The structure of HydABC
The heterologous production of apo-HydABC in Escherichia coli was described recently (12). Here, we have used this heterologously expressed HydABC to prepare the cryo-EM grids. Following the grid imaging, data collection, and processing, we obtained a 2.3 Å resolution map when D2 symmetry was enforced (fig. S1, fig. S2, A). Into this, an atomic model of HydABC was constructed, starting with a homology model based on homologous subunits in bacterial complex I (12, 13), together with ab initio model building in regions of the highest resolution. Initially, the last 91 and 61 C-terminal (CT) residues of HydA and HydB, respectively, could not be built as they were not present in the homology model and had a low resolution in the map, indicating regions of high heterogeneity (explored later).
The processed cryo-EM map shows that the heterotrimeric HydABC forms a tetrameric complex, Hyd(ABC)4 of HydABC heterotrimer units (protomers). Oligomerization of HydABC occurs through interactions between four HydA subunits in the core of the complex (Fig. 1, A see also Movie S1). Each HydA has extensive interactions with one adjacent HydA chain (buried surface area of 2,280 Å2), and minor interactions with another HydA chain (780 Å2) (Fig. 1, A). HydB is tightly bound to a single HydA (buried surface area of 1,232 Å2, table S1) but with minor interactions between HydB of one heterotrimer and HydA and HydB in another heterotrimer. HydB and HydC extend outward from the core and form the four lobes clearly visible in the 2D class averages (fig. S3). The HydA core is the best resolved part of the map, consistent with the core being rigid and homogenous (fig. S4).
Based on the density map, each HydABC protomer appears to contain nine redox cofactors including five [4Fe-4S] clusters (one of which forms the [4Fe-4S] subcluster of the H-cluster), three [2Fe-2S] clusters, and one FMN. However, based on published Fe quantitation as well as published sequence analysis predictions we expect a total of seven [4Fe-4S] clusters (including the subcluster of the H-cluster) and four [2Fe-2S] clusters in each HydABC protomer (14, 15). According to sequence predictions, these missing clusters should be located in the less well resolved CT regions of the HydA and HydB subunits (discussed below) (14). Interestingly, a high-density site, likely a monometallic center, is found in the resolvable part of the HydB-CT domain. Inductively-coupled plasma mass spectrometry on the separately produced and purified HydB subunit identified 0.99 ± 0.43 Zn/protein and ≈14.2 ± 1.5 Fe/protein. As the observed Fe-content matches with the estimated Fe-content of HydB, which is expected to contain three [4Fe-4S] clusters and one [2Fe-2S] cluster (14 Fe/protein), these results allow us to assign the metal center as zinc (Zn2+). This is further supported by the identities of the ligating residues: three cysteines and one histidine in a tetrahedral coordination geometry (fig. S5, C) (16).
Cofactor arrangement in HydABC
Electron transfer chains, often connecting distant active sites, are composed of redox-active cofactors usually less than 14 Å apart to allow sufficiently fast electron tunnelling through the protein dielectric to sustain catalysis (17). In each HydABC heterotrimer, the spatially distant H-clusters and FMN centers are electrically connected via a chain of four FeS clusters (A1, A2, A3, and B2, see Fig. 1, D for cluster nomenclature). The edge-to-edge distances between all these clusters are < 15 Å and within a distance for electron transfer at physiologically relevant rates (Fig. 1, D). Among the three remaining FeS clusters, the [4Fe-4S] cluster from HydA (A4) lies at the interface of the two tightly interacting HydA chains, and the two [2Fe-2S] clusters from HydC (C1) and HydB (B1) subunits lie in the vicinity, but on the opposite side, of the FMN toward a dangling helix at the N-terminus of HydB (fig. S6). This helix has locally lower resolution indicating some structural heterogeneity. Similar structural features have been suggested to function as a “fishing rod” for catching ferredoxin in a “fly casting” mechanism in cyanobacterial complex I (18) and ferredoxin NADP+ reductase (19, 20). Furthermore, a formate dehydrogenase enzyme that is closely related to HydABC but which does not bind ferredoxin, is specifically missing this helix (21). Thus, we suggest that the N-terminus of HydB may be the region where ferredoxin binds and transfers electrons to the B1 cluster. Electrons from ferredoxin would then be transferred from B1 to C1, and possibly to the FMN. However, the distance between C1 and FMN is ~16 Å, which is at the theoretical limit for fast electron transfer (17).
Within the Hyd(ABC)4 complex, there appear to be two redox networks, each composed of two electrically connected HydABC protomers, separated by at least 50 Å and held together by extensive HydA-HydA interactions (Fig. 1, B and C). The large distance between each network indicates there is no possibility for electrons to be exchanged and that they probably function independently (Fig. 1, C). The two tightly interacting HydABC protomers within the Hyd(ABC)2 unit are electrically connected through the His-ligated [4Fe-4S] cluster (A4) in HydA (Fig. 1, B), part of the so-called Y-junction of iron-sulfur clusters (22). This junction is well conserved in a wide number of structurally related enzymes, but its function is unknown. In HydABC it is clear that the Y-junction connects the NADH and ferredoxin oxidation sites to the hydrogenase active site and to the neighboring protomer. The two A4 clusters are separated by 9.0 Å and have the possibility to allow overflow of electrons from one protomer to the other. An electrical connection between two identical protomers has already been observed in cytochrome bc1 (23), called an electronic ‘bus-bar’, which is speculated to have a number of possible roles such as allowing the physiological function of the protein even after operational damage of one of the two protomers.
