feature articles
Epoxide hydrolysis as a model system for understanding
through a branched reaction schemeaDepartment of Chemistry, BMC, Uppsala University, Box 576, 751 23 Uppsala, Sweden, and bScience for Life Laboratory, Department of Cell and Molecular Biology, Uppsala University, BMC Box 596, 751 24 Uppsala, Sweden
*Correspondence e-mail: kamerlin@icm.uu.se, mikael.widersten@kemi.uu.se
The epoxide hydrolase StEH1 catalyzes the hydrolysis of trans-methylstyrene oxide to 1-phenylpropane-1,2-diol. The (S,S)-epoxide is exclusively transformed into the (1R,2S)-diol, while hydrolysis of the (R,R)-epoxide results in a mixture of product enantiomers. In order to understand the differences in the stereoconfigurations of the products, the reactions were studied kinetically during both the pre-steady-state and steady-state phases. A number of closely related StEH1 variants were analyzed in parallel, and the results were rationalized by structure–activity analysis using the available crystal structures of all tested enzyme variants. Finally, empirical valence-bond simulations were performed in order to provide additional insight into the observed kinetic behaviour and ratios of the diol product enantiomers. These combined data allow us to present a model for the through the catalyzed reactions. With the (R,R)-epoxide, ring opening may occur at either C atom and with similar energy barriers for hydrolysis, resulting in a mixture of diol products. However, with the (S,S)-epoxide, although either epoxide C atom may react to form the covalent enzyme intermediate, only the pro-(R,S) alkylenzyme is amenable to subsequent hydrolysis. Previously contradictory observations from kinetics experiments as well as product ratios can therefore now be explained for this biocatalytically relevant enzyme.
Keywords: epoxide hydrolase; stereoselectivity; empirical valence-bond simulations; biocatalysis; reaction flux; trans-methylstyrene oxide.
1. Introduction
et al., 1992). Therefore, it would be of great value to be able to control epoxide ring opening, and thus also to predict the stereoconfigurations of the products formed. In aqueous solution, the regioselectivity of ring opening by a depends on the of the substituents and on steric effects. For instance, the hydrolysis of trans-methylstyrene oxide (1a; Fig. 1a) is expected to occur via breakage of the bond between the benzylic C atom (C1) and the epoxide O atom. This prediction is based on electron donation from the phenyl substituent, which stabilizes the buildup of positive charge on the attacked C atom (Fig. 1b; Parker & Isaacs, 1959). The reaction is further facilitated by protonation of the leaving-group oxide. However, making such regioselectivity predictions is far less trivial in an enzyme-catalyzed epoxide ring-opening reaction, since the chiral nature of the active site and its particular microenvironment steer the reaction in directions that are uncoupled from the intrinsic chemical reactivities (Monterde et al., 2004) of the epoxide substrates. To correctly predict the resulting stereoselectivity therefore becomes a major and often unmet challenge in itself.
are important components in the toolbox of chiral building blocks for (SantanielloEpoxide hydrolase 1 from potato (StEH1; Stapleton et al., 1994) catalyzes the hydrolytic ring-opening reactions of monosubstituted and trans-disubstituted alkyl and aryl by an SN2 mechanism, resulting in inversion of the stereoconfiguration when the attacked epoxide C atom is asymmetric. The catalytic mechanism is that of a generic α/β-hydrolase (Heikinheimo et al., 1999), involving a combination of nucleophilic and acid/base catalysis: the carboxylate of Asp105 attacks one of the electrophilic epoxide C atoms, forming a negatively charged covalent alkylenzyme intermediate which is stabilized by two tyrosine (Fig. 1c). This ester intermediate is subsequently hydrolyzed by a general-base-activated water molecule, where the side chain of the active-site histidine, His300, acts as a general base (Pinot et al., 1995; Tzeng et al., 1996, 1998; Elfström & Widersten, 2005; Widersten et al., 2010).
The StEH1-catalyzed hydrolysis of (S,S)-1a proceeds with high regioselectivity for the benzylic C atom, which is in accordance with what would be expected from the intrinsic reactivities (Table 1; Lindberg et al., 2008; Lindberg, Ahmad, 2010). The regioselectivity in the hydrolysis of (R,R)-1a, however, is less strict, although with a modest preference for ring opening at the benzylic epoxide C atom. The hydrolysis of the epoxide thus generates both the (1R,2S)-2a and (1S,2R)-2a diol enantiomers, and the corresponding regiopreference is sensitive to environmental conditions such as pH and temperature (Lindberg, de la Fuente Revenga et al., 2010). This enantiomer-dependent regiopreference is not due to trivial causes such as steric clashes in the active site (Lindberg, de la Fuente Revenga et al., 2010; Reetz et al., 2009). However, since is a highly desired feature in an increased understanding of the underlying mechanisms is both motivated and necessary. It is known, for example, that StEH1 catalyzes the hydrolysis of styrene oxide (R = H in Fig. 1a), resulting in to (R)-phenylethanediol (Monterde et al., 2004; Janfalk Carlsson et al., 2012). This makes StEH1 an excellent model system for the design of novel enantioconvergent biocatalysts.
