Origins of catalysis by computationally designed retroaldolase enzymes

Article abstract of DOI:10.1073/pnas.0913638107

We have investigated recently reported computationally designed retroaldolase enzymes with the goal of understanding the extent and the origins of their catalytic power. Direct comparison of the designed enzymes to primary amine catalysts in solution revealed a rate acceleration of 10⁵-fold for the most active of the designed retroaldolases. Through pH-rate studies of the designed retroaldolases and evaluation of a Bronsted correlation for a series of amine catalysts, we found that lysine pK_a values are shifted by 3-4 units in the enzymes but that the catalytic contributions fromthe shifted pK_a values are estimated to be modest, about 10-fold. For the most active of the reported enzymes, we evaluated the catalytic contribution of two other design components: a motif intended to stabilize a bound water molecule and hydrophobic substrate binding interactions. Mutational analysis suggested that the bound water motif does not contribute to the rate acceleration. Comparison of the rate acceleration of the designed substrate relative to a minimal substrate suggested that hydrophobic substrate binding interactions contribute around 10³-fold to the enzymatic rate acceleration. Altogether, these results suggest that substrate binding interactions and shifting the pKa of the catalytic lysine can account for much of the enzyme's rate acceleration. Additional observations suggest that these interactions are limited in the specificity of placement of substrate and active site catalytic groups. Thus, future design efforts may benefit from a focus on achieving precision in binding interactions and placement of catalytic groups.

Full text of DOI:10.1073/pnas.0913638107

Origins of catalysis by computationally
designed retroaldolase enzymes
a
b
a,1
Jonathan K. Lassila , David Baker , and Daniel Herschlag
a
b
Department of Biochemistry, Stanford University, Stanford, CA 94305; and Department of Biochemistry, University of Washington, Seattle, WA 98195
Edited* by Stephen J. Benkovic, The Pennsylvania State University, University Park, PA, and approved January 27, 2010 (received for review November 25, 2009)
We have investigated recently reported computationally designed
retroaldolase enzymes with the goal of understanding the extent
and the origins of their catalytic power. Direct comparison of the
designed enzymes to primary amine catalysts in solution revealed
is catalyzed by amines through the formation of an iminium (or
Schiff base) intermediate. Analogous to the strategy used by
type I aldolases (10), formation of the iminium intermediate with
the enzymatic lysine side chain provides an electron sink,
facilitating the retroaldol cleavage. Several catalytic antibodies
and peptide systems have been developed that utilize this lysine
iminium strategy (11–16).
5
a rate acceleration of 10 -fold for the most active of the designed
retroaldolases. Through pH-rate studies of the designed retroaldo-
lases and evaluation of a Brønsted correlation for a series of amine
catalysts, we found that lysine pKa values are shifted by 3–4 units
in the enzymes but that the catalytic contributions from the shifted
pKa values are estimated to be modest, about 10-fold. For the most
active of the reported enzymes, we evaluated the catalytic contri-
bution of two other design components: a motif intended to sta-
bilize a bound water molecule and hydrophobic substrate binding
interactions. Mutational analysis suggested that the bound water
motif does not contribute to the rate acceleration. Comparison of
the rate acceleration of the designed substrate relative to a
minimal substrate suggested that hydrophobic substrate binding
The lysine side chain provides a key catalytic element in the
enzymatic reaction, yet the free lysine side chain in solution
can also catalyze the retroaldol reaction. Thus, to evaluate the
contribution of the computationally designed active sites to
catalysis, we first determined the rate acceleration of the enzymes
beyond that of the lysine side chain alone. Next, we asked how the
computational design procedure facilitates additional catalysis
beyond that of the lysine side chain alone. We investigated the
catalytic contribution of each of three elements used in designing
the enzymes: a hydrophobic pocket intended to lower the pK of
a
3
interactions contribute around 10 -fold to the enzymatic rate accel-
the catalytic lysine, a stabilized, positioned water molecule for
facilitation of proton transfer, and binding interactions with
the substrate (Fig. 2) (7). We provide an accounting for much
of the catalysis by the most active of the computationally designed
retroaldolases and use these results to evaluate strengths and
limitations of current enzyme design methodology.
eration. Altogether, these results suggest that substrate binding
interactions and shifting the pKa of the catalytic lysine can account
for much of the enzyme’s rate acceleration. Additional observa-
tions suggest that these interactions are limited in the specificity of
placement of substrate and active site catalytic groups. Thus, fu-
ture design efforts may benefit from a focus on achieving precision
in binding interactions and placement of catalytic groups.
Results and Discussion
Conditions for Kinetic Studies. To investigate mechanisms of catal-
ysis in an enzymatic system, it is necessary to identify assay con-
ditions in which the enzyme is stable and the substrate is soluble.
