INTRODUCTION

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

Dating back to the iceberg hypothesis of Frank and Evans (1945), molecular explanations of hydrophobic phenomena have been sought, based on the unique tetrahedral hydrogen-bonding structure of water. Manifestations of hydrophobic phenomena include the limited solubility and aggregation of nonpolar solutes in water (Blokzijl and Engberts, 1993; Tanford, 1980), as well as self-assembly processes, such as protein folding (Dill, 1990; Kauzmann, 1959) and micelle formation (Hunter, 1987; Israelachvili, 1992), that are thought to be driven in large part by hydrophobic interactions. Indeed, many of the successful molecular theories describing hydrophobic effects for simple nonpolar solutes have incorporated into their framework either the structure of bulk liquid water or the microscopic structure of waters of hydration in the vicinity of such solutes (Hummer et al., 1996b; Lazaridis and Paulaitis, 1992; Pohorille and Pratt, 1990; Pratt and Chandler, 1977; Pratt and Pohorille, 1992; Stillinger, 1973).

In the Pratt-Chandler theory, for example, the experimental water oxygen-oxygen radial distribution function is used within an Ornstein-Zernike-like formalism to calculate both solute-water and solute-solute correlations for nonpolar solutes in water (Pratt and Chandler, 1977). The solute-water correlations are subsequently used in a perturbative approach to calculate the thermodynamics of hydrophobic hydration. More recently, an information theory model of hydration has been used to calculate the free energy of forming a cavity the size and shape of the solute in waterthe dominant contribution to the solute chemical potentialby modeling water density fluctuations within molecular-sized volumes in liquid water (Hummer et al., 1996b). Applications of the information theory model provide a quantitative understanding of the dependence of hydrophobic effects on temperature and pressure that have led to insights into heat (Garde et al., 1999; Garde et al., 1996a) and pressure (Hummer et al., 1998) denaturation of proteins. This approach, however, has limited applicability for molecularly large and mesoscopic solutes, because the probability of spontaneously forming a large cavity in liquid water becomes vanishingly small, and, more fundamentally, Gaussian-like descriptions of molecular scale fluctuations become inaccurate. Subtle effects, such as the cavity expulsion potential (Hummer and Garde, 1998), and, at larger length scales, dewetting phenomena (Stillinger, 1973; Wallqvist and Berne, 1995; Lum and Luzar, 1997; Lum and Chandler, 1998; Lum et al., 1999) must be considered for a complete description of the hydration of molecularly large hydrophobic solutes.

Lazaridis and Paulaitis, on the other hand, demonstrated that the large negative entropy and large positive heat capacity increments characteristic of hydrophobic hydration of simple solutes at ambient temperature can be calculated from solute-water translational and orientational pair correlations alone (Lazaridis and Paulaitis, 1992, 1993, 1994; Paulaitis et al., 1994). Their approach provides a direct connection between the molecular structure and the macroscopic thermodynamics of hydration. However, the accurate calculation of solute-water translational and orientational correlations becomes increasingly difficult for solutes of increasing size and structural complexity, and prohibitive for solutes with internal conformational degrees of freedom.

Recent simulations of a variety of hydrated nonpolar solutes with different geometries show that water structure in the vicinity of these solutes is only locally sensitive to the structural details of the solute (Ashbaugh and Paulaitis, 1996; Garde et al., 1996b). Hence, a hierarchy of n-site proximity approximations has been proposed to describe water structure near a molecular (i.e., multisite) solute in terms of water correlations with only the nearest n sites of the solute (Garde et al., 1996b). These proximity approximations represent an extension of the proximity criterion introduced by Ben-Naim (1974) in the context of generalized molecular distribution functions, and implemented by Beveridge and coworkers (Mehrotra and Beveridge, 1980; Mezei and Beveridge, 1986) to characterize water structure around hydrated molecular solutes. Pettitt and coworkers (Pettitt et al., 1998; Makarov et al., 1998) review the implementation of proximal radial distribution functions to protein hydration.

The proximity approximations, in some sense, have parallels to linear surface area correlations for the free energy of hydration (Chothia, 1974; Hermann, 1972; Reynolds et al., 1974; Sitkoff et al., 1994), although these correlations do not describe the solute-water interface in sufficient detail to provide a unique relationship between free energy and molecular topology (Ashbaugh et al., 1998; Ashbaugh et al., in preparation). Water structure, however, can be highly sensitive to local solute structure (Cheng and Rossky, 1998), and it is precisely these details of molecular structure that are accounted for in the proximity approximations.