‡Data from Janfalk Carlsson et al. (2012). §Calculated from the steady-state expression for kcat and the determined values of kcat and (k−2 + k3). ¶Experimentally determined in a single-turnover experiment. |
In a recent computational study using the empirical valence-bond (EVB) approach, we managed to both qualitatively and quantitatively mimic the experimentally determined enantiomeric ratios of the hydrolysis products (Bauer et al., 2016). An important insight provided by this study was that despite the completely different regioselectivities of epoxide ring opening for the different enantiomers, both enantiomers displayed the same preferred substrate-binding mode (of two possible binding modes; see Fig. 2). That is, rather than being driven by radical changes in binding mode, the differences in selectivity appeared to be owing to subtle variations in binding interactions between the enzyme and the substrate for the different enantiomers. In combination with differences in the interactions with active-site water molecules, this resulted in distinct configurations for attack, ultimately leading to to the (R)-diol product.
The kinetic scheme for the hydrolysis of styrene oxide can be described by a simple linear Michaelis–Menten model, and the formation and decay of only one alkylenzyme can be observed during the pre-steady-state phase of the reaction (Bauer et al., 2016). However, the case of the hydrolysis of 1a is more complex, and the fact that two enantiomeric product diols can be formed from an optically pure starting epoxide by necessity invokes a branched reaction scheme. We have proposed a model (Fig. 3) that builds on earlier experimental results (Elfström & Widersten, 2005, 2006; Thomaeus et al., 2007; Lindberg et al., 2008; Lindberg, Ahmad et al., 2010; Lindberg, de la Fuente Revenga et al., 2010; Widersten et al., 2010). The model invokes a substrate-independent conformational step (E to E′) as well as reversible isomerizations of the respective Michaelis complexes (ES to E′S). The catalytic mechanism includes covalent alkylenzyme(s) (E-alkyl and E′-alkyl, respectively), which are considered not to be interconvertible. If the formed alkylenzyme is readily hydrolyzed to a diol, the regioselectivity of the nucleophilic attack on the epoxide ring also decides the product configuration. However, if the barrier for hydrolysis is higher than that for decomposition to the Michaelis complex(es), then interconversion between ES and E′S is conceivable. Hence, an alkylenzyme with the opposite configuration may be formed, albeit at a lower steady-state concentration. If, however, this (less favoured) alkylenzyme intermediate is efficiently hydrolyzed, the diol product will inherit its configuration. This model of the different states of productive enzyme–substrate complexes describes the simplest cases and does not explicitly include the concept of different substrate-binding modes within the active sites of either E or E′. The assumption is that the different binding modes are rapidly equilibrated on a timescale that is not rate-limiting for catalysis.
Experimental validation of the model can be achieved by monitoring the transient buildup and decay of the respective alkylenzymes and by measuring the appear to describe a linear mechanism, this does not prove the formation of a single product The diol products of either of trans-stilbene oxide (1b) will in all cases be meso-hydrobenzoin (2b), and therefore product analysis neither supports nor disproves the proposed model.
ratios of the product diols. If one alternative pathway leading to product dominates, or if the buildup of one of the alkylenzymes is low in comparison to the other, the model reduces to a simple linear Michaelis–Menten mechanism. It follows, therefore, that even if the kineticsUsing the epoxide 1a as a substrate, however, allows the kinetic model to also be tested experimentally, since the different product enantiomers are readily individually quantifiable. As mentioned above, the StEH1-catalyzed hydrolysis of (S,S)-1a produces only (1R,2S)-2a, even though two transient rates with distinct amplitudes are detected during the pre-steady state phase of the reaction, suggesting the formation of two distinct alkylenzyme intermediates. It was originally speculated that only one of the alkylenzymes was hydrolyzed to form product (for unknown reasons; Lindberg et al., 2008). The second rate of alkylation could also come from a different substrate-binding mode, allowing a different rate but with the same regioselectivity. In contrast, the pre-steady-state phase of the hydrolysis of the (R,R)-enantiomer of 1a displays only one amplitude, indicating that a single alkylenzyme intermediate is being formed, although a mixture of diol enantiomers are produced (see the data in Table 1).
In the present work, our goal is to provide a general explanation for these observations and for the stereoselectivities exhibited by this enzyme. To address this goal, we have both experimentally and computationally analyzed wild-type StEH1 together with four related laboratory-evolved StEH1 offspring. These enzyme variants were originally isolated for their ability to catalyze the production of enantio-enriched (2R)-3-phenylpropane-1,2-diol from a of benzyloxirane (Gurell & Widersten, 2010; Janfalk Carlsson et al., 2012, 2016). We have utilized them, together with the wild-type enzyme, as a series of structurally closely related isoenzymes (Fig. 4). Together, they represent excellent models to probe the linkage between changes in active-site structure and the corresponding effects of these changes on the observed substrate selectivity.
1a and 1b were specifically chosen as model substrates since they also allow data to be gathered during the transient pre-steady-state phase of the reactions. The experimentally determined rates could thus complement theoretical calculations and protein structure analyses. StEH1 is also unusually well suited as a model enzyme system for clarifying some fundamental aspects of enzyme catalysis and substrate recognition because (i) the chemical mechanism is well understood and its generality increases the scope of possible applicable related systems, as highly similar catalytic mechanisms are found in a wide range of hydrolytic enzymes, representing both related (α/β-hydrolases) as well as more structurally unrelated systems (for example enzymes such as the pancreatic and subtilisin); (ii) the different alternative mechanistic possibilities for epoxide ring opening provide opportunities for explaining and regioselectivity in the transformation of asymmetric substrates and, finally; (iii) crystal structures have been solved for all of the isolated variants (Mowbray et al., 2006; Bauer et al., 2016; Janfalk Carlsson et al., 2016). To further rationalize the functional data, we have performed detailed EVB calculations on the hydrolysis reaction catalyzed by both the wild-type enzyme and the variants studied here. Together, our results allow us to construct a model for through a branched reaction scheme that is able to explain both previous contradictory observations from the kinetic experiments and the unusual product ratios that are observed.
