|Home | About | Journals | Submit | Contact Us | Français|
IR and Raman frequency shifts have been reported for numerous probes of enzyme transition states, leading to diverse interpretations. In the case of the model enzyme ketosteroid isomerase (KSI), we have argued that IR spectral shifts for a carbonyl probe at the active site can provide a connection between the active site electric field and the activation free energy (Fried et al. Science, 2014, 346, 1510–1514). Here we generalize this approach to a much broader set of carbonyl probes (e.g. oxoesters, thioesters, and amides), first establishing the sensitivity of each probe to an electric field using vibrational Stark spectroscopy, vibrational solvatochromism, and MD simulations, and then applying these results to re-interpret data already in the literature for enzymes such as 4-chlorobenzoyl-CoA dehalogenase and serine proteases. These results demonstrate that the vibrational Stark effect provides a general framework for estimating the electrostatic contribution to the catalytic rate and may provide a metric for the design or modification of enzymes. Opportunities and limitations of the approach are also described.
The physical origins of the remarkable catalytic properties of enzymes have been the subject of intense interest in many labs. We have recently provided a framework for extracting information on the electrostatic contribution to the catalytic rate of the model enzyme ketosteroid isomerase (KSI) by using the vibrational Stark effect (VSE).1 While many labs using well-chosen inhibitors as probes have obtained a large body of high quality infrared and Raman data, the absence of a physically based interpretation of the observed spectral shifts and their relationship to the catalytic rate has led to considerable confusion. In this paper we extend the application of the VSE approach to the analysis of vibrational data in the literature for several other enzymes suggesting that this approach is generalizable, and also noting its limits.
The vibrational Stark effect is a relatively new method to probe electric fields in condensed matter. Vibrational Stark spectroscopy (VSS) in a defined external electric field, , provides a calibration step quantifying the sensitivity of a vibrational transition to an electric field, known as the Stark tuning rate or difference dipole for the probe oscillator, , expressed in units of cm−1/(MV/cm). The VSE is then the more general application of such a calibrated vibrational probe to obtain information on the absolute electric field (MV/cm):
where is the observed vibrational frequency shift (in cm−1) from a reference state calibrated to zero electric field and is the electric field experienced by the vibrational probe due to a particular environment or matrix. For localized bond modes, the angle between and the transition dipole moment is expected to be zero as has been confirmed in several cases, and since the transition dipole for high frequency vibrational modes is typically parallel to the oscillator bond axis, the direction of in a complex system like a protein is known by x-ray crystallography.2 As such, these vibrational modes serve as calibrated and directional probes of the projection of the electric field exerted by the environment along the bond axis. Furthermore, vibrational solvatochromism can be used to systematically vary the field experienced by the probe in fluid solution, and the resulting spectral shifts can be related to the absolute electric field through MD simulations.3–5 The separately measured Stark tuning rate (by VSS) provides a bridge between these two approaches. Additional computational methods have also been developed to determine the environmental effects on vibrational frequencies.6–10 A key assumption is that is unaffected by , that is, spectral shifts are interpreted entirely as being due to a classical electrostatic interaction, not a change in the probe vibration's bond order. We return to this assumption later.
The nitrile (CN) vibration has been widely used due to its absorption in an uncluttered region of the IR spectrum.11–15 However, the nitrile probe's electrostatic response is complicated by other effects in hydrogen bonding (H-bonding) solvents.16–19 The carbonyl (C=O) vibration, which is generally more intense than nitriles, has been shown to vary linearly with electrostatic field in both H-bonding and non-H-bonding environments.3, 5, 7–8, 20 The main limitation of the carbonyl probe is that its frequency overlaps with the densely populated amide I region in many biological systems. Fortunately, this can often be overcome by careful selection of a reference sample that is nearly identical to the sample of interest as well as isotopic labeling, or by using Raman spectroscopy. The earliest work used CO bound to the heme iron of myoglobin mutants,21–22 and this has recently been extended by us to acetophenone and the carbonyl group on 19-nortestosterone, an inhibitor of KSI, to provide an estimate for the electric field in the active site and its contribution to lowering the activation free energy.1, 23
There already exists a large body of work describing perturbed carbonyl frequency shifts in the active site of enzymes, several of which document a relationship between the catalytic rate and C=O frequency; some examples are listed in Table 1. These findings have been the subject of diverse interpretations including but not limited to bond polarization, distortion, and ground state destabilization.24–31 We believe that the VSE can provide a more general, consistent, and quantitative framework with which to interpret these findings and explore the role of electric fields in catalysis.