We have shown in previous work that the proximity approximations provide an accurate description of water structure in the vicinity of molecular hydrophobic solutes, and further, that this water structure can be used to accurately predict the large negative entropies of hydrophobic hydration (Ashbaugh and Paulaitis, 1996; Garde et al., 1996b), as well as hydrophobic driving forces for the aggregation of nonpolar solutes in water (Garde et al., 1996c). In this paper, we explore the accuracy of water structures predicted by the two-site proximity approximation for calculating conformational free energies of n-alkanes in water. Our approach is to compare conformational free energies of hydration calculated from the free energy perturbation method of Zwanzig (1954) using explicit water simulations to those calculated using the proximity approximation. The accuracy and enhanced computational speed of the proximity approximation relative to the explicit water simulations demonstrate the utility of this simple molecular theory of hydration from which thermodynamic properties of hydrophobic interactions can be derived.

	THEORY AND METHODS

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

Molecular simulations

We calculate the hydration contribution to the conformation-dependent free energy of three linear alkanes in water: butane, pentane, and hexane. Exploration of the entire dihedral angle space of pentane and hexane using explicit water simulations is computationally too expensive. Hence, we calculate only the change in hydration free energy corresponding to well-defined reaction paths for these two n-alkanes.

Canonical ensemble Monte Carlo (MC) simulations (Allen and Tildesley, 1987) were carried out for aqueous solutions consisting of a single n-alkane molecule in a bath of 216 water molecules at 25°C and a density of 0.997 g/cm³. Lennard-Jones (LJ) interactions for the alkane CH₃ and CH₂ groups were modeled using the optimized potentials for liquid simulations (OPLS) united-atom parameters (Table 1) (Jorgensen et al., 1984). Alkane internal bond lengths and bond angles were constrained to 1.53 Å and 109.5°, respectively. Water was modeled using the simple point charge (SPC) potential (Table 1) (Berendsen et al., 1981). The LJ parameters for solute-water interactions were obtained from geometric mean combining rules. Electrostatic interactions were evaluated using the generalized reaction-field method introduced by Hummer et al. (1994). Both LJ and generalized reaction-field minimum image interactions were truncated on a site-by-site basis at one-half the simulation box length. Conformational reaction paths were considered that connect the extended all-trans conformations to the alternating gauche conformations of pentane (tt  gg') and hexane (ttt  gg'g). [Note: The letters t and g denote the ideal trans ( = 180°) and gauche ( = ±60°) backbone dihedral conformations, respectively. g' differentiates between the two mirror image gauche conformers, g⁺ and g, along the same alkane chain. For example, the designation tg denotes the set of equivalent pentane conformers {tg⁺, tg, g⁺t, gt}, whereas gg' denotes the conformers {g⁺g, gg⁺}.] These reaction paths are given in Table 2, and the initial and final conformational states of pentane and hexane are depicted in Fig. 1. Two independent reaction paths corresponding to the ttt  gg'g transition for hexane were considered to evaluate the path independence of the calculated free energy changes. The compact gg' and gg'g conformations of pentane and hexane, respectively, are inaccessible in reality due to unfavorable intrachain overlap, i.e., the pentane effect (Mattice and Suter, 1994). We note, however, that solvent contributions to the free energy, which are our interest here, are, by definition, independent of the intrachain energy.

View this table: [in this window] [in a new window]   TABLE 1   Optimized potentials for liquid simulations (Jorgensen, et al., 1984) and simple point charge (Berendsen, et al., 1981) interaction parameters for the alkane united-atom groups and water oxygens

View larger version (32K): [in this window] [in a new window]   FIGURE 1   Schematic illustration of the all trans (tt and ttt) and alternating gauche (gg' and gg'g) conformations for pentane and hexane.

