Anda di halaman 1dari 21

J. Phys. Chem.

B 2001, 105, 7887-7907


Energetics and Dynamics of Enzymatic Reactions
Jordi Villa ` and Arieh Warshel*
Department of Chemistry, UniVersity of Southern California, Los Angeles, California 90089 ReceiVed: March 20, 2001; In Final Form: June 6, 2001

This review considers the advances made in using computer simulations to elucidate the catalytic power of enzymes. It is shown that some current approaches, and in particular the empirical valence bond approach, allow us to describe enzymatic reactions by rigorous concepts of current chemical physics and to estimate any proposed catalytic contribution. This includes evaluation of activation free energies, nonequilibrium solvation, quantum mechanical tunneling, entropic effects, and other factors. The ability to evaluate activation free energies for reactions in water and proteins allows us to simulate the rate acceleration in enzymatic reactions. It is found that the most important contribution to catalysis comes from the reduction of the activation free energy by electrostatic effects. These effects are found to be associated with the preorganized polar environment of the enzyme active site. The use of computer simulations as effective tools for examining different catalytic proposals is illustrated by two examples. First, we consider the popular proposal that enzymes catalyze reactions by special dynamical effects. It is shown that this proposal is not supported by any consistent simulation study. It is also shown that the interpretation of recent experiments as evidence for dynamical contributions to catalysis is unjustified. Obviously, all chemical reactions involve motion, but unless this motion provides non-Boltzmann probability for reaching the transition state there are not dynamical effects. Vibrationally enhanced tunneling is shown to be a well understood phenomenon that does not lead to special catalytic effects. Similarly, it is shown that nonequilibrium solvation effects do not constitute dynamical contributions to catalysis. Second, the effectiveness of simulation approaches is also demonstrated in studies of entropic contributions to catalysis. It is found that the corresponding contributions are smaller than previously thought.

1. Introduction Enzymes are the catalysts in most biological processes. Thus, there is a major practical and fundamental interest in finding out what makes enzymes so efficient. Although many crucial pieces of this puzzle were provided by biochemical and structural studies, the sources of the catalytic power of enzymes are not widely understood. The main question is related to the rate constant, kcat, for the reaction in the enzyme active site being usually much larger than the rate constant, kcage, for the corresponding reaction in a solvent cage (see below). In other words, the activation barrier g cat (that corresponds to kcat) is much smaller than g cage (see Figure 1), and thus gcat ) gcat - gcage < 0. The fact that the activation barrier is reduced by the enzyme was stated by Pauling long ago.1 However, the way by which this reduction is accomplished was not elucidated (the actual proposal involved ground-state steric effects). There are many proposals for the origin of the rate acceleration by enzymes (a partial list is given in refs 2 and 3) and in many cases it is assumed in a democratic way that evolution can use any factor to accelerate reactions. However, as will be shown below, most of the catalytic proposals involve
* To whom correspondence should be addressed. e-mail:

Figure 1. Comparing the free energy surfaces for an enzyme reaction and the corresponding reaction in solution. The substrate is designated by L (ligand) and the reactive part of the enzyme by L. The main question addressed in the present paper is the origin of the difference between gcat and gcage (or, alternatively, g w).

contributions that cannot be used effectively in enzyme catalysis. One of the main problems in elucidating the origin of the catalytic power of enzymes is the difficulty of dissecting different energy contributions by direct experiments. Even

10.1021/jp011048h CCC: $20.00 2001 American Chemical Society Published on Web 07/24/2001

7888 J. Phys. Chem. B, Vol. 105, No. 33, 2001 mutation experiments, which were found to be extremely useful in identifying catalytic residues, cannot tell us in a unique way what the origin of the catalytic effect of such residues is (see discussion in ref 3). Probably, the most effective way of determining the origin of enzyme catalysis is the use of computer simulation approaches. Such approaches should allow us (at least in principle) to evaluate the rate constants of reactions in enzymes and in solutions and to determine what are the reasons for the given rate acceleration. This review demonstrates the power of computer simulations in determining the origin of catalytic effects. Simulations of different feasible catalytic effects are reviewed by focusing on studies of g cat and the finding that this quantity is usually reduced by electrostatic effects. Specific examples of the use of computer simulations will be provided by focusing on recent proposals that attributed enzyme catalysis to dynamical effects.4,5 These proposals were met with significant excitement6 and have become a focus of current interest. Yet, as will be shown below, the available experiments and simulation studies have not provided any direct evidence of dynamical contributions to enzyme catalysis. This will be done by analyzing the experimental observations by critical simulation studies and also by examining directly what is (if any) the magnitude of the dynamical effects. The proposal that enzyme catalysis is due to entropic effects will also be considered. We hope that both the general review of studies of different catalytic proposals and the specific studies of dynamical and entropic proposals will illustrate the potential of computer simulations in elucidating the molecular origin of enzyme catalysis. 2. The Factors that Determine the Rate Constant Any study of structure-function correlation of an enzyme must be based on the evaluation of the corresponding rate constant. Thus, we start by discussing this critical quantity in the framework of transition-state theory (TST), and also consider the corrections to this theory. Transition-state theory (TST), although originally developed for reactions in the gas phase (e.g., ref 7), has been successfully employed for explaining chemical processes in the condensed phase (e.g., see refs 8-13). TST is based on the so-called fundamental assumption that there exists a hypersurface in phase space (which is usually called transition state, TS) with two properties7 (see also ref 13 and references therein): (a) it divides the phase space into a reactant region (R) and a product region (P), and (b) it assumes that trajectories crossing this surface coming from the reactant region will not recross again before being thermalized or captured in the products region. It is important to clarify at this point that what is called transition state is an ensemble of structures, and not an individual topological saddle point on the potential energy surface. As inferred from the above definition, and in contrast to some recent suggestions5 there is nothing in TST that implies the existence of a rigid barrier (see Figure 2 below). It is now generally agreed that the rate constant for chemical reactions in the condensed phase can be written as (e.g., refs 8, 9, and 11-14):

Villa ` and Warshel

1 kTST ) |x |TS exp{-g}/ 2 1 |TS exp{-G} ) |x 2

-xexp{-g(x)} dx


where g(x) is the free energy functional that describes the probability of being at different points along the reaction coordinate, x; x is the value of x at the transition state; g is is the velocity, and TS g(x); G is defined by eq 2; x designates an average over TS configurations. In other words, g measures the difference between the value of g(x) at the TS and the minimum of g(x). On the other hand, G gives the ratio between the probability of being at the TS to the entire probability of being in the reactant state. Finally, ) 1/kBT, where kB is the Boltzmann constant and T is the temperature. The free energy functional g(x) reflects an average over the entire solute-solvent configurational space, which is perpendicular to the reaction coordinate. The corresponding potential energy surface should involve proper coupling between the solute and solvent charge distribution.10 The evaluation of g(x) and g will be described in the next section. The transmission coefficient depends on two interrelated factors: the probability that a trajectory arriving at the TS from R will end up at P rather than return to R, and the average number of times that a productive trajectory passes back and forth (see ref 15) across x before it moves permanently to P. These factors can be evaluated by examining a family of MD trajectories that start in the TS.8,9,11,12,14 In an approach first developed by Keck,14 the transmission coefficient can be written as

1 |TS )x TS( H(x )x TS(/ |x ) H(x )x TS(/H(x 2 ( (3)

Here, ( denotes an average over many trajectories that start on either the reactant or product side of the barrier, climb to the TS, possibly oscillate in this region for a certain time, and are terminated when they leave the TS region. TS designates an average over periods when a trajectory is in the crossing (t)) is a Heavyside step region (x ) x ( x/2). Further, H(x function that is unity for positive x and zero for negative x . The function is defined as follows: if a trajectory crosses x n times with positive x and (n - 1) times with negative, then ) 1/n; for all other trajectories, ) 0. This function counts only trajectories that begin in R and end in P, and it weights each of these inversely to the number of times the trajectory crosses x in the forward direction. A practical way to evaluate is to propagate downhill trajectories from the TS. Combining the forward and backward trajectories from a given starting point gives a complete trajectory with the desired properties. Alternatively, the reactive flux approach8 can be used. Finally, when the reactive process involves degrees of freedom of light atoms, we should also consider the quantum mechanical corrections to g. The evaluation of such corrections will be considered in section 5.4. 3. Simulations of g of Enzymatic Reactions The most important factor in the rate constant in eq 2 is the exp{-g} term. This factor expresses the relative probability that the reactants will be at the TS. To evaluate g in enzymes, it is necessary to have a model that is sufficiently detailed to capture the complexity of such systems. Available options for modeling enzymatic reactions range from hybrid quantum-

k ) kTST


where kTST is the transition-state theory rate constant and is the transmission factor, which accounts for the correction of the nonrecrossing part of the fundamental assumption of the TST. In a multidimensional system, the TST rate constant can be expressed as2,10

Feature Article mechanical/molecular-mechanics (QM/MM) methods2,16-27 to the empirical valence bond (EVB) method.2,10,11,23,28-36 At present, it is still difficult to evaluate activation free energies, nonequilibrium effects, and dynamical effects by QM/MM approaches. Another problem is the need to use accurate ab initio QM methods, which are computationally intensive, in particular for calculations of activation free energies. One interesting option has been considered by Kollman and co-workers,37 who used gas-phase ab initio calculations to obtain the substrate (solute) charges and force field, using those charges and force field in free energy calculations in solution and in the enzyme active site. This approach is referred to here as the solvated gas-phase (SGP) approach.38 The SGP approach, which was introduced by Jorgensen and co-workers39 in studies of solution reactions, has some useful aspects (see below) but involves major problems. In particular, the solute charges in the enzyme active site are very different than those in the gas phase. This point might not be appreciated by some who consider the missing effect to be rather small, and similar to the change in solvation of the reactants state of a polar molecule upon changing its charges from their value in the gas phase to their value in solution. However, the problem in using gas-phase charges can be quite serious if we are dealing with chemical reactions. In such cases the difference in solvation of the charges obtained in solution (that will resemble more the actual charge distribution in the enzyme) to the gas-phase charges can be more than 100 kcal/mol. For example, in the heterolytic cleavage of a C-O bond, the best gas-phase calculation will give at a large distance a C O system with very small solvation while the solution calculations give a C+ O- solvated ion pair system (see Figure 1 of Hwang et al.10 and Figure 2.3 in ref 2) with solvation energy of about -80 kcal/mol at 4 separation. Furthermore, the force field deduced from the gas-phase calculations constrains the reacting system along its reaction coordinate.37 This prevents the exploration of the crucial entropic effects that are associated with the motion of the reactants perpendicular to the reaction coordinate (e.g., the bending modes of the TS become translational motions in the ground state and this motion is not represented in the SGP methods). Nevertheless, the SGP method does allow us to perform proper free energy perturbation (FEP) calculations of the solvation of the gas-phase charges in enzymes and solutions. In this respect it has an advantage over those QM/MM approaches that do not perform a proper configurational average. The above problems do not exist in the EVB approach, which was designed from its early conception28 as a consistent way of transferring gas-phase charges and potential energy surfaces to solutions and proteins. The EVB method is perhaps the most effective way of performing proper FEP/umbrella sampling (FEP/US) calculations on such systems.10 As will be shown below, the EVB method correctly reproduces the effect of the solvent on the solute charges and allows us to perform consistent free energy calculations. This includes the evaluation of nonequilibrium solvation effects that are missing in SGP and in many other current treatments. It might be instructive to address the assertion that the SGP uncoupled approach has the advantage over EVB in that one does not have to calibrate the parameters for the solution reaction in order to study the enzymatic reaction.40 Actually, when the EVB Hamiltonian is calibrated by gas-phase ab initio calculations, it provides a consistent surface for both enzymes and solution calculations (see ref 41). The SGP approach is also calibrated by fitting a force field to the gas-phase surface, but this fitting cannot be

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7889 SCHEME 1: VB Structures Used in This Work for Describing the Hydride-Transfer Step in the Reaction Catalyzed by ADH

transferred consistently to solutions or to enzymes since the change in the solute polarization changes its surface. In addition to the above advantages, the EVB approach provides an extremely effective way of evaluating ab initio QM/ MM free energy surfaces.23 In this approach one starts by running trajectories over the EVB mapping potential and evaluating the corresponding free energy profile. Next, trajectories over the EVB potential surface, in the reactant state and the transition state, are used to evaluate the time dependent differences between the ab initio (ai) and EVB surfaces. These differences and the EVB mapping free energy are used to evaluate the ab initio QM/MM activation free energy. This approach is referred to as the QM(ai)/MM-FEP(EVB) method. Another promising alternative has been used by Truhlar, Gao, and co-workers, who applied variational transition state (VTST) with semiclassical multidimensional tunneling (MT) calculations to enzymatic reactions.24,42 They employed a QM/MM potential energy surface, treating the reactive part with a semiempirical Hamiltonian. The use of a QM/MM approach allows a consistent calculation of solute charges along the reaction coordinate. However, the molecular orbital semiempirical methods used may not provide accurate charge distribution and potential energy surfaces for the solute system. The use of an ab initio solute Hamiltonian can help in obtaining more reliable results, but in this case, the calculations would become too expensive to allow for the proper phase space sampling, which is essential for obtaining reliable free energy profiles. Current VTST/MT studies have not involved sufficient sampling of the protein configurations to allow a proper evaluation of the relevant potential of mean force (PMF), although this PMF was used for the initial exploration of the potential energy surface. The VTST/MT approach has included the vibrational entropy of the solute within the harmonic approximation, but this does not provide a realistic estimate of the entropic contributions in the shallow potential surfaces of reactions in solution. Another problem is associated with the use of PMFs. Although this is a reasonable approach, it does not include the contribution of nonequilibrium solvent effects (see section 5.3). Other promising approaches, which should help in reliable evaluations of activation free energies in enzymes, were reported recently. Mo et al.43 elegantly followed the EVB idea and used diabatic QM states in free energy calculations. This approach should allow one to explore nonequilibrium solvation effects. An interesting approach of Zhang et al.,25 does take into account the proper effect of the environment on the solute charges in ab initio free energy calculations. This approach, however, still constrains the reacting fragments to move along a predefined reaction coordinate obtained from ab initio QM/MM calculations. Thus, this approach does not consider the effect of the solute fluctuations. In our opinion, although other approaches

