Crystallographic and kinetic analyses of the FdsBG subcomplex of the cytosolic formate dehydrogenase FdsABG from Cupriavidus necator

Formate oxidation to carbon dioxide is a key reaction in one-carbon compound metabolism, and its reverse reaction represents the first step in carbon assimilation in the acetogenic and methanogenic branches of many anaerobic organisms. The molybdenum-containing dehydrogenase FdsABG is a soluble NAD+-dependent formate dehydrogenase and a member of the NADH dehydrogenase superfamily. Here, we present the first structure of the FdsBG subcomplex of the cytosolic FdsABG formate dehydrogenase from the hydrogen-oxidizing bacterium Cupriavidus necator H16 both with and without bound NADH. The structures revealed that the two iron-sulfur clusters, Fe4S4 in FdsB and Fe2S2 in FdsG, are closer to the FMN than they are in other NADH dehydrogenases. Rapid kinetic studies and EPR measurements of rapid freeze-quenched samples of the NADH reduction of FdsBG identified a neutral flavin semiquinone, FMNH•, not previously observed to participate in NADH-mediated reduction of the FdsABG holoenzyme. We found that this semiquinone forms through the transfer of one electron from the fully reduced FMNH−, initially formed via NADH-mediated reduction, to the Fe2S2 cluster. This Fe2S2 cluster is not part of the on-path chain of iron-sulfur clusters connecting the FMN of FdsB with the active-site molybdenum center of FdsA. According to the NADH-bound structure, the nicotinamide ring stacks onto the re-face of the FMN. However, NADH binding significantly reduced the electron density for the isoalloxazine ring of FMN and induced a conformational change in residues of the FMN-binding pocket that display peptide-bond flipping upon NAD+ binding in proper NADH dehydrogenases.

The oxidation of formate to carbon dioxide is a key reaction in the metabolism of one-carbon compounds. In many anaerobic organisms, the reverse reaction represents the first step in carbon assimilation in both the acetogenic and methanogenic branches of the Wood-Ljungdahl pathway. The genome of Cupriavidus necator H16 (formerly known as Ralstonia eutropha) (1, 2) encodes four putative formate dehydrogenases, of which two are soluble NAD ϩ -dependent formate dehydrogenases. One of them is the molybdenum-containing dehydrogenase, FdsABG.
Formate dehydrogenase FdsABG forms a dimer of heterotrimers, (␣␤␥) 2 , with the 105-kDa FdsA subunit containing the active site molybdenum center and five iron-sulfur clusters, one Fe 2 S 2 and four Fe 4 S 4 clusters; the 55-kDa FdsB containing an Fe 4 S 4 cluster, FMN, and a binding site for NADH/NAD ϩ ; and finally, the 19-kDa FdsG subunit containing one Fe 2 S 2 cluster. Although the physiological function of FdsABG is to oxidize formate to CO 2 by reducing NAD ϩ to NADH (3)(4)(5)(6), the enzyme has been shown to be fully capable of catalyzing the reverse reaction, the reduction of CO 2 to formate, using NADH as an electron source with steady-state kinetic parameters that conform to the required Haldane relationship (7).
Formate dehydrogenase FdsABG is a member of the NADH dehydrogenase superfamily, which is one of the most broadly distributed families of enzymes (5,(8)(9)(10). Members of this family typically couple electron transfer to generate a transmembrane proton or sodium gradient. The water solubility of FdsABG formate dehydrogenase greatly simplifies its biochemical study. Whereas the protein does not pump cations, it does contain redox-active centers that have been lost in other members of the superfamily over the course of evolution, including a bis(enedithiolate) molybdenum center that is the site of the fully reversible oxidation of formate to CO 2 . As such, FdsABG is thought to represent, along with a separate hydrogenase and membrane-integral ion channel, an evolutionarily primitive, functional module of NADH dehydrogenases.
The FdsA, FdsB, and FdsG subunits have strong sequence homology to corresponding subunits in the cytosol-exposed arm of the membrane-integral and proton-pumping NADH dehydrogenase (3)(4)(5)(6). FdsA is cognate to the Nqo3 subunit of the Thermus thermophilus NADH dehydrogenase (11,12), despite the absence of the molybdenum center and of one of the Fe 4 S 4 clusters (5). The close homology between FdsA and Nqo3 extends to the presence of a histidine ligand to one of the Fe 4 S 4 clusters. The C terminus of FdsA, containing the molybdenum center, also has an ϳ60% sequence similarity to the structurally characterized FdhF formate dehydrogenase of Escherichia coli, with cysteine, Cys-378, equivalent to the molybdenum-coordinating selenocysteine, Sec-140, in FdhF (13). This C-terminal domain is also present in Nqo3, although the molybdenum center has been lost over the course of evolution.
The FdsB subunit has ϳ45% sequence identity to the Nqo1 subunit of the T. thermophilus NADH dehydrogenase and, like Nqo1, contains an Fe 4 S 4 cluster, FMN, and a binding site for NADH/ NAD ϩ . As demonstrated here, it also possesses an N-terminal thioredoxin-like domain that lacks the typical iron-sulfur cluster and the disulfide bond of thioredoxin. The FdsG subunit has ϳ34% sequence identity to the Nqo2 subunit of the T. thermophilus NADH dehydrogenase and possesses an Fe 2 S 2 cluster.
Under physiological conditions, electrons enter the FdsABG holoenzyme at the molybdenum center in the reductive half of the catalytic cycle and depart at the FMN in the oxidative halfreaction, with electron transfer between the two sites (thought to be separated by ϳ55 Å (9)) mediated by intervening ironsulfur clusters.
Because FdsB-and FdsG-like domains are found in all members of the NADH dehydrogenase superfamily (12,14), we have undertaken a study of the FdsBG subcomplex of C. necator formate dehydrogenase, with the aim of putting the oxidative half-reaction of the enzyme in a structural context. Here, we present the structure of FdsBG complex and characterize its kinetic properties.

Results
Previously, the FdsBG homologous, stand-alone complex of NADH dehydrogenase from Paracoccus denitrificans has been characterized and shown to be capable of NADH oxidation (15). We therefore set out to characterize the electron transfer reaction of the FdsBG complex from C. necator.
The recombinant expression of C. necator FdsABG holoenzyme in E. coli produces not only monomeric and dimeric holoenzyme, but also significant amounts of FdsBG complex (16). A similar observation was reported for the expression of Rhodobacter capsulatus FdsABG holoenzyme (17).
To streamline the purification of FdsBG complex, we redesigned the FdsABG holoenzyme expression system to produce only FdsBG complex with two different N-terminal tags on FdsG, His 6 -linker-His 6 and Twin-Strep tag. The resulting constructs produce FdsBG complexes that are substantially identical in their NADH/NAD ϩ redox activity and their ability to crystallize. A detailed description of the two constructs and the purification of the overexpressed FdsBG complexes is provided under "Experimental procedures."