In order to expand the applications of the carbonyl vibrational probe to diverse problems, we report experimental Stark tunings rates for various oxoesters, thioesters, and amides of biologically relevant model compounds. Additionally, we measure vibrational solvatochromism for these probes and use MD simulations to calculate absolute electric fields, finding that all exhibit a linear field-frequency relationship. These results, summarized in Part A of Results & Discussion, should be useful for further studies using carbonyl probes, following in the same spirit as an earlier paper on VSS of diverse oscillator types.50 We follow this in Part B by reinterpreting IR and Raman data obtained by others for several enzymes in terms of the vibrational Stark effect to quantify the contributions of electric fields to catalysis.
The carbonyl containing compounds were acquired from the following sources: N,N-dimethylacetamide (DMA, 99.9%, Sigma Aldrich), N-methylacetamide (NMA, 99% Sigma Aldrich), S-butyl thiobenzoate (>95%, TCI America), ethyl thioacetate (98%, Sigma Aldrich), ethyl acetate (99.8%, Sigma Aldrich), methyl acetate (>98%, Sigma Aldrich), methyl propionate (>98%, Sigma Aldrich), methyl benzoate (99%, Sigma Aldrich), methyl trans-cinnamate (99%, Sigma Aldrich), ethyl 4-aminocinnamate (97%, Sigma Aldrich), ampicillin sodium salt (Sigma Aldrich), β-nicotinamide adenine dinucleotide 2'-phosphate reduced tetrasodium salt (NADPH, 95%, Sigma Aldrich). Solvents were purchased as follows: 2-methyltetrahydrofuran (2MeTHF, >99%, Sigma Aldrich), toluene (99.8%, Sigma Aldrich), ethanol (99.5%, Acros Organics), methanol (99.9%, Fisher Scientific), D2O (99.8% atom D, Cambridge Isotope), glycer(ol-d3) (99% atom D, Sigma Aldrich), ethylene glycol-(OD)2 (98% atom D, Sigma Aldrich). All reagents were used without further purification.
Carbonyl-containing solutes were dissolved in glass forming solvents as specified for each experiment with typical concentrations of 50–200 mM. Solutions were then loaded in a sample-cell consisting of two offset CaF2 optical windows (1 mm thickness, 12.7 mm diameter, FOCtek Photonics) coated with a 4.5 nm layer of nickel metal and separated by approximately 26 μm Teflon spacers. The filled sample cell is then immediately immersed into liquid N2 in a custom-built cryostat,51 and Stark spectra were recorded on a Bruker Vertex 70 with fields of 0.5–1.8 MV/cm (applied voltages of ca. 1.5–4.0 kV) from a Trek 10/10 high-voltage power amplifier, 1 cm−1 resolution, 64 scans apiece of field-on/off transmission spectra, and repeated with increasing fields to ensure the intensity of the Stark spectra scales linearly with the square of the field strength as expected for an isotropic sample.52 The linear Stark tuning rates were determined as previously described from the second derivative contribution to the Stark spectrum of numerical fits of the 0th, 1st, and 2nd derivative contributions of the best-fit Voigt profile of the experimental absorbance spectrum.52 This analysis assumes an isotropic, immobilized sample and that the angle (ζ) between and the transition dipole is zero; the experimentally set angle (χ) between the incident light polarization and the external electric field direction was 90° as previously described.52 In cases where overlapping peaks were observed (e.g where both H-bonding and non-H-bonding species occur), the derivative contributions were determined from the full absorbance spectrum in the spectral region of interest unless otherwise noted. Note that while the applied electric field () is known accurately from the applied voltage and separation of the electrode, the actual field experienced by the probe () is modified by the local field correction factor (f), .2, 52 While the exact value of the local field factor is not known, it is expected to have a value between 1 and 2.1, 4, 23, 53 As such, the Stark tuning rates are reported in terms of and this is discussed in detail with the results.
FTIR spectra were recorded on a Bruker Vertex 70 spectrometer with a liquid nitrogen-cooled mercury cadmium telluride (MCT) detector using methods described previously.3 Briefly, a demountable liquid cell was prepared from two CaF2 optical windows (0.75 in diameter, 0.25 in thickness, Red Optronics), separated by two semicircular Mylar spacers (75 and 100 μm thickness), to which 20–30 μL of approximately 10 mM solute solution was added.
Transmission spectra were acquired by averaging 128–256 scans after 5–10 mins of purging with a nitrogen flow to remove atmospheric CO2 and water vapor. Spectra were recorded from 2000–1400 cm−1 with 1 cm−1 resolution and aperture settings from 2-6 mm in order to maximize signal intensity. The absorption spectra were calculated by taking the negative logarithm of the difference between solute and neat solvent transmission spectra. All solvatochromism measurements were repeated in triplicate, and all frequencies determined using the OPUS software's peak picker (Bruker) as well as Voigt fitting with the Levenberg-Marquardt algorithm.