2. Materials and experimental details
2.1. Enzyme expression and purification
Wild-type StEH1 and variants were expressed in Escherichia coli XL1-Blue (Stratagene) cells carrying a plasmid with genes coding for the GroEL/ES chaperonins (Dale et al., 1994) and were purified by Ni2+–IMAC and as described previously (Elfström & Widersten, 2005). Protein concentrations were determined from the UV absorption at 280 nm using an extinction coefficient based on the value for wild-type StEH1 (Elfström & Widersten, 2005). The extinction coefficients for the R-C1B1, R-C1B1D33 and R-C1B1D33E6 variants were adjusted as described in (1),
to correct for the W106L and L109Y substitutions in these proteins.
2.2. Enzyme kinetics
The current working model of the kinetic mechanism for this enzyme is shown in Fig. 3 and is based on previous analysis of kinetics and product ratios. Rate constants are numbered accordingly.
2.2.1. Steady-state kinetics
Initial rates of hydrolysis of 1b (Fig. 1a) were recorded by following the decrease in absorbance at 229 nm (Wixtrom & Hammock, 1988), reflecting epoxide depletion. Reaction conditions were as described previously (Elfström & Widersten, 2005), with assays being performed at pH 6.8 and 30°C. Steady-state parameters were determined by nonlinear regression of the Michaelis–Menten equation to the experimental data using SIMFIT (https://www.simfit.org.uk).
2.2.2. Pre-steady-state kinetics: multiple-turnover conditions
The buildup of steady-state levels of alkylenzyme intermediates formed during the catalyzed hydrolysis of either 1a and 1b was followed by monitoring the decrease in the intrinsic tryptophan (wild type and R-C1) or tyrosine (R-C1B1, R-C1B1D33 and R-C1B1D33E6) fluorescence of the enzyme, as described previously (Elfström & Widersten, 2005). To detect tyrosine fluorescence, an excitation wavelength of 274 nm was applied and light emitted above 305 nm was detected. The apparent rates, kobs, were determined by fitting either a single exponential function with a floating endpoint,
ofor a double exponential function with a floating endpoint,
to the averaged progression curves. Averages of at least six traces were used in all cases. When used, the validity of the higher-order equation was verified by an F-test. The substrate concentrations were 30–1500 and 3–100 µM 1a and 1b, respectively. The enzyme concentrations were kept more than tenfold lower than the substrate concentration at all times in order to ensure conditions. Example averaged traces of fluorescence quenching are shown in Figs. 5(a) and 5(b). Parameter values were obtained after fitting the determined kobs values to either (4) or (5), where the latter describes a two-step process composed of a substrate-independent transition followed by a considerably faster step (Fersht, 1999). In the current model, the substrate-independent step is the E to E′ transition, with rate sums of k0 + k−0, followed by rapid equilibration of the respective Michaelis complexes as described in Fig. 3.
In cases where kobs displayed a hyperbolic substrate-concentration dependence, k2 and KS were determined after fitting (4) to the observed rates. In cases where the values of KS were considerably larger than the solubility limit of the substrate, the estimated values of k2 and KS consequently show large standard errors owing to the required extrapolation (Table 1). k−2 and k3 were calculated from the determined value of their sum (at [S] = 0), applying the derived expression for kcat (the numerator in equation 6). The relatively large error margins in the estimated values are owing to the standard errors from the regression analysis connected with the values of k2 and (k−2 + k3). Values for k2, k−2 and k3 were calculated from the experimentally determined value of (k−2 + k3) and from the derived expressions for kcat and kcat/Km and their determined values. In cases where kobs decreased with increasing [S], (5) was fitted to extract the sum of k5 + k−5 (Lindberg et al., 2008; Fig. 6).
2.2.3. Pre-steady state kinetics: single-turnover conditions
The rate of hydrolysis, k3, was directly determined by measuring the tryptophan or tyrosine fluorescence recovery following the alkylation step. This experiment was only performed with the enzyme variants in which a sufficiently low KS1b allowed fulfilment of the criteria [S] and [E] ≫ KS without exceeding the solubility limit of the substrate, so that [S]tot ≃ [ES]. Hence, the recorded observed rate of fluorescence recovery reflected the rate of the process ES→E+P and thus kobs ≃ k3. The enzyme concentrations used were in all cases 60 µM in the presence of 50 µM substrate, except for the hydrolysis of (R,R)-1b catalyzed by the wild type, where 115 µM enzyme was mixed with 100 µM substrate. These concentrations correspond to 5 × KS(R,R)-1b and 9 × KS(S,S)-1b for the wild-type enzyme, 3.8 × KS(R,R)-1b and 3.1 × KS(S,S)-1b for R-C1 and 2.4 × KS(S,S)-1b for R-C1B1. Example averaged traces are presented in Figs. 5(c), 5(d) and 5(e).
2.3. Empirical valence-bond calculations
All empirical valence-bond (EVB; Warshel & Weiss, 1980; Warshel, 1980) calculations in the present study were performed using a similar protocol to that used in our previous work on the StEH1-catalyzed hydrolysis of styrene oxide and trans-stilbene oxide (Amrein et al., 2015; Bauer et al., 2016). Full details of the simulations, as well as all EVB parameters used in the present work, are provided as Supporting Information. In brief, the EVB approach employs valence-bond (VB) theory to model chemical reactivity using a force-field description of different reacting states, with the propagation of the reaction being achieved using linear mixing of the potentials for the different states that are described. The use of Morse rather than harmonic potentials to describe bonds that are broken or formed during the reaction allows for a physically meaningful description of chemical reactivity, and this classical description is then moved to a quantum-chemical framework through the construction of an n × n Hamiltonian matrix based on the energies of the different reacting states, where n denotes the number of reacting states (for details, see Warshel, 1980; Shurki et al., 2015).