Characterization of the two iron-sulfur clusters of the FdsBG complex
We first examined the EPR signals attributable to the complex's two iron-sulfur centers. Upon extended incubation of FdsBG complex with dithionite, two signals are observed (Fig.  1A, black trace).
The first signal is seen at liquid nitrogen temperatures with g-values of g 1,2,3 ϭ 2.000, 1.948, and 1.920 and line widths of 1.4, 1.7, and 1.6 mT, 3 respectively. These values are in good agree-ment with those described previously for the cluster designated as Fe/S 1 for the FdsABG holoenzyme (Fig. 1B, blue dashed trace) (6). This EPR signal is observed well into the 200 K regime.
The second signal is detected below 20 K with g-values of g 1,2,3 ϭ 2.039, 1.955, and 1.891 and line widths of 4.5, 1.4, and 5.3 mT, respectively (Fig. 1B, red dashed trace). This cluster has not been described previously in the holoenzyme (6), and we now designate it as Fe/S 5 .
In the homologous FdsBG complexes from NADH dehydrogenase (Aquifex aeolicus NuoEF or T. thermophilus Nqo1⅐2), only EPR signals of the Fe 2 S 2 clusters are detectable above 77 K, whereas those for the Fe 4 S 4 clusters appear only below 50 K (18). The g-values g 1,2,3 ϭ 2.004, 1.945, and 1.917 for the N1a Fe 2 S 2 cluster of the 24-kDa subunit of respiratory complex I from bovine mitochondria (the homolog to FdsG) are in excellent agreement with the Fe/S 1 signal. Accordingly, we assign the Fe/S 1 signal to the Fe 2 S 2 cluster of FdsG (19). The signal Figure 1. EPR spectra of the iron-sulfur clusters of the FdsBG complex. A, observed iron-sulfur EPR spectrum (black trace) and simulated composite spectrum (red trace) of the dithionite-reduced FdsBG complex, collected at 9 K with modulation amplitude of 8 gauss and microwave power of 2 microwatts. The sample was prepared by incubation of 125 M of FdsBG complex in 100 mM KPO 4 , pH 7.0, with 2 mM buffered sodium dithionite under anaerobic conditions for 1 h at room temperature prior to freezing. B, the individual component spectra resulting from the simulation of the composite spectrum in A: the spectrum corresponding to the previously assigned Fe/S 1 (blue dashed trace (6)) and an additional Fe/S 5 signal (red dashed trace) resolved only in the FdsBG complex. C, superposition and graphical alignment of the spectrum presented in A (red trace) with the spectrum of dithionite-reduced FdsABG collected at 20 K (blue trace, expanded to match the amplitude of the g 1 feature in the Fe/S 1 component of the FdsBG spectrum; reprinted with permission (6)). Dashed lines mark the location of g 1 and g 3 features of the Fe/S 3 component of the FdsABG holoenzyme spectrum.

Characterization of FdsBG complex
assigned to the N3 Fe 4 S 4 cluster in the NADH dehydrogenase system, with g-values of g 1,2,3 ϭ 2.037, 1.945, and 1.852, is also in good agreement with the Fe/S 5 signal of the FdsBG complex. Therefore, we assign the Fe/S 5 signal to the Fe 4 S 4 cluster of FdsB.
We had initially assigned the Fe/S 1 signal to the His-coordinated Fe 4 S 4 cluster contained in the FdsA subunit of the FdsABG holoenzyme (6), but that is clearly not the case as the signal is observed in the FdsBG complex. We note that the Fe/S 5 signal has very similar g-values to that reported for the previously observed Fe/S 3 , although the Fe/S 5 line widths are considerably broader (see the g 1 and g 3 features of red and blue EPR spectra, centered around the dashed lines in Fig. 1C). However, the Fe/S 3 cluster has been shown to be magnetically coupled to the molybdenum center and must accordingly arise from the Fe 4 S 4 cluster proximal to the molybdenum center in FdsA. Therefore, the Fe/S 5 signal must be a newly identified signal.
The assignment of the Fe/S 5 signal to the Fe 4 S 4 cluster of the FdsB subunit is strengthened by the below described structure of the FdsBG complex and by our previously published structural model of the FdsABG holoenzyme (9). This model places the Fe 4 S 4 cluster in the proximity of FdsA in the intact holoen-zyme and in the electron path between the molybdenum and FMN. It is likely that the large line widths, in conjunction with the relatively low level of reduction and overlapping g-values with Fe/S 3 , obscured its detection in the earlier work with the holoenzyme. It is also possible that the presence of additional, higher-potential clusters in the FdsA subunit prevents the reduction of the Fe/S 5 cluster in the holoenzyme.

Crystal structure of FdsBG complex: Overview
We further characterized the FdsBG complex by determining its structure by X-ray crystallography to 2.3 Å resolution (Fig. 2, B-E). The final phased electron density allowed us to build nearly all residues of the FdsBG heterodimer (for data and refinement statistics see Table S1). The heterodimer has dimensions of 82 ϫ 64 ϫ 62 Å, and the two heterodimers of the asymmetric unit are related by a noncrystallographic rotation of ϳ179°( Fig. 2A). The axis of rotation passes through the interface formed by the N-terminal thioredoxin-like domains of the two FdsB molecules of the asymmetric unit (residues 3 B -104 B , where the superscript B indicates that the residue belongs to FdsB). The unusual relationship of the two FdsBG complexes in the asymmetric unit notwithstanding, their structures are nearly identical, with an RMSD of 0.35 Å for

Characterization of FdsBG complex
all the atoms of the 661 residues of the core of the 735-amino acid FdsBG complex. Each FdsBG complex contains two iron-sulfur clusters (one Fe 2 S 2 and one Fe 4 S 4 cluster), one FMN, and one K ϩ ion. The final model also contains 498 water molecules.
For the detailed analysis of the structural differences between FdsBG complex and its homologous subcomplex from NADH dehydrogenase, we used the recently published, high-resolution structure from A. aeolicus, NuoEF complex (20). The NuoE subunit is homologous to the Nqo1 and FdsG subunits, and the NuoF subunit is homologous to the Nqo2 and FdsB subunits (Fig. 3, A and B).

Structure of the FdsB subunit of the FdsBG complex
Except for the N-terminal thioredoxin-like domain (residues 3 B -104 B ), the FdsB subunit structure is highly similar to that of the NuoF subunit of the A. aeolicus NADH dehydrogenase (RMSD of 1.48 Å for 394 C␣ atoms). Like NuoF, FdsB contains a Rossmann-like fold (residues 138 B -332 B , also classified as Complex I_51K domain (21)), a ubiquitin-like domain (residues 338 B -423 B ), and a four-helical bundle (residues 426 B -511 B , NADH_4Fe-4S domain). The Rossmann-like fold contains the FMN-binding site, and the C-terminal four-helical bundle contains the Fe 4 S 4 cluster (22) (Fig. 3A).
In the Rossmann-like domain, the main structural differences between the FdsB and NuoF subunits are in loops that are solvent-exposed. The only loop with a structural difference that is buried in the complex is positioned between the bound FMN and the Fe 2 S 2 cluster of the FdsG subunit (183 B -190 B loop), thus enabling the Fe 2 S 2 cluster to be ϳ0.2 Å closer to the FMN in the FdsBG complex than in the NuoEF complex (Fig. 4A).
The ubiquitin-like domain of FdsB is tilted toward the FMN bound in the Rossmann-like domain. This tilt toward the FMN propagates into the four-helical bundle domain, resulting in an ϳ0.2-Å distance closer between the FMN and the Fe 4 S 4 cluster compared with that seen in the NuoEF complex (Fig. 4B). The Fe 4 S 4 cluster is located in a hydrophobic environment close to the surface and is coordinated by Cys-443 B , Cys-446 B , Cys-449 B , and Cys-489 B (Fig. 2E).

