A safety cap protects hydrogenase from oxygen attack

[FeFe]-hydrogenases are efficient H2-catalysts, yet upon contact with dioxygen their catalytic cofactor (H-cluster) is irreversibly inactivated. Here, we combine X-ray crystallography, rational protein design, direct electrochemistry, and Fourier-transform infrared spectroscopy to describe a protein morphing mechanism that controls the reversible transition between the catalytic Hox-state and the inactive but oxygen-resistant Hinact-state in [FeFe]-hydrogenase CbA5H of Clostridium beijerinckii. The X-ray structure of air-exposed CbA5H reveals that a conserved cysteine residue in the local environment of the active site (H-cluster) directly coordinates the substrate-binding site, providing a safety cap that prevents O2-binding and consequently, cofactor degradation. This protection mechanism depends on three non-conserved amino acids situated approximately 13 Å away from the H-cluster, demonstrating that the 1st coordination sphere chemistry of the H-cluster can be remote-controlled by distant residues. [FeFe]-hydrogenases catalyze the conversion of protons and electrons to molecular hydrogen, but upon exposure to oxygen, their catalytic cofactor is irreversibly inactivated. Here, the authors determine the crystal structure of hydrogenase CbA5H and identify a cysteine residue, which acts as a safety cap that shields the active site from oxygen.


[Fe
Fe]-hydrogenases catalyze the reversible reduction of protons to dihydrogen at low overpotential and high turnover rates [1][2][3] . Their catalytic cofactor consists of a [4Fe-4S]-cluster ([4Fe] H ) that is covalently bound to a diiron complex ([2Fe] H ) by a bridging cysteine 4 . The two Fe-sites of the [2Fe] H complex, referred to as proximal (Fe p ) and distal (Fe d ) according to their distance to the [4Fe] H -cluster, are coordinated by three carbon monoxide (CO) and two cyanide (CN -) ligands 1 . They exhibit strong vibrational absorption signals at discrete infrared frequencies, which respond to changes in redox and protonation states of the H-cluster [5][6][7] . The diatomic ligands are responsible for the low-spin state of the [2Fe] H sub-cluster and stabilize it in a configuration that creates an open coordination site at Fe d , allowing the binding and heterolytic splitting of dihydrogen 8,9 . The open coordination site at Fe d is also the major target of inhibitors such as CO and dioxygen [10][11][12][13][14] . The pending bridgehead amine-group of the azadithiolate-ligand which connects Fe p and Fe d in the [2Fe] Hcluster shuttles protons between Fe d and a nearby located cysteine residue (C367 in CbA5H), which is part of the highly conserved proton transfer pathway 4,6,[15][16][17][18] .
Dioxygen irreversibly damages the H-cluster of most [FeFe]hydrogenases [10][11][12]14 . After reaching the active center by diffusion via packing defects 11,19-21 , O 2 binds to the open coordination site at Fe d , forming a transient adduct which is further transformed by successive protonation and reduction steps 11,14,22,23 . To a limited extent it can be fully reduced to water, partial reduction and protonation however, lead to the production of reactive oxygen species (ROS) that cause H-cluster destruction 11,12,22-26 . To the best of our knowledge, among the [FeFe]-hydrogenases that have been characterized so far, only the recently isolated CbA5H of Clostridium beijerinckii, a gram-positive, anaerobic bacterium isolated from soil and feces, has been shown to resist long-term exposure to O 2 , by reversibly switching from the active oxidized ready state H ox to the O 2 -protected but inactive Hcluster state H inact 27 . This could be demonstrated by monitoring the spectroscopic signature of the enzyme that is repeatedly oxidized in air and reduced by H 2 (ref. 27 and Supplementary Figure 1). The H inact -state was originally identified as the 'as isolated' form of [FeFe]-hydrogenase DdH from Desulfovibrio desulfuricans after homologous expression and aerobic purification 28,29 . H inact provides full protection against O 2 prior to the first reductive activation to H ox and can be identified spectroscopically by a characteristic set of IR-signals. However, DdH becomes O 2 -sensitive after gaining catalytic activity 10,19 . For DdH and some other [FeFe]-hydrogenases, sulfide-addition under oxidizing conditions (either aerobic or anaerobic) also produces the H inact -state; the anaerobic formation of H inact is detected in cyclic voltammetry by the early onset of oxidative inactivation 30,31 . Rodríguez-Maciá et al. concluded that H inactformation under oxidizing conditions is the consequence of sulfide-binding to Fe d 30 , but this protected state is also formed in CbA5H 27,31 and CpIII ([FeFe]-hydrogenase III from Clostridium pasteurianum) 32 in the absence of exogenous sulfide. The sulfideindependent, reversible transition between H ox and the inactive, O 2 -resistant H inact -state is therefore an uncharacterized and intriguing feature of certain [FeFe]-hydrogenases, which significantly enhances their utilization potential ( Fig. 1; CbA5H).