All substrate compounds were parameterized first by modeling the molecule in GaussView and then geometry optimized by DFT at the B3LYP/6-311++G(d,p) level in Gaussian09.54 The resulting structure was input into the Antechamber program of AmberTools14 and parameterized using the GAFF forcefield55–56 and the AM1-BCC method57 to assign atomic charge parameters. Parameters for organic solvents were taken from Caleman et al.58 and water was modeled using the TIP3P parameters59. All simulations were setup and carried out in GROMACS version 18.104.22.168 A single solute molecule was solvated in a cube expanding 2.0 nm around the solute with parameterized solvents as previously described.3, 58 Note that pre-polarized, fixed charge force fields were used for these simulations. There are subtle differences when a polarizable force field is used, in particular the field in hexane is not zero (see Ref. 4 for a detailed discussion; for the purpose of this work these much more computationally expensive simulations were not needed).4 Simulations were performed with periodic boundary conditions, long-range electrostatics were treated by Particle-Mesh-Ewald (PME) with 1 nm cut-off, temperature coupling with V-rescale thermostat, and Berendsen pressure coupling. The system was first energy minimized by steepest descent, then equilibrated in the NPT-ensemble at 300 K over 200 ps, and finally MD simulations were carried out for 2.0 ns with snapshots of the forces and positions recorded for a total of 10,000 snapshots per trajectory.
Electric fields were calculated as previously described.3 Briefly, using each recorded snapshot from the MD simulations, the forces and positions were calculated on the carbon and oxygen atoms of the carbonyl moiety. Then a charge-neutralized trajectory was generated keeping all positions at each snapshot constant for both solute and solvent, but setting the charge of all solvent atoms to zero. Subtraction of the charge neutralized forces at each snapshot yields the electrostatic forces due to intermolecular solute-solvent interactions. With the forces and positions of each snapshot the electric field experienced by the carbon and oxygen atoms of the carbonyl were determined by dividing by the atom's partial charge, and the total electric field experienced by the carbonyl was calculated from the projection of the carbonyl bond vector on the atom's electric field and then averaging the two field projections as previously described.3 Repeating this for each snapshot of the MD trajectory provides the ensemble-averaged electric field and its standard deviation.3
The vibrational absorption and Stark spectra of several representative oxoesters, thioesters, and amides are shown in Figure 1 along with a larger collection in Figures S1 & S2, and the results are summarized in Table 2. Due to the unique properties of oxoesters and thioesters, the two are easily distinguishable by IR and the Stark spectra of comparably conjugated C=O compounds are shown in Figure 1A–D. By measuring the Stark tuning rates of these different probes simultaneously in the same sample, we minimize any systematic differences in the frozen glass, solvent, applied field (e.g. voltage and electrode separation), which could affect the local field factor (f) as it is expected to be about the same for a series of compounds in the same solvent.2, 52 At least for cases where there is no band overlap, the Stark spectra are entirely dominated by the second derivative contribution from which is obtained. The local field correction factor, treated here as a scalar, distinguishes the field felt by the probe () from the known applied electric field, , and is a common feature of any spectroscopy that involves the application of electric fields.23, 52, 61 While expected to have a value between 1–2,23, 62–63 this is a source of uncertainty and our best estimate for its value is discussed further below in the context of comparisons with the solvatochromism data. However, as seen in Table 2, there is negligible variation in the measured Stark tuning rate for a solute across different solvents, suggesting that the local field factor is not a solvent-dependent property. This enables quantitative comparison between different systems for a single probe using the VSE.
In earlier work, the Stark tuning rates of ketones have been observed to be 0.7–1.0 cm−1/(MV/cm), which further increases upon additional conjugation towards values of 1.0–1.8 cm−1/(MV/cm) or greater.3, 22, 50 As summarized in Table 2, we observe that has values between 1.05–1.48 cm−1/(MV/cm) for various oxoesters, 1.39–1.47 cm−1/(MV/cm) for thioesters, and 0.52–2.2 cm−1/(MV/cm) for the amides studied herein. The results with model oxoesters are in close agreement with that observed by Takashima et al. for poly(methyl methacrylate) films of .64 Pazos et al. predicted the Stark tuning rate for the methyl ester analogs of aspartic and glutamic acid from vibrational solvatochromism and MD simulations, but these were over-estimates (1.3 cm−1/(MV/cm)) of those reported in Table 2 for methyl acetate and methyl propionate, likely due to a correction for the local field factor that was implicit in the calculations and carried over from earlier studies with acetophenone.3, 5