7890 J. Phys. Chem. B, Vol. 105, No. 33, 2001 are surely to provide reliable and powerful tools in the future, at present the EVB approach is the method of choice. Because the EVB method has been discussed extensively elsewhere,10,28,44 we will describe it here only briefly. This approach represents a reaction in terms of resonance structures, or more precisely, mixtures of diabatic states corresponding to classical valence-bond (VB) structures. In the reaction catalyzed by alcohol dehydrogenase (ADH), for example, transfer of a hydride from the alcoholate to the NAD+ cofactor could be modeled by the two VB structures shown in Scheme 1. Note that in Scheme 1 and in the rest of the paper we actually deal with the reversed reaction, the hydride transfer from NADH to the substrate. We study the reversed reaction because we are interested in the origin of catalytic enhancement by enzymes and there is no apparent enzyme catalysis in the reaction of hydride transfer from benzyl alcoholate to NAD+ (see Figure 4 below). Note also that the above drawings do not correspond to pure VB states. This is because each of the two states shown above represent a mixture of pure states, and when these states are of sufficient high energy, we can consider them in an implicit way without using a large Hamiltonian (see Appendix of ref 45). Thus, we can reduce the number of representative VB structures while still retaining the effect of the rest in an effective way.2,10 The diagonal matrix elements of the EVB Hamiltonian are represented by

Villa ` and Warshel one can expect that an EVB Hamiltonian that is calibrated on reliable gas-phase ab initio (or experimental) results will provide reliable potential surfaces in solution. This is apparently the feature missing in the above-mentioned SGP approach. The reliability of the EVB treatment has been demonstrated by comparing EVB electrostatic free energies to the corresponding ab initio quantum-mechanical values in solution.23,46 The EVB approach evaluates the activation energy g by running series of trajectories on potential surfaces that drive the system gradually from one VB state to another. In the simple case of two VB states, these mapping potentials, m, can be written as linear combinations of the reactant and product potentials, 1 and 2:

) (1 - m)

+ m

(0 e m e 1)


If is changed from 0 to 1 in fixed increments (m ) 0, 1, 2, ..., M), potentials with one or more of the intermediate values of will force the system to fluctuate near the TS. The difference between the free energy minimum for a system on mapping potential m and the free energy minimum of the reactant state is the sum of the free energy changes associated with incrementing from 0 to m:

Gm )

Hii )




+ Uss(r,q) (4)

G(n-1)fn n)1


which resembles a classical force field. Here R and Q represent the atomic coordinates and charges of the VB structures, and r and q are those of the surrounding protein and solvent. Rigas is the gas-phase energy of the ith VB structure at infinity separation between the fragments: Uintra(R,Q) is the intramolecular potential of the solute system (relative to its minimum), USs(R,Q,r,q) represents the interaction between the solute (S) atoms and the surrounding (s) solvent and protein atoms, and Uss(r,q) represents the potential energy of the protein/solvent system (ss designates surrounding-surrounding). The offdiagonal elements of the EVB Hamiltonian are represented by simple functions (usually exponential or Gaussian) of the solute coordinates

These changes can be evaluated by a free-energy perturbation procedure that uses the relationship47

G(n-1)fn ) --1 lnexp[-(




where, as before, n-1 denotes an average over trajectories on n-1. The free energy functional g for any particular value of the reaction coordinate, x, then is10

g(x) ) Gm - -1 ln(x - x) exp[-(Eg -


where m is the mapping potential that keeps x in the region of x. If the changes in are sufficiently gradual, the free energy functionals g(x) obtained with several values of m overlap over a range of x, and patching together the full set of g(x) gives the complete free energy curve for the reaction. Analogously, we can obtain the probability of being at x on the diabatic surfaces by replacing Eg by i (i ) 1, 2) in eq 10. The overall procedure is referred to as a free energy perturbation/ umbrella sampling (FEP/US) approach. The seemingly simple appearance of the EVB model may have led to the initial impression that this is an oversimplified model. However, the model has been eventually widely adopted as a powerful tool for studies of reactions in condensed phases (e.g., refs 31-35). Some workers (e.g., refs 11, 48, and 49) confused the EVB model with early gas-phase VB studies or with the use of the VB picture in discussing reactions in solution. For example, refs 48 and 49 attributed the EVB idea to Mulliken.50 It is important to clarify in this respect that the main and most important element of the EVB model is the realization that the addition of the interaction between the solvent and each VB state to the corresponding VB Hamiltonian, and the diagonalization of this Hamiltonian, should lead to quantitative potential surfaces for reactions in solutions. We are not aware of an earlier approach with this crucial feature or, in fact, of earlier models for quantitative calculations of chemical reactions

Hij ) f(R)


and the diabatic states are assumed to be in a representation that makes them orthogonal so that the ground-state potential surface, Eg, is obtained by

HCg ) EgCg


where Cg is the ground-state coefficients vector. This Eg provides the actual potential surface for the given reaction in the specific environment (solution or protein). The EVB treatment provides a natural picture of intersecting electronic states that is useful for exploring environmental effects on chemical reactions in condensed phases.28,44 The groundstate charge distribution of the reacting species (solute) polarizes the surrounding (solvent), and the charges of each resonance structure of the solute then interact with the polarized solvent.2,29 This coupling enables the EVB model to capture the effect of the solvent on quantum-mechanical mixing of ionic and covalent states of the solute. When, for example, the solvent stabilizes the ionic state to a greater extent, the resulting ground state has more ionic character and more solvation energy. Thus,

Feature Article

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7891 generated for a hypothetical low barrier reaction (this was done by drastically lowering the R(2) gas term in eq 4) in order to generate the reactive trajectory without investing the extremely long time needed to generate such a rare event with the actual barrier. As seen from the figure, the barrier oscillates with the fluctuations of the polar environment surrounding the bound substrate. Once in a while the environment reaches a configuration where the barrier is small and the hydride can be transferred. Thus, the rate of hydride-transfer reactions, or any charge-transfer process, depends on the fluctuations of the surrounding active site and the resulting fluctuations of the potential energies 1 and 2 of 1 and 2 (nuclear quantum mechanical aspects of the process will be considered in section 5.4). To examine the relationship between the protein fluctuations to the activation free energy, we consider the situation in Figure 2 and approximately estimate the potential energy at the transition state and reactant state by

1 Eg(R) [ 2

gas 1 (R (r))

gas 2 (R (r))]

- H12(R) + 1 [U (Q ) + U2 (Q2)] 2 1 1


gas 1 (RRS)

+ U1(Q1)


Figure 2. Illustrating the effects of the fluctuations of 2 and 1 on the reaction of ADH. The figure depicts the time dependence of the groundstate energy profile Eg. This energy is plotted at reactants, transition state, and products. The data have been obtained along an artificially generated (see below) reactive trajectory that involves hydride transfer from the NADH to the benzyl aldehyde substrate. The panels below the main figure show the ground-state potential and activation barrier at different points along the trajectory. The panels are colored in order to indicate the state of the reacting system at each panel, moving from reactants (red) to products (blue). The chance for the reaction depends on the average of the fluctuating activation barrier E. E is determined approximately by ( 1 + 2)/2 - H12 - 1 (see text) and thus its trend is determined by the fluctuations of the electrostatic contributions to ( 1 - 2)/2. The reaction has been forced to be very exothermic and almost barrierless by having a low value of R(2) gas in order to generate the rare reactive trajectory. Because the actual average potential energy barrier is around 16 kcal/mol in the actual enzyme reaction, this rare event will only occur in a time that is 109 times longer than the time depicted in the figure. In the rest of the paper we will always consider the actual potential energy surface of ADH.

where R and r are the solute and solvent coordinates respectively, Ui is the solute-solvent interaction potential for the charge distribution (Qi) of the ith resonance structure. Here we assume that H12 is independent of the solvent coordinate. Next we have

1 E(R) ) Eg(R) - Eg(RRS) E gas(R ) + U 2



1 E gas(R ) ) [ 2

gas 1 (R (r))

gas 2 (R (r))]

gas 1 (RRS)


U ) U2(Q2) - U1(Q1)
Now we can estimate the activation free energy by


exp [-g] ) exp [-E]ERS

exp [-E gas(R (r))]ERS

in solutions. In fact, the EVB has allowed for the first realistic MD simulations of the energetics and dynamics of chemical reactions in solutions51 and enzymes.52 Finally, some instructive variants of the EVB models are not much different than the original EVB. For example, the so-called Kim-Hynes model31 seems to be an EVB model where the solvent is modeled by means of a continuum approach and is quite similar to the original EVB model, which used a Langevin dipole solvent model28 (see also chapter 2 in ref 2). Although the EVB mapping procedure of eq 10 provides a rigorous way of obtaining g, we can also use the EVB model in an approximated way in order to clarify the relation between fluctuations of the protein structure and the activation free energy for an enzymatic reaction. To do so, it is instructive to consider the transfer of a hydride from NADH to benzyl aldehyde in the reaction catalyzed by ADH (see Scheme 1). Figure 2 shows the calculated barrier (E) for the hydridetransfer process as a function of time. The plot has been

1 exp - U 2



Here we assumed that the solute and solvent contributions are separated and thus established that the solvent contribution to the activation barrier is given approximately by exp[-(1/2)U]ERS. For reactions that involve relatively small changes in geometry but substantial changes in polarity, simulation studies have shown that electrostatic interactions usually make a much larger contribution to exp[-(1/2)U]ERS than van der Waals interactions. Thus, we focus on the electrostatic effect and show (Figure 3) the time-dependent electrostatic energy difference (1) el(t) ) (2) el - el for the hydride transfer from benzyl alcoholate to NAD+ in ADH and in water. The minima in el correspond to points with relatively high probabilities of hydride transfer.

7892 J. Phys. Chem. B, Vol. 105, No. 33, 2001

Villa ` and Warshel energy gained from charge-dipole interaction, GQ, is spent on changing the dipole-dipole interaction, G, so that the free energy of solvation of the transition state is given by the following equation:2

Gsol = GQ + G = GQ - 1/2GQ ) 1/2GQ (15)

In proteins, however, the active site dipoles associated with polar groups, internal water molecules, and ionized residues are already partially oriented toward the transition-state charge center. Thus, G is smaller than in water, and less free energy is spent on creating the oriented dipoles of the protein transition state. The free energy term G is basically the so-called reorganization energy2,80 for the process of forming the transition-state charges. For example, in water we have to break water-water interactions to form good hydrogen bonds to the TS. In the enzyme, on the other hand, the hydrogen bonds are already partially oriented toward the transition-state charges. A calculation of the reorganization energies for the hydridetransfer reaction in ADH and in solution is illustrated in Figure 4b. Larger effects are found in other cases, e.g., refs 44, 64, and 69. The idea that enzymes use preorganized dipoles to catalyze their reactions should not be confused with the regular considerations of the Marcus reorganization energy. That is, the activation energy for chemical reactions can be approximated by2,81

Figure 3. Fluctuations of the electrostatic contribution to the energy gap ( 2 - 1) for the hydride-transfer step in ADH ( 1 and 2 correspond to RS1 and RS2 in Scheme 1) and for the corresponding reaction in water. The plots for the protein (lower plot) and water (upper plot) systems have been obtained by running trajectories on the mapping potential m ) 1. A value of R(2) gas ) 28 kcal/mol has been used and the system was initially relaxed during 30 ps.

4. Electrostatic Effects Play a Major Role in Enzyme Catalysis The EVB method and related techniques provide ways to evaluate both the activation barrier for the reaction steps in an enzyme active site (g cat) and the corresponding barriers for a reference reaction in a water solvent cage (g cage). Activation free energies gcat and gcage (and in some cases related energy contributions) have been calculated in simulation studies of a variety of enzymes, including lysozyme,16,53,54 serine proteases,29,55-57 acetylcholinesterase,58,59 ribonuclease,60 staphyloccocal nuclease,61 carbonic anhidrase,62 DNA polymerase,63 triosephosphate isomerase,18,64 Ras,65 aldose reductase,66-68 lactate dehydrogenase,69 malate dehydrogenase,70 isocitrate dehydrogenase,71 chorismate mutase,26,72 tyrosine phosphatase,73 orotidine 5-monophosphate decarboxylase (ODCase),74,75 and others (for a partial list, see ref 76). In virtually all the studies, in which the activation energy of the enzymatic reaction was compared consistently with the same reaction in solution, it has been found that g cat is considerand that this reduction results mainly ably smaller than g cage from electrostatic interactions that solvate the TS in the enzyme more strongly than the corresponding TS in solution (a recent discussion of this issue in ODCase is given in ref 75). Figure 4a shows a typical example for the reaction catalyzed by ADH, although this case does not involve a significant catalysis. The preferential electrostatic stabilization of the TS in an enzyme seems paradoxical because the electrostatic interaction energy of the reacting atoms with their surrounding typically is similar to the corresponding interaction energy in solution. A solution to this puzzle emerged from early computer simulations and energy considerations,77 which showed that the folded enzyme provides a preorganized dipolar environment that is already partially oriented so as to stabilize the charge distribution in the TS. In reactions in water, the solvent must pay a significant reorganization energy to orient the polar environment toward the TS charges. The enzyme has already paid much of its reorganization energy to create the active site during the folding process.29,55,64,69,77-79 This point can be quantified by considering the fact that in polar solvents, about half of the

g (G0 + )2/4 - H12 + H122/( + G0) (16)

where G0 is the free energy of moving from the reactant to the product (or an intermediate) and is the solvent reorganization energy, which mainly reflects the changes in the solventsolvent interaction during the reaction. H12 describes the mixing between the reactant and product states. The first term in this equation is the well-known Marcus expression80 for electrontransfer reactions, whereas the other terms allow the treatment of regular chemical reactions. At any rate, one can use the first term in equation 16 to correlate small changes of g with the effect of mutations on G0. Consideration of the Marcus term led different workers82 to argue correctly that g can be reduced by reducing G0 or by reducing .83 Unfortunately, this argument does not help in elucidating the origin of enzyme catalysis. That is, instead of stating that g cat is reduced, we state now that G0 and/or are reduced. However, we are still left without explaining how these free energies are reduced. As seen from Equation 16 the reduction of in cases when G0 is small leads to the reduction of g cat. Furthermore, is reduced in nonpolar environments. Thus, one may propose that proteins catalyze their reactions by using nonpolar active sites. This, however, does not resolve the problem of enzyme catalysis. That is, although the nonpolar environment does reduce , it would lead to an increase of G0 in typical charge separation reactions and to anticatalytic effects (see discussion in ref 69). What is missing in the above proposal is the idea that enzymes reduce both and G0 by a preorganized polar (rather than nonpolar) environment as established by simulation studies.2,44,69 The origin of this reduction is directly connected to the abovementioned preorganized polar environment. Although is reduced in enzyme active sites44,69 many of the cases with the largest reduction of g cat involve major reduction of G0, and this reduction cannot be accomplished in nonpolar environments (e.g., see ref 75 and a general discussion in ref 2). The reduction of G0 is accomplished by a preorganized polar

