Understanding Biological Hydrogen Transfer Through the Lens of Temperature Dependent Kinetic Isotope Effects.

Introduction:

For decades, this laboratory has been focused on the study of enzymatic C-H bond cleavage reactions that occur by quantum-mechanical tunneling mechanisms. These processes have unique kinetic features that provide insight into the complete shape, not just the height of the barrier for catalysis. The ability to stably label C-H bonds with hydrogen isotopes enables the magnitude of the kinetic isotope effect (KIE) to be determined and related to perturbations of the reaction coordinate arising from changes in e.g., temperature, pressure or protein mutation.1–10 Despite years of experimental findings that have led to the current prevailing theoretical constructs for hydrogen transfer from carbon, common misconceptions frequently still arise from non-specialists, including the elimination of tunneling mechanisms based on the magnitude of experimental KIEs. As is highlighted herein, one of the most robust kinetic parameters to emerge from such studies of hydrogen transfer is the difference between the enthalpic barriers for protium and deuterium wavefunction overlap, (i.e. the temperature dependence of the kinetic isotope effect, ΔEa = Ea(D) – Ea(H)). Because of the greater mass of deuterium and its less diffuse wavefunction, the magnitude of Ea(D) provides a measure of an enzyme’s ability to sample the short H-donor-acceptor distances (2.7 Å) conducive to tunneling. This mini-review will further emphasize the catalytic role of hydrophobic side chains in enzymes whose mutational impacts can, in certain instances, only be evident from the ΔEa parameter.

There is a rich history regarding mechanistic interpretations of ΔEa in condensed phase, beginning with the original expectation that the Arrhenius plot of rates for H vs D transfer would converge on the y-axis, AH/AD = 1, if a reaction proceeds classically, and that the KIE would arise predominantly from differences in isotopically sensitive zero-point energies (cf. Figure 1A).11 As introduced by Bell,12 room temperature through-the-barrier tunneling can also occur near the top of the classical barrier, leading to AH < AD (Figure 1B). Lastly, in the context of enzyme-catalyzed hydrogen tunneling, another property has become apparent for native enzymes, in which ΔEa values are close to zero (AH/AD = KIE >> 1) across the experimentally accessible temperature range (Figure 1C).

Importantly, as revealed from decades of study of protein enzyme catalysis, a distinctive pattern for the temperature dependence of the KIE emerges when a native enzyme is subjected to either intrinsic or extrinsic perturbations (Figure 2). Beginning with the most active (wild-type, WT) form of enzyme, ΔEa is small, often close to zero (Figure 2A). Subsequently, as the enzyme system is impaired, the kinetic behavior can appear to “mimic” semi-classical behaviors where AH/AD = 1 (Figure 2B). And finally, when perturbation to the enzyme becomes extreme, ΔEa will greatly exceed the semi-classical limit, resembling tunneling correction models where AH/AD << 1 (Figure 2C). In select instances, these trends in ΔEa are accompanied by nearly constant KIE values under physiological conditions, raising the compelling question of how to arrive at a predictable physical model that incorporates the range of observed kinetic properties. From the perspective of traditional, semi-classical theory it is, simply put, impossible to explain the behavior of native enzymes (Figure 2A). Also, while the patterns in Figure 2 could suggest a progression to semi-classical behavior (Figure 2B), that is followed by a role for significant tunneling (Figure 2C), this would be in direct conflict with structural evidence of increased hydrogen donor-acceptor distances following increased enzyme impairment.13,14