The Stark tuning rates () of N-methylacetamide (NMA) and N,N-dimethylacetamide are both ca. 1.3 cm−1/(MV/cm). Assuming a local field factor of 2, this is in good agreement with the calculated value or 0.78 cm−1/(MV/cm) using either a distributed dipole or distributed charge coarse-grained model.65–66 Recent work by Oh et al.67 estimates the Stark tuning rate () of the amide I mode for a 13C=18O labeled peptide to be 1.4 cm−1/(MV/cm), which differs by the local field factor. Computational studies68 have shown that the vibrational frequency shifts of a localized amide I mode are dominated by the electrostatic interactions with the C=O, though there are other normal modes contributing to the amide I. For simplicity the value of the Stark tuning rate was calculated from the data assuming that the angle between the transition moment and difference dipole is 0° (note that the transition dipole has been reported to be ca.10–20° off the C=O bond axis68–70 and this will affect the interpretation of vibrational Stark effects for the amide I band in proteins). In support of the notion that the frequency response of this mode to an electric field is primarily due to the C=O, the observed Stark tuning rate of 1.3 cm−1/(MV/cm) for both the NMA monomer in 2MeTHF and NMA-d1 in D2O indicates a minimal effect of N-deuteration, which was expected to alter the normal mode composition of the amide I frequency (Figure S2C–F).68, 71 The Stark spectrum of NMA in 2MeTHF (Fig S2D) indicates the presence of several C=O bands which we tentatively assign to the monomer at 1671.2 cm−1 and an oligomer at 1637.1 cm−1, which is generally observed for NMA in low polarity organic solvents.72 The peak at 1655 cm−1 is observed in both NMA and DMA, in both cases without an observable Stark feature and as such was not considered in the fitting. Furthermore, the Stark feature corresponding to the oligomeric peak is broad and of much lower intensity than the monomer (Figure S2D). This feature may arise from coupling between the NMA oligomer C=O modes,73–76 which could lead to a breakdown in the assumptions of our Stark analysis and therefore was not characterized further. Beyond simple amides, the Stark tuning rate for the amide and β-lactam of ampicillin are significantly greater than those observed for NMA or DMA, suggesting that intramolecular H-bonding from the neighboring amine and ring strain, respectively, can have large influences on the magnitude of the Stark tuning rate.
When comparing the Stark tuning rates of the simple oxoesters, thioesters, and amides there is a general trend that increases such that O < N < S. This indicates that the heteroatom (X) in the R-C(O)X-R moiety is the main contributor distinguishing the Stark tuning rates, with larger effects from less electronegative heteroatoms. These results suggest that there are only small differences in the linear Stark tuning rate with increasing conjugation (e.g. methyl acetate, methyl benzoate, and methyl cinnamate) and that changing the substitution (e.g. ethyl 4-aminocinnamate), ring strain (β-lactam), or intramolecular hydrogen bonding (e.g. amide mode of ampicillin) can significantly increase . Furthermore, there is little to no variation in the Stark tuning rate of a given vibrational probe between different hydrogen-bonding (H-bonding) and non-hydrogen-bonding glass-forming solvents (Table 2, Figures 1, S1, and S2). As observed with acetophenone,3 this suggests that each carbonyl's intrinsic sensitivity to an electric field is the same in both H-bonding and non-H-bonding environments. In other words, the constant value of observed across multiple solvents suggests that both and f are independent of environment.
In order to further develop the carbonyl group as an electric field probe, we carried out vibrational solvatochromism studies on ethyl acetate, methyl benzoate, ethyl thioacetate, butyl thiobenzoate, and DMA as model compounds.
Similar to earlier work from our lab on acetophenone,3 the C=O frequency in all solutes was observed to red-shift in solvents of increasing polarity from hexanes to aqueous solution as shown in Figure 2. In general, the observed solvatochromic shift in peak position between hexanes and water () reflects the different magnitudes of their respective Stark tuning rates. This is evidenced by comparison of butyl thiobenzoate (), methyl benzoate (), and acetophenone (),3 which are all similarly conjugated but have decreasing solvatochromic shifts (34.0, 28.6, and 27.0 cm−1, respectively) across the solvent series. However, the opposite trend is observed with respect to the Stark tuning rate for ethyl acetate and thioacetate. This may be due to the broad absorbance of the C=O peak of ethyl acetate in water, which has a prominent shoulder and complicates the analysis. This observation is consistent with studies on similar alkyl acetates where the solvents exhibit altered conformations of the terminal methyl group, resulting in differences in the E/Z-rotamer populations,77–79 or the presence of different H-bonding configurations as observed by MD simulations and 2D IR.80–82 Consistent with multiple H-bonding configurations, post-processing of MD trajectories by Pazos et al. showed that the fields and frequencies of these 1- and 2-H-bonding configurations fall on the same field-frequency calibration line as other non-H-bonding solvents, consistent with a linear Stark effect.5 For simplicity, all C=O frequencies were determined using the peak picking method and are in close agreement with those from curve-fitting except in cases where multiple peaks were observed (Table S1). Surprisingly, the solvatochromic response of the amide C=O of DMA has the largest frequency shifts of those studied (Figure 2C) of 66.9 cm−1. This result is discussed further in the context of the MD simulations presented below. Consistent with previous studies,3 we observe that there is a strong correlation between the C=O line widths, which are related to the inhomogeneous distribution of electric fields in solution, and peak positions, which are related to the average electric field experienced by the bond (Table S1).
