The N‐terminal domains of the paralogous HycE and NuoCD govern assembly of the respective formate hydrogenlyase and NADH dehydrogenase complexes

Formate hydrogenlyase (FHL) is the main hydrogen‐producing enzyme complex in enterobacteria. It converts formate to CO 2 and H2 via a formate dehydrogenase and a [NiFe]‐hydrogenase. FHL and complex I are evolutionarily related and share a common core architecture. However, complex I catalyses the fundamentally different electron transfer from NADH to quinone and pumps protons. The catalytic FHL subunit, HycE, resembles NuoCD of Escherichia coli complex I; a fusion of NuoC and NuoD present in other organisms. The C‐terminal domain of HycE harbours the [NiFe]‐active site and is similar to other hydrogenases, while this domain in NuoCD is involved in quinone binding. The N‐terminal domains of these proteins do not bind cofactors and are not involved in electron transfer. As these N‐terminal domains are separate proteins in some organisms, we removed them in E. coli and observed that both FHL and complex I activities were essentially absent. This was due to either a disturbed assembly or to complex instability. Replacing the N‐terminal domain of HycE with a 180 amino acid E. coli NuoC protein fusion did not restore activity, indicating that the domains have complex‐specific functions. A FHL complex in which the N‐ and C‐terminal domains of HycE were physically separated still retained most of its FHL activity, while the separation of NuoCD abolished complex I activity completely. Only the FHL complex tolerates physical separation of the HycE domains. Together, the findings strongly suggest that the N‐terminal domains of these proteins are key determinants in complex assembly.