Significantly, nonadiabatic theoretical models for tunneling mechanisms (Figure 3) are able to explain both the native behavior of enzyme (Figure 2A) as well as the changes that arise when a system undergoes perturbation (Figure 2B-C).2,5,10,15–18 On the surface, these theoretical constructs are extensions of Marcus models for electron transfer, that describe the reaction barrier in terms of the environmental reorganization energy, λ, and reaction free energy, ΔGo (Figure 3B); this coordinate controls the achievement of transient, reactant-product degeneracy (Figure 3C), a pre-requisite for wavefunction overlap.19 In addition, another aspect uniquely emerges in hydrogen tunneling reactions, as a result of the heightened sensitivity of nuclear wavefunction overlap to small changes in DAD. Illustrated in Figure 3D, local dynamical sampling over an ensemble of donor-acceptor distances (DADs) can become important when the distance between the H-donor and acceptor is outside the regime of a dominant tunneling distance, Rdom, of 2.7 Å.2,20,21 The experimental manifestation of the behavior in Figure 3D is ΔEa. In the context of the model, ΔEa values of zero in native enzyme systems indicate a tunneling ready state (TRS) where the DAD has been reduced from Req to a sufficiently short distance (R0) to promote efficient wavefunction overlap of both protium and deuterium, leading to a minimal contribution of local DAD sampling to the reaction barrier (Figure 3D, top). The introduction of protein packing defects through, for example, the strategic mutagenesis of hydrophobic side chains causes ΔEa values to become enlarged (Figure 3D, bottom), at first appearing to approximate semi-classical “over the barrier” catalytic behavior and then, with increasing impairment, appearing to approximate “tunneling correction” models. The beauty of this model is its ability to link regular increases in ΔEa to structurally-induced increases in DADs. In almost all instances this can be compensated by local dynamical sampling that enhances wavefunction overlap. The latter plays a disproportionate role for D-transfer, arising from an isotope-dependent change in wavefunction overlap that is completely distinct from the zero point energy-based, semi-classical origin of primary hydrogen KIEs.11

In the context of the behavior illustrated in Figure 2 and the theoretical model in Figure 3, we now introduce three different HAT systems that have yielded a combination of KIEs and/or ΔEa parameters. These include taurine/α-ketoglutarate dioxygenase, the dual copper β-monooxygenases and soybean lipoxygenase. Each system has been chosen to illustrate different aspects of the experimental challenges that have been encountered in the course of mechanistic studies and the resulting insights that lead to our current perspective. The reactions catalyzed by each enzyme system are illustrated in Scheme 1, while Figure 4 presents representative X-ray structures.

TauD illustrates the impact of isotope- and mutation-dependent reaction uncoupling during C-H oxidation.

TauD, responsible for the hydroxylation of taurine, is a member of the α-ketoglutarate (αKG) dependent mono- and dioxygenases. The diverse reactions in this family are initiated by the decarboxylation of αKG leading to the production of a non-heme oxyferryl intermediate that catalyzes the cleavage of substrate C-H bonds (Scheme 1A).25 TauD had been reported to exhibit a large, intrinsic deuterium KIE, k(H)/k(D) = Dkcat ≅ 60 under the partial reaction of hydrogen atom transfer from C1 of taurine to the oxygen acceptor on FeIV=O, and has served as a model system for extensive spectroscopic and structural characterization of iron intermediates within this family of enzymes.26,27 The X-ray structure of substrate-bound tauD (Figure 4A) displays a metal-bound αKG and includes substrate taurine sandwiched approximately equidistant between iron (ferrous state) and F159. Following the initial report of the large isotope effects suggestive of tunneling behavior,26 our group turned to an investigation of the possible role of sidechain bulk at position 159 in tuning tunneling efficiency. The expectation was that replacement of F159 by smaller hydrophobic sidechains would concomitantly expand the active size cavity and reduce the ability of protein to promote hydrogenic wavefunction overlap.

As shown in Table 1, substitution of F159 by L results in modest changes to rate constants, with a significantly greater impact under pre-steady-state conditions.28 Importantly the intrinsic KIE, as reflected in the pre-steady-state Dk(prod), Table 1, is nearly the same (within error) for WT and F159L. The fact that the KIEs are within error, while the rate of hydrogen atom abstraction is reduced ca 20-fold, suggested a successful recovery for F159L to the native donor-acceptor distance via the distance sampling term represented in Figure 3D. Proceeding through the series to the less bulky sidechain, F159V, the fact that a 50-fold reduction in the pre-steady-state k(prod) leads to an elevation of Dk(prod) to ca. 200, further suggested that F159V had crossed a threshold whereby DAD sampling was no longer capable of recapturing the close DAD of WT.28

In principle, the predicted trends in the DAD sampling modes for the F159X could have been tested by measurement of the corresponding values for ΔEa.28 However, while WT tau D catalyzes a tightly coupled reaction, in which O2 uptake and product formation occur in a 1:1 ratio, perturbations via substrate deuteration and/or mutation lead to increasingly large degrees of reaction uncoupling, Figure 5. As a result, our ability to determine precise intrinsic rate constants became limited to the conditions of Table 1 with estimates of Dk(prod) not even attempted for the Ala and Gly mutants.29 TauD thus illustrates some of the potential complications that can arise in experimental efforts to estimate ΔEa values. As shown in this case, pre-steady state conditions may be required to separate the C-H cleavage step from other rate determining steps (Table 1) while reaction uncoupling (Figure 5) may compromise the ability to obtain sufficiently precise intrinsic rate constants for a reliable determination of trends in the ΔEa parameter.