We performed molecular dynamics (MD) simulations to calculate the average solvent reaction field based on previous work by Fried et al.3 When directly correlated to the solvatochromism measurements this provides a calibration curve that relates the C=O frequency to the absolute electric field experienced by the carbonyl in the solvent series. Furthermore, the slope of this correlation provides a further measure of the Stark tuning rate () for direct comparison to that measured by VSS () as discussed below.
Utilizing the same solvent series as the solvatochromism measurements, we have calculated the average absolute electric field experienced by the carbonyl of ethyl acetate, methyl benzoate, ethyl thioacetate, butyl thiobenzoate, and DMA. Consistent with previous studies, the average electric field (Figure 3, horizontal axis) and standard deviation (Table S1) increase with increasing solvent polarity. Additionally, for two molecules in the same solvent of similar size and conjugation, the average electric field experienced by the C=O is comparable as observed between ethyl acetate and ethyl thioacetate as well as methyl benzoate and butyl thiobenzoate. This may suggest that the solute geometry and steric considerations around the C=O are the primary determinants for the magnitude of the solvent electric field since there are only small variations between oxoesters and thioesters of similar shape. However, for DMA, the average solvent electric field is consistently larger than that of similarly sized ethyl acetate and thioacetate, suggesting that another factor may give rise to this difference between amides and other carbonyl probes. Furthermore, these results are consistent with DMA having the largest solvatochromic frequency shifts among the probes studied, indicating that the MD simulations are able to recapitulate the trends observed by solvatochromism. While multiple peaks are occasionally observed with certain solutes and solvents, in the case of DMA the observed shoulder at ca. 1655 cm−1 may be attributed to either Fermi resonances83 (previously observed with 19-nortestosterone,1 where shoulders appear at the same frequency across different solvents) or the noncoincidence effect,84–85 though we observe no appreciable concentration dependence between 1–100mM for the shoulder. We do not observe two separate peaks each with their own solvatochromic shifts and the simulations exhibit single populations based on the electric field distributions. No effort was made to further deconvolve these peaks, and the major peak's center frequency was used for all field-frequency correlations, which may introduce small errors into the fits.
The linear correlation between the C=O frequency and the calculated electric field (Figure 3) is consistent with the solvent shifts being due to a linear VSE. Despite this observed linear correlation for each molecule, the slopes of the best-fit lines are not equivalent to the independently measured Stark tuning rate, and the difference is assumed to reflect subtle differences in the magnitude of the local field factor (f). This result highlights an on-going source of uncertainty in the analysis of solvatochromism data and its connection to VSS results (Table 1).23 The VSS data is taken with frozen solvent glasses (to prevent solute and solvent reorientation in the applied field and dielectric breakdown), while the solvatochromism data and MD simulations apply to fluid solutions. There is no applied field in the case of solvatochromism, thus the local field factor does not enter, whereas the VSS data is obtained in an applied electric field, hence the data is given as in Table 1. The slope of the field-frequency correlation should equal the Stark tuning rate, , thus the simplest interpretation of the variations in observed slopes (Figure 3) is that the local field correction factor is somewhat different for each compound. The observed range, f = 1.6–2.6 based on the differences in slopes of the field-frequency calibration curves () relative to the experimentally measured linear Stark tuning rates (; Table S2), suggests that is approximately half that measured by vibrational Stark spectroscopy, that is f ≈ 2, consistent with earlier work and recent calculations.1, 3, 53 On the other hand, there may be subtle but systematic discrepancies in the simulations of solute-solvent pairs that are not accounted for at the current level of approximation. This uncertainty makes quantitative comparison between probes difficult, but a single probe can be used consistently to study a system of interest using the VSE. More sophisticated treatments of the electric fields in different solvents may be possible;9 however, these treatments involve large numbers of parameters and likely would not affect the main conclusions on the magnitude of the electric field as the carbonyl response to different environments has been shown to be dominated by electrostatic interactions.9, 86–90
In order to further utilize carbonyl vibrational probes and interpret frequency shifts in terms of electric fields in condensed matter, we have measured the Stark tuning rate of oxoesters, thioesters, and amides through vibrational Stark spectroscopy and vibrational solvatochromism correlated with MD simulations. These findings further indicate that the frequency response of the C=O stretch is primarily electrostatic in nature as indicated by the linear field-frequency calibration and the consistency of the measured Stark tuning rates in both H-bonding and non-H-bonding frozen glasses. Therefore, the C=O vibrational probe is well suited to interrogate diverse molecular interactions such as catalysis and binding using the vibrational Stark effect as exemplified in the following section, where we have selected representative examples from Table 1.
