Tracking the route of molecular oxygen in O₂-tolerant membrane-bound [NiFe] hydrogenase

In our protein crystallography experiments, we applied the soak-and-freeze method, recently established at the European Synchrotron Radiation Facility (ESRF) (24), to investigate whether O₂ reaches the [NiFe] active site through a dedicated hydrophobic tunnel network. This technique differs fundamentally from previous crystal derivatization methods as the crystals are flash-frozen in the corresponding gas in its liquid form while still under pressure. This facilitates crystal derivatization at high pressure using difficult-to-handle, highly mobile gases like krypton and molecular oxygen. The specially designed sample supports allow for higher throughput and direct use at synchrotron beamlines. MBH crystals were grown and subsequently exposed to molecular oxygen in a gas pressure cell at 56 bar and 70 bar, which is equivalent to dissolved O₂ concentrations of 72.8 mM and 91 mM, respectively. After incubation for time periods ranging from 15 to 70 min, the derivatized crystals were flash-frozen in liquefied O₂ at 77 K (Table 1). Subsequently, the gas pressure cell was depressurized and the crystals were transferred to liquid nitrogen (LN2) for analysis at the synchrotron beamline. For all O₂-derivatized crystals, a nonderivatized reference (REF) dataset was collected using MBH crystals from the same crystallization batch (Table 2). The O₂ positions were determined by examining the $F{o}^{O_2}-F{o}^{\text{reference}}$ electron density map calculated using structure factors of the O₂-derivatized (OxD) and REF datasets (Fig. 2 and Fig. S1, including additional final 2mFo − DFc, Fo − Fc difference and simulated-annealing omit electron density maps for all four O₂ molecules). The dioxygen positions were discriminated from water molecules by (i) their residence in hydrophobic tunnel cavities (Table 3), (ii) comparison with the REF dataset showing no difference in electron density at the same positions, (iii) inspection of geometry and distances to surrounding amino acids and (iv) correlation with the krypton sites of Kr-derivatized MBH (Fig. S2). Using the soak-and-freeze technique, we identified seven O₂ locations (Oxy1–7) in the best dataset (Fig. 1), which were located within the hydrophobic gas tunnel network. Oxy1–4 were modeled into the protein structure because of an occupancy of at least 50%. The real-space correlation coefficients for all four are above 91% (Table 3). Three of the O₂ molecules (Oxy5–7) showed occupancies of less than 40%, and were therefore not modeled in the structure [Protein Data Bank (PDB) ID code 5MDL]. The majority of the O₂ sites found in the structure describe a continuous pathway between the protein exterior and the [NiFe] active site. One O₂ molecule (Oxy2) resides in a hydrophobic pocket surrounded by amino acids M30_L, V46_L, S47_L, F569_L, E570_L, L573_L, and L591_L (subscript indicates subunit: L, large subunit; S, small subunit) (Figs. 1 and 2 and Table 3). Owing to the 2D representation of Figs. 1 and 2, Oxy2 appears to be close to the [NiFe] active site, but the actual distance is 16.3 Å. A maximum occupancy of 100% was observed for Oxy1 at 70 bar after 15 min of soaking in pressurized O₂ before freezing. Using Oxy1 as a benchmark, we can clearly deduce a correlation of O₂ occupancy with pressure and incubation time (Table 1). To inspect the positions of the O₂ molecules with respect to the gas tunnel network, the latter was calculated with the program CAVER 3.0 (25), using the highly conserved residue R530_L as the initial starting point (Fig. 1). The program predicts two subunit-crossing tunnels, which merge before reaching the [NiFe] active site. Tunnel A, with a minimal diameter (d_min) of ∼1.4 Å and a length of ∼33 Å, is categorized as the primary gas tunnel, whereas tunnel B (d_min of ∼1.3 Å and length of ∼52 Å) may serve as a secondary pathway. Approximately 56% of the amino acids forming the tunnel surface are of hydrophobic character (Fig. S3). The hydrophobic tunnel entrances presumably facilitate access of the gaseous substrates. The highest occupied O₂ site (Oxy1), located in branch 1, has no obvious connection to the protein exterior and might serve as a substrate storage cavity. A similar storage cavity has been proposed by Montet et al. (26) for the O₂-sensitive hydrogenase from Desulfovibrio gigas, assuming that some of the hydrophobic cavities serve as gas reservoirs. They have suggested that the concentration of the gaseous substrate in the enzyme affects the enzyme activity. Comparable conclusions have been drawn from the MD simulation results from [NiFeSe] hydrogenase (27). Baltazar et al. (27) deduce from cavities with trapped H₂ molecules that they either hinder H₂ transport to the active site or promote access by storing gas molecules. Considering the low solubility of H₂ in water, the hydrophobic tunnel system might aid in elevating the H₂ concentration above the surrounding medium level. To further inspect a gas reservoir feature, MD simulations of simultaneous H₂ and O₂ transport in the corresponding ratio through the tunnel network, as well as amino acid exchanges within the proposed tunnel wall, might shed further light on this topic. Oxy3 is located within tunnel B near the bottleneck (smallest tunnel diameter) consisting of the amino acids A127_L, V131_L, and A202_L. Oxy4 is positioned close to the bottleneck of tunnel A and the catalytic center (Figs. 2 and 3). The bottleneck in tunnel A has a distance of <3 Å from the [NiFe] active site and consists of amino acids E27_L, R530_L, P596_L, and C597_L.