Characterization of FdsBG complex
The N-terminal thioredoxin-like domain of FdsB is structurally similar to the C-terminal thioredoxin-like domain of FdsG (RMSD 1.97 Å for 71 C␣ atoms of the core of the domain). However, unlike the domain in FdsG, that in FdsB has lost its capacity to bind an Fe 2 S 2 cluster due to mutation of the corresponding cysteines to proline, alanine, serine, and phenylalanine, respectively (Pro-10 B , Ala-15 B , Ser-45 B , and Phe-49 B ).

Structure of the FdsG subunit of the FdsBG complex
The structure of the FdsG subunit is also highly similar to the NuoE subunit of T. thermophilus NADH dehydrogenase. Like NuoE, FdsG consists of an N-terminal four-helical bundle (residues 29 G -74 G ) and a C-terminal thioredoxin-like domain (residues 79 G -159 G ). The domains are separated by a 4-aminoacid-long linker, around which the two domains are hinged by ϳ26°compared with NuoE. This domain arrangement is stabilized by a K ϩ ion, which is held in place by residues from both domains (His-71 G , His-112 G , and Glu-123 G ; Fig. 4C).
Although the N-terminal domains of FdsG and NuoE consist of a four-helical bundle, the first helix of FdsG packs in parallel to the second and third helix, whereas in NuoE it runs across the interface formed by these two helices (Fig. 4C). The C-terminal thioredoxin-like domain of FdsG is structurally similar to the N-terminal thioredoxin-like domain of FdsB but contains a spinach ferredoxin-like Fe 2 S 2 cluster (22). The Fe 2 S 2 cluster is located in a hydrophobic environment close to the surface and is coordinated by Cys-86 G , Cys-91 G , Cys-127 G , and Cys-131 G (Figs. 2D and 3B). Compared with Nqo2, FdsG's and NuoE's thioredoxin-like domains are truncated at their C termini and are missing a disulfide bond (Fig. 3B).

Characterization of FdsBG complex The interface between FdsB and FdsG subunits in the FdsBG complex
Two helices of FdsG's four-helical bundle interact with the surface of the Rossmann-like domain of FdsB, similar to the interactions seen in the NuoEF interface (Fig. 4D). The C-ter- With an edge-to-edge distance of nearly 21 Å, direct electron transfer between the Fe 2 S 2 cluster and Fe 4 S 4 cluster is unlikely to be kinetically significant. However, both clusters are within 12 Å of the isoalloxazine ring of FMN (Fig. 4E), which evidently mediates electron equilibration between the two iron-sulfur clusters. Whereas the Fe 4 S 4 cluster of FdsB is clearly a part of the electron transport chain between the molybdenum center of FdsA and the FMN of FdsB, the Fe 2 S 2 cluster of FdsG lies off-path and appears to function as a temporary repository for electrons in the course of electron egress out of the flavin (see "Discussion").

The FMN-binding site of the FdsB subunit
The FMN is bound by a network of interactions in an extended, solvent-accessible cavity that also accommodates NAD ϩ /NADH. The isoalloxazine ring of FMN is within 12 Å of each iron-sulfur cluster. The C8-methyl of the dimethylbenzene moiety of the flavin points toward the Fe 4 S 4 cluster, and the N5-containing edge faces the Fe 2 S 2 cluster, whereas the ribityl phosphate tail extends away from both iron-sulfur clusters (Fig. 4E).
The direct and water-mediated interactions between the FMN isoalloxazine ring and the ribityl phosphate tail with FdsB's Rossmann-like fold are conserved in the high-resolution structures of other members of the NADH oxidoreductase family (14, 20) (Fig. 5, A-C). (For a detailed description and the 2D representation of these interactions, see the supporting material and Fig. S1, respectively. For comparison, we also provide the 2D representation of the FMN interactions with the NADH dehydrogenase, NuoE, and the NAD ϩ -reducing hydrogenase, HoxF, as Figs. S2 and S3, respectively. These 2D representations of the interactions were prepared with LigPlotϩ version 2.1 (23)).

Biochemical characterization of the bound FMN
In the absence of the FdsA subunit, which contains five ironsulfur clusters and a molybdenum active site, the UV-visible spectrum of the isolated FdsBG complex is dominated by the spectral features of the FMN centered around 350 and 450 nm (Fig. 6A, red spectrum). In addition, a feature around 550 nm, attributed to the Fe 4 S 4 and Fe 2 S 2 clusters of the FdsBG complex, is significantly more pronounced due to the lack of interference of the additional iron-sulfur clusters present in the FdsABG holoenzyme (6). Removal of the FMN via TCA precipitation indicates that the FdsBG complex is ϳ50% saturated with FMN. This level of FMN saturation is comparable with the level of flavin saturation in the isolated FdsABG holoenzyme. 4 This incomplete FMN saturation in FdsBG suggests that we are visualizing in our crystal structures an average of at least two kinds of FdsBG, one with and one without FMN-bound, deflavo FdsBG.
A reductive titration with sodium dithionite results in a systematic bleaching of the absorbance throughout the UV-visible region (Fig. 6A, blue spectrum). At the end of the titration, the bleaching in the 550-nm region is incomplete, suggesting that the cluster principally responsible for the residual 550-nm absorption is the Fe 4 S 4 cluster that is not fully reduced by dithionite. A plot of the fractional absorbance changes at 550 nm against that at 450 nm provides information about the relative order of reduction during titration. The changes at 550 and at 450 nm are dominated by the reduction of the iron-sulfur clusters and FMN, respectively. A concurrent reduction of iron-sulfur clusters and FMN will result in a diagonal in this plot. A deflection of the diagonal to the top left suggests an earlier reduction of iron-sulfur clusters, whereas deflection to the bottom right suggests an earlier reduction of FMN (24). The inset of Fig. 6B shows a deflection to the top left from the diagonal, indicating that at least one iron-sulfur cluster is reduced sooner than the FMN during titration with dithionite. This suggests that the FMN has a lower reduction potential relative to at least one of the iron-sulfur clusters.

