Rational control of enzymatic enantioselectivity through solvation thermodynamics

Article abstract of DOI:10.1021/ja961394q

The enantioselectivity of cross-linked crystals of γ-chymotrypsin in the transesterification of the medicinally important compound methyl 3-hydroxy-2-phenylpropionate (1) with propanol has been examined in a variety of organic solvents. The (k(cat)/K(M))(

Full text of DOI:10.1021/ja961394q

J. Am. Chem. Soc. 1996, 118, 10365-10370
10365
Rational Control of Enzymatic Enantioselectivity through
Solvation Thermodynamics
Charles R. Wescott,† Hidetaka Noritomi,§ and Alexander M. Klibanov*
Contribution from the Department of Chemistry, Massachusetts Institute of Technology,
Cambridge, Massachusetts 02139
ReceiVed April 29, 1996X
Abstract: The enantioselectivity of cross-linked crystals of γ-chymotrypsin in the transesterification of the medicinally
important compound methyl 3-hydroxy-2-phenylpropionate (1) with propanol has been examined in a variety of
organic solvents. The (kcat/KM)S/kcat/KM)R ratio in this enzymatic process can be forced to span a 20-fold range
simply by switching from one solvent to another; in fact, while the enzyme strongly prefers the S-enantiomer of 1
in some solvents, the R-antipode is more reactive in others. These striking observations are quantitatively rationalized
by accounting for the energetics of desolvation of S-1 and R-1 in the enzyme-bound transition states. In order to
accomplish this, explicit rules have been established for the modeling and thermodynamic quantification of the partially
desolvated substrate’s transition state moieties.
Introduction
to-water partition coefficients, which were either measured
experimentally4a or calculated4b using the UNIFAC5 computer
algorithm. However, since variations in the partition coefficients
arose solely from chemical differences in the substrates, this
initial model could not account for solvent-induced changes in
selectivities involving chemically identical compounds (as in
the case of prochiral selectivity or enantioselectivity).
Recently, we have further developed this methodology to
explain the solvent effect on enzymatic prochiral selectivity by
accounting for the role of partial desolvation of the enzyme-
substrate transition state.6 Transition states leading to the
opposite enantiomers of a product may be desolvated to different
extents, resulting in a nonzero differential desolvation energy
even though the substrates are chemically identical. Imple-
mentation of this model to predict the solvent dependence of
prochiral selectivity employed molecular modeling to determine
the desolvated portions of the substrate in the pro-R and pro-S
transition states, followed by the calculation of the thermo-
dynamic activity coefficients of these moieties. However, due
to conceptual difficulties in modeling partially desolvated
surfaces of the transition states, application of this methodology
was restricted to substrates specifically designed to interact with
the enzyme in the desired manner.
The exquisite stereoselectivity of enzymes is their most
valuable attribute to the organic chemist.1 Ironically, this same
trait also limits the generality of enzymatic synthesis, because
enzymes that catalyze the reaction of interest with the desired
stereochemistry are not always available. Nonaqueous enzy-
mology,2 and especially the discovery that enzymatic selectivity
can be markedly altered by the reaction medium,3 thus greatly
enhances the utility of enzyme-catalyzed syntheses. Our
objective is to elucidate the mechanisms by which the solvent
influences enzymatic stereoselectivity, and thereby enable the
rational design of stereoselective systems on the basis of
physicochemical properties of the substrate and solvent, as well
as of enzyme structure.
Our pursuit of this goal first led to a thermodynamic model
which explained the solvent dependence of the substrate
specificity of subtilisin Carlsberg on the basis of the free energies
of desolvation of the substrates.4 The differential desolvation
energy between two substrates was derived from their solvent-
* To whom correspondence should be addressed.
† An NIH Biotechnology Predoctoral Trainee.
§ Permanent address: Department of Industrial Chemistry, Tokyo
Metropolitan University, Minami-ohsawa, Hachioji, Tokyo 192-03, Japan.
X Abstract published in AdVance ACS Abstracts, October 15, 1996.
(1) (a) Simon, H.; Bader, J.; Gunther, H.; Neumann, S.; Thanos, J. Angew.
Chem., Int. Ed. Engl. 1985, 24, 539. (b) Yamada, H.; Shimizu, S. Angew.
