The plant pathogen Pectobacterium atrosepticum contains a functional formate hydrogenlyase‐2 complex

Summary Pectobacterium atrosepticum SCRI1043 is a phytopathogenic Gram‐negative enterobacterium. Genomic analysis has identified that genes required for both respiration and fermentation are expressed under anaerobic conditions. One set of anaerobically expressed genes is predicted to encode an important but poorly understood membrane‐bound enzyme termed formate hydrogenlyase‐2 (FHL‐2), which has fascinating evolutionary links to the mitochondrial NADH dehydrogenase (Complex I). In this work, molecular genetic and biochemical approaches were taken to establish that FHL‐2 is fully functional in P. atrosepticum and is the major source of molecular hydrogen gas generated by this bacterium. The FHL‐2 complex was shown to comprise a rare example of an active [NiFe]‐hydrogenase‐4 (Hyd‐4) isoenzyme, itself linked to an unusual selenium‐free formate dehydrogenase in the final complex. In addition, further genetic dissection of the genes encoding the predicted membrane arm of FHL‐2 established surprisingly that the majority of genes encoding this domain are not required for physiological hydrogen production activity. Overall, this study presents P. atrosepticum as a new model bacterial system for understanding anaerobic formate and hydrogen metabolism in general, and FHL‐2 function and structure in particular.


Introduction
Many members of the γ-proteobacteria are facultative anaerobes with the ability to switch their metabolisms to exploit the prevailing environmental conditions. Aerobic or anaerobic respiration is generally preferred, depending on the availability of respiratory electron acceptors. In this phylum, and specifically under anaerobic conditions, the three-carbon product of glycolysis, pyruvate, is often further metabolised by the oxygen-sensitive pyruvate formatelyase enzyme to generate acetyl CoA and the one-carbon compound formic acid (Pinske and Sawers, 2016). Studies of the model bacterium Escherichia coli have established that endogenously produced formate is initially excreted directly from the cell using a dedicated channel (Suppmann and Sawers, 1994;Hunger et al., 2014;Mukherjee et al., 2017). Under respiratory conditions this formate would be used as an electron donor through the activity of periplasmic enzymes, but under fermentative conditions the formate accumulates in the extracellular milieu until its rising concentration begins to cause a drop in extracellular pH. This is thought to trigger formate re-uptake, which in turn induces synthesis of formate hydrogenlyase (FHL) activity in the cell (Rossmann et al., 1991;McDowall et al., 2014;Sargent, 2016). FHL activity then proceeds to detoxify the formic acid by disproportionation to carbon dioxide and molecular hydrogen (H 2 ).
While FHL activity has been characterised in E. coli (Sargent, 2016), it is not confined to enteric bacteria and has been reported across the prokaryotic domains, including in hyperthermophilic archaea where it is not only involved in pH homoeostasis but also in generating transmembrane ion gradients (Kim et al., 2010;Lim et al., 2014;Bae et al., 2015). The ion-pumping activity stems from an evolutionary link between FHL and the respiratory NADH dehydrogenase Complex I (Bohm et al., 1990;Friedrich and Scheide, 2000;Batista et al., 2013;Schut et al., 2016). Like Complex I, FHL comprises a cytoplasm-facing catalytic domain (termed the peripheral arm in Complex I terminology) linked to an integral membrane arm. In FHL, the peripheral arm contains a [NiFe]-hydrogenase of the 'Group 4' type, which is primarily dedicated to H 2 production (Greening et al., 2015), and is linked by [Fe-S]-cluster-containing proteins to a molybdenum-dependent formate dehydrogenase (Maia et al., 2015). The FHL membrane arm is predicted to take two different forms allowing the enzyme to be further sub-classified as either 'FHL-1' or 'FHL-2' (Friedrich and Scheide, 2000;Marreiros et al., 2013;Sargent, 2016;Finney and Sargent, 2019). The FHL-1 is the predominant archetypal FHL activity of E. coli (McDowall et al., 2014) and comprises [NiFe]-hydrogenase-3 (Hyd-3), formate dehydrogenase-H (FdhF), and a relatively small membrane arm compared to Complex I that contains only two proteins (Fig. 1A). Genes for the much less wellunderstood FHL-2 enzyme are also found in E. coli (Andrews et al., 1997). This isoenzyme is predicted to comprise a [NiFe]-hydrogenase-4 (Hyd-4), an as-yet undefined formate dehydrogenase, and a much larger membrane arm than FHL-1, containing at least five individual integral membrane subunits and more closely resembling the Complex I structure (Fig. 1B). Understanding the structure, function and physiological role of E. coli Hyd-4 and FHL-2 has been hindered by poor native expression levels (Skibinski et al., 2002;Self et al., 2004); a missing important accessory gene from the E. coli hyf cluster (Sargent, 2016); and a lack of consensus on the appropriate experimental conditions to test (Bagramyan et al., 2001;Mnatsakanyan et al., 2004). Thus, in order to bring fresh impetus to understanding the physiology and biochemistry of the FHL-2 complex, it was considered important that an appropriate alternative biological model system was established. In this work, Pectobacterium atrosepticum SCRI1043 was chosen (Bell et al., 2004;Babujee et al., 2012). However, very recently an operon encoding a Hyd-4 isoenzyme was cloned from Trabulsiella guamensis, which is a bacterium previously mistaken for a subspecies of Salmonella (McWhorter et al., 1991), and found to be functional in E. coli (Lindenstrauß and Pinske, 2019).