We found that substrate solubility was exceeded in previously
used conditions for the designed enzymes (7,17). Thus, we tested
buffers and cosolvents and identified conditions that provided
improved solubility while protein stability was maintained (see
Materials and Methods). For the variants tested herein, the en-
zymes could not be saturated with substrate under conditions
where the substrate remained soluble. We therefore focused
amine ∣ Brønsted correlation ∣ catalytic mechanism ∣
computational enzyme design ∣ retroaldol reaction
ecause natural enzymes catalyze reactions with tremendous
B
rate accelerations and specificities, a long-standing goal in
enzymology and protein engineering has been to reliably design
new enzyme catalysts for chemical reactions of interest. Compu-
tational protein design offers a promising tool for achieving this
goal. Computational protein design methods use specialized
potential energy functions in combination with search algorithms
to optimize an amino acid sequence for a given protein structure
and function (1–3). These methods have been adapted to the
challenge of designing enzyme active sites (4–8).
exclusively on second-order rate constants (kcat∕K ), mea-
M
sured under subsaturating conditions where initial rates in-
creased linearly with both enzyme and substrate concentration
(
ν ¼ k ∕K ½Eꢀ½Sꢀ). The resulting values of k ∕K are within
o
cat
M
cat
M
Although new enzymatic activities have been designed com-
putationally, the resulting catalysts have catalytic efficiencies
2
-fold of those from the original report for the variants tested
(
Table 1) (7).
The second-order rate constant kcat∕K reports on the differ-
−
1 −1
(
kcat∕K values) of about 1–100 M
s
(4,7,8), considerably less
M
5
9
−1 −1
M
than values of 10 –10 M
s
typical of natural enzymes, and
ence in energy between the free enzyme and substrate in solution
and the rate-limiting transition state in the steps up to and includ-
ing formation of the 6-methoxy-2-naphthaldehyde product that is
monitored by fluorescence (steps 1–3, Fig. 1). These rate con-
stants are especially useful for our purposes because they can
be directly compared to the second-order rate constants for
similar to those of early catalytic antibodies. In contrast to cata-
lytic antibody methods and other stochastic processes, however,
the potential for success of computational enzyme design is tied
to the predictive power of the computational model. Thus, future
improvement of our ability to computationally design enzymatic
activity will require ongoing rigorous assessment of the successes
and failures of the design process.
We therefore sought to investigate the origins of the catalytic
power of recently reported computationally designed retroaldo-
lases (7). One of few examples of catalytic activities designed using
computational modeling alone, these enzymes were developed to
catalyze the retroaldol cleavage of 4-hydroxy-4-(6-methoxy-
Author contributions: J.K.L. and D.H. designed research; J.K.L. performed research; and
J.K.L., D.B., and D.H. analyzed data.
The authors declare no conflict of interest.
*This Direct Submission article had a prearranged editor.
1
To whom correspondence should be addressed. Email: herschla@stanford.edu.
2
-naphthyl)-2-butanone (Fig. 1), a fluorogenic substrate devel-
This article contains supporting information online at www.pnas.org/cgi/content/full/
0913638107/DCSupplemental.
oped for catalytic antibody studies (9). This retroaldol reaction
www.pnas.org/cgi/doi/10.1073/pnas.0913638107
PNAS ∣ March 16, 2010 ∣ vol. 107 ∣ no. 11 ∣ 4937–4942

Lys
Lys
STEP 1
STEP 2
Lys
O
OH
O, + H+
+HN
OH
HN
OH
- H
2
+
H N
OH
2
O
4
2
-hydroxy-4-(6-methoxy-
-naphthyl)-2-butanone
O
O
STEP 3
Lys
N
Lys
STEP 5
H O
STEP 4
Lys
O
O
O
+
2
O
+
+
+
+
HN
H N
2
O
O
O
6
-methoxy-2-naphthaldehyde
Fig. 1. Major steps in amine-catalyzed retroaldol reaction. Proton transfers, binding events, and some intermediate steps are omitted for clarity.
the corresponding nonenzymatic reactions of primary amines in
solution. This comparison provides a direct evaluation of the
transition state stabilization provided by the designed enzyme ac-
tive sites.
lost upon binding to the enzyme. For all of the retroaldolase vari-
ants tested, an exponential loss of fluorescent signal was evident
when 6-methoxy-2-naphthaldehyde was incubated with enzyme in
the absence of substrate (SI Appendix).
For practical reasons described below, we obtained kcat∕KM
values from initial rates under presteady-state conditions, before
the first turnover. Under subsaturating conditions, kcat∕KM
values obtained before the first turnover are the same as those
obtained under multiple-turnover conditions. For the designed
retroaldolases, however, tight binding of the fluorescent product
creates a special complication that can make multiple-turnover
conditions difficult to achieve. This complication is described
in the following paragraphs because it results in unexpected assay
behavior that could be misinterpreted as a kinetic burst, and be-
cause it has interesting implications for the specificity of the ac-
tive site. However, the present studies used initial rates measured
under conditions where only low concentrations of product were
formed, so the complication described below does not affect the
kinetic results reported herein.
As the 6-methoxy-2-naphthaldehyde product begins to build
up, it can return to the active site to form a covalent Schiff base
species with the active site lysine (SI Appendix). This covalent en-
zyme-product species is not a part of the reaction pathway, and its
stability and rate of formation may be enhanced by the aldehyde
functionality of the product relative to the ketone of the sub-
strate (18).
During a reaction time course beginning with substrate, this re-
binding of product can manifest as an apparent burst of fluores-
cence (SI Appendix). Although the reductions in the rate of
fluorescence development arising from this mechanism can ap-
pear burst-like, they need not follow normal expectations for burst
kinetics. Reductions in apparent rate arising from this mechanism
can occur under second-order conditions where the enzyme is
almost entirely unsaturated by substrate, when the enzyme is in
excess, and well after or before a full turnover of the enzyme.
