Active site plasticity of a computationally designed retro-aldolase enzyme

Article abstract of DOI:10.1002/cctc.201300933

RA110 is a computationally designed retro-aldolase enzyme that utilizes amine catalysis to convert 4-hydroxy-4-(6-methoxy-2-naphthyl)-2-butanone to 6-methoxy-2-naphthaldehyde and acetone. The original design accelerated substrate cleavage by a factor of 12 000 over background, and its activity was subsequently increased more than a thousand-fold by directed evolution. The X-ray structure of the evolved catalyst covalently modified with a 1,3-diketone inhibitor deviates substantially from the design model, however, with the ligand adopting a completely different orientation than predicted. Moreover, significant activity was maintained even after relocation of the reactive lysine within the apolar binding pocket. These results suggest that the success of the original design is not ascribable to atomically accurate molecular recognition, but rather to successful placement of a reactive lysine adjacent to an apolar binding pocket. Nevertheless, the stabilizing interactions observed at the active site of the evolved variant suggest that improvements in the precision of design calculations will afford enzymes with higher catalytic activities. What's that in your pocket? A computationally designed and experimentally optimized retro-aldolase enzyme utilizes amine catalysis for substrate cleavage. However, substantial differences between the original design model and experimental structure highlight the need for improved computational protocols. Generating catalysts with true enzyme-like activities will require more than simply placing a reactive lysine adjacent to a hydrophobic pocket.

Full text of DOI:10.1002/cctc.201300933

CHEMCATCHEM
FULL PAPERS
DOI: 10.1002/cctc.201300933
Active Site Plasticity of a Computationally Designed Retro-
Aldolase Enzyme
[
a]
[a]
[a]
[a]
[b]
Richard Obexer, Sabine Studer, Lars Giger, Daniel M. Pinkas, Markus G. Grꢀtter,
[c]
[a]
David Baker, and Donald Hilvert*
RA110 is a computationally designed retro-aldolase enzyme
that utilizes amine catalysis to convert 4-hydroxy-4-(6-me-
thoxy-2-naphthyl)-2-butanone to 6-methoxy-2-naphthaldehyde
and acetone. The original design accelerated substrate cleav-
age by a factor of 12000 over background, and its activity was
subsequently increased more than a thousand-fold by directed
evolution. The X-ray structure of the evolved catalyst covalent-
ly modified with a 1,3-diketone inhibitor deviates substantially
from the design model, however, with the ligand adopting
a completely different orientation than predicted. Moreover,
significant activity was maintained even after relocation of the
reactive lysine within the apolar binding pocket. These results
suggest that the success of the original design is not ascriba-
ble to atomically accurate molecular recognition, but rather to
successful placement of a reactive lysine adjacent to an apolar
binding pocket. Nevertheless, the stabilizing interactions ob-
served at the active site of the evolved variant suggest that im-
provements in the precision of design calculations will afford
enzymes with higher catalytic activities.
[8]
Aldolases are broadly useful catalysts for the stereoselective
formation and cleavage of carbon–carbon bonds. Class I en-
ficial aldolases known, additional hydrogen bond networks
likely facilitate key proton transfer events in the catalytic
[9]
zymes, which are members of the large (ba) -barrel superfami-
cycle.
8
ly, share a common mechanism involving amine catalysis by
Recently introduced computational methods have also been
[
1]
[10]
a Schiff base-forming lysine. Because natural aldol substrates
are usually quite polar, extensive hydrogen bonding interac-
tions and salt bridges are used to preorganize them for reac-
tion at the active site. These hydrogen bonding networks are
also crucial for mediating mechanistically important proton
transfers along the multistep reaction coordinate.
used to design artificial aldolases. A motif that combines a re-
active lysine, a substrate-binding site, and water-mediated sta-
bilization of the b-hydroxyl group was successfully modeled
[11]
into 13 protein scaffolds, affording functional catalysts for
the retro-aldol cleavage of methodol (4-hydroxy-4-(6-methoxy-
2-naphthyl)-2-butanone, 1) to give 6-methoxy-2-naphthalde-
hyde (3) and acetone (Scheme 1). Although the initial activities
Efforts to create artificial aldolases have focused on installing
reactive amino groups into different scaffolds. A wide range of
were not high, they could be significantly increased by labora-
[
2]
[3]
[11–13]
amino acid derivatives, peptides and other sequence-specif-
tory evolution.
The best catalysts achieve efficiencies ap-
[
4,5]
[6]
ic oligomers,
engineered proteins, and catalytic antibod-
proaching those of natural enzymes. Although evolutionary
optimization of the originally programmed mechanism is the
norm, the unexpected does sometimes occur, as evidenced by
extensive active site remodeling, such as a catalytic residue
[
7,8]
ies
have been shown to provide significant rate accelera-
tions and stereochemical control of aldol reactions. They typi-
cally exploit either Coulombic interactions or a hydrophobic
[
12]
environment to lower the pK value of the catalytic amine. In
switch, observed in the case of the RA95 retro-aldolase.
a
[1]
the case of catalytic antibodies, which are among the best arti-
Like natural aldolases and many other computationally de-
[10,14,15]
signed enzymes,
RA95 belongs to the (ba) -barrel super-
8
[11]
+
+
+
+
family. The small and easily produced RA110 retro-aldolase
[
a] R. Obexer, S. Studer, Dr. L. Giger, Dr. D. M. Pinkas, Prof. D. Hilvert
Laboratory of Organic Chemistry
ETH Zurich
(Figure 1) represents an interesting alternative starting point
for exploring fold evolvability. It is based on a ketosteroid iso-
merase from the NTF2-like superfamily, whose cone-shaped
fold consists of a curled b sheet packed on one face with three
a helices. A deep hydrophobic binding pocket, originally con-
figured for ketosteroid recognition, is well suited for binding
large hydrophobic substrates such as methodol. RA110 was
among the best computationally designed retro-aldolases, ac-
celerating substrate cleavage by a factor of 12000 over back-
ground, and this starting activity was subsequently improved
8
093 Zurich (Switzerland)
E-mail: hilvert@org.chem.ethz.ch
[
b] Prof. M. G. Grꢀtter
Department of Biochemistry
University of Zurich
8
057 Zurich (Switzerland)
[
c] Prof. D. Baker
Department of Biochemistry
University of Washington
Seattle, WA 98195 (USA)
[11]
+
more than a thousand-fold by directed evolution. In the cur-
rent study, we determined the structure of the evolved
[
] These authors contributed equally to this work.
