Hydride bridge in [NiFe]-hydrogenase observed by nuclear resonance vibrational spectroscopy

The metabolism of many anaerobes relies on [NiFe]-hydrogenases, whose characterization when bound to substrates has proven non-trivial. Presented here is direct evidence for a hydride bridge in the active site of the 57Fe-labelled fully reduced Ni-R form of Desulfovibrio vulgaris Miyazaki F [NiFe]-hydrogenase. A unique ‘wagging' mode involving H− motion perpendicular to the Ni(μ-H)57Fe plane was studied using 57Fe-specific nuclear resonance vibrational spectroscopy and density functional theory (DFT) calculations. On Ni(μ-D)57Fe deuteride substitution, this wagging causes a characteristic perturbation of Fe–CO/CN bands. Spectra have been interpreted by comparison with Ni(μ-H/D)57Fe enzyme mimics [(dppe)Ni(μ-pdt)(μ-H/D)57Fe(CO)3]+ and DFT calculations, which collectively indicate a low-spin Ni(II)(μ-H)Fe(II) core for Ni-R, with H− binding Ni more tightly than Fe. The present methodology is also relevant to characterizing Fe–H moieties in other important natural and synthetic catalysts.

The Ni-R state represents a special challenge for spectroscopy in that it is EPR-silent, photolabile (and thus difficult to study with Raman spectroscopy) and expected to feature an active site M-H moiety (being notoriously difficult to observe using infrared methods) 33 . The present study instead employs a synchrotron radiation technique called nuclear resonance vibrational spectroscopy (NRVS), which involves X-ray excitation of a Mössbauer-active nuclide [34][35][36][37] . The raw NRVS data are commonly translated into partial vibrational density of states (PVDOS) spectra 38 , which show vibrational energy contributions specifically from the Mössbauer-active nuclei, such as 57 Fe. PVDOS can also be predicted using density functional theory (DFT) or empirical force field calculations, assisting in confident spectral assignments.
NRVS is uniquely suited to detailed investigations of 57 Felabelled enzyme active sites, avoiding interference from the thousands of protein modes present in a typical infrared or Raman spectrum. NRVS has enabled the observation of Fe-CN and Fe-CO bending and stretching modes for the active sites in [NiFe]-hydrogenases 39,40 , despite the presence of 11 (or more) Fe centres in clusters of the electron transport chain. This is because Fe-CN and Fe-CO modes are strongest in the region from 440 to 640 cm À 1 , while Fe-S cluster modes only make significant contributions below 440 cm À 1 (refs 41-43). Another recent application of NRVS to a trans-H/D- 57 Fe-(CO) compound shed light on coupling of Fe-H/D and Fe-CO bending modes 33 .
The present study discloses the first spectroscopic evidence for the bridging hydride in the Ni-R active site and the unprecedented NRVS observation of a Fe-H stretching mode in a synthetic Ni-H-Fe system. A combined experimental/theoretical analysis of both Ni-R and its synthetic mimics is presented here, an approach that we anticipate to be of broad utility for the characterization of (bio)inorganic Fe-hydride catalysts.

Results
NRVS of model complexes. The Ni(II)(m-thiolate) 2 (m-H)Fe(II) core proposed for Ni-R is reproduced by the diamagnetic H 2 -evolving catalyst [(dppe)Ni(m-pdt)(m-H)Fe(CO) 3 ] þ ([1H] þ , dppe ¼ 1,2-bis(diphenylphosphino)ethane ¼ 1,2-Ph 2 PCH 2 CH 2 P Ph 2 , pdt 2 À ¼ 1,3-propanedithiolate ¼ À SCH 2 CH 2 CH 2 S À ) 44,45 shown in Fig. 2a, which has recently been studied using resonance Raman spectroscopy 46 51 for other species, the 1,532 and 1,468 cm À 1 bands for [1 0 H] þ are the first Fe-H stretching modes detected using this technique, as well as being the highestfrequency bands observed using NRVS to date. The band at 954 cm À 1 for [1 0 H] þ , previously assigned to a n Ni-H mode 46 , is evidently also Fe-coupled, given its detection using NRVS. In expected from a harmonic oscillator in which H/D binds a much heavier nucleus, the n Fe-H /n Fe-D and n Ni-H /n Ni-D frequency ratios are B2 ½ . This difference in the H/D and 57 Fe nuclear masses also results in pure hydride bands having low NRVS intensities, as such vibrations involve only small displacements of the 57 Fe centre. In addition, our successful use of NRVS to detect the welldefined n Ni-D stretch in 57 Fe-labelled [1 0 D] þ contrasts Raman studies of natural Fe abundance [1D] þ , in which the Ni-D band was obscured by solvent modes 46 .
