Biotin (vitamin H) is an essential cofactor for numerous biotin-dependent carboxylases in a variety of microorganisms 19. The strict control of biotin biosynthesis is mediated by the bifunctional BirA protein, which acts both as a biotin-protein ligase and a transcriptional repressor of the biotin operon. The consensus binding signal of BirA is a palindromic sequence TTGTAAACC-[N_14/15]-GGTTTACAA 20. Consistent with the presence of the biotin repressor BirA, all bacteria in this study have one or two candidate BirA-binding sites per genome, depending on the operon organization of the biotin genes (Table 1). In the Desulfovibrio species, the predicted BirA site is located between the divergently transcribed biotin operon and the birA gene. In other genomes, candidate binding sites for BirA precede one or two separate biotin biosynthetic loci, whereas the birA gene stands apart and is not regulated.

All δ-proteobacteria studied possess genes for de novo biotin synthesis from pimeloyl-CoA precursor (bioF, bioA, bioD, bioB) and the bifunctional gene birA, but the initial steps of the biotin pathway are variable in these species (Figure 1). The Geobacter species have the bioC-bioH gene pair, which is required for the synthesis of pimeloyl-CoA in Escherichia coli. The Desulfuromonas species contain both bioC-bioH and bioW genes, representing two different pathways of pimeloyl-CoA synthesis. In contrast, D. psychrophila is predicted to synthesize a biotin precursor using the bioC-bioG gene pair, where the latter gene was only recently predicted to belong to the biotin pathway 20. Both Desulfovibrio species have an extended biotin operon with five new genes related to the fatty-acid biosynthetic pathway. Among these new biotin-regulated genes not present in other δ-proteobacteria studied, there are homologs of acyl carrier protein (ACP), 3-oxoacyl-(ACP) synthase, 3-oxoacyl-(ACP) reductase and hydroxymyristol-(ACP) dehydratase. From positional and regulatory characteristics we conclude that these genes are functionally related to the biotin pathway. The most plausible hypothesis is that they encode a novel pathway for pimeloyl-CoA synthesis, as the known genes for this pathway, bioC, bioH, bioG and bioW, are missing in the Desulfovibrio species.

Riboflavin (vitamin B₂) is an essential component of basic metabolism, being a precursor to the coenzymes flavin adenine dinucleotide (FAD) and flavin mononucleotide (FMN). The only known mechanism of regulation of riboflavin biosynthesis is mediated by a conserved RNA structure, the RFN-element, which is widely distributed in diverse bacterial species 21. The δ-proteobacteria in this study possess a conserved gene cluster containing all genes required for the de novo synthesis of riboflavin (ribD-ribE-ribBA-ribH), but lack this regulatory element. The only exception is D. psychrophila, which has an additional gene for 3,4-dihydroxy-2-butanone-4-phosphate synthase (ribB2) with an upstream regulatory RFN element.

Vitamin B₁ in its active form, thiamine pyrophosphate, is an essential coenzyme synthesized by the coupling of pyrimidine (HMP) and thiazole (HET) moieties in bacteria. The only known mechanism of regulation of thiamine biosynthesis in bacteria is mediated by a conserved RNA structure, the THI-element 22. Search for thiamine-specific regulatory elements in the genomes of δ-proteobacteria identified one or two THI-elements per genome that are located upstream of thiamine biosynthetic operons (Figure 1 in Additional data file 1). The δ-proteobacteria possess all the genes required for the de novo synthesis of thiamine (Figure 2) with the exception of Geobacter species, which lack some genes for the synthesis and salvage of the HET moiety (thiF, thiH and thiM), and D. psychrophila, which has no thiF. In most δ-proteobacteria there are two paralogs of the thiamine phosphate synthase thiE, and Geobacter and Desulfuromonas species have fused genes thiED. In D. psychrophila, the only THI-regulated operon includes HET kinase thiM and previously predicted HMP transporter thiXYZ 22, whereas other thiamine biosynthetic genes are not regulated by the THI-element (Figure 2).

In most cases, downstream of a THI-element there is a candidate terminator hairpin, yielding regulation by the transcription termination/antitermination mechanism. The two exceptions predicted to be involved in translational attenuation are THI-elements upstream of genes thiED in Desulfuromonas and thiM in D. psychrophila. In the Desulfovibrio species, the thiSGHFE operon is preceded by two tandem THI-elements, each followed by a transcriptional terminator. This is the first example of possible gene regulation by tandem riboswitches.

