How a Formate Dehydrogenase Responds to Oxygen: Unexpected O2 Insensitivity of an Enzyme Harboring Tungstopterin, Selenocysteine, and [4Fe-4S] Clusters

The reversible two-electron interconversion of formate and CO2 is catalyzed by both non-metallo and metallo-formate dehydrogenases (FDHs). The latter group comprises molybdenum-or tungsten-containing enzymes with the metal coordinated by two equivalents of a pyranopterin cofactor, a cysteinyl or selenocysteinyl ligand supplied by the polypeptide, and a catalytically essential terminal sulfido ligand. In addition, these biocatalysts incorporate one or more [4Fe-4S] clusters for facilitating long-distance electron transfer. But an interesting dichotomy arises when attempting to understand how the metallo-FDHs react with O2. Whereas existing scholarship portrays these enzymes as being unable to perform in air due to extreme O2 lability of their metal centers, studies dating as far back as the 1930s emphasize that some of these systems exhibit formate oxidase (FOX) activity, coupling formate oxidation to O2 reduction. Therefore, to reconcile these conflicting views, we explored context-dependent functional linkages between metallo-FDHs and their cognate electron acceptors within the same organism vis-à-vis catalysis under atmospheric conditions. Here, we report the discovery and characterization of an O2-insensitive FDH2 from the sulfate-reducing bacterium Desulfovibiro vulgaris Hildenborough that ligates tungsten, selenocysteine, and four [4Fe-4S] clusters. Notably, we advance a robust expression platform for its recombinant production, eliminating both the requirement of nitrate or azide during purification and reductive activation with thiols and/or formate prior to catalysis. Because the distinctive spectral signatures of formate-reduced DvH-FDH2 remain invariant under anaerobic and aerobic conditions, we benchmarked the enzyme activity in air, identifying CO2 as the bona fide product of catalysis. Full reaction progress curve analysis uncovers a high catalytic efficiency when probed with an artificial electron acceptor pair. Furthermore, we show that DvH-FDH2 enables hydrogen peroxide production sans superoxide release to achieve O2 insensitivity. Direct electron transfer to cytochrome c in air also reveals that electron bifurcation is operational in this system. Taken together, our work unambiguously proves for the first time the coexistence of redox bifurcated FDH and FOX activities within a metallo-FDH scaffold. These findings have important implications for engineering O2-tolerant FDHs and bio-inspired artificial metallocatalysts, as well as for the development of authentic formate/air biofuel cells, modulation of catalytic bias, assessing the limits of reversible catalysis, understanding directional electron transfer, and discerning formate bioenergetics of gut microbiota.


INTRODUCTION 9
The simplest carboxylic acid (formic acid), and its conjugate base (formate) are 10 normal products of metabolic activity in living organisms, including bacteria and humans. 1 11 However, bacterial aerobic respiration of formate derived from human gut microbiota 12 drives inflammatory dysbiosis. 2 Although formic acid is primarily used as a food 13 preservative (E236) or as silage additive for maintaining the nutritive value of animal feed, 3 14 it is a highly sought-after electron-mediator and feedstock in (electro)microbial 15 bioproduction, 4 as well as a low carbon-footprint molecule that serves as a chemically 16 robust hydrogen storage medium. 5 In addition to being a carbon and energy source for 17 the (an)aerobic growth of disparate bacteria, 6 archaea, 7 and syntrophic consortia, 8 18 formate can be generated abiotically from CO 2 and renewable electricity. 5 19 Formate oxidation and CO 2 reduction are interconvertible processes that are 20 carried out by prokaryotic formate dehydrogenases (FDHs) (Reaction 1). 9, 10 21 HCOO − + H + ⇔ CO 2 + 2e − + 2H + E m,7 = -430 mV (1) 22 There are two phylogenetically distinct FDH families that can be distinguished by their 1 transition metal ion requirement for enzyme activity. 11 Metallo-FDHs are thought to be 2 highly sensitive to O 2 , 12,13 necessitating catalytic measurements under anaerobic 3 conditions. However, available data in the primary literature are more confusing than 4 definitive. For example, DvH-FDH3 has been reproducibly shown to be O 2 sensitive 14, 15 5 while its ortholog from D. desulfuricans ATCC 27774 (Dd) can be purified in air. 16, 17 6 Similarly, D. gigas (Dg) FDH1 is readily isolated and stored under atmospheric conditions 18, 7 19 but its counterpart from DvH requires the presence of 10 mM sodium nitrate to prevent 8 O 2 inactivation. 20 Regardless of the purification protocols, the resulting enzymes are 9 'dead on arrival' in that they must be resurrected by lengthy incubations with high 10 concentrations of thiols (10 -50 mM dithiothreitol 20 for DvH-FDH1 and 130 mM β-11 mercaptoethanol 16,17,19 for Dd-FDH3 and Dg-FDH1) and/or formate 17,21 prior to catalytic 12 measurements under anaerobic conditions (a representative example can be found in 13 Figure S5a of Oliveira et al 20 ). A satisfactory molecular explanation for these 14 phenomenological observations has not been forthcoming for over three decades. 18 The 15 situation is equally unclear with FDHs isolated from organisms other than sulfate-reducing 16 bacteria (SRB). Escherichia coli Fdh-H has been purified and characterized in the presence 17 of sodium azide to minimize O 2 inactivation. 22, 23 10 mM sodium nitrate 24, 25 or azide 26, 27 18 has been added as stabilizers during the isolation of other bacterial metallo-FDHs as well. 19 Very little is known about how these small molecules protect the enzyme from O 2 . 20 Because stability in air does not enable aerobic catalysis, the inhibitors are either removed 21 prior to measurements under anaerobic conditions 20,25 or allowed to remain while the 22 activity is probed anaerobically 23 or in air. 28 To our knowledge, no FDH has been shown 23 to reversibly interconvert formate and CO 2 in air. Claims to the contrary have been 24 considered as experimental artifacts. 25 Moreover, mechanistic details regarding how O 2 25 reacts with these metalloenzymes are not available. 26 Largely overlooked, however, is the fact that the present-day claims of FDH O 2 1 sensitivity fail to recognize or rationalize the findings reported between the late 1920s 2 and early-1990s vis-à-vis existence of metallo-FDHs capable of oxidizing formate with 3 oxygen (Reaction 2). 4 Starting with the first purification of a bacterial FDH by Stickland in 1929, O 2 uptake served 6 as a proxy for measuring enzyme activity. 29 Subsequently, Stephenson and Stickland 7 isolated Escherichia coli hydrogenlyase (Fdh-H), revealing that it was not responsible for 8 the FOX activity. 30 A key insight regarding the latter came from Gale's demonstration that 9 formate dependent O 2 consumption by E. coli was higher in aerobically grown cells. 31 10 Pinsent's groundbreaking study relied on Gale's O 2 utilization assay to convincingly show 11 that not only E. coli FDH requires molybdenum and selenium for function, but more 12 importantly, that it retained robust FOX activity. 32 This has since been independently 13 confirmed by several research laboratories. [33][34][35][36][37] Pichinoty went on to show that FOX 14 activity was broadly distributed across bacteria. 38