Chem., Int. Ed. Engl. 1988, 27, 622. (c) Jones, J. B. Tetrahedron 1986,
42, 3351. (d) Faber, K. Biotransformations in Organic Chemistry; Springer-
Verlag: Berlin, 1992. (e) Poppe, L.; Novak, L. SelectiVe Biocatalysis; VCH
Publishers: New York, 1992. (f) Sheldon, R. A. Chirotechnology:
Industrial Synthesis of Optically ActiVe Compounds; M. Dekker: New York,
1993. (g) Margolin, A. L. Enzyme Microb. Technol. 1993, 15, 266. (h)
Wong, C.-H.; Whitesides, G. M. Enzymes in Synthetic Organic Chemistry;
Pergamon: Oxford, 1994. (i) Roberts, S. M.; Turner, N. J.; Willetts, A. J.;
Turner, M. K. Introduction to Biocatalysis Using Enzymes and Microorgan-
isms; Cambridge University Press: New York, 1995. (j) Drauz, K.;
Waldmann, H. Enzyme Catalysis in Organic Synthesis; VCH Publishers:
New York, 1995.
In the present work, a set of rules is derived and validated
for the modeling of the partially desolvated transition state
(4) (a) Wescott, C. R.; Klibanov, A. M. J. Am. Chem. Soc. 1993, 115,
1629. (b) Wescott, C. R.; Klibanov, A. M. J. Am. Chem. Soc. 1993, 115,
10362.
(5) UNIFAC is a computational method for the estimation of Rault’s
law activity coefficients. Being a group contribution method, it can calculate
activity coefficients in systems for which there is no experimental data by
assessing the individual contribution of each chemical group which makes
up the system. Use of this method requires three types of parameters for
each group in the system: the group’s surface area, the volume of the group,
and empirically determined parameters which reflect the free energy of
interaction between a given group and every other group in the system. (a)
Fredenslund, A.; Gmehling, J.; Rasmussen, P. Vapor-Liquid Equilibria
Using UNIFAC; Elsevier: New York, 1977. (b) Steen, S.-J.; Ba¨rbel, K.;
Gmehling, J.; Rasmussen, P. Ind. Eng. Chem. Process Des. DeV. 1979, 18,
714. (c) Rasmussen, P.; Fredenslund, A. Ind. Eng. Chem. Process Des.
DeV. 1982, 21, 118. (d) Macedo, E. A.; Weidlich, U.; Gmehling, J.;
Rasmussen, P. Ind. Eng. Chem. Process Des. DeV. 1983, 22, 678. (e) Teigs,
D.; Gmehling, J.; Rasmussen, P.; Fredenslund, A. Ind. Eng. Chem. Res.
1987, 26, 159. (f) Hansen, H. K.; Rasmussen, P.; Schiller, M.; Gmehling,
J. Ind. Eng. Chem. Res. 1991, 30, 2355.
(2) (a) Klibanov, A. M. Trends Biochem. Sci. 1989, 14, 141. (b) Chen,
C.-S.; Sih, C. J. Angew. Chem., Int. Ed. Engl. 1989, 28, 695. (c) Dordick,
J. S. Enzyme Microb. Technol. 1989, 11, 194. (d) Klibanov, A. M. Acc.
Chem. Res. 1990, 23, 114. (e) Gupta, M. N. Eur. J. Biochem. 1992, 203,
25. (f) Faber, K.; Riva, S. Synthesis 1992, 895. (g) Halling, P. J. Enzyme
Microb. Technol. 1994, 16, 178. (h) Koskinen, A. M. P.; Klibanov, A.
M., Eds. Enzymatic Reactions in Organic Media; Blackie: London, 1996.
(3) For reviews, see: (a) Wescott, C. R.; Klibanov, A. M. Biochim.
Biophys. Acta 1994, 1206, 1. (b) Carrea, G.; Ottolina, G.; Riva, S. Trends
Biotechnol. 1995, 13, 63.
(6) Ke, T.; Wescott, C. R.; Klibanov, A. M. J. Am. Chem. Soc. 1996,
118, 3366.
S0002-7863(96)01394-7 CCC: $12.00 © 1996 American Chemical Society

10366 J. Am. Chem. Soc., Vol. 118, No. 43, 1996
Wescott et al.
Scheme 1
moieties. Moreover, we expand our predictive repertoire to the
solvent dependence of a new, key type of stereoselectivity,
namely enantioselectivity. Consequently, we have been able
to successfully apply our structure-based thermodynamic treat-
ment to the quantitative analysis of the chiral resolution of a
nondesigned, medicinally significant compound, methyl 3-hy-
droxy-2-phenylpropionate (1).
Results and Discussion
Figure 1. Dependence of the activity of γ-chymotrypsin cross-linked
crystals on the fraction of active enzyme in the crystal (f). Activity is
measured as the rate of enzymatic transesterification of the S- (b) or
R- (9) enantiomers of 1 with propanol in cyclohexane. f is controlled
by co-crystallizing native and inactivated (by diisopropyl fluorophos-
phate) γ-chymotrypsin in varying proportions (see Methods for details).
