A two-hybrid system reveals previously uncharacterized protein–protein interactions within the Helicobacter pylori NIF iron–sulfur maturation system

Iron–sulfur (Fe–S) proteins play essential roles in all living organisms. The gastric pathogen Helicobacter pylori relies exclusively on the NIF system for biosynthesis and delivery of Fe–S clusters. Previously characterized components include two essential proteins, NifS (cysteine desulfurase) and NifU (scaffold protein), and a dispensable Fe–S carrier, Nfu. Among 38 proteins previously predicted to coordinate Fe–S clusters, two proteins, HP0207 (a member of the Nbp35/ApbC ATPase family) and HP0277 (previously annotated as FdxA, a member of the YfhL ferredoxin-like family) were further studied, using a bacterial two-hybrid system approach to identify protein–protein interactions. ApbC was found to interact with 30 proteins, including itself, NifS, NifU, Nfu and FdxA, and alteration of the conserved ATPase motif in ApbC resulted in a significant (50%) decrease in the number of protein interactions, suggesting the ATpase activity is needed for some ApbC-target protein interactions. FdxA was shown to interact with 21 proteins, including itself, NifS, ApbC and Nfu, however no interactions between NifU and FdxA were detected. By use of cross-linking studies, a 51-kDa ApbC-Nfu heterodimer complex was identified. Attempts to generate apbC chromosomal deletion mutants in H. pylori were unsuccessful, therefore indirectly suggesting the hp0207 gene is essential. In contrast, mutants in the fdxA gene were obtained, albeit only in one parental strain (26695). Taken together, these results suggest both ApbC and FdxA are important players in the H. pylori NIF maturation system.

. Main components of the H. pylori NIF system. a HP number refers to strain 26695 26 . b Based on chromosomal mutagenesis attempts (as found in this study, or as reported by 28,29,34 ). c As reported by 29 . d H. pylori strain specific (as found in this study, and as reported by 34 ). e Before the current study.
HP number a ID Uniprot entry Name (proposed function) Essential? b Amino acids (size in kDa) Interactions

Results and discussion
HP0207, a protein belonging to the Nbp35/ApbC family, is essential in H. pylori. Analysis of the amino-acid sequence of H. pylori HP0207 suggests the protein belongs to the Nbp35/ApbC family, a wellstudied class of proteins with ATPase activity, known to bind and transfer Fe-S clusters in vitro as well as in vivo, as shown in bacteria (Salmonella enterica serovar Typhimurium ApbC) 35,36 , archaea (Methanococcus maripaludis MMP0704) 37,38 and eukarya (Sacharomyces cerevisiae Cfd1 and Nbd35) 39,40 . A sequence alignment of HP0207 and S. Typhimurium ApbC is shown in Fig. 1. Importantly, a Walker A motif (GKGGhGKS/T, with h being hydrophobic residue), described as "deviant" because it contains two conserved lysine residues instead of one, is present in both HP0207 and S.T. ApbC (Fig. 1, shaded box). This motif has previously been shown to be involved in ATP binding and hydrolysis in Salmonella 36 . In addition, both HP0207 and S.T. ApbC proteins contain a ferredoxin-type CxxC motif ( Fig. 1 38 . A survey of 100 sequenced H. pylori genomes showed that the ApbC sequence is highly conserved in H. pylori (98.37-100% identity). As expected, both the deviant Walker A motif and the ferredoxin-type CxxC motif are highly conserved in all H. pylori genomes (data not shown). At this stage, it is worth noting that the hp0207 (apbC) gene is wrongly annotated in many H. pylori genome sequences and several databases (e.g. Genbank, Biocyc), with a putative misassigned ATG start codon located 132 bp upstream of the actual start (see Uniprot for the actual start).