Pectobacterium atrosepticum SCRI1043 is a phytopathogenic γ-proteobacterium that can grow under anaerobic conditions (Babujee et al., 2012). A global transcriptomic study identified a chromosomal locus (Fig. 1C) that was transcribed under anaerobic conditions in this organism (Bell et al., 2004;Babujee et al., 2012). This locus neatly collects together almost all of the known genes for hydrogen metabolism (Fig. 1C), including genes for a bidirectional Hyd-2-type [NiFe]hydrogenase; genes for specialist metallo-cofactor biosynthesis; a putative formate-responsive transcriptional regulator; a predicted formate dehydrogenase gene; and an 11-cistron operon apparently encoding a Hyd-4 isoenzyme and its associated accessory proteins (Babujee et al., 2012).
In this work, a molecular genetic approach was taken to characterise the hydrogen metabolism locus of P. atrosepticum. A bank of un-marked and in-frame gene deletion mutants was constructed and used to demonstrate unequivocally that the unusual FHL-2 identified in the genome is functional in P. atrosepticum and responsible for the majority of H 2 production under anaerobic conditions. The complex was shown to contain an active Hyd-4 and, unusually, a version of formate dehydrogenase that does not rely on selenocysteine. Surprisingly, it was shown that many of the genes encoding the large membrane arm of FHL-2 can be removed without adversely affecting H 2 production activity. This has potential implications for the molecular architecture of the membrane arm. Overall, this work introduces P. atrosepticum as a tractable model system and presents important genetic, biochemical and physiological characterisation of FHL-2 and [NiFe]-hydrogenase-4.

Results
P. atrosepticum produces molecular hydrogen under anaerobic conditions P. atrosepticum SCRI1043 (Bell et al., 2004) contains the genes for potentially H 2 -evolving enzymes (Babujee et al., 2012). Therefore, the initial goal of this study was to establish the growth conditions under which molecular hydrogen could be evolved. First, the SCRI1043 wild-type strain was grown under anaerobic fermentative conditions in a minimal medium supplemented with 0.8% (w/v) glucose. The culture headspace was sampled at periodic intervals and the amount of H 2 present quantified by gas chromatography (GC). Under these conditions, H 2 evolution activity was found to be temperature dependent, with H 2 accumulation in the headspace observed to be maximal when the phytopathogen was incubated at 20 or 24°C ( Fig. 2A). Taking forward 24°C as standard incubation temperature, H 2 evolution was observed and found to level off after 40 h incubation (Fig. 2B).
When anaerobic respiratory conditions were tested, comprising 0.5% (v/v) glycerol and 0.4% (w/v) nitrate, H 2 production was found to cease with no H 2 detectable after 48 h growth (Fig. 2C). However, replacement of nitrate with 0.4% (w/v) fumarate as a terminal electron acceptor allowed the generation of low, but detectable, levels of H 2 (Fig. 2C). Maximal H 2 production is observable under fermentative conditions (Fig. 2C).