On the basis of previous studies, one would expect d OC-Fe-H/D modes to appear in the 530-750/410-640 cm À 1 regions, respectively 33,46 . The hydride [1 0 H] þ exhibits a well-defined NRVS feature at 758 cm À 1 consistent with a d Fe-H mode, this region being obscured by solvent bands in Raman data 46 . As will become clear, this 758 cm À 1 mode observed for [1 0 H] þ is of particular relevance to the interpretation of NRVS data for [NiFe]-hydrogenase. A distinct assignment for the corresponding d Fe-D mode in [1 0 D] þ is prevented by its mixing with the Fe-CO bending modes, such that the NRVS intensity is redistributed throughout the 440-630 cm À 1 region.
Analysis of [1 0 H] þ revealed several intense Fe-CO bands in the 440-630 cm À 1 region (Fig. 2b), the energies of which are almost identical to those of the conjugate base (dppe)Ni(m-pdt) 57 Fe(CO) 3 (1 0 ) 47 , which lacks the hydride bridge. This is exemplified by the d Fe-CO NRVS triplets for [1 0 H] þ (558, 587 and 617 cm À 1 ) and 1 0 (557, 588 and 613 cm À 1 ) being virtually coincident (Supplementary Table 1), suggesting that the presence of H À does not significantly perturb the d Fe-CO dynamics. In contrast, the d Fe-CO region for [1 0 D] þ collapses to a pair of bands at 580 and 608 cm À 1 , consistent with significant coupling to the d Fe-D bending motion. Thus, although Fe-D modes have intrinsically low NRVS intensity, the Fe-D/Fe-CO coupling allows for the high-intensity d Fe-CO region to serve as an indicator of whether or not a Fe-D moiety is present 33 .  ARTICLE Given that the broad 4700 cm À 1 region is solely populated by Ni/Fe-H/D bands, DFT also allows for a confident assignment of these NRVS features (Fig. 2b,c) despite their low intensities and the difficulties recognized in the accurate theoretical prediction of M-H vibrational frequencies 46 .
In line with previous DFT calculations and Raman analyses 46 , the key distinction between calculated NRVS data for the two [1 0 H] þ Ni/Fe-flippamers results from splitting of the n Fe-H ¼ 1,479/1,447 cm À 1 and n Ni-H ¼ 1,022/1,061 cm À 1 modes, respectively, as indicated in Fig. 2c. For the Ni-flippamer, the calculated n Fe/Ni-H modes are shown correspondingly in Fig. 2f Table 3) gives rise to the inverted character of these B30-40 cm À 1 flippamer-dependent splittings.