To test our methodology with enzymatic enantioselectivity,
we have examined the transesterification of racemic 1 with
propanol catalyzed by cross-linked crystals7 (CLCs) of γ-chy-
motrypsin (Scheme 1). Various esters of the acid moiety of 1
(3-hydroxy-2-phenylpropionic acid, 2) are potent anticholin-
ergics, including atropine, hyoscyamine, and scopolamine.8
While most synthetic methods produce racemates of these drugs,
only the S-antipodes are pharmaceutically active.8 γ-Chymo-
trypsin CLCs are employed as the catalyst herein because the
crystalline form of the enzyme has been found to retain its native
conformation in organic solvents,9 thus allowing the use of
structure-based molecular modeling.
Because enzyme CLCs (as well as nearly all other enzyme
preparations) are insoluble in organic solvents, the transesteri-
fication in Scheme 1 is catalyzed in a heterogeneous system
and thus is susceptible to rate limitation by diffusion of the
substrate into the solid catalyst particle. To ensure that the initial
velocities measured reflect the true kinetic constants of the
enzyme, and not the mass transfer rates of the substrate through
the crystals, enzymatic activity was examined as a function of
the loading of the biocatalyst particles.4a,7f,11 To this end, active
γ-chymotrypsin was co-crystallized with varying amounts of
this enzyme inactivated with diisopropyl fluorophosphate. In
the absence of diffusional limitations, a plot of catalytic activity
Table 1. Solvent Dependence of the Enantioselectivity of
γ-Chymotrypsin Cross-Linked Crystals for the Transesterification of
1 with Propanol
(kcat/KM)S/
(kcat/KM)R
product ee, %b
(preferred enantiomer)
a
solvent
cyclohexane
octane
hexane
13
8.8
8.0
5.6
2.4
1.8
1.5
0.91
0.80
0.74
0.73
0.64
85 (S)
79 (S)
77 (S)
69 (S)
40 (S)
28 (S)
20 (S)
4.6 (R)
11 (R)
15 (R)
15 (R)
21 (R)
toluene
isopropyl acetate
tetrahydrofuran
tert-butyl acetate
tert-butyl alcohol
tert-amyl alcohol
dioxane
propanol
acetone
a See Methods for details on the measurement of (kcat/KM)S/(kcat/KM)R.
b Enantiomeric excesses (ee) were calculated from the enantioselec-
tivities for a 5% conversion as described by Chen et al.12
vs the fraction of active enzyme in the crystal should yield a
straight line which passes through the origin.11 If, however,
the mass transfer contribution to the reaction rate is not
negligible, a convex dependence should be observed,11 because
incremental increases in the enzyme loading do not produce
similar amplifications in the catalytic efficiency of the CLC.
Such a plot (Figure 1) for the transesterification of both
enantiomers of 1 in cyclohexane reveals a linear dependence
between catalytic activity and the fraction of active enzyme in
the crystal, thus ruling out the possibility that the measured
reactions are affected by mass transfer of the substrate.
To explore the effect of the solvent on the kinetic resolution
of 1, the enantioselectivity of γ-chymotrypsin CLCs for the
reaction depicted in Scheme 1 was measured in a variety of
organic solvents. The enantiomeric excess (ee) at 5% conver-
sion was subsequently calculated12 to quantify the efficiency
of the resolution in each solvent (Table 1). Inspection of Table
1 reveals that the enantioselectivity, expressed as (kcat/KM)S/
(kcat/KM)R, can be forced to span a 20-fold range simply by
switching from one organic solvent to another under otherwise
identical conditions. Perhaps even more striking is the fact that
the enantioselectivity can actually be reversed through the choice
(7) Such cross-linked crystals of various enzymes have been employed
as robust catalysts in synthetic transformations: (a) St. Clair, N. L.; Navia,
M. A. J. Am. Chem. Soc. 1992, 114, 7314. (b) Sobolov, S. B.; Bartoszko-
Malik, A.; Oeschger, T. R.; Montelbano, M. M. Tetrahedron Lett. 1994,
35, 7751. (c) Persichetti, R. A.; St. Clair, N. L.; Griffith, J. P.; Navia, M.
A.; Margolin, A. L. J. Am. Chem. Soc. 1995, 117, 2732. (d) Lalonde, J. J.;
Govardhan, C.; Khalaf, N.; Martinez, A. G.; Visuri, K.; Margolin, A. L. J.