Feature Article

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7893 explained above, low dielectric proteins would destabilize rather than stabilize ionic transition states. Krishtalik and co-workers then suggested that the interprotein field stabilizes the ionic states but, in contrast to their statements, were unable to estimate the effect and origin of such fields (see ref 87). In fact, they thought that ion pairs are stabilized in a low dielectric,85,87 while the opposite is true. The idea that enzymes stabilize the TS primarily by electrostatic effects is consistent with numerous mutation experiments, which have shown that mutations of residues that stabilize the TS charge distribution in the native protein lead to decreases in kcat. Examples can be found in the studies of tyrosyl t-RNA synthase,88 carboxypeptidase,89 and subtilisin.90 The proposal that the catalytic effect results largely from paying part of the reorganization energy during the folding process is also consistent with mutation experiments. Shoichet et al.91 have found that mutations that increase the stability of T4 lysozyme increase g cat. Evidently, the active site of the native protein is not optimized for maximum stability, but rather for minimizing g cat. As stated above, to evaluate the catalytic effect of an enzyme, it is important to calculate g cage (or gw) in addition to gcat. Consider, for example, a reaction in which the charges of the reacting groups are more localized in the reactant ground state (RS) than in the TS. Calculations of g cat alone might suggest that the enzyme solvates the RS more strongly than it does the TS. But this conclusion would overlook the fact that water also solvates the RS better than the TS. A proper comparison would show that the enzyme stabilizes the TS more than water does and thus decreases g cat relative to gcage. This argument does not depend on the choice of the reference state that is used to define g cat and gcage. One could just as well calculate the energies of the TS in the enzyme and solution relative to the energy in a vacuum, rather than relative to the RS. This point can be illustrated in the case of haloalkane dehydrogenase, which catalyzes the reaction Asp-CO2- + ClCH2CH2Cl f Asp-CO2CH2CH2Cl + Cl-. Simulations of the reaction indicated that the enzyme interacts more strongly with the RS, where a negative charge is localized on the Asp carboxyl group than with the TS, where the charge is delocalized.92 However, a comparison with the solution reaction probably would lead to the conclusion that the enzyme stabilizes its TS more than water does. Isocitrate dehydrogenase, which catalyzes oxidative decarboxylation of isocitrate in the presence of Mg2+, provides another instructive example. Hurley and Remington71 calculated g cat for this reaction and found that the TS was much more stable in the enzyme than in water. They noted, however, that although the protein interacted favorably with the Mg2+ ion in the TS, the electrostatic interactions of the protein with a substrate carboxylate group were unfavorable, which led them to conclude that the enzyme destabilized the TS. This conclusion reflects inconsistent separation of the Mg2+ ion from the rest of the surrounding. If one likes to consider Mg2+ as part of the solute, then including Mg2+ in the reference reaction in water would increase g cage, showing again that the extended TS is more stable in the enzyme than in water. The important quantity is the difference between the total free energies of forming the TS in the enzyme and in solution. It is hard to underemphasize the importance of using a proper thermodynamic cycle and proper reference states in analyzing enzymatic reactions. Additional instructive examples about a proper analysis of various ground-state destabilization proposals are given in refs 2, 3, and 75.

Figure 4. (a) Ground-state free energy profiles (the g of eq 10) on the EVB ground-state potential energy surface for the hydride-transfer reaction from NADH to benzyl aldehyde in ADH and in water. (b) Solvent contribution to the diabatic free energy functionals (gelec 1 and gelec 2 ) for a G0 ) 0 version of the corresponding free energy functions. The free energy functions of the product states have been shifted vertically (by changing R(2) gas in order to have G0 ) 0 and thus to simplify the illustration of the solvent reorganization energy . The reorganization energy is shown to be smaller in the protein than in water. The experimental g and G0 are 15 and -1.4 kcal.mol for the reaction in the enzyme159 and 20 and 4 kcal/mol for the reaction in water,66 respectively.

environment. Here, what counts is not the reduction of the Marcus reorganization energy, which corresponds to the reaction coordinate (or eq 16) but the preorganization of the active site permanent dipoles that allow the enzyme to stabilize charges more than water does (the above-mentioned reduction of G). Having preorganized dipoles also reduces , but what counts here is the reduction of G0. It might be useful to comment here on a related proposal of Krishtalik and co-workers.84-86 These workers emphasized the importance of reducing the reorganization energy in the low dielectric environment of the protein but could not reproduce or examine the corresponding catalytic effect. That is, those studies of Krishtalik and co-workers that can be considered as well-defined studies, were based on a continuum model of nonpolar spherical proteins (e.g., refs 84 and 86). Since these studies obtained small s, it was stated that such a model reproduces the catalytic effect of chymotrypsin. However, as

7894 J. Phys. Chem. B, Vol. 105, No. 33, 2001 5. What Is the Role of Dynamical Effects in Enzyme Catalysis? 5.1. Defining Dynamical Contributions to Catalysis. As stated in the Introduction, it is frequently suggested that dynamical effects contribute significantly to enzyme catalysis.4,5,93-100 The appeal of this idea may stem from the fact that all molecular processes depend on thermal fluctuations. As we have discussed in section 3, fluctuating interactions of a system with its surrounding occasionally create configurations in which the energies of the reactant and product states overlap. Similar fluctuating interactions dissipate energy to the surrounding as the product relaxes. However, the fundamental importance of random thermal motions does not imply that enzymes use particular vibrational modes to enhance reaction rates. To address this issue in a meaningful way, it is crucial to define what is meant by a dynamical effect. As discussed in section 2, the exponential factor exp{-g} in the TST expression for the rate constant is simply a probability factor that could be evaluated by a nondynamical approach such as Monte Carlo simulations. Thus, if we can estimate correctly the rate constant by Monte Carlo approaches, we do not haVe dynamical effects. It is generally accepted that any dynamical effects on the rate must lie in the transmission coefficient .9,12,14 The main question, therefore, is whether is much greater in some enzymatic reactions than in the corresponding reactions in solution. We will address this question in the following section. Although g is a nondynamical property, there are cases where g reflects only a small part of the overall configurational space of the system. This could be the case in fast photochemical processes, where an excited reactant is created impulsively.101 The effective activation barrier for a reaction of the excited species reflects only configurations that are accessible during the reaction time. The rates of such reactions could depend on the dynamics with which the system explores new configurations and equilibrates with the surrounding. Dynamical effects of this nature, however, are not likely to be important in most reactions with significant activation barriers, even if large regions of conformational space are effectively inaccessible. For example, in enzymatic reactions, we do not need to consider contributions to g from the unfolded protein. In principle, an enzymatic reaction could involve a different mixture of solute and environmental coordinates than the corresponding reaction in solution. The solvent coordinates could, for example, be more frozen in the enzyme than in solution. However, there is no reason to assume a priori that this will be the case. We will examine this point in section 5.3. There are of course reactions where the TST requires substantial modifications or restrictions. For example, the classical partition functions that underlie the TST cannot be used safely for reactions that involve proton or hydride transfer, because they neglect both the zero-point vibrational energies and the possibility of nuclear tunneling. The treatment of these quantum effects is discussed in section 5.4. 5.2. Transmission Coefficients Are Similar for Corresponding Reactions in Enzymes and Water. As discussed in section 2, the transmission coefficient () for a reaction can be evaluated rigorously by running a family of downhill MD trajectories starting at the TS and using eq 3. These procedures require extensive simulations and have been applied to only a small number of enzymatic reactions. As much as the role of in enzyme catalysis is concerned, one can use a more general approach by considering the

Villa ` and Warshel interactions of a reacting substrate with its surrounding at the TS.2,10,29 The starting point for this treatment is the consideration of the average time, +, that it takes to move away from the TS region. This time can be evaluated by generating a set of trajectories that start at x and recording the time dependence of the average position of the productive trajectories along the reaction coordinate, x+(t).10 + is taken to be twice the time required for x+(t) to change from x to x + x/2. The average velocity of productive trajectories away from x then is x/+, and the transmission coefficient can be viewed as the ratio of this velocity to the mean velocity of crossing x:
-1/2 1 1 x |TS+ x+ /|x + (2/m)


where m is the relevant reduced mass. According to this expression, any significant difference between the transmission factors for an enzymatic reaction and the corresponding reaction in solution must reflect a difference in this time constant. Further, since the solute is the same in the enzyme and solution, such a difference in + must originate in the interactions of the reacting groups with their surrounding in the two transition states. We define our generalized, time-dependent reaction coordinate x(t) as the energy gap between the reactant and product VB states ( (t) ) 2(t) - 1(t)).2,51 We can divide this coordinate into a solute coordinate (xR(t)) reflecting internal bonds of the reacting solute, and a solvent coordinate (xr(t)) representing interactions of the solute with the solvent. We use the term solvent in a general sense for the surrounding of the reacting atoms in either the enzyme or the solution.2,77 ,102 The solvent coordinate usually depends mainly on the difference in electrostatic energies, el:10

xr(t) el(t) )




Thus, in the common case that the relaxation time for solvent motions is equal to or longer than that for the solute dipole, + is given to a good approximation by10
1 + )

x(t)TS/t x

xr(t)TS/t x r


where x r is the width of the TS region along the solvent coordinate. By starting with coupled equations for the time dependence of the solvent and solute coordinates and using the linearresponse approximation,103 Hwang et al.10 obtained the following relationship between the electric dipole of the solute () and the average time dependence of the solvent reaction coordinate after a system enters the TS:

el(t)TS )

max el

0t el(0) el()TS(t - )TS d

el(0) el(0)TS (20)


In this expression, TS denotes an average over a trajectory on a mapping potential that keeps the system in the region of the TS, max is the difference between the solute dipoles in the product and TS (2 - TS), and el(t)TS is the average change in el between these two states el2 - elTS. The integral in the numerator depends on the response function

Feature Article

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7895 reaction, although reference calculations of for the same reaction in water were not described. Note in this respect that in order for to help in catalysis, it should be small in water and not in the enzyme, but as shown in our studies, this is not the case. qvist and Fothergill64 have calculated the activation free energies of the triosephosphate isomerase reaction in the enzyme and in water. These calculations reproduced the observed catalytic effect of the enzyme, supporting the conclusion that the rate of the enzymatic reaction is determined mainly by the activation energy and that any dynamical effects are relatively minor. The importance of dynamical contribution to catalysis could have been determined more quantitatively if we knew the overlap between the vibrational wave functions of the reactant and product states. In other words, we would like to know the Franck-Condon factors or the corresponding origin shifts (the di) between the reactant and product state. Although this might look hopeless in the case of chemical reactions in solution or proteins, we can obtain the relevant information by the dispersed polaron (or the equivalent spin-Boson) approach.105-107 This is done by using a Fourier transform of the time dependent energy gap, (t), between the two VB states that represent the reactant and product states of the rate-determining step. The relevant spectral density, J(), satisfies the relationship

Figure 5. Autocorrelation of the electrostatic contributions to the energy gap in ADH and in solution. The calculations are done by running trajectories with a mapping potential that corresponds to the transition-state region, m ) ( 1 + 2)/2. The oscillations in the protein autocorrelation function are due primarily to the movement of the Zn2+ ion.

el(0) el(t), which is related to the time derivative of the classical autocorrelation function of el by

C() ) el(t) el(t + ) ) el(t) el(t + ) ) t t - el(t) el(t + ) (21)

The last equality follows from the fact that the classical autocorrelation function C() is a real and even function of . Equations 20 and 21 indicate that we can deduce the importance of dynamical effects in an enzymatic reaction simply from the autocorrelation function of el. If the autocorrelation functions are similar in the enzyme and in solution, the transmission factors must also be similar. This analysis avoids the need to average a large number of downhill trajectories with a distribution of initial velocities, because all the time-dependent quantities in eq 20 can be obtained directly from a single trajectory on a mapping potential for the TS.10 This approach was verified in ref 10 (see footnote 15 of the present work) and used to estimate the ratio between the transmission factors in the catalytic reaction of subtilisin and the same reaction in solution.2,104 A similar analysis is done here by comparing the dynamical effects for hydride transfer in ADH and in solution, where both reactions are modeled using EVB potential surfaces. Figure 5 shows the relevant C(t) for the catalytic reaction of ADH and the corresponding reference solution reaction. The similarity of the two autocorrelation functions indicates that the transmission factors are similar. Thus, dynamical effects appear to make little contribution to enzymatic catalysis in this case. The same conclusions were obtained by comparing the autocorrelation functions for native subtilisin and the Asn155Ala mutant, in which the rate of the reaction is decreased by a factor of 2 10-3. Again the dynamical effects are similar for the native and mutant enzymes.29 Neria and Karplus11 have used the EVB method to simulate the first step in the catalytic reaction of triosephosphate isomerase. They calculated the transmission factor by the reactive flux method8 and found it to be 0.4. Considering that triosephosphate isomerase increases the rate of the reaction by an overall factor of 5 105 compared to general-base catalysis in solution,64 a transmission factor of 0.4 seems not far from unity. Thus, dynamical effects appear to be minimal in this

pjdj2 ) 2k T|-J()d| 2 j


where is the total reorganization energy. J() measures how strongly a vibrational mode (or modes) with frequency is coupled to the reaction coordinate. A comparison of the spectral density functions for the enzymatic and nonenzymatic reactions should highlight any vibrational modes that play particularly important roles in the catalysis. If the strongly coupled modes have significantly different frequencies in the enzymatic and solution reactions, the catalysis could depend partly on dynamical effects. Figure 6 shows calculated spectral density functions obtained from MD simulations of the reaction catalyzed by ADH and the corresponding reaction in solution. The spectral density for both systems is given separately for the solute and solutesolvent energy contributions. Although the peaks are not identical they occur in similar regions. In particular, the environmental contributions involve mainly low-frequency modes (phonons) in both the protein and solution cases, and the main difference between the two spectral densities is the integrated amplitude, which is proportional to the reorganization energy by eq 22.105 It appears that the amplitude is smaller for the enzymatic reaction than for the reaction in solution (although this difference is smaller here than in other related cases108). As discussed in section 5.2, a decreased reorganization energy appears to be a common feature of the reduction of g in enzymatic reactions and is not indicative of dynamical effects. Furthermore, the differences in the amplitudes of the modes in the enzyme and solution cases are unlikely to lead to differences in dynamical effects as long as the energies are distributed between the different modes (and within each mode) according to the corresponding Boltzmann distribution. Since we deal with a reaction with a significant activation barrier, the system has sufficient time to equilibrate at the reactant state before the very rare event where the thermal energy produces a combination of many modes that leads to the TS without any special dynamical effect. This point is demonstrated in Figure 7, where we generated rare productive trajectories by running downhill