In this work, we elucidate the mechanism that protects CbA5H from dioxygen employing a multidisciplinary approach. We describe the structural rearrangement of a polypeptide-loop close to the active site that attributes a 'safety cap' function to the conserved cysteine at position 367. It shields the open coordination site of Fe d from O 2 by enabling the thiol-group to reversibly bind to Fe d , thus providing an intrinsic source for the protective sulfide-ligand.

Results
Crystal structure of O 2 -exposed CbA5H WT Table 2); this stability of the active site in the air-exposed crystal contrasts with the observation that the H-cluster of standard [FeFe]-hydrogenases is destroyed under air 26 . Since we could not crystallize CbA5H under anaerobic conditions, we compared the structures of CbA5H air and CpI, a "standard" [FeFe]-hydrogenase. The structure of CbA5H air shows localized structural differences with anaerobically purified CpI (PDB ID: 4XDC) 9 (Supplementary Figure 3). The peptide-loop that spans T365, S366, and C367 in CbA5H (hereafter "TSC-loop", corresponding to T297, S298, and C299 in CpI) is shifted from the conformation observed in the structures of CpI, DdH, or HydA1 ( Fig. 2e and Supplementary Figure 5). The bulky side chain of the strictly conserved tryptophan 371 (W303 in CpI), adjacent to the TSCloop, exhibits a different conformation in CbA5H air , hinging away from the loop. The alpha helix carrying this residue is slightly shifted away from the H-cluster. Consequently, the orientation of the residue of the conserved cysteine C367 in CbA5H air is different from that of the corresponding side chain in standard [FeFe]-hydrogenases (Fig. 2d, e and Supplementary Figs. 3 and 5). This observation is consistent with a recently published hypothesis according to which the loop that bears C367 in CbA5H may be flexible enough to allow the binding of the cysteine to Fe d 31 . The strictly conserved cysteine residue is involved in long-range proton transfer to and from the H-cluster (Supplementary Figure 6) 4,16,18 . In CbA5H air , the distance between the sulfur atom of C367 and Fe d is only 3.1 Å, compared to 5.9 Å in CpI (Fig. 2e)  for a covalent bond between iron and sulfur 34 , but short enough to prohibit the insertion of any molecule (including H 2 and O 2 ) between the sulfur atom and Fe d . This observation suggests that the structure of CbA5H air is that of an inactive state and explains why the H-cluster is stable under air. That the infrared spectrum of air-exposed CbA5H WT ( Fig. 3c and Supplementary Figure 1) shows the typical signature of the H inact state suggests that the Xray structure of CbA5H air , with C367 attached to Fe d , is actually that of H inact .
The crucial role of C367 for H inact -state formation. We compared the properties of CbA5H WT and two site-directed variants in which C367 is replaced with either aspartate or alanine. Below, we demonstrate that each of the three distinctive properties of CbA5H is dependent on the presence of a cysteine at position 367: (1) the non-standard anaerobic oxidative inactivation which occurs at low oxidative potential, (2) the resistance to O 2 that results from this inactivation, and (3) the reversible formation of the H inact -state upon exposure to O 2 . In standard [FeFe]-hydrogenases, the replacement of the conserved cysteine in the proton transfer pathway with aspartate is the only substitution that preserves a significant level of enzyme activity 16,18 . Consistent with previous results with CpI and HydA1, we observed that replacing C367 with alanine renders CbA5H inactive while variant C367D retains 20% of the H 2 -production activity measured for CbA5H WT and exhibits a shifted pH-optimum (Supplementary Figure 7) 18 .
