Enantiospecificity of Chloroperoxidase-Catalyzed Epoxidation: Biased Molecular Dynamics Study of a Cis-β-Methylstyrene/Chloroperoxidase-Compound I Complex

Abstract

Molecular dynamics simulations of an explicitly solvated cis-β-methylstyrene/chloroperoxidase-Compound I complex are performed to determine the cause of the high enantiospecificity of epoxidation. From the simulations, a two-dimensional free energy potential is calculated to distinguish binding potential wells from which reaction to 1S2R and 1R2S epoxide products may occur. Convergence of the free energy potential is accelerated with an adaptive biasing potential. Analysis of binding is followed by analysis of 1S2R and 1R2S reaction precursor structures in which the substrate, having left the binding wells, places its reactive double bond in steric proximity to the oxyferryl heme center. Structural analysis of binding and reaction precursor conformations is presented. We find that 1), a distortion of Glu¹⁸³ is important for CPO-catalyzed epoxidation as was postulated previously based on experimental results; 2), the free energy of binding does not provide significant differentiation between structures leading to the respective epoxide enantiomers; and 3), CPO's enantiospecificity toward cis-β-methylstyrene is likely to be caused by a specific group of residues which form a hydrophobic core surrounding the oxyferryl heme center.

Introduction

Chloroperoxidase (CPO), a heme-thiolate protein secreted by the marine fungus Caldariomyces fumago, is one of the most versatile enzymes of the heme protein superfamily. CPO-catalyzed oxidative reactions proceed after two-electron oxidation of the ferric heme center, in the presence of a suitable hydroperoxide, to an oxyferryl porphyrin-radical cation intermediate, Compound I. Compound I is common to CPO, cytochrome P450 (P450), most traditional heme peroxidases, and catalases. The normal function of CPO is to utilize chloride, bromide, and iodide ions in the halogenation of electron-rich organic substrates (1–3). CPO also catalyzes a variety of other reactions, including the dehydrogenations characteristic of classical heme peroxidases (4,5), dismutation of hydrogen peroxide characteristic of catalase (4,6), and monooxygenation of many organic molecules characteristic of monooxygenase (7). Recently CPO has been the target of intense experimental (8–15) and theoretical research (16–19). Of particular current interest are stereoselective reactions such as the oxidation of sulfides (20), epoxidation of olefins (21), and hydroxylation of alkynes (22).

The active site of CPO is considered to be a peroxidase-P450 hybrid (23) responsible for the protein's versatile catalytic activity. Like P450, CPO possesses a cysteine-derived thiolate ligand to the heme iron and has channels connecting the surface to the heme pocket. However, unlike P450, CPO has a relatively polar active site, placing it in the family of peroxidase enzymes. A unique feature of CPO catalysis is the use of a glutamic acid residue, Glu¹⁸³, as the general acid-base catalyst in facilitating the formation of the Compound I form of CPO (CPO-I) (16,17,24); most traditional heme peroxidases use a histidine residue for this purpose. A histidine is also functionally important for CPO, but less directly, as part of a His¹⁰⁵-Glu¹⁸³ proton shuttle (17,25). In peroxidases, substrates are limited to electron-transfer reactions at the heme edge with no direct access to the center of the heme. There is mounting evidence that one-electron oxidations catalyzed by CPO occur predominantly outside the heme active site as well, presumably at the surface of the enzyme, whereas P450-type reactions involving two-electron oxidations proceed through binding of substrates to the heme pocket of CPO (12,26,27).

As early as 1993, Allain et al. (21) reported that CPO-catalyzed epoxidations of some olefin substrates have very high enantiomeric excesses. Further such studies followed (22,28) and attracted much attention, as chiral epoxides are useful synthons for many practical purposes (13,29). (It should be noted that P450-catalyzed epoxidation depends on both molecular oxygen and electron transfer proteins, whereas CPO's ability to utilize hydroperoxides for the same reactions makes it a promising candidate for practical applications.) Shaik and co-workers (30–32) established that epoxidation is characterized by a “two-state reactivity” in which reaction can proceed on either of two closely spaced spin surfaces. However, little is known about the structural basis for the enantiospecificity exhibited by CPO-catalyzed epoxidations. The factors giving rise to this pronounced enantiospecificity are the subject of this article. The substrate cis-β-methylstyrene (CBMS) was chosen for this study because CPO-catalyzed epoxidation of CBMS has a 96% enantiomeric excess and because of its use in prior experimental and computational studies (21,27).