PHM illustrates estimation of intrinsic deuterium KIEs and their temperature dependence via precise measurement of tritium KIEs.

PHM is a member of a family of eukaryotic monooxygenases that include dopamine β-monooxygenase (DβM) and tyramine β-monooxygenase (TβM) and contain two protein-bound, uncoupled copper atoms,30 separated by a ≥10 Å solvent cleft (Figure 4B).31 Despite differences in substrate/product structures, these reactions occur by a highly conserved chemical mechanism that is initiated via hydrogen atom abstraction at an activated oxygen within the CuM site (Figure 4B).3 In contrast to tauD, and of importance for characterization of HAT in this family of enzymes, substrate and oxygen activation occur in a fully coupled manner.3,32 However, similar to the tauD reaction, steady-state rate constants are largely rate limited by substrate binding and product release. In the case of tauD, pre-steady state decay rates determined from the spectral signal of FeIV=O enabled direct determination of the intrinsic values for k(prod) and Dk(prod). The lack of an amenable spectral handle for active site chemistry in the copper monoxygenases required an alternate kinetic strategy. Isolation of the intrinsic isotope effects on C-H bond cleavage was made possible by developed methods for the estimation of highly precise competitive kinetic measurements of primary tritium isotope effects, kH/kT and kD/kT that allow calculation of the magnitude of kH/kD.1,33

Using this approach, PHM has been shown to display an intrinsic primary isotope effect, ~11,1 elevated from the semi-classical limit (ca. 7), that compares well to values estimated for DβM33 and TβM34 (Table 2). Extension of this intrinsic primary isotope effect measurement to a range of physiological temperatures (Figure 6),1 indicates a relatively flat dependence with an isotope effect on the Arrhenius prefactor, AH/AD, that is well above the semi-classical limit of unity and an accompanying value for ΔEa that is essentially zero (Table 2). We note that while the size of the intrinsic isotope effects for the hydrogen atom transfer catalyzed by PHM and the related copper dependent monooxygenases are near the Bell tunneling correction limit (KIE ~ 10),12 it was only after interrogation of the ΔEa (~0 kcal/mol) that such behavior could be ruled out. It was shown possible to model this behavior of PHM using the analytical equations described in greater detail for SLO below.1,2,10,15,18,22

SLO provides insight into role of hydrophobic sidechain bulk on active site precision through mutational impact to KIE and ΔEa

SLO oxidizes its substrate, linoleic acid (LA), via an initial, rate-limiting abstraction of a hydrogen atom from position C11 of the substrate by an active site ferric-hydroxide, in which a proton is transferred to the iron-bound hydroxide and an electron reduces the ferric center to ferrous (Scheme 1C). The very large Dkcat (ca. 80) near room temperature that is weakly temperature dependent (ΔEa = 0.9 ± 0.2 kcal/mol) has enabled comparative measurement of KIEs under a wide variety of conditions. Because the active site is deeply buried and surrounded by bulky hydrophobic residues, SLO has further established itself as an especially attractive model to probe the contributions of individual amino acids in mediating active site tunneling-appropriate DADs.

Single alanine “scanning” was initially applied to hydrophobic sidechains, including two proximal, conserved leucines, L546 and L754 and a more distal isoleucine, I553.2 The reduction in sidechain volume at either 546 or 754 results in expanded active site cavities and altered kinetic features, such as decreased kcat values with concomitant increases to the activation energy parameters, Ea and ΔEa (Table 3).

For I553A alone, the kcat and Ea are comparable to WT with the only parameter impacted being the ΔEa (4.0 kcal/mol), whereas the double mutant I553A/L546A results in a value for ΔEa that is midway between WT and I553A, with an “apparent” semi-classical pattern for AH/AD of unity. This combination of results can be readily explained by mutationally-produced changes in DAD that are accompanied by enhanced local dynamic sampling to restore catalytic proficiency (Figure 3D).2

Subsequently, the mutational analysis of I553A was followed by preparation of a series of aliphatic mutants of varying sidechain volume at this position, Table 4. Extensive X-ray analyses of this series in the absence of substrate have ruled out significant structural changes with decreasing sidechain volume, while kinetic analyses highlight a remarkable trend in the size of ΔEa5. According to developed analytical expressions for the temperature dependence of the KIE,2,15,18,22 both a pre-tunneling donor-acceptor (R0) and a frequency for the DAD sampling coordinate (Ω,) can be estimated, illustrating the significant impact of hydrophobic side chain mutations on trends in both the R0 and Ω.