Characterization of FdsBG complex NADH binding to the FdsB subunit: Overview
To establish the NADH/NAD ϩ -binding site of the FdsBG complex, we solved the structure of FdsBG with bound NADH. To this end, preformed and stabilized FdsBG crystals were soaked with NADH at a final concentration of 10 mM in stabilization solution for 2 h, before flash freezing in liquid nitrogen. The crystal structure was solved to 2.0 Å by molecular replacement using the above described structure of the FdsBG complex as the initial search model. (See Table S1 for diffraction and refinement statistics.) Electron density for the adenosine diphosphate moiety of NADH is visible in the unbiased (F o Ϫ F c )-map of both FdsBG complexes of the asymmetric unit (Fig.  7A, green mesh). The overall protein structures of the FdsBG complexes are virtually identical in the presence and absence of NADH (RMSD 0.19 Å for the 1435 C␣ atoms shared between the FdsBG complex on its own and NADH-bound structures).
As in the FdsBG structure by itself, the two NADH-bound FdsBG complexes of the asymmetric unit are highly similar (RMSD 0.34 Å for the 660 C␣ atoms that the two FdsBG complexes share). However, the density for the isoalloxazine ring of FMN in each of two FdsBG complexes of the asymmetric unit is distinct. In one FdsBG complex, most of the FMN has significantly lower electron density than its terminal phosphate (Fig.  7A, light blue versus blue meshes). Due to our crystallization conditions, we speculate that the terminal phosphate is actually a sulfate bound to the fraction of FdsBG with no FMN bound.

FMN binding to FdsBG in the presence of NADH
Comparing the FMN-binding pocket of both FdsBG complexes in the asymmetric unit suggests that the peptidyl bond between Asp-184 B and Glu-185 B is pointing toward the O4 carbonyl oxygen of the isoalloxazine ring of FMN, in particular when less FMN is bound. The refinement of the FdsBG structure without FMN places the carbonyl oxygen of the Asp-184 B -Glu-185 B peptide bond so far into the FMN-binding pocket that it would encroach on the O4 of the isoalloxazine ring (Fig.  7B). This suggests that the change of the peptide bond position correlates with the loss of FMN.

NADH/NAD ؉ -binding site on FdsBG complex: The adenosine diphosphate moiety
More electron density is visible for the adenosine diphosphate of NADH than for the nicotinamide portion. This observation implies that the nicotinamide moiety of NADH contributes less to NADH affinity to FdsBG.
The ␤-phosphate of NADH is held in place by a salt bridge to FdsB and by two hydrogen bonds to the O2Ј and O4Ј hydroxyl groups of FMNЈs ribityl moiety (Fig. 7D). The adenosine base of NADH binds into a hydrophobic pocket. The O2Ј hydroxyl and O3Ј hydroxyl adenosine are within hydrogen bond distance of the carboxyl group of Glu-275 B . This places a negative charge adjacent to the O 2 Ј, ensuring FdsBG's specificity for NAD ϩ / NADH over NADP ϩ /NADPH (25). (A more detailed description of the ADP-binding interactions with FdsB is provided in the supporting material.) The specificity for the adenine base is brought about by the hydrogen bond from the side-chain hydroxyl group of Thr-168 B , which is specific to FdsB. In the NADH dehydrogenases and NAD ϩ -reducing hydrogenases, the threonine is replaced by either phenylalanine or leucine. This hydrogen bond interaction between the adenine base and threonine contributes to the more than 0.5-Å distant binding of the adenosine diphosphate relative to the FMN isoalloxazine ring compared with its binding in NuoF (Fig. 7C).

Binding of the nicotinamide ring of NADH relative to the isoalloxazine ring of FMN
Although only weak electron density is evident for the nicotinamide ring of NADH in the structure of both FdsBG complexes of the asymmetric unit, the direction of the two phosphates visible in the electron density suggests that the nicotinamide is projected toward the solvent-exposed re-face of the isoalloxazine ring of FMN. We therefore suggest that the nicotinamide moiety of NAD ϩ /NADH stacks onto the re-face of the isoalloxazine ring during the actual electron transfer step, similar to what is observed in the NuoEF complex with bound NAD ϩ and NADH (Fig. 7C) (20). However, the more distant binding of the adenosine relative to the isoalloxazine ring suggests that the nicotinamide ring will also be displaced over the

Characterization of FdsBG complex
isoalloxazine ring during the electron transfer reaction as compared with that in NuoF.

Rapid-reaction kinetics of the electron transfer reaction of the FdsBG complex
Due to the high velocity of the electron transfer reaction, we performed the rapid-reaction kinetics of FdsBG reduction by NADH at 5°C. Although this is the reverse of the physiological reaction for the formate dehydrogenase, it is the physiological direction seen in NADH dehydrogenase and most other members of this enzyme family. As illustrated by a typical kinetic trace in the inset of Fig. 8, the reaction is biphasic. The fast phase of the reaction is complete within 120 ms. A plot of the observed rate constant for the fast phase as a function of NADH concentration is hyperbolic (Fig. 8), yielding a limiting k red of 680 s Ϫ1 at high concentration of NADH and a K d NADH of 190 M. Assuming a doubling of the rate constant for every 10°C increase in temperature, this corresponds to a limiting k red of 3850 s Ϫ1 at 30°C, a rate that is more than 6-fold faster than the limiting rate of reduction of FdsABG holoenzyme at high concentration of formate, some 20-fold faster than the k cat of the FdsABG holoenzyme for formate oxidation (6) and 350-fold faster than the k cat for CO 2 reduction (7).

Characterization of FdsBG complex Formation of a neutral FMNH ⅐ semiquinone intermediate in the course of the reaction of FdsBG complex with NADH
At the completion of the fast phase of the reaction of 10 M FdsBG complex with 5 M NADH, the transiently observed UV-visible spectrum exhibits absorption in the 500 -550-nm region consistent with the appearance of a neutral flavin semiquinone (FMNH ⅐ ) (Fig. 9A, red 0.3-s spectrum). This species, not seen with the FdsABG holoenzyme, is the result of transfer of a single electron from the fully reduced flavin hydroquinone formed in the initial reaction with NADH to one or the other of the iron-sulfur clusters, leaving an FMNH ⅐ semiquinone behind; as described further below, this iron-sulfur cluster is the Fe 2 S 2 cluster of FdsG giving rise to the Fe/S 1 EPR signal (see also Scheme 1).
On a longer timescale, the system disproportionates to some degree in a second-order process (Fig. 9A, blue 10-s spectrum), with an electron passing between two equivalents of FdsBG 2eϪ to give one equivalent each of FdsBG 1eϪ and FdsBG 3eϪ . The FdsBG 1eϪ will have its Fe 2 S 2 cluster (Fe/S 1 ) reduced and its FMN oxidized, and FdsBG 3eϪ will have both Fe/S 1 and the FMN reduced. The reaction shown in Fig. 9A was carried out with substoichiometric NADH so that the FdsBG 1eϪ formed will not be further reduced by reaction with a second equivalent of NADH. The inset of Fig. 9A displays the kinetic time course obtained at 450 nm (extracted from the full data set collected with a photodiode array detector), showing both the fast reduction and the much slower disproportionation reactions.
We note that FdsBG may also involve a comproportionation reaction of FdsBG 2eϪ with the deflavo form of FdsBG complex (which accounts for ϳ50% of the total protein as described above). Whereas the deflavo form cannot be reduced by NADH due to the absence of the FMN, it can accept electrons from the fully constituted and reduced FdsBG 2eϪ , resulting in one equivalent of FdsBG 1eϪ with an oxidized FMN and reduced Fe/S 1 and a deflavo FdsBG 1eϪ containing only a reduced Fe/S 1 .
As the two processes are expected to proceed on at least approximately the same time scale, at the end of the redox reaction, there will be an excess of enzyme containing more oxidized than reduced flavin. An analysis of the slow phase of the reaction (described under "Experimental procedures") yields a second-order rate constant for this redox reaction on the order of 10 5 M Ϫ1 s Ϫ1 . It is noteworthy that, particularly at higher NADH concentrations, that the amplitude of the dis-/comproportionation phase decreases as the oxidized FMN thus generated becomes re-reduced by reaction with a second equivalent of NADH (data not shown).