From a bioengineering point of view, it is interesting to consider a hypothesis that the favorability of substrate binding conformations parallels the favorability of epoxide enantiomers (i.e., the binding hypothesis). If this is true, the molecular mechanical (MM) level of theory can be used to assess quantitatively the potential of site-directed mutagenesis for enhancing CPO as a chiral epoxidation catalyst. The binding hypothesis gets support from an early study on P450_cam-catalyzed epoxidation of styrene and cis/trans-β-methylstyrene (33). Using molecular dynamics (MD) simulations and analysis of energy-minimized binding conformations, Fruetel et al. (33) found that the high enantiospecificity of P450_cam toward these substrates can be explained on the basis of specific steric interactions involved in substrate/active site binding (33). Subsequently, Sundaramoorthy et al. (27) tried to provide a structural basis for the enantiospecificity of CPO toward CBMS on the basis of a similar conceptual framework. In their study (27), it was postulated that the electrostatic interaction between the substrate methyl and Glu¹⁸³ carboxyl groups in a bound complex leads to the preference for the 1S2R epoxide enantiomer. To our knowledge, these studies (27,33) are the only available literature addressing the cause of the enantiospecificity of P450-type epoxidation. Both studies, in effect, utilized the binding hypothesis to gain a qualitative understanding of this problem.

A quantitatively reliable test of the binding hypothesis requires calculation of the free energy difference between enantiomer-related binding conformations. When the aforementioned studies (27,33) were performed (more than a decade ago), such calculations were not computationally feasible even though a simplified simulation protocol was used for MD. In this simplified protocol, the peptide backbone is kept constrained after minimization of a crystal structure. The only water molecules included are those present in the crystal structure, and the water oxygen atoms, in addition to the peptide backbone, were constrained after minimization. The protocol relies on an assumption that the enzyme's active site is rigid (27,33). Given CPO's versatile catalytic activity, this assumption may be not realistic for CPO. Indeed, in the study of Yi et al. (24), the Glu¹⁸³His mutant of CPO indicates that a distortion of the His¹⁰⁵-Glu¹⁸³ proton shuttle favors epoxidation.

We have chosen to revisit the binding hypothesis and assess it quantitatively, on the basis of free energies, without rigid active site restriction. Even with the improvement in computational resources since the above referenced studies (27,33), extracting such thermodynamic information remains a challenging task (34). In biomolecular systems, mapping of a free energy potential at room temperature is hindered by the low statistical weights of the transitional states between conformers and by a rugged energy landscape. To alleviate these issues, several different implementations of a nonphysical sampling approach have been proposed in conjunction with MD or Monte Carlo simulations (35–43). The MD simulations presented here have been carried out using an adaptive biasing potential method (38). To reach satisfactory sampling of CBMS/CPO conformational space, we modified the adaptive biasing potential method as described in the next section. Previous free energy calculations using the adaptive biasing method were performed mainly for small peptides (42,44). There are numerous publications on the adaptive biasing method devoted to method development (e.g., see the introduction to Maragakis et al. (42)).

Here we present a free energy map of CBMS/CPO binding interactions calculated from a set of six 10-ns MD simulations that use a successively refined biasing potential. We are able to characterize the binding interactions structurally but find that, despite structural diversity, binding interactions do not provide sufficient differentiation between structures leading to the respective epoxide enantiomers to explain the observed enantiomeric excess. For this reason, we performed a further analysis that focused on structures more closely related to reactive events but still in the regime where molecular mechanics is applicable.

In this second part of our analysis, the available MD trajectories were analyzed to identify 1S2R and 1R2S precursor events in which the substrate is out of binding wells and the substrate's reactive double bond undergoes steric proximity to the oxyferryl heme center. Such a reaction precursor event is assumed to represent a structure intermediate between a binding and a reactive event.

Consequently, reaction precursor structures were expected to differentiate 1S2R and 1R2S epoxide enantiomers, although the degree of stereoselectivity might be less than that of corresponding transition states. Sampling of precursor events was not sufficient for quantitatively accurate thermodynamic averaging over the conformational space of the solvated CBMS/CPO-I complex, as it was for binding. Therefore, following previous examples (27,33), we focused on the limited number of active site residues that are likely to influence stereoselectivity. These are residues having steric contacts with the substrate (33) and residues of the His¹⁰⁵-Glu¹⁸³ proton shuttle (27). Based on this analysis, we postulate a structural basis for the enantiospecificity of CPO toward the substrate CBMS.