Supporting information for this article is available on the WWW under
http://dx.doi.org/10.1002/cctc.201300933.
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1043

CHEMCATCHEM
FULL PAPERS
www.chemcatchem.org
which replaced the active site
residue Thr118 with an arginine.
RA110.4 was further evolved
over six rounds of random muta-
genesis and screening until no
further improvement was ach-
[11]
ieved.
Eight additional muta-
tions were identified (Figure S1).
Except V38I, all are located on
the protein surface. The best cat-
alyst, RA110.4-6, exhibits a 1000-
fold higher k /K than the origi-
cat
m
nal design.
Scheme 1. Catalysis by retro-aldolase RA110. a) The amino group of the catalytic lysine accelerates the retro-aldol
cleavage of 1 to 3 and acetone via the formation of the iminium intermediate 2. b) The 1,3-diketone inhibitor 4 is
a mechanism-based inhibitor that reacts with the catalytic lysine to form a Schiff base adduct (5). Deprotonation
of the intermediate affords a stable vinylogous amide (6).
Biochemical characterization
of RA110.4 and RA110.4-6 re-
vealed that both proteins form
disulfide-linked oligomers under
aerobic conditions. As shown in
enzyme and explored how the location of the reactive lysine
modulates catalytic activity.
Figure 2, RA110.4 produces a range of multimers. During di-
rected evolution, however, the distribution of oligomers shifted
toward the dimeric state. This change can be attributed pri-
marily to the C69Y mutation, which eliminated the most reac-
[16]
tive surface cysteine residue.
Although protein yield de-
Results
creased by approximately 50% over the course of optimiza-
tion, protein stability increased, judging by the melting tem-
RA110 was previously improved by a two-step evolutionary
[
11]
method. Initial optimization of the active site by targeted
cassette mutagenesis yielded RA110.4, a catalyst with a 10-fold
higher kcat value and an 80-fold higher catalytic efficiency (kcat/
Km). This variant contains four mutations (Figure S1), one of
peratures (T ) of 70 and 808C for RA110.4 and RA110.4-6, re-
m
spectively (Table 1).
To avoid kinetic ambiguities associated with variable oligo-
meric states, we characterized the evolved enzymes after re-
duction with dithiothreitol (DTT; Figure 3 and Table 1). The
overall catalytic efficiency of RA110.4, reflected by the apparent
second-order rate constant k /K , remained constant; howev-
cat
m
er, the kcat and Km parameters both increased by a factor of 3
after treatment. In the case of RA110.4-6, Km also increased 3-
fold under reducing conditions but kcat increased only 1.3-fold,
Figure 1. Computational design of the retro-aldolase RA110. a) Idealized
retro-aldolase active site. b) Model of RA110 designed with the Rosetta soft-
Figure 2. Sodium dodecyl sulfate polyacrylamide gel electrophoresis of
RA110.4 and RA110.4-6 under reducing and nonreducing conditions.
RA110.4 has a higher tendency to oligomerize than the evolved variant
RA110.4-6. LMW=Low molecular weight markers.
[
11]
ware suite. The catalytically relevant residues (yellow carbon atoms) and
the composite transition states (purple carbon atoms) are indicated as
sticks.
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1044

CHEMCATCHEM
FULL PAPERS
www.chemcatchem.org
The mechanism-based 1,3-diketone inhibitor 4 reacts stoi-
chiometrically with reduced RA110.4-6 to give vinylogous
amide 6 (Scheme 1). The structure of this complex was deter-
mined by X-ray crystallography at a resolution of 2.5 ꢂ. As ex-
pected, the protein adopts the canonical NTF2-like fold, al-
though a slight displacement of the b strands harboring most
of the mutations relative to the original design model was ob-
served (Figure 4a). The root-mean-square deviation for all Ca
atoms is 2.24 ꢂ.
[
a]
Table 1. Characterization of RA110 variants.
[
b]
Variant
k
cat
K
[mm]
M
k
cat/K
M
pK
a
T
m
Protein yield
ꢁ
1
ꢁ1
ꢁ1
ꢁ1
[
min
]
[m min
]
[8C] [mgL
]
[
c]
RA110
0.005
0.07
0.23
0.23
0.29
1600
278
780
69 3300
210 1400
3.1
250
290
[
[
c]
RA110.4
RA110.4
RA110.4-6
RA110.4-6
d]
6.3
6.5
70
80
130
60
[
[
c]
d]
[
a] All measurements were performed at 298C in 25 mm HEPES (pH 7.5)
The inhibitor is covalently linked to Lys88 as envisioned by
design. However, the orientation adopted by this ligand devi-
ates significantly from the Rosetta model. Instead of occupying
the designed hydrophobic pocket, the naphthyl group binds
snugly in a surface crevice in an orientation that is rotated ap-
proximately 1808 from the intended binding mode (Figure 4b
and c). Although this site is not present in the original design,
only minor adjustments in the backbone and loop conforma-
tions are needed to form it. A hydrogen bonding interaction
between the b-carbonyl group of the inhibitor and the argi-
nine at position 118 that was introduced during active site op-
timization favors the new binding mode (Figure 4d). A second
hydrogen bond with a water molecule that also hydrogen
bonds to the side chain of Trp120 may also be important
during catalysis (Figure 4d). In contrast to the most evolved
containing 100 mm NaCl and 2.7% acetonitrile using the UV/Vis spectro-
scopic assay. [b] The sequences of all variants are shown in Figure S1.