The modest ΔEa value for WT SLO supports a highly evolved, compacted active site. In contrast, the ΔEa data for I553X indicate a DAD sampling coordinate that begins to contribute significantly when packing defects both alter/enlarge the initial DAD and decrease the force constant of the local DAD sampling mode.

The I553X variants offer a unique opportunity to investigate the relationship between dominant ground state enzyme-substrate structures and the kinetic detection of changes in active site compaction.

However, there has been the perennial inability to capture co-crystal structures of SLO with substrate, leading to a search for alternative methods to examine enzyme-substrate complexes. ENDOR is often regarded as “EPR-detected NMR”, because of its ability to measure interactions of nuclei, such as 1H, 2H, 13C and 15N, within close proximity to unpaired electron spins.37,38Application of ENDOR to SLO leads not only towards the depiction of the ground state structure of the native enzyme (Figure 7A), but also towards a sensitive assay for structural impacts from seemingly modest active site mutations.14

The desired structure of the WT SLO-LA complex was generated using a structurally-faithful, manganese (S = 5/2) substituted enzyme samples and selectively 13C-labeled substrates (at the reactive carbon C11 as well as the adjacent carbon, C10 as a reference point). With a focus on C11, the 13C ENDOR response of SLO is summarized in Figure 7.14 The data (Figure 7A) indicate two conformations, with the larger shifted feature showing a shoulder at ~0.4 MHz representing the “active” a substrate conformation that places the hydrogen donor atom (C11) within van der Waals distance of the acceptor oxygen. The second b conformation reflects a substrate conformation positioned nearly 1 Å further from the hydrogen acceptor oxygen and is anticipated to be non-catalytic (cf. Figure 7B). We expanded the investigation of WT to include the impact of site specific mutations, with studies of the most extreme I553 mutant, I553G, leading to a complete loss of the a conformer (Figure 7B). In short, the I553 variant with the largest ΔEa value (Table 4) has reduced the population of the near, active site enzyme-substrate conformer to a point where it can no longer be detected experimentally. While an observation of ΔEa = 5.3 ± 0.7 for I553G could, on the surface, have been compatible with tunneling correction models as originally introduced by Bell, it seems inconceivable that a mutant that increases the effective DAD distance would be capable of increased tunneling relative to WT. In the context of the critical dependence of tunneling on a close approach of the H-atom donor and acceptor, this combination of kinetic data and biophysical probes for SLO provides some of the strongest validation of the originally proposed multi-dimensional analytical model for HAT illustrated in Figure 3.

Comparison of analytical constructs for nonadiabatic HAT to recent quantum computations.

The remarkable simplification of the analysis of the tunneling process using isotopically-labeled substrates, when combined with various biophysical probes, can provide great insight into catalysis, as illustrated above for the I553X series in SLO. While the majority of other single site mutations in enzymes behave like I553X, showing a combination of largely unchanged KIE values to WT with elevated ΔEa values, a new double mutant of SLO (DM), constructed from simultaneous mutation of the highly conserved leucine active site residues, L546A and L754A, has emerged as an outlier.7,10 This variant is characterized by a significantly reduced rate (kcat down 104-fold), an enormous KIE of 661 ± 27 across five temperatures, and a ΔEa that is essentially zero (ΔEa = 0.3 ± 0.7 kcal/mol).7,10 With these extreme kinetic parameters, the DM variant provides a benchmark to examine the limits of the nonadiabatic tunneling construct. Impressively, statistical analysis of the composite data for this enzyme has shown that the combination of an elongated DAD at the tunneling-ready state and a modified DAD sampling mode that prevents restoration of tunneling efficient configurations, is sufficient to explain all of the experimental parameters.10