Thus, there are two sources of burst-like kinetics to consider in
this system. First, traditional burst kinetics may arise under
saturating conditions due to the occurrence of a rate-limiting step
subsequent to product formation, such as proton transfer or re-
lease of acetone in steps 4 and 5 (Fig. 1). Second, the apparent
burst described above relates to product rebinding to the enzyme
in an off-pathway state and does not provide kinetic information
about steps 4 and 5. For example, RA61 shows continuous
reaction progress curves over multiple turnovers at low enzyme
concentration but curvature in the fluorescence time course at
higher enzyme concentration where product rebinding is more
pronounced (SI Appendix). We estimate a K for product disso-
d
ciation from RA61 of about 10–30 μM, considerably lower than
Formation of the covalent enzyme-product species reduces the
reaction rate in the fluorescence-based assay in two ways. First,
the concentration of free enzyme available to react is reduced;
this reduces the actual rate of the reaction. Second, the measured
rate is reduced because the fluorescent signal of the product is
the K , which is greater than 500 μM (SI Appendix).
M
Because the present studies were performed under conditions
where little product builds up during the reactions and with
enzyme concentrations below the Kd for product rebinding, the
fluorescence loss due to product binding does not complicate our
Fig. 2. Three elements comprise the catalytic motif used for the retroaldolases investigated in this paper (7), illustrated here with the designed model for the
most active enzyme, RA61. The experimental crystal structure for RA61 did not contain ligand, and the binding orientation of the substrate is not known.
(A) A hydrophobic pocket was intended to lower the pKa of the catalytic lysine side chain. (B) Hydrogen-bonding interactions to a bound water molecule were
incorporated into the design models. In RA61, the side chains modeled in contact with the bound water are Tyr78 and Ser87. (C) Hydrophobic side chains line
the active site cavity, providing a binding surface for the hydrophobic substrate.
4
938
∣
www.pnas.org/cgi/doi/10.1073/pnas.0913638107
Lassila et al.

Lys
Table 1. Second-order rate constants at pH 7.5 and rate
acceleration relative to lysine in solution
A
B
O
OH
OH
(kcat/K
M
)
max
+
+
products
Catalyst
ðkcat∕KMÞobs, (M− s−1) *
1
Rate acceleration
H N
2
O
O
RA61
RA60
RA45
RA34
0.49
0.16
0.11
0.22
2.0 × 105
Ka
4
6.7 × 10
4.6 × 104
9.2 × 104
(1)
Lys
O
−
6
Lysine in solution
2.4 × 10
+
H N
3
*
Observed second-order rate constants at 25 °C, in 50 mM sodium
phosphate, 300 mM NaCl, and 5% DMSO, pH 7.5. Standard
deviations were less than 20% over multiple assays and protein
preparations.
2
1
0
efforts to investigate contributions to catalysis of the
retroaldol reaction (steps 1–3). Nevertheless, this unexpected and
tight product inhibition illustrates how one hallmark of natural
enzymes, their exquisite specificity, remains a challenge for
enzyme engineering.
-1
-
2
-3
-
4
Rate Acceleration Relative to Lysine Side Chain in Solution. In the
protein engineering literature, rate acceleration is frequently re-
ported as kcat∕kuncat, where kuncat often represents the first-order,
nonenzymatic water-mediated reaction. However, mechanisti-
cally appropriate small-molecule catalysts in solution can provide
another useful comparison (19–21). In the case of the retroaldo-
lases, the enzymatic reaction is catalyzed by a lysine side chain
that is required for activity (7), whereas the reaction in water
is specific-base catalyzed through a different chemical mechan-
ism. Thus, for understanding the catalytic contribution of the
designed active sites, a more direct measure of rate acceleration
is provided by comparing the second-order rate constant for the
enzymatic active site lysine with that of the lysine side chain
alone, free in solution.
4
5
6
7
pH
8
9
1 0
Fig. 3. Dependence of second-order rate constant on pH. (A) Scheme for pH
dependence. The deprotonated, neutral form of lysine is required for
reaction, whereas protonated, charged lysine cannot form the iminium
and progress to products. Thus, in this model, the concentration of active
enzyme is controlled by the pKa of the catalytic lysine. (B) pH dependence
for the RA61-catalyzed reaction. The fit to a single titratable group gives
a pKa of 6.8. Plots for the other variants are included in the SI Appendix.
Table 2). The assignment of the observed pK values to the cat-
a
alytic lysine is discussed further in SI Appendix and is supported
by a pH-binding study for RA61. Relative to the solution pK of
a
the lysine side chain of 10.6, observed pK values are considerably
a
shifted in the retroaldolases, with measured values from 6.8–7.5.
For comparison to the retroaldolase rate constants, Table 1 in-
cludes a value for the nonenzymatic, second-order rate constant
of the lysine side chain in solution at pH 7.5, taken from the value
for a primary alkylamine of pKa 10.6, as described in a later
section. The most active of the reported retroaldolases has a rate
These perturbed pKa values are consistent with prior observa-
tions of lysine pK shifts in similar enzymes, including acetoace-
a
tate decarboxylase and catalytic antibodies and helical peptides
that catalyze the same or related reactions (11, 22–24).
5
Our goal was to estimate the catalytic contribution of these
observed pKa shifts, and for this we needed more information.
Lowering the pKa by one unit can increase the reactivity by
raising the concentration of active, neutral lysine by up to 10-fold.
However, lowering the pKa also reduces the reactivity of the
neutral amine species, as the amine becomes less nucleophilic.