[c] From Ref. [11]. [d] Values determined for reduced protein by using the
UV/Vis spectroscopic assay.
[12]
RA95 variants and catalytic antibodies elicited by 1,3-dike-
[8]
tones or b-ketosulfones, RA110 lacks residues that might facil-
itate formation and breakdown of the carbinolamine inter-
mediate that arises upon attack of the catalytic lysine on the
substrate carbonyl group. This difference may explain why
RA110.4-6 is more than an order of magnitude less active than
the best artificial retro-aldolases.
To explore how the position of the catalytic lysine in
RA110.4-6 influences enzyme efficiency, we shifted this residue
deeper into the steroid-binding pocket (Figure 5). To that end,
two suitable residues Val101 and Leu103, whose side chains
point into the pocket, were individually replaced by lysine, and
the original lysine at position 88 was randomized. The most
active variants isolated from these libraries, called RA110.V and
RA110.L, contain the K88C mutation. Relocating the Schiff
base-forming lysine in this way decreased the activity of the
evolved retro-aldolase 8 to 15-fold (Table 2); however, RA110.V
[
a]
Table 2. Kinetic parameters of RA110.V and RA110.L variants.
[
b]
[b]
Variant
k
cat
K
M
k
cat/K
M
Figure 3. Michaelis–Menten kinetics and pH profiles of RA110.4 (*) and
RA110.4-6 (~), with corresponding fits, determined with the UV/Vis spectro-
scopic assay and (ꢀ)-methodol.
ꢁ1
ꢁ1
ꢁ1
[
min
]
[mm]
[m min
]
[
c]
RA110.4-K88C/V101K
RA110.V
RA110.V-4
RA110.4-K88C/L103K
RA110.L
RA110.L-1
0.009ꢀ0.0005
0.08ꢀ0.004
0.18ꢀ0.005
0.03ꢀ0.006
0.08ꢀ0.003
0.23
240ꢀ50
37
180
750
20
90
250
450ꢀ60
240ꢀ20
1500ꢀ500
908ꢀ50
908
[c]
hence the overall catalytic efficiency of the reduced enzyme
decreased by a factor of 2. Thus, the evolved variant per-
formed somewhat better under the aerobic conditions that
were used in the screening procedure. Both catalysts exhibited
sigmoidal pH–rate profiles with apparent ionizations of 6.3 and
[
a] Measurements were performed at 298C in 25 mm HEPES (pH 7.5) con-
taining 100 mm NaCl and 2.7% acetonitrile and reduced protein using
the UV/Vis spectroscopic assay unless otherwise indicated; [b] The se-
quences of all variants are shown in Figure S1; [c] Determined by using
the fluorescence assay.
6.5, consistent with the participation of a catalytic lysine with
a significantly perturbed pK value (Figure 3b).
a
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1045

CHEMCATCHEM
FULL PAPERS
www.chemcatchem.org
RA110.L, which has the most
buried lysine, is the least active,
whereas RA110.4-6, which has
the most accessible lysine, is the
most active. Reversion of muta-
tions identified during the evolu-
tionary optimization of RA110.4
further reduces the efficiency of
RA110.V and RA110.L by an addi-
tional factor of approximately 5.
Because this effect roughly cor-
responds to the advantage ob-
served for RA110.4-6 over
RA110.4, these substitutions
appear to provide
a generic
rather than a specific catalytic
benefit, consistent with their rel-
atively exposed locations on the
protein surface.
Both RA110.V and RA110.L
were subjected to multiple
rounds of random mutagenesis
and screening. Apart from 2 to
3-fold activity boosts in the first
round, only minor improvements
were observed for RA110.V and
no further improvement was
achieved for RA110.L (Figure 6).
In neither case was the full activ-
ity of RA110.4-6 recovered. For
RA110.V, four mutations accumu-
lated over four rounds of evolu-
Figure 4. Structure of the evolved retro-aldolase RA110.4-6. a) Superimposition of the RA110 design model
(yellow) and the crystal structure of RA110.4-6 (cyan). b) Surface representation of RA110.4-6 with snugly bound
inhibitor (orange). c) The inhibitor (orange) adopts a completely different orientation than the carbinolamine inter-
[
11]
mediate (purple) in the Rosetta model of the RA110.4 variant. d) The inhibitor is covalently bound to Lys88 and
also forms hydrogen bonds with an ordered water molecule and the side chain of Arg118. The water molecule
also hydrogen bonds to the indole ring of Trp120, although the latter residue can adopt a second conformation
in which the indole ring points out of the binding pocket and participates in crystal contacts with a symmetry-
related molecule.
tion; they were all located in surface loops and, together, yield-
ed the 4-fold improved catalyst RA110.V-4. The 3-fold improve-
ment in RA110.L can be ascribed to a single F16L mutation
close to the catalytic lysine. Two additional focused libraries
targeting His56 and Trp60, which shield the entrance of the
steroid-binding pocket, and Met105 and Arg118, which are in
proximity to the catalytic lysine at position 103, failed to pro-
vide any additional improvement.