7896 J. Phys. Chem. B, Vol. 105, No. 33, 2001

Villa ` and Warshel

Figure 6. Spectral density, J(), for the reaction of ADH and the corresponding solution reaction. These J()s were obtained from the autocorrelation function of the total time dependent energy gaps for trajectories at the TS region. (a) depicts the contribution of the modes of the reacting part where the peaks at 700 cm-1 reflect hydrogen motions in the transition-state region. (b) depicts the contribution of the generalized solvent coordinate.

trajectories from the TS (the time reversal of these trajectories gives us the actual reactive trajectories). As is clear from the figure, the nature of the reactive trajectories and the corresponding time dependence of the solute and solvent coordinates are similar in the protein and solution cases. The recrossing effect (the transmission factor) is rather trivial and almost identical in both cases (see also discussion on C(t) and Figure 5). To further clarify the above point, we note that it is hard to attribute significant dynamical contributions to the small difference between the peaks of J() of the protein and water reactions. Basically, the reactive trajectories of Figure 7 appear to correspond to superpositions of the many available reactive modes, rather than to coherent excitations of selective modes. Furthermore, the average speed of these trajectories (even out of the TS region) appears to be similar in the protein and solution cases (the velocity can be estimated by looking at the distance traversed in the times depicted in Figure 7). Berendsen and others have suggested that important collective motions can be identified by a simulation technique called essential dynamics, in which one examines the covariance of positional fluctuations of CR atoms in different regions of the protein (see, e.g., ref 93). The low-frequency global vibrational modes extracted by diagonalization of the covariance matrix have been suggested to have special functional significance. In proteins with several distinct domains, these modes can reflect

Figure 7. Downhill classical trajectories on the ground-state EVB surface for (a) the hydride-transfer reaction catalyzed by ADH and (b) the corresponding reaction in solution. The time reversal of the downhill trajectories correspond to the actual reactive trajectories. The figure represents the reactive trajectories in terms of the corresponding solute int elec and solvent coordinates, which are defined as int 1 - 2 and 1 elec , respectively, where int and elec correspond to the intramo2 lecular and electrostatic contributions to the energy gap. Each downhill trajectory runs for a total time of 250 fs. The figure also shows the position of each trajectory after 50 fs. These times can help the reader in estimating the average velocity of the reactive trajectories.

movements of one domain relative to another and sometimes can be correlated with structural variations seen in X-ray structures determined under different conditions.93 However, the availability of a conformational transition with a small barrier does not mean that fluctuations between the two conformations have any functional importance. The structural variations are usually an integral part of a flat free energy surface for the transition. Only if the conformational changes are coupled to the reaction coordinate and associated with coherent (nonBoltzmann) vibrational modes could we attribute special dynamical significance to them. Such a non-Boltzmann coupling has never been demonstrated by Berendsen and co-workers, nor are they likely to be found in the future in reactions with significant activation barriers. Furthermore, essential dynamics studies have not been related to the corresponding reaction coordinate. In fact, essential dynamics in a functional sense is really obtained by our dispersed polaron approach (which tells us about the functionally relevant motions). Although several simulation studies have been interpreted as pointing to dynamical effects in enzyme catalysis, the evidence, in our view, is not very compelling. For example, Radkiewicz

Feature Article and Brooks100 have described ground-state MD simulations of three ternary complexes of dihydrofolate reductase. In these simulations, motions that occurred in the reactive complex disappeared in the product. The authors suggested that these particular motions reflect gate-opening processes with special dynamical significance. It was also pointed out that mutations that slow the reaction also change the rate of conformational changes.94 This is related to the argument94 that if catalysis requires the sampling of a population of active site conformations and only a subset of these conformations are catalytically competent, then by changing the dynamics of this process conformational changes and chemistry would be necessarily affected. However, there are several problems with the above proposals. First, the calculations did not examine the transition states in the catalytic reaction and did not actually demonstrate any dynamical effects on the enzymatic rate constant. Again, as stated above, to show that a given motion of an enzyme plays a dynamical role or that a particular mutation modifies a dynamical effect of this motion, it is essential to establish that the motion has a projection on the reaction coordinate and that modifying the motion changes the transmission factor in a significant way. Second, basically all reactions involve conformational sampling where only some conformations lead to catalysis.52 This is, for example, the case in Figure 2. Changing the fraction of accessible ground-state configurations with high activation barriers is simply a probabilistic entropic effect, which does not qualify to be classified as a dynamical effect. It seems very likely that the different motions of the three ternary complexes of dihydrofolate reductase simply reflect the coupling of enzyme-substrate interactions to enzyme-enzyme interactions, which is common to all proteins. Changes in the motions of the protein are likely to occur during almost any enzymatic reaction in concert with the redistribution of charge in the reacting system. Such changes in motions could contribute to the activation entropy of the reaction but in most cases cannot be considered as dynamical effects. The effects of a mutation on and g must be evaluated by direct simulations of the reactions in the enzyme and solution, and our experience with related systems suggests that these effects will always be associated with g and not with . Also note that large changes occur in the configurations of the solvent in charge-transfer reactions in solutions. This, however, does not mean that there are any dynamical effects. Mutations or other structural modifications at considerable distances from the active site sometimes modify enzyme activities.109 Such effects could, in principle, reflect perturbations of functionally important global motions of the protein. In most cases, they more likely reflect a coupling of g to a free energy change in the distant site. Long-range coupling of free-energy changes has been demonstrated in computer simulations of several systems, including allosteric effects in hemoglobin110 and the effects of GTPase-activating proteins (GAPs) on the activity of Ras.65 In the case of Ras, the GAP was found to stabilize a catalytic configuration in which binding of GDP was favored over binding of GTP.65 In ribonuclease A, the uridyl base of the substrate has an analogous long-range effect that also appears to be energetic in origin.60 Miller et al.111 have given a particularly clear experimental demonstration of a similar energetic effect in orotidine 5-phosphate decarboxylase, where removing the ribose 5-phosphate moiety of the substrate decreases kcat/KM by more than 12 orders of magnitude. All the aboVe effects are associated with changes in the interaction between the TS and the actiVe site and do not reflect dynamical effects.

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7897

Figure 8. Illustrating the nature of nonequilibrium effects by a schematic representation of solute-solvent potential surface. If we fix the solute at R and let the solvent find its equilibrium position, it will get trapped around r3, and in order to pass to the product region, it will still have to climb to r. The corresponding barrier is the nonequilibrium solvation barrier. This barrier cannot be found by PMF calculations since such calculations involve the change of the potential from V(Ri) to V(Ri+1) and the evaluation of eq 23. Now, for example, the step that gives G(R2 f R3 R) involves an average on trajectories with V(R2). Such trajectories will mainly tell us about contributions between r2 and r3 and will not reflect the very rare contributions from r. Thus, G(R1 f R) will underestimate g[(R1,r1) f (R,r)]. The figure also demonstrates the fact that the PMF mapping with V(R f R4) gives contributions from both the reactant region (r r3) and the product region (r r4).

The fluctuations of the electrostatic energy gap in ADH shown in Figure 3 reflect conformational oscillations that clearly are a key factor in the bond-breaking process (see also ref 52). This does not make such fluctuations dynamical factors, neither does it make them catalytic factors, because, as shown in the figure, similar fluctuations occur in solution. Although it is tempting to describe structural fluctuations that occasionally give a configuration in which el , 0 as a time-dependent gateopening, such a description has very little predictive power. On the other hand, using a nondynamical expression to evaluate the Boltzmann probability of finding the system at the TS provides a prescription for predicting rate constants and catalytic effects. 5.3. Nonequilibrium Solvation Effects Do Not Constitute Dynamical Cntributions to Catalysis. Studies of chemical processes in solution have suggested that nonequilibrium solvation can have some effect in the transmission factor for a reaction11, 112 and, more importantly, can contribute to the activation barrier.46,104,113 Although the term nonequilibrium suggests a dynamical effect, it refers primarily to a simple probability factor. The concept is that the solvent must overcome an activation barrier in order to move from the reactant configuration to the product configuration when the solute is in its TS. The barrier is associated with the solvent reorganization energy and is, in fact, the only barrier in outer-sphere electron-transfer reactions, but its contribution to g can be significant in other types of reactions as well.46 The nonequilibrium contribution to g is missing in many treatments of reaction kinetics, including standard QM/MM methods that consider the solvent and solute coordinates separately and allow the solvent to relax at each step along the solute reaction coordinate. This problem is demonstrated schematically in Figure 8 (see also ref 46). As seen from the figure the solvent coordinates might not be at their TS value, r, when the solute coordinate is at its TS value, R. Regular QM/MM approaches

7898 J. Phys. Chem. B, Vol. 105, No. 33, 2001

Villa ` and Warshel

Figure 9. Demonstration of the nonequilibrium solvation effect in (a) the catalytic reaction of subtilisin and in (b) the corresponding reaction in water. The upper part of each figure depicts the diabatic free energy functions for the case where the solute coordinates are fixed at R. The intersection of these curves gives the nonequilibrium solvent reorganization effect (g dia(R)R ) ne/4 for G0 ) 0). The light gray background insets give the nonequilibrium solvation adiabatic free energy profiles and the corresponding nonequilibrium solvation free energies of activation, g ne. In the case of the reference reaction in water (b), the adiabatic free energies are evaluated by using eq 10 with R ) R . In the case of the reaction in subtilisin, the nonequilibrium free energy surface is relatively flat and eq 10 does not provide a sufficiently good approximation (this case could be better treated with the mapping approach used in eq 11 in ref 29). Fortunately, since the surface is flat, we can evaluate the adiabatic free energy by simply running a long trajectory on Eg(R)R) and evaluating the probability of being at different values of .

evaluate the free energy profile by determining the corresponding PMF. This is done by running MD simulations on potentials that constrain the solute coordinate to different values and using the relationship

h is some arbitrary where rmin is the minimum of gne and A constant that cancels in eq 24 (and that can be chosen to be, for r)). With this expression we can use example, equal to gne(j the approximation
g gPMF + g ne

g(R1 f R2)PMF ) --1 lnexp[-[Eg(R2,r) Eg(R1,r)]]Eg(R1,r) (23)

Now, the free energy of reaching R can be lower than the free energy of reaching the correct TS (R,r) (see discussion in the caption of Figure 8). The EVB/FEP/US of section 3 automatically determines the correct activation barrier (g(R,r)) without missing the solvent contribution at R ) R. This is due to the fact that the EVB mapping potential forces the solvent coordinate to change by changing the solute charge (rather than only as a response to the change in the solute coordinate) and also because the function in eq 10 sorts the probability distribution in terms of the energy gap that reflects both the solute and the solvent coordinates. Using the EVB approach, it is also relatively simple to determine the nonequilibrium solvation effect. This is done by constraining the solute to be at R and performing our EVB mapping procedure. The nonequilibrium activation barrier is now given by
g ne ) gne(R ,r ) - gne(R ,rmin)




r) ) --1 ln(j r - r) exp{-[Eg(R,r) - A h ]Eg(R,r) gne(j (25)

Figure 9 demonstrates the evaluation of g ne for the nucleophilic attack of the ionized Ser221 on the carbonyl carbon of the substrate in the catalytic reaction of subtilisin (see ref 2 for a discussion of this reaction). The figure gives the diabatic free energy function for R ) R and also the corresponding adiabatic free energy (see figure caption). As seen from the figure and, particularly, from the diabatic free energies, the nonequilibrium diabatic barrier is lower in the enzyme than in solution. The corresponding nonequilibrium adiabatic barrier is 4 kcal/mol in the water reaction and 1 kcal/mol in subtilisin. Basically, the reduction of g ne is a manifestation of the reduction in the reorganization energy in the enzyme (see also below). Neria and Karplus11 have used EVB simulations to study dynamical effects in the proton-transfer step in triosephosphate isomerase. Their studies focused mainly on the transmission coefficient (), which they calculated by the reactive-flux method. More importantly, from the perspective of the present work, they concluded that, unlike charge-transfer reactions in polar solvents, the barrier modulation is not due to nonequilibrium solvation by the environment; [contrarily] the height of the barrier is determined by the configuration of the intramolecular subsystem whose dynamics is coupled to lowfrequency fluctuations in the enzyme.11 This and their discussion (see also below) might have created an impression that nonequilibrium solvation effects lead to dynamical effects that

Feature Article are very different in enzymes and in solution. Unfortunately, ref 11 has not compared the simulations in TIM to the EVB simulations of the corresponding reaction in water. Instead, it compared the EVB results for TIM to those obtained by Hynes and co-workers112 for an SN2 reaction in water. This SN2 study was performed with an inconsistent SGP38,39 potential energy surface that did not consider the coupling of the solute charges to the electric field of the solvent (see Hwang et al.10 for additional discussion of this point). Other fundamental problems with the frozen bath approach112 will be considered below. Basically, the problem with the analysis of Neria and Karplus is the focus on the very small effect of nonequilibrium solvation on , rather than on the much larger effect on g To further clarify that the so-called nonequilibrium solvation effects do not reflect significant dynamical contributions, it is useful to reconsider the proposed relationships between to the nonequilibrium solvation.112 That is, ref 11 assumed that the TST is based on the potential of mean force and that thus the additional effect due to nonequilibrium solvation is given by . Combining this assumption and the so-called frozen bath approximation112 lead to
ln k ) ln froz - gPMF + ln(1/h)