Hyd-4 is the predominant hydrogen-producing enzyme in P. atrosepticum To determine the molecular basis of the observed H 2 production activity, a molecular genetic approach was taken. Initially, the genes encoding the catalytic subunits of the [NiFe]-hydrogenases were targeted. First, a strain PH001 (Table 1) was constructed carrying an unmarked in-frame deletion of the hyfG gene, predicted to encode the catalytic subunit of a Hyd-4 isoenzyme. When cultured fermentatively in the presence of glucose, the PH001 (ΔhyfG) strain produced less than 5% of the total H 2 accumulated by the wild-type control under the same conditions (Fig. 3A). Next, the gene encoding the catalytic subunit of Hyd-2 (hybC) was tested. Mutant strain PH002 (Table 1) was prepared carrying only a ΔhybC allele and, in this case, H 2 evolution under fermentative conditions was essentially indistinguishable from the wild-type strain (Fig. 3A). Finally, a ΔhybC ΔhyfG double mutant (PH003, Fig. 1. Biochemistry and genetics of formate hydrogenlyase. Structural models of (A) formate hydrogenlyase-1 (FHL-1) from Escherichia coli and (B) formate hydrogenlyase-2 (FHL-2) from Pectobacterium atrosepticum. Subunits related at the primary and tertiary levels are coloured similarly. Structural modelling of the formate hydrogenlyases complexes was performed using Phyre 2 predictions of respective subunits (Kelley and Sternberg, 2009). Using Chimera (Pettersen et al., 2004) and the Cryo-EM structure of the Pyrococcus furiosus Membrane Bound Hydrogenase, MBH (PDB: 6CFW), individual FHL-2 subunits were manually assembled. FdhF, which is not present in Pyrococcus furiosus MBH, was positioned principally to align its [4Fe-4S] cluster with that of the surface-exposed [4Fe-4S] cluster from HyfA. C. The genetic organisation of the hydrogen metabolism gene cluster of P. atrosepticum (ECA1225-ECA1252). Predicted gene product functions are indicated and the operon for Hyd-4 is colour coded to match the structure model in panel (B). A. Anaerobic hydrogen production is optimal at lower temperatures. The P. atrosepticum SCRI1043 parent strain was incubated in M9 medium supplemented with 0.8% (w/v) glucose for 168 h at the temperatures indicated before gaseous H 2 accumulation was quantified. B. A time course of H 2 accumulation. P. atrosepticum SCRI1043 was incubated in M9 medium supplemented with 0.8% (w/v) glucose at 24°C and gaseous H 2 accumulation was measured every 24 h. C. P. atrosepticum SCRI1043 was incubated in M9 medium supplemented with either 0.5% (v/v) glycerol and 0.4% (w/v) nitrate ('Gly Nit'); 0.5% (v/v) glycerol and 0.4% (w/v) fumarate ('Gly Fum'); 0.5% (v/v) glycerol only (Gly); or 0.8% (w/v) glucose only ('Glc') at 24°C for 48 h. In all cases, the levels of molecular H 2 in the culture headspace were quantified by GC and normalised to OD 600 and culture volume. Error bars represent SD (n = 3). Table 1) was constructed and was found to be completely devoid of the ability to produce gaseous H 2 (Fig. 3A).
P. atrosepticum can be stably transformed and plasmids encoding either hyfG or hybC were constructed. In the case of PH001 (ΔhyfG) and PH003 (ΔhybC ΔhyfG), H 2 evolution could be rescued in the mutant strains by supplying extra copies of hyfG on a plasmid (Fig. 3B).
Taken altogether, the data presented in Figs 2 and 3 demonstrate that Hyd-4 is responsible for the majority of physiological H 2 production by P. atrosepticum, and that this activity is present under fermentative conditions at temperate growth temperatures ≤ 24°C.

P. atrosepticum contains an active FHL-2 with a selenium-free formate dehydrogenase
Having established that Hyd-4 was active, the next task was to test the hypothesis that Hyd-4 could be part of a wider FHL-2 complex (Fig. 1). First, formate dependence on H 2 production was tested by growing the wildtype parental strain, the PH001 (ΔhyfG) strain, and the PH002 (ΔhybC) strain anaerobically in the presence of increasing amounts of exogenous formate (Fig. 4A). A correlation was observed between the amount of H 2 produced and the amount of formate added to the growth medium, and this was particularly clear when the uptake hydrogenase activity was inactivated ( Fig. 4A). High levels of H 2 production remained dependent upon the presence of an active Hyd-4 ( Fig. 4A), providing initial evidence for a link between formate and H 2 metabolism.
The P. atrosepticum SCRI1043 genome contains a gene encoding a putative formate dehydrogenase close to those for Hyd-4 (Fig. 1C). The gene product shares 85% overall sequence identity with E. coli FdhF but interestingly contains a cysteine residue at position 140 ( Fig. S1), which is occupied by a critical selenocysteine in the E. coli version and other related enzymes (Axley et al., 1991). A mutant strain was therefore constructed (PH004 , Table 1) carrying a ΔfdhF allele. The PH004 (ΔfdhF) strain produced very low, but detectable, levels of H 2 gas under fermentative conditions (Fig. 4B). Addition of a ΔhybC allele to the ΔfdhF strain to generate a double mutant (PH005 ,  Table 1) had no further effect on the amount of H 2 that could be produced (Fig. 4B). It is interesting, however, that a strain devoid of both hybC (Hyd-2 activity) and hyfG (Hyd-4 activity) could not produce any H 2 gas (Fig. 3B). This suggests that the low levels of H 2 evolved from the ΔhybC, ΔfdhF strain are derived from Hyd-4 but that an alternative electron donor may be operating. Notably, two further genes encoding homologs of FdhF are encoded on the P. atrosepticum SCRI1043 chromsome (Bell et al., 2004). The ECA1507 gene encodes a protein with 65% overall sequence identity with FdhF, and the ECA1964 gene encodes a protein with 22% overall sequence identity with FdhF. Deletion of the genes encoding ECA1507 or ECA1964 alone (Table 1) had no influence on the H 2 production capability of the bacterium (Fig. 4C). Moreover, when the genes were supplied in multicopy on an expression vector, neither was able to rescue the phenotype of the ΔfdhF mutant back to native levels of H 2 production (Fig. 4D). However, it is clear that extra levels of ECA1507 in the cell result in a slight increase in H 2 accumulation over the time course of the experiment (Fig. 4D). The increase in H 2 production is statistically significant (P < 0.0001) and suggests ECA1507 could function as an alternative electron donor subunit for P. atrosepticum Hyd-4. Interestingly, extra copies of ECA1964 on a plasmid had the opposite effect (Fig. 4D). In this case, H 2 production in the ΔfdhF strain was pushed to a statistically significant (P < 0.0001) even lower level (Fig. 4D), suggesting ECA1964 was either largely inactive or interfering with the interaction of Hyd-4 with other redox partners.