During the simulations, each solute was held in a fixed conformation along a given reaction path, and the relative hydration free energy difference between this reference conformation and a perturbed solute conformation, A_hyd(   + ), was evaluated using the free energy perturbation expression (Zwanzig, 1954),

(1)

where ¹ = kT is the product of Boltzmann's constant and the absolute temperature, and E is the solute-water potential energy difference between the perturbed,  + , and the reference, , conformations. The angle brackets, ···, denote thermal averaging over solvent configurations generated for the solute in the reference conformation. A simulation for one conformation of butane in water consisted of 40,000 MC passes for equilibration followed by 80,000 MC passes for sampling (1 MC pass = 216 attempted water and 4 attempted solute moves). A simulation for one conformation of pentane or hexane in water consisted of 40,000 MC passes for equilibration followed by 200,000 MC passes for sampling. Twenty-four solute conformations from  = 172.5° to +172.5° in 15° increments were simulated along each reaction path with  = ±7.5°.

Path independence of the free energy guarantees that A_hyd(180°  180°) = 0. However, this does not necessarily hold for free energies calculated from simulation because only a finite portion of the total configurational space is sampled. Therefore, we report A_hyd(180°  180°) as a simple measure of the uncertainty in the simulation calculations. This hysteresis error was subtracted from the reported hydration free energies assuming a linear dependence of the error on . The rotation of a single dihedral angle through 360° along each of the reaction paths in Table 2 required approximately 7 CPU days on a DEC AlphaStation 255/300.

Free energy calculations using the proximity approximation

The perturbation expression, Eq. 1, can be expanded in terms of cumulants of the distribution of E in the reference state (Hummer et al., 1995b, 1996c; Kubo, 1962; Smith and van Gunsteren, 1994; Zwanzig, 1954). For sufficiently small perturbations in , i.e., for E kT, the first order term provides an accurate approximation to the corresponding change in free energy,

(2)

which is exact in the limit,  0 (Chandler, 1987). The ensemble average for E in this equation is related to the water structure surrounding the solute in its reference conformation by

(3)

where (r_W|{r₁, ... , r_n}) is the number density of water molecules at r_W given that the sites of an n-site solute in conformation  are situated at {r₁, ... , r_n}. For an n-site solute,

(4)

where (r_W, r_i,) is the interaction energy between a water molecule at r_W and the ith solute site at r_i, in conformation . In the following, we provide details of the proximity approximation and solute-water interactions used to calculate (r_W|{r₁, ··· , r_n}) and E, respectively, in the integrand of Eq. 3.

Proximity approximation

The inhomogeneous density of water in the vicinity of a solute is obtained from the solute-water potential-of-mean force (PMF), which, for a multisite solute, can be expressed as an expansion in terms of one-particle, two-particle, and higher-order multiparticle PMFs between the solute sites and a water molecule (Green, 1952; Hummer and Soumpasis, 1994a, 1994b). This expansion, truncated at the level of two- or three-particle correlations, has been applied successfully to strongly associating inhomogeneous systems, such as ion density distributions near nucleic acids (Klement et al., 1991; Soumpasis, 1984), the ice-water interface (Hummer and Soumpasis, 1994a), and biomolecular hydration (García et al., 1997; Garde et al., 1997; Hummer et al., 1995a, 1996a; Hummer and Soumpasis, 1994b). However, for the hydration of hydrophobic solutes, where volume exclusion is the dominant solute-water interaction, it was found that the PMF expansion, truncated at the level of three-particle correlations, substantially overestimates water densities at contact (Garde et al., 1996b). Unreasonably large water densities (more than 50 times the bulk density) were likewise noted near the alanine dipeptide using the PMF expansion truncated at the level of two-particle correlations (Pelligrini and Doniach, 1995; Pelligrini et al., 1996). On the other hand, the proximity approximations correctly account for the excluded volume of overlapping solute sites, thereby providing an accurate representation of the solute-water PMF for weakly interacting, hydrophobic solutes (Garde et al., 1996b)

In the one-site proximity approximation, the solute-water PMF is given by the PMF between water and only the nearest site of the solute, assuming solute sites of equal size. Thus,

(5)

where j is the solute site closest to a water molecule at r_W, _b is the bulk water density, and g(r_W, r_j) is the pair correlation function between water and a lone solute site. Similarly, for the two-site proximity approximation,

(6)

where k is the second nearest solute site to a water molecule at r_W, and g(r_W|r_j, r_k) is the conditional triplet correlation function between a water molecule and a lone pair of solute sites located at r_j and r_k. These approximations are readily generalizable to n sites. It is straightforward to confirm that, for a solute composed of n sites, the n-site proximity approximation is exact. We use the two-site proximity approximation for water density predictions throughout this work and refer to it hereafter simply as the proximity approximation.