Adenosylcobalamin (Ado-CBL), a derivative of vitamin B₁₂, is an essential cofactor for several important enzymes. The studied genomes of δ-proteobacteria possess nearly complete sets of genes required for the de novo synthesis of Ado-CBL (Figure 3). The only exception is the precorrin-6x reductase, cbiJ, which was found only in Desulfuromonas but not in other species. The occurrence of CbiD/CbiG enzymes instead of the oxygen-dependent CobG/CobF ones suggests that these bacteria, consistent with their anaerobic lifestyle, use the anaerobic pathway for B₁₂ synthesis similar to that used by Salmonella typhimurium 23.

Ado-CBL is known to repress expression of genes for vitamin B₁₂ biosynthesis and transport via a co- or post-transcriptional regulatory mechanism, which involves direct binding of Ado-CBL to the riboswitch called the B12-element 2425. A search for B12-elements in the genomes of δ-proteobacteria produced one B12-element in D. desulfuricans, D. psychrophila and G. metallireducens, two in D. vulgaris and G. sulfurreducens, and four in Desulfuromonas (Figure 2 in Additional data file 1). In Geobacter species these riboswitches regulate a large locus containing almost all the genes for the synthesis of Ado-CBL (Figure 3). One B12-element in the Desulfovibrio species regulates both the cobalamin-synthesis genes cbiK-cbiL and the vitamin B₁₂ transport system btuCDF, whereas three such regulatory elements in Desulfuromonas precede different vitamin B₁₂ transport loci. In D. psychrophila, a B12-element occurs within a large B₁₂ synthesis gene cluster and precedes the cbiK-cbiL genes.

The most interesting observation is that genes encoding the B₁₂-independent ribonucleotide reductase NrdDG are preceded by B12-elements in D. vulgaris and Desulfuromonas. Notably, all δ-proteobacteria have another type of ribonucleotide reductase, NrdJ, which is a vitamin B₁₂-dependent enzyme. We propose that when vitamin B₁₂ is present in the cell, expression of the B₁₂-independent isozyme is inhibited, and a relatively more efficient B₁₂-dependent isozyme is used. This phenomenon has been previously observed in other bacterial genomes 26.

The sulfur-containing amino acid methionine and its derivative S-adenosylmethionine (SAM) are important in protein synthesis and cellular metabolism. There are two alternative pathways for methionine synthesis in microorganisms, which differ in the source of sulfur. The trans-sulfuration pathway (metI-metC) utilizes cysteine, whereas the direct sulfhydrylation pathway (metY) uses inorganic sulfur instead. All δ-proteobacteria in this study except the Desulfovibrio species possess a complete set of genes required for the de novo synthesis of methionine (Figure 4). The Geobacter species and possibly Desulfuromonas have some redundancy in the pathway. First, these genomes contain the genes for both alternative pathways of the methionine synthesis. Second, they possess two different SAM synthase isozymes, classical bacterial-type MetK and an additional archaeal-type enzyme 27. Moreover, it should be noted that the B₁₂-dependent methionine synthase MetH in these bacteria lacks the carboxy-terminal domain, which is involved in reactivation of spontaneously oxidized coenzyme B₁₂.

In Gram-positive bacteria, SAM is known to repress expression of genes for methionine biosynthesis and transport via direct binding to the S-box riboswitch 28. In contrast, Gram-negative enterobacteria control methionine metabolism using the SAM-responsive transcriptional repressor MetJ. The δ-proteobacteria in this study have no orthologs of MetJ, but instead, we identified S-box regulatory elements upstream of the metIC and metX genes in the genomes of the Geobacter species and Desulfuromonas (see Figure 3 in Additional data file 1). A strong hairpin with a poly(T) region follows all these S-boxes, implying involvement of these S-boxes in a transcriptional termination/antitermination mechanism.

Both Desulfovibrio species have genes involved in the conversion of homocysteine into methionine (metE, metH and metF), which could be involved in the SAM recycling pathway, but not those genes required for de novo methionine biosynthesis. The ABC-type methionine transport system (metNIQ), which is widely distributed among bacteria, was also not found in these δ-proteobacteria. The Desulfovibrio species appear to have the single-component methionine transporter metT 28.