8
Error bars represent standard deviations from three independent measurements.

10
sulfate medium ( Figure 1B). We also constructed deletion strains harboring different 11 combinations of fdh genes for functional analyses, including JW2111 (∆fdh3) and JW2121 12 ( ∆fdh1 and ∆fdh3; see Tables S1 and S2). The latter two served as controls in this study 13 ( Figure S1). Subsequently, we used JW2127 for the homologous expression of FDH2. 14 Introduction of a Strep-tag II at the C-terminus of the large subunit facilitated one-step 15 affinity purification. Whereas Oliveira et al 20 used DvH cells derived from 300 L 16 fermentation to purify FDH1, we have streamlined our workflow to produce 1.8 mg of 17 highly-pure heterodimeric FDH2 from a gram of wet cell paste (Figures 2A and 2B). Thus, 18 our 10 L culture (biomass yield of ~ 8 g) generates sufficient protein to tackle even the 19 most demanding experiments. And our method can be readily scaled up.

8
Notably, there is a fundamental difference between prevailing strategies for metallo-FDH 9 isolation and what we have advanced. Our purification workflow ( Figure 2) and 10 downstream handling steps (including storage) occur in air without involving nitrate, 11 azide, thiols, or formate at any stage of the process. 12 Aerobic In-Gel Catalysis of Recombinant DvH-FDH2. Literature precedents exist for 13 anaerobic activity staining of FDHs in native polyacrylamide gels using 2,3,5-14 triphenyltetrazolium chloride 56, 59 or phenazine methosulfate (PMS) / nitroblue 15 tetrazolium chloride (NBT). 60-62 However, this has not been achieved for any FDH in air. 16 Because O 2 -insensitive group 5 [NiFe]-hydrogenases have been zymographically 17 visualized using redox dyes, 63 we asked whether a similar approach could work with DvH-18 FDH2. When native polyacrylamide gel strips containing recombinant DvH-FDH2 were 1 incubated aerobically with NBT and formate, a single dark blue colored band appeared 2 within two minutes ( Figure 2C). In the absence of formate, this band was not observed. 3 The same pattern was recapitulated in the spot assay where the blue color developed 4 within 15 s ( Figure S2). These observations demonstrate that electrons released from 5 enzymatic aerobic formate oxidation are readily transferred to an artificial electron 6 acceptor with positive reduction potential (E m,7 = +50 mV 24, 63 ), resulting in the generation 7 of insoluble reduced NBT-formazan precipitates. Furthermore, our results establish that 8 both nitrate-assisted purification of FDH and/or reductive activation with high 9 concentration of thiols are not essential for maintaining redox activity under anaerobic or 10 atmospheric conditions. 11 [4Fe-4S] Metalloclusters, Tungstopterin, and Selenocysteine Remain Unaffected by 12 O 2 During Catalytic Turnover. As correctly pointed out by Hagen 64 , the metal specificity 13 profiles of SRB FDHs remain incompletely described. Moreover, the nature of redox 14 centers in DvH-FDH2 is unknown. 56 Because DvH-FDH1 and DvH-FDH2 exhibit 61% 15 protein sequence identity (large subunit) and share all the metal coordination sites within 16 the two subunits ( Figure S3), we surmised that a similar complement of redox centers 17 must exist in both systems. Since the DvH biomass was derived from a medium containing 18 Mo (1.24 µM) and W (0.15 µM), we predicted a metal ratio of 1Mo/W:16Fe:1Se. Consistent 19 with this, inductively coupled plasma mass spectrometry (ICP-MS) revealed that for every 20 mole of 182 W present, another 17± 1 moles of 56 Fe and 0.7 ± 0.1 moles of 78 Se were also 21 found (Table S3) 16,20,24,65 Although aerobic spectra exist for an O 2 -tolerant Mo-Cys-9 FDH stabilized by 10 mM nitrate, 28 their utility remains unclear, for the addition of formate 10 did not afford a characteristic spectral change. Here, we offer the first functional 11 validation of a W-Sec-FDH in air via electronic spectroscopy. Aerobically purified DvH-12 FDH2 is brown in color and shows a broad S → Fe 3+ charge transfer transition at 412 nm 13 ( Figure 3A, blue trace), which is characteristic of [4Fe−4S] 2+ clusters. 66 Addition of formate 14 leads to a substantial loss of this signal, indicating reduction to the [4Fe−4S] + state ( Figure  15 3A, green trace). Independently, reduction with dithionite yields a similar result ( Figure  16 3A, orange trace). Employing anaerobic conditions makes no difference to the outcome 17 ( Figure 3C). The virtually identical lineshape and amplitude of the difference spectra 1 ( Figure 3B,D) illustrate that formate completely reduces (six reducing equivalents; four 2 [4Fe−4S] 2+ clusters and one W center) the majority of catalytically competent FDH2 in 3 solution. As dithionite would be expected to reduce both functional and non-functional 4 metal centers, we conclude that >94% of DvH-FDH2 is functionally fit. We have also 5 obtained the source DvH-FDH1 spectrum ( Figure S4, pink trace, of Oliveira et al (2020)) 6 and compared it with our as-isolated DvH-FDH2 counterpart acquired under anaerobic 7 conditions ( Figure S4). The A 400 /A 280 ratio -an indicator of the extent of cluster loading 67 -8 estimated from these spectra are 0.18 (DvH-FDH2) and 0.17 (DvH-FDH1), affirming that 9 the two orthologs exhibit comparable protein purity and cofactor integrity. 