Triplet correlations between a water oxygen and two solute sites, g(r_W|r_j, r_k), were evaluated from a series of MC simulations of two OPLS united-atom methanes in SPC water with the methanes held at separations varied in increments of 0.2 Å from 1.2 to 6.6 Å. Triplet correlations at intermediate separations were approximated using tri-linear interpolation. Details of these simulations are given elsewhere (Garde et al., 1996b).

Interaction potentials

In the explicit water simulations, OPLS united-atom potentials were used for the CH₂ and CH₃ interactions with SPC water. Water structures predicted using the proximity approximation, however, assume that all the alkane sites are OPLS united-atom methanes. Because liquid structure is determined primarily by repulsive interactions and the OPLS repulsive core of all the alkane sites are similar, differences in water structure between the various solute sites are not expected to be significant. However, small differences in the free energy calculated using the proximity approximation are expected if methane-water interactions are used in the energy calculations. To account for differences between the OPLS LJ well depths, we used modified solute-water interactions in Eq. 4 following a strategy similar to that in Weeks-Chandler-Andersen perturbation theory (Weeks et al., 1971):

(7)

where  = _CH4O, r =r_CH4O, and i = 2 or 3 specifies the CH₂ and CH₃ sites of an n-alkane, respectively. Thus, the repulsive core of CH_i-water interactions are assumed to be the same as that for CH₄-water interactions, whereas the attractive CH₄-water interactions have been normalized to the potential minimum for CH_i-water interactions.

The free energy changes for perturbations along a particular reaction path were calculated by integrating Eq. 3 using the solute-water potentials in Eq. 7 with water densities predicted from the proximity approximation in Eq. 6. This integration was performed using a grid placed within a rectangular box that extended 10 Å from the minimum and maximum x, y, and z coordinates of the solute in its extended conformation. The calculated free energy values are sensitive to the integration parameters, the grid spacing, and the value of along a reaction path. We found that, for a grid spacing of ~0.25 Å and  = ±3.75°, the calculated free energies are independent of the integration parameters. This grid spacing is close to the bin width of 0.20 Å used for calculating triplet correlations. Increasing the dimensions of the box did not change our results because the long-range contributions to the LJ interaction potential are small and essentially independent of conformation, and thereby cancel when calculating E in Eq. 3. Approximately one CPU minute was required to calculate the free energy profile along each reaction path on an SGI R10K processor. This corresponds to an increase in computational speed over the explicit water simulations of approximately four orders of magnitude.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

Calculated free energies of hydration are shown in Figs. 2 through 5 as a function of n-alkane conformation for the reaction paths in Table 2. For butane (Fig. 2), the free energy of hydration increases with increasing backbone dihedral angle from the compact cis (₁ = 0°) to the extended trans (₁ = 180°) conformation. This characteristic behavior has been reported in a large body of simulation (Ashbaugh et al., 1998; Beglov and Roux, 1994; Jorgensen, 1982; Jorgensen and Buckner, 1987; Kaminski et al., 1994; Rosenberg et al., 1982; Tobias and Brooks, 1990) and theoretical (Hummer et al., 1996b; Pratt and Chandler, 1977; Zichi and Rossky, 1986) studies of butane conformations in water, and is consistent with the notion embodied in linear free energy-surface area correlations that hydrophobic interactions tend to minimize the extent of solvent-accessible surface area of hydrophobic solutes. Differences in the methods and interaction potentials used here and in previous studies, however, preclude quantitative comparisons with previous work.

View larger version (16K): [in this window] [in a new window]   FIGURE 2   The conformation-dependent hydration free energy of butane following reaction path 1 (Table 2). , simulation result; line, prediction using the proximity approximation including attractive interactions (Eq. 7); , the proximity approximation predictions excluding attractive interactions (Eq. 8). The simulation results are referenced to Ahyd(180°) = 0. The proximity approximation predictions excluding attractive interactions have been shifted downward for clarity. The arrow denotes the location of the free energy maximum in both the simulation results and proximity approximation predictions. The simulation hysteresis error is smaller than the symbols.