H inact can be accumulated anaerobically by incubating CbA5H WT with oxidants (DCIP or thionine) 27 . Such oxidative treatment can be mimicked in protein film electrochemistry (PFE) experiments by applying high enough electrode potentials 35,36 . The voltammetric response of CbA5H WT (black curve in Fig. 3a, d) resembles that of CpIII 32 and markedly differs from that of all other characterized [FeFe]-hydrogenases 37 . As expected, a negative current is seen at electrode potentials lower than the Nernst potential of the H + /H 2 redox couple 38 , which reveals proton reduction under reductive conditions. Upon increasing the electrode potential, the current becomes positive, but in contrast to other [FeFe]-hydrogenases, the current strongly decreases above −0.3 V vs SHE (standard hydrogen electrode, all potentials refer to SHE) as a result of anaerobic, oxidative inactivation, as recently observed in another electrochemical investigation of CbA5H 31 . This inactivation at relatively low potential is reminiscent of the shift in the onset of anaerobic inactivation to lower potentials observed for DdH and HydA1 in the presence of exogenous sulfide 30 . When subsequently sweeping the potential down, a faint increase in current, starting below -0.25 V, reveals the beginning of enzyme reactivation, which is complete only when the potential reaches −0.6 V. The C367D exchange causes a striking effect on the voltammetry (blue trace in Fig. 3a): the onset of inactivation occurs at a significantly ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-20861-2 higher potential than for CbA5H WT , thus restoring a "standard" voltammetric signature. The latter is illustrated for CpI and HydA1 in Fig. 3a and Supplementary Figure 8, respectively: inactivation occurs at a much higher potential (above 0 V) than for CbA5H, and reactivation begins below −0.1 V. The standard oxidative inactivation of [FeFe]-hydrogenases is fully dependent on the presence of chloride ions acting as uncompetitive inhibitors 39 . Inhibition by chloride is also observed for variant C367D, whereas chloride has only minor effects on CbA5H WT (Supplementary Figure 9). Anaerobic inactivation of standard hydrogenases and CbA5H-variant C367D on the one hand, and CbA5H WT on the other hand, therefore, result from distinct mechanisms.
To quantify and compare the O 2 -sensitivity of CbA5H WT and variant C367D, we employed a PFE chronoamperometric procedure where the enzyme is exposed to O 2 under oxidizing conditions, and the level of activity is measured under reductive conditions before and after exposure to O 2 (Fig. 3b). The measured catalytic H 2 -evolution currents were normalized by the initial value recorded at the end of the 1st step at low potential (−0.3 V). Upon switching the potential to 0 V, CbA5H WT instantly inactivates whereas C367D retains >95% of its H 2oxidation activity. Injecting 50 µM O 2 into the electrochemical cell fully inhibits the C367D variant. After removal of O 2 from the system by a five-fold buffer exchange and subsequently shifting the potential back to −0.8 V, CbA5H WT returns to its original H + reduction activity, unlike variant C367D, which appears to have been completely and irreversibly inactivated by O 2exposure. These experiments demonstrate that anaerobic inactivation at relatively low potential and resistance to O 2 -induced damage both depend on C367.
As observed by IR-spectroscopy, exposure of CbA5H WT to air induces a quantitative transition from H ox to H inact ( We have therefore shown that (1) anaerobic inactivation, (2) O 2 -resistance, and (3) aerobic formation of H inact all occur due to the presence of C367 which according to our structure of CbA5H air binds to Fe d in the H inact state. This saturates the coordination sphere of Fe d and thus prevents the binding of substrate (H 2 ) and inhibitor (O 2 ) (Fig. 2d-e), rendering the enzyme inactive but protected from O 2 -induced degradation. That CbA5H is quantitatively locked in the H inact configuration is not only obvious from the homogenous IR-spectra of O 2 treated enzyme but also from the low b-factor values observed for the Hcluster environment in CbA5H air (6TTL) including the fully shifted TSC-loop ( Supplementary Fig. 16).