To determine how much the reactivity decreases, we deter-
acceleration of 2 × 10 relative to the lysine side chain in solution,
4
5
and the other enzyme rate constants range from 10 - to 10 -fold
above lysine in solution.
†
How do the designed enzymes achieve these rate accelerations
beyond those of the catalytic lysine side chains within their active
sites? The computational design strategy included three elements
intended to contribute to catalysis: a hydrophobic pocket to lower
mined the relationship between pK and reactivity for a series of
a
the pK of the catalytic lysine, a bound water motif designed to
a
nonenzymatic amines in solution. This information, combined
facilitate proton transfers, and binding interactions with the sub-
strate (Fig. 2) (7). We investigated the contributions provided by
each of these elements to the rate acceleration of the designed
retroaldolases, with a focus on the most active of these enzymes,
RA61.
with the measured enzymatic pK values, allowed us to estimate
a
the catalytic contribution from lowering the pK of the enzymatic
a
lysine side chains.
We measured rate constants for the retroaldol reaction of
4
-hydroxy-4-(6-methoxy-2-naphthyl)-2-butanone catalyzed by a
series of primary amines of the structure RCH NH2 across a
2
Catalytic Contribution of Shifted Lysine pKa. One of the strategies
broad range of pK values (Table 3). Plotting the log of the max-
a
employed in the computational design process was to attempt
imal rate constants for the amines against their pK values yields
a
to lower the pK of the catalytic lysine residue by placing it within
a
the linear relationship with slope of þ0.54 ꢁ 0.03 (Fig. 4A). This
a hydrophobic pocket (Fig. 2A). Because the deprotonated, neu-
slope represents a Brønsted value, a quantity related to changes
tral form of lysine is required to form the iminium intermediate
and progress to products, lowering the pK of the catalytic lysine
a
effectively increases the concentration of active enzyme available
to react, thereby increasing the observed second-order rate con-
stant for reaction of free enzyme and substrate (Fig. 3A).
To determine whether the computational design process was
Table 2. pKa values for the designed retroaldolases obtained from
pH-rate profiles and estimated catalytic contribution calculated
with a Brønsted value of þ0.54
Retroaldolase
pKa
Estimated catalytic contribution
successful in lowering the pK of the catalytic lysine, we measured
a
pH-rate profiles for several of the retroaldolases (Fig. 3B and
RA34
RA45
RA60
RA61
>7
7.5
7.2
6.8
≤14-fold
14-fold
13-fold
10-fold
†
These values are reduced at higher pH; for example, the rate accelerations for RA61
would be 2.4 × 10 and 2.6 × 10 at pH 8.5 and 9.5, respectively.
4
3
Lassila et al.
PNAS
∣
March 16, 2010
∣
vol. 107
∣
no. 11
∣
4939

Table 3. pH-independent second-order rate constants for amine-
catalyzed retroaldol reactions of 4-hydroxy-4-(6-methoxy-2-
naphthyl)-2-butanone
Several observations suggest that the nonenzymatic Brønsted
value provides a reasonable estimate of the relationship between
reactivity and pKa in the enzyme. First, studies of nucleophilic
kmax (M−1 s−1) *
substitution reactions have suggested that changes in pK values
a
Amine
pKa
5.3
5.7
8.2
8.5
8.8
9.5
10.6
10.6
arising from changes in substituent groups (as used to obtain the
nonenzymatic Brønsted value here) have similar effects on reac-
tivity to changes in pKa values arising from different solvation
environments (as the enzymatic pKa shift presumably entails),
and equivalent Brønsted values have been measured for solvation
and substituent pKa perturbations (27, 28). Second, although
differences in the rate-limiting step between nonenzymatic and
enzymatic reactions are possible, a Brønsted value of þ0.8 re-
ported for iminium formation with primary amines (29) suggests
that if iminium formation rather than the retroaldol step limited
either the enzymatic or nonenzymatic reactions, a value larger
than þ0.54 might be expected. A larger Brønsted value, whether
arising from a different rate-limiting step, or as is sometimes ob-
served, from placement in a low dielectric environment, would
result in smaller catalytic contributions from the pKa shifts
(see SI Appendix for catalytic contributions calculated using dif-
ferent Brønsted values). Finally, because the crystallographic
structure of RA61 shows a relatively open, partially solvent-
exposed active site (Fig. 2C), dramatic changes in mechanism
for the enzymatic retroaldol cleavage relative to the solution
amine-catalyzed reaction are not anticipated. Altogether, it
seems likely that the catalytic contribution of shifting the pKa of
the active site lysine in the retroaldolases is around 10-fold and
does not account for the majority of their rate acceleration.
What other elements of the designed retroaldolase active sites
provide their rate acceleration above that of the lysine side chain
Cyanomethylamine
Trifluoroethylamine
Propargylamine
Bromoethylamine
Chloroethylamine
Ethanolamine
5.8 × 10−
7.1 × 10−
1.0 × 10−
6
6
4
1.5 × 10−
4
4
4
3
3
−
4.0 × 10
6.9 × 10−
−
Butylamine
Methylamine
5.3 × 10
2.6 × 10−
*
kmax refers to the rate constant for the fully deprotonated, neutral species,
whereas kobs is used elsewhere for the observed rate constant for an amine
at a given pH.
in charge distribution between the ground state and transition
state of the reaction (25, 26). The Brønsted slope provides an
empirical relationship between reactivity and pK for the amine-
a
catalyzed retroaldol reaction. The relationship can be replotted,
as in Fig. 4B, to account for the concentration of neutral amine
present at the pH value of the reaction conditions, here
pH 7.5. The plot in Fig. 4B relates the observed rate constant
(
kobs) at pH 7.5 to the pKa of the amine and illustrates that
the optimal amine catalysts have pKa values similar to the pH
of the reaction conditions.