For pentane (Fig. 3), A_hyd(₁, 180°) is symmetric about ₁ = 0° with each half of the free energy profile essentially the same as that obtained for butane. Moreover, the depth of the free energy minimum at ₁ = 0° is approximately the same as that obtained for the cis conformation of butane, as might be expected, because holding ₂ fixed at 180° minimizes the interaction between the terminal methyl groups of pentane along this reaction path. In contrast, holding ₁ fixed at 60° while rotating ₂ results in a partial overlap of the terminal methyl groups along the gt  gg' reaction path. Consequently, A_hyd(60°, ₂) is asymmetric about ₂ = 0°, and the free energy minimum at ₂  30° is more than twice that for A_hyd(₁, 180°). In this case, the terminal methyl groups make their closest approach in the gg' conformation (Fig. 1), which roughly corresponds to this free energy minimum.

View larger version (20K): [in this window] [in a new window]   FIGURE 3   The conformation-dependent hydration free energy of pentane following reaction paths 2a and b (Table 2).  and , simulation results for reaction paths 2a and b, respectively; The lines are the two-site proximity approximation predictions. The locations of the tt, gg, and gg' conformations are identified on this figure for clarity. The simulation results are referenced to Ahyd(180°, 180°) = 0. The cumulative simulation hysteresis errors, Ahyd(180°  180°), are indicated by the error bars at  = 0°.

For hexane along reaction path 3 (Fig. 4), ₃ is held fixed at 180° in steps 3a and 3b, which effectively reduces the influence of the terminal methyl group adjacent to this dihedral angle. As a consequence, the free energy profiles A_hyd(180°, ₂, 180°) and A_hyd(₁, 60°, 180°) are virtually indistinguishable from A_hyd(₁, 180°) and A_hyd(60°, ₂) for pentane, respectively, with the exception that A_hyd(₁, 60°, 180°) and A_hyd(60°, ₂) are mirror images of one another because of solute symmetry. When ₃ is rotated in step 3c of this reaction path, a minimum in the free energy profile is observed at ₃ = 30° that is much lower than the minima observed for butane or pentane. We attribute this deep minimum to the partial overlap of the terminal methyl groups, as well as the hydration contribution to the attraction between the methyl group adjacent to ₃ and the methylene group attached to the methyl group at the other end of the chain.

View larger version (21K): [in this window] [in a new window]   FIGURE 4   The conformation-dependent hydration free energy of hexane following reaction paths 3a-c (Table 2). Shown are the simulation results for reaction paths 3a (), 3b (), and 3c (). The lines are the two-site proximity approximation predictions. The locations of the ttt and gg'g conformations are identified on this figure for clarity. The simulation results are referenced to Ahyd(180°, 180°, 180°) = 0. The cumulative simulation hysteresis errors, Ahyd(180°  180°), are indicated by the error bars at  = 0°.

For hexane along reaction path 4 (Fig. 5), ₂ is held fixed at 180° in steps 4a and 4b, which has the effect of isolating conformations at each end of the alkane chain. Hence, changes in the hydration free energy for rotations of ₁ and ₃ are virtually independent of one another and A_hyd(₁, 180°, 180°) and A_hyd(60°, 180°, ₃) are qualitatively similar to the free energy profile of butane. The largest free energy change occurs in step 4c due to the overlap of the opposing CH₃CH₂ groups, which gives a favorable hydration free energy contribution. In this case, the deep minimum at ₂ = 60° accounts for most of the change in free energy along the entire reaction path.

View larger version (23K): [in this window] [in a new window]   FIGURE 5   The conformation-dependent hydration free energy of hexane following reaction paths 4a-c (Table 2). Shown are the simulation results for reaction paths 4a (), 4b (), and 4c (). The lines are the two-site proximity approximation predictions. The locations of the ttt, ggg, and gg'g conformations are identified on this figure for clarity. The simulation results are referenced to Ahyd(180°, 180°, 180°) = 0. The cumulative simulation hysteresis errors, Ahyd(180°  180°), are indicated by the error bars at  = 0°.

All free energy calculations using the water structure predicted from the proximity approximation are in excellent agreement with the profiles obtained from simulation. Systematically larger discrepancies are noted for butane and the butane-like reaction paths of pentane and hexane (reaction paths 1, 2a, 3a, 4a, and 4b). However, the absolute differences between the simulation and proximity results for the butane-like reaction paths are, in general, small compared to the overall change in free energy for pentane and hexane.