The framework used to apply VSE probes to parse the electrostatic contribution to catalysis is shown in a simplified reaction coordinate diagram in Figure 4 for some of the major classes of carbonyl chemical reactions found in enzymes. Within this framework, we are specifically focusing on the degree to which the transition state is stabilized by interaction with the total field, , from the surrounding protein. This includes very local interactions such as specific hydrogen bonds, but also more global contributions that are less easily visualized (these effects need not work in concert). For simplicity, we show the bond(s) involved in the substrate as having a small dipole moment, (often not the case), with a substantial increase in the dipole moment of the transition state (e.g. , a common situation) associated with the reaction, and . The effect of the protein field on the free energy of activation is then given by:
Note that the effect of is illustrated in Figure 4 as stabilizing the transition state, but this depends on the relative orientation of and . By placing a calibrated vibrational probe that closely mimics the presumed transition state into the enzyme active site, as close as possible to the bond experiencing , the observed frequency shifts detected by IR or Raman spectroscopy for a series of variants whose free energies of activation has been measured separately can provide information on the relationship, if any, between the free energy of activation and the protein field. As shown in Part A, the frequency response of carbonyl vibrational probes can be explained entirely using the linear VSE, providing a means for quantifying the catalytically-relevant electric fields in reactions in which the rate-determining step involves charge rearrangements occurring at the carbonyl, assuming that the probe is oriented such that it experiences some projection of the relevant active site electric field. While chemical positioning and distal binding interactions with the substrate may be relevant for catalysis, these effects are not part of electric field catalysis and cannot be explored using the VSE method.1, 91 This strategy has been described earlier for KSI,1, 49 and in the following we use this framework and the results from Part A to re-interpret IR and Raman data already in the literature for several enzymes. Note that while a relationship between the free energy and protein field may have significance, a large electric field, in and of itself, is not necessarily indicative of a good catalyst. As extensively discussed elsewhere,91–96 water (where the “mechanism-filtered” reference reaction takes place) is a poor catalyst, though it has a large electric field, for reactions which involve large changes in the magnitude and/or reorientation of the dipoles over the reaction coordinate.
4-chlorobenzoyl-CoA dehalogenase (Dehalogenase) is a member of the Crotonase superfamily, which possesses common structural and catalytic features to stabilize enolate intermediates in the oxyanion hole.97 Dehalogenase catalyzes the hydrolytic dehalogenation of 4-chlorobenzoyl-CoA to 4-hydroxybenzoyl-CoA through a step-wise reaction mechanism that proceeds from the initial addition of Asp145 to the C4 position of the benzoyl ring, forming the Meisenheimer intermediate, followed by removal of the chloride ion and subsequent hydrolysis of the arylated enzyme (Figure 5A).45 This enzyme is proposed to be a recent evolutionary response to increased synthetic chlorinated organic pollutants in the environment as part of the 4-chlorobenzoate degradation pathway in Pseudomonas species CBS3.98
Using Raman spectroscopy, Dong et al.45 observed that there is a linear relationship between the single-turnover rate (kobs), where the formation of the Meisenheimer intermediate is assumed to be rate-limiting, and the thioester C=O vibrational frequency using both the natural substrate and inhibitors for a series of mutants.45 The rate changes by 2.4×106 and the frequency varies by 61 cm−1 across the mutant series, clearly suggesting a role for electrostatics, but there was no way to go beyond an empirical description correlating frequency shifts with hydrogen-bonding strength.45–46
Assuming that the Stark tuning rate of the Dehalogenase inhibitor, 4-methylbenzoyl-CoA (Figure 5B), has a similar Stark tuning rate as that of butyl thiobenzoate, (Table 1), we can interpret these frequency shifts in terms of electric fields (Figure 5C). This analysis relies on the following: the field-frequency relationship of butyl thiobenzoate (f ≈ 2) is a reasonable estimate for the inhibitor, the calculated electric field of ethyl 4-methylthiobenzoate in water is −58.5 MV/cm, and the known frequency of the C=O stretch of 4-methylbenzoyl-CoA in water is 1651 cm−1.46 Plotted in Figure 5C are the free energies of activation obtained by Dong et al. on the y-axis and their observed Raman frequencies,45 converted into electric fields through the combination of VSS, solvatochromism, and MD simulations on the x-axis. These data fit very well to a line, where . The slope of the least squares regression line, 2.06 D (see units on the upper x-axis), is the reaction difference dipole, , which is a measure of the change in charge distribution between the ground and transition states. This value is about twice as large as that observed for KSI,1 as may be expected given the larger charge rearrangement between the ground state and Meisenheimer intermediate. The intercept at 27.5 kcal mol−1 corresponds to the hypothetical free energy barrier if Dehalogenase did not exert any stabilizing electric field, i.e. , suggesting that 11.9 kcal mol−1 of the enzyme's catalytic power comes from a large stabilizing electric field, a rate enhancement of 108.7-fold relative to the reaction in the absence of an electric field. As noted previously,1 rescaling of the calculated electric fields (e.g. due to the local field factor) will not affect the primary finding that the active site electric field is directly correlated with lowering the activation barrier, though it will affect the magnitude of . To our knowledge the enthalpy of activation has not been measured for the wild-type enzyme, though the intercept is in close agreement with the calculated enthalpic contribution to catalysis of ΔH‡ = 27.4 kcal mol−1 for the nucleophilic aromatic substitution reaction in aqueous solution.99 Furthermore, based on the predicted uncatalyzed rate in solution of 2.3×10−15 sec−1,99 it is estimated that approximately 55% of the rate enhancement comes from the presence of a large electric field at the C=O bond. The remaining significant portion of the enzyme's rate enhancement can be attributed to other factors, largely entropic, based on the predicted rate in solution. There is likely a significant contribution from chemical positioning (entropic) of Asp145 and the active site water necessary for formation of the Meisenheimer intermediate and subsequent hydrolysis.