Characterization of the neutral semiquinone intermediate, FMNH ⅐
The spectral signature of the FMNH ⅐ , which has not been characterized previously in the FdsABG holoenzyme, can be best observed in the difference spectra presented in Fig. 9B. At the completion of the fast phase described by the absorbance difference between the spectrum of the fully oxidized enzyme subtracted from the spectrum collected at 0.3 s after the initial reaction of FdsBG with NADH (black difference spectrum). A positive feature consistent with the formation of the FMNH ⅐ is observed in the 500 -650-nm region, although attenuated by the decrease in the absorbance due to the concurrent reduction of the iron-sulfur clusters. The complex negative feature in the 300 -500-nm region is dominated by the loss of absorbance associated with iron-sulfur reduction as well.
The disappearance of the FMNH ⅐ is also evident in the end of the second, slow phase of the reaction (and is even more distinct, being unaffected by any further reduction of the ironsulfur clusters). In the difference spectrum obtained by subtracting the spectrum collected at 0.3 s from that at 10 s after initial reaction with NADH (red difference spectrum), the 500 -650-nm region is defined by two broad peaks centered at 570 and 605 nm, characteristic of the presence of FMNH ⅐ (26). The features are negative as they reflect the disappearance of the FMNH ⅐ .

Characterization of FdsBG complex
Further evidence for the presence of a neutral semiquinone intermediate FMNH ⅐ comes from EPR measurements. To capture the transiently formed FMNH ⅐ , the enzyme was reacted with NADH at 0°C and rapid freeze-quenched with a quenching time of ϳ40 ms. Fig. 10A shows the resulting spectrum collected at 150 K.
The 150 K EPR spectrum includes contributions from both FMNH ⅐ and Fe/S 1 , demonstrating that it is the Fe 2 S 2 cluster of FdsG that has been reduced. Graphical subtraction of the Fe/S 1 signal from the measured EPR spectrum yields an isotropic EPR signal with a g iso ϭ 2.003 and a line width of ϳ1.9 mT, which is indicative of a neutral rather than an anionic semiquinone (27,28). Integration of both the scaled Fe/S 1 spectrum and the FMNH ⅐ spectrum from the graphical subtraction yielded a spin density ratio Fe/S 1 :FMNH ⅐ of 1:0.7, consistent with the formation of FMNH ⅐ and reduction of only a single iron-sulfur cluster at the end of the fast phase of the NADH reduction of the FdsBG complex. No additional iron-sulfur signal was seen at 9 K either (data not shown).
To ascertain whether reduction of the Fe/S 5 cluster occurred during the slow, second phase, an additional sample was prepared by incubating FdsBG complex with an excess of NADH for 10 s under anaerobic conditions. The resulting spectrum recorded at 9 K is shown in Fig. 10B. The spectrum illustrates that only the spectrum corresponding to a single iron-sulfur cluster, Fe/S 1 , is present at the end point of the second, slow phase of the NADH reduction reaction. Taken together, these results indicate that the fully reduced FMNH 2 initially formed on reaction with NADH breaks down by electron transfer of a single electron to the Fe 2 S 2 cluster (Fe/S 1 ), with the second electron remaining as FMNH ⅐ ; the Fe 4 S 4 cluster (Fe/S 5 ) remains oxidized (see Scheme 1).

Discussion
The genome of C. necator H16 (1, 2) encodes four putative formate dehydrogenases, of which two are soluble NAD ϩ -dependent formate dehydrogenases. One of them is a molybdenum-containing dehydrogenase (FdsABG), and the other is a tungsten-containing dehydrogenase (FdwAB) (29). Whereas the FdsB of the molybdenum-containing dehydrogenase has an additional N-terminal thioredoxin-like domain, the FdsB homolog of the tungsten-containing dehydrogenase is N-terminally fused to its FdsG homolog to form FdwB. Two other NADH oxidoreductases with homology to FdsABG and FdwAB are also present in C. necator-NADH dehydrogenase and NAD ϩ -reducing hydrogenase.

Functional relevance of the N-terminal thioredoxin-like domain of FdsB
The N-terminal thioredoxin-like domain of FdsB has structural similarity to the thioredoxin-like Fe 2 S 2 ferredoxin domain from A. aeolicus (21) (RMSD of 1.67 Å of 89 C␣ atoms between both domains). The A. aeolicus domain is known to dimerize. The largest contact interface between two FdsBG complexes within the asymmetric unit is formed between the two thioredoxin-like domains. This suggests that the thioredoxin-like domain of FdsB contributes to the dimerization of the FdsABG holoenzyme.
We suggest, however, because of (a) the relative small size of ϳ560 Å 2 of the buried area, (b) the presence of only a few interactions across the interface (Fig. S4, A-D), (c) the low conservation of the residues at the dimer interface (Fig. S4, E and F), and (d) the absence of a dimeric form in solution during purification (data not shown), that the dimer formation in the asymmetric unit is merely due to crystal packing (30,31) and does not participate in the dimerization of the FdsABG holoenzyme.
Although the N-terminal thioredoxin-like domain of FdsB is absent in other members of the NADH dehydrogenase family, it may nevertheless contribute to the stability of FdsB within the FdsABG holoenzyme. Protein stability is apparently a concern for some members of the family, as FdwB and HoxF have their FdsG and FdsB homologs fused. The linker between the FdsB-FdsG homologs overlaps structurally with the N-terminal thioredoxin-like domain of FdsB. The absence of the Fe 2 S 2 cluster in the FdsG homologous region of FdwB and HoxF might destabilize this region to such an extent that its stability is only maintained by the fusion to their FdsB homologous regions.
Finally, we cannot exclude the possibility that the N-terminal thioredoxin-like domain provides an interface for an unknown partner. This partner may either participate in the redox reaction or provide a connection to the cell's metabolic status, such as ACP does in the case of NADH dehydrogenase (32,33).