That both, aerobic and anaerobic oxidative conditions induce the formation of H inact suggests that the sensing of oxidative conditions occurs via the two accessory [4Fe-4S]clusters (FS4B and FS4A in Fig. 2b). They mediate electron transfer between external redox partners and the H-cluster and can be oxidized either by the electrode at high potential or by transferring electrons to molecular oxygen. This reaction with O 2 appears to be reversible.
Kinetics and mechanism of reversible anaerobic inactivation. We examined the kinetics and potential-dependence of anaerobic (in)activation, which can be quantitatively probed using PFE by potential-step experiments 40 . Figure 4a shows a typical sequence of steps and the resulting change in current. Each initial current peak (e.g., at 50 or 100 s) results from the turnover frequency of the fully active enzyme instantly changing after the potential is stepped up or down; the subsequent slow change in current results from the accordingly slow change in the concentration of the H inact state. The data can be interpreted by assuming various kinetic models, which we tested by fitting them to the corresponding current traces 41 . A model that considers just two species (corresponding to the cysteine being either unbound or bound to Fe d ) and resulting in mono-exponential changes in current after each potential step is not satisfactory (green trace in Fig. 4a): the kinetic traces are multiphasic, and any good kinetic model must therefore include the conversions between at least three species (red trace in Fig. 4a). After having analyzed the data recorded at different potentials and pH values, we concluded that the simplest good model is the following, where A means "active", and A 1 and A 2 are two distinct active species.  Figure 11).
Since the C367D mutation prevents the formation of H inact and gives back CbA5H standard catalytic properties, we assume that the structure of the active form A 1 of CbA5H is similar to that of standard hydrogenases, and that the difference between the structures of CpI and CbA5H shown in Fig. 2e illustrates the conformational change that occurs upon the formation of H inact from the active form A 1 of the enzyme. The conversion between the active forms A 1 and A 2 occurs on the time scale of seconds (1/(k 1 + k −1 ) =10 s) which implies that there must be a large activation energy barrier to overcome, probably due to a conformational change. In state A 2 , C367 cannot be bound to Fe d , since A 2 is still active (about half as active as A 1 , Fig. 3 Electrochemical and spectroscopic features of CbA5H WT and mutagenesis variants. a Comparison of cyclic voltammograms of CbA5H WT (Cb-WT), CpI (CpI-WT), and Cb-variant C367D (T = 5°C, pH 7, 1 atm. of H 2 , scan rate 3 mV/s, electrode rotation rate 3000 rpm, currents normalized at E = −0.56 V). b Potential step chronoamperometry of CbA5H WT and C367D (5°C, pH 7, 1 atm. of H 2 , 1000 rpm). P1: H 2 -production current at −0.8 V prior to O 2 -exposure; P2: potential step to 0 V; P3: injection of 50 μM O 2 ; P4: five-fold buffer exchange to re-establish anaerobic conditions; P5: potential step to −0.8 V to measure the residual H 2 -production current. c ATR-FTIR-spectroscopy of CbA5H WT Figure 12). The binding of the cysteine sulfur to Fe d must therefore occur in the final step (A 2 to I). That this bond is labile 1=ðk inact þ k react Þ ' 1 ð s at high potential) is consistent with the relatively long distance of 3.1 Å between Fe d and the Satom of C367 observed in the structure of air-oxidized CbA5H WT . The reactivation from H inact to A 2 is the only step whose rate constant (k react ) depends on pH and the applied potential, showing that reactivation is triggered by a reduction step coupled to a protonation. At high potential, the reactivation rate constant k react is lower than the inactivation rate constant k inact , locking down the enzyme in the H inact state. The fact that k inact is potential-independent points to a classical "CE" mechanism for step 2, in which one or a series of chemical step(s) ("C") precede(s) proton-coupled oxidation ("E"). Strong coupling implies that the sites of deprotonation and oxidation are very close to one another. We therefore consider likely that this deprotonation occurs from the nitrogen atom of the azaditiolate bridge, or from the bound cysteine, to produce a thiolate ligand.