A general characteristic of all the free energy profiles is the presence of weak maxima observed near the rim of the free energy wells (e.g., see simulation results in Fig. 2 near ₁  120°). This feature is observed for both the explicit water simulation results and the proximity approximation predictions, although it is not as prominent for the proximity approximation. To gain insight into the origin of this characteristic behavior, we calculated the conformational free energy of hydration for butane using the proximity approximation and the purely repulsive Weeks-Chandler-Andersen interaction potential for methane (Weeks et al., 1971),

(8)

The results are plotted in Fig. 2. For the purely repulsive interaction potential, the predicted free energy minimum for the cis conformation is deeper than that obtained when attractive interactions are included. More importantly, the maximum at ₁  120° disappears. Similar observations were made for the hydration free energy profiles of pentane and hexane (not shown). We conclude, therefore, that the weak maxima result mainly from the opposing effects of solute-water attractive and repulsive interactions. The attractive interactions favor extended alkane conformations, whereas the repulsive hydrophobic interactions favor compact alkane conformations. The free energy maxima correspond to conformations for which these solute-water interactions offset one another.

We noted above that hysteresis errors arise in explicit simulations due to insufficient sampling of phase space in simulations of finite length. Similar hysteresis errors can arise in the free energies calculated using the proximity approximation. One reason for this error is that the triplet correlation functions themselves are evaluated from explicit water simulations of finite length, and as such, contain statistical uncertainties. A second source of hysteresis error arises because the triplet correlation functions are available only at discrete water-methane and methane-methane separations. Finally, hysteresis errors can arise even if the triplet correlation functions are known exactly, and result from the inherent approximations of using only lower order correlation functions to calculate the inhomogeneous density of water surrounding the solute. These errors, when combined with the numerical integration of Eq. 3 and discrete perturbations, , lead to a systematic error at each step along a reaction path.

Nonetheless, we find that, for most of the reaction paths considered here, the net change in free energy for rotating a single dihedral angle through 360° is zero. The one exception is step 4c in which a CH₃CH₂ group of hexane is moved as a rigid body. In this case, we observe a net decrease of 0.6 kcal/mol in the free energy for ₂ = 180°  180°. Moreover, this hysteresis error can be reproduced by rotating a CH₃CH₂ group along this reaction path in the absence of the rest of the hexane molecule. Thus, we conclude that the hysteresis error in our free energy predictions is amplified along reaction paths for which multiple sites of the solute are moved as a rigid body. This error can be reduced by choosing reaction paths such that only a limited number of solute sites are perturbed in each step. We also suggest making quantitative evaluations of the hysteresis error for reaction paths involving multisite perturbations, as was done here. For the free energy calculations reported here for step 4c, we have subtracted the corresponding estimate of the hysteresis error to obtain the free energy profiles in Fig. 5, which are in excellent agreement with the explicit water simulation results.

The calculated change in hydration free energy for the conformational rearrangement of each alkane chain from the extended all-trans conformation (t, tt, and ttt) to the alternating gauche (gg' and gg'g) or the helical gauche (g, gg, and ggg) conformation is reported in Table 3. Overall, the free energies calculated using the proximity approximation are in excellent agreement with those obtained from simulation. With the exception of butane and the pentane tt  gg transition, this agreement is within the simulation uncertainties; for butane the free energy change is within twice the uncertainty, and for the pentane tt  gg transition, the difference is only 0.07 kcal/mol or approximately 0.1 kT. Moreover, we find that the free energy changes for the hexane ttt  gg'g transition calculated for the two independent reaction paths are in close agreement with each other and their differences are within the simulation uncertainties, thus verifying the path independence of the hydration free energy. We note that the proximity approximation also gives slightly different free energies for this transition along the two hexane reaction paths. Nonetheless, the agreement is excellent.

A point-by-point comparison of hydration free energies calculated from explicit water simulations and from the proximity approximation is shown in Fig. 6 for the two hexane reaction paths. All points fall close to the diagonal line with a root mean square difference of 0.1 kT between the free energies over the entire range of hexane conformations. This root mean square difference is more than three times smaller than the statistical uncertainties reported in Table 3 for the explicit water simulations along either reaction path. Thus, the two methods are in excellent agreement.