Characterization of FdsBG complex Electron transfer mechanism in FdsBG
In the FdsABG formate dehydrogenase of C. necator, the reducing equivalents from the oxidation of formate pass through a chain of iron-sulfur clusters from the molybdenum center in FdsA to the FMN in FdsB, which ultimately reduces NAD ϩ . The FdsBG subcomplex can perform this ultimate step of reducing NAD ϩ , as can the corresponding subcomplexes from both NADH dehydrogenase (15,20,34) and NAD ϩ -reducing hydrogenase (35).
In our FdsBG structure, the Fe 4 S 4 cluster of FdsB is in close proximity to the C8-methyl of FMN and to highly conserved surface residues that are implicated in the interaction of FdsB with FdsA (i.e. Glu-441 B -Ser-487 B , Ile-451 B -Gly-452 B , and Lys-292 B -Leu-298 B ). This is consistent with the Fe 4 S 4 cluster being on-path between the FMN and the molybdenum center of the FdsABG holoenzyme. The Fe 2 S 2 cluster of FdsG, on the other hand, is farther away from residues implicated in FdsA binding and is also too far from the Fe 4 S 4 cluster to directly take part in electron transfer from the molybdenum center to the bound FMN.
Nevertheless, it is clear from the evidence presented here that the initial electron transfer event out of the FMN upon reduction by NADH is to the off-path Fe 2 S 2 cluster (having the higher reduction potential) rather than to the on-path Fe 4 S 4 cluster (with much lower potential), as shown in Scheme 1. This process has not been observed previously in the reduction of either the FdsABG holoenzyme or the NADH dehydrogenase by NADH. It is thought that a similar off-path electron transfer minimizes the formation of reactive oxygen species in NADH dehydrogenase by minimizing the accumulation of neutral flavin semiquinone, FMNH ⅐ (22,36). In the case of FdsBG, the on-path electron transfer from the fully reduced flavin hydroquinone to the Fe 4 S 4 cluster, to the extent that it occurs, leaves a neutral flavin semiquinone, FMNH ⅐ . This semiquinone would then rapidly transfer the second electron to the Fe 2 S 2 cluster, thereby minimizing accumulation of the neutral semiquinone in the NADH dehydrogenase.

Implications for electron transfer in other members of the NADH dehydrogenase superfamily
Detailed structural analysis of the reduced and oxidized NuoEF complex from A. aeolicus revealed that the peptide bond between Glu-95 F and Ser-96 F of NuoF undergoes a reversible flip, depending on the oxidation state of the protein (20). In the oxidized state, the carbonyl group of Glu-95 F points to the FMN, whereas in the reduced state, it points to the Fe 2 S 2 cluster (Fig. 11D).
The equivalent peptide bond in the FdsBG complex (Glu-185 B -Gly-186 B ) adopts the same orientation in either state, oxidized or reduced (i.e. FdsBG as isolated and in the presence of NADH, respectively; Fig. 11C). However, in a subpopulation of the reduced FdsBG complex, the peptide bond prior to the oxidation state-sensitive peptide bond (Asp-184 B -Glu-185 B ) is pointing toward the isoalloxazine ring of FMN, even encroaching on its O4 carbonyl oxygen. We have attributed this encroachment to the low FMN occupancy of the NADH-reduced FdsBG.
We note that this region of the protein is not particularly well-conserved between NuoF and FdsB (94 F -DESEP-98 F versus 184 B -DEGDS-188 B ; the residues of the peptide that reorients dependent on the oxidation state in NuoEF are shown in boldface type; see also Fig. 11B). The equivalent region in HoxF (232 F -DEGEP-236 F ) appears to be flexible as well. As HoxF is devoid of an Fe 2 S 2 cluster (one cysteine coordinating the Fe 2 S 2 cluster is an alanine, and another cysteine is a serine), it is more likely that the absence of FMN in the reduced state of the Hox complex allows this peptide region to undergo the conformational change (14). These observations highlight the inherent flexibility of this region but also suggest that the oxidation state-dependent flipping of the peptide bond may be unique to the proper NADH dehydrogenase and not shared across the family.

Conclusion
In the present work, we have determined the X-ray crystal structure of the FdsBG portion of the FdsABG holoenzyme and characterized its interaction with NADH. The flexible region between the FMN and the Fe 2 S 2 cluster in NADH dehydrogenase has recently been reported to undergo a conformational change dependent on the oxidation state in proper NADH dehydrogenases; however, an oxidation state-dependent conformational change is not seen here with the formate dehydrogenase.
In addition, we see for the first time the formation of a neutral semiquinone, FMNH ⅐ , in the course of reduction by NADH as a result of an electron transfer from the fully reduced flavin hydroquinone to the Fe 2 S 2 cluster. The hydroquinone is generated upon reaction with NADH. The electron transfer to the Fe 2 S 2 cluster is off-path with respect to the electron transfer pathway to the molybdenum center, where the enzyme reduces CO 2 to formate in the FdsABG holoenzyme. The electron transfer to the Fe 2 S 2 cluster occurs over a distance of some 12 Å, with a rate constant that must exceed that for the NADH concentrationdependent rate of FMN reduction. The rate of FMN reduction is the rate-limiting step for electron transfer and occurs with a rate of at least 680 s Ϫ1 at 5°C. This electron transfer process to the Fe 2 S 2 cluster, observed here for the first time, has previously been proposed to play a role in limiting reactive oxygen species generation in the case of NADH dehydrogenase.

Cloning, protein expression, and protein purification of FdsABG holoenzyme and FdsBG subcomplex
The fdsGBACD operon with a His 6 -linker-His 6 tag fused to the N terminus of the FdsG subunit was cloned into a pTrcHisB vector to construct the pTrc12HLB-FdsGBACD plasmid as described previously (16).
For protein expression, the plasmid was transformed into E. coli DH5␣ cells. By varying expression conditions, protein synthesis was optimized for production of FdsBG complex at the expense of FdsABG holoenzyme. For the optimized procedure, a 50-ml starter culture of Luria broth medium with 100 g/ml ampicillin was inoculated from a glycerol stock of the DH5␣ transformed cells and grown overnight at 37°C. The overnight culture was used to inoculate three 6-liter Erlenmeyer flasks, each containing 2 liters of TB medium supple-

Characterization of FdsBG complex
mented with 100 g/ml ampicillin and grown at 30°C and 200 rpm to an OD 600 ϳ0.4, at which time they were induced with 0.25-0.5 mM isopropyl 1-thio-␤-D-galactopyranoside. The induced cells were grown for an additional 44 -48 h until OD 600 reached ϳ10. Harvested cells were frozen in liquid nitrogen and stored at Ϫ80°C until further use.
The plasmid pTrc12HLB-FdsGBACD was modified by removing the gene encoding the FdsA subunit, to yield pTrc12HLB-FdsGBCD. The pTrc12HLB-FdsGBCD plasmid was transformed into E. coli DH5␣ cells, and the protein FdsBG was expressed using the same procedure as described above. In addition, the plasmid pTrc12HLB-FdsGBCD was further modified by replacing the His 6 -linker-His 6 tag with a Twin-Strep tag (IBA, Gottingen, Germany; Twin-Strep tag sequence: WSHPQFEKGGGSGGGSGGSAWSHPQFEKSG) to yield pTrc-strep-FdsGBCD. The construction of this plasmid was outsourced (Epoch Life Science, Sugar Land, TX).
For overexpression of the FdsBG complex, the pTrc-strep-FdsGBCD plasmid was transformed into E. coli DH5␣ cells. A 50-ml starter culture of Luria broth medium supplemented with 100 g/ml ampicillin was grown overnight at 37°C. The overnight culture was used to inoculate two 6-liter Erlenmeyer flasks, each containing 2 liters of TB medium supplemented with 100 g/ml ampicillin and grown at 30°C and 200 rpm to an OD 600 of ϳ0.4, at which time they were induced with 0.25 mM isopropyl 1-thio-␤-D-galactopyranoside. The induced cells were grown for an additional 16 h at 30°C and 160 rpm (final OD 600 ϳ12-16). Harvested cells were frozen in liquid nitrogen and stored at Ϫ80°C until further use.
Purification of His 6 -linker-His 6 -tagged FdsBG complex-All steps were performed on ice or at 4°C. Frozen cells were thawed and suspended in 40 mM KPO 4 , 10 mM KNO 3 , pH 7.2, in the presence of lysozyme, DNase I, 0.5 mM PMSF, 1 mM benzamidine HCl, and 1 mM NaF. The resuspended cells were passed 1-2 times through a French Press at 10,000 -15,000 p.s.i. Cell debris was removed by centrifugation for 45 min at 200,000 ϫ g.