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7899

(27) (28)

froz ) e-Vr

r)), where R is the point where Here V(r) ) (Eg(R,r) - Eg(R,j V reaches a maximum for the given fixed r, and j r is r only in the special case of a frozen solvent model (see Figure 10b) (in other cases the prescription for choosing j r is not welldefined). In other words, R is the highest point for moving from R toward R4 in Figure 8. Now the above expression involves several problems. First, k is determined by g (that includes g ne) and not by gPMF. This is due in part to the fact that eq 23 does not have the (r - r) term of eq 10 so that it might underestimate g ne. Thus, the correction to the PMF is not ln froz. Second, lne-V is not equal to -g ne, which is the proper correction to the - gPMF term (see eq 26). The problem will be particularly serious when Hij is small (the extreme case being electron transfer-type reactions). Now in the special frozen solvent case when r3, r and r4 are almost identical (Figure 10), froz can be evaluated by eq 28, but this case is not so relevant to most charge-transfer reactions. Furthermore, even in the frozen solvent case, what is attributed to in eq 28 is related to the width of the bottleneck region. This is customary related to the entropic effects13 associated with the reaction coordinate. It is not obvious that this correction is identical to the . In fact, it is not clear that has a simple correlation with the nonequilibrium solvation effect. For example, as seen from Figure 7 the solute and solvent coordinate change in a similar way, and this occurs both in the protein and in solution. In particular, the minor recrossing effects occur in a similar way in the enzyme and solution cases. This recrossing has apparently little to do with g ne, since this barrier changes significantly between the water and protein cases, while the recrossings are similar. Thus, we conclude that if the nonequilibrium solVation effect is significant, it is due to g ne and not to dynamical effects. Cannon et al.95 have discussed the fact that the solvent reorganization energy plays a major role in chemical reactions and have suggested that enzymes may catalyze reactions by removing slow components of the reorganization energy of the

Figure 10. Illustrating the frozen solvent limiting case and the approach used in eq 28. Here the solvent is frozen at its transition-state structure and the potential surface for R ) R has a minimum at r. Now the nonequilibrium solvation effect is actually zero. Here the free energy correction to the least energy path is given by eq 28. It is not clear to us that this correction is exactly equal to even in the frozen solvent limit (see text). However, in the more general case of Figure 8, it is not clear how to evaluate eq 28 and, more importantly, this equation does not tell us much about the nonequilibrium solvation effect (see text).

solvent coordinate. Obviously, a reduction of the solvent reorganization energy can clearly contribute to enzyme catalysis (and this is what was probably meant by Cannon et al.). However, as much as dynamics is concerned, it appears to us that in most cases this reflects a decrease in g, not a dynamical effect. Experimental studies of solvation dynamics have shown that the motions associated with reorganization of the first solvation sphere in water are extremely fast.114, 115 The same fast motions are also found in simulations of reactions in enzymes and solutions (e.g., refs 2 and 29; see also Figure 5). The dynamical effects are determined by the fast motions (they determine the autocorrelation decay) and the fast motions are similar in enzymes and in solutions. Thus, removal of slow motions does not change the dynamical effects, although it changes g. Overall, the results discussed above indicate that the thermal fluctuations of solvent and reactant coordinates probably are similar in enzymes and solution, in accordance with the view

7900 J. Phys. Chem. B, Vol. 105, No. 33, 2001 that dynamical effects do not play a major role in enzyme catalysis. Nonequilibrium solvation effects evidently are simply one aspect of the reduction of g in the enzymes active site, rather than a reflection of special dynamical effects. As stated in section 4, the solvent reorganization energy is probably reduced in most enzyme active sites, where a preoriented polar environment appears to be the main key to catalysis.2 5.4. Vibrationally Enhanced Tunneling Does Not Provide a Major Catalytic Advantage. Recent studies have suggested that vibrationally enhanced tunneling (VET) plays a major and special role in enzyme catalysis. In fact, the excitement about this effect may have led sometimes to the impression that we have here an entirely new effect that makes TST inapplicable to enzymatic reactions.116 However, the VET effect is not new and is common to many chemical reactions in solution. This effect, which has been introduced and conceptualized in terms of solvent fluctuations in the work of Warshel,51, 117 Hynes,118 Bialek,119 and co-workers as well as in the works of others,120 is strongly related to TST. That is, when the solvent fluctuates and changes the energy gap (see Figure 3) the light atom sees a fluctuating barrier that allows in some cases for the larger rate of tunneling. As shown in refs 51 and 52 these fluctuations are taken into account in the statistical factor in the classical TST and the same is true when quantum effects are taken into account. Thus, the recent finding that the solvent coordinates should be considered in tunneling studies is not new and does not mean that this effect is important in catalysis. At any rate, it is obviously important to explore the contributions of quantum mechanical tunneling to enzyme catalysis. To examine the question if vibrationally enhanced tunneling provides a major catalytic advantage in enzyme reactions, we should properly address this issue. In particular, we need to find out whether tunneling is important for catalysis and whether there is any evidence of coherent vibrations coupled to the reaction coordinate. Furthermore, we should determine whether those coherent vibrations are essentially different in water and in protein. These issues will be considered below. As pointed out in the previous sections, quantum effects (in particular tunneling effects) are clearly important in reactions involving transfer of light particles. This is the case, of course, for electron-transfer reactions, and to a great extent also in hydrogen-transfer reactions (H, H+, or H-) quantum effects should be taken into account. Experimentally, traditionally kinetic isotope effects (KIE) have been used as proof of the manifestation of quantum effects in a given reactive process. The idea is that quantum effects are demonstrated when some approximated rules,121 obtained from algebraic manipulations of the general TST expression, are not fulfilled. In particular tunneling effects can be manifested by122

Villa ` and Warshel isotope effect of the hydride-transfer step in the reaction catalyzed by ADH.109 However, a note of caution must be taken when applying eq 29 to secondary KIEs. The value of primary KIEs is usually several times bigger than the values of secondary KIEs, which are usually close to 1. Thus the use of the Swain-Schaad rule, as a proof of the presence of tunneling effects, must be taken with care. The proper computational exploration of KIEs for a given process should involve evaluations of the quantum corrections of the rate constants of this process for the relevant isotopes. Several semiclassical techniques, where quantum corrections are added in some way to classical simulations, have been developed in the past few years for the inclusion of quantum effects in reactions in the condensed phase.24,108,123-128 Computer simulations of quantum mechanical nuclear effects in electrontransfer reactions in solution and proteins have started in the mid 80s.105 These and subsequent studies106, 107 have been based on the dispersed polaron-spin boson approach105,106,123,129 and exploit the fact that the off-diagonal electronic coupling (the H12 in the EVB notation) is small. The simulation of quantum corrections for adiabatic reactions such as PT and HT is more challenging since the off-diagonal are large (see discussion in ref 117). The first realistic simulation of quantized PT and HT in solutions and proteins were performed by Warshel and coworkers,108 ,117 who formulated different effective ways of obtaining the quantum mechanical activation free energies for adiabatic reactions in the condensed phase. These and subsequent studies made use of the centroid path integral approach introduced by Gillan126 and extended by Voth and co-workers.130 However, the emphasis in the development of Warshel and coworkers was placed on the evaluation of activation free energies and the introduction of FEP/US approaches to centroid calculations of adiabatic reactions. This led to the development of the quantized classical path (QCP) method,108 ,127 which uses the configurations obtained from classical simulations as a reference potential for centroid calculations (see below). The efficiency and stability obtained by this approach was exploited in studies of quantum mechanical rate constants in condensed-phase reactions.127, 131 In particular, the QCP has been applied successfully in the inclusion of quantum effects in calculations of rate constants of enzymatic reactions.128, 132 Studies of reactions in the condensed phase that involve the centroid approach without the QCP trick have been reported in recent years (e.g., refs 133 and 134). Recently, Alhambra et al. have successfully applied VTST/ MT to the calculation of KIEs of the hydride-transfer step in the LADH-catalyzed reaction.42 The dynamical effects in that calculation (as seen from the comparison of TST and VTST calculations in Table 1 in ref 42) were found to be extremely small. However, the VTST/MT approach, in its current form, does not yet sample the fluctuations of the protein. Obviously, this makes it hard to examine possible dynamical effects of such fluctuations. The protein fluctuations are taken into account automatically by the QCP approach although the transmission factor is evaluated classically. Other promising approaches for treating quantum effects in condensed phases were reported recently.135 ,136 Here we will focus, however, on the QCP approach. Since the details of the centroid path integral approaches and the QCP version have been given elsewhere,127 ,133 we will only emphasize below the main points. The starting point of the QCP approach is the approximated expression126, 130


kH/kT > 3.3 kD/kT


where kH is the rate constant for the nonisotopic reaction and kD and kT are the rate constants obtained by substitution of protium by deuterium and tritium, respectively. The value of 3.3 is the value predicted by the Swain-Schaad rule in the absence of tunneling effects. Equation 29 has been applied to both primary and secondary KIEs. Klinman and co-workers have made extensive use of the Swain-Schaad relationship,121 and they have been successful in demonstrating the presence of tunneling effects in some enzymatic reactions, and in particular, they have found an unusually high secondary kinetic

Feature Article

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7901

kq )

q exp(-g q) h


where kq is the quantum mechanical rate constant and g q is the quantum mechanical activation free energy. The problem is, of course, the evaluation of g q. In general, one can calculate some quantum mechanical properties by Feynmans path integral formulation,137 where each quantum particle is represented by a ring of P quasiparticles (beads) that are subjected to the effective quantum mechanical potential.

Uq )

M2xk2 + U(xk) P k)1 2P


where xk ) xk+1 - xk, ) P/(p), M is the mass of the particle, and U is the actual potential surface of the system. In addition, xk+P ) xk. The total quantum mechanical partition function is obtained by running classical trajectories of the quasiparticles with the potential Uq. The problem is that there is no rigorous way of obtaining the probability of being along a reaction coordinate. Fortunately, an excellent approximation was obtained by Gillan126 and expanded by Voth and coworkers.130 In this approximation one finds the probability of being at different points along the reaction coordinate by evaluating the probability distribution for the center of mass of the quasiparticles (the centroid). In some respects, the introduction of the centroid approach is similar to the extension of the TST expression to a larger configuration space that includes the degrees of freedom of the quantum chain. Note also that the probability of accessing the transition-state region is larger when obtained with the centroid approximation than when using only the classical expression, because now every quasiparticle experiences only the potential 1/PU(xk) rather than U(x0). Although the above centroid path integral approach is formally simple, it does not tell us how to determine the probability distribution. This is particularly serious when we have high activation barriers in multidimensional systems so that trajectories at the reactant state never reach the TS. The QCP approach provides a very robust way of overcoming this challenge. Instead of pruning trajectories on all the quasiparticles, we propagate classical trajectories (of the classical particles only) on the classical potential surface and use the positions of the classical atoms to generate the centroid positions for the quantum mechanical partition function. This treatment, which has been introduced in ref 108 and verified in ref 127, allows one to use the same efficient FEP/US approach used in regular EVB classical studies (see section 3). Thus, the main point of the QCP approach is that it allows one to evaluate the quantum mechanical free energy profile by a centroid approach, which is constrained to move on the classical potential. Apparently, the QCP provides a very convenient and powerful way of calculating KIEs for reactions with large activation barriers in the condensed phase. This is done by running the centroid calculations using different isotopic substitutions on the same classical trajectory, which saves great amounts of computer time.128, 132 Similar information can be obtained by the VTST/ MT approach,138 ,139 but the QCP offers a simple way of exploring the effect of the protein fluctuations on kq and the corresponding KIEs. Obtaining this information by VTST/MT is not so simple. Finally, we should note that the QCP approach

Figure 11. Snapshot of a QCP calculation in the TS of the hydride transfer from NADH to benzaldehyde (BA) in LADH. In the EVB/ QCP calculation all region I atoms are treated quantum mechanically by the path integral formalism, while the rest of the protein + solvent is treated classically. These include the atoms in the nicotinamide part of NADH (that becomes NAD+ during the reaction) and the substrate benzaldehyde (that becomes benzyl alcoholate during the reaction studied). This is a total of 31 atoms (of course, all the protein atoms are included in the classical MD simulation). Each quantum particle has been split into 20 quasiparticles (or quantum beads). As seen in the figure, the quantum beads are more spread in lighter particles (H) than in heavier particles (C, N, and O), due to the direct dependence of the force constant connecting the quasiparticles on the mass of the classical particles. The calculation has been performed both in water and in protein, but only the snapshot in the protein is shown.