Formate hydrogenlyase (FHL) is the main hydrogen-producing enzyme complex in enterobacteria. It converts formate to CO 2 and H 2 via a formate dehydrogenase and a [NiFe]-hydrogenase. FHL and complex I are evolutionarily related and share a common core architecture. However, complex I catalyses the fundamentally different electron transfer from NADH to quinone and pumps protons. The catalytic FHL subunit, HycE, resembles NuoCD of Escherichia coli complex I; a fusion of NuoC and NuoD present in other organisms. The C-terminal domain of HycE harbours the [NiFe]-active site and is similar to other hydrogenases, while this domain in NuoCD is involved in quinone binding. The N-terminal domains of these proteins do not bind cofactors and are not involved in electron transfer. As these N-terminal domains are separate proteins in some organisms, we removed them in E. coli and observed that both FHL and complex I activities were essentially absent. This was due to either a disturbed assembly or to complex instability. Replacing the N-terminal domain of HycE with a 180 amino acid E. coli NuoC protein fusion did not restore activity, indicating that the domains have complex-specific functions. A FHL complex in which the N-and C-terminal domains of HycE were physically separated still retained most of its FHL activity, while the separation of NuoCD abolished complex I activity completely. Only the FHL complex tolerates physical separation of the HycE domains. Together, the findings strongly suggest that the N-terminal domains of these proteins are key determinants in complex assembly.
Protein tertiary structure describes the autonomously folding regions of a protein as a domain. Domains can serve different functions, for example binding of cofactors, recognition of a motif or catalytic activity. Evolution has shuffled domains to create the huge diversity of proteins that occur today [1]. Taken on their own, a domain sometimes behaves completely different compared to the holo-protein, as observed for fibril formation by the isolated N-terminal acylphosphatase domain of the HypF hydrogenase maturation factor [2,3]. Artificial fusion proteins have been used for many years as biochemical tools, and in many cases, these fusions have no or little impact on the function of the original protein.
It has been observed that proteins possessing coordinate functions sometimes become fused during evolution, allowing greater efficiency, based on colocation. One prominent example is the alcohol dehydrogenase fusion with acetaldehyde dehydrogenase, which couples two sequential reactions without releasing the toxic intermediate acetaldehyde [4,5]. The most straightforward way for these fusion events to occur is when neighbouring genes of a polycistronic operon are joined through frameshift mutations. Clearly, evolutionary selective pressure is required for the proteins to remain fused, for example when the activity is enhanced or assembly is augmented. Protein complexes with several subunits are likely to assemble via ordered pathways, and natural selection tends to favour gene fusions to optimize assembly [6]. Gene fusion seems also to have happened to the hycE gene of the Escherichia coli hydrogenase three large subunit, HycE. Together with a formate dehydrogenase (FdhH), the HycE hydrogenase forms the catalytic components of the formate hydrogenlyase (FHL) complex in E. coli and other Enterobacteriacea. FHL catalyses the oxidation of formate with the concomitant reduction of protons to produce hydrogen to detoxify formate during mixed-acid fermentation [7]. Along with the FdhH and HycE, three iron-sulfur proteins, including the small subunit HycG, HycF and HycB, form the soluble part of the FHL complex. This large soluble domain is anchored to the membrane by two membrane subunits [8]. The complex is phylogenetically and structurally related to the respiratory complex I, with the NuoCD protein being the homologue of HycE [9][10][11][12]. Although the relationship is less obvious on the primary structural level, where HycE and NuoCD share only 26% (145/563) identical and 43% (244/563) similar residues, it is clearly apparent when the secondary structural elements are compared between the two proteins ( Fig. 1). Complex I, or NADH:ubiquinone oxidoreductase (Nuo), is the primary energy-conserving complex of many respiratory chains and couples NADH oxidation to the translocation of four protons across the membrane [13]. NuoC varies significantly in length in different organisms [14]. Unlike most organisms, which have separate NuoC and NuoD proteins, both are fused in complex I of E. coli and Bacteroides fragilis. Moreover, the NuoC subunit is homologous to the N-terminal domain of HycE and its paralogue in E. coli, HyfG [14]. The crystal structures of both the soluble portion and the entire complex from Thermus thermophilus have been determined [15,16]. NuoC is an orthologue of Nqo5 in T. thermophilus and of the 30-kDa protein in Bos taurus [13]. Unlike its counterpart in HycE, which contains the [NiFe]-active site, NuoD is not known to harbour any cofactors, but it does have a quinone-binding site [17].
The C-terminal domain of HycE extends from amino acids 175-537 and harbours the four conserved cysteinyl residues that coordinate the bimetallic [NiFe]-cofactor required for catalysis (scheme in Fig. 1 and alignment in Fig. 2). These cysteinyl residues are not present in NuoD (Figs 1 and 2). Six pleiotropic Hyp proteins are required for synthesis of the [NiFe]cofactor including its diatomic ligands attached to the Fe atom [18,19]. Of these proteins, the HypC and HypA proteins interact directly with HycE for Fe (CN) 2 CO and Ni 2+ insertion, respectively [20,21]. In addition, HycE has a 32-amino acid C-terminal extension from amino acids 538-569 that is endoproteolytically processed after cofactor insertion and is assumed to facilitate this step by keeping the empty, cofactorfree apo-protein in an open conformation [22]. In this context, the N terminus of the unprocessed hydrogenase large subunit from Thermococcus kodakarensis was found to extend into the nickel-delivery protein HypA, while the C-terminal extension replaces the position of the N terminus prior to proteolytic processing [23]. A recent DFT calculation on HycE identified conserved glutamate residues in both the N-and C-terminal domains, which are proposed to govern the insertion of the Ni 2+ ion of the cofactor, but the function of these residues has yet to be validated experimentally for HycE [24]. This strongly suggests that both domains act in concert for cofactor insertion.
The N-terminal domain of HycE, corresponding to NuoC, extends from amino acids 9-194 and thus is predicted to have a short overlap with the C-terminal domain [25]. This domain/protein bears a cofactor neither in the FHL complex nor in complex I, and its role remains enigmatic.
The N-and C-terminal domains of HycE and NuoCD are predicted to be fused via a long unstructured loop region that contacts the formate dehydrogenase or diaphorase subunits in FHL and complex I, respectively (Fig. 1). Previous work has shown that conserved residues (Glu-138, Glu-140, Asp-143) in the N-terminal domain NuoC are required for the stability of not only the NuoCD protein, but also all other subunits [26]. These residues are also conserved in the HycE protein (Figs 1 and 2). Amino acid exchanges within the N-terminal domain of HycE were better tolerated than exchanges in these conserved residues; for example, the addition of a 10-His residue stretch after Gly-83 was shown to have no effect on the stability and activity of the FHL complex because they are exposed to the surface [27]. Moreover, engineering of the HycE protein revealed that truncation of the C terminus increased H 2 production from the FHL complex, despite the fact that these deletions remove the [NiFe]-binding site [28].
Therefore, in this study, we investigated the possibility of splitting the HycE and NuoCD domains into separate proteins in order to address the role of the N-terminal domain and the potential evolutionary significance of the fusion.