Am. Chem. Soc. 1995, 117, 6845. (e) Sobolov, S. B.; Leonida, M. D.;
Bartoszko-Malik, A.; Veivodov, K. I.; McKinney, F.; Kim, J.; Fry, A. J. J.
Org. Chem. 1996, 61, 2125. (f) Schmitke, J. L.; Wescott, C. R.; Klibanov,
A. M. J. Am. Chem. Soc. 1996, 118, 3360.
(8) Reynolds, J. E. F., Ed. Martindale, the Extra Pharmacopoeia, 28th
ed.; Pharmaceutical Press: London, 1982; p 304.
(9) It is reasoned that if the structure of uncross-linked chymotrypsin is
nearly the same in water and in hexane,10 then it also should be the same
in other organic solvents because the differences in physicochemical
properties among them are less than those between water and hexane.
Furthermore, cross-linking followed by placement in acetonitrile does not
affect the structure of two other serine proteases, subtilisin Carlsberg
(Fitzpatrick, P. A.; Steinmetz, A. C. U.; Ringe, D.; Klibanov, A. M. Proc.
Natl. Acad. Sci. U.S.A. 1993, 90, 8653. Fitzpatrick, P. A.; Ringe, D.;
Klibanov, A. M. Biochem. Biophys. Res. Commun. 1994, 198, 675) and
porcine pancreatic elastase (Allen, K. N.; Bellamacina, C. R.; Ding, X.;
Jeffery, C. J.; Mattos, C.; Petsko, G. A.; Ringe, D. J. Phys. Chem. 1996,
100, 2605).
(10) (a) Yennawar, N. H.; Yennawar, H. P.; Farber, G. K. Biochemistry
1994, 33, 7326. (b) Yennawar, H. P.; Yennawar, N. H.; Farber, G. K. J.
Am. Chem. Soc. 1995, 117, 577.
(11) Boudart, M.; Burwell, R. L. In Techniques in Chemistry, 3rd ed.;
Lewis, E. S., Ed.; Wiley: New York, 1973; Vol. 6, Chapter 12.
(12) Chen, C.-S.; Fujimoto, Y.; Girdaukas, G.; Sih, C. J. J. Am. Chem.
Soc. 1982, 104, 7294.

SolVent Control of Enzymatic EnantioselectiVity
J. Am. Chem. Soc., Vol. 118, No. 43, 1996 10367
Figure 2. Molecular models of S- (A) or R- (B) 1 in the transition state for acylation of γ-chymotrypsin. See Methods for details on creation of
the models.
of the solvent. For instance, in cyclohexane the enzyme
preferentially transesterifies the S-enantiomer of 1, while in
acetone the R-antipode is preferred.
trypsin’s S1 binding pocket, while the phenyl group extends
away from the enzyme toward the solvent. Figure 2B depicts
the opposite situation for the R transition state: the aryl moiety
is buried in the active center of the enzyme, while the hydroxyl
group is oriented toward the solvent. The solvated surface areas
for the transition states, calculated using the method of Con-
nolly,14 are displayed as dot surfaces in Figure 3. One can see
that, for example, in the S transition state the hydroxyl group is
desolvated, while the phenyl group is not. In contrast, the
surface for the R transition state indicates the inverse desolvation
pattern for these two groups.
It has been shown that the solvent exerts its effect on the
selectivity of enzyme-catalyzed processes through the dif-
ferential Gibbs free energy of desolvation for the transition states
of the reactions.4,6 A quantitative analysis6,13 of such a system,
when applied to enantioselectivity herein, affords an equation
relating the latter in a given solvent to the thermodynamic
activity coefficients of the desolvated moieties of the substrate
transition states (γ′) in that solvent:
With the desolvated portions of the transitions states ascer-
tained, the next step in the calculation of γ′ is the construction
of molecular fragments, based on UNIFAC groups,5 which
approximate the desolvated portions of the substrates in the
transition states. To this end, the enantiomers of 1 have been
modeled in terms of the smallest possible UNIFAC groups, and
the percent of desolvation of each such group is tabulated in
Table 2. Groups are then included in the molecular fragment
for a given substrate enantiomer if they are at least 50%
desolvated. Groups desolvated to a lesser extent are considered
solvated and thus not part of the desolvated substrate moiety.
According to these rules, the desolvated portion of the S
transition state is represented by one hydroxyl group, one aryl
carbon, two aryl methine groups, one carbonyl group, and one
aliphatic methine group. Similarly, the corresponding R mo-
lecular fragment consists of one aryl carbon, three aryl methine
groups, one carbonyl group, and one aliphatic methine group.