The relationship in Fig. 4B shows that the dependence of kobs
on pK is not steep, and the difference in observed rate constant
a
between an amine of pKa 10.6 and 6.8, as in RA61, is modest,
about 10-fold. Thus, an enhancement of ∼10-fold is estimated
for RA61, assuming that the enzymatic and nonenzymatic
Brønsted values are similar, and similar rate enhancements are
estimated for the other designed retroaldolases (Table 2).
in solution? In RA61, the shifted pKa appears to contribute about
5
1
0-fold out of a total of 10 -fold rate acceleration relative to the
lysine side chain in solution. To understand the remaining rate
acceleration of this most active variant, we investigated the other
design elements in RA61.
0
A
Catalytic Contribution of Designed Water Interactions. A hydrogen
bond network was included in the design calculations used to
develop the retroaldolases. This network was intended to stabi-
lize a water molecule that in turn would facilitate proton transfer
in multiple steps. In RA61, the computational design model in-
cluded hydrogen-bonding interactions between a water molecule
and Tyr78 and Ser87 (Fig. 2B). To evaluate whether the modeled
water motif contributes to the rate acceleration by RA61, we
mutated these residues and assessed the effects on catalysis.
In contrast to expectations for catalytic residues, mutation of
Tyr78 and Ser87 increased activity by 2- to 3-fold, and a double
mutation led to a 5-fold increase in activity over wild type
-
-
-
-
2
4
6
8
4
6
8
10
12
pK
a
of amine
B -1
(
Table 4). No significant change was seen in the pH-rate profile
of the double mutant relative to RA61.
The increase in activity upon loss of these designed interac-
tions indicates that either these interactions do not facilitate
reaction at the active site or that the predicted interactions
are not made. The crystal structure determined for this enzyme
does not contain bound ligand and therefore cannot be used to
evaluate the presence of the predicted water molecule. The active
site of RA61 is relatively open and solvent-exposed, so bulk
-
-
-
-
3
5
7
9
4
6
8
10
12
Table 4. Second-order rate constants for RA61 mutants at pH 7.5
a
pK of amine
Variant
ðkcat∕KMÞobs (M− s−1
1
)
Mutant∕WT
Fig. 4. (A) Relationship between the pH-independent second-order rate
constant of the fully deprotonated, neutral amine ðkmaxÞ and pKa for the
retroaldol reactions of 4-hydroxy-4-(6-methoxy-2-naphthyl)-2-butanone
catalyzed by amines listed in Table 3. The slope obtained by linear fit is
Wild type RA61
Tyr78Phe
Tyr78Ala
Ser87Ala
Ser87Gly
0.49
1.56
0.98
0.79
1.8
(1)
3.2
2.0
1.6
3.7
5.3
0
.54 ꢁ 0.03, where the error is the standard error of the slope. (B) Relation-
ship between the observed second-order rate constant at pH 7.5 (kobs) and
pKa for the same reactions.
Tyr78Phe/Ser87Ala
2.6
4
940
∣
www.pnas.org/cgi/doi/10.1073/pnas.0913638107
Lassila et al.

solvent may act to facilitate proton transfers. It is possible that
mutations of Tyr78 and Ser87 increase activity slightly by allowing
greater access to bulk solvent, although this model remains to be
tested. Regardless, we can conclude that the designed water motif
does not contribute to catalysis by RA61.
an estimate of around 500-fold for the catalytic contribution of
enzymatic binding interactions with the hydrophobic naphthyl
side chain of the substrate. A designed water hydrogen-bonding
network did not appear to make any catalytic contribution
in RA61.
Thus, our results suggest that around 5,000-fold of the rate ac-
celeration of RA61 can arise through a combination of binding
the substrate into the active site and increasing the amount of
reactive primary amine available to catalyze the reaction. The
40-fold remaining rate acceleration may arise through addi-
tional binding energy with the hydroxybutanone scaffold or the
4-hydroxy-4-methyl-2-pentanone substrate used for comparison
herein. Other elements of catalysis not directly tested herein such
as desolvation of the substrate and the reactive amine as well as
entropic effects of localizing the substrate and the amine can also
be linked to binding energies (30).
Catalytic Contribution of Hydrophobic Binding Interactions. A third
and final element included in the design calculations was binding
interactions with the hydrophobic surface of the substrate
molecule. While specific binding interactions were not explicitly
required in the design calculations, hydrophobic binding interac-
tions with substrate are favored by van der Waals and solvation
terms in the energy function (7). The design model of RA61
shows close proximity of the hydrophobic naphthyl rings with hy-
drophobic side chains (Fig. 2C). These types of interactions are
expected to stabilize the transition state bound within the active
site and would thus contribute to catalysis. In particular, binding
interactions with the transition state would be expected to in-
crease the enzymatic second-order rate constant examined here,
which corresponds to the difference in energy between the tran-
sition state and the free enzyme and substrate in solution.