The rest of the article is organized as follows. In next section, we describe the calculational method. In Results, we present analysis of binding and reaction precursor interactions. Discussion contains a discussion of the results and a hypothesis concerning the structural basis of CPO's enantiospecificity toward the substrate CBMS.

Method

Simulation details

Classical MD simulations were performed with the biomolecular simulation program CHARMM (45), using the CHARMM22 (46) force field for proteins including CMAP corrections (47). The partial charges and other MM parameters for an oxyferryl heme-thiolate (Compound I) were those developed by Park and Harris (48). An all-atom MM potential for CBMS was developed employing a procedure given by Vanommeslaeghe et al. (49). CPO is heavily glycosylated, which renders it highly soluble in aqueous solvent. MM parameters for needed sugars not previously parameterized were created by analogy, using the recommendations of Vanommeslaeghe et al. (49). Patches were also created to connect sugars with other sugars and with amino-acid residues, as found in the 1CPO PDB file, an x-ray structure of CPO in the resting state (23). The parameters developed in this study are available in Section A in the Supporting Material.

The long-range effect of Coulomb interactions was calculated using the particle-mesh Ewald method (50), with a cutoff of 10 Å for real-space interactions and a 1 Å grid with sixth-order B-spline interpolation for reciprocal-space interactions. Lennard-Jones forces were treated with the force switch method with a switching range of 8–10 Å. The lengths of all bonds involving hydrogens were constrained using the SHAKE algorithm (51) with allowed relative deviations of 10⁻¹⁰. For NPT dynamics, an extended system constant pressure and temperature algorithm (52) was chosen for keeping the pressure at 1 atm and the temperature at 298 K. A structure of CPO-I was created from the resting state structure of CPO (23) by changing the heme to the oxyferryl form and relaxing the active site. A structure of CBMS complexed with CPO-I was obtained previously in our group by docking (53). This structure was solvated in a truncated octahedral box of ∼7000 water molecules represented with the TIP3P model (54) as modified for the CHARMM force field (46).

The protonation states of titrable amino-acid residues were chosen as follows: +1 for His, Lys; −1 for Glu, Asp; and neutral for Tyr. These values are appropriate for pH 5, the pH of the target experimental study (21). The dynamics was propagated with the Verlet leapfrog algorithm using a time step of 1 fs. The structure was neutralized by adding sodium cations, minimized, heated, and equilibrated to generate the initial coordinates and velocities. An adaptive bias method, described below, was coded using the user energy facility of the CHARMM program. With the c34b1 version of CHARMM, the parallel simulations gave a performance of ∼1 ns per day on a cluster of 32 Xeon 2.5 GHz processors.

Adaptive bias sampling

A free energy potential function, F(s), describes the relative thermodynamics underlying the conformational changes of a system as a function of a collective variable s. Usually s is a function of a limited number of degrees of freedom related to a conformational change of interest. F(s) is obtained by thermodynamic averaging in a state s (55),

$$F\left(\mathbf{s}\right)=-{\beta }^{-1}\ln W\left(\mathbf{s}\right)=-{\beta }^{-1}\ln \int \delta \left(\mathbf{s}-{\mathbf{s}}^{\prime }\left(\mathbf{x}\right)\right)w\left(\mathbf{x}\right)d\mathbf{x},$$ (1)

where W(s) is the probability density for observing the state s at temperature T, x is a point in configuration space, w(s) is the microscopic probability density in the x space, β = (k_BT)⁻¹, k_B is Boltzmann's constant, and δ is the Dirac delta. Like Fruetel et al. (33) in work on P450_cam, we use the two-dimensional collective variable s = (r, ϕ), as explained in Fig. 1, to distinguish ligand-bound conformations that will lead to different product enantiomers upon reaction. If a system is embedded in a heat reservoir, w(x) is proportional to exp(−βH(x)), where H(x) is the Hamiltonian function of the system. Hence, using Eq. 1 it is easy to connect perturbed and unperturbed thermodynamics,

$$F\left(\mathbf{s}\right)=-V_b\left(\mathbf{s}\right)-{\beta }^{-1}\ln W_b\left(\mathbf{s}\right),$$ (2)

where the subscript b denotes the thermodynamics of a system with a perturbed (biased) Hamiltonian

$$H_b\left(\mathbf{x}\right)=H\left(\mathbf{x}\right)+V_b\left(\mathbf{s}\right)\text{.}$$