10 To evaluate the predictions made via electronic spectroscopy, we pursued EPR 11 measurements. Figure 4 shows the EPR spectra seen with reduced DvH-FDH2 under a 12 variety of conditions, as well as the spectra for the as-isolated protein (panels (i), (v)). At 13 15K, we observed predominantly two distinct EPR signals ( Figure 4A(ii)-(iv)), the relative 14 intensities of which are essentially independent of reductant (formate or dithionite) and 15 environment (anaerobic or air). The signals are consistent with the presence of reduced 16 iron sulfur clusters. At 26K, one of the signals is significantly broadened ( Figure 4A(vi)-17 (viii)) while by 40K both signals have disappeared (data not shown). This behavior is 18 consistent with fast relaxing [4Fe-4S] clusters. There is also some indication of additional 19 signals ( Figure 4A, red arrows), which are described in more detail below. 20 The simulated spectrum for the formate-reduced DvH-FDH2 prepared under 21 aerobic conditions and collected at 15K from Figure 4(iii) is shown in Figure S5, and the 22 simulation parameters given in Table S4.  Table S4.  Figure 4B(ii) and (iii) present the component spectra scaled to their 5 contribution to the composite simulation in Figure 4B(i). The simulation parameters are 6 presented in Table S4 and include the well-resolved tungsten I=1/2 hyperfine splittings 7 originating from the 14.3% natural abundance 183 W isotope. The presence of the I=1/2 8 hyperfine splitting is further evidence that these signals arise from the tungsten center 9 rather than additional Fe/S clusters. The simulations indicate that the two species are in 10 an approximate ratio of 1:0.54 and the g-values (g 1-3 = 1.982, 1.876, 1.849 and 1.904, 1.849, 11 1.914, respectively) are in good agreement with those seen from other W-containing 12 enzymes. Somewhat surprisingly, the large anisotropy of the W(V) g-values more closely 13 resembles the "low potential" signal for the P. furiosus aldehyde ferredoxin 14 oxidoreductase (AOR), which is a member of a different family of tungsten-containing 15 enzyme than the FDHs. 70 The presence of multiple W(V) signals in a single sample has 16 been seen with a number of W-containing enzymes and may be due to the presence of 17 inactive species in addition to the catalytically competent one, which is a rather common 18 feature of W-containing enzymes. 64 19 surrogates to report on catalytic robustness. Although cautions have been raised against 26 trusting kinetic parameters derived from the use of these "inefficient and slow redox 27 mediators" 71, 72 , they continue to be favored. Mo-Cys-FDHs offer an alternative by making 1 it possible to track NAD + reduction or NADH oxidation. 24, 28 Unfortunately, this strategy 2 can be extended only to select metallo-FDHs and it is prone to yield false-positive results 3 when interrogating aerobic CO 2 reduction with aerotolerant enzymes. 25 To further 4 complicate matters, FDHs from SRB are in a class of their own (Table S5). Unlike other 5 bacterial FDHs, these retain little to no activity after purification, requiring lengthy 6 reductive activation with high concentration of thiols and/or formate. 17,20 Finally, it is 7 impossible to assess the validity or robustness of the published results when experimental 8 data remains inaccessible -we are not aware of a report on metallo-FDH enzymology that 9

Full Progress Curves Reveal High Catalytic Efficiency Under Atmospheric Conditions
has disclosed a complete set of raw absorbance versus time data used to extract kinetics 10

parameters. 11
To resolve these uncertainties, we explored rigorous and reproducible solution 12 enzyme kinetics approaches capable of yielding results with functional information 13 content. First, we assessed catalytic efficiency with two chemically distinct artificial 14 electron acceptors, one each from the low-(BV) and high-potential (PES/DCPIP) E m,7 = 15 +217 mV 24 ) categories. Only the latter afforded the ability to acquire kinetics data both 16 in air and under argon. Second, we identified conditions under which full progress curves 17 could be measured. Such an approach is only possible for stable enzymes that catalyze a 18 single-substrate irreversible reaction in the absence of enzyme inactivation or product 19 inhibition (see below). 73-76 And third, we have simultaneously analyzed several full 20 progress curves using dynamic simulation-based global fitting 75 to extract k cat and 21 k cat /K m . This strategy overcomes the limitations of classical steady-state analysis, such as 22 the use of only the first few seconds of data, unreliable initial velocity values, and 23 overparameterization. 77 To benchmark our models (Schemes 1 and S1), we obtained 24 source BV enzyme kinetics data that formed the basis of Figure S3 and Figure 1C   The rate of product release (k +3 ) is set to a high value, facilitating the reaction chemistry to define k cat .