In the present study, we aimed at generating apbC deletion mutants in H. pylori to determine whether the H. pylori ApbC protein plays a role in coordinating and transferring Fe-S clusters, like its Salmonella ApbC homolog. We used a (deletion-insertion) mutagenesis method successfully used in our lab to generate numerous mutants, including most recently nfu mutants 29 . However, repeated attempts to generate hp0207 mutants in three different H. pylori parental strains (i.e. 26695, 43504 and X47) were unsuccessful, suggesting the apbC gene is essential in H. pylori, as previously observed for nifS and nifU 28,29 . Additional experimental approaches will be needed to confirm the essentiality of the H. pylori apbC gene. For instance, generating conditional apbC mutants or constructing mero-diploids apbC strains, as previously done for other essential H. pylori genes, such as nifU and tatC 28,29,44 , could be informative. Nevertheless, a likely reason for the failure to obtain apbC mutants (or nifS and nifU mutants) is the fact that H. pylori only relies on the NIF system to deliver Fe-S clusters to proteins, some of which perform vital roles in the cell; for instance components of the essential NADH:ubiquinone respiratory pathway, such as NuoB (HP1261), Nqo3 (HP1266) and Nqo9 (HP1268). In contrast, mutants in Apbc/   34,47,48 . In C. jejuni, the FdxA homolog (39% identity with HP0277) is iron-induced and plays a role in oxidative stress resistance 48 . In E. coli, FdxA has been shown to be required for incorporation of [4Fe-4S] clusters into two hydrogenases, Hyd-1 and Hyd-2. Consequently, E. coli fdxA mutants are deficient in hydrogen (H 2 )-oxidizing hydrogenase activity 47 . In the present study, attempts were made to disrupt the hp0277 gene, using a previously described method 29 . Whereas we could not obtain any mutant in strains 43504 or X47, fdxA mutants were isolated in strain 26695. The concomitant deletion of the fdxA gene and the insertion of the cat marker in the chromosome of strain 26695 were confirmed by PCR followed by agarose gel analysis (data not shown). The fact that we could only generate fdxA mutants in some, but not all, H. pylori strains was not unexpected, as similar results had previously been reported. Indeed, Mukhopadhyay et al. found a correlation between the metronidazole (Mtz) resistance/sensitivity status of H. pylori strains, and their readiness to have fdxA inactivated 34 . Thus, the fdxA gene can usually be inactivated in strains expressing the frxA nitroreductase gene (hp0642 in strain 26695), i.e., Mtz S strains, such as strain 26695; whereas fdxA can rarely (or never) be inactivated in strains with low or no frxA expression, i.e., Mtz R strains, such as strains 43504 and X47 34 . In fact, the authors of the study showed that fdxA could be disrupted in some Mtz R strains, but only after experimental inactivation of frxA; these strains then became Mtz S . In summary, contrary to other members of the H. pylori NIF maturation system, such as nifS, nifU and apbC, the hp0277 gene encoding for the ferredoxin-like FdxA is dispensable, at least under certain (genetic background) conditions. The 26695 fdxA mutant was further studied; more specifically, we hypothesized hydrogenase activity would be affected in this mutant. Indeed, H. pylori possesses one (H 2 )-oxidizing heterotrimeric hydrogenase, encoded by the hp0631-hp0632-hp0633 genes in strain 26695 (previously known as HydABC, and recently renamed HynABC 49 ). The enzyme is predicted to contain a [4Fe-4S] cluster 29 . We hypothesized the hydrogenase activity would be affected in the fdxA mutant, based on (weak) protein-protein interaction between FdxA and the [4Fe-4S] subunit HP0631 (as shown by BACTH, see below), and deficiency in H 2 -oxidizing hydrogenase activity, as previously reported for the E. coli fdxA mutant 47 . Whole cell H 2 -uptake hydrogenase assays were carried out with H. pylori WT and fdxA mutant cells, using a previously described method 50 . Surprisingly, hydrogenase activity in the fdxA mutant was comparable to that of the WT (71 ± 15 and 59 ± 9 nmoles H 2 per min per 10 9 cells, respectively). Therefore, it does not appear that the FdxA protein is required for hydrogenase [4Fe-4S] cluster. Further characterization of the fdxA mutant will be needed to fully understand the physiological role of the ferredoxin protein in the gastric pathogen.