Discussion
Structural characterization of RA110.4-6 complexed with a 1,3-
diketone inhibitor has provided mechanistic insight into the
origins of catalysis in this computationally designed and exper-
imentally optimized retro-aldolase. As per design, the enzyme
exploits amine catalysis by a reactive lysine at position 88,
proximal to a hydrophobic substrate-binding site. Nevertheless,
shifts in the b sheet and some of the loops, as well as a com-
pletely unanticipated binding mode for the ligand, evidence
significant discrepancies relative to the original design model.
The naphthyl group of the inhibitor does not occupy the
steroid-binding pocket as intended, but docks instead at an
adjacent hydrophobic crevice on the surface of the protein.
Structural differences between the inhibitor and the composite
transition state used for design may account, at least in part,
Figure 5. Cross section of RA110.4-6 with the bound mechanism-based in-
hibitor. The cyan surface indicates the hydrophobic cavity present within the
protein; two helices in the foreground were removed for clarity. The original-
ly designed catalytic lysine (Lys88; shown in cyan) is covalently modified
with the inhibitor molecule. The positions of the V101K (red) and L103K
(yellow) mutations in the former steroid-binding pocket are indicated.
and RA110.L are still 60 and 30 times more active, respectively,
than the original RA110 computational design.
Simply placing a lysine residue in or near a hydrophobic
pocket thus affords retro-aldolase activities comparable to
those achieved with in silico design. Efficiency correlates inver-
sely with the accessibility of the catalytic residue. Accordingly,
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1046

CHEMCATCHEM
FULL PAPERS
www.chemcatchem.org
resulting retro-aldolases (RA110.V and RA110.L) are at least as
good as the original RA110 design, suggesting that the advant-
age provided by computation was relatively modest. This find-
ing echoes the results of a recent report showing that active
retro-aldolases could be isolated from random libraries that are
[19]
compositionally similar to active computational designs.
Directed evolution of the starting RA110 design was ulti-
mately essential for achieving high activity. The largest increase
(
ꢂ100-fold) in activity was obtained by cassette mutagenesis
of the designed active site. Subsequent random mutagenesis
of the entire protein by a combination of error-prone poly-
merase chain reaction (PCR) mutagenesis and DNA shuffling af-
forded an additional 12-fold (5-fold for the reduced protein) in-
crease. In the evolved variant, interactions between the b-hy-
droxyl group of the substrate and Arg118, which was intro-
duced during active site optimization, and a water molecule
held in place by Trp120 are likely responsible for the increased
efficiency over the original design. Interestingly, these features
recapitulate the catalytic motif originally targeted by the
design model, suggesting that enzymes with higher catalytic
activities will be accessible once the accuracy of the design
cycle has been improved.
The results with RA110 also indicate that the choice of a scaf-
fold could ultimately limit catalyst evolvability. The improve-
ments achieved per round of laboratory evolution leveled off
faster than for retro-aldolase designs based on (ba) -barrel pro-
8
teins. This finding may reflect the small size of the NFT2-like
scaffold, which is dominated by regular secondary structural el-
ements. In contrast, the active sites of (ba) -barrel proteins and
8
immunoglobulins are constructed from flexible loops that can
be molded around the substrate and enable convergent as-
sembly of multiple functional groups. These features may
make such proteins especially evolvable, accounting for the
higher activities of the corresponding artificial retro-aldolases.
Figure 6. Michaelis–Menten kinetics of a) RA110.V and b) RA110.L variants.
*
: RA110.4-K88C/V101K and RA110.4-K88C/L103K based on RA110.4, ~: evo-
lutionary starting points RA110.V and RA110.L, : best-evolved variants
&
RA110.V-4 and RA110.L-1, each with a corresponding fit. Michaelis–Menten
kinetics were determined with the UV/Vis spectroscopic assay, except for
RA110.4.K88C/V101K and RA110.4.K88C/L103K, for which the fluorescence
assay was used.
The large number of natural catalytic activities supported by
[20]
(
ba) -barrel proteins is consistent with this view.
8
for this altered binding mode; however, the divergence be-
tween the model and the experiment is much larger than that
observed for other computationally designed retro-aldolas-
Conclusions
[
12,13]
es.
The mutations introduced during design and subse-
The RA110.4-6 structure underscores the importance of using
mechanistically relevant ligands for the structural characteriza-
tion of computationally designed proteins. In this instance, dif-
ferences between the design model and the experimental
quent optimization are more likely responsible for the back-
bone and loop changes that enabled the emergence of the
new surface groove into which the ligand docks. Such structur-
al perturbations could not be predicted by the original calcula-
[21–23]
structure reveal the limitations of current design efforts.
[
17]
tions with the Rosetta software suite, which assumed a rigid
protein backbone. Subsequent modeling of the RA110.4 var-
iant, which contains all the active site mutations, similarly
Reducing the scaffold to anchor points for installing catalytic
motifs greatly oversimplifies protein complexity. Achieving
atomically accurate molecular recognition will require taking
the dynamic nature of these molecules into account. That the
RA110 retro-aldolase was evolvable but less so than designs
[11]
failed to identify the new pocket or altered binding mode.
Catalysis in this system thus reflects the successful installa-
[
18]
tion of a reactive lysine in the NTF2-like scaffold rather than
atomically accurate ligand recognition. It is notable in this con-
text that the catalytic lysine could be successfully relocated
within the steroid-binding pocket without recourse to calcula-
tion. Although less active than the most evolved variant, the
based on (ba) -barrel scaffolds is also noteworthy. A better un-
8
derstanding of how a scaffold structure and the initial design
constrain the chemistry that can be performed at a particular
active site should increase the prospects of generating cata-
lysts with true enzyme activities.
