Cytochrome P450s Using Decoy Molecules: Oxygenation of Non-
(26) Acevedo-Rocha, C. G.; Agudo, R.; Reetz, M. T. Directed native Substrates by Bacterial Cytochrome P450s. Isr. J. Chem. 2015,
evolution of Stereoselective Enzymes Based on Genetic Selection as 55, 32−39.
Opposed to Screening Systems. J. Biotechnol. 2014, 191, 3−10. (42) Noble, M. A.; Miles, C. S.; Chapman, S. K.; Lysek, D. A.;
(27) Geronimo, I.; Denning, C. A.; Rogers, W. E.; Othman, T.; MacKay, A. C.; Reid, G. A.; Hanzlik, R. P.; Munro, A. W. Roles of Key
Active-Site Residues in Flavocytochrome P450 BM3. Biochem. J. 1999,
Huxford, T.; Heidary, D. K.; Glazer, E. C.; Payne, C. M. Effect of
339, 371−379.
Mutation and Substrate Binding on the Stability of Cytochrome
(43) Zhao, B.; Lei, L.; Kagawa, N.; Sundaramoorthy, M.; Banerjee, S.;
P450BM3 Variants. Biochemistry 2016, 55, 3594−3606.
Nagy, L. D.; Guengerich, F. P.; Waterman, M. R. Three-Dimensional
(28) Reetz, M. T. The Importance of Additive and Non-Additive
Structure of Steroid 21-Hydroxylase (Cytochrome P450 21A2) with
Mutational Effects in Protein Engineering. Angew. Chem., Int. Ed. 2013,
Two Substrates Reveals Locations of Disease-Associated Variants. J.
52, 2658−2666.
Biol. Chem. 2012, 287, 10613−10622.
(29) (a) Firth, A. E.; Patrick, W. M. GLUE-IT and PEDEL-AA: New
(44) Agematu, H.; Matsumoto, N.; Fujii, Y.; Kabumoto, H.; Doi, S.;
Programmes for Analyzing Protein Diversity in Randomized Libraries.
Machida, K.; Ishikawa, J.; Arisawa, A. Hydroxylation of Testosterone
Nucleic Acids Res. 2008, 36, W281−W285. (b) Patrick, W. M.; Firth, A. by Bacterial Cytochromes P450 Using the Escherichia coli Expression
E. Strategies and Computational Tools for Improving Randomized System. Biosci., Biotechnol., Biochem. 2006, 70, 307−311.
Protein Libraries. Biomol. Eng. 2005, 22, 105−112. (c) Denault, M.; (45) (a) Bracco, P.; Janssen, D. B.; Schallmey, A. Selective Steroid
Pelletier, J. N. Protein Library Design and Screening: Working Out the Oxyfunctionalisation by CYP154C5, a Bacterial Cytochrome P450.
Probabilities. In Protein Engineering Protocols; Humana Press: Totowa, Microb. Cell Fact. 2013, 12, 95. (b) Makino, T.; Katsuyama, Y.;
NJ, 2007; pp 127−154. Otomatsu, T.; Misawa, N.; Ohnishi, Y. Regio- and Stereospecific
(30) Nov, Y. When Second Best is Good Enough: Another Hydroxylation of Various Steroids at the 16α Position of the D Ring
Probabilistic Look at Saturation Mutagenesis. Appl. Environ. Microbiol. by the Streptomyces Griseus Cytochrome P450 CYP154C3. Appl.
2012, 78, 258−262. Environ. Microbiol. 2014, 80, 1371−1379.
(31) Kille, S. Flavoproteins in Directed Evolution: Iterative CASTing to (46) Directed evolution studies using ISM followed by epPCR or
Evolve YqjM and P450BM3, Ph.D. Dissertation, Ruhr-Universitaet DNA shuffling: (a) Kwan, D. H.; Constantinescu, I.; Chapanian, R.;