As EVB is a parameterized approach, the quality of the results provided by this approach depends on two things: (i) the quality of the parameterization used in the work (and thus the resulting EVB potentials) and (ii) the amount of conformational sampling performed and whether this is sufficient to obtain convergent results. A well parameterized EVB potential with adequate conformational sampling can easily give errors of less than 1 kcal mol−1 on the calculated energetics compared with experiment (see, for example, Barrozo et al., 2015; Blaha-Nelson et al., 2017; Kulkarni et al., 2017). In the case of the computational data presented in this work, obtaining reliable absolute energies for each through calculations is difficult owing to the uncertainities in the energetics for the corresponding uncatalyzed reaction in aqueous solution, as also discussed in Amrein et al. (2015) and Bauer et al. (2016), hence the larger deviations between the experimental and calculated values. However, the standard deviations on each calculated value (0.7 kcal mol−1 or less) are smaller than the calculated differences between the different enantiomers. Therefore, despite the limitations when considering the absolute values from the calculations, the relative values can still be compared in a meaningful way.
The EVB simulations of the hydrolysis of 1a by wild-type StEH1 and the R-C1 variant were performed using the two different proposed binding modes for (R,R)-1a and (S,S)-1a illustrated in Fig. 2 and shown using the valence-bond description in Supplementary Fig. S1. Specifically, we have modelled here first the nucleophilic attack of the Asp105 side chain on substrate 1a, followed by the subsequent activation of a water molecule to form the tetrahedral that will ultimately decompose to yield the final product diol. As this final step is likely to be extremely fast (McClelland et al., 1990) we have not modelled it in the present work, focusing our computational resources instead on the steps that are likely to be rate-limiting.
All systems were initially equilibrated for 10 ns using standard 2cjp; Mowbray et al., 2006) as well as the R-C1 variant (PDB entry 4uhb; Janfalk Carlsson et al., 2016). The corresponding uncatalyzed reaction in aqueous solution was modelled by truncating the reacting atoms to propionate, methylimidazole and 1a, and calibrated to high-level quantum-chemical calculations using density functional theory, as in our previous work on this enzyme (Amrein et al., 2015; Bauer et al., 2016). All MD and EVB simulations were performed using the Q simulation package (Marelius et al., 1998) and the OPLS-AA force field (Jorgensen et al., 1996). For further simulation details, including an in-depth description of the system setup, we refer the reader to the Supporting Information.
to provide starting points for subsequent EVB simulations. The EVB free-energy perturbation/umbrella sampling (EVB-FEP/US) procedure was then performed in 51 mapping frames of 200 ps in length to give a total of 10.2 ns of simulation time per trajectory. The EVB simulations were performed 30 times for each system, leading to a total simulation time of 14.6 µs over all systems, taking into account both the initial equilibration and subsequent EVB simulations. The EVB simulations of the enzyme-catalyzed reactions were based on the crystal structures of wild-type StEH1 (PDB entryAll subsequent energy analysis was performed using the QCalc module of Q (Marelius et al., 1998), and all structural analysis was performed using GROMACS 5.0.5 (Abraham et al., 2015). All root-mean-square deviation (r.m.s.d.) and clustering calculations were performed without mass weighting, and r.m.s.d. calculations were performed relative to the initial crystal structures in complex with the substrate.
3. Results and discussion
The StEH1 variants tested here contain between two and six substitutions, depending on the variant, in residues putatively involved in substrate binding (Fig. 4, Table 2; Janfalk Carlsson et al., 2012, 2016). The StEH1-catalyzed hydrolysis of (R,R)-1a results in a close-to-racemic mixture of the diol enantiomers (at the assay temperature used here, 30°C; Lindberg, de la Fuente Revenga et al., 2010; regiopreference shown in Table 1). The EVB simulations similarly predict this low degree of regioselectivity with (R,R)-1a, with very similar barriers for hydrolysis of either the (R,S)- or (S,R)-alkylenzymes (C1 or C2; mode 1 in Table 3 and Figs. 7a and 8a). In addition, the calculated energy barriers suggest that in order to result in productive Michaelis complex(es), the substrate needs to be bound to StEH1 in mode 1 (Fig. 2 and Supplementary Fig. S5), in which the phenyl substituent points towards His300. This is the same preferred binding mode as has previously been suggested for the analogous StEH1-catalyzed hydrolysis of styrene oxide (Bauer et al., 2016).
|
‡Calculated from kcat. §Calculated from k2. ¶Calculated from k3. |
Our EVB calculations for the hydrolysis of (R,R)-1a suggest the following through the pathway modelled in Fig. 3. (i) Nucleophilic attack occurs preferentially at C2, but attack at C1 is also possible, albeit to a much lower degree. (ii) The barriers for hydrolysis of the alkylenzymes are equally high, which is in line with the observed product ratio (Figs. 7 and 8). The finding that the formation of one alkylenzyme is favoured over the other fits well with the experimental observations, where only one amplitude is detected although two product enantiomers are formed. Still, the formation of two product diols by necessity invokes the formation of an alkylenzyme after attack at C1 also, albeit at a lower, undetectable level. We note as an aside that as in our previous calculations of the hydrolysis of styrene oxide (Bauer et al., 2016), as well as previous DFT calculations of epoxide hydrolase-catalyzed epoxide hydrolysis (Lind & Himo, 2016), we obtain unphysically free energies for the initial ring opening (i.e. the alkylenzyme intermediate is far too exothermic compared with what would be expected from considering the experimental rates; see Table 1, which shows both k2 and k−2 where available). As discussed previously (Bauer et al., 2016), we believe this to be a simulation artifact from the DFT calculations used to perform the fitting (see the Supporting Information). We have thus considered the energetics of the two reacting steps individually in all cases.