Taken altogether, these data establish that P. atrosepticum SCRI1043 has functional formate hydrogenlyase activity where molecular hydrogen production is clearly linked to both formate availability and the presence of a formate dehydrogenase gene. Importantly, the predominant electron donor for the reaction is an unusual version of formate dehydrogenase that does not require selenocysteine at its active site, and the enzyme responsible for proton reduction is a [NiFe]-hydrogenase-4.

The role of the FHL-2 membrane arm in hydrogen production
One clear defining structural difference between the FHL-1 type formate hydrogenlyase found in E. coli and the FHL-2 type of P. atrosepticum SCRI1043 is the number of genes encoding components of the membrane arms ( Fig. 1). An FHL-1 enzyme is predicted to contain two different membrane proteins, HycC (related to HyfB in FHL-2) and HycD (related to HyfC) (Fig. 1A). Alternatively, an FHL-2 enzyme is predicted to contain three additional membrane proteins, including HyfE (not present in FHL-1) and two further homologs of HycC/HyfB, namely HyfD and HyfF (Fig. 1B).
To explore the roles of the extra hyfDEF genes located within the FHL-2 locus, mutant strains were constructed (Table 1). First, versions of the ΔhybC strain PH002, lacking either the genes encoding the entire FHL-2 membrane arm (PH007: ΔhybC, ΔhyfBCDEF - Table 1) or lacking the extra membrane components not found in FHL-1 (PH008: ΔhybC, ΔhyfDEF -Table 1) were constructed. In addition, the ΔhybC strain PH002, producing Hyd-4 as the only active hydrogenase, was modified by addition of a 10-His sequence between codons 82 and A. Hyd-4 is responsible for fermentative H 2 production. P. atrosepticum parental strain SCRI1043 and mutants PH001 (ΔhyfG), PH002 (ΔhybC) and PH003 (ΔhybC ΔhyfG) were incubated in M9 medium supplemented with 0.8% (w/v) glucose at 24°C for 48 h. B. Complementation of the mutant phenotype in trans. Strains PH001 (ΔhyfG), PH002 (ΔhybC) and PH003 (ΔhybC ΔhyfG) were separately transformed with plasmids encoding either HyfG or HybC under the control of constitutive promoters. Levels of molecular H 2 in the culture headspace were quantified by GC and normalised to OD 600 and culture volume. Error bars represent SD (n = 3). 83 of the hyfG gene. This new epitope-tagged strain was called PH009 (ΔhybC, hyfG His - Table 1). Finally, versions of PH009 lacking either the genes encoding the entire FHL-2 membrane arm (PH020: ΔhybC, hyfG His , ΔhyfBCDEF - Table 1) or lacking the extra membrane components not found in FHL-1 (PH021: ΔhybC, hyfG His , ΔhyfDEF - Table 1) were constructed.
Deletion of the genes encoding the entire membrane arm reduced the FHL-2-dependent H 2 accumulation levels to around 5% of that observed in the parent strain (Fig. 5A). The addition of the 10-His tag to HyfG allowed the Hyd-4 catalytic subunit to be visualised in whole cell extracts by western immunoblotting (Fig. 5B). The polypeptide was clearly detectable when P. atrosepticum was cultured under anaerobic fermentative conditions (Fig. 5C). Interestingly, the amount of cellular HyfG His was seen to increase when the genes encoding the membrane arm were removed (Fig. 5). This is particularly pertinent for the PH020 strain (ΔhybC, ΔhyfBCDEF), which is essentially devoid of FHL-2 activity (Fig. 5A), since it can be concluded that genetic removal of the complete membrane arm does not destabilise the Hyd-4 catalytic subunit, but instead leads to a physiologically inactive enzyme. It is also notable that in the absence of the genes A. Addition of exogenous formate increases H 2 production. P. atrosepticum parental strain SCRI1043 and mutants PH001 (ΔhyfG) and PH002 (ΔhybC) were incubated in low-salt (5 g l -1 ) LB (LSLB) rich medium supplemented with 0.2% or 0.4% (w/v) formate at 24°C for 48 h. B. The formate dehydrogenase encoded within the gene cluster is responsible for FHL-2 activity. Strains SCRI1043, PH004 (ΔfdhF), PH005 (ΔhybC ΔfdhF) were incubated in M9 medium supplemented with 0.8% (w/v) glucose at 24°C for 48 h. C. Alternative formate dehydrogenase homologues do not have a major role in H 2 production. Strains SCRI1043, PH002 (ΔhybC), PH019 (ΔhybC ΔECA1964), PH028 (ΔhybC ΔECA1507) and PH005 (ΔhybC ΔfdhF) were incubated in M9 medium supplemented with 0.8% (w/v) glucose