The amino acid lysine is produced from aspartate through the diaminopimelate (DAP) pathway in most bacteria. The first two stages of the DAP pathway, catalyzed by aspartokinase and aspartate semialdehyde dehydrogenase, are common for the biosynthesis of lysine, threonine, and methionine. The corresponding genes were found in δ-proteobacteria where they form parts of different metabolic operons. Four genes for the conserved stages of the lysine synthesis pathway (dapA, dapB, dapF and lysA) were further identified in δ-proteobacteria, whereas we did not find orthologs for three other genes (dapC, dapE and dapD), which vary in bacteria using different meso-DAP synthesis pathways. The lysine synthesis genes are mostly scattered along the chromosome, and in only some cases are dapA and either dapB, dapF or lysA clustered. All δ-proteobacteria studied lack the previously known lysine transporter LysP. However, in D. desulfuricans and D. psychrophila we found a gene for another candidate lysine transporter, named lysW, which was predicted in our previous genomic survey 29.

In various bacterial species, lysine is known to repress expression of genes for lysine biosynthesis and transport via the L-box riboswitch 30. In addition, Gram-negative enterobacteria use the lysine-responsive transcriptional factor LysR for control of the lysA gene. Among the δ-proteobacteria studied, we found neither orthologs of LysR, nor representatives of the L-box RNA regulatory element. In an attempt to analyze potential lysine regulons in this phylogenetic group, we collected upstream regions of all lysine biosythesis genes and applied SignalX as a signal detection procedure 31. The strongest signal, a 20-bp palindrome with consensus GTGGTACTNNNNAGTACCAC, was observed upstream of the lysX-lysA operons in both Desulfovibrio genomes and the candidate lysine transporter gene lysW in D. desulfuricans (Table 2). The first gene in this operon, named lysX, encodes a hypothetical transcriptional regulator with a helix-turn-helix motif (COG1378) and is the most likely candidate for the lysine-specific regulator role in Desulfovibrio. To find new members of the regulon, the derived profile (named LYS-box) was used to scan the Desulfovibrio genomes. The lysine regulon in these genomes appears to include an additional gene (206613 in D. vulgaris, and 394397 in D. desulfuricans), which encodes an uncharacterized membrane protein with 14 predicted transmembrane segments. We predict that this new member of the lysine regulon might be involved in the uptake of lysine or some lysine precursor.

Iron is necessary for the growth of most bacteria as it participates in many major biological processes 32. In aerobic environments, iron is mainly insoluble, and microorganisms acquire it by secretion and active transport of high-affinity Fe(III) chelators. Under anaerobic conditions, Fe(II) predominates over ferric iron, and can be transported by the ATP-dependent ferrous iron transport system FeoAB. Genomes of anaerobic δ-proteobacteria contain multiple copies of the feoAB genes, and lack ABC transporters for siderophores. Regulation of iron metabolism in bacteria is mediated by the ferric-uptake regulator protein (FUR), which represses transcription upon interaction with ferrous ions. FUR can be divided into two domains, an amino-terminal DNA-binding domain and a carboxy-terminal Fe(II)-binding domain. The consensus binding site of E. coli FUR is a palindromic sequence GATAATGATNATCATTATC 33.

In all δ-proteobacteria studied except D. psychrophila, we identified one to three FUR orthologs that form a distinct branch (FUR_Delta) in the phylogenetic tree of the FUR/ZUR/PerR protein family (see below). One protein, FUR2 in D. desulfuricans, lacks an amino-terminal DNA-binding domain and is either non-functional or is involved in indirect regulation by forming inactive heterodimers with two other FUR proteins. Scanning the genomes with the FUR-box profile of E. coli did not result in identification of candidate FUR-boxes in δ-proteobacteria. In an attempt to analyze potential iron regulons in this phylogenetic group, we collected upstream regions of the iron-transporter genes feoAB and applied SignalX to detect regulatory signals. The strongest signal, a 17-bp palindrome with consensus WTGAAAATNATTTTCAW (where W indicates A or T), was observed upstream of the multiple feoAB operons and fur genes in all δ-proteobacteria except D. psychrophila (Table 3). The constructed search profile (dFUR-box) was applied to detect new candidate FUR-binding sites in these five genomes (Figure 5 and Table 3).