Following from this, the hydrolysis of (S,S)-1a yields only (R,S)-2a, suggesting that attack at C1 is favoured. The paradox here is that two distinct amplitudes are observed during the pre-steady state, with lifetimes that are described by two different observed rates (Lindberg et al., 2008). Once again, our EVB calculations provide a plausible explanation for the contradictory observations of product ratios and the (postulated) presence of two alkylenzymes. The calculated energy barriers suggest that (S,S)-1a is also bound productively in mode 1 for the formation of alkylenzyme(s). Attack by the Asp105 carboxylate on the epoxide ring is possible at either C atom (Fig. 2 and Supplementary Fig. S6), with closely matched activation energies, thereby allowing the formation of either the pro-(1R,2S)-diol or pro-(1S,2R)-diol alkylenzyme (Table 3, Figs. 7 and 8). The energy barriers resulting from the simulation agree with the presence of two experimentally observed amplitudes in the pre-steady-state phase of the reactions. The barriers for the subsequent hydrolysis step, however, will only allow reaction of the pro-(1R,2S)-diol alkylenzyme; the activation energy for hydrolysis of the other alkylenzyme is >4 kcal mol−1 higher (Table 3, Figs. 7 and 8). This can explain the exclusive formation of the (1R,2S)-2a diol. Furthermore, the calculated barriers agree reasonably well with those back-calculated from determined microscopic rates, or kcat in the case of (R,R)-1a, and also emphasize that hydrolysis is indeed the rate-determining step in the overall catalytic cycle with both enantiomers.
3.1. R-C1
The two substitutions in the R-C1 variant (V141K and I155V) cause only small effects on kcat or kcat/Km, with the exception of the kcat for (R,R)-1a, which is increased by fivefold (Table 1). As the Km for (R,R)-1a is also elevated to a comparable extent, the overall enzyme efficiency is unaltered. This increase in turnover number may be owing to an elevated interconversion rate between the ES and E′S species (k5 + k−5 in Fig. 3), supported by the sixfold-increased value of the sum of these rates. The estimated value is similar to that previously determined for the wild-type enzyme at elevated temperatures (Lindberg, de la Fuente Revenga et al., 2010). In the wild type, the value of this sum of rates is similar to the value of kcat and may be rate-limiting at 30°C, but not at higher temperatures. The increase in the isomerization of the Michaelis complexes with this substrate may suggest that the mutations in R-C1 introduce some level of increased structural flexibility.
expressed as eitherThe activity with (S,S)-1a measured at the is essentially unchanged. Effects from the substitutions can, however, be seen on the transient rates and on the dissociation constant of the Michaelis complex, KS (Table 1). Similarly to wild-type StEH1, the transient alkylenzyme buildup and decay in the catalyzed hydrolysis of (S,S)-1a can only be modelled satisfactorily by a second-order exponential function. Hence, two distinct rates and amplitudes separated in time are recorded (Fig. 5a). The faster observed rate (kobs1 in Fig. 6a) follows a (`normal') hyperbolic dependence on substrate concentration and is modelled by (4), which describes rapid (pre-equilibrated) formation of the Michaelis complex, followed by alkylenzyme formation and decay. The determined values of the microscopic rates show an approximately fivefold increase in the alkylation rate (k2), while the decay rates (k−2 and k3) are essentially unchanged (Table 1). The estimated (k5 + k−5) rate from the concentration-dependence of the slower observed transient rates with (S,S)-1a (kobs2 in Fig. 6a) is not significantly different from the corresponding wild-type value.
Structurally, we have considered the changes in the root-mean-square fluctuations (r.m.s.f.) of all backbone atoms during the equilibration of the Michaelis complexes for both possible binding modes of (R,R)-1a and (S,S)-1a in complex with the wild-type and R-C1 variant of StEH1 (Supplementary Fig. S2). We have also calculated the corresponding dynamic changes in active-site volume during these equilibration runs using the POVME 2.0 pocket-volume measurer (Durrant et al., 2011, 2014; Table 4). From the r.m.s.f. plots, it can be seen that globally the backbone of the R-C1 variant appears to be more flexible than that of wild-type StEH1. The exception to this, however, appears to be in the regions around Tyr154 and in some instances Tyr235, where the incorporation of the I155V substitution appears to dampen the flexibility of these catalytically important tyrosine residues. We note that the overall flexibility trends mostly appear to be independent of the or binding mode (i.e. the four plots are qualitatively similar). Tying in with this, the calculated active-site volumes shown in Table 4, as well as the corresponding standard deviations (which indicate how much the active-site volume changes over the course of the corresponding equilibration runs), show that for both enantiomers and binding modes of 1a the active-site volume is more compact in the Michaelis complexes with the R-C1 variant than those with wild-type StEH1. We note, however, that these R-C1 volumes should be assessed also taking into account the flexibility of the Lys141 side chain, which is not present in the wild-type enzyme, and which will impact the POVME calculations based on whether the side chain is positioned in the active site or out of it in a given snapshot.