The observations based on the crystallographic structure were compared with MD simulations of MBH in aqueous solution, as described in further detail in Materials and Methods. The resulting 300-ns-long trajectories showing the diffusion of O₂ and H₂ molecules suggested a hopping-like movement of O₂ and H₂ molecules (examples for diffusion pathways of four individual O₂ molecules are shown in Movies S1–S4) through the main tunnels (Figs. 2 and 3 and Figs. S3, S5, and S6). Notably, diffusion outside the tunnel system and no permanent tunnel opening was detected in the simulations, which is in agreement with previous calculations performed on a structurally similar O₂-sensitive “standard” [NiFe] hydrogenase (21). Diffusion of gas molecules through the previously proposed water channel (19) was predicted to be unfavorable compared with the gas diffusion through tunnels A and B. O₂ diffusion (hopping process) through the gas tunnel network resulted in high O₂ densities at the Oxy1 and Oxy3 sites (molecule occupancies of >75%) (Fig. 3), which is in excellent agreement with the results of the soak-and-freeze experiment. Furthermore, O₂ molecules were detected in the vicinity of all other experimentally proposed locations, as well as within a 3-Å radius from the active site. Due to the side-chain flexibility at room temperature, the positions of the O₂ sites change during the course of the simulations, particularly in tunnel A. This impedes a direct comparison with the experimentally determined values. The assignment of these sites, however, was computationally predicted on the basis of residence times and site occupancies by the O₂ molecules (Fig. 3). Analysis of transition probabilities between the preferred O₂ locations showed that, except for Oxy2, all O₂ positions in the MBH were connected by the tunnel network, which was also predicted by CAVER 3.0 (Fig. 2). In addition, Oxy5 and Oxy6, located close to the proposed tunnel entries, showed the highest transition probabilities (28.6% and 23.2%, respectively) to the “bulk region” (solvent and undefined regions within the protein), making these positions potential entry sites for O₂ (Fig. 3B and Table 4). The probability for a transition from the bulk region to branch 1 (Oxy1 and Oxy7) was much lower compared with transitions from the bulk region to positions Oxy5 and Oxy6 at the potential entry points (Table 4). Deeply buried sites, such as Oxy3 and the branch 1 region, showed higher self-transition probabilities (Table 4) and occupancies (Fig. 3B) than regions close to the protein surface (Oxy5 and Oxy6), indicating longer residence times for O₂ at these sites. The only difference between MD and CAVER 3.0 predictions was that in the MD simulations, the observed O₂ densities in tunnel A were lower than in tunnel B (Fig. 3), while CAVER 3.0 predicted tunnel A as the main access pathway (Fig. 1). This finding may result from the increased flexibility of the amino acid side chains narrowing tunnel A in the simulations.

The spatial probability density of H₂ in the gas tunnel network was similar to that predicted for O₂ (Fig. S6). However, distinct differences in the diffusion properties of H₂ and O₂ were encountered. For the H₂ diffusion, tunnel A was strongly favored over tunnel B, as reflected by (i) the high transition probabilities between Oxy5 and Oxy4 (Table S1) and (ii) the increased occupancies of these sites (Fig. S6). Concomitantly, the low occupancy of Oxy3 (45.7% vs. 76.9% for H₂ and O₂, respectively) reflects a reduced diffusion of H₂ through tunnel B. Because of the smaller size and lower molecular weight of H₂ compared with O₂, higher H₂ densities were found in close vicinity (2.7-Å distance) to the [NiFe] center, as well as outside the MBH tunnel system in the protein matrix (Fig. S5).