at 24°C for 48 h. D. Complementation of the mutant phenotype in trans. Strains PH002 (ΔhybC) and PH005 (ΔhybC ΔfdhF) were separately transformed with plasmids encoding either FdhF, ECA1964 or ECA1507 under the control of constitutive promoters. In all cases, the levels of molecular H 2 in the culture headspace were quantified by GC and normalised to OD 600 and culture volume. Error bars represent SD (n = 3). In panel (D) a onetailed t-test was used to determine statistical significance (*P < 0.0001). encoding membrane proteins that the HyfG His protein migrates as two electrophoretic species during SDS-PAGE (Fig. 5B). This is a common observation for catalytic subunits of [NiFe]-hydrogenases as they are synthesised as precursors that undergo proteolytic processing at the C-terminus following cofactor insertion (Bock et al., 2006). In this case, the faster migrating species was calculated as 56.4 kDa, while the slower migrating species was estimated as 62.6 kDa by SDS-PAGE. The predicted molecular mass of HyfG His prior to proteolytic processing is 67,559 Da, and the predicted molecular weight of the 32-residue C-terminal tail that is removed is 3,821 Da.
in the biosynthesis and hydrogen production activity of FHL-2.

A requirement for accessory genes in anaerobic hydrogen production
FHL-2 is a multi-subunit metalloenzyme and assembly of such enzymes is often carefully co-ordinated by dedicated chaperones, sometimes called accessory proteins or 'maturases'. Maturation of molybdenum-dependent formate dehydrogenases has been reported to require the action of an FdhD protein, which is believed to supply an essential sulfur ligand to the active site metal (Arnoux et al., 2015). In P. atrosepticum SCRI1043, fdhD (ECA0093) is not part of the FHL-2 locus but is located elsewhere on the chromosome next to a gene encoding superoxide dismutase (sodA or ECA0092) (Bell et al., 2004). Genetic modification of the PH002 strain, containing only Hyd-4 and FHL-2 activities, by the incorporation of a ΔfdhD allele (PH013: ΔhybC, ΔfdhD - Table 1) led to a defect in physiological H 2 production under fermentative conditions (Fig. 5D). This phenotype could be rescued by the provision of extra copies of fdhD in trans (Fig. 5D).
Maturation of [NiFe]-hydrogenases requires the activity of a network of proteins involved in metal homoeostasis and cofactor maturation and insertion (Sargent, 2016). The P. atrosepticum FHL-2 locus (Fig. 1C) contains a hoxN gene (ECA1252) encoding a putative membrane-bound nickel ion transporter (Eitinger and Mandrand-Berthelot, 2000). Deletion of the hoxN gene in P. atrosepticum SCRI1043 (strain PH011, Table 1) reduced hydrogen evolution levels to around 50% of that observed for the parental strain (Fig. 5D). Note that there is no other homologue of hoxN encoded on the P. atrosepticum SCRI1043 genome, but there are several uncharacterised ABC transporters that could be related to the high-affinity nikA system (Wu et al., 1991), which could account for the continued availability of nickel for hydrogenase biosynthesis in this experiment.
Once inside the cell, nickel is processed into the Ni-Fe-CO-2CN − cofactor through the action of several enzymes and chaperones. One key step in the biosynthesis of the cofactor is the first step in the generation of CNfrom carbamoyl phosphate by HypF (Sargent, 2016). Deletion of the hypF gene from P. atrosepticum (PH010 ,  Table 1), which is located in the hydrogen metabolism gene cluster under investigation here (Fig. 1C), led to the complete abolition of all detectable H 2 evolution (Fig. 5D). The mutant phenotype could be rescued by supply of hypF in trans, but note that full H 2 evolution levels were not restored (Fig. 5D).
Finally, it was observed that a member of the HyfR family of transcriptional regulators was encoded in the hydrogen metabolism gene cluster (Fig. 1C). The HyfR protein is predicted to be related to FhlA, which is a formate-sensing transcriptional activator (Skibinski et al., 2002). A ΔhyfR strain devoid of the HyfR protein has very low formate hydrogenlyase-2 activity (Fig. 5D).
Taken altogether, it can be concluded that all of the genes required for biosynthesis of FHL-2 are functional in P. atrosepticum SCRI1043, which is entirely consistent with the physiological data reported here.