The smallest FUR regulons were observed in the Geobacter and Desulfuromonas species, where they include the ferrous iron transporters feoAB (one to four copies per genome), the fur genes themselves (one copy in the Geobacter species and two copies in Desulfuromonas), and two hypothetical porins. The first one, named psp, was found only in G. metallireducens and Desulfuromonas genomes, where it is preceded by two tandem FUR-boxes. The psp gene has homologs only in Aquifex aeolicus and in various uncultured bacteria, and in one of them (a β-proteobacterium) it is also preceded by two FUR-boxes (GenBank entry AAR38161.1). This gene is weakly similar to the family of phosphate-selective porins (PFAM: PF07396) from various Gram-negative bacteria. The second hypothetical porin was found only in G. sulfurreducens (383590), where it is preceded by a FUR-box and followed by feoAB transporter. This gene, absent in other δ-proteobacteria, has only weak homologs in some Gram-negative bacteria and belongs to the carbohydrate-selective porin OprB family (PFAM: PF04966). Thus, two novel genes predicted to fall under FUR control encode hypothetical porins that could be involved in ferrous iron transport.

Another strong FUR-box in the G. sulfurreducens genome precedes a cluster of two hypothetical genes located immediately upstream of the feoAB-containing operon. The first gene in this operon, named genX (383594), has no orthologs in other bacteria and the encoded protein has a heme-binding site signature of the cytochrome c family (PFAM: PF00034). The second gene, named genY (383592), encodes a two-domain protein that is not similar to any known protein. In Desulfuromonas, an ortholog of the genY amino-terminal domain (391875) is divergently transcribed from a predicted ferric reductase (391874), and their common upstream region contains a strong FUR-box. Moreover, orthologs of the genY C-terminal domain were identified in Desulfovibrio species, where they are again preceded by two tandem FUR-boxes and form a cluster with the hypothetical gene, genZ, encoding a protein of 100 amino acids with two tetratricopeptide repeat domains that are usually involved in protein-protein interactions (PFAM: PF00515). From genomic analysis alone it is difficult to predict possible functions of these new members of the FUR regulon in δ-proteobacteria.

Two Desulfovibrio species have significantly extended FUR regulons that are largely conserved in these genomes and include ferrous iron transporter genes feoAB and many hypothetical genes. Another distinctive feature of the FUR regulon in Desulfovibrio species is a structure of two partially overlapping FUR-boxes shifted by 6 bp. Interestingly, the flavodoxin gene, fld, is predicted to be regulated by FUR in both Desulfovibrio species. In addition to this iron-repressed flavodoxin (a flavin-containing electron carrier), the Desulfovibrio species have numerous ferredoxins (an iron-sulfur-containing electron carrier). One possible explanation is that in iron-restricted conditions these microorganisms can replace ferredoxins with less-efficient, but iron-independent alternatives. A similar regulatory strategy has been previously described for superoxide dismutases in E. coli, Bordetella pertusis and Pseudomonas aeruginosa 343536 and predicted, in a different metabolic context, for B₁₂-dependent and B₁₂-independent enzymes 26; see the discussion above.

Other predicted regulon members with conserved FUR-boxes in both Desulfovibrio species are the hypothetical genes pep (Zn-dependent peptidase), gdp (GGDEF domain protein, PF00990), hdd (metal dependent HD-domain protein, PF01966), and a hypothetical P-type ATPase (392971) that could be involved in cation transport, and a long gene cluster starting from the pqqL gene (Zn-dependent peptidase). The latter cluster contains at least 10 hypothetical genes encoding components of ABC transporters and biopolymer transport proteins (exbB, exbD and tonB). In D. vulgaris, the first gene in this FUR-regulated cluster is an AraC-type regulator named foxR, since it is homologous to numerous FUR-controlled regulators from other genomes (foxR from Salmonella typhi, alcR from Bordetella pertussis, ybtA from Yersinia species, pchR from Pseudomonas aeruginosa), which usually regulate iron-siderophore biosynthesis/transport operons 33. An ortholog of foxR, a single FUR-regulated gene, was identified in D. desulfuricans located about 30 kb away from the FUR-regulated pqqL gene cluster. Given these observations, we propose that this gene cluster is involved in siderophore transport and is regulated by FoxR.