Use of BATCH system reveals novel interactions involving ApbC and FdxA.
To investigate protein-protein interactions between ApbC or FdxA, and all proteins identified as putative Fe-S clusters-containing proteins, we used a bacterial (E. coli) adenylate cyclase-based two-hybrid system (BACTH) 33 . This system was previously used in our lab to decipher interactions between H. pylori NifS, NifU, Nfu, and 36 proteins predicted to coordinate Fe-S clusters, including ApbC and FdxA 29 . Although E. coli contains an ApbC homolog, it has limited homology with the H. pylori ApbC protein (38% identity / 60% similarity). Thus, we did not anticipate H. pylori HP0207 shares 45% identity and 64% similarity with E. coli Fdx. A conserved ferredoxin CX 2 CX 2 CX 3 C(P) motif is shown in the white box and a ferredoxin-like CX 2 CX 9 CX 3 CP motif (hallmark of the YfhL family) is shown in the shaded box. The sequence alignment was done using BlastP (https:// blast. ncbi. nlm. nih. gov/). www.nature.com/scientificreports/ its presence in the background strain to be a problem for using H. pylori ApbC in the BACTH system. Briefly, interaction between plasmid-encoded T18 and T25 peptides is required to turn on cAMP-CRP dependent operons, such as lac or mal operons. The activation can be monitored on screening media, such as MacConkey-Mal (MC-Mal) or LB-X-Gal, or on a selection medium, such as M63-Mal. In the present study, apbC (hp0207) was cloned in plasmid pKT25, thus generating a T25-ApbC chimeric protein; fdxA (hp0277) was cloned in plasmid pKNT25, thus producing a FdxA-T25 fusion protein. Both genes (e.g. apbC and fdxA), as well as nifS, nifU, nfu and 34 other genes encoding for putative Fe-S-containing proteins, were previously cloned in plasmid pUT18C, generating T18-target fusion proteins 29 . After E. coli cya mutants (BTH101 , Table S1) were co-transformed with both pK(N)T25 and pUT18C derivatives, cells were spotted on MC-Mal and LB-X-Gal plates, which were then incubated for 36 to 48 h at 30 °C under aerobic conditions. Strong interactions between H. pylori proteins yielded red and blue colonies on MC-Mal and LB-X-Gal, respectively (see supplementary Fig. S1 and S2 for ApbC interactions, and Fig. S3 for FdxA interactions). T18-fusions with T25 (vector only) were also included as negative controls in each screening (see supplementary Fig. S4). Results obtained on both chromogenic media were in good agreement (e.g., clones turning red on MC-Mal were blue on LB-X-Gal, while white clones on MC-Mal were also white on LB-X-Gal, see Fig. S1, S2 and S3). ApbC was found to interact with 19 putative Fe-S-containing proteins, including itself (HP0207), NifU (HP0221), FdxA (HP0277) and Nfu (HP1492) ( Fig. S1 and S2). Other proteins strongly interacting with ApbC included the fumarate reductase subunit FrdB (HP0191), the 2-oxoglutarate oxidoreductase subunit OorD (HP0588), the pyruvate-ferredoxin oxidoreductase subunit (HP1109), as well as two components of the essential NADH:ubiquinone respiratory pathway, Nqo3 (HP1266) and Nqo9 (HP1268) (Fig S1 and S2). The hydrogenase subunit HynA (HP0631) and aconitase (HP0779) were also among the T18-protein fusions able to activate the BACTH system in presence of T25-ApbC ( Fig. S1 and S2). Likewise, FdxA was shown to interact with 12 predicted Fe-S-containing proteins, including itself (HP0277), ApbC (HP0207) and Nfu (HP1492) (Fig. S3). Furthermore, proteins such as the dual-specificity RNA methyltransferase RlmN (HP1428) and the [2Fe-2S] Ubiquinol cytochrome c oxidoreductase Rieske (HP1540) were found to interact with FdxA (Fig. S3).