One of the most interesting and useful results from the present calculations on [1 0 H] þ is the prediction of a 57 Fe PVDOS band at 774/767 cm À 1 for the Ni/Fe-flippamers, respectively. With only a small flippamer-dependent splitting of 7 cm À 1 , this mode gives rise to the most intense feature above 700 cm À 1 and aligns well with the NRVS band observed at 758 cm À 1 (Fig. 2b,c). Inspection of the DFT-calculated nuclear displacements (see  Table 1). The results of such mixing are evident in the intense 440-630 cm À 1 Fe-CO region (Fig. 2b,c). Thus, the NRVS signatures of the Ni-H-Fe moiety in the enzyme mimic are the weak wag band observed for [1 0 H] þ and the change in the amplified features in the Fe-CO region when comparing spectra of Hydrogenase NRVS results. NRVS data for Ni-R in H 2 /H 2 O and in D 2 /D 2 O are compared in Fig. 3a. While vibrations of the three electron transport Fe-S clusters exclusively populate the o420 cm À 1 region 39,40,43 , this work instead focuses on the Fe-CO/CN region and higher-energy NRVS features to assign spectroscopic markers characteristic of the bridging hydride. Analysis of Ni-R in H 2 O revealed a sharp band at 549 cm À 1 previously assigned to a n Fe-CO mode 39 , with additional features at 454, 475 and 502 cm À 1 arising from a mixture of Fe-CO and Fe-CN modes. Compared with our previous results 39 , the absence of shoulders and additional features around the n Fe-CO band indicates a higher level of sample purity. The NRVS band positions are similar to but nevertheless distinct from Raman peaks for the Ni-L Ni(I)Fe(II) state of [NiFe]-hydrogenase, for which n Fe-CO was observed at 559 cm À 1 (ref. 53). Samples of Ni-R in H 2 O exhibit bands at 590 and 609 cm À 1 that collapse to a single peak at 609 cm À 1 when D 2 O is instead used. Qualitatively, one can attribute the differences in this Fe-CO/CN region to a different coupling to Fe-H and Fe-D motion in the respective samples, as discussed above for [1 0 H/D] þ . Other details about the active site, such as whether cysteine ligands are unprotonated or protonated, cannot be addressed on the basis of the NRVS data alone.
NRVS analysis of [NiFe]-hydrogenase in H 2 O also revealed a weak but well-resolved band at 675 cm À 1 not observed for other samples. This band is presumably related to the Ni-H-Fe wag exposed for [1 0 H] þ at 758 cm À 1 (Fig. 2b,d), making this the first assignment of a Fe-H-related mode in any enzyme by NRVS and the first direct spectroscopic evidence for a Ni(m-H)Fe core in Ni-R. Deuteration of Ni-R is expected to red-shift this mode into the 420-620 cm À 1 Fe-CO/CN region, in agreement with the changes observed and calculated for [1 0 D] þ . Analogous to the model complex, the Ni-D-Fe wag is strongly mixed with Fe-CO/ CN modes, which, in the case of Ni-R, makes its unique assignment very difficult.
Hydrogenase DFT results. To interpret NRVS measurements in terms of suitable structural candidates for Ni-R, we performed DFT calculations on a series of active site models featuring different binding modes of the H 2 substrate or its heterolysis products (see Supplementary Fig. 19). Limiting our models to the [NiFe] site is appropriate in that Fe-S clusters do not feature NRVS bands in the 4420 cm À 1 region of interest 39,40 . Two main structures were considered: one in which substrate is present in the form of a dihydrogen ligand ((Z 2 -H 2 )NiFe, I or NiFe(Z 2 -H 2 ), II) and another where a bridging hydride is present (Ni(m-H)Fe, III or HNi(m-H)Fe, IV). In addition, variants of III, in which a terminal Cys ligand is protonated ((Cys546)SHNi(m-H)Fe, V and (Cys81)SHNi(m-H)Fe, VI), were also studied. Taking into account the Ni(II)Fe(II) Ni-R active site 16,18 , and assuming Fe(II) remains low-spin, each model may exist in electronic singlet (S ¼ 0) or triplet (S ¼ 1) Ni(II) states, both of which were evaluated computationally. The DFT-calculated 57 Fe PVDOS for selected models were compared with the NRVS data for Ni-R over the range 400-750 cm À 1 (Fig. 3a- (Fig. 3d,e, superscript S denotes singlet Ni(II)) match experimental data remarkably well, with the number and positions of absorption bands being in accordance. Relative intensities of the calculated peaks are also in good agreement with our measurements.
According to the normal mode analysis of model V S (Supplementary Figs 28 and 29; animated representations of vibrational modes for models V S and VI S are provided in Supplementary Movies 4-34), vibrations in the 440-504 cm À 1 region predominantly involve Fe-CN bending and stretching, while higher-energy bands (543-613 cm À 1 ) are derived from Fe-CO vibrations. The feature calculated at 543 cm À 1 has significant n Fe-CO character, while that at 613 cm À 1 is assigned to a d Fe-CO mode. Likewise, the 588 cm À 1 band can be assigned to an H-Fe-CO bend in which H, Fe and C remain nearly collinear. The above normal modes are highly mixed, which make the assignment of individual fragments complicated.