Key differences between FHL-2 and FHL-1
Formate hydrogenlyases can be classified into two structural classes, FHL-1 and FHL-2 (Finney and Sargent, 2019). The most obvious structural difference between an FHL-1, such as the best-characterised E. coli enzyme (McDowall et al., 2014;Pinske and Sargent, 2016), and an FHL-2, such as the P. atrosepticum enzyme characterised here, is the predicted size and composition of the membrane arm (Fig. 1B). Indeed, this large membrane arm is thought to be the ancient progenitor to the ion-pumping membrane arm of respiratory Complex I Marreiros et al., 2013;Yu et al., 2018). Although eukaryotic Complex I, prokaryotic Complex I, and Group 4 hydrogenases such as FHL-1, FHL-2, Ech and MBH are clearly evolutionarily related, the gene and protein names for each type of enzyme are different. Some review articles contain useful tables to highlight the relatedness of the individual subunits (Friedrich and Scheide, 2000;Marreiros et al., 2013;Schut et al., 2016). FHL-1 includes only two membrane proteins, which are a single HycD/HyfC-type protein together with a single HycC/HyfB. This is sufficient to anchor the peripheral arm close to the membrane and, in the case of Thermococcus onnurineus FHL-1 (Lim et al., 2014) and the related Ech hydrogenase from Methanosarcina mazei (Welte et al., 2010), will also allow initial generation of a proton gradient (Yu et al., 2018). Operons encoding FHL-2 complexes encode at least three further integral membrane proteins. In P. atrosepticum these are HyfD and HyfF, which are extra versions of the HycC/HyfB putative ion channels, and the HyfE protein, which is more closely related to a region of NuoK in Complex I. Interestingly, if FHL-2 is modelled based on the Complex I structure (Fig. S2), the extra HyfDEF proteins would be placed between HyfBC, thus separating them and pushing HyfB to the most distal point in the peripheral arm . Alternatively, if FHL-2 is modelled based on the Hyd-4like MBH structure from Pyrococcus furiosus (Yu et al., 2018), then HyfBC remain in contact with each other and HyfDEF form the distal region of the membrane arm (Figs 1 and S2). The experimental evidence presented in this work suggests P. atrosepticum FHL-2 adopts a membrane arm architecture similar to the Pyrococcus furiosus MBH hydrogenase (Fig. 1). This is because removal of all of the extra HyfDEF membrane proteins from FHL-2 had no discernible effect on the physiological activity of the P. atrosepticum system (Fig. 5A), suggesting an active FHL-1-like core enzyme remains. Clearly if HyfB was normally separated from HyfC by the extra proteins they would be unlikely to come together to form a complex when placed in a ΔhyfDEF background, and E. coli FHL-1, for instance, is completely inactive when lacking its HyfB homolog HycC (Pinske and Sargent, 2016). This highlights the principle of modularity in metalloenzyme evolution, since it is clear that the HyfDEF module may be added or removed depending on both selective pressure and also the, as yet undefined in terms of hydrogenases, biochemical function of these membrane proteins (Friedrich and Scheide, 2000). Indeed, it is notable that distal components of the Pyrococcus furiosus MBH membrane arm (MbhABC) could also be genetically removed with only minor effects on cellular hydrogenase activity (Yu et al., 2018). Taken together, this perhaps points to Hyd-3 from FHL-1 as the minimal module of a Group 4 hydrogenase.
Western immunoblotting pointed towards either stabilisation or upregulation of the catalytic subunit HyfG in the absence of hyfDEF or hyfBCDEF (Fig. 5B). This is unlikely to be caused by an accumulation of formate in the cells, perhaps leading to maximal transcription, because the ΔhyfDEF strain retained normal levels of formate hydrogenlyase activity (Fig. 5A). It is more likely that the removal of genes encoding large membrane proteins from immediately upstream of hyfG relaxes some restrictions on the rates of transcription and translation. In bacteria, transcription, translation and membrane insertion of the nascent chain are thought to be coupled together in a process called transertion (Roggiani and Goulian, 2015), and removal of some or all of the elaborate membrane integration step could have an effect on translation of downstream genes.
At native levels, the HyfG His protein can be detected as a single species migrating at 56.4 kDa in SDS-PAGE (Fig. 5B), and occasionally a slower migrating form is detectable migrating at 62.6 kDa (Fig. 5C). These two forms of HyfG His become prominent when the membrane arm of FHL-2 is genetically modified (Fig. 5B). It is known that almost all [NiFe]-hydrogenases are proteolytically processed at their C-termini following successful insertion of the Ni-Fe-CO-2CNcofactor (Bock et al., 2006). In P. atrosepticum HyfG, processing is expected to occur at Arg-546 and would remove 32 amino acids. Thus, in theory, HyfG should be processed from a 67.6 kDa inactive precursor to a 63.8 kDa active mature form. In practice, the motility of HyfG in SDS-PAGE does not match precisely the theoretical values ( Fig. 5B and C); however, only the mature form of HyfG could contribute to physiological formate hydrogenlyase activity.