5
Reaction of E red with A ox (k +4 ) is assumed to be diffusion limited. Rationale for setting reverse rate constants 6 to zero has been explained by Johnson. 77 The last equation is relevant only under aerobic conditions. Also, see 7 Scheme S1. results from the initial velocity data, we were able to extract five full progress curves from 13 their source data and perform de novo analysis ( Figure S7A-C). In addition to finding k cat 14 values in the reported range, our method redefines the K m of DvH-FDH1 to be 4.6 ± 0.3 15 µM rather than 17 µM ( Figure S7D-F). These observations illustrate that our catalytic 16 models are poised to extract reliable kinetic parameters from DvH-FDH2 progress curves. 17 Because the original characterization of native DvH-FDH2 -by the same laboratory 18 that has reported extensively on DvH-FDH1-was done using 2 mM BV (see Table S5 Table   10 1 lists rate constants, as well as best fit parameters derived from this analysis.

12
we attempted to reproduce the published results with aerobically purified recombinant 13 DvH-FDH2. However, our enzyme was added to the reaction mix without any pre-14 activation with thiols or formate. Although DvH-FDH2 displays redox activity in air ( Figure  15 2C), abiotic reaction of reduced BV + with O 2 made us employ strict anaerobic conditions. 16 Since our measurements were not made inside an anaerobic chamber, we ensured anoxic 17 conditions by adding 1 unit/mL of glucose oxidase (GO). Catalase was also 18 Electron acceptor k cat (s -1 ) K m (µM) ‡ k cat /K m (µM -1 s -1 )   Table 1). Furthermore, our progress curves revealed that two 4 molecules of BV + are generated for every formate molecule oxidized by DvH-FDH2. 5 Despite elimination of the reductive activation step, the enzymatic parameters derived 6 from our standard steady-state kinetics analysis were virtually identical to the published 7 values (Table S6), suggesting that the pre-activation step introduced by da Silva et al 8 (2011) had no effect on the outcome. In our hands, DvH-FDH2 exhibits catalytic 9 parameters that are roughly an order of magnitude smaller than their DvH-FDH1 10 counterparts (Table S6 and Figure S7F). A closer inspection of product stoichiometry at 11 high formate concentrations suggested that BV concentration could be limiting ( Figure  12 S8). Therefore, we repeated the experiments at 20 mM BV. This restored 2BV + :1F 13 stoichiometry across the board but catalytic parameters did not change appreciably 1 ( Figure S9, Table S6). We have also confirmed that addition of GO and catalase do not 2 interfere with the results of activity measurements ( Figure S10). 3 Collectively, the observations above suggest that BV is not a good electron 4 acceptor for DvH-FDH2. To test this hypothesis, we independently pursued activity assays 5 with PES/DCPIP. Consistent with literature precedents on dehydrogenases, 78 phenazine 6 ethosulfate (PES; E m,7 = +65 mV 79 ) was required for transferring electrons from FDH to 7 DCPIP ( Figure S11A). By varying the concentrations of DCPIP and PES, we were able to 8 identify optimal conditions that would support activity measurements both in air ( Figure  9 S11B-F and Figure S12) argon ( Figure S13). At DCPIP concentrations below 100 µM, global 10 fitting of anaerobic full progress curves (Scheme 1 and Figure 5D-F) resulted in roughly 11 five-fold higher turnover number (TN) than what we obtained with 2 mM BV (Table 1). K m 12 values did not show a significant difference, however. The same pattern was reproduced 13 when the measurements were made in air ( Figure 5G-I). A key difference between the 14 two conditions is that the slow reoxidation (k +5 = 3 ± 0.5 x 10 -6 µM -1 s -1 ; see last equation 15 in Scheme 1) of reduced DCPIP by O 2 caused the post reaction region to slope slightly 16 upward ( Figure 5G, Figure S11D, and Figure S12B). However, this should not be confused 17 with alterations to the shape of progress curves stemming from product inhibition, 75 18 which we did not observe when DCPIP ( Figure 5G) or BV ( Figure 5A) served as electron 19 acceptors. In fact, a product stoichiometry of one reduced DCPIP for every formate 20 oxidized was reproducibly found in our measurements (Table 1). Since addition of fresh 21 substrate at the end of a progress curve cleanly reproduced the original trace ( Figure  22 S12F), we can further ascertain that the enzyme was stable and fully active during catalysis 23 in air. Taken together, PES/DCPIP-dependent catalytic parameters obtained from global 24 fits are in excellent agreement with those from our initial velocity calculations (Table S6). 25 And the TN with PES/DCPIP remains virtually the same under anaerobic and aerobic 26 conditions. Notably, the catalytic efficiency of DvH-FDH2 in air is in the range of 7 x 10 7 1 M -1 s -1 (Table 1), which is comparable to that reported for DvH-FDH1 20 (~ 8 x 10 7 M -1 s -2 1 ) when BV serves as the electron acceptor under anaerobic conditions. Moreover, our 3 PES/DCPIP-based TN and k cat /K m for formate oxidation are roughly an order of magnitude 4 and 500-fold higher, respectively, than what has been reported for the aerotolerant Mo-5 Cys-FDH from Rhodobacter capsulatus using the natural (NAD + ) electron acceptor. 28 6 Finally, we have found that the inhibition profiles of DvH-FDH2 in the presence of azide 7 or nitrate are not significantly impacted by O 2 (Figures S14 and S15). Whereas azide 8 blocks the enzyme with an IC 50 of about 0.8 mM, nitrate is far less effective. 9