Characterization of FdsBG complex
were then transferred into an empty column and washed with five column volumes of resuspension buffer containing 45 mM imidazole. The FdsABG holoenzyme and FdsBG complex were eluted from the column with 300 mM imidazole. Ammonium sulfate was added to the eluent to 50% saturation, and the precipitate was collected. To reduce the contact time (ϳ1.5 h) between FdsBG complex and the nickel-nitrilotriacetic acid resin, the chromatography was performed with a head pressure of ϳ2 p.s.i. of N 2 .
The ammonium sulfate precipitate was treated in one of two ways. For preparations containing only FdsBG complex (the product of the overexpression of pTrc12HLB-FdsGBCD plasmid), no further purification was performed. The precipitate was resuspended and buffer-exchanged into 10 mM HEPES-NaOH, pH 7.2, with a PD-10 column (GE Healthcare) or by ultrafiltration with an Amicon Ultra with an molecular weight cuf-off (MWCO) of 50,000 (MilliporeSigma, Burlington, MA).
For preparations containing a mixture of FdsABG holoenzyme and FdsBG complex (the product of the overexpression from pTrc12HLB-FdsGBACD plasmid), the ammonium sulfate precipitate was dissolved in resuspension buffer containing 150 mM NaCl. The FdsBG complex was separated from FdsABG holoenzyme by gel filtration (Superdex 200 PG, GE Healthcare). The concentration and quality of FdsBG complex for each fraction of the gel filtration chromatography were assessed by the ratio of the absorption at 280 and 450 nm; pure FdsBG complex has a 280/450 nm ratio of 3.4. Fractions of sufficient purity and quality were combined, concentrated by ultrafiltration, and buffer-exchanged into 10 mM HEPES-NaOH, pH 7.2, with a PD-10 column or by ultrafiltration with an Amicon Ultra with an MWCO of 50,000. The concentrated protein from both preparations was aliquoted, flash-frozen in liquid nitrogen, and stored in liquid nitrogen until further use.
Purification of Twin-Strep-tagged FdsBG complex-All steps were performed on ice or at 4°C. Frozen cells were thawed and resuspended in 40 mM KPO 4 , pH 7.5, in the presence of lysozyme, DNase I, 0.5 mM PMSF, 1 mM benzamidine HCl, and 1 mM NaF. The resuspended cells were passed 1-2 times through a French Press at 10,000 -15,000 p.s.i. Cell debris was removed by centrifugation, and the clarified supernatant was loaded onto a 10-ml Strep-Tactin XT Superflow column (IBA, Gottingen, Germany) with a 2.5-cm diameter at 0.5 ml/min. The flow-through was loaded an additional time over the Strep-Tactin column before washing the column with 40 column volumes of resuspension buffer. FdsBG complex was eluted with resuspension buffer containing 50 mM biotin. The eluent was concentrated by ultrafiltration and buffer-exchanged into 10 mM HEPES-NaOH, pH 7.2, with a PD-10 column or by ultrafiltration using an Amicon Ultra with an MWCO of 50,000. The protein was aliquoted, flash-frozen in liquid nitrogen, and stored in liquid nitrogen until further use.

Structure determination of the FdsBG complex by X-ray crystallography
Crystallization-His 6 -linker-His 6 -tagged FdsBG complex was crystallized under anaerobic conditions by vapor diffusion in a sitting-drop format. 3 l of FdsBG complex at 2.5 mg/ml were mixed with 1 l of reservoir solution and set up over 500 l of reservoir solution. Oxygen was removed from all solutions used for crystallization by bubbling ISP-grade nitrogen through them. 5 mM DTT was added to the FdsBG stock solution prior to the removal of the oxygen. Crystals for native and anomalous data collection were grown over 5-6.5% (v/v) isopropyl alcohol, 0 -1.25% (v/v) n-propanol, 1.6 M (NH 4 ) 2 SO 4 , pH 7.5, at 16 -18°C and appeared within 2 days. Twin-Strep-tagged FdsBG complex was crystallized over 2-3% (v/v) isopropyl alcohol, 1.4 M (NH 4 ) 2 SO 4 , pH 7.5, at 14°C and appeared within 3 days.
Crystals were stabilized by the successive addition of reservoir solution containing no isopropyl alcohol but increasing concentrations of glycerol to a final concentration of 24% (v/v). The thus stabilized crystals were mounted and flash-frozen in liquid nitrogen. For NADH-bound FdsBG complex, stabilized Twin-Strep-tagged FdsBG crystals were soaked for an additional 45 min in final stabilization solution containing 10 mM NADH, before mounting and flash-freezing in liquid nitrogen.
Data collection-X-ray diffraction data of native FdsBG crystals were collected at the Se-absorption edge (0.97-1.00 Å) on the 5.0.1 and 5.0.2 beamlines at the Advanced Light Source (ALS) at Lawrence Berkeley National Laboratory (LBNL) and on the 24 ID-C and 24 ID-E beamlines at the Advanced Photon Source (APS) at Argonne National Laboratory (ANL). Anomalous X-ray diffraction data were collected at the Fe absorption peak wavelength of the intrinsic iron-sulfur clusters at 1.7388 Å on the 5.0.2 beamline at ALS at LBNL, on the 24 ID-C beamline at APS at ANL, and on the BL7-1 beamline at the Stanford Synchrotron Radiation Lightsource (SSRL) at the SLAC National Accelerator Laboratory. Collected data were processed with HKL-3000 (37), DIALS (38), XDS (39), and Mosflm (40).
Structure determination-The structure was solved to a resolution of ϳ3.5 Å by single anomalous diffraction-molecular replacement using the Crank2-SAD/MR pipeline (41) as implemented in CCP4 (42). The search model for molecular replacement was generated based on the homology of FdsB and FdsG to Nqo1 and Nqo2 from NADH dehydrogenase of T. thermophilus (chain 1 and 2 in the 2FUG PDB file (11)) using the SWISS-MODEL web server (43). To improve the completeness of our high-resolution data set, two native data sets of equivalent diffraction limit were combined using the "blend" program as implemented in CCP4. The initial phases generated by single anomalous diffraction-molecular replacement were extended and refined to 2.3 Å of our high-resolution native FdsBG data set using AutoBuild and Refine as implemented in the PHENIX program suite (44,45).
The solution of the X-ray structure shows two molecules of FdsBG complex in the asymmetric unit, which are related by an improper noncrystallographic rotation axis of ϳ179°. This improper rotation axis is nearly perpendicular (ϳ92°) to the crystallographic 2-fold axis (Fig. 2A).
The model of the FdsBG complex was built into the highresolution phased electron density with COOT (46) and refined with the PHENIX Refine subroutine (44). The restraints for the FMN and the Fe 2 S 2 cluster were generated with eLBOW as implemented in the PHENIX program suite. The restraints file for the noncubane Fe 4 S 4 cluster was taken from the GeoStd Characterization of FdsBG complex restraints file library (https://sourceforge.net/projects/geostd/ (47)). 5 The distance and angle restraints for the covalent bonds between the iron-sulfur clusters and the cysteine residues of their binding sites in the FdsBG complex were derived from high-resolution structures of well-characterized iron-sulfur complexes. Iterative cycles of building and refining were performed until R-factor values converged to an R work of 18.2% and R free of 22.3% (see Table S1 for detailed data and refinement statistics).
The data set of the NADH-soaked crystal was solved by molecular replacement using the above solved FdsBG structure as a starting model. Iterative cycles of building and refining were performed until R-factor values converged to an R work of 15.9% and R free of 19.9% (see Table S1 for detailed data and refinement statistics).
Structure alignments were performed using secondary structure matching as incorporated in COOT, and all molecular structure figures were prepared with PyMOL (version 1.7.4, Schrödinger LLC).