Growth conditions
Bacterial growth was generally carried out either aerobically on LB agar plates and in shaking liquid LB cultures or anaerobically as standing liquid cultures in TGYEP medium, pH 6.5, with 0.8% (w/v) glucose included [29] at 37°C. Alternatively, for complex I activity, aerobic cultures were grown in glycerol medium [1% (w/v) peptone, 0.5% (w/v) yeast extract, 0.8% (w/v) NaCl, 0.4% (w/v) glycerol and separately prepared 15

DNA modifications and strain construction
All strains and plasmids are listed in Table 1 and oligonucleotides in Table 2. For constructing the N-terminal truncation of HycE in strain DhycEN, the 514-bp DNA upstream region up to the encoded methionine start codon was amplified using the oligonucleotides HycEup_XbaI and HycEup_BamHI leaving the ribosome binding site intact, while the downstream 1254 bp was amplified from methionine codon 153 to the stop codon using the oligonucleotides HycEdown_BamHI and HycEdown_EcoRI. The two fragments were digested with XbaI/BamHI and BamHI/EcoRI, respectively, and ligated into pBSK(+) before being moved by XbaI/EcoRI restriction digestion to pMAK705 for chromosomal integration. The chromosomal integration into strain MG1655 was performed according to Ref. [30] and yielded the strain DhycEN. Similarly, the above pMAK705 construct was modified by introducing the nuoC DNA sequence into the BamHI site after amplification with the oligonucleotides nuoC_HA_FW_BamHI and nuoC_RW_BamHI. The directionality of the insert was verified by DNA sequencing, and the integration onto the chromosome of MG1655 was done independently and resulted in strain nuoC-hycE.
Separation of the hycE gene into two distinct genes on plasmid pHycE was done by the NEBaseChanger method (NEB, Ipswich, MA, USA) using the oligonucleotides HycEN+C_FW and HycEN+C_RW and resulted in plasmid pHycEN+C. This construct was subcloned onto pMAK705 and moved to the chromosome as described [30].
In trans complementation was done by cloning the DNA fragment encoding the N-terminal domain including the connecting alpha helix with AA1-168 in pJET1.2 (Thermo  Scientific, Waltham, MA, USA) using the primers HycEup_XbaI and HycEN_RW_EcoRI.
Strain BW25113 Dndh has been described before [31]. This strain was used for further modification of complex I. Deletion of the DNA sequence encoding the NuoC domain was performed by initially replacing the sequence encoding the domain with the cat gene from vector pKD3 (oligonucleotides DnuoC_5 0 _FW and DnuoC_3 0 _RW_KpnI) and subsequently removing the resistance cassette using vector pCP20 as described [32,33]. The DNA sequence encoding Table 1. Strains and plasmids used in this study.

Genotype
Reference  Table 2. Oligonucleotides used in this study.
Oligonucleotide the downstream NuoD domain received a novel ribosome binding site and start codon. The deletion of the entire region from nuoC-L was performed similarly, but using the oligonucleotides DnuoC_5 0 _FW and DnuoLM_3 0 _RW [33].
The nuoC+D strain was constructed by ordering the corresponding region as gBlocks fragment (DNA sequence in Fig. 3), which contained a stop codon after the codon encoding Y169 in nuoC, a ribosome binding site and a start codon for nuoD (IDT DNA, Coralville, IA, USA), cloning it as BamHI/BamHI fragment into pMAK705 and introducing it into BW25113 Dndh.
The pJET-nuoCD plasmid was constructed by amplification of the chromosomal nuoCD gene, including its own ribosome binding site, using the oligonucleotides nuoCD_FW_BamHI and nuoCD_RW_HindIII and Q5 DNA polymerase (NEB) and cloning it blunt-end into pJET1.2 vector according to the manufacturer's instructions (Thermo Scientific). Expression of the gene is under the control of the lacUV5 promoter of the vector.
All authenticity of all DNA sequences in the plasmid inserts and strains was verified by sequencing.

RT-PCR
The RT-PCR was performed according to previously published methods [34]. RNA was isolated from strains in the exponential growth phase using the SV Total RNA Isolation System according to the manufacturer's instructions (Promega, Fitchburg, WI, USA). An extra DNase I digestion step was performed using RQ1 RNase-Free DNase (Promega). The RT-PCR to generate cDNA was performed using M-MLV Reverse Transcriptase, RNase H Minus and random hexamer primers according to the manufacturer's instructions (Promega). The presence of transcripts was tested with gene-specific oligonucleotides, which are listed in Table 2, by using 17 PCR cycles to remain within the semiquantitative part of the amplification reaction.