The idea of the adaptive bias method is to speed up sampling on a complex energy landscape by an iterative accumulation of the thermodynamic information into the biasing function, i.e.,

$$V_b^{(i+1)}\left(\mathbf{s}\right)=-F^{\left(i\right)}\left(\mathbf{s}\right),$$ (3)

where the superscript (i) stands for the iteration number, and F⁽ⁱ⁾(s) is the approximation of the free energy potential F(s) at the i^th iteration:

$$F^{\left(i\right)}\left(\mathbf{s}\right)=-V_b^{\left(i\right)}\left(\mathbf{s}\right)-{\beta }^{-1}\ln W_b^{\left(i\right)}\left(\mathbf{s}\right)\text{.}$$ (4)

Usually, instead of a continuous function W_b⁽ⁱ⁾(s), an outcome of an MD simulation is a piecewise histogram of visits, Hist_b⁽ⁱ⁾(s). In this case, the biasing potential V_b⁽ⁱ⁺¹⁾(s) is constructed by replacing F⁽ⁱ⁾(s) in Eq. 3 by a smooth approximation (e.g., see (43)). For sufficiently long trajectories, a flat sampling will eventually be achieved, indicating that the biasing function, to within a constant, is the negative of the free energy potential:

$$W_b^{\left(i\right)}\left(\mathbf{s}\right)=const\Rightarrow V_b^{\left(i\right)}\left(\mathbf{s}\right)=-F\left(\mathbf{s}\right)\text{.}$$ (5)

However, for explicitly solvated systems the size of CBMS/CPO-I, achieving flat sampling is still a computationally expensive task. In our experience, three consecutive 10-ns iterations, using Eqs. 2–5, did not result in a smooth approach to flat sampling; valleys and hills in F⁽ⁱ⁾(s) may be overcompensated from one iteration to the next. Thus, the information directly inherent in F⁽ⁱ⁾(s) may be lost at iteration i+1 if the duration of iteration i+1 is not sufficiently long. To prevent such losses of the information, we modified Eq. 3 as

$$V_b^{(i+1)}\left(\mathbf{s}\right)=-\left[\gamma /(i+1)\right]\sum _{k=0}^i{F}^{\left(k\right)}\left(\mathbf{s}\right),$$ (6)

where i = 0 corresponds to the initial iteration, and the reason for scaling by the system-specific factor γ is described later. Instead of testing for flat sampling, we monitored the standard error in the mean for the distribution of F⁽ⁱ⁾(s) over i where all iterations have equal time-length. It should be noted that multidimensional biased sampling of biomolecular systems is known to be very sensitive to the choice of biasing potential (56), and some system-specific modifications of a sampling method are often necessary. In the case of CBMS/CPO-I, by running trial simulations, we found that efficient sampling requires a scaling factor on the right-hand side of Eq. 6 to keep net energy contributions of the biasing potential to transitions in s space close to kT. In this way, the results of biasing are kept close in value to thermal bath excitations, whereas in our experience, too-strong biasing destroys the active site structure.

Figure 1.

Two-dimensional collective variable s = (r, ϕ); r is the distance O-C_β; ϕ is the dihedral angle O-C_β- C_α-R. The values ϕ = −90° and ϕ = +90° correspond, respectively, to the 1R2S and 1S2R epoxide.

We started the first iteration using the biasing function

$$V_b^{\left(0\right)}\left(r\right)={\left(r/R\right)}^{12},$$

where R = 7 Å. This creates a potential wall that prevents the substrate from moving far away from the active site. At the same time, the distance R ∼ 7 Å is large enough to ensure an appropriate sampling of possible binding conformations by allowing the substrate to exit the sterically crowded immediate neighborhood of the heme iron, undergo reorientation, and reenter this neighborhood. A binning scheme with discretization widths Δr = 0.05 Å and Δϕ = 2° was used to construct the histograms Hist_b⁽ⁱ⁾ (r,ϕ). The logarithm of Hist_b⁽⁰⁾ (r,ϕ) accumulated over a 10-ns trajectory is shown in Fig. 2 a. The free energy potential F⁽⁰⁾ (r,ϕ) was calculated using Eq. 2 and is shown in Fig. 2 b. The value F ⁽⁰⁾(r,ϕ) was then approximated by a smooth function ${\tilde{F}}^{\left(0\right)}\left(r,\varphi \right)$ according to the following formula,

$${\tilde{F}}^{\left(i\right)}\left(r,\varphi \right)=\sum _{k,l}{\alpha }_{kl}^{\left(i\right)}{T}_k\left(r\right)Y_l\left(\varphi \right),$$ (7)