Three distal residues determine reversible H inact -formation. The formation of H inact as a result of cysteine binding to Fe d is a unique property of CbA5H, and yet C367 and the other residues of the shifted TSC-loop are strictly conserved (Supplementary Figure 5). Structural differences in their environment must therefore account for the loop-flexibility that enables the translocation of C367 in CbA5H WT . The most prominent differences between standard hydrogenases and CbA5H WT near the TSCloop concern the residues at positions 364 (leucine), 561 (alanine) and 386 (proline) (Fig. 2e, f and Supplementary Figure 5): their backbones align with those of their counterparts in standard hydrogenases, but the latter exhibit bulkier residues.
To assess whether the residues at these positions determine the possibility of a conformational change and are responsible for the unique properties of CbA5H, we produced the CbA5H-variants L364F, A561F, P386L and the double-exchange variant L364F-A561F, and characterized them using the same combination of experiments as described above. Only variant P386L shows a reduced H 2 -evolution activity (55% compared to CbA5H WT , Supplementary Figure 7).
The cyclic voltammograms recorded with the four variants show reversible anaerobic inactivation, but at a significantly higher potential than observed for CbA5H WT (Fig. 3d). We could fit the above kinetic model to the chronoamperometric data recorded with each of the mutants; the results in Table 1 confirm that the mutations hinder the A 1 to H inact conversion and inform about their individual effects on each of the two steps. The L364F and P386L substitutions decrease k inact and increase k react (the P386L substitution also increases k −1 , Table 1). The A561F substitution only significantly affects the 1 st step (A 1 to A 2 ): it decreases k 1 and increases k −1 ( Table 1). The double-exchange variant L364F-A561F combines the individual effects of the two single exchanges. Figure 3e shows that each of the substitutions also decreases O 2 -resistance. Each of the variants exhibited some residual level of activity just after the first potential step from −0.8 V to 0 V, followed by an inactivation that is slower than that of CbA5H WT . After buffer exchange and the final step to −0.8 V, the residual activities of all single-exchange variants are down to 76-64% and the double-exchange variant merely reaches 34% of the activity recorded prior to O 2 -exposure, compared to 100% for CbA5H WT (also in Fig. 3b). These results show that resistance to O 2 correlates with the overall rate of formation of the inactive state Fig. 4 Anaerobic conversion between active and inactive CbA5H WT monitored by PFE. a Chronoamperograms recorded to analyze the kinetics of anaerobic (in)activation of CbA5H WT by stepping between −0.31 and −0.21 V. The boxes along the abscissa depict the sequence of applied potential steps and the corresponding current response (black line). The plot also shows the best fits of models that consider the interconversion between two (green dotted line) or three (red dotted line) species. Experimental parameters: 5°C, pH 7, 3000 rpm. b Dependence of the rate constants of the "AAI" model (A1 ↔ A2 ↔ H inact ) on potential and pH, based on the analyses of chronoamperograms recorded for CbA5H WT at pH 10 (squares), 8.5 (triangle), and 7 (circle). Only k react significantly depends on pH and electrode potential. Source data are provided in a source data file. Table 1 Values of k 1 , k -1 , k inact , k react (the latter at −0.196 V, all at 5°C) for CbA5H WT and variants. Rate constants were obtained by analyzing kinetic data of anaerobic inactivation such as those shown in Fig. 4a with the "AAI" model ( Supplementary Fig. 14 shows the dependence of k react on E). The CbA5H WT reference for the kinetic parameters of the variant proteins, measured at pH7 is marked in bold. The % accuracy is our estimate of the maximal error for the determination of each rate constant.
Monitoring the IR-signatures of the H-cluster confirms that the residual O 2 -resistance of the variants results from the formation of a residual fraction of H inact (Fig. 3f and Supplementary Figure 13). Prior to O 2 -treatment, all variants show the signature of the H ox state, with minor contributions from H ox -CO. However, unlike the case of CbA5H WT (Fig. 3c), these signals disappear upon exposure to air. The only remaining signals after O 2 -treatment result from H ox -CO and a small fraction of H inact which is just above noise level for L364F and P386L, and barely visible for L364F-A561F, thus contrasting with the clear and homogenous H inact spectrum of CbA5H WT .