The P. atrosepticum HyfG catalytic subunit from the Hydrogenase-4 component of FHL-2 shares 74% overall sequence identity (85% similarity) with the E. coli HycE protein from Hydrogenase-3/FHL-1. The sequence variation between these two Group 4A hydrogenases is therefore small with only subtle notable differences. For instance, each protein is known or predicted to undergo cleavage during cofactor insertion and maturation leaving a C-terminal arginine residue in the mature form of the proteins. The cleavage sites themselves are slightly differently conserved in an FHL-1-type enzyme compared to an FHL-2, for example, …R*MTVV… for HycElike proteins compared to …R*VTLV… for HyfG. This may reflect the need for a different maturation protease for each type of hydrogenase, however, this remains to be tested experimentally. In addition, it is notable that both E. coli and P. atrosepticum hyfG initiate translation with a GUG start codon, which may have a role in controlling cellular levels of the enzyme (Belinky et al., 2017).
Phylogenetic analysis of the Group 4A [NiFe]-Hydrogenase subunits, including HycE and HyfG, shows that the enzymes associated with FHL-1 separate into a clearly distinct evolutionary clade from those associated with FHL-2, which form their own distinct clade (Fig. S3). Examples of species that encode both FHL-1 and FHL-2 are rare (Fig. S3).

A selenium-free formate dehydrogenase
Arguably one of the best-studied FdhF enzymes is the E. coli version, which contains selenocysteine at its active site (Axley et al., 1991;Gladyshev et al., 1994;Boyington et al., 1997). Selenocysteine is incorporated co-translationally at a special UGA 'nonsense' codon within the coding sequence (Zinoni et al., 1987), and replacement of selenocysteine with cysteine in the E. coli enzyme resulted in a dramatically reduced turnover number (Axley et al., 1991). One surprising aspect of P. atrosepticum SCRI1043 is that it contains none of the biosynthetic machinery to synthesise selenocysteine (Babujee et al., 2012) and the fdhF gene studied in this work contains a cysteine codon where selenocysteine would be encoded in the E. coli enzyme (Fig. S1). Certainly, the discovery of an active FHL-2 with no need for selenocysteine would benefit scientists interested in engineering this activity into other biological systems. Indeed, an in-frame deletion in the fdhF gene located in the FHL-2 gene cluster (Fig. 1C) resulted in a ~ 500 times reduction in H 2 production (Fig. 4), indicating the majority of H 2 production from P. atrosepticum is dependent on this formate dehydrogenase engaging with Hyd-4 to form an FHL-2 complex. However, the ΔhybC ΔfdhF double mutant still produced low, but quantifiable, levels of H 2 (Fig. 4). Compare that with the behaviours of the ΔhybC ΔfdhF strain (Fig. 3B) and the ΔhypF mutant (Fig. 5C), neither of which produced any detectable H 2 gas. This genetic approach points to the residual H 2 emitting from Hyd-4, perhaps through the activity of alternative electron donors. Certainly for the E. coli FHL, it is known that FdhF is only loosely attached (Boyington et al., 1997) and this may be because the enzyme is 'moonlighting' in other biochemical pathways (Iwadate and Kato, 2019). It raises the possibility that other FdhF-like enzymes in particular could 'plug in' to Hyd-4 and pass excess reducing electrons on to protons. In this work, ECA1507 was found to partially rescue the phenotype of a ΔfdhF strain (Fig. 4D) suggesting it could be an alternative redox partner: note well, however, that the potential substrates and kinetics of ECA1507 cannot be reliably predicted and should be determined empirically.
The FdhF formate dehydrogenase from P. atrosepticum shares 65% overall sequence identity (and 85% similarity) with the well-known E. coli enzyme (Fig. S1). Interestingly, phylogenetic analysis suggests that > 50% of bacterial species that contain FHL genes utilise a cysteinedependent, rather than selenocysteine-dependent, formate dehydrogenase (Fig. S4). P. atrosepticum ECA1507 and ECA1964 were identified here as two FdhF-like proteins that could potentially interact with Hydrogenase-4 to generate novel FHL-like complexes. Sequence analysis revealed ECA1507 and ECA1964 share 65% and 22% overall sequence identity with FdhF, respectively, and phylogenetic analysis determined that ECA1964 is more similar to E. coli YdeP than any other predicted molybdenum dependent oxidoreductases in P. atrosepticum (Fig. S5). YdeP has a putative role in acid resistance in E. coli (Masuda and Church, 2003).

A role for formate metabolism in a plant pathogen
In the potato pathogen P. atrosepticum, FHL-2 activity was found to be expressed at lower growth temperatures (Fig. 2). This suggests that FHL-2 may be produced in planta during the infection or colonisation event. Formate is produced endogenously by enteric bacteria under fermentative conditions, but plants and tubers have multiple metabolic pathways that generate and consume formate. Potato tubers produce a NAD + -dependent formate dehydrogenase (FDH), and the levels of this enzyme are boosted under stress conditions (Hourton-Cabassa et al., 1998). Indeed, proteomic experiments have identified FDH as a differentially produced protein during wound healing in potato tuber slices, with order of magnitude level changes in protein during this process (Chaves et al., 2009). It could be hypothesised that the expression of FDH in the potato tuber could be coordinated with the initial secretion of formate by a fermenting pathogen. Potentially, this would generate NADH from formate in stressed or damaged plant tissues. Recently, it was shown that FDH co-ordinates cell death and defence responses to phytopathogens in Capsicum annum (Bell pepper) (Choi et al., 2014). There is also indication that formate and other molecules that lead to the generation of formate, such as methanol and formaldehyde, induce the production of the NAD + -dependent FDH, perhaps suggesting there is a signalling response to these C1 compounds in plants (Hourton-Cabassa et al., 1998).