Paradigm to Seek Insights into How FDHs May Have Evolved to Achieve Aerobic 11
Catalysis. Our full progress curve analysis establishes that both k cat and k cat /K m are 12 severely underestimated when BV is used as the electron acceptor. It also reveals a 13 preference for the latter viz., whereas DvH-FDH2 favors high-potential acceptors, such as 14 PES/DCPIP or NBT, DvH-FDH1 is highly active with BV 20 . Although DCPIP data is not 15 available for the latter enzyme, a close ortholog (Dg-FDH1) only shows 10% of BV activity 16 with the high-potential acceptor. 18 Such linkages take on special significance when

4
is involved in anaerobic respiration and that the PES/DCPIP-linked FDH2 plays a role in 5 aerobic respiration. It has already been established that FDH1 is essential for anaerobic 6 sulfate respiration when formate serves as the electron donor. 50 Biological function of 7 FDH2 remains to be elucidated. Our study proves that catalytic parameters derived from 8 viologen-based measurements lack functional information content to make predictions 9 about how well a given FDH would perform under aerobic conditions. Instead, high 10 catalytic performance on BV only guarantees activity under anaerobic conditions. 11 Confirmation bias has boosted reliance on viologen-based kinetics and stymied efforts to 12 uncover O 2 -immune FDHs that can reversibly function in air. This is best exemplified by 13 DvH-FDH2, which exhibits the lowest TN with BV (Table S5) and yet is the most O 2 -14 insensitive of all metallo-FDHs characterized to date from any bacterium. Therefore, 1 biological context must factor critically into future search efforts aimed at discovering air 2 insensitive FDHs. In all remaining works, product formation is implied based on the reduction of a natural 7 (NAD + ) or artificial electron acceptor, which is often BV. Although we have used two 8 different artificial electron acceptors in this study, we made sure to leave no stone 9 unturned where product analysis in air is concerned. At pH 7.5, combining DvH-FDH2 10 with isotopically labeled 13 C-formate readily yields a discernible H 13 CO 3 − resonance at 11 NMR spectroscopy 83 and not optimizing data collection for 5x T 1 (relaxation time), we 1 succeeded in demonstrating that CO 2 is the true product of aerobic catalysis. 2 FOX Activity Generates H 2 O 2 , Enabling Oxygen Insensitivity of DvH-FDH2. To 3 understand how DvH-FDH2 deals with O 2 , we used a Clark-type O 2 electrode to ask 4 whether formate oxidation under atmospheric conditions is coupled to O 2 reduction. 5 Addition of enzyme to formate-containing aerobic buffer led to robust O 2 consumption 6 ( Figure 7A). Once the signal plateaued, catalase was added. This led to O 2 evolution 7 followed by O 2 uptake until a plateau was reached, suggesting the production of H 2 O 2 8 via 2e − reduction of O 2 (reaction 3). 9 We attempted a calculation of the electron flux that led to H 2 O 2 formation. 84, 85 The x/y 11 value in Figure 7A would imply that roughly one quarter of the electrons from formate 12 were ending up in H 2 O 2 . However, this is likely to be an underestimate because formate 13 was in large excess and was continuing to be oxidized post dismutation of H 2 O 2 by 14 catalase, manifesting as the second O 2 uptake step. Instead, if we considered the u/v 15 value, ~ 80% of the electron flux goes towards peroxide generation. To resolve this 16 uncertainty, we pursued the kinetics approach developed by Lu and colleagues. 86 By 17 comparing the initial velocities of O 2 uptake in the absence ( Figure 7A) and presence 18 ( Figure 7B) of catalase, we found that it was 50% lower in the latter ( Figure 7C). And the 19 rates did not vary significantly between pH 6.0 and 8. This outcome suggested that H 2 O 2 20 was the major product of O 2 reduction during aerobic formate oxidation. Appropriate 21 controls were built into our experimental design for evaluating alternate endpoints. H 2 O 2 22 addition in the absence of exogenous catalase showed that DvH-FDH2 lacks catalase 23 activity ( Figure S24A). To rule out the possibility of abiotic O 2 consumption, DvH-FDH2 24 was heat denatured and subjected to oximetry. Neither O 2 uptake nor H 2 O 2 generation 1 was found ( Figure S24B). Moreover, inclusion of 1 mM EDTA minimized artifacts arising 2 from transition metal contaminants. Even with these controls in place, we could not 3 eliminate the possibility that atmospheric O 2 was completely excluded during oximetry. 