To estimate the catalytic contribution of binding interactions to
the rate acceleration of RA61, we compared the reactions of
the full substrate to those of an alternative substrate that lacks
the hydrophobic naphthyl side chain, 4-hydroxy-4-methyl-2-
pentanone (Table 5). The second-order rate constants for the
nonenzymatic reactions differ by 4- to 5-fold from those of the
full substrate, suggesting that intrinsic differences in reactivity
are small. In contrast, the rate constants for enzyme-catalyzed re-
Binding substrate and providing reactive chemical groups
represent fundamental strategies of natural enzymes. However,
natural enzymes offer highly specific interactions that provide
optimal stabilization in the transition state of the reaction. Two
observations suggest that the designed retroladolases lack the
precision and specificity in their interactions with substrate that
are seen in natural enzymes. The observed tight inhibition by the
naphthaldehyde product indicates that the enzymes favorably ac-
commodate the naphthyl rings in a position that is translated
toward the enzyme by 2–3 Å (SI Appendix). Furthermore, the sub-
strate 4-hydroxy-4-(6-methoxy-2-naphthyl)-2-butanone has
a
chiral center but we see no evidence of significant stereospecifi-
city in RA61 (SI Appendix), suggesting that this enzyme has
little preference for specific orientations of the naphthyl rings
with respect to iminium bond. These results suggest that RA61
achieves its rate acceleration without taking advantage of the
precise positioning and specific interactions that help natural en-
actions of the larger and smaller substrates differ by more than
3
1
0 -fold. The differential reactivity between large and small
substrates in RA61 compared to the amine-catalyzed reactions
provides an estimate of around 500-fold for the contribution
of the enzyme’s interaction with the naphthyl side chain. As
the smaller substrate has two additional methyl groups that could
make hydrophobic interactions in the active site and both
substrates retain the hydroxybutanone scaffold, the full catalytic
contribution from binding interactions is expected to be some-
what larger.
‡
zymes to achieve selective stabilization of transition states. Thus,
an important next step in the development of computational
enzyme design methodology may be a careful assessment of
the extent to which computational design can yield precise place-
ment of side chains and ligands on an atomic level.
Our results demonstrate that the retroaldolases utilize funda-
mental elements of catalysis—binding interactions and stabiliza-
tion of charge through reactive chemical groups—in achieving
rate acceleration. The observed lack of a catalytic contri-
bution from the designed water interaction with Tyr78 and
Ser87 illustrates how designed elements may not always work
as anticipated based on structural models. Thus, quantitative, ex-
perimental dissection of the catalytic contributions of designed
enzymes provides necessary feedback for future improvement.
By critically evaluating catalytic contributions in designed en-
zymes, we stand to gain a deeper insight into both catalytic me-
chanisms and future possibilities for engineering practical and
beneficial enzymes.
Conclusions and Implications. Computational enzyme design meth-
ods represent a promising tool for engineering new enzymes. Yet
the success of these methods depends on their predictive power,
so it remains crucial to experimentally evaluate computationally
designed enzymes and to determine the strengths and limitations
of computational design procedures.
The results presented here illustrate that the computationally
5
designed retroaldolases provide around 10 -fold rate accelera-
tion beyond catalysis by the lysine side chain alone in solution.
Studies of pH-rate profiles and a Brønsted analysis of nonenzy-
matic reactions suggest that about 10-fold of this rate accelera-
tion arises from shifted pK values for the catalytic lysines in the
a
enzymes. For RA61, the most active of the retroaldolases,
comparison of reactivities with a smaller aldol substrate provides
Materials and Methods
Materials. The compounds 4-hydroxy-4-methyl-2-pentanone, 6-methoxy-2-
naphthaldehyde, 2-aminoacetonitrile, 2,2,2-trifluoroethylamine, propargyla-
mine, 2-bromoethylamine, 2-chloroethylamine, ethanolamine, methylamine,
and butylamine were obtained from Sigma-Aldrich/Fluka at the highest
available purity.
Table 5. Observed second-order rate constants at pH 7.5 for
retroaldol cleavage of 4-hydroxy-4-methyl-2-pentanone and full
substrate
The fluorogenic substrate 4-hydroxy-4-(6-methoxy-2-naphthyl)-2-buta-
none was synthesized from acetone and 6-methoxy-2-naphthaldehyde as
previously described (9) and further purified by silica gel chromatography
in hexanes/ethyl acetate. Substrate quantitation was checked by comparison
O
OH
O
OH
1
of H NMR signal to an internal standard.
O
Proteins were obtained using the previously described procedures (7),
including an autoinduction expression protocol followed by purification
1
kobs (M−1 s−1
Catalyst
kobs (M− s−1
)
)
Ratio*
−
−
5
6
−6
Propargylamine
Trifluoroethylamine
RA61
1.7 × 10
7.0 × 10
0.49
3.3 × 10
1.8 × 10
5
4
−6
−4
‡
A recent molecular dynamics study concluded that a related retroaldolase not studied in
this work, RA22, lacked sufficient constraining interactions to maintain effective catalytic
geometry during reaction (17).
2.3 × 10
2,100
*
kobs for the full substrate divided by kobs for the smaller substrate.
Lassila et al.