We therefore conclude that each of the three substitutions that were expected to alter the flexibility of the TSC-loop slows down oxidative inactivation, limits the extent of H inact -formation and, consequently, decreases the resistance of CbA5H to O 2 . The exchanges have distinct effects on the two steps of the inactivation process (Table 1), and this information is used hereafter to elucidate the conformational change occurring upon formation of H inact .
In summary, we deduce the events leading to oxidative inactivation and accumulation of H inact according to the AAI model (Table 1 and Fig. 5) as follows.
The first step of the conformational change (from state A 1 , as observed in the structure of CpI, to A 2 ) cannot move C367 significantly closer to Fe d since A 2 is still active. We consider likely that the A 1 /A 2 transition which involves a turn of the segment including CbA5H-T365 and -S366 in the TSC-loop (equivalent to T297 and S298 in CpI) (Fig. 5a, c, step 1), consistent with this first step being slowed only by the A561F substitution (Table 1) which induces a steric clash with T365 of CbA5H air (Fig. 5a). This movement of T365 and S366 requires a reorganization of the H-bond network that stabilizes TSC loop configuration A 1 , including a disruption of the H-bond contact between the N-atom of the indole ring and the carbonyl oxygen of T365 (Supplementary Fig. 15) and provides the necessary space for the subsequent translocation of W371 (see clash between Trp (red) and Ser (blue) in its original position of state A 1 in Fig. 5a) and of the alpha helix that bears it (Fig. 5b, c; step 2). The substitution L364F favors state A 2 over H inact (Table 1) because it removes the slight steric clash between L364 and the W371 in state A 2 (Fig. 5b, blue Trp). The conformational shift of W371 in step 2 should be slowed when P386 is replaced with leucine (see clashes between W371 of CbA5H air and L318 of CpI in Fig. 5b), which is consistent with the transition from A 2 to H inact being disfavored in variant P386L (Table 1).
Step 2 brings C367 close enough to Fe d to allow its coordination (Fig. 5c, step 3), which should favor cluster oxidation and deprotonation (probably of the bound cysteine). This oxidation step finally locks down the enzyme in the H inact -state.
The protective function of the thiolate group described here is analogous to that of exogenous sulfide demonstrated for several standard [FeFe]-hydrogenases 30 . Sulfide, the main product of the sulphate metabolism of D. desulfuricans, could be trapped near the H-cluster and be quickly relocated under oxidative conditions to the open coordination site of Fe d in order to shield it against O 2 -attack. To make use of this effect, the presence of extrinsic sulfide as a potential-dependent Hcluster inhibitor would be mandatory. Likewise relying on the capability of a sulfur ligand to reversibly occupy the substrate coordination site of the H-cluster, CbA5H adopts the same protected state as DdH but independent of the presence of external S 2− . The intrinsic security cap may have been the consequence of evolutionary pressure to cope with incidental O 2 -exposure in the absence of abundant sulfide.
Our study exemplifies how amino acid residues that are located far from the active site, still may determine the ability to undergo a conformational change which controls chemistry in the 1st coordination sphere, allowing e.g. the enzyme to resist O 2 attack. Enzymes benefit from the multiplicity of structural and functional features of the protein scaffold in ways that are far more elaborate than merely providing a cofactor cavity that stabilizes and tunes the properties of the active site, and substrate/product pathways. The intrinsic cofactor protection mechanism of CbA5H is an impressive showcase for the level of complexity that such additional contributions may reach under corresponding evolutionary pressure.

Methods
Protein preparation. The cDNA used for heterologous overexpression of CbA5H was codon optimized for Escherichia coli strain K12 and synthesized with a Cterminal spacer and strep-tagII sequence 27 before being cloned into the pET21b vector, where gene expression is controlled by the T7 promotor. QuikChange PCR was carried out to generate the constructs for the expression of site directed mutagenesis (SDM) variants using mismatch primers (Supplementary Table 3) 18 . Expression constructs were characterized via sequencing.