(kcat/KM)S
(kcat/KM)R
γ′S
log
) log
+ constant
(1)
( )
[
]
γ′R
where subscripts S and R indicate parameters related to the
respective enantiomers of the substrate, and “constant” contains
kinetic and thermodynamic terms in an arbitrarily selected
reaction medium.6
Unlike the situation for substrate specificity, where solvent-
dependent variation in the activity coefficient ratio for the two
substrates is primarily driven by chemical differences between
them,4 γ′ for enantioselectivity differs for each reaction pathway
only due to differences in transition state solvation. In this work,
we calculate γ′ for both the R and S transition states using a
three-step procedure. First, the desolvated portion of each
enantiomer of the substrate in the transition state is determined
using molecular modeling based on the crystal structure of the
enzyme. Second, this desolvated moiety is approximated in
terms of individual UNIFAC groups. Finally, the thermo-
dynamic activity coefficient of this desolvated fragment is
calculated using the UNIFAC group contribution method and
then equated to γ′. According to eq 1, knowing only the ratio
of γ′R and γ′S for a series of solvents, it should be possible to
explain the observed solvent dependence of enantioselectivity.
As a first step in the calculation of γ′, molecular models have
been constructed for the S and R transition states for the
acylation of γ-chymotrypsin by 1 (see Methods for details).
Examination of the S transition state model (Figure 2A) reveals
that the hydroxyl group of the substrate is buried in chymo-
Finally, the activity coefficients for the S and R model
fragments are calculated using UNIFAC and equated with the
activity coefficients of the desolvated portions of the corre-
sponding transition states, γ′S and γ′R, respectively.
Equation 1 predicts that a double logarithmic plot of
enantioselectivity vs the γ′ ratio should be linear, with a slope
of unity. Such a plot, presented in Figure 4A, does indeed
follow the expected dependence: a least-squares fit to a linear
model with a slope of 1 yields a correlation coefficient of 0.87.
Furthermore, when the data are plotted in linear coordinates
(Figure 4B), linear regression yields a correlation coefficient
of 0.93. Thus eq 1 correctly predicts the solvent dependence
of enzymatic enantioselectivity in a near quantitative fashion.
(13) This analysis assumes the absence of specific enzyme-solvent
interactions; such interactions have been deemed to be negligible under
similar circumstances.4,6,7f
(14) Connolly, M. L. Science 1983, 221, 709.

10368 J. Am. Chem. Soc., Vol. 118, No. 43, 1996
Wescott et al.
Figure 3. Solvent-accessible surface areas of 1 in the S (left) and R (right) transition states with γ-chymotrypsin. See Methods for details.
Table 2. Percent of Desolvation of Component Groups for S- or
R-Enantiomers of the Transition States for the Acylation of
γ-Chymotrypsin by 1
yields an unchanged correlation coefficient of 0.87. These
results lead to the conclusion that, for the reaction in Scheme
1, the solvent dependence of the γ′ ratio is quite insensitive to
those moieties that happen to be partially desolvated and is
determined only by those that happen to be essentially desol-
vated in the transition state (Table 2).
desolvation (%)b
groupa
S
R
hydroxyl
aryl C 1
100
100
86
0
11
42
67
100
49
100
13
11
100
35
100
54
In order to rationalize why the groups partially desolvated in
our case do not appreciably affect the solvent dependence of
the γ′ ratio, we have calculated the activity coefficients of all
the individual component groups. As seen in Table 3, the
activity coefficients of the partially desolvated groups vary
relatively little throughout the series of solvents tested (0.26,
0.67, and 0.44 for the aryl CH, methyl, and methylene groups,
respectively). In contrast, two of the groups for which the extent
of desolvation is clearly defined, namely the hydroxyl and
carbonyl groups, exhibit much larger respective activity coef-
ficient changes of 8.0 and 2.9. Although the calculations of
activity coefficients for individual groups are inherently qualita-
tive,15 they suggest that the solvent dependence of the γ′ ratio
is dominated by the interaction of the hydroxyl and carbonyl
groups with the organic solvents. Consistent with this conclu-
sion is the finding that γ′ ratios calculated by including the
hydroxyl in, or excluding it from, both transition states produce
a correlation coefficient below 0.1 when the data are plotted as
in Figure 4A. To generalize, different types of groups impact
the free energy of desolvation of the transition states to varying
extents. Therefore, in choosing a biocatalyst for the resolution
of a chiral compound, one should seek an enzyme with an active
center that maximizes the difference in desolvation of the
“impactful” groups (such as OH) between the two enantiomers.