Scientific Reports
Since neither screening medium was sensitive enough to detect weak interactions or to discriminate between weak interactions and negative controls (even after prolonged incubation times), we monitored the growth of (co-transformed) E. coli cells in M63-Mal minimal medium, as previously described 29 . Briefly, the duration and/or strength of T18-protein/T25-protein interactions leads to the cAMP-CRP activation of the mal operon, allowing E. coli cya mutants to use maltose and grow in the minimal medium. The growth (OD 595 ), recorded after 72 h incubation at 30 °C under aerobic condition, was scored as follows: OD 595 ≤ 0.05 (white boxes), no growth, e.g. no detectable interaction, including vector-only negative controls; 0.05 < OD 595 < 0.1 (yellow boxes), weak interactions; 0.1 ≤ OD 595 < 0.2 (blue boxes), intermediate interactions; OD 595 ≥ 0.2 (green boxes), strong interactions (Table 2). Overall, most protein-protein interactions detected on (MC-Mal and LB-X-Gal) chromogenic media correlated with detectable growth (e.g. OD595 > 0.05) in the M63-Mal selection medium, although there were a few discrepancies between media; for instance we observed ApbC-NifU interactions on both MC-Mal and LB-X-Gal ( Fig. S1 and S2) however cells harboring this combination failed to grow in M63-Mal liquid medium (Table 2) Taken together, those results suggest that ApbC is a major actor of the NIF maturation system, since the protein appears to interact with NifS (Table 2), NifU ( Fig. S1 and S2) and with most (29 out of 37) of the Fe-S proteins, including itself (HP0207), FdxA and Nfu (Table 2, Fig. S1 and S2). The number of proteins interacting with FdXA was slightly more limited compared to ApbC's network; nonetheless FdxA was found to interact with NifS and 20 putative Fe-S proteins, including itself (HP0277), ApbC and Nfu. Interestingly, no interaction between NifU and FdxA was observed, suggesting the ferredoxin-like protein FdxA might acquire its [4Fe-4S] cluster content from another source; ApbC is a good candidate, since both proteins appear to strongly interact together, as shown on both solid media ( Fig. S1 and S2), as well as with the M63-Mal liquid growth assay ( Table 2). Table 3 summarizes all interactions between the five proteins identified so far as core components of the NIF maturation pathway (e.g. ApbC, FdxA, Nfu, NifS, NifU), based on results herein as well as previously published data 29 .
A functional ATPase motif is needed for some ApbC-(Fe-S) protein interactions. Previous studies have shown that neither the presence of ATP nor the functional role of the ATPase domain is needed for the [Fe-S] cluster transfer activity of Salmonella ApbC 35 . However, the ATPase activity might be needed for the actual interaction of ApbC with [Fe-S] cluster donors (such as NifU) or with [Fe-S] cluster recipients (such as FdxA, or any other target Fe-S protein). Such possibility was addressed in this study, by mutagenizing the conserved ATPase motif known as the deviant Walker box (GK 106 GGVGK 111 S, see Fig. 1). Site-directed mutagenesis of the apbC gene was carried out, leading to the replacement of both conserved lysine residues (e.g., K106 and K111) by two alanine residues. The mutated gene was cloned in pKT25 and the recombinant plasmid was introduced in E. coli BTH101 (cya mutant), along with each of the previously described pUT18C fusion plasmids 29 . Protein-protein interactions involving the T25-ApbC K106A, K111A variant were investigated using the same approach as that used to study T25-ApbC WT . Results obtained with MC-Mal and LB-X-Gal solid media are shown in Fig. S1 and S2 (alongside T25-ApbC WT interaction results, to allow for direct comparison). Of interest, strong interactions between ApbC WT and some Fe-S target proteins, such as HP0191 (fumarate reductase subu- www.nature.com/scientificreports/ nit) or HP0779 (aconitase), were not present anymore with the variant (Fig. S1 and S2). These decreased levels of interaction were confirmed when co-transformed BTH101 cells were grown in M63-Mal minimal medium growth experiments (Table 1). Overall, alteration of the ApbC ATPase motif resulted in a substantial decrease in protein-protein interactions: indeed, almost 50% of the interactions observed between (T25-) ApbC WT and (Fe-S) target proteins were not observed with the (T25-ApbC K106A, K111A variant ( Table 2). These results strongly suggest that ATP binding and/or hydrolysis is important for H. pylori ApbC's ability to interact with (Fe-S) target proteins.  