Inductively coupled plasma-optical emission spectrometry (ICP-OES)
ICP-OES (Optima 7300DV (PerkinElmer Life Sciences) at the Environmental Sciences Research Laboratory, University of California, Riverside) was used to identify the metal ions present in a sample of FdsBG. Approximately 220 l of 240 M FdsBG complex was buffer-exchanged into 5 mM triethanolamine, pH 7.7, and then diluted with nitric acid to 65% (v/v). The sample was boiled for 10 min and then diluted to 15 ml for analysis (final nitric acid concentration ϳ2-3%).

EPR spectroscopy
EPR spectra were recorded using a Bruker EMX spectrometer equipped with a Bruker ER 4119HS high-sensitivity X-band cavity and gaussmeter, operated with WinEPR version 4.33 acquisition software. Temperature was controlled using a Bruker variable temperature unit and liquid nitrogen or liquid helium cryostat. For purposes of comparison, all spectra were adjusted to a microwave frequency of 9.3871 GHz. Detailed instrument settings are included in the figure legends.
Samples were made anaerobic at 4°C and transferred to argon-flushed, septum-sealed EPR tubes. Samples prepared in this way were subsequently frozen in an ethanol/dry ice bath and transferred to liquid nitrogen. Further details of sample preparation are included in the figure legends. Simulations were performed using the EasySpin 4.5.5 software package (48). Prior to simulation of the composite spectrum described in Fig.  1A, parameters for the Fe/S 5 signal were estimated by graphically subtracting the spectrum of Fe/S 1 . During simulations, g-values for the Fe/S 5 were not allowed to vary by more than 0.005, whereas anisotropic line broadening (simulated as HStrain in the EasySpin software package) varied by no more than 20 MHz. Every multicomponent simulation also included a "weight" term, which could be used to estimate the relative contribution of each component to the composite spectrum.
Rapid freeze-quench EPR experiments were carried out using an apparatus described previously (49) except that the sample was not quenched by an isopentane bath at Ϫ140°C. Instead, the reaction mixture was sprayed unto rotating silverplated wheels immersed and equilibrated in liquid nitrogen (Ϫ196°C) and driven by a DC motor. The large surface area of the wheels served to efficiently freeze and simultaneously grind the reaction mixture for subsequent packing into an EPR tube. The quenching time of the experiment was determined as described by Ballou and Palmer (50), taking advantage of the well-characterized pseudo-first-order kinetics of the reaction of equine metmyoglobin with azide, the Mb⅐N 3 complex being readily detected by EPR. Double-integration of the EPR signal yields the total spin density for the Mb⅐N 3 complex, which is then correlated with the expected kinetics. Under our experimental conditions, the quenching time was estimated to be ϳ40 ms.

UV-visible absorbance measurements
Activity measurements and absorbance spectra were performed using a Hewlett-Packard 8452A diode array spectrophotometer equipped with a temperature-controlled cell holder. Reductive titrations were performed with 20 mM buffered sodium dithionite and 50 M enzyme in 100 mM KPO 4 , pH 7.0, at room temperature using a custom fabricated anaerobic cuvette with sidearms (51). The level of flavin saturation was determined as described previously, by treatment of protein with 5% TCA and quantification of released flavin at 446 nm using ⑀ ϭ 11,100 M Ϫ1 cm Ϫ1 for acidified FMN (52).

Rapid-reaction kinetics
The reaction of oxidized FdsBG complex with NADH was followed using a SX-20 stopped flow spectrophotometer (Applied Photophysics Ltd., Leatherhead, Surrey, UK) equipped with a photodiode array and a photomultiplier tube detection and operated with ProData SX 2.2.5.6 acquisition software. Time courses for the reaction were monitored at 450 nm at 5°C and fitted to a single-exponential equation by nonlinear least squares regression analysis using the software ProData Viewer 4.2.0 (Applied Photophysics Ltd., Leatherhead, Surrey, UK).
Only the first 15-120 ms of each time course was analyzed. The observed rate constants, k obs , were plotted against substrate concentrations to obtain the limiting rate constant for reduction, k red , and the dissociation constant, K d , using the following equation.
k obs ϭ k red ͓S͔/͑K d ϩ ͓S͔͒ For analysis of the second-order disproportionation reaction (the slow phase of the reaction of the enzyme with NADH), a 60-s trace collected at 450 nm was used to first estimate the end point of the reaction. A background value was then subtracted to bring the beginning of the second-order process to zero. The trace was then subtracted from the end point value, and a ⌬⑀ ϭ 7200 M Ϫ1 cm Ϫ1 was estimated empirically, assuming all 5 M of NADH were converted to the neutral semiquinone at the end of the fast phase and that disproportion proceeded to completion 5 Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party hosted site.

Characterization of FdsBG complex
by the end of the slow phase. The data were then converted to molar concentrations using the ⌬⑀ calculated above, and time versus reciprocal concentrations was plotted. The second-order rate constant was calculated by linear regression analysis of the data in the 2-5-s range.

Data availability
All atomic coordinates and structure factors can be found in the Research Collaboratory for Structural Bioinformatics Protein Data Bank under accession numbers 6VW8 for the FdsBG complex and 6VW7 for the NADH-bound FdsBG complex. All data are included in the article.