Although Dehalogenase has only a modest catalytic rate and a smaller active site electric field compared with KSI, a larger allows for a greater overall rate-enhancement due to electric field catalysis (as seen in Fig. 4, the effect scales linearly with and ). In other words, proteins that catalyze reactions with larger reaction difference dipoles can achieve the same catalytic rate-enhancement with smaller active site electric fields, a concept we have termed the “catalyzability” of the reaction91; this likely has important implications in protein evolution and catalyst design and dehalogenase would be a good target.
Serine proteases have been extensively studied, often highlighting the importance of a pre-organized active site on catalytic function.100–102 The reaction mechanism is illustrated in Figure 6A and showcases the catalytic triad which participates in peptide bond hydrolysis leading to the acyl-intermediate. The acyl-intermediate is then attacked by an activated water molecule during the rate-limiting step to form the anionic tetrahedral transition state, which proceeds to form the hydrolyzed product. Despite much study, there has been continuing debate on the molecular origins of the catalytic power in serine proteases, either from low-barrier hydrogen bonds (LBHB's) in the catalytic triad (e.g. Ser-His-Asp as observed in chymotrypsin) and/or transition state stabilization from the oxyanion hole.100–101, 103
In an effort to elucidate the physical origins of catalysis in serine proteases, Tonge et al.29 utilized resonance Raman spectroscopy to study the frequency shifts of acyl-intermediates of Chymotrypsin, Subtilisin BPN', and Subtilisin Carlsberg. Using 5-methylthienyl acrylate (5-MeTA), these authors observed that there was a linear relationship between the C=O frequency of the acyl-intermediate (Figure 6B) and the rate-limiting deacylation kinetics (k3).29 In order to rationalize this result, vibrational solvatochromism was employed along with empirical solvent polarity correlations to relate these frequency shifts to changes in hydrogen-bonding strength and C=O bond length in the active site. Note that it is common to relate red shifts with changes in bond length and double bond character, i.e. to a change in the force constant for the oscillator (often called “bond polarization”). However, we have shown in earlier work1, 3, 23 and for the compounds discussed here that these frequency shifts can be entirely accounted for by changes in the electric field exerted by different solvents as they interact with the solute oscillator; this treatment explicitly does not alter the physical properties of the solutes, and a recent theoretical analysis by List et al.7 is consistent with this assumption. Note also that solvent polarity, while convenient and widely used as a description of bulk solvent properties, does not provide a molecular picture and does not even have useful units, in contrast to the electric field.
Solvatochromic measurements with the methyl ester of 5-MeTA showed that the C=O frequency shifts by 36 cm−1 from hexanes to water.37 A similar solvatochromic shift of ca. 38 cm−1 was observed with methyl cinnamate in the same solvent series (Figure S3). Additionally, the calculated electric fields of methyl cinnamate and 5-MeTA in water are 60.3±1.9 and 58.1±1.3 MV/cm respectively, in agreement with the observation in Part A that two molecules of similar size and conjugation will experience electric fields of the same magnitude. Based on these results it can be reasonably assumed that the Stark tuning rate of 5-MeTA is approximately that of methyl cinnamate ( and f =2), and we use this value to reinterpret these findings in terms of the VSE.