Concluding remarks
In this work, P. atrosepticum SCRI1043 has been established as a tractable new model organism for studying hydrogen metabolism in general and FHL function in particular. The organism is a rare example of a bacterium with an active hydrogenase-4-containing FHL-2 complex, however, in the course of this work, hydrogenase-4 activity was reported in T. guamensis, another γ-proteobacterium (Lindenstrauß and Pinske, 2019). Interesting, the T. gaumensis Hyd-4 was found to be active in vivo but very poorly reactive in vitro in standard enzymatic assays with redox-active dyes (Lindenstrauß and Pinske, 2019). This again highlights the need for development of new approaches to characterise FHL-2 and its component parts. In P. atrosepticum, the active hydrogenase-4 enzyme operates in tandem with an unusual selenium-free formate dehydrogenase, which may be more amenable to biotechnological engineering than selenium-dependent isoenzymes. In evolutionary terms, the FHL-2 complex has been discussed as a key intermediate in the evolution of the NADH dehydrogenase (Complex I) from a structurally simpler membrane-bound hydrogenase (Friedrich and Scheide, 2000;Marreiros et al., 2013;Schut et al., 2016). The most obvious difference in the predicted quaternary structures inferred from the genetics is the large membrane arm present in FHL-2 compared to FHL-1, and data presented here points to the extra membrane proteins being not essential for formatedependent hydrogen evolution in vivo. The role of the FHL membrane arm in generating a transmembrane ion gradient remains to be fully explored in enteric bacteria.

Bacterial strains
The parental P. atrosepticum strain used in this study was SCRI1043 (Bell et al., 2004). In-frame deletion and insertion mutants were constructed using pKNG101 suicide vector in E. coli strain CC118λpir (Kaniga et al., 1991;Coulthurst et al., 2006). Briefly, upstream and downstream regions (≥ 600 bp) of the target gene(s) was amplified and inserted into pKNG101 using a three fragment Gibson assembly reaction (HiFi Assembly, NEB). For the insertion of a deca-His encoding sequence into hyfG, primers were designed using the NEBuilder online tool to include the deca-His encoding sequence in the overlapping region of the two fragments containing the respective 3ʹ and 5ʹ sequences of hyfG. After successfully assembly and sequencing of pKNG101 plasmids, the CC118λpir strain with desired plasmid, a HH26 pNJ5000 helper strain, and the desired P. atrosepticum strain were grown in rich media, with antibiotics as necessary. Equal volumes of the stationary phase cultures were mixed and 30 µl was spotted on a non-selective rich media plate for 24 h at 24°C. P. atrosepticum cells with the pKNG101 plasmid were initially selected for on minimal media agar with streptomycin (100 µg ml -1 ). After this, single colonies were re-streaked on the fresh minimal media agar with streptomycin. Co-integrants were then grown to stationary phase in rich medium with no selection before the culture was diluted 1/500 with phosphate buffer. Then, 30 µl of this diluted culture was plated on minimal media agar with sucrose. These colonies were patch screened for sensitivity to streptomycin before PCR screens were performed to check for presence of the desired mutation(s).

Plasmids and complementation
All plasmids were cloned using Gibson assembly (HiFi Assembly, NEB) using DNA amplified from P. atrosepticum SCRI1043 genomic DNA (Table 1). Genes were cloned into pSU-PROM (Kan R ), which includes the constitutive tatA promoter from E. coli (Jack et al., 2004). Complementation plasmids were used to transform electrocompetent P. atrosepticum cells using a 2 mm electroporation cuvette (Molecular BioProduct) with application of an electrical pulse (2.5 kV voltage, 25 µF capacitance, 200 Ω resistance and 2 mm cuvette length) via a Gene Pulsar Xcell electroporator (BioRad). Post recovery, cells were plated on LB Lennox agar plates with 50 µg ml -1 kanamycin.

Hydrogen quantification
Hydrogen was directly quantified from 5 ml cultures grown in sealed Hungate tubes (Pinske and Sargent, 2016). Gas headspace samples were collected using a syringe with Luer lock valve (SGE), Samples were analysed using gas chromatography (Shimadzu GC-2014, capillary column, TCD detector). A hydrogen standard curve was used to quantify sample hydrogen content, this was then normalised to optical density (OD 600 ) and culture volume (Pinske and Sargent, 2016).