To investigate whether the pressure treatment, which comes along with the soak-and-freeze method, had an impact on the overall structure of MBH and its gas tunnels, we compared the O₂-derivatized structure and a structure obtained after high-pressure (2,000 bar of helium) treatment of MBH crystals with the nonderivatized reference structure of MBH at ambient pressure (28). The crystal structures were aligned using the program PyMOL (29), and we obtained root mean square deviations (rmsds) of 0.043 Å and 0.195 Å when comparing the O₂-derivatized structure and the high-pressure structure with the reference MBH structure, respectively (Fig. S7). The greater differences with the high-pressure structure are also reflected by the B-factors (temperature factors). The B-factor comparison between the three structures was visualized using PyMOL in a range between 10 Ų and 50 Ų (Fig. 4). All O₂-derivatized MBH structures revealed no significant changes in the overall temperature factors upon pressurization compared with the REF datasets. Nevertheless, a further increase of the pressure to 2,000 bar led to a marked rise in the B-factors. The pressure-induced higher flexibility of amino acids even led to an unresolved disordered loop structure close to the entrance of tunnel A and in an enlargement of the surrounding area (Fig. 4). Furthermore, branch 1 of the hydrophobic tunnel is capped by a surface-located α-helix, which does not show an increase in flexibility through pressurization either at 70 bar or 2,000 bar. This supports the proposal of branch 1 being mainly a dead end and not a transient gas tunnel with a lockable lid. Notably, the [NiFe] active site, as well as the medial [3Fe4S] cluster and the distal [4Fe4S] cluster of the electron relay, maintained their spatial structure upon pressurization. In all three MBH structures, the Ni and Fe atoms have distances of 2.9 Å (Table 5), and they are bridged by a hydroxide ligand, which is entirely in line with the oxidation state of the catalytic center (18, 19). They also showed the typical open conformation of the proximal [4Fe3S] cluster with a bond length of ∼2.1 Å between iron 4 and the backbone nitrogen of cysteine 20 (Fig. S8), which is compatible with the as-isolated, superoxidized state of the enzyme (19). Despite the structural conformity, the [4Fe3S] cluster of the O₂-derivatized MBH showed two additional structural features compared with the reference structure. First, the carboxylate functional group of E76_S is shifted toward Fe4 of the cluster (Fig. 5A), and, second, Fe1 carries a hydroxyl ligand (Fig. 5B). While the distance of E76_S and Fe4 amounted to 3.4 Å and 3.2 Å in the reference and high-pressure structure, respectively, it was reduced to 2.6 Å in the O₂-derivatized MBH (Fig. 5A and Tables 2 and 5). The E76_S shift toward the proximal cluster was observed in all of the O₂-derivatized MBH structures (Tables 1 and 2). In the as-isolated structure of the membrane-bound O₂-tolerant [NiFe] hydrogenase 1 from Escherichia coli (EcHyd-1), the E76_S was found in a double conformation with distances of 4.9 Å and 2.4 Å of the glutamate-derived carboxylate O_Ɛ to Fe4 (30). Volbeda et al. (30) proposed that E76_S functions as a base for deprotonation of the carboxamido N of cysteine 20, which further enables the Fe4 to bind to the backbone of cysteine 20, and thus stabilizes the high-potential form of the [4Fe3S] cluster. A similar cluster stabilization was proposed for the Fe1-bound hydroxyl ligand in MBH. The presence of this ligand with an Fe-O distance of 1.8 Å between the hydroxyl oxygen and the Fe1 was confirmed in all of the O₂-derivatized MBH structures. The hydroxyl ligand has an average occupancy of ∼34%, with no obvious relation to the O₂ concentration (Table 1). In the REF datasets, however, the Fe1-bound hydroxyl ligand was not visible. In the previously published crystal structures of as-isolated, superoxidized MBH, the hydroxyl ligand occurred with high variation in occupancy, reaching from 0% in PDB ID code 4TTT, to 30% in PDB ID code 4IUC (B-factor: 9 Ų), and up to 88% in PDB ID code 4IUB (B-factor: 11 Ų) (19). At present, it remains unclear whether the Fe1-bound hydroxyl ligand and the E76_S shift are related to the 5+/4+ redox state transition of the [4Fe3S] cluster (19) or represent damage caused by high O₂ concentrations. Nevertheless, it is important to note that no remarkable damage due to high O₂ pressure and/or radiation was observed for the remaining metal cofactors in the MBH. Moreover, the two structural features might point to a dedicated function. In the course of MBH preparation, the enzyme was oxidized by ferricyanide and air. However, the occurrence of the OH⁻ ligand varied significantly (0–88%) in these preparations (19). Upon derivatization under pressurized pure O₂, the redox potential was presumably further raised, which is why we are able to see stable occupancies of the OH⁻ ligand and the E76_S shift in