As with KSI and Dehalogenase, the Stark tuning rate, frequency-field calibration combined with solvatochromism, Raman and rate data from Tonge et al.37 allow for quantification of the electric fields experienced by the acyl-intermediate in the protein active site as shown in Figure 6C. The intercept of the least-squares regression line is 23.2 kcal mol−1 and represents the projection onto the probe transition dipole (approximately parallel to probe bond axis) in the inhibitor used to collect the Raman data. Because of the geometry change going to the transition state, this projection is a lower limit to the hypothetical free energy barrier if these proteases possessed no stabilizing electric field, a shortcoming that is discussed below. The slope, which gives an estimate for of 1.09 D, is comparable to that observed in KSI. We can also add data for similar measurements made with 5-MeTA in α-Chymotrypsin and Subtilisin Carlsberg which catalyze the same reaction via a shared mechanism. As seen in Figure 6C, dotted line, these data lie on essentially the same least-squares regression line as that of the Subtilisin BPN' series, suggesting that is an intrinsic property of the reaction rather than enzyme-dependent.
In contrast to the inhibitors used to probe KSI and Dehalogenase, whose ground-state and transition-state C-O geometries are expected to be very similar, this will not be true for serine proteases and many other enzymes that proceed through tetrahedral transition states. A change in the orientation between and means that the electric field experienced in the two states could be significantly different. Based on the model in Figure 4, an enzyme can achieve the greatest rate enhancement by maximizing the transition state and minimizing the ground-state electric fields, which will reflect the preorganization of the protein (Figure S4). Enzymes that can utilize these geometry changes to preferentially stabilize or anticipate the transition state, would be able to achieve larger rate enhancements by electric field catalysis.91 However, the experiment reports on the projection of the electric field on the ground state C=O bond, which is not necessarily the orientation with the maximum electric field. Therefore, the extent to which catalytic information can be parsed from the VSE measurements is limited by the availability of probes that bind in an orientation resembling the transition state. While there are single bond transition state analogs for proteases, these probes would have frequencies in a more cluttered region of the infrared and, by virtue of being low-frequency single bond modes, they may exhibit increased mode-mixing in the normal modes and non-linear Stark effects.
Based on this discussion, the analysis likely provides a lower limit for the actual magnitude of the electric field and its role in catalysis. Furthermore, the apparent as shown in Figure 6C is actually a measure of the change in the free energy barrier as a function of the change in electric field experienced by the C=O probe across a series of mutants () and may be expected to differ for a transition state analog. There is experimental evidence for this notion in the literature. Doran et al.44 observed a negative linear correlation between the C=O frequency of the acyl-intermediate and catalytic rate in several cysteine proteases, in contrast to all previously observed frequency-rate correlations. While this was previously rationalized on the basis of changes in polarization or resonance forms in the active site,44 in the context of VSE, we can interpret this result to suggest that the active site electric field has evolved to minimize the ground state electric field and increase the catalytic advantage from transition state stabilization expected from a pre-organized enzyme active site (for further discussion see Figure S4). It is worth noting that these observed fields are still stabilizing the ground state C=O, since , and cannot be taken as an example of ground state destabilization based on the VSE model and in agreement with Warshel et al.94 Although untested, this may suggest that the electric field varies to a greater extent in cysteine versus serine proteases over the angle displacement between the ground state (C=O) and transition state (C-O−) bond axes. Though we hypothesize that the fields experienced in the transition state would be significantly greater than those observed for the acyl-intermediate, the current analysis clearly indicates a significant role for electric fields in catalysis. These differences in the electric field in the transition versus ground state would reflect the preorganization of these proteases, suggesting a role for these electric fields to guide enzyme evolution and protein design.
KSI is the most extensively studied enzyme using vibrational Stark effects,1, 49 but many other examples can be described within the same framework. The generalizability of the VSE model provides a quantitative framework for discussing and interpreting vibrational frequency shifts in terms of electric fields. The characterization of oxoester, thioester, and amide model compounds expands upon the available carbonyl vibrational probes that can be used for exploring electric fields. Through the combined methods of vibrational Stark spectroscopy, vibrational solvatochromism, and MD simulations these carbonyl probes all exhibit a linear field-frequency relationship and constant Stark tuning rate across multiple environments, suggesting that the response to an electric field is primarily due to the linear Stark effect.
These findings enable a more comprehensive method for rationalizing non-covalent interactions in many enzymatic systems as illustrated in Part B, which illustrates the role of electric fields to facilitate catalysis. Measurement of VSE's in proteins and other systems provides a method for relating frequency shifts to electric fields, a general physical quantity, which may have significant applications in catalyst design.
The authors would like to thank Dr. Stephen Fried and Yufan Wu for helpful discussions on methods, results, and interpretations. This work was supported in part by grants from the NIH (Grant GM27738 and GM118044).
Supporting Information Available: Additional vibrational Stark spectrum of oxoesters and amides, IR solvatochromism of trans-methyl cinnamate, discussion of relation between the reaction coordinate geometry and the active site electric field, tabulated frequencies, FWHM, and electric fields from solvatochromism and MD simulations, and comparison of the Stark tuning rates from VSS and field-frequency correlations. This information is available free of charge via the Internet at http://pubs.acs.org.