This optimization of differential desolvation of the impactful
groups can be performed at the expense of the nonimpactful
groups without consequence.
aryl CH 2
aryl CH 3
aryl CH 4
aryl CH 5
aryl CH 6
carbonyl
methylene
methine
33
100
100
36
100
35
methyl
a The aryl units make up the phenyl group of 1. b The percent of
desolvation of a group is calculated as [1 - (AB/AF)] × 100%, where
AB is the solvent-accessible surface area of the substrate group in the
enzyme-bound transition state (Figure 3), and AF is the solvent-
accessible surface area of the same substrate group in the free transition
state.
The methodology employed above is surprisingly effective
despite the fact that, while most substrate groups are in reality
only partially desolvated in the transition state (Table 2), for
the construction of the model fragments all groups are ap-
proximated as either 100% or 0% desolvated. Why does this
approach work even though a continuous range of desolvated
surface areas are approximated by a simple “all-or-nothing”
model? To answer this question, we have examined how the
method of treatment of partially desolvated groups (herein for
simplicity defined as groups which are 20% to 80% desolvated)
affects the predictive performance of our model. One extreme
alternative to the partially desolvated groups is to include all
of them in the model fragments. When these model fragments
are used to calculate the activity coefficients of the desolvated
portions of the S and R transition states, and the data are plotted
as in Figure 4A, the resultant correlation coefficient (0.89) has
been found to be virtually unaffected. A second extreme
recourse to the treatment of the partially desolvated groups is
to exclude all of them from the model fragments. Once again,
fitting these data to a linear dependence with a slope of unity
An insight into the nature of the solvent-solute interactions
can also be gained from Table 3. The activity coefficients for
most of the groups in the table are below unity, indicating
(15) It should be cautioned that these comparisons are qualitative and
indicative of general trends only. This is because the standard states of
the individual group activity coefficients are different, and the contribution
of each group to the overall activity coefficient of the substrate model is
affected by inter-group interaction parameters.5

SolVent Control of Enzymatic EnantioselectiVity
J. Am. Chem. Soc., Vol. 118, No. 43, 1996 10369
Figure 4. Dependence of the enantioselectivity of the transesterification in Scheme 1 catalyzed by cross-linked crystals of γ-chymotrypsin on the
activity coefficient ratio for the desolvated portions of the substrates in the enzyme-bound transition states. (A) Double logarithmic plot with a
least-squares fit to a linear dependence with the slope of unity, used to assess predictive ability of eq 1 (correlation coefficient 0.87). (B) Linear
plot with linear regression (correlation coefficient 0.93). Solvents: (a) propanol, (b) tert-butyl alcohol, (c) tert-amyl alcohol, (d) dioxane, (e)
acetone, (f) tetrahydrofuran, (g) isopropyl acetate, (h) tert-butyl acetate, (i) toluene, (j) hexane, (k) octane, (l) cyclohexane. See Methods for
experimental details.
Table 3. Solvent Dependence of the Activity Coefficients of Component Groups of the Model Fragments for the Enantiomers of 1
activity coefficientsa
solvent
hydroxyl
aryl C
aryl CH
carbonyl
methyl
methylene
methineb
cyclohexane
octane
hexane
9.3
6.8
7.7
5.4
3.0
3.6
3.0
1.3
1.4
2.1
1.3
2.5
0.35
0.28
0.35
0.34
0.37
0.51
0.33
0.43
0.37
0.55
0.51
0.56
0.37
0.28
0.35
0.35
0.35
0.46
0.31
0.42
0.37
0.47
0.50
0.54
4.1
3.3
3.8
1.7
1.6
1.7
1.6
2.5
2.3
1.2
2.7
1.6
0.56
0.43
0.51
0.63
0.65
0.76
0.56
0.71
0.61
0.90
0.86
1.1
0.42
0.33
0.40
0.45
0.48
0.60
0.42
0.54
0.46
0.69
0.64
0.77
0.34
0.27
0.33
0.35
0.38
0.51
0.34
0.43
0.37
0.56
0.51
0.59
toluene
isopropyl acetate
tetrahydrofuran
tert-butyl acetate
tert-butyl alcohol
tert-amyl alcohol
dioxane
propanol
acetone
a Activity coefficients were calculated using UNIFAC5 (see Methods for details). The aryl units make up the phenyl group of 1. b Aliphatic.
thermodynamic stabilization of most groups by the solvent. The
hydroxyl and carbonyl groups, however, consistently feature
activity coefficients greater than unity, i.e., these groups are
destabilized by the solvents. Solvent control of the γ′ ratio (and
thereby the enantioselectivity) for this system is thus exercised
through thermodynamically unfavorable interactions between
the solvents and the hydroxyl and carbonyl groups of the
substrate.