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1047

CHEMCATCHEM
FULL PAPERS
www.chemcatchem.org
XhoI (New England Biolabs, Ipswich, MA, USA). After dephosphory-
lation of the vector with Antarctic phosphatase (New England Bio-
labs), the fragments were gel purified and ligated with T4 DNA
ligase (New England Biolabs). The ligation mixture was purified
with the DNA Clean & Concentrator-25 kit (Zymo Research, Irvine,
CA, USA) and transformed into electrocompetent E. coli BL21-gold-
Experimental Section
Active site variants RA110.V and RA110.L
To create the active site variants RA110.V and RA110.L, the RA110.4-
6
gene was modified and recloned into a commercially available
pET-29b(+) vector (Novagene). Cloning primers (restriction sites
are underlined) were obtained from Microsynth AG (Balgach,
Switzerland):
(
(
DE3) cells. The transformation mixture was plated on Luria Broth
LB) agar containing kanamycin sulfate (30 mgmL ). The resulting
ꢁ1
libraries were analyzed by sequencing plasmids isolated from
single colonies.
forward primer: 5’-TGA TAA CAT ATG AAC CTG CCG ACC GC,
reverse primer: 5’-TTA T CA CTC GAG ACT GCC CTG CGG T.
The codon for the original catalytic lysine at position 88 was
This method was repeated with the best first-round RA110.L var-
iant. For RA110.V, plasmids encoding the best variants from each
[26]
round (up to 12) were pooled and further diversified by epPCR
[24]
randomized by overlap extension PCR:
[27]
and DNA shuffling.
The epPCR product was digested with
ꢁ
1
forward primer: 5’-GGC ACG ATG CTG TTG CGT NNS GAG TGG GTT
TGG AAT GGT CAG C,
bovine pancreatic DNase I (90 mgmL , Roche, Basel, Switzerland)
to give 50–100 bp fragments. The purified fragments were reas-
sembled by performing the PCR with the cloning primers. The full-
length DNA library was digested, ligated into pET-29b(+), and used
to transform BL21-gold(DE3) cells as described above.
reverse primer: 5’-ACG CAA CAG CAT CGT GCC
in which N represents an equimolar mixture of all four nucleotides
and S an equimolar mixture of G and C. The resulting full-length
PCR fragments served as templates for introducing V101K and
L103K mutations through site-directed mutagenesis:
Focused libraries
The best first-round RA110.L-1 variant was used to create focused
libraries by cassette mutagenesis, targeting two pairs of residues
V101K forward primer: 5’-CAG CCT TGC GCA ACG GAT AAA ATT
CTG GTT ATG CGC TTC GAT GA,
(His56 and Trp60 at the entrance of the binding pocket and
V101K reverse primer: 5’-ATC CGT TGC GCA AGG CTG;
Met105 and Arg118 in proximity to Lys103). The DNA libraries were
constructed by one or two overlap extension PCRs:
L103K forward primer: 5’-CCT TGC GCA ACG GAT GTT ATT AAA
GTT ATG CGC TTC GAT GAA CAA,
His56/Trp60 forward primer: 5’-CGT GAA CAG ATT GCG GCG NNS
TAC CGT AAG NNS CTG GGC GGG GGT AAA CTG,
L103K reverse primer: 5’-AAT AAC ATC CGT TGC GCA AGG.
His56/Trp60 reverse primer: 5’-CGC CGC AAT CTG TTC ACG;
The gene was reassembled by PCR, digested, ligated into the pET-
2
9b(+) vector, and transformed into Escherichia coli BL21-gold(DE3)
Met105 forward primer: 5’-T TGC GCA ACG GAT GTT ATT AAA
GTT NNS CGC TTC GAT GAA CAA GGT CG,
cells (Agilent Technologies, Santa Clara, CA, USA) as described
below. The plasmid of the most active variant was isolated, se-
quenced, and used for further mutagenesis. Mutations found in
active variants were also introduced into RA110.4 by two sequen-
tial overlap extension PCRs, first substituting Lys88 by cysteine:
Met105 reverse primer: 5’-AAC TTT AAT AAC ATC CGT TGC GCA
A;
Arg118 forward primer: 5’-CGC ATC CAG ACC GAA CAG NNS TAT
TGG AGC GAA GTG AAT CTG TGT GAC,
forward primer: 5’-GCG ATG CCG TTG CGT TGC GAG TGG GTT TGG
AAT GGT CAG,
Arg118 reverse primer: 5’-CTG TTC GGT CTG GAT GCG AC.
reverse primer: 5’-CTG ACC ATT CCA AAC CCA CTC GCA ACG CAA
CGG CAT CGC
The resulting libraries were digested, ligated into pET-29b(+), and
used to transform BL21-gold(DE3) cells as described above.
and then introducing a lysine at position 101 or 103:
V101K forward primer: 5’-CAG CCT TGC GCA ACG GAT AAA ATT
CTG GTT ATG CGC TTC GAT GA,
Screening
In each round, approximately 800 clones were screened as de-
scribed previously.
[11–12]
V101K reverse primer: 5’-ATC CGT TGC GCA AGG CTG;
Single-colony transformants were used to
inoculate LB medium (150 mL) containing kanamycin sulfate
L103K forward primer: 5’-GCA ACG GAT GTT ATT AAA GTT ATG
CGC TTC GAT GAA C,
ꢁ1
(30 mgmL ). After incubation at 308C and 900 rpm overnight, the
precultures were used to inoculate prewarmed expression cultures
(1.5 mL) in 96-well plates. The resulting cultures were incubated at
L103K reverse primer: 5’-AAT AAC ATC CGT TGC GCA AGG.