Western blotting and signal quantification
Aliquots of 50 lg total proteins of anaerobically grown cells were separated on a 10% (w/v acrylamide) SDS/PAGE, transferred to a nitrocellulose membrane and challenged with antibodies raised against HycG (1 : 3000, a kind gift from A. B€ ock/R. G. Sawers) [35,36]. The secondary antibody used was conjugated to HRP enzyme (Bio-Rad, Hercules, CA, USA), and the light signal generated by the ECL reaction (Thermo Scientific) was recorded with a film. The signal intensities were analysed using IMAGEJ software [37].

Enzymatic assays
Hydrogenase activity was determined as H 2 -dependent benzyl viologen (BV) reduction using 50 mM MOPS buffer, pH 7.0, with 4 mM BV in a stoppered 1.6-mL cuvette with 0.8 mL H 2 gas headspace as described [38]. For kinetic determination of FHL activity, cell suspensions were applied to a modified Clark-type electrode (Oxytherm, Hansatech Instruments, Norfolk, UK), which was equipped with an OXY/ ECU module to reverse the polarizing voltage to À0.7 V, where the electrode disc is only sensitive to H 2 (Hansatech Instruments). A volume of 2 mL degassed MOPS buffer, pH 7.0, was used, and the reaction was started with 15 mM formate. The H 2 content of the gas headspace of a 15-mL Hungate tube filled with 7 mL culture was determined by gas chromatography using a GC2010 Plus Gas Chromatograph (Shimadzu, Ky oto, Japan) as described [39].
To measure complex I activity, the NADH-dependent oxidase activity was determined on a Clark-type electrode sensitive to O 2 (Hansatech Instruments) and was performed essentially identical to the H 2 measurements, with the exception that 1.25 mM NADH was used as substrate [31]. Similarly, the succinate oxidase activity was monitored in the presence of 10 mM succinate. Protein concentrations were determined by the method of Ref. [40].

Sucrose gradient centrifugation
The sucrose gradients were prepared and NADH/ferricyanide oxidoreductase activity was determined basically as described [31]. Briefly, membrane extracts from cells grown under the appropriate growth conditions were prepared after cell rupture with an Emulsiflex (EF-C5; Avestin Europe, Mannheim, Germany) in the presence of PMSF (0.1 mM) and DNase at 160 000 g. The sediment was resuspended in buffer A (50 mM MES/NaOH, 50 mM NaCl) with 1% (w/v) DDM and 5 mM MgCl 2 . After solubilization, the membrane fraction was briefly centrifuged and subsequently applied to a freshly poured 24 mL linear sucrose gradient ranging from 5% to 30% (w/v) sucrose in buffer A supplemented with 5 mM MgCl 2 . After centrifugation for 16 h at 140 000 g, the gradient was fractionated into 24 9 1 mL fractions. The NADH/ ferricyanide oxidoreductase activity of 50 lL from each fraction was assayed in 1 mL buffer A containing 1 mM ferricyanide (AppliChem, Darmstadt, Germany). The reaction was started by an addition of 0.2 mM NADH, and the reaction was followed by monitoring the slope at 410 nm in an Ultrospec photometer (Amersham Pharmacia Biotech, Amersham, UK). The rate was calculated using an e 410 of 1 mM À1 Ácm À1 . Protein concentration was determined by biuret assay [41].