Figure 12. QCP free energy profiles. The figure presents preliminary results of the quantized free energies (open symbols) and the corresponding classical results (bold symbols) for the hydride-transfer step in the ADH-catalyzed reaction circles and for the corresponding reaction in solution (squares). The results have been plotted in reference to the same zero of energy for clarity. As seen from the figure, preliminary results yield essentially the same reduction in g in water than in protein due to quantum effects. Additional calculations testing the influence of the number of beads and calculating kinetic isotope effects are in process in our lab.

assumes that the classical and quantum mechanical transmission factors ( and q, respectively) are identical. The range of validity of this approximation requires further studies. Figures 11 and 12 depict preliminary results obtained with the QCP approach for the reaction in LADH. Centroid calculations reflect all the so-called dynamical induced tunneling effects, yet nothing special to catalysis when comparing the same

7902 J. Phys. Chem. B, Vol. 105, No. 33, 2001 reaction in water and in the protein active site (see also ref 128). In view of this and related studies132 we must question the impression that recent experimental studies have provided evidence for the importance of quantum mechanical nuclear effects in enzyme catalysis.109,116,140 It seems to us that at present we do not have any experimental or theoretical evidence of a significant quantum mechanical contribution to enzyme catalysis. More specifically, the instructive experiment of Bahnson et al.,109 which studies different mutations in LADH in order to unmask the tunneling effect of the hydride-transfer step, has not shown any significant difference between the kcat for the different mutants. This is significant because of the observation that the distance between the donor and the acceptor atoms varies for the different mutants. Interestingly, the increase of kcat/KM with the decrease of the distance between the donor and the acceptor was brought as a support for the idea that the enzyme works by pushing the donor and acceptor together.141 However, this assertion overlooked the fact that the chemical step is entirely determined by kcat and kcat is basically constant in the above mutants (the relevant catalytic effect is associated with g cat). In a related theoretical study of the hydridetransfer reaction catalyzed by formate dehydrogenase, Torres et al.142 used MD simulations and showed a large dispersion of donor-acceptor distances in the ground state. Unfortunately, they have not studied the same reaction in water, which would tell us if the different dispersion of distances in the two environments have something to do with enzyme catalysis. The relationship between the activation barrier and the fluctuations is a well-known fact, as has been pointed out repeatedly through this and our earlier papers. Finally, it is useful to comment about a recent theoretical study of Antoniou and Schwartz,143 where it was implied that dynamical effects are important in enzyme catalysis. Unfortunately, this interesting work, that invoked some of the dispersed polaron123 ideas, studied aspects of hydride transfer in acetonitrile rather than in enzymes. It is also useful to mention here cases where a crossover temperature from the thermally activated over the barrier regime at relatively high temperatures to thermally assisted quantum tunneling through the barrier at relatively lower temperatures exist. For example, in the B-H complex in silicon, this temperature has been shown to be around 60 K.134 Interestingly, the trend in these studies of highly ordered systems is opposite to the trend proposed by the recent works that are being used as a proof of the importance of VET in enzyme catalysis. Thus, it is unlikely that the enzymes act in the same way as highly ordered solids do, and if this would be the case one would expect the temperature crossover related with this change in regime to be even lower than 60 K. In conclusion: fluctuations exist and they can bring the enzyme-substrate system to situations where the energy barrier decreases or even becomes zero (see, e.g., Figure 2). The tunneling contribution becomes more important when the barrier is lower, and in the enzyme the barrier to cross is lower because of preferential binding of the transition state relative to what happens in solution. This would lead us to think erroneously that tunneling contributions should be more important in enzymes than in solution, but this overlooks the fact that first there exists a barrier decrease (the actual reason for catalysis) and only then a minor side effect will be the slightly higher tunneling effect that would be obtained in the enzyme relative to solution. 5.5. Temperature Dependence of Activation Barrier Does Not Reflect Dynamical Contributions to Catalysis. Kohen et al.4 have examined the temperature dependence of kinetic

Villa ` and Warshel isotope effects on kcat in alcohol dehydrogenase from a thermophilic microorganism, Bacillus stearothermophilus. Primary and secondary isotope effects were seen, and both decreased with increasing temperature. (A primary isotope effect in this case is an effect of substituting deuterium or tritium for the hydride that is transferred from C1 of ethanol to NAD+; a secondary isotope effect is an effect of modifying the other hydrogen bound to C1.) This is in accord with previous studies of alcohol dehydrogenase from mesophilic organisms. With the B. stearothermophilus enzyme, however, both the primary and the secondary isotope effects gave a larger ratio of ln(kH/kT) to ln(kD/kT) than predicted by eq 29, and the ratio calculated from the secondary isotope effects increased with temperature.4 The authors interpreted the latter observation to mean that hydride tunneling becomes increasingly important at elevated temperatures, which is opposite the pattern usually seen in enzymes from mesophilic organisms. They concluded that protein fluctuations are crucial for catalysis and that, in the enzyme from B. stearothermophilus, the fluctuations needed for effective tunneling increase with temperature. Kohen et al. suggested further that a tuning of particular vibrational modes could explain why enzymes are more effective than catalytic antibodies. As mentioned above, however, the magnitudes of the primary isotope effects, which should reflect the actual quantum corrections to the rate constant from tunneling and zero-point energies, decreased with temperature. This seems difficult to reconcile with a major role of thermally induced quantum mechanical effects. Kohen et al.4 also noted that the activation enthalpy (H) for the B. stearothermophilus alcohol dehydrogenase reaction decreased with increasing temperature between 283 and 333 K. They viewed this as supporting the view that tunneling becomes more important at high temperatures. The activation free energies (g in our notation), however, were almost independent of temperature. (Kohen et al. did not give values for g, but these can be calculated from the data in their Figure 4 by the expression g (12.8 - log kcat)(4.6 10-3)T, which follows from the TST expression. The calculated values are 15.0-15.6 and 16-17 kcal/mol for the protonated and deuterated substrates, respectively.) The decrease in H with temperature evidently is accompanied by a compensating increase in -TS. Such changes in H andS are consistent with hydride transfer from a bound ethanolate ion to NAD+. Because the TS probably is less polar than the reactant state (see Figure 13) S is expected to be positive, making -TS negative. Raising the temperature is likely to decrease both H and S, because thermal motion opposes both the tight binding of the ionic substrates and electrostatic polarization of the protein and solvent. It appears to us that the weak temperature dependence of the kinetics probably can be explained in this way without invoking special dynamic effects. In another recent study, Scrutton and co-workers5 observed a primary deuterium isotope effect of about a factor of 17 on the hydrogen-transfer reaction catalyzed by a mesophilic methylamine dehydrogenase. This exceptionally large effect was independent of temperature, although the rates of the reaction with both the protonated and the deuterated substrate increased with temperature. Basran et al.5 concluded that the reaction occurs by hydrogen tunneling driven by thermally induced vibrational motions. Tunneling presumably occurs mainly when the donor-acceptor distance is minimal. Raising the temperature thus could possibly facilitate tunneling by increasing the amplitude of the vibrations. However, it is much more likely

Feature Article

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7903 However, we must keep in mind that the effect in the enzyme should be compared to the corresponding effect in solution. Without doing so, one cannot ask any question about enzyme catalysis. Finally, we might rationalize all the above arguments about dynamical effects by pointing out back to the rate expression of eq 1. Since the activation barrier g is in the exponent, it is much simpler to change the rate constant by changing g than by using dynamical effects through the preexponential term . 6 What Is the Role of Entropic Effects in Enzyme Catalysis? 6.1. Defining the Problem. Many proposals37,40, 144-150 invoked entropic contributions as major factors in enzyme catalysis. These proposals, which are intuitively appealing (e.g., see ref 150), have assumed that the large configurational space available for the reacting fragments in water would be drastically restricted in the enzyme active site. It has been thus deduced that this should lead to large entropic contributions to the difference between activation barrier in the enzyme and in the reference solution reaction. However, the validity of these proposals is far from being obvious.2,3,151 First, the relationship between the binding entropy and the entropic contribution to catalysis has not been correctly defined in most proposals and the implicit assumption that the entropic contribution to catalysis is equal to the negative of the binding entropy is not justified (see below). Second, the influential proposal of Page and Jencks152 that the formation of the transition state in bimolecular reactions in solution involves the complete loss of three rotational and three translational degrees of freedom is not justified.153, 154 Apparently, two or more of these degrees of freedom are almost free at the transition state.151 Unfortunately, the early entropic proposals have not involved a proper thermodynamic cycle. That is, to determine the contribution of the substrate entropy to catalysis we have to find the difference between the contribution in the enzyme, (Scat), and in water, (Sw). Instead of considering free energy diagrams that involve these contributions (see below), it has been customary to talk in a vague way on the negative of the binding entropy as the entropic contributions to catalysis. This approach seems to be implicitly assumed in a recent implementation by Kollman and co-workers.37,40 These workers used the energy diagram in Figure 13a, where they apparently considered the following quantity:
g cat ) Hgas + (gsol,p)cat g w ) Hgas + (gsol,w) + Gcratic

Figure 13. (a) Thermodynamic analysis of the data of Kohen et al.4 and (b) a consistent rationalization of these data. As shown in the upper panel the activation free energy is temperature independent while the changes of H and -TS compensate each other. The lower panel rationalizes the above trend by pointing out that protein dipoles are more free to fluctuate in the nonpolar reactant state than in the partially polar transition state. Increasing temperature increases the thermal fluctuations in both transition state and reactants states. This picture is consistent with the decrease of |TS| (see text).


that the temperature dependence of the rate has little to do with the frequency or amplitude of a particular vibrational mode, but rather simply reflects the Boltzmann factor for populating a configuration in which the hydrogen can tunnel from the donor to the acceptor with no change in energy. Furthermore, as we emphasized above and in clear contrast to the implication of Scrutton and co-workers,5,116 the fluctuations that help the tunneling process are not at all special to enzymes. Path-integral simulations that include nuclear tunneling indicate that quantum nuclear effects are fundamentally similar in enzymatic reactions and in solution and probably do not contribute greatly to enzyme catalysis.108,128,132 In making this statement we are not saying that quantum effects are not interesting, the opposite is true. These effects in proton- and hydride-transfer processes have been studied by us and others by microscopic simulations24,42,108,128,132 and further studies should be instructive.

Here, H gas is the activation enthalpy of the reaction in the gas phase, while g sol,p and gsol,w are the contributions of the solvation free energies of the gas-phase charges to the indicated gq for the reaction in the protein and in water, respectively. The Gcratic term has different definitions (e.g., refs 37, 154, and 155), but ref 37 considers this term as the work necessary to constrain the reacting molecules in a productive geometry, or equivalently, the free energy contribution obtained by restricting the motion of the reacting system in solution to the same space they retain in the enzyme. The problem is not with this definition but with the thermodynamic considerations of Figure 13a. Unfortunately, Figure 14a and eq 32 do not consider properly the activation entropies in the enzyme and in solution, and it is assumed implicitly (without stating so) that the motion in the transition states in water are frozen. The best way to see

7904 J. Phys. Chem. B, Vol. 105, No. 33, 2001 the problem is to describe the entropic effect by a correct thermodynamic treatment as it is done in Figure 13b. The correct treatment gives
g cat ) Hcat + (gsol,p)cat - T(S TS,p - S RS,p) g w ) Hw + (gsol,w) - T(S TS,w - S RS,w)

Villa ` and Warshel


Here, the H correspond to the reaction in solution and the gsol to the actual charges in the given environment rather than in the gas phase. The solute entropy, S, in water and protein are denoted here according to the corresponding system (protein and water) and state but in the case of the water reaction we use the notation S RS,w to designate the entropy of the reacting fragments in their standard state. We also find it convenient (but not essential) to use the relationship:
S TS,w - S RS,w ) Scage - RT ln(Vcage/V0)


where Vcage is the volume of a water cage where the reacting fragments are at interaction distance and V0 is the molar volume. This definition allows us to separate the trivial fact that the reactants in the enzyme active site are at a contact distance (this is the effect of being at 55 M concentration, which was known by all early workers in the field) from the much more controversial orientational effects (see below). Now, according to eq 33, the contribution of S to g cat is
-T(S) ) -T[(S cat) - (Sw)]

) -T[(S TS,p - S RS,p) (S TS,w - S RS,w)] ) -T[(S TS,p - S TS,w) - (S RS,p - S RS,w)] ) -T[(S TS,p - S TS,w) - S bind] (35)

Figure 14. Different treatments of the entropic contributions to enzyme catalysis. (a) The free energy considerations of ref 37. These considerations reflect the implicit assumption of early workers144 that the entropic contribution to catalysis is given by the binding energy (which is the main contribution to Gcratic according to the definition of ref 37. gcat is assumed to have no entropic contribution while g w is assumed to involve the cratic entropy (see text). (b) A consistent and complete analysis of the entropic contribution to catalysis from the solute motions. gp and gw are the free energy functions for the reactions in enzyme and in solution respectively, while g p and g w are, respectively, the free energy functions for the protein and solution reactions when the solute motions are frozen (see text and ref 151 for more details).

Thus, we establish that the expression used in eq 32 overestimates the actual entropic contribution by -T(S TS,p STS,w ). In other words, the assumption that Sbind (which is equal to Scratic in Kollmans definition) gives the entropic contribution to catalysis, misses the transition-state contributions. Apparently, S TS,w is quite large and can be similar to S RS,w or - S and thus cancel most of its effect (see below). bind In other words, the assumption that the entropic contribution to catalysis is given by the binding entropy or related quantities is unjustified. It basically misses the fact that the activation entropy in water can be much smaller than the binding entropy since the transition state in water usually has much larger entropic contributions than the transition state in the enzyme (the two are simply unrelated). 6.2. Actual Estimates of (S). Despite the fact that most entropic proposals have been poorly defined, these proposals might be still correct and thus it is crucial to determine how large (S) is. However, it is not so clear how to accomplish this experimentally since the observed activation entropies include solvation contributions. Thus, it is essential to use theoretical approaches to determine (S). This should be done on specific test cases, considering the actual potential surfaces of the system studied. The first attempt to consistently determine the entropic contribution to enzyme catalysis was reported in ref 2. This was done for the nucleophilic attack step in subtilisin by determining the configurations with energy of less than -1 on the corresponding surfaces and by determining the entropy associated with forming the solvent cage. Kollman and co-workers appear to overlook this work and argued that

Warshel has not considered the Gcratic, arguing that it was appropriate to consider a preorganized reference state for both solution and enzyme-catalyzed reactions.40 Actually, not only the cratic entropy is not the relevant quantity (see above), but more importantly, the orientational entropy in the solvent cage has been calculated explicitly in ref 2 and the reacting fragments in this cage were never preoriented. Similarly, our free energy simulations involve sampling the motions within the reactants state in the solvent cage and in the enzyme. Stanton et al.37 recognized the importance of evaluating the entropic contributions but have not developed approaches that allow one to consistently calculate these contributions, and end up overestimating their values. That is, ref 37 used MD runs to estimate the force constants necessary for keeping the system in the reactant conformation with a rather arbitrary standard deviation of 0.2 and 20 for distances and angular dependence, respectively. These force constants were used for deriving the entropy contributions using equations derived by Hermans and Wang.155 However, this does not amount to the evaluation of the ground-state entropy but rather to assuming the corresponding effective volume. There is a direct relationship between the volume and the entropy that does not require any force constants. The use of the relationship between the volume (or the corresponding force constants) and the entropy does not amount to using Hermans approach, which is actually a restraint release approach (see below). Stanton et al.37 have also used very reasonable qualitative considerations to estimate - TS bind (the solute contribution to the cratic entropy of ref 37). However, as shown below, S bind is not equal to -Swfcat. Reference 40 used gas-phase entropy calculations to estimate the entropy cost of bringing the reactants in water to the corresponding configuration in the enzyme-substrate (ES) complex. However, the gas-phase vibrational entropy may be drastically modified

Feature Article
TABLE 1: Summary of Proposals for the Origin of Enzyme Catalysis
mechanisms desolvation144,160-162 strain1,144 entropy40,144 orbital steering165 dynamics4,5,96,99 LBHB166, 167 electrostatics16,77 tunneling4,5,122 diffusion169 ts stabilization + + + consistent with mutations ? ? ? + + + +