Both bridging hydride models V S and VI S are predicted to exhibit a weak Ni-H-Fe out-of-plane wagging band (at 727 and 692 cm À 1 , respectively, see Fig. 3; Supplementary Movies 11 and 28 for mode animations) whose intensity is comparable to that of the 675 cm À 1 feature observed for Ni-R. When compared with the Fe-CO/CN region, both theory and experiment predict lower NRVS intensity of the wag in Ni-R than that observed and calculated for [1 0 H] þ (at 758 and 774/767 cm À 1 , respectively, see Fig. 2). Moreover, simulations for V S and VI S accurately reproduce the disappearance of the 675 cm À 1 band in data for Ni-R in D 2 O. The difference between the experimental and calculated frequencies of the wagging mode are likely due to limitations in our model, which does not take into account direct contacts between the protonated cysteines and surrounding residues, as well as anharmonicity effects. Moreover, one would expect an intrinsic error of the chosen functional/basis set combination. While hydride bands are extremely sensitive to (electronic) structure, we note that the observed error is still well within normal limits 54-58 expected for this methodology. However, since the full spectral information is considered in the interpretation, the present conclusions can be made with confidence. Finally, calculated H/D isotope shifts for the two representative models (Supplementary Tables 4 and 5) in the lowenergy region are also fully consistent with the observed data. The overall analysis here identifies the 675 cm À 1 feature in the Ni-R NRVS spectrum as the Ni-H-Fe wag mode.

Discussion
Our NRVS measurements on synthetic bridging hydrides and Ni-R, combined with DFT calculations, provide new constraints on the structure of this key catalytic state of [NiFe]-hydrogenase. Spectra of [1 0 H/D] þ feature characteristic Fe-H/D stretches whose energies (1,532/1,468 cm À 1 for the Ni/Fe-flippamer, respectively) are comparable to those for bridging hydrides in other structures, including another recently reported Ni(m-thiolate) 2 (m-H)Fe species for which infrared spectroscopy revealed a n Fe-H band at 1,687 cm À 1 (ref. 28). Symmetrical m 2 -H À bridges, such as those in Fe 4 H 4 þ clusters, give rise to symmetric stretches at around 1,400 cm À 1 (ref. 59), while purely terminal hydrides have n Fe-H B1,700-2,300 cm À 1 (refs 33,60). The sensitivity of hydride vibrations to structural perturbations underscores their enormous diagnostic value in understanding catalyst structure and function.
Unfortunately, even the strongest of these relatively pure stretches, n Fe-D , is predicted to have much lower NRVS intensity than the n Fe-CO modes. Thus, while [1 0 H/D] þ allowed for direct observation of n Fe-H/D and n Ni-H/D modes, the resolution of similar bands for [NiFe]-hydrogenase is beyond our current capabilities.
Of special significance is the DFT prediction of a Ni-H-Fe wag (Fig. 2d), this vibration being assigned to an observed NRVS band at 758 cm À 1 for [1 0 H] þ , a mode likely obscured by solvent bands in Raman spectra 46 . A key advantage of NRVS is thus demonstrated in that its sole detection of modes coupled to the Mössbauer-active 57 Fe nucleus makes it unaffected by solvent or matrix modes. The NRVS intensity of the Ni-H-Fe wag is at least four times greater than those of the Ni-H/Fe-H stretches. The Ni-H-Fe wag is a valuable diagnostic probe of the Ni-R structure, with NRVS data for Ni-R in H 2 O featuring a weak but reproducible band at 675 cm À 1 that is absent when a D 2 O medium was used (Fig. 3). Given that 57 Fe NRVS-active modes necessarily involve motion of this metal centre, observation of an H/D isotopically sensitive band at 675 cm À 1 is strong evidence for the presence of a Fe-H moiety in Ni-R.