Because a given substrate molecule can be represented by
several different combinations of groups, it is important to assess
the effect of approximating the substrate using different types
of fragments. To this end, substrate 1 has been modeled using
a series of incrementally larger groups, and the ability of each
member of the series to predict the solvent dependence of
enantioselectivity has been assessed via the ensuing correlation
coefficient. One can see in Table 4 that as the size of the groups
increases, the ability to approximate the desolvated portion of
the substrate in the transition state deteriorates. This results in
increasing error in the calculation of the γ′ ratio, which erodes
and ultimately destroys the predicted power of eq 1. Therefore,
to optimize the performance of this methodology, one should
model the substrate in terms of the smallest possible fragments
for the determination of γ′.
Materials and Methods
Enzymes. Native and diisopropyl-fluorophosphate-inactivated
R-chymotrypsins (EC 3.4.21.1) were purchased from Sigma Chemical
Co. γ-Chymotrypsin crystals were created from the R-form of the
enzyme, following the method of Stoddard et al.16 Partially inactivated
γ-chymotrypsin crystals (used in the diffusional limitation experiments)
were grown from mother liquors containing varying ratios of the
diisopropyl-fluorophosphate-inactivated enzyme. Co-crystallization
of the native and inhibited forms of the enzyme was confirmed by
measuring the activity of single crystals dissolved in water.7f For use
in organic solvents, crystals were cross-linked and prepared for catalysis
as described previously.6
Chemicals and solvents were purchased from Aldrich Chemical
Co. The organic solvents were of the highest purity available from
that vendor (analytical grade or better) and were dried prior to use to
a water content below 0.01% (as determined by the Karl Fischer
titration17) by shaking with Linde’s 3-Å molecular sieves.
1 was synthesized by refluxing 1 g of 2 in 25 mL of anhydrous
methanol containing 5 drops of concentrated H2SO4 for 12 h. The
reaction mixture was subsequently concentrated by rotary evaporation,
dissolved in 50 mL of diethyl ether, and then washed with five 10-mL
aliquots of 5% NaHCO3 and with 10 mL of deionized water. The crude
product was recovered from the organic phase by rotary evaporation
In closing, the present study constitutes a further step toward
the ultimate goal of predicting enzymatic selectivity on the basis
of the enzyme structure and physicochemical properties of the
substrates and solvents.
(16) Stoddard, B. L.; Bruhnke, J.; Porter, N.; Ringe, D.; Petsko, G. A.
Biochemistry 1990, 29, 4871.
(17) Laitinen, H. A.; Harris, W. E. Chemical Analysis, 2nd ed.; McGraw-
Hill: New York, 1975; pp 361-363.

10370 J. Am. Chem. Soc., Vol. 118, No. 43, 1996
Wescott et al.
Table 4. Correlation Coefficients for Representations of the
Transition State of 1 Using Successively Larger Groups
withdrawn from the top of the solution and the water content was
determined to be 1.46 mg (i.e., approximately 0.3% (v/v)) by Karl
Fischer titration. The hydrolysis product 2 was not detected during
any of the reactions studied. The effect of the added water on γ′ is
accounted for by the explicit inclusion of 0.2% (v/v) water in the
UNIFAC calculations (see Activity Coefficient Calculation below for
details). Note that any competing hydrolysis would merely reduce the
concentration of the acyl-enzyme available for reaction with propanol,
equally reducing the rate of production of both enantiomers of the propyl
ester product, and thus leaving the enantioselectivity unaffected. The
suspensions were shaken at 45 °C and 300 rpm. Periodically, a 10-µL
sample was withdrawn and assayed by chiral HPLC. Because the
transesterifications which lead to the R and S products take place in
the same reaction mixture and the substrate enantiomers compete for
the same population of free enzyme, the ratio of initial velocities of
the reactions is equal to (kcat/KM)S/(kcat/KM)R.3a,4a
desolvation (%)a
group
S
R
corr coeffb
0.87
representation 1
see Table 2
representation 2
hydroxyl
phenyl
methyl
methyl acetate
representation 3
methanol
0.80
100
38
49
11
74
36
59
54
0.78
0.68
0.00
70
38
54
26
74
59
phenyl
methyl acetate
representation 4
ethanol
73
38
50
36
74
56
Chiral HPLC separations were performed using a Chiralcel OD-H
column and a mobile phase of 95:5 (v/v) hexane:2-propanol. A flow
rate of 0.5 mL/min separated the R and S enantiomers of 1 with retention
times of 17 and 19 min, respectively. The products were quantified
using a UV absorbance detector at 220 nm.