where the coefficients α⁽ⁱ⁾_kl are determined by the least-squares method, and $T_k\left(r\right),k=0,\dots ,10$ are Chebyshev polynomials of the first kind;

$$Y_l\left(\varphi \right)=\text{cos}\left(l\varphi \right),l=0,\dots ,5,$$

and $Y_l\left(\varphi \right)=\sin \left[\left(1-5\right)\varphi \right],l=6,\dots ,10$. The biasing potential for the next iteration, $V_b^{\left(1\right)}\left(r,\varphi \right),$ was constructed according to the formula

$$V_b^{(i+1)}\left(r,\varphi \right)=-\left[\gamma /(i+1)\right]\sum _k{\tilde{F}}^{\left(k\right)}\left(r,\varphi \right)+{\left(r/R\right)}^{12},$$ (8)

where γ = 1/3. The result is shown in Fig. 3. We accumulated a 10-ns trajectory with the biasing potential V_b⁽¹⁾ (r,ϕ). The histogram and the free energy potential are shown in Fig. 2, c and d, respectively. The histograms in Fig. 2, a and c exhibit a large difference in the sampling patterns. In particular, the biasing potential V_b⁽¹⁾ (r,ϕ) enforces better sampling of reaction precursor regions (see below) with r close to 3 Å and ϕ close to ±90°. The stable reproduction in Fig. 2 d of the binding wells A, B, and C present in Fig. 2 b confirms that a 10-ns timescale is sufficient for representative sampling of the important active site degrees of freedom, such as side-chain rotations. These three binding wells are also reproduced in the other four simulations (see Sections B and D in the Supporting Material), further confirming that they are real and not artifacts of a particular biasing potential.

Figure 2.

Two-dimensional maps a and c are the visit histograms in logarithmic scale for the zeroth and first iteration, respectively; panels b and d are the corresponding free energy potentials calculated using Eq. 4. The contour lines are drawn according to the grayscale code (color code online) given.

Figure 3.

Two-dimensional map of the biasing potential for iteration 1 (i = 1) calculated using Eq. 9. The contour lines are drawn according to the grayscale code (color code online) given.

Results

Binding structures

The free energy potential F(r,ϕ) resulting from the six 10-ns iterations (see Fig. 4) confirms that CBMS can form a reaction precursor complex within CPO's active site by binding into one of the three potential wells denoted in the previous section as A, B, and C. Averaged binding structures (Fig. 5) were calculated by averaging the structures within the ranges 3.75 < r < 3.8 Å and 149.5 ° < ϕ < 150.5° for minimum A, 4.65 < r < 4.7 Å and 44.5 ° < ϕ < 45.5 for minimum B, and 3.75 < r < 3.8 Å and −150.5 ° < ϕ < −149.5 ° for minimum C.

Figure 4.

Free energy potential averaged over six 10-ns MD simulations with the adaptive biasing potential. The contour lines are drawn according to the grayscale code (color code online) given. The statistical error in the free energy as a function of r and ϕ is generally less than the spacing between contour levels. A map of the statistical error is given in Section B in the Supporting Material.

Figure 5.

Averaged structures for binding wells A, B, and C. The side chains of residues Phe¹⁰³, Ile¹⁷⁹, and Phe¹⁸⁶ undergo rotational transitions (see Table 1) upon binding. The His¹⁰⁵-Glu¹⁸³ proton shuttle undergoes distortion (see Table 2) upon binding.

For these structures, the root mean-square deviations (RMSD) relative to the resting-state crystal structure (23) and the root mean-square fluctuations (RMSF) during MD simulation were calculated for active site residues within 6 Å of the substrate. These are residues Cys²⁹, Val⁶⁷, Ile⁶⁸, Leu⁷⁰, Ala⁷¹, Phe¹⁰³, Ile¹⁷⁹, Val¹⁸², Glu¹⁸³, Phe¹⁸⁶, Ile²³⁴, Ala²⁶⁵, and the heme moiety. The RMSFs averaged over these residues are 1.2 Å, 1.0 Å, and 0.75 Å for structures A, B, and C, respectively. For all three averaged binding structures, the backbone RMSDs for these residues are less than the corresponding backbone RMSFs. However, the side-chain RMSDs of Phe¹⁰³, Ile¹⁷⁹, Glu¹⁸³, and Phe¹⁸⁶ turned out to be substantially larger than the corresponding RMSFs. These results demonstrate that substantial motions of the side chains are required for CBMS binding at the active site of CPO.