View larger version (18K): [in this window] [in a new window]   FIGURE 6   Comparison of the hydration free energies of hexane reaction paths 3a-c and 4a-c determined from simulation and the proximity approximation. The root mean square difference between the free energies calculated using these techniques is 0.06 kcal/mol.

In Fig. 7, the conformational free energy of hydration of hexane, obtained from simulation, is plotted as a function of its van der Waals surface area. The van der Waals surface was chosen rather than the solvent-accessible surface or the molecular surface (Lee and Richards, 1971; Richards, 1977) because it was found to give the most accurate linear free energy-surface area correlation for n-alkane conformations in water (Ashbaugh et al., 1998). The strong correlation between free energy and van der Waals surface area depicted in Fig. 7 is clearly not a linear correlation. Indeed, we find that the free energy-surface area coefficient, , in the linear relation,

(9)

differs by a factor of two for the all-trans conformation and the gg'g conformation of hexane. Thus, to achieve accuracies in the conformational free energy of hydration comparable to those obtained from the proximity approximation, typical linear free energy-surface area correlations would require a conformationally dependent  which, in fact, cannot be extracted from measured transfer free energies for the n-alkanes. Moreover, we expect the conformational dependence of  found in Fig. 7 to be specific to hexane, and therefore, not transferable to a wider range of hydrocarbons.

View larger version (16K): [in this window] [in a new window]   FIGURE 7   Conformational free energy of hydration of hexane (reaction paths 3a-c and 4a-c) obtained from simulation as a function of the van der Waals surface area of hexane. The surface area and Ahyd are referenced to zero in the all trans conformation.

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

Most simulation studies of hydrophobic effects to date have focused mainly on water-mediated interactions between simple nonpolar solutes, such as the PMF between methane pairs (Lüdemann et al., 1997, 1996; New and Berne, 1995; Pangali et al., 1979; Payne et al., 1997; Smith and Haymet, 1993; van Belle and Wodak, 1993; Young and Brooks, 1997) or the conformational equilibrium of a single butane chain in water (Ashbaugh et al., 1998; Beglov and Roux, 1994; Jorgensen, 1982; Jorgensen and Buckner, 1987; Kaminski et al., 1994; Rosenberg et al., 1982; Tobias and Brooks, 1990). Although these studies add substantially to our understanding of the hydrophobic driving forces for small nonpolar solutes in aqueous solution, few unifying concepts have resulted which permit the extension of molecular hydrophobic effects to larger clusters or assemblies of nonpolar solutes.

The proximity approximation presented herein may provide such a connection. We have shown that the two-site proximity approximation leads to quantitative predictions of the conformational equilibria of n-alkanes in water, in contrast to conventional free energy-surface area correlations, which are at best approximate in nature, or the truncated PMF expansion, which, in previous work, was shown to substantially overestimate water densities around nonpolar solutes. We conclude, therefore, that the underlying assumption of the proximity approximation is correct, at least for relatively simple molecular solutes, such as the n-alkanes considered here; that is, the inhomogeneous structure of water is only locally sensitive to the molecular details of a hydrophobic solute. Although we expect this assumption to break down for macroscopic interfaces where water structure is strongly perturbed, the length scale at which this occurs has not yet been determined. Indeed, our preliminary results for much larger clusters of 135 methanes in water indicate that the proximity approximation gives a quantitatively accurate description of water structure around these assemblies as well (Ashbaugh and Paulaitis, unpublished results).

Although the one-site proximity approximation has been shown, in previous work, to account largely for the water structure in the vicinity of a single nonpolar solute in water (Ashbaugh and Paulaitis, 1996; Garde et al., 1996b), the basic unit for accurate determination of water-mediated forces is the methane dimer in aqueous solution (Garde et al., 1996c). Indeed, the two-site proximity approximation used in conjunction with free energy perturbation yields the exact PMF between a methane pair in water, whereas the one-site approximation does not (Garde et al., 1996c). Although the triplet correlation functions required for the two-site proximity approximation represent an initial computational cost that is significant, these correlation functions are determined only once. Subsequent applications of the proximity approximation can be accomplished much faster by referencing this existing library of triplet correlation functions. For the n-alkane solutes studied here, we observed an enhancement in computational speed of nearly four orders of magnitude.