Similar to the wild type, the equal barriers for R-C1-catalyzed ring opening of (S,S)-1a contradict the fact that only (R,S)-2a is formed after its hydrolysis catalyzed by either the wild type or R-C1. However, as in the wild-type parent, the barrier for hydrolysing the intermediate formed after ring opening at C2 is 5 kcal mol−1 higher (Table 3). Thus, the stereoconfiguration of the diol product again becomes exclusively (1R,2S)-2a. The fact that the selectivity is determined at the hydrolysis step and not the alkylation step is in good agreement with our previous computational studies of the hydrolysis of other substrates by these enzymes (Amrein et al., 2015; Bauer et al., 2016). The calculated energy barriers for the hydrolysis of (S,S)-1a nicely explain the experimental observations of two transient amplitudes, indicative of two alkylenzymes, but resulting in only one product Also, although the magnitude is overestimated, the faster alkylation rate (k2 in Table 1) determined for R-C1 is caught by the calculations as lower barriers for ring opening compared with the wild type (Table 3).
The fluorescence-quenching signal recorded with all enzyme variants and with both enantiomers of 1b could be fitted by a first-order exponential function in all cases. The catalytic activities of R-C1 are comparable to those of the wild-type enzyme, with the exception of the alkylation rate with (S,S)-1b which, similar to (S,S)-1a, is fivefold faster (Table 1). The dissociation constants with both these are also elevated to approximately the same degree, which may be connected to the increased alkylation rates.
3.2. R-C1B1
The R-C1B1 variant contains two additional substitutions: W106L and L109Y. Trp106 is a relatively conserved residue in epoxide α/β-hydrolase family (The Epoxide Hydrolase Database; https://www.led.uni-stuttgart.de; Barth et al., 2004) and contributes its main-chain amide to the oxyanion hole, stabilizing the tetrahedral intermediate formed during the hydrolytic decomposition of the alkylenzyme (Fig. 1c). The indole of this residue has also been suggested to contribute favourably to the binding of aryl-substituted such as 1b and styrene oxide (Amrein et al., 2015; Bauer et al., 2016). The second substitution, L109Y, spatially located adjacent to Trp106, does not reintroduce an aromatic ring to the substrate-binding site since the phenolic side chain swings out of the active site and engages in a hydrogen bond to the side chain of Asn241 (Fig. 4b).
of theThe structural changes in this variant cause relatively larger effects on the catalytic rates (Table 1), with a 30-fold decrease in kcat/Km with (R,R)-1a compared with R-C1. The decrease in is primarily manifested owing to a lower turnover number (kcat) when compared with R-C1. Owing to the absence of a clear dependence of the observed rates (kobs) on substrate concentration, we have been unable to determine the individual rate constants under multiple-turnover conditions for the R-C1B1-catalyzed hydrolysis of either (R,R)-epoxide. This can occur if two different alkylenzymes are produced on the same timescale. Therefore, we did not attempt to perform EVB calculations for this or the following variants, as the missing information about the individual rates would make the comparison between simulation and experiment meaningless. Under such circumstances the different signal amplitudes can mix, with the observable result of an apparent overall lack of substrate-concentration dependence. The inset in Fig. 6(c) shows the behaviour of the observed transient rates as a function of (R,R)-1a concentration. A possible increase in the observed rates may be present at higher substrate concentrations, rather than the decrease seen for the wild-type and R-C1 enzymes. The activity with (S,S)-1a was less affected, with a mere fourfold decrease in kcat/Km compared with R-C1. There was no indication of two observed transient rates (Fig. 5b), suggesting a rapid equilibrium between the two Michaelis complexes. The most notable effect on the kinetics is an approximately 20-fold decrease in the decomposition rate from the alkylenzyme back to the Michaelis complex (k−2). Since the hydrolysis rate is also slower compared with the wild type and R-C1, this is expected to stabilize the alkylenzyme to a greater degree, which may be the cause of the lower hydrolysis rate (and kcat).
The corresponding rates with (S,S)-1b were also altered; the alkylation rate was almost 15-fold lower compared with R-C1, while the rates of alkylenzyme decay were similar for the two variants. This lower alkylation rate explains the tenfold increase in the Km of (S,S)-1b in this variant (see the expression for the Km term in the denominator of equation 6), and also identifies the alkylation step to be partially rate-limiting in this case.
The relatively tight binding of the 1b enantiomers to these enzymes made single-turnover experiments possible, which allows the direct extraction of the hydrolysis rates, k3. With (S,S)-1b, the observed single-turnover rates were close to the cognate kcat values (Table 1), underlining that hydrolysis is indeed rate-determining in these reactions. On the other hand, for wild-type- and R-C1-catalyzed hydrolysis of (R,R)-1b the determined hydrolysis rates were substantially faster than kcat, suggesting that another, yet unidentified, step on the reaction pathway may be rate-limiting and contributing to the value of kcat. The alkylation rates (k2 in Table 1) clearly showed that this step is not rate-limiting for kcat, which is corroborated by the calculated barriers of the wild-type reaction (Amrein et al., 2015). Hence, a step downstream of the hydrolysis step is suspected to influence kcat in these reactions.