PNAS
∣
March 16, 2010
∣
vol. 107
∣
no. 11
∣
4941

by His6-tag affinity to Ni-NTA resin (QIAGEN). Proteins were extensively
dialyzed into a storage buffer of 25 mM sodium phosphate, pH 7.5, 100 mM
NaCl or 50 mM sodium phosphate pH 7.5, 300 mM NaCl. Protein concentra-
tion was determined by absorbance at 280 nm in 6 M guanidinium HCl.
Mutants of RA61 were constructed using inverse PCR mutagenesis.
D2O for lock signal. Reactions were kept in a temperature-controlled water
bath between data collection time points. The acetone peak was integrated
and compared to an internal standard, sodium 2,2,3,3-tetradeutero-
3
-trimethylsilylpropionate and a standard curve of acetone integration
under identical buffer conditions and delay times. All reactions were first-
order with respect to both enzyme or amine and substrate.
Cosolvent and Buffer Conditions. The low solubility of substrate 4-hydroxy-
4
-(6-methoxy-2-naphthyl)-2-butanone in water dictates that cosolvents be in-
cluded in kinetic assays. Initial investigations of the designed proteins used
.7% acetonitrile in assays (7). However, under these conditions, we observed
pH-Rate Profiles. Second-order rate constants were measured at different
pH values using the buffer conditions described above (50 mM sodium
carbonate, phosphate, or acetate, 300 mM NaCl, and 5% DMSO). Control ex-
periments with enzyme incubated at pH extremes and subsequently assayed
at pH 7.5 indicated that the protein was not irreversibly denatured during
the course of the experiments. Data were fit to the following relationship,
derived from the kinetic scheme shown in Fig. 3:
2
substantial departure from linearity in plots of nonenzymatic rate versus
substrate concentration suggestive of a loss of substrate through insolubility.
We therefore tested other cosolvent conditions. Several prior studies of anti-
body and peptide catalysis of the retroaldol reaction of the same substrate
used 5% cosolvent (15). In 5% DMSO, substrate concentrations of ∼500 μM
could be reached before significant curvature in nonenzymatic rates was
evident.
In testing conditions for pH-rate profiles, we found that many organic
buffers either inhibited the enzymatic reaction or accelerated the nonenzy-
matic reactions. Enzymatic and nonenzymatic reactions were not affected by
varying the concentration of acetate, phosphate, and carbonate, so these
buffers were used in all assays. Reaction pH values were checked by tests
of mock reactions with a pH meter.
ðkcat∕KMÞmax
:
pKa−pH
ðkcat∕KMÞobs
¼
1
þ 10
Nonenzymatic, Amine-Catalyzed Reactions. Second-order rate constants for
amine-catalyzed reactions were measured under conditions in which rates
remained linear with both substrate and amine concentration over at least
1
0-fold ranges. Spot checks were performed to ensure that no significant
third-order effects were observed with buffer components. pH values of
aqueous amine stock solutions were adjusted to match the reaction pH be-
fore addition to buffer solutions, and pH values of the final reaction condi-
tions were tested in mock reactions. Amine-catalyzed rate constants were
measured at several different pH values and average values for maximal rate
constants were determined using the above equation. While the presence of
DMSO cosolvent in reaction conditions has effects on both pH and pKa
values, at 5% DMSO, these effects will be small (<0.5 units) and will not
significantly affect the correlation (31–34).
Kinetic Assays. Second-order rate constants were determined under condi-
tions in which rates remained linear with enzyme and substrate concentra-
tion over at least 10-fold ranges. Standard assay conditions were 25 °C,
5
0 mM sodium phosphate, 300 mM NaCl, and 5% DMSO at pH 7.5. Stocks
of the 4-hydroxy-4-(6-methoxy-2-naphthyl)-2-butanone substrate were made
in DMSO and stored at −20 °C. Reaction kinetics were monitored by following
fluorescence of the product 6-methoxy-2-naphthaldehyde on a FluoroLog
3
spectrofluorometer (HORIBA Jobin Yvon) with excitation at 330 nm and
emission at 452 nm. Product concentration was calibrated by measurement
of a standard curve of 6-methoxy-2-naphthaldehyde fluorescence. Quartz
cuvettes were used for all assays, and evaporation was controlled by use
of excess reaction volume and tight-sealing caps.
Reactions of 4-hydroxy-4-methyl-2-pentanone were monitored by proton
NMR on a 600 MHz system with PRESAT water suppression. Reaction condi-
tions were 25 °C, 50 mM sodium phosphate, pH 7.5, 300 mM NaCl, and 5–8%
ACKNOWLEDGMENTS. We thank Jason Schwans, Stephen Lynch, Eric Althoff,
and members of the Baker and Herschlag labs for helpful discussions.
This work was supported by a National Institutes of Health postdoctoral
fellowship to J.K.L. (F32 GM080865), by NIH Grant GM64798 to D.H., and
by the DARPA Protein Design Processes program.
1
2
3
4
5
. Dahiyat BI, Mayo SL (1997) De novo protein design: Fully automated sequence
selection. Science 278:82–87.
. Boas FE, Harbury PB (2007) Potential energy functions for protein design. Curr Opin
Struct Biol 17:199–204.
. Suarez M, Jaramillo A (2009) Challenges in the computational design of proteins.
J R Soc Interface 6:S477–S491.
. Bolon DN, Mayo SL (2001) Enzyme-like proteins by computational design. Proc Natl
Acad Sci USA 98:14274–14279.