A hypothetical gene in D. vulgaris (209207) has the strongest FUR-box in this genome; however, its orthologs in D. desulfuricans are not predicted to belong to the FUR regulon. Another operon in D. desulfuricans (392971-392970-392969), encoding three hypothetical proteins, is preceded by two candidate FUR-boxes, but these genes have no orthologs in other δ-proteobacteria. Thus, FUR-dependent regulation of these hypothetical genes is not confirmed in other species, and their possible role in the iron homeostasis is not clear.

The transition metal nickel (Ni) is an essential cofactor for a number of prokaryotic enzymes, such as [NiFe]-hydrogenase, urease, and carbon monoxide dehydrogenase (CODH). Two major types of nickel-specific bacterial transporters are represented by the NikABCD system of E. coli (the nickel/peptide ABC transporter family) and the HoxN of Ralstonia eutropha (the NiCoT family of nickel/cobalt permeases). Nickel uptake must be tightly regulated because excessive nickel is toxic. In E. coli and some other proteobacteria, nickel concentrations are controlled by transcriptional repression of the nikABCD operon by the Ni-dependent regulator NikR 37.

The genomes of δ-proteobacteria studied so far contain multiple operons encoding [NiFe] and [Fe] hydrogenases and Ni-dependent CODH, but lack urease genes. Both known types of nickel-specific transporters are absent in δ-proteobacteria, but these genomes contain orthologs of the nickel repressor nikR. In an attempt to identify potential nickel transporters in this taxonomic group, we analyzed the genome context of the nikR genes. The nikR gene in Desulfuromonas is co-localized with a hypothetical ABC transport system, which is weakly homologous to the cobalt ABC-transporter cbiMNQO from various bacteria. Orthologs of this system, named here nikMNQO, are often localized in proximity to Ni-dependent hydrogenase or urease gene clusters in various proteobacteria (data not shown). Among δ-proteobacteria, the Geobacter species have a complete nikMNQO operon, whereas operons in D. desulfuricans and D. psychrophila lack the nikN component but include two additional genes, named nikK and nikL, which both encode hypothetical proteins with amino-terminal transmembrane segments (Figure 6). Desulfovibrio vulgaris has a nikMQO cluster and separately located nikK and nikL genes. Since various other proteobacteria also have the same clusters including nikK and nikL, but not nikN (data not shown), we propose that these two genes encode additional periplasmic components of the NikMQO ABC transporter, possibly involved in the nickel binding.

By applying SignalX to a set of upstream regions of the nikMQO operons, we identified de novo the NikR binding signal in all δ-proteobacteria except D. psychrophila (Table 4). This signal has the same structure as in enterobacteria (an inverted repeat of 27-28 bp), but its consensus (GTGTTAC-[N_13/14]-GTAACAC) differs significantly from the consensus of NikR binding signal of enterobacteria (GTATGAT-[N_13/14]-ATCATAC) 37. Using the derived profile to scan the genomes of δ-proteobacteria we identified one more candidate NikR-binding site in D. desulfuricans. Thus the nickel regulon in this bacterium includes the hydAB2 operon, encoding periplasmic iron-only hydrogenase. Altogether, D. desulfuricas has three paralogs of [NiFe] hydrogenase and two paralogs of [Fe] hydrogenase. We predict that an excess of nickel represses a nickel-independent hydrogenase isozyme using the Ni-responsive repressor NikR. Regulation of hydrogenase enzymes by NikR has not been described previously. A closer look at the upstream region of the putative nickel transport operon in D. psychrophila revealed similar NikR consensus half-sites but in the opposite orientation to each other (GTAACAC-[N_13/14]-GTGTTAC). Searching the genomes with this reversed NikR signal, we observed one more hypothetical gene cluster in D. psychrophila which has two high-scoring NikR-sites in the upstream region, and a NikR-site upstream of the single nikK gene in D. vulgaris (Figure 6).