3.3. R-C1B1D33
This variant has picked up one additional substitution: F189L (Fig. 4). The replacement has not caused any major effects on the catalytic efficiencies with either epoxide 1a or 1b compared with its parent R-C1B1 (Table 1). There are, however, some noteworthy changes: kcat with either (S,S)-epoxide is increased. The observed rates of alkylenzyme formation and decay during the pre-steady-state phase could, as in the case of R-C1B1, be satisfactorily modelled by a single exponential. The estimated microscopic rates with (S,S)-1a point towards a lower energy barrier for alkylenzyme formation coupled with a decreased affinity for the substrate [KS(S,S)-1a is >6 mM]. This suggests that the inserted substitution allows a higher degree of flexibility of the Michaelis complex, possibly facilitating productive binding modes for nucleophilic attack. The microscopic rates with (S,S)-1b were increased to values that escaped detection outside the of the stopped-flow equipment.
3.4. R-C1B1D33E6
This final, fourth-generation variant contains an additional L266G substitution (Fig. 4). This mutation is detrimental to the activity with either of the (R,R)-enantiomers tested here (Table 1). The activities with the (S,S)-epoxides are also lower (20–60-fold with 1a and 1b, respectively). The steady-state parameters with (S,S)-1b indicate that both turnover and Km have been affected. As discussed in Janfalk Carlsson et al. (2016), this variant has the largest active-site cavity: ∼250 Å3 larger than its parent. This larger volume may allow new binding modes that are favourable for the initial binding step but unfavourable for the subsequent formation of the alkylenzyme. We note that this variant was originally isolated for its ability to transform benzyloxirane into (2R)-3-phenylpropane-1,2-diol; hence, the of the product was prioritized over catalytic efficiency in this case (Janfalk Carlsson et al., 2012).
4. Conclusions
It has been established that the epoxide ring opening and hydrolysis catalyzed by StEH1 proceeds via an SN2 mechanism, causing inversion of configuration at the attacked epoxide C atom (Lindberg et al., 2008; Janfalk Carlsson et al., 2012). A mixture of diol products after hydrolysis therefore requires the presence of branched reaction pathways. We show here for the first time that through such complex reaction pathways is affected by kinetic rates between noncovalent Michaelis complexes, the rates of formation of the respective covalent intermediates and the rates of their subsequent hydrolyses. Specifically, by employing a series of related StEH1 variants, evolved in vitro, we have been able to link structural effects, in particular changes in the active-site cavity as well as preferred binding modes across different variants, to stepwise changes in catalytic function. Providing a molecular understanding of how such subtle structural effects can manipulate complex kinetic pathways is a major advance towards the efficient generation of novel enzymes with tailored enantioselectivities and regioselectivities.
Supporting information
Details of computational methodology, supplementary figures and tables. DOI: https://doi.org/10.1107/S2052252518003573/jt5024sup1.pdf
Footnotes
‡These authors contributed equally to this work.
Acknowledgements
All computational work was performed on the Abisko and Kebnekaise clusters at HPC2N through the generous allocation of computational time by the Swedish National Infrastructure for Computing (SNIC; 2015/16-12 and 2016/34-27).
Funding information
This work was supported by the Swedish Research Council (grant 621-2011-6055) and COST Action 1303, Systems Biocatalysis to Mikael Widersten. The European Research Council provided financial support under the European Community's Seventh Framework Programme (grant No. FP7/2007–2013 to S. C. Lynn Kamerlin). Finally, we thank the Sven and Lily Lawski Foundation for a doctoral fellowship to Paul Bauer.
References
Abraham, M. J., Murtola, T., Schulz, R., Páll, S., Smith, J. C., Hess, B. & Lindahl, E. (2015). SoftwareX, 1–2, 19–25. CrossRef Google Scholar
Amrein, B. A., Bauer, P., Duarte, F., Janfalk Carlsson, Å., Naworyta, A., Mowbray, S. L., Widersten, M. & Kamerlin, S. C. L. (2015). ACS Catal. 5, 5702–5713. Web of Science CrossRef CAS Google Scholar
Barrozo, A., Duarte, F., Bauer, P., Carvalho, A. T. P. & Kamerlin, S. C. L. (2015). J. Am. Chem. Soc. 137, 9061–9076. Web of Science CrossRef CAS Google Scholar
Barth, S., Fischer, M., Schmid, R. D. & Pleiss, J. (2004). Bioinformatics, 20, 2845–2847. Web of Science CrossRef CAS Google Scholar
Bauer, P., Janfalk Carlsson, Å., Amrein, B., Dobritzsch, D., Widersten, M. & Kamerlin, S. C. L. (2016). Org. Biomol. Chem. 14, 5639–5651. Web of Science CrossRef CAS Google Scholar