Two recent quantum calculations39,40 extend the nonadiabatic analytical theory developed for SLO in new directions by introducing anharmonicity into the potential surface defining the DAD sampling mode. Both studies deal with theoretical aspects of rate, isotope effects and their temperature dependencies that rely on non-adiabatic models, differing in their primary focus on gas phase DFT39 or QM/MM40 computations. The former has a primary goal of assessing the physical parameters relevant to the observed PCET, whereas the latter is targeted toward an atomistic picture of the protein’s role in catalysis. Despite the different strategies, an important agreement between the works is the elongation of the ground state equilibrium DAD from 3.2–3.3 Å in WT to 3.6–3.8 Å in DM, values that approach ENDOR-determined values of 3.1 Å (WT) and 3.9 Å (DM).14

The QM study by Champion has investigated the impact on reactivity of an increasingly negative charge on the iron-bound hydroxide ion that can be modulated by hydrogen bonding to the C-terminal ligand to iron (I839). This analysis is simplified by use of an active site structure represented by iron, the ligands to the iron (without backbone), and methane as a model for the reactive carbon of the substrate LA. The computations begin with either “conformer a” for WT or “conformer b” for DM, using ENDOR-derived distances as a guide. These ground state configurations are the starting point to reach the tunneling ready states, via the application of a relatively large (pN) force that is “extrinsic” to the QM region and primarily arises from an electric field. As discussed extensively in other publications, the processggg of reaching the reduced distances within the TRS involves a series of stochastic conformational fluctuations that can originate far from the active site. The detailed analysis by Champion and co-workers is fundamentally similar to a model first put forth by Kuznetsov and Ullstrup,20 further developed in our Berkeley laboratory,2 and subsequently elaborated in numerous publications.5,10,15–18,36

A caveat to the above described work is that the quantum treatment of substrate was limited to methane, an issue addressed by Hammes-Schiffer and co-workers, who have extended the DFT computation to a 2,5-heptadiene moiety and to linoleic acid itself in their QM/MM analysis. One surprising conclusion is that the electrostatic force field needed to reach the Rdom (Figure 3D) is relatively local rather than global and depends largely on the double bonds flanking the reactive carbon of substrate. This conclusion arises in both their simplified DFT treatment of the reaction coordinate as well as the full QM/MM analysis, raising the question of what aspects of the reaction trajectory have been influenced by the full protein structure. In fact, the MM portion of the protein is shown to uncover an altered substrate binding mode and a reaction driving force in the double mutant enzyme; however, the QM/MM approach has thus far been unable to capture an experimentally determined long-range dynamical network that has recently been shown to communicate thermal activation between the protein-water interface and the active site of WT SLO (cf. Ref 23).

From the perspective of this Account, it is also notable that the new QM/MM study of SLO is unable to readily reproduce the experimental trends in ΔEa: while this analysis of SLO is able to match computed to experimental KIEs at select temperatures, via the introduction of small corrections to the computed potentials of mean force, it remains unable to reproduce well the weakly temperature dependent KIEs for WT and the DM variant. The extensively characterized system, for which analytical models allow a precise fit of both KIE and ΔEa, thus may provide valuable perspective into the long recognized and persistent difficulty in fitting temperature independent KIEs by way of QM/MM computations.41 In the recent analysis by Li et al.,40 the productive E-S complex in SLO is the starting point for movement along an all atom-generated potential energy surface. This is accomplished using a non-adiabatic approximation, in which all protein motions have been combined together into a collective work term that is both anharmonic in nature and expected to be comprised of a hierarchy of modes that spans global and local motions; the latter is the primary determinant of the ΔEa term, Figure 3D. The net result is a single, aggregate protein sampling mode that will have a softer potential than anticipated for an isolated, local DAD sampling process. As shown repeatedly using non-adiabatic analytical approaches (cf. Table 4), reduced force constants for DAD sampling produce more temperature-dependent KIEs. Thus, as it now stands, the QM/MM analysis is unable to easily rationalize or reasonably model the very small ΔEa values routinely observed for native enzymes. It will be of great interest to see if continued refinement of all atom computations can solve this problem, perhaps via methodology that permits resolution of the hierarchy of protein motions that is expected to be transiently populated as the potential energy surface progresses from ground state to tunneling-ready state.

Conclusions.

Kinetic isotope effects have played an important role in our analysis of hydrogen transfer processes. In the context of enzyme reactions, this review highlights the unique information that can be gleaned from an interrogation of the temperature dependence of the KIE. Once appropriate controls have been set in place, to rule out complications such as multiple rate limiting steps and reaction uncoupling, the magnitude of ΔEa provides key information regarding the catalytic role of reduced barrier width and its dependence on environmentally-controlled active site alignment between the hydrogen donor and acceptor atoms.