A second observable indicating the presence of a bridging H À /D À in both Ni-R and its mimics [1 0 H/D] þ stems from the coupling of Fe-CO and Fe-CN stretches and bends with the Ni-D-Fe wag. While coupling to the Fe-CO/CN modes makes resolution of the Ni-D-Fe wag impossible, it also results in marked changes in band position and intensity in the 440-630 cm À 1 region on H/D substitution (Figs 2 and 3). In line with the marked isotope effects described above for [1 0 H/D] þ , NRVS spectra of Ni-R display similar shifts in position and intensity around 450-480 cm À 1 , splitting of a band at 502 cm À 1 , and Taken together, NRVS analysis of Ni-R, in combination with NRVS and DFT data for [1 0 H/D] þ , indicates that models VI S and V S provide a consistent and detailed picture of Ni-R. This cohesive study thus represents the first evidence from vibrational spectroscopy for the presence of a bridging H À in the active site of Ni-R, a state central to the function of [NiFe]-hydrogenase. The combined experimental and theoretical approach described here has applicability far beyond the Ni-R and [NiFe]hydrogenase. Ideal for the detailed study of Fe-H fragments, we envisage that such methods will also be of importance for unravelling the mechanisms of [FeFe]-hydrogenase and nitrogenase 61 , as well as for the development of synthetic catalysts inspired by these metalloenzymes. The isotopic purity of the hydrides and deuterides was confirmed using multinuclear ( 1 H, 2 H, 13 C and 31 P) NMR spectroscopy, infrared spectroscopy and ESI mass spectrometry. NRVS analysis was conducted on a solid sample of each of the four model complexes (vide infra).

D. vulgaris Miyazaki F [NiFe]-hydrogenase preparation.
[NiFe]-hydrogenase expressed in D. vulgaris was isolated and purified as described earlier 62 . The as-isolated protein was transferred from 25 mM Tris-HCl (pH ¼ 7.4) buffer to 100 mM MES (pH ¼ 5.0). The solution was placed in a tube, which was sealed, degassed and then purged with H 2 (1.2 bar) for 8 h. In the case of the sample in D 2 O, the buffer was replaced by 100 mM MES (pD ¼ 5.0) in D 2 O and the mixture placed under D 2 (1.3 bar) for 8 h. Samples were transferred to an anaerobic chamber and loaded into NRVS cells. FT-IR spectra were recorded on a Bruker IFS66v/S FT-IR spectrometer with a 2 cm À 1 spectral resolution at 293 K ( Supplementary Fig. 31).
NRVS measurements and data analysis. The NRVS data were collected according to a published procedure 39 at SPring-8 BL09XU (with flux B1.4 Â 10 9 photons s À 1 ) and BL19LXU (B6 Â 10 9 photons s À 1 ) using 14.4 keV radiation at 0.8 meV resolution. The spectral maximum counts/second (cts s À 1 ) at BL19 is B2.6-3 times of that at BL09. To compare the data from different beamlines, we rescale the BL19 counting time on the basis of its max cts s À 1 versus the max cts s À 1 at BL09 and create BL09 equivalent seconds, for example, the 10 (s) per point (s/pt) at BL19 is corresponding to 26 or 30 equivalent s/pt at BL09. Delayed nuclear and Fe K fluorescence (from internal conversion) were recorded with a 2 Â 2 APD (avalanche photodiode) array in either beamline, and raw NRVS data were converted to single-phonon 57 Fe PVDOS using the PHOENIX software 39 . Sample temperatures were maintained at 30-50 K during analysis.
Model complex calculations. Initial coordinates for the DFT calculations on [1 0 H/D] þ were extracted from the X-ray structure of [1H]BF 4 Á 3THF (ref. 44). The methodology applied was mostly equivalent to our earlier set-up on [1 0 ] 0/ þ (ref. 52). Structural optimizations and subsequent normal mode analyses were performed using GAUSSIAN 09 on the basis of the densities exported from single point calculations using JAGUAR 7.9. The BP86 functional and the LACV3P** basis set were employed. The environment was considered using a self-consistent reaction field model. 57 Fe PVDOS spectra were generated using Q-SPECTOR, successfully applied earlier 33 . Simulated spectra were broadened by convolution with a full-width at half-maximum ¼ 12 cm À 1 Lorentzian.