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7905

supported by simulations ? + -

inconsistencies a b c d e f h h

quantitative analysis 75, 163 2, 16 2, 151, 164 2 2, 29 168 2, 76, 75 108, 128 58

a Groups that are supposed to be ionized are not ionized in the assumed nonpolar environment, and the desolvation energy is not taken into account.2,75 b Enzymes are too flexible.144 c Original support from model compounds was found to be questionable2,164 and irrelevant,2,151 while simulation studies show that the actual catalytic effect is small.151 d The bending force constants at the TS are much smaller than the value needed for the proposed effect.2 e Computer simulations produce very similar dynamical effects in enzymes and the corresponding solution reactions (this work, and refs 2 and 52). f Shown to be anticatalytic due to loss of solvation energy.168 g Simulation studies produce a very small catalytic effect.128 h Observed diffusion effects are negligible, contributing around 1 kcal/mol to the reduction of g in cases where the overall reduction is larger than 10 kcal/mol.58

in solution. More importantly, the assumption144 that the estimated entropy is equal to -Sw fcat is not justified. Systematic evaluation of (S ) is extremely challenging due in part to convergence problems. To overcome this challenge, we developed recently a restraint release approach (RRA)151 ,156 that has some elements in common with the approach used by Hermans and Wang in studies of binding entropies.155 Our approach evaluates S for the RS and TS of the given system by imposing Cartesian restraints on the system and then evaluating the free energy associated with the release of these restrains. The resulting free energy (G) includes entropic and enthalpic contributions but the enthalpic contribution makes it hard to evaluate the entropic part. Fortunately, H can be eliminated by performing series of simulations with different restraint coordinates and finding the one that yields the smallest G.156 The application of our approach to the catalytic reaction of subtilisin produced much smaller entropic contributions to catalysis than usually assumed.151 This is due to the fact that many of the motions that are free at the RS in the reference reaction are also free at the TS (S TS,w * 0). Furthermore, the binding to the enzyme does not completely freeze the motion of the reacting fragments. The above point can be nicely illustrated by considering the reaction of serine protease that involves a proton transfer from a serine to histidine and a nucleophilic attack of the ionized serine on the carbonyl of the substrate. It is natural to assume that the histidine residue loses its motion at the TS in solution. However, the protonated histidine can move as freely at the TS as the unprotonated histidine at the RS. In summarizing this section, we note that although entropic contributions to catalysis cannot be ignored, the magnitude of these contributions is smaller than previously assumed. The actual evaluation of such contributions requires both careful definition of the relevant quantities and very challenging simulation approaches. 7. Concluding Discussions In the previous sections we demonstrated the use of simulation studies in examining the origin of enzyme catalysis and in examining the importance of several catalytic proposals. Because of space limitation we will not discuss other catalytic proposals and only summarize the corresponding considerations in Table 1. The table lists different proposals and considers their feasibility in view of the available experimental and theoretical studies. This table is not brought as an attempt to discredit proponents of different proposals but to point out that each case can be examined using well-defined energy considerations and

computer simulations. Although more simulation studies are clearly needed, we feel that they provide the most effective way of finding out whether a given experimental finding can be used to support a specific catalytic proposal. It seems to us that the main problem in reaching concensous in the search for the origin of enzyme catalysis is the attempts to address this issue without well-defined energy considerations. Unfortunately, there is no way to resolve the open questions in this field without some form of structure-energy correlation. As an example to this problem, we can take a recent work5 that attributed the fact that catalytic antibodies are much less effective than enzymes to the corresponding difference in dynamical effects. Here the starting point might be the observation that many people find it hard to understand why catalytic antibodies, which are supposed to work by TS stabilization, are less effective than enzymes. Accepting this presumed problem as a real problem may lead one to propose that enzymes must use a trick other than TS stabilization and the best candidates are dynamical effects. Of course, such assumptions ignore two major points. First, it is quite easy to find the reason for the ineffectiveness of catalytic antibodies while using TST (see below). Second, unless we actually calculate the catalytic effect in antibodies and in the corresponding enzymes, we have no base for invoking any special non-Boltzmann effect as the origin of the difference between these systems. In fact, the problems with catalytic antibodies can be easily understood once we use energy-based concepts. Basically, we do not have good TS analogues. That is, most enzymatic reactions involve several transition states of similar activation energies, and the enzyme must reduce these barriers simultaneously. This is done by evolution where the active site is designed to stabilize the different transition states. In constructing the corresponding catalytic antibodies, one raise antibodies to what one thinks is a proper TS analogue. Now, not only that the selection of the proper charge distribution of the TS analogue is not obvious but more importantly there is no single TS analogue that possess the charge distribution of two TSs simultaneously. Thus, the antibody that is elicited against an analogue of one TS cannot stabilize the second TS in an effective way. In summary, we emphasize that the search of the origin of the catalytic power of enzymes must involve dissection of partial energy contributions, which is very hard to accomplish experimentally. Thus, the best way to accomplish this is by computer simulation approaches. At present it appears from the accumulating simulation studies that electrostatic effects are the primary source of enzyme catalysis. Other factors such as entropic effects might still play a significant role in reducing

7906 J. Phys. Chem. B, Vol. 105, No. 33, 2001 g cat, but the largest effect is probably due to the preorganized polar environment of protein active sites. Acknowledgment. This work was supported by the NIH grant GM24492. J.V. acknowledges EMBO fellowship ALTF 509-1998. We are grateful to Dr. William Parson for insightful discussions and colaboration. We also acknowledge Dr. Mats Olsson, Dr. Marek S trajbl, Dr. Gongyi Hong, Dr. Jan Floria n, and Ms. Claudia N. Schutz for insightful discussions; Dr. Avital Shurki for providing data for the nonequilibrium solvation study; Dr. Francis C. Jen for his preliminary work on the QCP calculations; and especially Dr. Zhen T. Chu for his continuous effort to maintain the software used in this work. References and Notes
(1) Pauling, L. Chem. & Eng. News 1946, 263, 294. (2) Warshel, A. Computer Modeling of Chemical Reactions in Enzymes and Solutions; John Wiley & Sons: New York, 1991. (3) Warshel, A. J. Biol. Chem. 1998, 273, 27035-27038. (4) Kohen, A.; Cannio, R.; Bartolucci, S.; Klinman, J. P. Nature 1999, 399, 496-499. (5) Basran, J.; Sutcliffe, M. J.; Scrutton, N. S. Biochemistry 1999, 38, 3218-3222. (6) Wilson, E. K. C&EN 2000, 78 (29), 42-45. (7) Glasstone, S.; Laidler, K. J.; Eyring, H. The Theory of Rate Processes; McGraw-Hill: New York, 1941. (8) Chandler, D. J. Chem. Phys. 1978, 68, 2959. (9) Grimmelmann, E. K.; Tully, J. C.; Helfand, E. J. Chem. Phys. 1981, 74, 5300-5310. (10) Hwang, J.-K.; King, G.; Creighton, S.; Warshel, A. J. Am. Chem. Soc. 1988, 110, 5297-5311. (11) Neria, E.; Karplus, M. Chem. Phys. Lett. 1997, 267, 23-30. (12) Anderson, J. B. AdV. Chem. Phys. 1995, 91, 381-431. (13) Truhlar, D. G.; Garrett, B. C.; Klippenstein, S. J. J. Phys. Chem. 1996, 100, 12771-12800. (14) Keck, J. C. AdV. Chem. Phys. 1966, 13, 85-121. (15) The fact that the transmission factor depends on the number of times the trajectory moves back and forth on the transition state was questioned in ref 157. However, this recrossing effect is an integral part of current reaction rate formulations (e.g., refs 9 and 13). Reference 157 also questioned the fact that ref 10 verified the validity of the linear response approximation of eq 20 by averaging downhill trajectories, suggesting that no such calculation was reported in ref 10. Unfortunately, the author overlooked Figure 15 of ref 10. (16) Warshel, A.; Levitt, M. J. Mol. Biol. 1976, 103, 227-249. (17) Singh, U.; Kollman, P. J. Comput. Chem. 1986, 7, 718-730. (18) Bash, P.; Field, M.; Davenport, R.; Petsko, G.; Ringe, D.; Karplus, M. Biochemistry 1991, 30, 5826-5832. (19) Waszkowycz, B.; Hillier, I.; Gensmantel, N.; Payling, D. J. Chem. Soc. Perkin Trans. 2 1991, 225-231. (20) Mulholland, A.; Grant, G.; Richards, W. Protein Eng. 1993, 6, 133147. (21) Hartsough, D.; Merz, K. M. J. Phys. Chem. 1995, 99, 1126611275. (22) Gao, J. Acc. Chem. Res. 1996, 29, 298-305. (23) Bentzien, J.; Muller, R.; Floria n, J.; Warshel, A. J. Phys. Chem. B 1998, 102, 2293-2301. (24) Alhambra, C.; Gao, J.; Corchado, J. C.; Villa ` , J.; Truhlar, D. G. J. Am. Chem. Soc. 1999, 121, 2253-2258. (25) Zhang, Y. K.; Liu, H. Y.; Yang, W. T. J. Chem. Phys. 2000, 112, 3483-3492. (26) Mart , S.; Andre s, J.; Moliner, V.; Silla, E.; Tun on, I.; Bertra n, J. Theor. Chem. Acc. 2001, 3, 207-212. (27) Garc a-Viloca, M.; Gonza ` lez-Lafont, A.; Lluch, J. M. J. Am. Chem. Soc. 2001, 123, 709-721. (28) Warshel, A.; Weiss, R. J. Am. Chem. Soc. 1980, 102, 6218-6226. (29) Warshel, A.; Sussman, F.; Hwang, J.-K. J. Mol. Biol. 1988, 201, 139-159. (30) Chang, Y.-T.; Miller, W. H. J. Phys. Chem. 1990, 94, 5884-5888. (31) Kim, H. J.; Hynes, J. T. J. Am. Chem. Soc. 1992, 114, 1050810537. (32) Grochowski, P.; Lesyng, B.; Bala, P.; McCammon, J. Int. J. Quantum Chem. 1996, 60, 1143-1164. (33) Hinsen, K.; Roux, B. J. Chem. Phys. 1997, 106, 3567-3577. (34) Schmitt, U. W.; Voth, G. A. J. Phys. Chem. B 1998, 102, 55475551. (35) Vuilleumier, R.; Borgis, D. Chem. Phys. Lett. 1998, 284, 71-77. (36) Kim, Y.; Corchado, J. C.; Villa ` , J.; Xing, J.; Truhlar, D. G. J. Chem. Phys. 2000, 112, 2718-2735.

Villa ` and Warshel

(37) Stanton, R.; Pera kyla , M.; Bakowies, D.; Kollman, P. A. J. Am. Chem. Soc. 1998, 120, 3448-3457. (38) Kollman and co-workers called this approach QM-FE. However, since the QM calculations are done in the gas phase, we believe that such a name should be preserved to true coupled QM/MM-FEP approaches. (39) Chandrasekhar, J.; Jorgensen, W. J. Am. Chem. Soc. 1984, 106, 3049-3059. (40) Kollman, P. A.; Kuhn, B.; Donini, O.; Pera kyla , M.; Stanton, R.; Bakowies, D. Acc. Chem. Res. 2001, 34, 72-79. (41) The fact that the EVB surface is frequently recalibrated using theoretical and experimental information about the reference reaction in solution is an advantage rather than a disadvantage if reliable information about the reaction in the enzyme is desired. This is because the enzyme and solution surfaces are much more similar than the gas-phase and the solution surfaces. Not using information about solution reactions is similar to attempts to refine solvation models using only gas-phase calculations rather than solvation energies. (42) Alhambra, C.; Corchado, J. C.; Sa nchez, M. L.; Gao, J.; Truhlar, D. G. J. Am. Chem. Soc. 2000, 122, 8197-8203. (43) Mo, Y.; Gao, J. J. Phys. Chem. A 2000, 104, 3012-3020. (44) A qvist, J.; Warshel, A. Chem. ReV. 1993, 93, 2523-2544. (45) Warshel, A.; Russell, S. T. J. Am. Chem. Soc. 1986, 108, 6569. (46) Muller, R.; Warshel, A. J. Phys. Chem. 1995, 99, 17516-17524. (47) Valleau, J. P.; Torrie, G. M. Modern Theoretical Chemistry; Vol. 5 of A Guide to Monte Carlo for Statistical Mechanics. 2. Byways; Plenum Press: New York, 1977. (48) Day, T. J. F.; Schmitt, U. W.; Voth, G. A. J. Am. Chem. Soc. 2000, 122, 12027-12028. (49) Thompson, W. H.; Hynes, J. T. J. Phys. Chem. A 2001, 105, 25822590. (50) Mulliken, R. S. J. J. Chim. Phys. (Paris) 1964, 61, 20. (51) Warshel, A. J. Phys. Chem. 1982, 86, 2218-2224. (52) Warshel, A. Proc. Natl. Acad. Sci. U.S.A. 1984, 81, 444-448. (53) Sun, D.-P.; Liao, D.-I.; Remington, S. J. Proc. Natl. Acad. Sci. U.S.A. 1989, 86, 5361-5365. (54) Warshel, A. Biochemistry 1981, 20, 3167-3177. (55) Rao, S. N.; Singh, U. C.; Bash, P. A.; Kollman, P. A. Nature 1987, 328, 551. (56) Soman, K.; Yang, A.; Honig, B.; Fletterick, R. Biochemistry 1989, 28, 9918-9926. (57) Warshel, A.; Naray-Szabo, G.; Sussman, F.; Hwang, J.-K. Biochemistry 1989, 28, 3629. (58) Fuxreiter, M.; Warshel, A. J. Am. Chem. Soc. 1998, 120, 183194. (59) Vagedes, P.; Rabenstein, B.; A qvist, J.; Marelius, J.; Knapp, E. W. J. Am. Chem. Soc. 2000, 122, 12254-12262. (60) Glennon, T. M.; Warshel, A. J. Am. Chem. Soc. 1998, 120, 1023410247. (61) A qvist, J.; Warshel, A. Biochemistry 1989, 28, 4680. (62) A qvist, J.; Warshel, A. J. Mol. Biol. 1992, 224, 7. (63) Fothergill, M.; Goodman, M. F.; Petruska, J.; Warshel, A. J. Am. Chem. Soc. 1995, 117, 11619-11627. (64) A qvist, J.; Fothergill, M. J. Biol. Chem. 1996, 271, 10010-10016. (65) Glennon, T.; Villa ` , J.; Warshel, A. Biochemistry 2000, 39, 96419651. (66) Va rnai, P.; Warshel, A. J. Am. Chem. Soc. 2000, 122, 3849-3860. (67) Lee, Y. S.; Hodoscek, M.; Brooks, B. R.; Kador, P. F. Biophys. Chem. 1998, 70, 203-216. (68) Va rnai, P.; Richards, W. G.; Lyne, P. D. Proteins: Struct. Func. Gen. 1999, 37, 218-227. (69) Yadav, A.; Jackson, R. M.; Holbrook, J. J.; Warshel, A. J. Am. Chem. Soc. 1991, 113, 4800-4805. (70) Cunningham, M. A.; Ho, L. L.; Nguyen, D. T.; Gillilan, R. E.; Bash, P. A. Biochemistry 1997, 36, 4800-4816. (71) Hurley, J. H.; Remington, S. J. J. Am. Chem. Soc. 1992, 114, 47694773. (72) Lyne, P.; Mulholland, A.; Richards, W. J. Am. Chem. Soc. 1995, 117, 11345-11350. (73) Hansson, T.; Nordlund, P.; A qvist, J. J. Mol. Biol. 1997, 265, 118127. (74) Wu, N.; Mo, Y.; Gao, J.; Pai, E. Proc. Natl. Acad. Sci. U.S.A. 2000, 97, 2017-2022. (75) Warshel, A.; S trajbl, M.; Villa ` , J.; Floria n, J. Biochemistry 2000, 39, 14728-14738. (76) Na ray-Szabo , G.; Fuxreiter, M.; Warshel, A. Electrostatic basis of enzyme catalysis. In Computational approaches to biochemical reactiVity; Na ray-Szabo , G., Warshel, A., Eds.; Kluwer Academic Publishers: Dordrecht, 1997. (77) Warshel, A. Proc. Natl. Acad. Sci. U.S.A. 1978, 75, 5250-5254. (78) Cannon, W.; Benkovic, S. J. Biol. Chem. 1998, 273, 26257-60. (79) Hwang, J.-K.; Warshel, A. Biochemistry 1987, 26, 2669. (80) Marcus, R. A. J. Chem. Phys. 1956, 24, 966-978. (81) Schweins, T.; Warshel, A. Biochemistry 1996, 35, 14232-14243.