Formate hydrogenlyase complex retains activity after separation of the N-and C-terminal domains of HycE
In order to investigate the effect of dividing the HycE protein into two separate domains on the activity and stability of the FHL complex, the efficiency of conversion of formate to H 2 and CO 2 was quantified by measuring the headspace H 2 content of the culture after fermentative growth. This activity represents the total yield of H 2 after growth on glucose medium, but also provides an accurate indication of the activity of the FHL complex, and is particularly useful for assessing low FHL activities. The parental strain produced 7.03 lmol H 2 ÁmL culture À1 ÁOD 600 nm À1 , while H 2 production was basically absent in the negative control strain JW2691, which lacks the hycE gene encoding the catalytic subunit HycE. This strain can be partially (71%) complemented for H 2 production by introducing plasmid pHycE carrying the hycE gene. We have previously observed that hycE cannot fully complement a DhycE strain when provided in trans, and we assume this to be due to the altered chromosomal context [42]. Surprisingly, the DhycE strain can also be complemented by the plasmid pHycEN+C, which consists of the same insert sequence as on pHycE with the exception that it includes a stop codon within hycE (Table 3). This 'split', or separated, hycE gene in the chromosomal context in strain hycEN+C also resulted in H 2 accumulation to a level similar to that of the parental strain. Strain DhycEN, which carried a truncation of the N-terminal domain of HycE, but retained a protein fragment of the C-terminal domain, was unable to evolve H 2 , a phenotype like that of the hycE deletion strain. The DhycEN strain could be complemented in trans, however, by introducing a plasmid encoding the missing N-terminal portion of HycE, which restored almost 50% of the H 2 production (Table 3). Complementation could not be achieved by using a fusion of the related NuoC to the C-terminal domain of HycE. Hence, despite sharing structural similarity, they fulfil distinct roles in both complexes. These assays also showed that the HycE protein, when physically separated into N-and C-terminal domains, was as efficient as the native HycE protein in converting formate to H 2 and CO 2 during mixed-acid fermentation.
In addition to the headspace analysis, which measures accumulation of H 2 over 16 h, cell suspensions were assayed on a modified Clark-type electrode with formate as substrate allowing quantitative measurement of the specific activity of the FHL complex. These measurements verified the absence of activity in strains JW2691 (DhycE), DhycEN and nuoC-hycE. As could be shown for the H 2 -headspace measurement, the pHycE and pHycEN plasmids were functional, but were only able to complement the corresponding deletion strains JW2691 (DhycE) and DhycEN, to 30% and 11% of the level of the wild-type strain, respectively (Table 3). While determination of the accumulated H 2 revealed 100% complementation of H 2 production, the direct FHL assay led to approximately 50% of the FHL activity rate of the parental strain when the domain-split HycE protein was introduced either in trans on plasmid pHycEN+C in strain JW2691 (DhycE) or after chromosomal integration in strain hycEN+C (Table 2). This result revealed that the two physically separated domains of HycE retain half of the activity of the fused, native protein, despite being fully functional in conversion of glucose during growth. Moreover, the findings demonstrate that providing each component in multiple copies did not result in higher activity, indicating that the amount of the two domains in the complex did not limit the reaction rate.
The samples used in the electrode experiments were subsequently used to generate crude extracts, allowing spectrophotometric quantification of total hydrogenase activity, which is a composite of Hyd-1, Hyd-2 and the Hyd-3 component of the FHL complex. Hyd-1 and Hyd-2 are both H 2 -oxidizing hydrogenases, which allow the cell to scavenge some of the FHL-produced H 2 . Under these conditions, Hyd-3 is generally the major contributor to the total hydrogenase activity [42]. A fourth hydrogenase, predicted to form a second FHL complex, is usually not expressed under the conditions used in this study and does not contribute to the activity [43,44]. Here, the residual total hydrogenase activity in the DhycE strain JW2691, representing the combined activities of Hyd-1 and Hyd-2, was approximately 30% of that of the wild-type strain MG1655 (Table 3). Strains DhycEN and nuoC-hycE showed levels of total hydrogenase activity of 17% of the parental, which, surprisingly, is lower than when only Hyd-1 and Hyd-2 are present. Using this assay, the HycE protein separated into two domains complemented to a level of 60% of the wild-type activity, which was slightly higher than the activity recovered by complementation with a plasmid, pHycE13, encoding the native HycE polypeptide. The formate dehydrogenase H activity, which is the other catalytic component of the FHL complex, followed these trends, but with slightly greater variation (Table 4). This indicates that the formate dehydrogenase polypeptide was able to stably interact with the complex carrying the separated domains of HycE.