Structural comparison of HydABC with homologous proteins
The spatial arrangement of subunits HydA, B and C in the HydABC protomer is similar to that of subunits Nqo3, Nqo1, and Nqo2, respectively, in the NADH-oxidation (N) module of Thermus thermophilus (Tt) respiratory complex I (fig. S7). This comparison is useful because complex I is structurally well-characterised, but does not oxidize ferredoxin or carry out electron-bifurcation. Therefore, structural differences between the subunits of complex I an HydABC may reveal important insight into the mechanism of electron transfer in the latter. The individual subunits are structurally highly similiar, with the highest similarity between HydB and Nqo1 (rmsd 1.040 Å) (24), followed by HydC and Nqo2 (rmsd 1.152 Å) and the lowest similarity between HydA and Nqo3 (rmsd 1.294 Å) (Fig. 2, A). The remarkable structural similarities between HydB and Nqo1 subunits agree with their common evolutionary origins (25), and suggest that NADH oxidation follows a similar mechanism in both enzymes (Fig. 2, B). The structural differences between Nqo3 and HydA likely reflect the fact that the latter accommodates the hydrogenase H-cluster and facilitates oligomerization of the Hyd(ABC)4 complex.
The structural similarities between HydABC and Tt respiratory complex I are also reflected by the FeS clusters positioning that is in excellent agreement in these two proteins (Fig. 2, C). However, in contrast to the Tt complex I, the HydABC protomers contain five additional FeS clusters. One of these additional clusters is a [4Fe-4S] cluster (A3) that electrically connects the [4Fe-4S] subcluster of the H-cluster (analogous to the cluster N7 in Tt complex I) with the rest of the electron transfer network. Another additional cluster is a [2Fe-2S] cofactor in HydB (B1) that is electrically connected to the [2Fe-2S] cluster in HydC (C1, analogous to N1a in Tt complex I); due to this connection and the proximity of HydC to the “bridge” domains (discussed later) it is likely that the [2Fe-2S] cluster in HydC has an important role in the mechanism of electron bifurcation, this is in contrast to its analogous N1a cluster in complex I, the role of which is unclear but is certainly not part of the main catalytic electron transport pathway (26, 27).
The HydA subunit has close structural homology (35% sequence identity) to the well-characterized monomeric non-bifurcating [FeFe] hydrogenase from Clostridium pasteurianum, CpI. In contrast to electron bifurcating [FeFe] hydrogenases, non-bifurcating [FeFe] hydrogenases use a single redox partner, typically ferredoxin. Aligning the two enzymes shows high similarity (rmsd 1.119 Å) and excellent conservation of the FeS clusters, including the A4 cluster, which connects neighboring HydA subunits in HydABC (Fig. 3). However, in CpI, for which ferredoxin is the only redox partner, the cluster homologous to A4 is thought to lead to the ferredoxin binding site (28). The multimerization of HydA blocks this site, so the two enzymes must have different ferredoxin binding sites. This re-arrangement is an example of how closely related systems may have different electron-transfer pathways formed by different multimerization of their subunits.
A bridging domain formed by the flexible C-termini of the HydA and HydB subunits
The core of the tetrameric HydABC complex is very well resolved, reaching a local resolution of 2.2 Å. However, the lobes formed from HydA and HydB subunits have substantially lower local resolution (~3 Å), due to increased heterogeneity (fig. S4) and low intensity, blurred map density was observed between the lobes of electrically connected HydABC protomers (fig. S8, A). To investigate the blurred regions, symmetry expansion followed by classification was explored to separate the different conformations into classes. Initial attempts to use D2 symmetry, to match the core, resulted in maps no better than before, however, using C2 symmetry revealed two classes with bridging density between the HydB lobes (Fig. 4, A) with local resolution similar to the lobes formed from HydA and HydB (Fig. 4, B). This bridging density breaks the rotational symmetry between the protomers in the Hyd(ABC)2 unit, explaining why D2 symmetry expansion was ineffective. The two classes correspond to the bridge domain being formed between different HydB lobes: when rotated by 180°, the bridges are identical (Fig. 4, A and C). Despite extensive attempts, we were unable to find a class with both bridges in the closed conformation. The observation that both bridges cannot close simultaneously suggests that these behave as reciprocating elements. A similar observation was made for the Rieske domains in the bifurcating bc1 complex (30).
To further explore the particles without a bridge a further classification was used (Fig. S8, B). It was possible to obtain a low-resolution map of a class where the HydB CT domain was found in an ‘open’ conformation (Fig. 4, D). The movement of the HydB C-terminal domain between the bridge-open and bridge-closed classes is shown in Fig. 4, E and Movie S2.
In the bridge-containing structure, the two C-terminal [4Fe-4S] clusters (named B3 and B4, Fig. 4, F) of HydB are close enough to exchange electrons with each other but are too far from the next nearest FeS clusters, such as cluster C1 (≈ 35 Å away) or cluster A5 (≈ 32 Å away). Furthermore, cluster A5 is completely electrically isolated with all the nearest clusters being >30 Å away. Thus, unless the HydA and HydB bridge domains undergo substantial conformational changes, the FeS clusters A5, B3 and B4 cannot participate in electronic exchange with the rest of the enzyme.
The bridge structure is particularly interesting as it appears that the C-terminal cysteine residues of HydB responsible for coordinating [4Fe-4S] clusters in the bridge are conserved in all biochemically characterized electron-bifurcating [FeFe] hydrogenases (31, 32). However, they all lack the analogous part of the bridge domain in HydA, which contains the A5 cluster.