The structural changes of residues Phe¹⁰³, Ile¹⁷⁹, and Phe¹⁸⁶ can be described as rotations of the side chains around the C_α-C_β and C_β-C_γ bonds. The side-chain rotations for these residues as well as the rotations of the phenyl ring substituent of CBMS are quantified in Table 1. For binding structures A, B, and C, within the fluctuation ranges given in Table 1, the side chains of Phe¹⁰³, Ile¹⁷⁹, and Phe¹⁸⁶ undergo essentially the same rotations relative to the resting-state crystal structure (23). The main difference is that the phenyl rings of Phe¹⁰³ and Phe¹⁸⁶ are flexible upon binding in wells A and B whereas binding structure C shows lack of flexibility. The substrate parallels this behavior; rotations of the phenyl ring observed for structures A and B are prohibited for structure C (see Table 1). As described above, the average RMSF for the active site residues of structure C is smaller than the RMSFs for structures A and B. It is important to note that wells A and C correspond to the van der Waals minimum of the substrate β-carbon and heme oxyferryl oxygen interaction, i.e., r ∼ 3.8 Å, whereas well B is characterized by the steric interaction of the substrate methyl group with the oxyferryl oxygen; the O-C distance is ∼3.4 Å.

Table 1.

Side-chain dihedrals of the crystal, binding, and reaction precursor structures

	Phe103		Ile179		Phe186		Substrate
	γ1	γ2	γ1	γ2	γ1	γ2	γ
Crystal	81	−20	172	174	160	−15	—
A	174 ± 8	−60 ± 25	−80 ± 33	175 ± 14	147 ± 9	−150 ± 25	23 ± 25
		95 ± 25				−75 ± 25	145 ± 30
						50 ± 25
B	170 ± 16	−65 ± 30	−87 ± 33	174 ± 20	145 ± 8	−130 ± 30	26 ± 27
						−60 ± 20	145 ± 25
						45 ± 25
C	173 ± 8	−69 ± 12	−71 ± 9	170 ± 21	145 ± 10	−60 ± 16	6 ± 30
1R2S	165 ± 20	−80 ± 16	83 ± 40	103 ± 40	154 ± 10	−51 ± 45	38 ± 30
1S2R	172 ± 6	61 ± 37	89 ± 31	91 ± 23	139 ± 22	−81 ± 10	150 ± 45
						89 ± 19

Note that all ranges representing 20% or more of all structures are shown. The values γ₁ and γ₂ (in degrees) are side-chain dihedral angles C-C_α-C_β-X_γ and C_α-C_β-X_γ-X_δ, respectively; γ (in degrees) is dihedral angle C_β-C_α-R1-R2 of the substrate, where R1 and R2 are heavy atoms of the R moiety (Fig .1).

A final remark on the binding structures concerns the functionally important side chain of Glu¹⁸³ (16,23,24), which occupies the immediate vicinity of the iron of the heme moiety in the resting-state crystal structure (23). For the binding structures A, B, and C, the average RMSD of this side chain is ∼1.3 Å. Further analysis showed that the carboxyl group undergoes a distortion with a net increase in the distance to the iron relative to the crystal structure. The distortions presented in Table 2 are similar for all binding scenarios.

Table 2.

Distances (in Å) between Glu¹⁸³ carboxyl group oxygen atoms and the heme iron

	Crystal	A	B	C	1R2S	1S2R
Fe-O1	6.4	8.2 ± 0.4	8.1 ± 0.4	8.3 ± 0.3	8.8 ± 0.8	8.4 ± 0.7
Fe-O2	5.1	7.3 ± 0.3	7.2 ± 0.3	7.3 ± 0.3	8.4 ± 0.8	8.1 ± 0.9

Reaction precursor events

To relate the binding wells to the enantiomer-related transition states, reaction precursor events (a further precursor event) were defined as visits of a simulation trajectory to the histogram entries where

$$3.15<r<3.25\AA \mathrm{and}\varphi =\pm \mathrm{90}\pm \mathrm{2°,}$$

and where the first plus and minus signs are for precursors to 1S2R and 1R2S epoxides, respectively. The criteria for ϕ were chosen on the basis of quantum mechanical (QM) calculations for the P450-catalyzed epoxidation of ethene (32). The criterion for r implies that a thermal excitation toward one of the transition states was strong enough to bring the substrate β-carbon and heme oxyferryl oxygen into steric contact, with a van der Waals energy ∼kT/2 above the corresponding minimum for this pair of atoms. Note that such an excitation does not bring the substrate into the vicinity of the transition state (the transition state for ethene epoxidation has an r of ∼2 Å and a barrier height one order of magnitude larger than kT (see Figs. 2 and 3 in de Visser et al. (32)). Because these precursor events correspond to the regime of van der Waals repulsion, they represent the closest approach for which meaningful information may be gleaned from molecular mechanics-based MD.