4 Therefore, we pursued two orthogonal approaches to directly quantify H 2 O 2 . In the first 5 method, horseradish peroxidase (HRP) catalyzed formation of fluorescent resorufin from 6 H 2 O 2 and amplex red was monitored. We observed that for every mole of formate 7 oxidized, roughly 0.75 mole of H 2 O 2 was produced during aerobic DvH-FDH2 catalysis 8 ( Figure 7D). Inclusion of catalase abolished the fluorescence signal and denatured 9 enzyme failed to yield H 2 O 2 ( Figure S25). However, the inability of amplex red assay to 10 quantify H 2 O 2 beyond 5 µM made it impossible for us to investigate the consequences of 11 O 2 reduction at formate concentrations approaching 10 -20K m . Furthermore, despite 12 being considered the gold standard, this assay is prone to artifacts. 87, 88 For example, 13 interferences stem from interaction between redox enzymes and resorufin 89 . To 14 overcome these limitations, we resorted to a method independent of HRP and amplex 15 red. Here, we followed the direct reaction of non-fluorescent coumarin-7-boronic acid 16 (CBA) with H 2 O 2 , leading to the production of fluorescent 7-hydroxycoumarin (COH). 90 17 Although this reaction is slow 90 (k on 1.5 M -1 s -1 ), the assay is linear over a much broader 18 range of H 2 O 2 . Therefore, we quantified H 2 O 2 production during aerobic DvH-FDH2 19 catalysis, varying formate concentration between 1 -10K m . It amounted to 64 ± 6% and 20 did not show significant variation when higher enzyme concentrations were used. 21 Catalase addition eliminated the signal completely, confirming that H 2 O 2 is indeed the 22 major product of O 2 reduction by DvH-FDH2 ( Figure S26). 23 Next, we tried to assess superoxide (O 2 • ̅ ) generation 91 by DvH-FDH2. Because 24 addition of SOD had negligible effect on both direct quantification (Figures S25 and S26) 25 and oximetry ( Figure S27), we probed the reduction of partially acetylated cytochrome c. 1 The advantage of using the latter is that it is still reducible by O 2 • ̅ but not susceptible to 2 interferences arising from oxidase or reductase activities when unmodified cytochrome c 3 serves as the substrate. 92 Significant reduction was not observed, implying that O 2 • ̅ was 4 not released by DvH-FDH2 ( Figure S28). 5 Taken together, our results establish FOX activity of a highly pure metallo-FDH. To 6 the best of our knowledge, this has never been demonstrated before. Based on IUPAC-7 IUB nomenclature, the term "oxidase" is reserved for enzymes, which utilize O 2 as the 8 electron acceptor. In our case, formate oxidation is coupled to 2e − reduction of O 2 by 9 DvH-FDH2, resulting in 65 to 75% H 2 O 2 production (reaction 3). Hence, we project that anaerobic conditions, only ~80% underwent reduction in air ( Figure 7F). However, the 7 initial rates remained invariant ( Figure S29). Doubling the formate concentration resulted 8 in near stoichiometric reduction in air ( Figure S30). Moreover, inclusion of SOD or catalase 9 did not result in a noticeable difference ( Figure S31). Under the conditions employed, 10 dissolved O 2 (257 µM) is at a much higher concentration than cytochrome c. Yet, electrons 11 are readily delivered to the latter. These results prove for the first time that oxidase and 12 dehydrogenase activities coexist within the DvH-FDH2 scaffold. We advance a non-13 energy-conserving electron bifurcation (EB) 96 mechanism to rationalize how this is 14 accomplished (Figure 8). Here, electrons resulting from the oxidation of a 2e − donor 15 (formate) traverse two independent thermodynamically favorable electron transfer paths 16 (FOX and FDH) to reduce two different electron acceptors (O 2 and cytochrome c). 1 Although we do not know the identity of the cofactor engaged in EB, both the 2 tungstopterin and the proximal [4Fe-4S] cluster of the large subunit are strong candidates. 3 Apropos, arsenite oxidase is thought to achieve EB by cycling between the Mo(VI)-dioxo 4 and Mo(IV)-oxo states. 97 Whether tungsten can assume this role in DvH-FDH2 remains 5 to be investigated. Based on our observation that superoxide was not generated during 6 aerobic catalysis, we hypothesize that the bifurcating site likely favors normally ordered 7 over inverted potentials. 98, 99 Even though the interconversion of formate to CO 2 is 8 reversible, the bifurcation events in either direction (especially O 2 reduction) are 9 potentially irreversible. 100 Finally, EB is thought to be the dominant source of reactive 10 oxygen species in biological systems. 100 We have shown that redox metalloenzymes can 11 capitalize on it to achieve O 2 insensitivity. Here, we have narrowed the sequence space to just two closely related paralogs -one of 16 these (DvH-FDH1) is unable to achieve catalysis in the presence of O 2 while the other 17 (DvH-FDH2) thrives in air. To gain atomic insights, we constructed a highly accurate 18 heterodimeric structure of the latter using AlphaFold2.1 101 ( Figure 9A). We did not utilize 19 preexisting structure templates to model the structure. The resulting atomic coordinates 20