3
78C and 600 rpm for 2.25 h. After induction with isopropyl b-d-1-
thiogalactopyranoside (0.1 mm), the cells were incubated for an-
other 5 h, harvested by centrifugation, and stored overnight at
Library construction
ꢁ
208C. Cell pellets were resuspended in assay buffer (25 mm N-(2-
The genes for RA110.V and RA110.L were used as templates for di-
rected evolution experiments. Random mutagenesis libraries with
one to two amino acid changes per variant were constructed by
performing error-prone PCR (epPCR) with the GeneMorph II
Random Mutagenesis Kit (Agilent Technologies) as recommend-
hydroxyethyl)piperazine-N’-ethanesulfonic acid (HEPES), 100 mm
ꢁ1
NaCl, pH 7.5) containing egg white lysozyme (1 mgmL ) and eth-
ylenediaminetetraacetic acid (0.5 mm) and incubated at RT for 1 h.
The lysate was cleared by centrifugation and transferred to a black
fluorescence 96-well plate (FluoroNunc, Nunc, Penfield, NY, USA).
The substrate (ꢀ)-methodol (270 mm) was added and product for-
mation monitored at 298C with a Varioskan Flash multimode
[
25]
ed.
The purified linear epPCR fragments and the pET-29b(+)
vector containing a stuffer fragment were digested with NdeI and
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1048

CHEMCATCHEM
FULL PAPERS
www.chemcatchem.org
reader (Thermo Scientific, Waltham, MA, USA; lexcitation =330 nm;
lemission =452 nm). For reference, the best variant from the previous
round and a negative control were included on each plate.
in which v denotes the initial velocity, [E ] the enzyme concentra-
tion, and [S] the substrate concentration. The pH–rate profiles
0
0
were determined under subsaturating conditions ([S] was ꢂ10-fold
lower than K ). The following buffers were used: 25 mm 2-(N-mor-
M
pholino)ethanesulfonic acid, 100 mm NaCl, pH 5.2–6.8; 25 mm
HEPES, 100 mm NaCl, pH 6.8–8.5; and 25 mm 2-amino-2-methyl-1-
propanol, 100 mm NaCl, pH 8.5–10. The data were corrected for
the background reaction at the corresponding pH value and fitted
to Equation (2):
Protein production and purification
Proteins were produced with a C-terminal His tag in E. coli BL21-
gold(DE3) cells and purified by affinity chromatography. LB
medium (200 mL) containing kanamycin sulfate (30 mgmL ) was
inoculated with an overnight culture (200 mL) prepared from
a single colony and incubated at 378C and 230 rpm until the cell
density reached an optical density of 0.4 at a wavelength of
6
ꢁ1
ðkcat=K
Þ
obs ¼ ðkcat=K
Þ
max=ð1 þ 10pKaꢁpH
Þ
ð2Þ
M
M
DTT was removed from the protein samples immediately before
the assay with a NAP-5 column.
6
00 nm. Protein expression was induced with isopropyl b-d-1-thio-
galactopyranoside (0.1 mm), and the cell culture was incubated for
h at 378C and 230 rpm before harvesting. The cell pellets were
5
stored at ꢁ208C prior to purification. Upon thawing, the pellets
Circular dichroism spectroscopy
were resuspended in sonication buffer (25 mm HEPES, 300 mm
ꢁ
1
The circular dichroism spectra and thermal denaturation curves
were recorded in aqueous buffer (10 mm phosphate, 10 mm NaCl,
pH 7.5) with an Aviv 202 spectropolarimeter (Aviv Association,
Lakewood, NJ, USA) and a sealed 1 cm quartz cuvette. Three far-UV
wavelength scans were performed from 260 to 200 nm (step size=
NaCl, pH 7.5) containing egg white lysozyme (1 mgmL ) and incu-
bated at 48C for 30 min. The cells were sonicated, and cell debris
was removed by centrifugation. The filtered soluble protein frac-
tion was loaded onto Ni-NTA beads (Qiagen, Venlo, The Nether-
lands) equilibrated with washing buffer (25 mm HEPES, 300 mm
NaCl, 20 mm imidazole, pH 7.5) and washed with buffer containing
imidazole (20 and 40 mm). The protein was eluted with buffer con-
taining imidazole (200 mm). The purity of the eluted protein was
assessed by using sodium dodecyl sulfate polyacrylamide gel elec-
trophoresis. Pure protein fractions were pooled and dialyzed
against assay buffer (25 mm HEPES, 100 mm NaCl, pH 7.5) at
1
nm; bandwidth=1 nm) with an averaging time of 3 s at 258C.
The average values were corrected for buffer background. Thermal
denaturation curves were recorded at 222 nm (bandwidth=1 nm)
between 25 and 908C (step size=0.58C) with an averaging time of
ꢁ
1
1
min (heating rate: 18Cmin ). The experimental data were cor-
rected for the enzyme concentration by the following relationship
[Eq. (3)]:
4
8C. Protein concentration was determined by using absorption
with calculated extinction coefficients (http://expasy.org/tools/
protparam.html). For purification under reducing conditions, all
buffers were degassed and contained b-mercaptoethanol (2 mm;
for purification) or DTT (2 mm; for dialysis).
½
Vꢃ ¼ ðV ꢄ 100 ꢄ M Þ=ðl c nÞ
ð3Þ
W
where [V] denotes the mean residue ellipticity, V the observed el-
lipticity, n the number of amino acids, l the path length, c the con-
centration of the protein, and MW the molecular mass. The melting
temperature Tm was obtained by determining the inflexion point.