Fig. S1 and S2), ApbC and Nfu are expected to interact together. To investigate this possibility, purified ApbC-(His) 6 (predicted mass: 41.1 kDa) and Nfu (10.1 kDa) were incubated, by themselves or together, with or without addition of the homobifunctional dimethyl suberimidate (DMS) cross-linker. The protein complexes were subjected to SDS 4-20% gradient PAGE and stained with Coomassie blue (Fig. 3). Upon addition of DMS, Nfu homodimers of approximate mass 20-25 kDa can be seen (Fig. 3, lanes 3 and 4). Incubation of ApbC along with Nfu in the presence of DMS resulted in the formation of a unique adduct (Fig. 3, lane 3) with a molecular mass expected for a heterodimeric ApbC-Nfu complex (calculated mass: approximately 51.2 kDa). The complex was not seen when either of these proteins was tested individually, with or without DMS (Fig. 3, lanes 1, 2, 4 and 5). Interaction between Nfu and the ApbC K106A, K111A variant was not tested in the present study, since it appears the ATPase activity is not required for such interaction. This statement is based upon the fact that (i) the cross-linking reaction was done with purified proteins in absence of ATP; (ii) Nfu interactions with the ApbC K106A, K111A variant appear as strong with ApbC WT , as shown by BACTH ( Table 2, Fig. S1 and S2).
In summary it appears that ApbC and Nfu, both of which are putative Fe-S carriers, intimately interact to form a complex, which is likely involved in Fe-S cluster(s) transfer, although this remains to be experimentally proven. Attempts to show transfer Fe-S from one (holo) protein to the other (apo) protein, using Raman resonance, were unsuccessful (data not shown).   complemented by the previous one, 29 is summarized in Fig. 4. Regarding ApbC, we propose a role in scaffolding and/or transferring Fe-S clusters, based on (i) significant homology between HP0207 and other members of the Nbp35/ApbC previously shown to coordinate and transfer [4Fe-4S] clusters, such as S. Typhimurium ApbC (Fig. 1) or M. maripaludis MMP0704; (ii) protein-protein interactions between HP0207 and numerous Fe-S target proteins, as well as NifS, as revealed by BACTH results in the present study ( Table 2, Fig. S1); and (iii) interactions between ApbC and Nfu, as shown by cross-linking (Fig. 3). The possibility of ApbC being a standalone scaffold relies on the fact that ApbC appears to directly interact with the L-cysteine desulfurase NifS, as suggested by the following lines of evidence. First, using NifS as bait in our previously published BACTH study, we identified ApbC as one of the few interacting partners 29 . Second, the reverse is true: when ApbC was used as bait, as described in the present BACTH study, we identified NifS as interacting partner (Table 1). Attempts were made to reconstitute Fe-S clusters on purified recombinant ApbC, using Azotobacter vinelandii IscS as S-donor (an approach successfully used with HP1492/Nfu in the past 29 ) however they were unsuccessful (data not shown). More work will be required to prove (or disprove) that ApbC can play a role as a standalone scaffold in the NIF pathway. Alternatively, the possibility of an ApbC-NifS-NifU multi-component complex, on which Fe-S recipient proteins would bind to acquire their Fe-S clusters, should not be discarded. Previous results obtained in E. coli are in support of such hypothesis: for instance, the E. coli NifS homolog, IscS, has been shown to form complexes with the NifU homolog, IscU, as well as several other partners, such as ferredoxin 51 , TusA 52 , bacterial frataxin CyaY 53 and the ancillary protein IscX 54,55 . A model involving a ternary complex IscS-IscU-CyaY has even been proposed 55 . Likewise, a FdxA-NifS-NifU multi-component platform is also a possibility in H. pylori, On one hand, both NifS-NifU and NifS-FdxA have been identified (based on 29 and present study); on the other hand, NifU-FdxA interactions have not been identified (in the present study), however one has to remember that the BACTH system only allows for direct (protein-protein) interaction. Hence, a heterotrimeric complex involving FdxA, NifS and NifU (without direct interaction between NifU and FdxA) should not be ruled out.