(sans cofactors) include confidence metrics (predicted Local Distance Difference Test, 21
pLDDT) at the single residue level wherein higher scores on a scale of 1 -100 represent 22 greater confidence. Figure S32A shows that the heterodimeric structure of DvH-FDH2 is 23 modeled with high confidence; bulk of the polypeptide chain displays pLDDT scores >90. 24 Similarly, the predicted aligned error (PAE; color saturation found at any x,y coordinate in 25 Figure S32B) is a metric of how well a residue is positioned and oriented. Backbone RMSD variations are shown in Figures S33 and S34). Similarly, the RMSD between DvH-FDH1 and DvH-FDH2 heterodimers is 1.37 Å (952 Cα atoms). After 1 completing these validations, we computed a residue-residue (RR) distance map to 2 identify the regions of major variation between FDH2 and FDH1 ( Figure S35). 3 Although the active site residues are largely conserved between FDH2 and FDH1, 4 there are several differences in the vicinity of the tungsten center (compare Figure 9B and 5 S36B). Specifically, two highly conserved residues 20 have been substituted in FDH2. The 6 introduction of S186 and H187 are noteworthy because they replace H187 and Q188, 7 respectively, in FDH1. These changes are likely to influence Sec reactivity.  DvH-FDH2 will enable strategies for solving problems previously considered intractable 24 in bioenergetics. 25

Construction of fdh Deletion Strains
The naming schemes for the fdh genes follow the 3 convention established previously 56 . DvH deletion strains (Table S1) were constructed 4 using methods already described 125,126 . Briefly, for the deletion of each predicted operon, 5 two plasmids were constructed: one to create a marker-exchange deletion and another 6 to remove the marker. Both plasmids are suicide vectors and require at least one 7 homologous recombination event to occur to provide the selectable phenotypes. A 8 phenotypic screen was performed to determine if a double recombination event took 9 place, thereby increasing the likelihood of choosing isolates that had the desired 10 genotype. Each vector contained a cloned copy of at least 300 bp upstream and a similar 11 DNA region downstream of the operon targeted for deletion that were captured in a 12 vector backbone containing the pUC origin of replication and a gene conferring 13 spectinomycin-resistance. The plasmids were constructed by the sequence and ligation 14 independent cloning (SLIC) technique 127 with amplicons obtained from PCR using the 15 primers found in Table S2 (Integrated DNA Technologies were screened for sensitivity to 100 µg spectinomycin/mL (consistent with the double 26 homologous recombination event), sensitivity to 40 µg 5-fluorouracil/mL (5FU S ; to ensure 27 the counter-selection of 5FU resistance (5FU R ) would be effective) and maintenance of 1 resistance to G418. A putative marker-exchange deletion isolate was then chosen and 2 transformed with the marker-less deletion plasmid, as described above. The transformed 3 cells were recovered, plated on medium containing 5FU and the three phenotypic markers 4 again screened. For the marker-less deletion isolates, however, isolates were selected that 5 were 5FU-resistant and G418-sensitive showing that the marker exchange cassette had 6 been removed from the cell by double homologous recombination. Up to three isolates 7 with the desired antibiotic-resistance phenotype were further analyzed by Southern blot. 8 Once confirmed, one of these isolates was chosen as the marker-less deletion mutant.  Bacterial growth. DvH strains were grown on MOYLS4 medium (see Protocol S1), which 8 was adjusted to pH 7.2 with NaOH. Thioglycolate was added before bottling and after 9 equilibration to dinitrogen (Airgas NI NF200 or research grade). For generating inocula, Once the culture recovered and became densely turbid, transfers were made to fresh 2 medium containing 100 µg/mL spectinomycin HCl. After two rounds of growth with 3 spectinomycin selection, freezer stocks in 10% glycerol were made. For colony selection 4 the same medium supplemented with 1.5% agar, 5 mM cysteine, 1 mM sodium sulfide, 5 and 100 µg/mL spectinomycin was used and kept in gas tight jars with an AnaeroGen 3.5 6 L Gas generating system pack (Oxoid). Colonies were picked into selective medium using 7 a sterile 1 mL syringe (Becton Dickinson 309659) fitted with an 18-gauge needle. 8 10 L carboy growth of DvH-FDH2 producing strain CSR21271 For each carboy, the 9 strain was transferred from 10% glycerol freezer stock in MOYLS4 medium; ~0.5 mL of 10 stock added to a 50 mL bottle of anaerobic MOYLS4 medium, supplemented with vitamins 11 and 100 µg/mL spectinomycin hydrochloride. Transfers were made by nitrogen purged 12 syringe with 23-gauge needles (Becton Dickison 305190). The culture was incubated 13 overnight at 37 C or until mid-exponential phase of growth. 20 mL of the overnight culture 14 was used to inoculate a 500 mL bottle of MOYLS4 medium, containing vitamins and 100 15 µg/mL Spectinomycin HCl. The 500 mL culture was incubated overnight at 37 C. 10 Liters 16 of MOYLS4 medium in 2 L bottles, prewarmed, sterile, aerobic, with iron and EDTA 17 withheld, was poured into a sterile 10 L polypropylene carboy (Thermo 2250-0020). The  under nitrogen pressure ( Figure S37C). The carboy was placed in an incubator (Sanyo 1 MCO-17A1C) at 37 C and the optical density (OD) was monitored at 550 nm (Beckman 2 DU-800 spectrophotometer) via 1 mL samples removed from the same port. Once OD 550 3 nm plateaued, the carboy was chilled in the cold room overnight ( Figure S37D) and then 4 harvested by centrifugation at 8,000 x g (Beckman Avanti HP-26 XPI) in 1 L bottles. Cell 5 pellets were transferred to 50 mL conical centrifuge tubes, respun (7,500 x g; Corning 6 430828), and then froze at -80 C. 7