Our assumption is that MM simulations of precursor events may contain information pertinent to understanding the stereoselectivity of the reaction, although a stringent analysis would require a QM treatment of reactive events. There were no precursor events detected in the trajectory for iteration 0. Approximately 50 1R2S and 50 1S2R events were detected in the biased trajectories for iterations 1–5. Apparently, this difference is caused by the bias of the functions V⁽ⁱ⁾_b(r,ϕ) toward precursor events (e.g., see Fig. 3). After a precursor event was detected, the trajectory was traced back until one of the binding wells was reached. We found that subtrajectories toward transition state 1R2S originate from well C whereas the subtrajectories toward transition state 1S2R originate from wells A and B. It should be noted that the ratio of the number of 1R2S and 1S2R events during biased simulation will not necessarily be the same as the ratio for unbiased mechanics and cannot be used to estimate the enantiospecificity of the reaction under consideration. To enhance sampling of the precursor events, two additional 10-ns MD simulations were carried out with harmonic restraints to keep r near 3.2 Å and ϕ near ±90° for 1S2R and 1R2S precursors, respectively. This allowed accumulation of ∼400 1R2S and ∼700 1S2R events (see Section C in the Supporting Material).

Analysis of the reaction precursor events

For each of the precursor events, a representative structure was chosen with r in the range of 3.2–3.25 Å. Two of these structures are presented in Fig. 6. The averaged RMSDs for the individual active-site residues mentioned above are ∼1 Å larger for the 1R2S and 1S2R precursors than for the corresponding binding structures (the RMSDs are relative to the resting-state crystal structure). Thus, it is likely that distortion of the active site is necessary for the substrate to approach the oxyferryl moiety to within reaction distance. The displacement of the Glu¹⁸³ side chain characteristic of the binding structures becomes even larger in the case of the precursor events (see Table 2). A dihedral angle analysis of the reaction precursor structures shows that, as for the binding structures, the side chains of Phe¹⁰³, Ile¹⁷⁹, and Phe¹⁸⁶ undergo structural changes whereas the rest of the active site dihedrals remain intact. Dihedral angle C-C_α-C_β-X_γ (γ₁) differs from its crystal and binding structure values only for Ile¹⁷⁹ (see Table 1). Dihedral angle C_α-C_β-X_γ-X_δ (γ₂) shows flexibility for Phe¹⁰³, Ile¹⁷⁹, and Phe¹⁸⁶.

Figure 6.

Examples of reaction precursor structures (a) of a 1S2R event and (b) of a 1R2S event. The hydrophobic core consisting of Phe¹⁰³, Ile¹⁷⁹, Val¹⁸², and Phe¹⁸⁶ is likely to favor 1S2R relative to 1R2S reaction precursor events.

To remove biasing effects, the precursor structures were minimized over the substrate's and oxyferryl oxygen's degrees of freedom with the rest of the coordinates fixed, and with harmonic restraints applied to keep r near 3.2 Å and ϕ near ±90° for 1S2R and 1R2S precursors, respectively. Minimization was followed by analysis of steric contacts, which we defined as a positive van der Waals energy for at least one residue-substrate atom pair among all such pairs for a given residue. This analysis shows that for both 1R2S and 1S2R precursors, the steric contacts are primarily caused by a specific hydrophobic core surrounding the oxyferryl heme center (for the complete set of residues involved in steric contacts; see Section C in the Supporting Material).

The hydrophobic core residues making the most significant contacts were Phe¹⁰³, Ile¹⁷⁹, Val¹⁸², and Phe¹⁸⁶ (see Fig. 6). The interaction energies of CBMS with these residues and with the heme moiety were found collectively to favor 1S2R precursors by 0.8 ± 0.1 kcal/mol on average whereas the internal energy of the substrate conformations was found to be equivalent for 1S2R and 1R2S precursors. Note that this energy difference (0.8 kcal/mol) is subject to limitations of MM/MD simulations in the van der Waals repulsion regime and should be viewed as suggestive rather than an accurate assessment of the steric effect.