Bochum, 2010. Higgins, M. A.; Kötzler, M. P.; Samain, E.; Boraston, A. B.;
(32) (a) Acevedo-Rocha, C. G.; Reetz, M. T.; Nov, Y. Economical Kizhakkedathu, J. N.; Withers, S. G. Toward Efficient Enzymes for
Analysis of Saturation Mutagenesis Experiments. Sci. Rep. 2015, 5, the Generation of Universal Blood through Structure-Guided Directed
10654. (b) Sullivan, B.; Walton, A. Z.; Stewart, J. D. Library Evolution. J. Am. Chem. Soc. 2015, 137, 5695−5705. (b) Garrabou Pi,
Construction and Evaluation for Site Saturation Mutagenesis. Enzyme X.; Macdonald, D. S.; Wicky, B. I. M.; Hilvert, D. Stereodivergent
Microb. Technol. 2013, 53, 70−77. Evolution of Artificial Enzymes for the Michael Reaction. Angew.
(33) Bougioukou, D. J.; Kille, S.; Taglieber, A.; Reetz, M. T. Directed Chem., Int. Ed. 2018, DOI: 10.1002/anie.201712554.
Evolution of an Enantioselective Enoate-Reductase: Testing the Utility (47) (a) Damborsky, J.; Brezovsky, J. Computational Tools for
of Iterative Saturation Mutagenesis. Adv. Synth. Catal. 2009, 351, Designing and Engineering Enzymes. Curr. Opin. Chem. Biol. 2014, 19,
3287−3305. 8−16. (b) Ebert, M. C.; Pelletier, J. N. Computational Tools for
(34) (a) Dodani, S. C.; Kiss, G.; Cahn, J. K. B.; Su, Y.; Pande, V. S.; Enzyme Improvement: Why Everyone Can - and Should - Use them.
Arnold, F. H. Discovery of a Regioselectivity Switch in Nitrating P450s Curr. Opin. Chem. Biol. 2017, 37, 89−96. (c) Romero-Rivera, A.;
Guided by Molecular Dynamics Simulations and Markov Models. Nat. Garcia-Borràs, M.; Osuna, S. Computational Tools for the Evaluation
Chem. 2016, 8, 419−425. (b) Narayan, A. R. H.; Jiménez-Osés, G.; of Laboratory-Engineered Biocatalysts. Chem. Commun. 2017, 53,
Liu, P.; Negretti, S.; Zhao, W.; Gilbert, M. M.; Ramabhadran, R. O.; 284−297. (d) Sáez-Jiménez, V.; Acebes, S.; Guallar, V.; Martínez, A.
Yang, Y.-F.; Furan, L. R.; Li, Z.; Podust, L. M.; Montgomery, J.; Houk, T.; Ruiz-Dueñas, F. J. Improving the Oxidative Stability of a High
K. N.; Sherman, D. H. Enzymatic Hydroxylation of An Unactivated Redox Potential Fungal Peroxidase by Rational Design. PLoS One
Methylene C−H bond Guided by Molecular Dynamics Simulations. 2015, 10, e0124750. (e) Chen, M. M. Y.; Snow, C. D.; Vizcarra, C. L.;
Nat. Chem. 2015, 7, 653−660. (c) Ebert, M. C. C. J. C.; Guzman Mayo, S. L.; Arnold, F. H. Comparison of Random Mutagenesis and
Espinola, J.; Lamoureux, G.; Pelletier, J. N. Substrate-Specific Semi-Rational Designed Libraries for Improved Cytochrome P450

3409 DOI: 10.1021/acscatal.8b00389

ACS Catal. 2018, 8, 3395−3410
ACS Catalysis Research Article

BM3-catalyzed Hydroxylation of Small Alkanes. Protein Eng., Des. Sel.

2012, 25, 171−178.

3410 DOI: 10.1021/acscatal.8b00389

ACS Catal. 2018, 8, 3395−3410