Protein expression, purification, and in vitro maturation were executed as described before 6,18 43 . The [2Fe] H mimic used for in vitro maturation was synthesized as described before 44 .
In vitro activity assay. To perform H 2 -production activity assays, 400 ng enzyme was mixed with 100 mM NaDT (sodium dithionite) as sacrificial electron donor and 10 mM MV (methyl viologen) as electron mediator in 100 mM K 2 HPO 4 /KH 2 PO 4 buffer (pH 6.8). The sealed reaction vessel was kept at 37°C in a shaking incubator (100 rpm) 6,18,43 . The evolved H 2 was quantified by analyzing the headspace of the reaction tube via gas chromatography (Shimadzu). To monitor the pH-dependent H 2 -production activity, the phosphate buffer was substituted with the respective pH-selective buffers (pH 5 to pH 9) 6,18 . Each measurement was done at least three times (see mean values and standard deviations in Supplementary Figure 7).
Crystallization and structure determination. Despite the strictly anaerobic handling of CbA5H WT during purification and maturation, protein crystallization was done aerobically (under air). CbA5H crystals were obtained using the hanging drop vapor diffusion method. 2 µl protein (20-40 mg/ml) was mixed with 2 µl reservoir solution, containing 0.1 M HEPES pH 7.5 (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid), 0.4 M MgCl 2 and 26-28% PEG400 (polyethylene glycol of molecular weight c.a. 400 Da). The crystals were grown at 18°C and reached full size within one week. Before flash freezing in liquid N 2 , the crystals were soaked for a few seconds in cryobuffer, being a premix of reservoir solution and 100% glycerol at a 1:1 ratio. X-ray diffraction data were collected on beamline ID29 at the European Synchrotron Radiation Facility (ESRF, Grenoble, France) at 100 K. Diffraction images were processed with XDS and the processed data were reduced and combined by running XSCALE 45 . Molecular replacement was carried out in Phaser 46 to obtain the phase. Initially, no satisfactory solutions were obtained by using either of the three homologous structures CpI (4XDC), DdH (1HFE) or HydA1 (3LX4). However, a manually trimmed model of CpI (4XDC), in which the non-conserved part was removed based on a sequence alignment, yielded a solution with TFZ (translation function Z) and LLG (log likelihood gain) scores of 14.7 and 137, respectively. Subsequently, Phenix Autobuild was used to generate the model 47 . The first round of Autobuild resulted in an improved model with an R free of 0.38. The model was then further optimized, using alternative rounds of interactive building in Coot 48 and automatic refinement in Phenix. Group occupancy was refined for the iron sulfur clusters. The final model and structure factor were deposited at RCSB protein data bank (PDB ID: 6TTL). Statistics for data collection and refinement are summarized in Supplementary Tables 1, 2. To verify placement of critical parts of the model, simulated annealing omit maps were calculated in Phenix after removing the relevant parts from the input model.
ATR-FTIR-Spectroscopy. FTIR-spectra were collected on a Bruker Tensor 27 spectrometer, equipped with a BioATR cell II (Harrick) harboring a doublereflection ZnSe/Si crystal. Prior to all measurements, the sample (15 µL) was dried at 30°C on the ATR crystal under N 2 (10 L/min) to enrich the oxidized ready state H ox . A semi-dried protein film was usually obtained after 10 minutes. The film was re-humidified by purging with an aerosol (Tris buffer, pH 8) and the temperature was decreased to 25°C before the measurements were started (2 cm -1 resolution) 49 . The recorded spectra were baseline corrected via OPUS (Bruker GmbH) and then imported and plotted by Origin (Origin Lab). For each enzyme variant, the experiment was repeated two times, showing no significant deviations.