Modifications of the HycE N-terminal domain do not influence transcription of the hyc operon
Bacterial operons often contain internal promoters and regulatory elements [45]. In the case of the hyc operon, computational evidence predicts a further sigma 70-dependent promoter in front of hycI, the last gene of the operon [46]. Therefore, it was investigated whether the genomic modifications in strains DhycEN and nuoC-hycE, which completely lacked FHL activity, influenced the transcription of the downstream genes. Reverse transcriptase PCR on RNA isolated from those strains in comparison with cDNA obtained from the parental strain MG1655 showed similarly intensive bands for the upstream hycB gene, as well as for the downstream hycF gene (Fig. 4). Despite this analysis not being quantitative, the overall levels of transcripts were similar, as shown for the transcripts of the constitutive mRNA of the tatB gene (Fig. 4). It is concluded, therefore, that the introduced genetic modifications do not alter the transcription of up-and downstream genes within the operon.
Presence of the small subunit protein HycG correlates with the FHL activities Analysis of the hydrogenase small subunit HycG of the FHL complex by quantitative western blotting showed that the protein amount is subject to change [8]. The absence of HycE in the DhycE strain JW2691 caused an approximate 70% reduction in HycG levels compared to the wild-type (Fig. 5). A similar observation for a DhycE strains was made previously [8]. Generally, for hydrogenase assembly to occur, the large subunit must receive its [NiFe]-cofactor and therefore interact with various delivery proteins and undergo endoproteolytic processing [19,47]. Only then is the protein primed for interaction with its small subunit, and this allows the subsequent initiation of the assembly of the entire FHL complex. However, if processing of the large subunit is delayed or prevented, the small subunit is rapidly degraded [47]. This reduced amount of HycG was also observed in the DhycEN strain, both with and without in trans complementation, and the nuoC-hycE fusion strain (Fig. 5). The strains hycEN+C and, to a lesser extent, the complementation of DhycE with pHycE and pHycEN+C show high amounts of HycG, which correlates with their FHL activities. This indicates that the absence of hydrogenase activity, which was observed for some of the HycE variants, correlates with loss of complex integrity and suggests that the N-terminal domain could be the driving force for assembly of the core FeS-carrying proteins. It is notable in this context that the DfdhF strain JW4040, despite lacking FHL activity due to the absence of the electron-input subunit FdhH, showed high amounts of HycG (Fig. 5). As FdhH is only loosely associated with the FHL complex, this result indicates that this subunit is not required for the stable assembly of the core FHL complex [27] and that the core FHL complex remains stable even without electron flow through the complex.