Biochemistry 8
Protein Expression and Purification. Strep-tag II DvH-FDH2 was purified from strain 9 CSR21271 (see Table S1). Unless specified otherwise all the following steps were done at 10 4 C and under atmospheric conditions. Nitrate, azide, thiols, or formate were not used at 11 any step of the purification or storage. Cells (~ 18 g) were suspended in six volumes of 12 Aldrich 49770) and stored at -80 C for future use. The protein concentration was estimated 1 by BCA assay (Thermo Fisher) versus a BSA standard. 2 Gel Electrophoresis DvH-FDH2 was separated on a Nupage 4-12% Bis-Tris Gel (Thermo 3 Fisher). The running buffer was 1x MES-SDS. The sample was loaded as 5 µL of 12 µM 4 DvH-FDH2 in 62.5 mM Tris-HCl buffer, containing 1.5% SDS, 10% sucrose, 0.0075% 5 bromphenol blue, pre-incubated at room temp (23 o C) for 30 minutes and then heated 5 6 minutes at 50 C. The protein was run alongside Precision plus Kaleidoscope prestained 7 standards (Bio-Rad #1610375) for 100 minutes at 100 V (Invitrogen mini gel talk A25977). 8 The gel was fixed in 40% methanol, 10% acetic acid, stained in 30% methanol, 10% acetic 9 acid, and 0.05% Coomassie blue G-250, and destained in 8% acetic acid. Gels were 10 scanned with a gel doc imager (Bio-Rad). 11 Chromogenic Visualization (In-Gel Assay). DvH-FDH2 was separated on a standard Tris 12 buffered 5% polyacrylamide gel, 2.6% crosslinker gel supplemented with 0.05% triton X- Benzyl viologen assay In this assay BV (colorless) is one-electron reduced to BV+ 16 (purple) 130, 131 by DvH-FDH2. Our workflow is described in Figure S38. was used to perform global fitting of enzyme kinetics data to Schemes 1 and S1. This is 23 based on numerical integration of rate equations. Confidence contour analysis was 24 performed to assess whether the parameters were properly constrained by the data. Electronic FDH2 spectra were collected at 23 o C in 50 mM Tris pH 8.0 using a screw cap 2 1 cm pathlength quartz cuvette (Starna; 1-SOG_10_GL14s). Under aerobic conditions, the 3 spectrum of air equilibrated enzyme was collected first. Then, 10 mM formate was added, 4 and the spectrum was measured again. The cuvette was then capped with silicone septa 5 (Starna GL14/SI) and 10 µL of 2 mM of dithionite was added under argon before collecting 6 a spectrum. For anaerobic measurements, FDH2 was gassed with argon in the sealed 7 cuvette before addition or formate or dithionite. Reduced spectra were also measured 8 using dithionite as the sole reductant (in the absence of formate). phosphate buffer, pH 6 or 7.5) and 13 C-sodium bicarbonate (4.8 mM in 100 mM sodium 10 phosphate buffer, pH 6) were first collected using standard 5 mm thin-walled NMR tubes 11 (Wilmad). 10% D 2 O was used to obtain internal signal lock. Subsequently, 1.3 µM DvH-12 FDH2 was added to the tube containing 13 C-formate (pH 7.5), mixed, and spectra were 13 recollected. Upon completion, 2 mM PES was added to the same tube, mixed, and 14 remeasured. Independently, this process was repeated with 13 C-formate at pH 6. NMR 15 solutions were used throughout. 1 mL reactions were performed at 23 o C in a closed cell 25 using air-saturated 100 mM Tris-HCl, pH 8, containing 1 mM EDTA (Fisher BP120-1). The 26 latter was added to limit adventitious metal ions from mediating O 2 consumption. After 1 obtaining a stable baseline with the buffer, 10 mM formate was added, and the baseline 2 was allowed to stabilize. The reaction was started by the addition of 50 nM DvH-FDH2. 3 Once the O 2 consumption plateaued, 2 µM catalase (Sigma C1345-G) was added. 4 Catalase breaks down H 2 O 2 to water and dioxygen (2H 2 O 2 → 2H 2 O + O 2 ) To test the effect 5 of additives, the order of addition was changed. For example, to test whether DvH-FDH2 6 had catalase activity, H 2 O 2 was added to buffer first, followed by the enzyme. Similarly, 7 to assess the effect of SOD on O2 uptake, SOD was the first component to be added. O 2 8 consumption rates were calculated as described before. 136  pathlength cell (Shimadzu UV-2600i spectrophotometer). We used two-fold higher 23 concentration of acetylated cytochrome c to offset its slightly weaker reactivity with 24 superoxide. 92 The reaction mix (total volume 2.5 mL) was stirred (Cowie 001.1609) at 300 containing cytochrome c, enzyme (1.6 nM final) was added first to obtain the background 3 signal. Subsequently, 10 µM formate was added to start the reaction. Upon completion, 4 ~ 2 mM dithionite was spiked into the mix to estimate the amount of remaining oxidized 5 protein. Controls devoid of formate, enzyme, and cytochrome c were also employed. 6 The effect of superoxide dismutase (10 -100 U / mL) or catalase (100 -400 U / mL) was 7 tested independently. 8

Structural analysis 9
Protein alignments were constructed using MUSCLE or MAFFT. Structural alignments 10 were performed using Chimera v1. 16. Amino acid sequences of the large (DVU2482) and 11 small (DVU2481) subunits of DvH-FDH2 were input together for running structure 12 predictions using a modified version of AlphaFold2.1. 101 Because this algorithm does not 13 recognize Sec, a Cys was substituted and that Tat signal peptide (see Fig. 3A) was not 14 included.
A dedicated Google Colab notebook (AlphaFold.ipynb-Colaboratory 15 (google.com)), which does not utilize homologous structures for making predictions was 16 used with default settings. We also independently predicted the structures of DVU2482