Whether ApbC is a standalone scaffold or a Fe-S carrier protein, nonetheless it is expected to play a central role in the NIF maturation system, given that its interactome is larger than that of FdxA or Nfu (Fig. 4) and apbC mutants are not viable, in contrast to fdxA and nfu mutants. Since ApbC can interact with NifU and with FdxA, whereas these two latter proteins do not appear to interact with one another (Table 3, Fig.  S2), this observation would support a model where Fe-S clusters flow according to the following scheme: NifU → ApbC → FdxA → (Fe-S) cluster recipient proteins (Fig. 4). More work is needed to better understand NIF-mediated [Fe-S] cluster maturation in bacteria. For instance, it would be interesting to confirm the www.nature.com/scientificreports/ ApbC-FdxA interaction (as suggested by BACTH results in the present study), using purified ApbC and FdxA proteins, and cross-linking. This will be the subject of future studies. The fact that NIF is the only pathway in H. pylori makes the gastric pathogen a unique and attractive bacterial model, hence more research on (Fe-S) cluster homeostasis will be conducted on this organism in the future.

Experimental procedures
Bacterial strains and plasmids. E. coli and H. pylori strains, and plasmids used in this study, are listed in  Construction of H. pylori apbC and fdxA mutants. We attempted to construct apbC or fdxA deletion mutant strains by replacing either apbC (hp0207) or fdxA (hp0277) with a cat (chloramphenicol acetyl transferase) cassette. The strategy relies on a splicing-by-overlap-extension (SOE) polymerase chain reaction (PCR) method. It has been successfully used to construct mutants in our lab, including most recently nfu::cat mutants 29 . Briefly, genomic DNA from H. pylori WT strain 26,695 (Table S1) was used as a template for PCR to amplify fragments of DNA flanking either hp0207 or hp0277 26 . Primers apbC-1 and apbC-2 (Table S2) were used to amplify a 455 bp-long DNA sequence located upstream of apbC and primers apbC-3 and apbC-4 were used to amplify a 410 bp-long sequence located downstream of apbC. Similarly, primers fdxA-1 and fdxA-2 were used to amplify a 555 bp-long DNA sequence located upstream of fdxA and primers fdxA-3 and fdxA-4 were used to amplify a 500 bp-long sequence located downstream of fdxA. Final SOE-PCR amplification steps included the two PCR products obtained for each gene deletion (e.g. apbC or fdxA), a 720 bp-long cat cassette 56 , and primers apbC-1 and apbC-4, or primers fdxA-1 and fdxA-4, respectively. The final PCR products (1585 bp for apbC::cat and 1775 bp for fdxA::cat, respectively) were introduced by natural transformation into various H. pylori parental strains (X47, 43504 and 26695) and cells were plated first on plain BA medium, and then transferred after 12 h on BA supplemented with chloramphenicol (BA-Cm)., No mutant could be recovered after transformation with the PCR product harboring the apbC::cat construct, despite multiple attempts. In contrast, fdxA::cat mutants appeared after 3 to 5 days on BA-Cm plates, but only in strain 26695 26 ; no mutant could be generated in strain 43504 57 or X47 58 . The concomitant deletion of the hp0277 gene and the insertion of the cat marker in the chromosome of the 26695 fdxA::cat mutant were confirmed by PCR, using genomic DNA from mutants as template, and primers fdxA-1 and fdxA-4.