Blaha-Nelson, D., Krüger, D. M., Szeler, K., Ben-David, M. & Kamerlin, S. C. L. (2017). J. Am. Chem. Soc. 139, 1155–1167. CAS Google Scholar
Dale, G. E., Schönfeld, H.-J., Langen, H. & Stieger, M. (1994). Protein Eng. 7, 925–931. CrossRef CAS Web of Science Google Scholar
Durrant, J. D., de Oliveira, C. A. & McCammon, J. A. (2011). J. Mol. Graph. Model. 29, 773–776. Web of Science CrossRef CAS Google Scholar
Durrant, J. D., Votapka, L., Sørensen, J. & Amaro, R. E. (2014). J. Chem. Theory Comput. 10, 5047–5056. Web of Science CrossRef CAS Google Scholar
Elfström, L. T. & Widersten, M. (2005). Biochem. J. 390, 633–640. Google Scholar
Elfström, L. T. & Widersten, M. (2006). Biochemistry, 45, 205–212. Google Scholar
Fersht, A. (1999). Structure and Mechanism in Protein Science. New York: W. H. Freeman & Co. Google Scholar
Gurell, A. & Widersten, M. (2010). ChemBioChem, 11, 1422–1429. Web of Science CrossRef CAS Google Scholar
Heikinheimo, P., Goldman, A., Jeffries, C. & Ollis, D. L. (1999). Structure, 7, R141–R146. Web of Science CrossRef PubMed CAS Google Scholar
Janfalk Carlsson, Å., Bauer, P., Dobritzsch, D., Nilsson, M., Kamerlin, S. C. L. & Widersten, M. (2016). ChemBioChem, 17, 1693–1697. CAS Google Scholar
Janfalk Carlsson, Å., Bauer, P., Ma. H. & Widersten, M. (2012). Biochemistry, 51, 7627–7637. Google Scholar
Jorgensen, W. L., Maxwell, D. S. & Tirado-Rives, J. (1996). J. Am. Chem. Soc. 118, 11225–11236. CrossRef CAS Web of Science Google Scholar
Kulkarni, Y. S., Liao, Q., Petrović, D., Krüger, D. M., Strodel, B., Amyes, T. L., Richard, J. P. & Kamerlin, S. C. L. (2017). J. Am. Chem. Soc. 139, 10514–10525. Web of Science CrossRef CAS Google Scholar
Lind, M. W. S. & Himo, F. (2016). ACS Catal. 6, 8145–8155. Web of Science CrossRef CAS Google Scholar
Lindberg, D., Ahmad, S. & Widersten, M. (2010). Arch. Biochem. Biophys. 495, 165–173. Web of Science CrossRef CAS Google Scholar
Lindberg, D., de la Fuente Revenga, M. & Widersten, M. (2010). Biochemistry, 49, 2297–2304. Web of Science CrossRef CAS Google Scholar
Lindberg, D., Gogoll, A. & Widersten, M. (2008). FEBS J. 275, 6309–6320. Web of Science CrossRef CAS Google Scholar
Marelius, J., Kolmodin, K., Feierberg, I. & Åqvist, J. (1998). J. Mol. Graphics Model. 16, 213–225. Web of Science CrossRef CAS Google Scholar
McClelland, R. A., Kanagasabapathy, V. M. & Steenken, S. (1990). Can. J. Chem. 68, 375–382. CrossRef CAS Web of Science Google Scholar
Monterde, M. I., Lombard, M., Archelas, A., Cronin, A., Arand, M. & Furstoss, R. (2004). Tetrahedron Asymmetry, 15, 2801–2805. Web of Science CrossRef CAS Google Scholar
Mowbray, S. L., Elfström, L. T., Ahlgren, K. M., Andersson, C. E. & Widersten, M. (2006). Protein Sci. 15, 1628–1637. Web of Science CrossRef CAS Google Scholar
Parker, R. E. & Isaacs, N. S. (1959). Chem. Rev. 59, 737–799. CrossRef CAS Web of Science Google Scholar
Pinot, F., Grant, D. F., Beetham, J. K., Parker, A. G., Borhan, B., Landt, S., Jones, A. D. & Hammock, B. D. (1995). J. Biol. Chem. 270, 7968–7974. CrossRef CAS Web of Science Google Scholar
Reetz, M. T., Bocola, M., Wang, L.-W., Sanchis, J., Cronin, A., Arand, M., Zou, J., Archelas, A., Bottalla, A. L., Naworyta, A. & Mowbray, S. L. (2009). J. Am. Chem. Soc. 131, 7334–7343. Web of Science CrossRef CAS Google Scholar
Santaniello, E., Ferraboschi, P., Grisenti, P. & Manzocchi, A. (1992). Chem. Rev. 92, 1071–1140. CrossRef CAS Web of Science Google Scholar
Shurki, A., Derat, E., Barrozo, A. & Kamerlin, S. C. L. (2015). Chem. Soc. Rev. 44, 1037–1052. Web of Science CrossRef CAS Google Scholar
Stapleton, A., Beetham, J. K., Pinot, F., Garbarino, J. E., Rockhold, D. R., Friedman, M., Hammock, B. D. & Belknap, W. R. (1994). Plant J. 6, 251–258. CrossRef CAS Web of Science Google Scholar
Thomaeus, A., Carlsson, J., Åqvist, J. & Widersten, M. (2007). Biochemistry, 46, 2466–2479. Web of Science CrossRef CAS Google Scholar
Tzeng, H.-F., Laughlin, L. T. & Armstrong, R. N. (1998). Biochemistry, 37, 2905–2911. Web of Science CrossRef CAS Google Scholar
Tzeng, H.-F., Laughlin, L. T., Lin, S. & Armstrong, R. N. (1996). J. Am. Chem. Soc. 118, 9436–9437. CrossRef CAS Web of Science Google Scholar
Warshel, A. (1980). Computer Modeling of Chemical Reactions in Enzymes and Solutions. New York: Wiley. Google Scholar
Warshel, A. & Weiss, R. M. (1980). J. Am. Chem. Soc. 102, 6218–6226. CrossRef CAS Web of Science Google Scholar
Widersten, M., Gurell, A. & Lindberg, D. (2010). Biochim. Biophys. Acta, 1800, 316–326. Web of Science CrossRef CAS Google Scholar
Wixtrom, R. N. & Hammock, B. D. (1988). Anal. Biochem. 174, 291–299. CrossRef CAS Web of Science Google Scholar
This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.