Fluorescence assay
Kinetic assays were performed in black fluorescence 96-well plates
with a plate reader (SpectraMax M2, Molecular Devices). All meas-
urements were conducted at 298C in assay buffer (25 mm HEPES,
Substrate and inhibitor synthesis
Racemic 1 was prepared by using the method described in the lit-
1
00 mm NaCl, pH 7.5) containing acetonitrile (2.7%; 5.4% for sub-
[28]
[12,13]
erature. It was oxidized as described previously
,3-diketone inhibitor 1-(6-methoxy-2-naphthalenyl)-1,3-butane-
dione.
to give the
[12]
strate concentrations higher than 540 mm). The initial rates were
determined by monitoring the formation of the fluorescent prod-
uct 3 (lexcitation =330 nm; lemission =452 nm). The fluorescence signal
was corrected for the buffer-catalyzed background reaction under
identical conditions and converted into product concentration by
using a calibration curve. The data were fitted to the Michaelis–
Menten equation as described below. DTT was removed from the
protein samples immediately before the assay by using a NAP-5
column (GE Healthcare, LifeScience).
1
Crystallization of RA110.4-6 complexed with a mechanism-
based inhibitor
After nickel affinity chromatography, the retro-aldolase RA110.4-6
was buffer exchanged with crystallization buffer by ultrafiltration
(5 mm HEPES, 150 mm NaCl, 1 mm ethylenediaminetetraacetic acid,
pH 7.5, 20 mm b-mercaptoethanol) and purified by gel filtration
chromatography over Superdex 75 with the same buffer. Fractions
containing the protein were pooled and concentrated through ul-
trafiltration for inhibition. The retro-aldolase (100 mm) was mixed
with a 1:2.5 molar ratio of 1-(6-methoxy-2-naphthalenyl)-1,3-buta-
nedione (4) in crystallization buffer supplemented with acetonitrile
(2.5%). The slurry reaction was gently shaken for 36 h at 308C. The
precipitate was removed through centrifugation, and the com-
pleteness of the reaction was assessed by using the UV364/UV280
ratio as well as by MS. Acetonitrile and excess diketone were re-
moved by buffer exchange with fresh crystallization buffer by ultra-
filtration. The complex was crystallized through vapor diffusion at
UV/Vis spectroscopic assay
Reactions were performed at 298C in assay buffer (25 mm HEPES,
1
00 mm NaCl, pH 7.5) containing acetonitrile (2.7%; 5.4% for sub-
strate concentrations higher than 540 mm). Formation of 3 was
ꢁ
1
ꢁ1 [12]
monitored spectroscopically at 350 nm (e=5970m cm ). The
initial rates were corrected for the background reaction. The
steady-state parameters kcat and KM were determined by fitting the
data to the Michaelis–Menten equation [Eq. (1)]:
v =½E ꢃ ¼ kcat½Sꢃ=ðKM þ ½SꢃÞ
ð1Þ
0
0
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1049

CHEMCATCHEM
FULL PAPERS
www.chemcatchem.org
2
08C in sitting drops comprising the enzyme–diketone adduct
[1] T. Gefflaut, C. Blonski, J. Perie, M. Willson, Prog. Biophys. Mol. Biol. 1995,
ꢁ
1
6
3, 301–340.
2] S. G. Zlotin, A. S. Kucherenko, I. P. Beletskaya, Russ. Chem. Rev. 2009, 78,
37–784.
(
100 nL, 50 mgmL ) and mother liquor (100 nL, 50 mm C H NaO ,
2 3 2
[
pH 4.5, 1m NaCl, 45% (v/v) PEG-400). Crystals were cryoprotected
in the mother liquor complemented with glycerol (25%, v/v) and
flash frozen in liquid N2.
7
[
[
3] F. Tanaka, R. Fuller, C. F. Barbas, III., Biochemistry 2005, 44, 7583–7592.
4] M. M. Mꢀller, M. A. Windsor, W. C. Pomerantz, S. H. Gellman, D. Hilvert,
Angew. Chem. Int. Ed. 2009, 48, 922–925; Angew. Chem. 2009, 121,
940–943.
[
5] W. Tagaki, H. Yamamoto, Tetrahedron Lett. 1991, 32, 1207–1208.
Crystallographic methods
[6] B. Wçrsdçrfer, L. M. Henning, R. Obexer, D. Hilvert, ACS Catal. 2012, 2,
82–985.
9
Crystallographic data were collected at 100 K by using l=0.98 ꢂ at
the Swiss Light Source PXI-X06SA beamline (Paul Scherrer Institute,
Villigen, Switzerland). The diffraction data were indexed and inte-
[
[
7] J. Wagner, R. A. Lerner, C. F. Barbas, III., Science 1995, 270, 1797–1800.
8] G. F. Zhong, R. A. Lerner, C. F. Barbas, Angew. Chem. Int. Ed. 1999, 38,
3738–3741; Angew. Chem. 1999, 111, 3957–3960.
[29]
grated with XDS. Crystals belonged to the P6 22 space group
[9] X. Zhu, F. Tanaka, R. A. Lerner, C. F. Barbas, I. A. Wilson, J. Am. Chem. Soc.
2009, 131, 18206–18207.
10] L. Jiang, E. A. Althoff, F. R. Clemente, L. Doyle, D. Rothlisberger, A. Zan-
ghellini, J. L. Gallaher, J. L. Betker, F. Tanaka, C. F. Barbas, III, D. Hilvert,
K. N. Houk, B. L. Stoddard, D. Baker, Science 2008, 319, 1387–1391.
1
(a=51.58 ꢂ, b=51.58 ꢂ, c=198.64 ꢂ, a=b=908, g=1208). The
[30]
[
structure was solved by molecular replacement with PHASER,
starting with the deposited model of the design scaffold, ketoste-
roid isomerase from Pseudomonas putida (PDB accession code:
[31]
[11] E. A. Althoff, L. Wang, L. Jiang, L. Giger, J. K. Lassila, Z. Wang, M. Smith,
1
OHO). The structure was refined with PHENIX, with manual cor-
S. Hari, P. Kast, D. Herschlag, D. Hilvert, D. Baker, Protein Sci. 2012, 21,
[32]
rections made with the COOT program. Restraints at the pro-
tein–ligand interface were generated by MM2 energy minimization
with the ChemBio3D Ultra 12.0 program. The resolution was esti-
mated and the data were cut off by limiting Rmeas to approximately
717–726.