Feature Article
(82) Albery, W. J.; Knowles, J. R. Biochemistry 1976, 15, 5631-5640. (83) Gerlt, J. A.; Gassman, P. G. J. Am. Chem. Soc. 1993, 115, 1155211568. (84) Krishtalik, L. I. J. Theor. Biol. 1980, 86, 757-771. (85) Krishtalik, L. I.; Topolev, V. V. Mol. Biol. (Moscow) 1984, 18, 892-900. (86) Krishtalik, L. I. J. Theor. Biol. 1985, 112, 251-264. (87) Krishtalik and co-workers, in some of their works, used a welldefined continuum model that did not evaluate the effect of the protein polar groups (e.g., ref 84). In other works85 they indicated that they tried to estimate the effect of polar groups but were unable to consider simultaneously the effect of the solvent around the protein and the effect of the protein polar groups.86 Instead, they tried to use a poorly defined approach (it is practically impossible to see what was done except to judge the reported results) and to estimate the energy of the Asp-ImH+ ion pair in chymotrypsin.85 First, this ion pair is not a transition state of the relevant reaction. Second, Krishtalik and Topolev85 obtained very small effect (0.6 kcal/ mol) from the presumably preorganized hydrogen bonds and incorrectly considered the interaction between Asp- and ImH+ as the stabilizing force (this interaction is a part of the solute system which is the same in all environments). Thus, they concluded that the stabilization of the ion pair is due to the low dielectric of the protein85 while the opposite is true and ion pairs are distabilized at a low dielectric medium.158 Apparently, only the use of microscopic approaches such as those criticized by Krishtalik and Topolev85 without a valid ground (see discussion in footnote 56 of ref 69) can be used to explore the origin of the catalytic effect of enzyme active sites. (88) Leatherbarrow, R. J.; Fersht, A. R.; Winter, G. Proc. Natl. Acad. Sci. U.S.A. 1985, 82, 7840-7844. (89) Phillips, M. A.; Fletterick, R.; Rutter, W. J. J. Biol. Chem. 1990, 265, 20692-20698. (90) Carter, P.; Wells, J. A. Proteins: Struct. Func. Gen. 1990, 6, 240248. (91) Shoichet, B. K.; Baase, W. A.; Kuroki, R.; Matthews, B. W. Proc. Natl. Acad. Sci. U.S.A. 1995, 92, 452-456. (92) Lightstone, F.; Zheng, Y.-J.; Maulitz, A.; Bruice, T. Proc. Natl. Acad. Sci. U.S.A. 1997, 94, 8417-8420. (93) Berendsen, H. J. C.; Hayward, S. Curr. Opin. Struct. Biol. 2000, 10, 165-169. (94) Cameron, C. E.; Benkovic, S. J. Biochemistry 1997, 36, 1579215800. (95) Cannon, W.; Singleton, S. F.; Benkovic, S. Nat. Struct. Biol. 1996, 3, 821-833. (96) Careri, G.; Fasella, P.; Gratton, E. Annu. ReV. Biophys. Bioeng. 1979, 8, 69-97. (97) Karplus, M.; McCammon, J. A. Annu. ReV. Biochem. 1983, 53, 263-300. (98) Kurzynski, M. Cell. Mol. Biol. Lett. 1999, 4, 117-130. (99) McCammon, J. A.; Wolynes, P. G.; Karplus, M. Biochemistry 1979, 18, 927-942. (100) Radkiewicz, J. L.; Brooks, C. L., III. J. Am. Chem. Soc. 2000, 122, 225-231. (101) Warshel, A. Nature 1976, 260, 679-683. (102) Orozco, M.; Luque, F. J. Chem. ReV. 2000, 100, 4187-4226. (103) Kubo, R.; Toda, M.; Hashitsume, N. Statistical Physics II: Nonequilibrium Statistical Mechanics; Springer-Verlag: Berlin, 1985. (104) Warshel, A.; Bentzien, J. Energetics and Dynamics of Transition States of Reactions in Enzymes and Solutions. In Transition State Modeling for Catalysis; Truhlar, D., Morokuma, K., Eds.; ACS Sympsium Series 721; American Chemical Socidty: Washington, DC, 1999. (105) Warshel, A.; Hwang, J.-K. J. Chem. Phys. 1986, 84, 4938-4957. (106) Bader, J. S.; Kuharski, R. A.; Chandler, D. J. Chem. Phys. 1990, 93, 203. (107) Schulten, K.; Tesch, M. Chem. Phys. 1991, 158, 421-446. (108) Hwang, J.-K.; Chu, Z. T.; Yadav, A.; Warshel, A. J. Phys. Chem. 1991, 95, 8445. (109) Bahnson, B. J.; Colby, T. D.; Chin, K. J.; Goldstein, B. M.; Klinman, J. P. Proc. Natl. Acad. Sci. U.S.A. 1997, 94, 12797-12802. (110) Warshel, A.; Weiss, R. M. J. Am. Chem. Soc. 1981, 103, 446. (111) Miller, B.; Snider, M. J.; Short, S.; Wolfenden, R. Biochemistry 2000, 39, 8113-8118. (112) Bergsma, J.; Gertner, B.; Wilson, K.; Hynes, J. J. Chem. Phys. 1987, 86, 1356. (113) Chuang, Y. Y.; Truhlar, D. G. J. Am. Chem. Soc. 1999, 121, 10157-10167. (114) Lang, M. J.; Jordanides, X. J.; Song, X.; Fleming, G. R. J. Chem. Phys. 1999, 110, 5884-5892. (115) Stratt, R. M.; Maroncelli, M. J. Phys. Chem. 1996, 100, 1298112996. (116) Sutcliffe, M. J.; Scrutton, N. S. Trends Biochem. Sci. 2000, 25, 405-408. (117) Warshel, A.; Chu, Z. T. J. Chem. Phys. 1990, 93, 4003. (118) Borgis, D.; Hynes, J. T. J. Chem. Phys. 1991, 94, 3619-3628.

J. Phys. Chem. B, Vol. 105, No. 33, 2001 7907

(119) Bruno, W. J.; Bialek, W. Biophys. J. 1992, 63, 689-699. (120) German, E.; Kuznetsov, J. Chem. Soc., Faraday Trans. 1 1981, 77, 397-412. (121) Swain, C. G.; Strivers, E. C.; Reuwer, J. F.; Schaad, L. J. J. Am. Chem. Soc. 1958, 80, 5885-5893. (122) Cha, Y.; Murray, C.; Klinman, J. Science 1989, 243, 1325-1330. (123) Warshel, A.; Chu, Z. T.; Parson, W. W. Science 1989, 246, 112116. (124) Hammes-Schiffer, S. J. Chem. Phys. 1996, 105, 2236-2246. (125) Bala, P.; Grochowski, P.; Nowinski, K.; Lesyng, B.; McCammon, J. Biophys. J. 2000, 79, 1253-1262. (126) Gillan, M. J. J. Phys. Chem. Solid State Phys. 1987, 20, 3621. (127) Hwang, J.-K.; Warshel, A. J. Phys. Chem. 1993, 97, 1005310058. (128) Hwang, J.-K.; Warshel, A. J. Am. Chem. Soc. 1996, 118, 1174511751. (129) Hwang, J.-K.; Warshel, A. Chem. Phys. Lett. 1997, 271, 223225. (130) Voth, G.; Chandler, D.; Miller, W. J. Chem. Phys. 1989, 91, 7749. (131) Kong, Y.; Warshel, A. J. Am. Chem. Soc. 1995, 117, 6234. (132) Feierberg, I.; Luzhkov, V.; A qvist, J. J. Biol. Chem. 2000, 275, 22657-22662. (133) Voth, G. AdV. Chem. Phys. 1996, 93, 135-218. (134) Noya, J. C.; Herrero, C. P.; Ram rez, R. Phys. ReV. Lett. 1997, 79, 111-114. (135) Billeter, S. R.; Webb, S. P.; Iordanov, T.; Agarwal, P. K.; HammesShiffer, S. J. Chem. Phys. 2001, 114, 6925-6936. (136) Makri, N. Annu. ReV. Phys. Chem. 1999, 50, 167-191. (137) Feynman, R. Statistical Mechanics; Benjamin: New York, 1972. (138) Villa ` , J.; Truhlar, D. G. Theor. Chem. Acc. 1997, 1-4, 317-323. (139) Fast, P. L.; Corchado, J. C.; Truhlar, D. G. J. Chem. Phys. 1998, 109, 6237-6245. (140) Kohen, A.; Klinman, J. P. Chem. Biol. 1999, 6, 191-198. (141) Bruice, T. C.; Benkovic, S. J. Biochemistry 2000, 39, 6267-6274. (142) Torres, R. A.; Schitt, B.; Bruice, T. C. J. Am. Chem. Soc. 1999, 121, 8164-8173. (143) Antoniou, D.; Schwartz, S. D. J. Phys. Chem. B 2001, ASAP. (144) Jencks, W. P. Catalysis in Chemistry and Enzymology; Dover Publication: New York, 1986. (145) Cornish-Bowden, A. Fundamentals of Enzyme Kinetics; Butterworth and Co.: London, 1979. (146) Laidler, K. J.; Peternan, B. F. Temperature Effects in Enzyme Kinetics. In Contemporary Enzyme Kinetics and Mechanism; Purich, D. L., Ed.; Academic Press: New York, 1983. (147) Price, N. C.; Stevens, L. Fundamentals of Enzymology; Oxford University Press: New York, 1989. (148) Kyte, J. Mechanism in protein chemistry; Garland Publishing: Hamden CT, 1995. (149) McMurry, J.; Castellion, M. E. Fundamentals of Organic and Biological Chemistry; Prentice Hall: Englewood Cliffs, NJ, 1999. (150) Blow, D. Structure Fold. Des. 2000, 8, R77-81. (151) Villa ` , J.; S trajbl, M.; Glennon, T. M.; Sham, Y. Y.; Chu, Z. T.; Warshel, A. Proc. Natl. Acad. Sci. U.S.A. 2000, 97, 11899-11904. (152) Page, M. I.; Jencks, W. P. Proc. Natl. Acad. Sci. U.S.A. 1971, 68, 1678-1683. (153) Dunitz, J. D. Science 1994, 264, 670. (154) Amzel, L. M. Proteins: Struct. Func. Gen. 1997, 28, 144-149. (155) Hermans, J.; Wang, L. J. Am. Chem. Soc. 1997, 119, 2707-2714. (156) S trajbl, M.; Sham, Y. Y.; Villa ` , J.; Chu, Z. T.; Warshel, A. J. Phys. Chem. B 2000, 104, 4578-4584. (157) Karplus, M. Proteins Reactions and Conformational Change. Simple, Complex, or Both? In Dahlem Workshop on Simplicity and Complexity in Proteins and Nucleic Acids; Frauenfelder, H., Deisenhofer, H., Wolynes, P. G., Eds.; Dahlem University Press: Berlin, 1999. (158) Warshel, A.; Russell, S. T. Q. ReV. Biophys. 1984, 17, 283-421. (159) Sekhar, V. C.; Plapp, B. V. Biochemistry 1990, 29, 4289-4295. (160) Crosby, J.; Stone, R.; Lienhard, G. E. J. Am. Chem. Soc. 1970, 92, 2891. (161) Dewar, M.; Storch, D. Proc. Natl. Acad. Sci. U.S.A. 1985, 82, 2225-2229. (162) Lee, J. K.; Houk, K. N. Science 1997, 276, 942-945. (163) Warshel, A.; A qvist, J.; Creighton, S. Proc. Natl. Acad. Sci. USA 1989, 86, 5820-5824. (164) Lightstone, F. C.; Bruice, T. C. J. Am. Chem. Soc. 1996, 118, 2595-2605. (165) Storm, D. R.; Koshland, D. E. J. Am. Chem. Soc. 1972, 94, 5805. (166) Frey, P.; Whitt, S.; Tobin, J. Science 1994, 264, 1927-1930. (167) Cleland, W.; Kreevoy, M. Science 1994, 264, 1887-1890. (168) Warshel, A.; Papazyan, A. Proc. Natl. Acad. Sci. U.S.A. 1996, 93, 13665. (169) Sines, J. J.; Allison, S. A. Biochemistry 1990, 29, 9403-9412.