Discussion

Experimentally, 96% of the product epoxide is the 1S2R enantiomer (21), which translates into a 1.9 kcal/mol difference in favor of the 1S2R transition state. Because all trajectories proceeding to precursor events for 1S2R products originated in wells A and B, and all trajectories proceeding to events for 1R2S products originated in well C, we can determine whether the free energy differences between the basins could be responsible for the observed enantiospecificity. Integration over the binding wells gives a free energy difference of zero (0.0 ± 0.1 kcal/mol) between 1S2R (wells A and B) and 1R2S (well C) binding structures. The statistical error is much smaller than the difference needed to explain the enantiomeric excess (1.9 kcal/mol). Thus, the free energy potential calculated in this study does not support the binding hypothesis. (The statistical error is derived from independent integrations over the free energy maps for the different iterations; see Section D in the Supporting Material for details.)

Sundaramoorthy et al. (27) postulated that favorable interactions between the substrate methyl group and Glu¹⁸³ cause the preference for the 1S2R epoxide. This hypothesis was based on an argument that this particular favorable interaction is paralleled in MD simulations by a favorable orientation of the substrate's reactive double bond with respect to the oxyferryl oxygen. We do not observe such a correlation in our simulations. The energies of both substrate/proton shuttle and substrate methyl group/proton shuttle interactions were found to be slightly unfavorable for 1S2R relative to 1R2S binding and precursor structures. Such apparent disagreement is likely due to the aforementioned rigid active site restriction used in the earlier study (27). Because of this restriction, the distortion of the Glu¹⁸³ carboxyl group away from the heme center observed in the MD simulations presented here (Table 2) may have been prevented in the MD simulation presented in that study. This assumption is confirmed by previous MD simulations we carried out (53), in which the distortion of Glu¹⁸³ was not observed when only the active site region was allowed to be flexible.

Yi et al. (24) found that substituting Glu¹⁸³ with a histidine residue is detrimental to the chlorination and dismutation activities of CPO but increases the epoxidation activity. Based on this experiment, they postulated that the epoxidation activity of native CPO is favored by an increase in the hydrophobicity of the active site due to a distortion of the proton shuttle (24). Our results support this postulate. The distortion of the Glu¹⁸³ side chain away from the heme center, observed in our MD simulation, may well constitute the mechanism of evolving the active site environment from polar peroxidase-like to hydrophobic, of P450 type. However, we found proton shuttle/substrate interactions to be of no significance for the enantiospecificity. The substrate is quite mobile in the active site, as it is oriented by nonpolar interactions. This explains the general flatness of the free energy landscape (Fig. 4). In general, we do not observe specific binding interactions of significance.

Having recognized that binding is not correlated with observed enantiomeric preferences, we examined aspects of the structure and energetics upon closer approach of the substrate to the oxyferryl heme moiety. Our results suggest that, similarly to P450 epoxidation of styrenes (33), the product enantiomer ratio for CPO-catalyzed epoxidation is likely due to a subtle balance of steric interactions caused by the hydrophobic core (residues Phe¹⁰³, Ile¹⁷⁹, Val¹⁸², and Phe¹⁸⁶) around the oxyferryl heme center. The importance of the residues Ile¹⁷⁹ and Val¹⁸² for the epoxidation reaction has not been reported previously.

We find that substrate interactions with the hydrophobic core shown in Fig. 6 lead to stereoselectivity not in the binding environment but after thermal excitation of the CBMS/CPO-I complex out of the binding wells and toward one of the transition states. Although we cannot investigate the transition region per se at this level of theory, we expect that the influence of hydrophobic-core residues upon the reacting moieties (oxyferryl heme and substrate) will be similar in the regime of precursor events and in the transition state. Thus, based on the results presented, we postulate that the enantiospecificity of CPO-catalyzed epoxidation toward the substrate CBMS is caused by steric effects of the hydrophobic core on the enantiomer-related transition states.

Conclusions

The cause of the enantiospecificity of CPO-catalyzed epoxidation toward the substrate CBMS has been postulated at a residue-specific level in terms of substrate/active-site-residue steric interactions. The critical interactions were those experienced during reaction precursor events, not in the binding environment. The CBMS/CPO-I reactive complex was modeled using MD with an adaptive biasing potential formulated in terms of two coordinates critical for the reaction. A stringent test of the proposed hypothesis requires QM/MM transition state calculations to be performed. The results should prove useful for the bioengineering of site-directed mutants aimed toward enhancing CPO as a chiral epoxidation catalyst.