Electrochemical experiments
Cyclic voltammetry in Fig. 3 and chronoamperometry in Fig. 4. Cyclic voltammetry and the amperometry experiments employed to elucidate the kinetics of the conformational change leading to H inact -formation were carried out in a JACOMEX glovebox filled with N 2 in a thermostated (T = 5°C) electrochemical cell with two compartments. The main compartment contained the rotating working electrode (pyrolytic graphite edge, diameter~3 mm, typical rotation rate 3 krpm, mounted on a Pine MSR rotator) and the counter electrode (a platinum wire) as well as a tube for constant H 2 -bubbling. Unless stated otherwise, the main compartment was filled with a mixed buffered solution (consisting of 5 mM of each MES, HEPES, sodium acetate, TAPS, and CHES and 0.1 M NaCl). The main compartment was connected via a Luggin capillary to a second compartment, containing the calomel reference electrode in 0.1 M NaCl. All potentials are quoted with respect to the standard hydrogen electrode, calculated using E SHE = E calomel + 0.244 V.
To prepare the protein films, the pyrolytic graphite "edge" (PGE) rotating disc working electrode was polished with an aqueous alumina slurry (Buehler, 1μm), rinsed, then painted with 0.5 µl of an enzyme solution (~5 mg/mL in 100 mM Tris-HCl, pH 8 with 2 mM NaDT) and let dry for 2 min. The measured catalytic currents were between 1-and 10 µA, so that mass transport towards the rotating electrode was not limiting 50 .
CVs and chronoamperometric experiments were recorded with the program GPES and analyzed with the open source program QSoas 41 . Chronoamperometric experiments such as those presented in Fig. 4a were started and ended at a potential 30 mV below the equilibrium potential (pH 7: −0.456 V vs SHE, pH 8.5: −0.546 V, pH 10: −0.636 V) to ensure complete activation of the enzyme film. The intermediate steps at oxidative potential were of 100 mV amplitude. Before analyzing the H 2 -oxidation current, we subtracted the background capacitive current recorded in a control experiment without enzyme. Fig. 3. Experiments were carried out in an anaerobic chamber (Coy Laboratory Products, USA) under an atmosphere of 98% N 2 and 2% H 2 . A PalmSens potentiostat (PalmSens 4) was used, controlled by the PalmSens software PSTrace 5.2. The gastight electrochemical cell was water jacketed to control the temperature and a PGE rotating disc electrode was used as a working electrode and controlled by a rotator (Autolab). The reference electrode (Ag/AgCl, 3.5 M KCl) was kept in a non-isothermal side arm, being connected to the main cell compartment by a Luggin capillary. Platinum wire was used as a counter electrode. The reference potential was converted to the standard hydrogen electrode (SHE) scale, using the correction E SHE = E Ag/AgCl + 0.205 V at 20°C. For each experiment, defined gas flow rates (Air liquid, Germany) were adjusted, using mass flow controllers (Sierra Instruments). Enzyme films were prepared by drop-casting 3 µL of a 10 µM hydrogenase sample. After 3 minutes, the electrode was rinsed with MQ water to remove all unbound enzyme molecules. Prior to the potential step experiment, a cyclic voltammogram was recorded for each enzyme film (5°C, pH 7, 1000 rpm electrode rotation, 100% H 2 atmosphere, from −1 to 0 V, using a scan rate of 20 mV/s). For the chronoamperometric experiments, a potential of −0.8 V vs SHE was applied for 200 s (Phase 1, t = 0 s) to monitor H + reduction activity. A potential step to 0 V vs SHE (phase 2, t = 200 s) was followed by the injection of 50 μM O 2 (using air-saturated buffer, phase 3, t = 480 s). A subsequent five-fold buffer exchange (5 mL cell volume, 50 mL volume exchange buffer, incubated at 5°C) was performed to reestablish anaerobic conditions (phase 4, t = 900 s) prior to the final potential step back to −0.8 V vs SHE (phase 5, t = 1300 s) which was required to measure the residual H + reduction current. For each enzyme variant the experiment was repeated at least 4 times, showing very similar results.

Chronoamperometry in
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The coordinates and structure factors of CbA5H air have been deposited in protein data bank (PDB) under PDB-ID: 6TTL All data are available in the main text or the supplementary materials. Further data supporting findings of this study are available from the corresponding authors upon reasonable request. Further publicly available datasets used or indicated in this study comprise PDB-ID: 4XDC, PDB-ID: 3LX4 and PDB-ID: 1HFE. Source data are provided with this paper.