Removal or separation of NuoC causes complete loss of complex I activity
Based on the findings made with HycE, it was decided to test whether the homologous protein of complex I, NuoCD, also revealed similar properties when the two-domain fusion protein was separated into two  distinct proteins. Strain nuoC+D, in which the gene encoding NuoCD was physically separated into two genes, was generated. Additionally, the DNA encoding the NuoC domain was deleted in strain DnuoC to examine the consequences on complex I assembly and activity. All changes were introduced into a strain lacking the ndh gene, which codes for the alternative NADH dehydrogenase, and whose presence would interfere with the NADH oxidase activity assay. As a control, a double-deletion strain DnuoC-L was constructed, which, in addition to the Dndh deletion, was also devoid of most of the genes of the nuo operon, including nuoEFG encoding the diaphorase domain and part of the membrane domain (nuoHIJKL). The functionality of complex I was initially tested by monitoring the NADH oxidase activity, which is catalysed by complex I and the cytochrome bo 3 and cytochrome bd oxidases. The strain Dndh, lacking the alternative NADH dehydrogenase, showed an activity of 0.2 UÁmg À1 (Table 5), which also has been observed in a previous study [31]. The negative control strain lacking complex I (strain DnuoC-L) had only a residual oxidase activity of 4%. The two mutant strains DnuoC and nuoC+D had almost no NADH oxidase activity, which indicated a loss of complex I activity (Table 5). A previous study established that deletion of any of the nuo-genes encoding complex I abolished its ability to transfer electrons into the respiratory chain [48]. In order to rule out a general effect of the mutations on the respiratory chain, succinate oxidase activity was assessed, which uses succinate dehydrogenase, complex II, instead of complex I to channel electrons into the quinone pool. Succinate oxidase activity was very similar in these strains, ranging from 40 to around 60 mU per mg protein ( Table 5), demonstrating that the lack of NADH oxidase activity is specifically due to the absence of complex I.
The detergent-solubilized membrane fractions were subjected to a sucrose gradient centrifugation in order to determine the stability and the assembly of the complex in the mutant strains. The fractions were separately assayed for NADH/ferricyanide oxidoreductase activity, which indicates the presence of the FMN-containing subunit NuoF, which is part of the diaphorase domain of complex I. The activity peaks in fraction 14, corresponding to intact complex I as observed in the Dndh strain (shaded area in Fig. 6). Complex I mutant strain DnuoC-L, which lacks the diaphorase proteins NuoEFG, did not show activity in these fractions. Fractions 5-9 contained a broader, more disperse activity peak, which was originally thought to include the diaphorase activity that had dissociated from the complex. However, because it is also present in the membrane preparations of the DnuoC-L strain, this seems unlikely and rather represents an unspecific reaction of the ferricyanide. Furthermore, the activity profile of the extract from strains DnuoC and nuoC+D resembles the negative control, indicating that NuoF is absent.
Assuming that the mutations could destabilize complex I after extraction from the membrane, but nevertheless allowed retention of a residual activity in whole cells, the strains were subjected to growth experiments  in an in vivo assay (Fig. 7A). The cells were grown aerobically with acetate as sole nonfermentable carbon source, which requires activity of complex I in order to maintain the NADH/NAD + ratio within the cells. This in turn is necessary for the function of the TCA cycle and the glyoxylate shunt to produce amino acids [49]. The growth monitoring during 24 h showed that the Dndh strain had no growth defect compared to the parental strain, as observed before [31]. In contrast, both the DnuoC strain and the nuoC+D strain did not  grow within the first 16 h of the experiment, while eventually the DnuoC strain showed a slight increase in cell density (Fig. 7A). The absence of growth with acetate, as in the DnuoC and nuoC+D strains, typically indicates lack of complex I activity. A complementation of the strains DnuoC and nuoC+D with a plasmid encoding the nuoCD gene, including its ribosome binding site, was attempted (Fig. 7B). However, the growth of the DnuoC or nuoC+D strain on acetate-containing minimal medium could only be restored to a small degree with the nuoCD plasmid, attaining optical densities of 0.18 (nuoC+D), while the parental strain grew to an OD of 0.33. This could indicate a dominant phenotype of the chromosomal nuoCD modification; however, it also does not entirely exclude effects caused by disruption of the nuo operon. All strains were in parallel subjected to growth in glucose medium, where they exhibited no difference in growth behaviour (Fig. 7C). Furthermore, glycerol medium was also tested, where the strains still attained the same final optical density in the stationary phase (Fig. 7D). The Dndh strain and its BW25113 parental derivative grew very similarly, with a growth rate l of 1.45 and 1.74 h À1 , respectively. In contrast, the strains DnuoC, nuoC+D and DnuoC-L had significantly, but similarly, diminished growth rates of 0.54, 0.46 and 0.52 h À1 , respectively. Therefore, all of the growth experiments indicate that the nuoC deletion and separation of nuoCD into two genes abolish complex I activity in whole cells in in vivo experiments. The absence of activity in the DnuoC strain was not surprising, because even single amino acid exchanges in NuoC had a similar effect [26]. What was surprising, however, was the different effects of separating the paralogous genes into two parts in the case of FHL and complex I. The FHL complex tolerated the introduced changes, while complex I presumably did not assemble anymore. The region within the proteins where separation of the domains occurred was based on protein alignments and the presence of the Met153 residue in HycE, which was directly hijacked as initiation codon (see alignment in Fig. 2). However, the amino acid sequence in this region is not as conserved as other regions of the protein, and without further structural information of HycE and NuoCD, it is very difficult to predict the location of the loop region and its function in connecting the modules of the complexes.

Conclusions
The research conducted here on the function of the N-terminal domains of the HycE and NuoCD subunits of the E. coli FHL complex and complex I, respectively, reveals that these domains have been evolutionarily conserved and appear to have an important role in complex assembly. It is still not understood what the regulatory steps in the assembly of these large complexes are, but it is reasonable to assume that assembly must involve a highly ordered sequence of events, where accumulated subcomplexes represent intermediate steps [12,50]. The FeS proteins tend to be unstable in the absence of other components, and thus, the nucleation of the core complex (HycB, E, F, G or NuoB, CD, H, I, L) probably starts with the catalytic subunit HycE and its paralogue NuoCD, respectively. These subunits reside next to the membrane interface, while the formate dehydrogenase or NuoEFG proteins represent the last complex-specific steps in the separate evolution of the two complexes and are located further away from the membrane [12]. We identified the N-terminal domains of the HycE and NuoCD proteins as key determinants in initiating the assembly of this core complex. The different effects of splitting the proteins in two parts show that despite the existing phylogenetic relationship between the FHL and complex I, the latter tolerates splitting of the nuoCD gene less well.