. Lassila JK, Privett HK, Allen BD, Mayo SL (2006) Combinatorial methods for
small-molecule placement in computational enzyme design. Proc Natl Acad Sci USA
19. Guthrie JP (1972) Acetopyruvate. Enamine formation with aminoacetonitrile. Models
for acetoacetate decarboxylase. J Am Chem Soc 94:7024–7029.
20. Richard JP (1984) Acid-base catalysis of the elimination and isomerization reactions of
triose phosphates. J Am Chem Soc 106:4926–4936.
2
1. Zeng B, Pollack RM (1991) Microscopic rate constants for the acetate ion catalyzed
isomerization of 5-androstene-3,17-dione to 4-androstene-3,17-dione: A model for
steroid isomerase. J Am Chem Soc 113:3838–3842.
2
2
2
2. Schmidt DE, Westheimer FH (1971) pK of the lysine amino group at the active site of
acetoacetate decarboxylase. Biochemistry 10:1249–1253.
3. Barbas CF, et al. (1997) Immune versus natural selection: Antibody aldolases with
enzymic rates but broader scope. Science 278:2085–2092.
4. Ho M-C, Menetret J-F, Tsuruta H, Allen KN (2009) The origin of the electrostatic
perturbation in acetoacetate decarboxylase. Nature 459:393–397.
5. Jencks WP (1969) Catalysis in chemistry and enzymology (Dover, New York).
6. Williams A (1992) Advances in Physical Organic Chemistry, (Academic Press, London),
Vol 27, pp 1–55.
1
03:16710–16715.
6
7
8
9
. Zanghellini A, et al. (2006) New algorithms and an in silico benchmark for computa-
tional enzyme design. Protein Sci 15:2785–2794.
. Jiang L, et al. (2008) De novo computational design of retro-aldol enzymes. Science
2
2
3
19:1387–1391.
. Rothlisberger D, et al. (2008) Kemp elimination catalysts by computational enzyme
design. Nature 453:190–195.
. List B, Barbas CF, Lerner RA (1998) Aldol sensors for the rapid generation of
fluorescence by antibody catalysis. Proc Natl Acad Sci USA 95:15351–15355.
27. Buncel E, Um IH, Hoz S (1989) Solvent-independent transition-state structure for
acyl-transfer reactions. A novel strategy for construction of a Brønsted correlation.
J Am Chem Soc 111:971–975.
1
1
1
0. Gefflaut T, Blonski C, Perie J, Willson M (1995) Class I aldolases: Substrate specificity,
mechanism, inhibitors, and structural aspects. Prog Biophys Mol Biol 63:301–340.
1. Johnnson K, Alleman RK, Widmer H, Benner SA (1993) Synthesis, structure, and activity
of artificial, rationally designed catalytic polypeptides. Nature 365:530–532.
28. Tarkka RM, Park WKC, Liu P, Buncel E, Hoz
transition state-structures. Part III. Sulfonyl transfer reactions. J Chem Soc Perkin Trans
2439–2444.
S (1994) Solvent independent
2
2. Reymond J-L, Chen Y (1995) Catalytic, enantioselective aldol reaction using antibodies
against a quaternary ammonium ion with a primary amine cofactor. Tetrahedron Lett
2
3
3
9. Hine J, Via FA (1972) Kinetics of the formation of imines from isobutyraldehyde and
primary aliphatic amines with polar substituents. J Am Chem Soc 94:190–194.
0. Kraut DA, Carroll KS, Herschlag D (2003) Challenges in enzyme mechanism and
energetics. Annu Rev Biochem 72:517–571.
1. Mukerjee P, Ostrow JD (1998) Effects of added dimethylsulfoxide on pKa values of
uncharged organic acids and pH values of aqueous buffers. Tetrahedron Lett
3
6:2575–2578.
1
1
1
1
1
1
3. Wagner J, Lerner RA, Barbas CF (1995) Efficient aldolase catalytic antibodies that use
the enamine mechanism of natural enzymes. Science 270:1797–1800.
4. Hoffmann T, et al. (1998) Aldolase antibodies of remarkable scope. J Am Chem Soc
1
20:2768–2779.
3
9:423–426.
5. Tanaka F (2005) Development of protein, peptide, and small molecule catalysts using
catalysis-based selection strategies. Chem Rec 5:276–285.
6. Muller MM, Windsor MA, Pomerantz WC, Gellman SH, Hilvert D (2009) A rationally
designed aldolase foldamer. Angew Chem Int Ed 48:922–925.
7. Ruscio JZ, Kohn JE, Ball KA, Head-Gordon T (2009) The influence of protein dynamics
on the success of computational enzyme design. J Am Chem Soc 131:14111–14115.
8. March J (1992) Advanced Organic Chemistry: Reactions, Mechanisms, and Structure
3
3
2. Gowland JA, Schmid GH (1969) Two linear correlations of pKa vs. solvent composition.
Can J Chem 47:2953–2958.
3. Glover DJ (1965) Equilibria in Solution. II. Evaluation of pK and solvation numbers.
J Am Chem Soc 87:5279–5283.
34. Gutbezahl B, Grunwald E (1953) The effect of solvent on equilibrium and rate con-
stants. II. The measurement and correlation of acid dissociation constants of anilium
and ammonium salts in the system ethanol-water. J Am Chem Soc 75:559–565.
(
Wiley, New York), 4th Ed, pp 896–897.
4
942
∣
www.pnas.org/cgi/doi/10.1073/pnas.0913638107
Lassila et al.