[
12] L. Giger, S. Caner, R. Obexer, P. Kast, D. Baker, N. Ban, D. Hilvert, Nat.
Chem. Biol. 2013, 9, 494–498.
[13] L. Wang, E. A. Althoff, J. Bolduc, L. Jiang, J. Moody, J. K. Lassila, L. Giger,
1
00%. The backbone dihedral angles were distributed within the
D. Hilvert, B. Stoddard, D. Baker, J. Mol. Biol. 2012, 415, 615–625.
[
14] D. Rçthlisberger, O. Khersonsky, A. M. Wollacott, L. Jiang, J. DeChancie,
J. Betker, J. L. Gallaher, E. A. Althoff, A. Zanghellini, O. Dym, S. Albeck,
K. N. Houk, D. S. Tawfik, D. Baker, Nature 2008, 453, 190–195.
15] H. K. Privett, G. Kiss, T. M. Lee, R. Blomberg, R. A. Chica, L. M. Thomas, D.
Hilvert, K. N. Houk, S. L. Mayo, Proc. Natl. Acad. Sci. USA 2012, 109,
most favored (95.2%) regions as well as allowed (4.8%) regions of
the Ramachandran map. Details of the refinement statistics are
summarized in Table S1.
[
3
790–3795.
16] S. W. Kim, S. Joo, G. Choi, H. S. Cho, B. H. Oh, K. Y. Choi, J. Bacteriol.
997, 179, 7742–7747.
17] A. Zanghellini, L. Jiang, A. M. Wollacott, G. Cheng, J. Meiler, E. A. Althoff,
D. Rçthlisberger, D. Baker, Protein Sci. 2006, 15, 2785–2794.
[
[
[
Accession number
1
Coordinates for RA110.4-6 have been deposited in the RCSB Pro-
tein Data Bank with the accession code 4cdl.
18] J. K. Lassila, D. Baker, D. Herschlag, Proc. Natl. Acad. Sci. USA 2010, 107,
4937–4942.
[
[
19] Y. Kipnis, D. Baker, Protein Sci. 2012, 21, 1388–1395.
20] N. Nagano, C. A. Orengo, J. M. Thornton, J. Mol. Biol. 2002, 321, 741–
7
65.
Acknowledgements
[
[
21] D. Baker, Protein Sci. 2010, 19, 1817–1819.
22] G. Kiss, N. C¸ elebi-ꢃlÅꢀm, R. Moretti, D. Baker, K. N. Houk, Angew. Chem.
Int. Ed. 2013, 52, 5700–5725; Angew. Chem. 2013, 125, 5810–5836.
23] D. Hilvert, Annu. Rev. Biochem. 2013, 82, 447–470.
We thank Cꢁline Stutz and Beat Blattmann for help in protein
crystallization as well as the beamline staff at the Swiss Light
Source for support during data collection. We also thank Peer
Mittl and Christophe Briand for assisting with the crystallography
and Eric Althoff for providing the genes for RA110.4. We acknowl-
edge support by the Swiss National Science Foundation (SNSF),
the ETH Zurich, the National Center of Excellence in Research
[
[
24] S. N. Ho, H. D. Hunt, R. M. Horton, J. K. Pullen, L. R. Pease, Gene 1989,
7
7, 51–59.
[
25] C. Neylon, Nucleic Acids Res. 2004, 32, 1448–1459.
[26] D. W. Leung, E. Chen, D. V. Goeddel, Technique 1989, 1, 11–15.
27] W. P. Stemmer, Proc. Natl. Acad. Sci. USA 1994, 91, 10747–10751.
28] B. List, C. F. Barbas, III., R. A. Lerner, Proc. Natl. Acad. Sci. USA 1998, 95,
[
[
15351–15355.
(NCCR) Structural Biology program of the SNSF, and the Defense
[
29] W. Kabsch, Acta Crystallogr. Sect. D 2010, 66, 125–132.
Advanced Research Projects Agency (DARPA). L.G. was funded by
the Stipendienfonds der Schweizerischen Chemischen Industrie
[30] A. J. McCoy, R. W. Grosse-Kunstleve, P. D. Adams, M. D. Winn, L. C. Storo-
ni, R. J. Read, J. Appl. Crystallogr. 2007, 40, 658–674.
[
[
31] P. D. Adams, K. Gopal, R. W. Grosse-Kunstleve, L.-W. Hung, T. R. Ioerger,
A. J. McCoy, N. W. Moriarty, R. K. Pai, R. J. Read, T. D. Romo, J. C. Sacchet-
tini, N. K. Sauter, L. C. Storoni, T. C. Terwilliger, J. Synchrotron Radiat.
(SSCI).
2
004, 11, 53–55.
Keywords: computational chemistry
·
enzyme catalysis
·
32] P. Emsley, K. Cowtan, Acta Crystallogr. Sect. D 2004, 60, 2126–2132.
enzyme models
relationships
·
protein folding
·
structure–activity
Received: November 1, 2013
Published online on March 3, 2014
ꢁ
2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemCatChem 2014, 6, 1043 – 1050 1050