phenyl
methyl formate
representation 5
2-phenylethanol
methyl formate
46
50
62
56
Activity Coefficient Calculation. All activity coefficients were
calculated using the UNIFAC group contribution method.5 Calculations
explicitly included the effects of 100 mM propanol and 0.2% (v/v)
water.
a The percent of desolvation of a group is calculated as described in
footnote b to Table 2. b To assess the performance of each representation
of the substrate molecule in the prediction of enantioselectivity, a double
logarithmic plot of the latter vs the activity coefficient ratio of the
desolvated substrate moiety in the enzyme-bound transition state
(calculated for the indicated representation of the substrate) was fit to
a linear dependence with a slope of unity (as predicted by eq 1), and
the resultant correlation coefficient was calculated. By fixing the slope
to 1, both random and systematic errors (due to deviation of the slope
from unity) are reflected in the correlation coefficient.
Structural Modeling. Molecular models were produced using the
crystal structure of γ-chymotrypsin in hexane (Brookhaven data bank
entry 1GMC).10 Because the transition state for the acylation of a serine
protease is structurally similar to the corresponding tetrahedral inter-
mediate for the reaction,18 transition states were modeled as the
tetrahedral intermediates for the reactions. Such models were produced
using the two-step procedure outlined below and described in detail
previously.6 First, potential binding modes of each enantiomer of the
substrate were generated by performing molecular dynamics simula-
tions, followed by energy minimization. The carbonyl oxygen of the
substrate was tethered to the oxyanion binding site using a harmonic
potential with a force constant selected to allow widely different
conformations to be explored, while preventing the substrate from
diffusing too far from the enzyme. This substrate binding mode search
is necessary because the covalently bound tetrahedral intermediate is
sufficiently sterically constrained that molecular dynamics simulations
do not sample highly different conformations separated by large
energetic barriers. For the second step, each substrate binding mode
thus identified was used as a template for creating an initial model of
the tetrahedral intermediate. The low-energy conformation of each of
these starting models was found using molecular dynamics simulations
and energy minimizations. The lowest-energy conformer of the
tetrahedral intermediate was selected as the model of the transition state.
The solvent-accessible surface areas were calculated by the method of
Connolly.14
and subsequently purified by vacuum distillation. 1H NMR (CDCl3)
δ 7.3-7.4 (5 H, m), 4.1-4.2 (1 H, m), 3.8-3.9 (2 H, m), 3.7 (3 H, s),
2.2 (1 H, s).
Propyl 3-hydroxy-2-phenylpropionate, a racemic mixture used to
calibrate the HPLC instrument, was synthesized from 2 following the
procedure used for 1, except that methanol was substituted with 50
mL of propanol. 1H NMR (CDCl3) δ 7.3-7.4 (5 H, m), 4.1-4.2 (1
H, m), 4.1 (2 H, t, J ) 6.0 Hz), 3.8-3.9 (2 H, m), 1.8 (1 H, s), 1.6-
1.7 (2 H, m), 0.8-0.9 (3 H, t, J ) 7.7 Hz).
Propyl (S)-3-hydroxy-2-phenylpropionate, used to assign the
S-product HPLC peak, was synthesized by refluxing 0.5 g of
scopolamine‚HBr in 10 mL of anhydrous propanol containing 5 drops
of concentrated H2SO4 for 3 days. The crude product was recovered
as in the synthesis of 1 and purified by TLC. 1H NMR (CDCl3) δ
7.3-7.4 (5 H, m), 4.1-4.2 (1 H, m), 4.1 (2 H, t, J ) 6.0 Hz), 3.8-3.9
(2 H, m), 2.1 (1 H, s), 1.6-1.7 (2 H, m), 0.8-0.9 (3 H, t, J ) 7.7 Hz).
Kinetic Measurements. One milliliter of solvent containing 100
mM racemic 1 and 100 mM propanol was added to 10 mg of cross-
linked γ-chymotrypsin crystals. Then 0.2% (v/v) water was added to
the suspension to enhance the rate of enzymatic transesterification. To
verify that the added water was soluble in each of the solvent systems,
an even larger amount of water, 0.3% (v/v), was added to 1 mL of
octane (the most hydrophobic solvent used herein) containing 100 mM
ester substrate and 100 mM propanol. The mixture was shaken at 45
°C for 15 min and allowed to settle for 10 min. Then 0.5 mL was
JA961394Q
(18) Warshel, A.; Naray-Szabo, G.; Sussman, F.; Hwang, J.-K. Bio-
chemistry 1989, 28, 3629.
(19) This work was financially supported by NIH grant GM39794. H.N.
gratefully acknowledges a fellowship from the Tokyo Metropolitan Govern-
ment.