Expression and purification of Nfu and ApbC. The cloning, expression and purification of Nfu has been reported 29 . ApbC was expressed as recombinant hexahistidine-tagged proteins, using E. coli BL21 RIL as host strain. Briefly, primers ApbC-NdeI and ApbC-XhoI (Table S2)  Site-directed mutagenesis of the ApbC ATPase motif. SOE-PCR was used to substitute the two conserved lysine residues (K106 and K111) for alanine residues in the ApbC ATPase motif, as follows. Briefly, genomic DNA from strain 26695 and primers ApbC-XbaI and ApbC-mut1 (Supplementary Table S2) were used to amplify a 500 bp-long DNA sequence containing the apbC sequence upstream of the Walker box, as well as to incorporate a XbaI restriction site (on the 5' end) and both (K → A) mutations (on the 3' end). Likewise, primers ApbC-KpnI and ApbC-mut2 were used to amplify a 825 bp-long DNA sequence containing the apbC sequence downstream of the Walker box, as well as to incorporate both (K → A) mutations (on the 5' end) and a KpnI restriction site (on the 3' end). Finally, both PCR products (with overlapping sequences) were combined, along with primers ApbC-XbaI and ApbC-KpnI, to amplify a 1260 bp-long DNA sequence containing the whole (mutated) apbC sequence. This PCR product was digested with XbaI and KpnI, gel-purified and cloned into similarly digested pKT25, pKNT25, or pUT18C plasmid, for further use with the BACTH system.

Bacterial adenylate cyclase two hybrid (BACTH).
A BACTH-based kit (Euromedex, France) 33 was used to study protein-protein interactions between ApbC WT , or ApbC K106A K111A , or FdxA, and NifS, or 38 Fe-S target proteins (including NifU and Nfu), as previously described 29 . Using primers designed to introduce a XbaI restriction site on the 5' end and a KpnI restriction site on the 3' end, respectively (Table S2), a 1130 bp-long DNA sequence containing the apbC/hp0207 (WT or K106A, K111A mutant) ORF and a 250 bp-long DNA sequence containing the fdxA/hp0277 ORF (without start and stop codons) were PCR-amplified, digested with XbaI and KpnI and ligated into similarly digested pKT25, pKNT25 or pUT18C plasmid to generate in-frame gene fusions (Table S1). All other genes described in this study have been previously cloned into pUT18C plasmid (Table S1) 29 . Ligation mixtures were introduced into E. coli TOP10 and transformants were selected on LB plates supplemented with 100 µg/mL Amp (for pUT18C derivatives) or LB plates supplemented with 30 µg/ mL Kan (for pKT25 or pKNT25 derivatives). Recombinant plasmids were verified by restriction profiles and DNA sequencing. Finally, E. coli BTH101 (cya mutant, Euromedex) cells were co-transformed with a combination of one pUT18C derivative and one pK(N)T25 derivative; co-transformants were selected on LB plates supplemented with both Kan and Amp. Individual colonies were picked and grown overnight at 30 °C in LB supplemented with both antibiotics. This cell suspension was used as inoculum for all subsequent screening and growth experiments. Interactions between Cya T18-and Cya T25-fusion proteins were analyzed using three complementary methods: (1)   . Once the baseline was stable (OD 570 ~ 1), whole cells aliquots (100-200 µL) were added to the mixture, and H 2 oxidation was followed by measuring the reduction (decrease in OD) of oxidized MB at 570 nm. One mole of oxidized H 2 corresponds to 2 mol of reduced MB. Hydrogenase activity is expressed as nmoles H 2 oxidized per min per 10 9 cells. Hydrogenase assays were done in triplicate.