all corresponding MBH structures. This also leads to the question of whether the hydroxyl ligand at Fe1 of the proximal [4Fe3S] cluster arises from surrounding water molecules or from molecular oxygen. Previous resonance Raman experiments using ¹⁸O₂ labeling showed an isotopic shift for the Fe-OH stretching mode, leading to the conclusion of an O₂-derived ligand (19). Therefore, the additional ligand at Fe1 of the [4Fe3S] cluster could be part of an O₂-reducing mechanism, where His229, which is in hydrogen-bonding distance to the hydroxyl ligand, might play an important role in delivering protons produced in the H₂ oxidation reaction at the active site. A recent publication by Dance (31) describes a theoretical model of the communication between the catalytic center and the proximal cluster via conformational changes of residues terminating in this highly conserved His229. This histidine plays a crucial role in transmitting a proton to the opposite side of the proximal cluster for protonation of sulfur S3, followed by an opening of the Fe4-S3 bond. It has been stated that a close water molecule (4.5 Å away from Fe1) could be involved in an OH₂/OH⁻ turnover to provide the H⁺ for the separate proximal cluster proton cycle. This result is consistent with our structural findings for the proximal cluster in O₂-derivatized MBH. S3 protonation, however, does not lead to the Fe4-N20 bond according to Dance (31). Here, a second protonation/deprotonation cycle is necessary, which might be facilitated by Glu76. In our structural data, we can confirm an interaction of Glu76 with the proximal cluster under aerobic, oxygen-derivatized conditions. However, a sequence of protonation or different intermediates cannot be assigned in our structures and requires further experiments. It needs to be elucidated whether the OH⁻ ligand could also originate from the surrounding water network. In this case, the proton released from water dissociation might be transferred via His229 (Fig. 5) to the catalytic center to support O₂ reduction to water. Additionally, the OH⁻ ligand may stabilize the superoxidized state of the proximal cluster (19).

In summary, the technique of crystal derivatization (soak-and-freeze) enables unprecedented insight into the accessibility of the MBH tunnel network for gas molecules. Our data show that MBH contains several hydrophobic tunnels that can be traced via several O₂ molecules residing at defined sites. MD simulations further manifest the pathways used by the gas molecules traveling from the protein surface to the deeply buried active site. Furthermore, the MD simulations indicate that O₂ and the much smaller H₂ molecules likely use the same routes within the gas tunnel network. Additional gas diffusion studies might provide supplementary information on the migration behavior of H₂ and O₂ within the MBH gas tunnel network under ambient conditions. For example, a previous experimental study on [FeFe] hydrogenases revealed that high H₂ levels decreased the inhibition by O₂ (32). Our MD simulations already revealed differences in the distribution of O₂ and H₂. O₂ occupies the two transition sites of tunnel A to a much lesser extent than H₂ does. O₂ might be primarily directed toward branch 1 of gas tunnel B, which is located far away from the vulnerable [NiFe] active site. This indicates a mechanism that, at least to some degree, can separate the two gases H₂ and O₂ from each other, and would thus be another adaptation of O₂-tolerant hydrogenases toward prevention of damage by molecular oxygen. However, our data do not exclude accessibility of O₂ to the [NiFe] active site. Understanding the architecture of the tunnel network in MBH is crucial for future modifications of the enzyme functions through engineering of the tunnel characteristics. Site-directed amino acid exchanges allow changing the physical tunnel properties, such as the diameter and hydrophobicity. These modifications may have an influence on both the access of the substrate H₂ to the active site and the susceptibility of the enzyme toward O₂.

We thank Uwe Müller, Manfred Weiss, and the scientific staff of the BESSY-MX (Macromolecular X-ray Crystallography)/Helmholtz Zentrum Berlin für Materialien und Energie at beamlines BL14.1, BL14.2, and BL14.3 operated by the Joint Berlin MX-Laboratory at the BESSY II electron storage ring (Berlin-Adlershof, Germany) as well as the scientific staff of the ESRF (Grenoble) at beamlines ID30A-3, ID30B, ID23-1, ID23-2, and ID29, where the data were collected, for continuous support. We also thank the “Norddeutscher Verbund für Hoch-und Höchstleistungsrechnen” for providing HPC resources. This work was supported by grants from the ESRF (to G.G., D.v.S., P.v.d.L., A.R., and P.C.), Deutsche Forschungsgemeinschaft (DFG; Grants SFB740-B6 and SFB1078-B6 to P.S.), and DFG Cluster of Excellence “Unifying Concepts in Catalysis” (Research Field D3/E3-1 to J.K., A.S., S.F., T.U., M.A.M., O.L., and P.S.).