Zinc is an important component of many proteins, but in large concentrations it is toxic to the cell. Thus zinc repressors ZUR regulate high-affinity zinc transporters znuABC in various bacteria 38. An orthologous zinc transporter was found in δ-proteobacteria (Figure 7). In G. sulfurreducens and the Desulfovibrio species, this cluster also includes a hypothetical regulatory gene from the FUR/ZUR/PerR family, named zur_Gs and zur_D, respectively. Phylogenetic analysis of this protein family demonstrated that ZUR_Gs and ZUR_D are not close relatives and are only weakly similar to known FUR, ZUR, and PerR regulators from other bacteria (see below). The predicted ZUR-binding site located just upstream of the zur-znuABC operon in G. sulfurreducens is highly similar to the ZUR consensus of Gram-positive bacteria (TAAATCGTAATNATTACGATTTA). Another strong signal, a 17-bp palindrome with consensus ATGCAACNNNGTTGCAT, was identified upstream of the znuABC-zur operons in two Desulfovibrio genomes (Table 5). Although znuABC genes are present in all δ-proteobacteria, we observed neither candidate ZUR regulators, nor ZUR-binding sites in G. metallireducens, Desulfuromonas and D. psychrophila, suggesting either the absence of zinc-specific regulation or presence of another regulatory mechanism for these genes.

The previously described cobalt transport system CbiMNQO was found only in the Geobacter species, where it is located within the B₁₂-regulated cbi gene cluster close to the cobaltochelatase gene cbiX, responsible for incorporation of cobalt ions into the corrin ring (see the 'Cobalamin' section above). In contrast, other δ-proteobacteria, possessing a different cobaltochelatase (cbiK), lack homologs of any known cobalt transporter. It was previously suggested by global analysis of the B₁₂ metabolism that different types of cobalt transporters are interchangeable in various bacterial species 26. From genome context analysis and positional clustering with the cbiK gene, we predicted a novel candidate cobalt transporter in δ-proteobacteria, named cbtX (Figure 3), which was previously annotated as a hypothetical transmembrane protein conserved only in some species of archaea (COG3366).

Molybdenum (Mo) is another transition metal essential for bacterial metabolism. Bacteria take up molybdate ions via a specific ABC transport system encoded by modABC genes. Mo homeostasis is regulated by the molybdate-responsive transcription factor ModE, containing an amino-terminal DNA-binding domain and two tandem molybdate-binding domains. Orthologs of ModE are widespread among prokaryotes, but not ubiquitous 39. All δ-proteobacteria have one or more homologs of the modABC transporter (Figure 8). However, full-length modE genes containing both DNA- and molybdate-binding domains were observed only in G. sulfurreducens and Desulfuromonas. In G. sulfurreducens, the molybdate transport operon is co-localized with modE and is preceded by a putative ModE-binding site (Table 6), which is similar to the E. coli consensus of ModE (ATCGNTATATA-[N₆]-TATATANCGAT). In contrast, we could not identify E. coli-type ModE-binding sites upstream of the mod operons in Desulfuromonas, indicating that these operons may be regulated by a different, unidentified signal.

Three other δ-proteobacteria (two Desulfovibrio species and D. psychrophila) have genes encoding a single DNA-binding domain of ModE (Figure 8). Searching with the E. coli-type profile did not reveal candidate binding sites of ModE in these species. To predict potential ModE sites de novo, we collected upstream regions of all molybdate transport operons and applied SignalX. In both Desulfovibrio genomes, we identified a common inverted repeat with consensus CGGTCACG-[N₁₄]-CGTGACCG, which is considerably different from the E. coli consensus of ModE (Table 6 and Figure 8). The modABC gene cluster in these species includes an additional chimeric gene encoding a fusion of phage integrase family domain (PF00589) and one or two molybdate-binding domains (MOP). The functions of these chimeric molybdate-binding proteins, and the mechanism of Mo-sensing by DNA-binding ModE domains in the Desulfovibrio species, are not clear.

Under aerobic conditions, generation of highly toxic and reactive oxygen species such as superoxide anion, hydrogen peroxide and the hydroxyl radical leads to oxidative stress with deleterious effects 40. Strictly anaerobic sulfate-reducing bacteria are adapted to survive in transient oxygen-containing environments by intracellular reduction of oxygen to water using rubredoxin:oxygen oxidoreductase (Roo) as the terminal oxidase 41. The main detoxification system for reactive oxygen species in aerobic and anaerobic bacteria involves superoxide dismutase (Sod), catalase (KatA, KatG) and nonspecific peroxidases (for example, AhpC). In addition to these enzymes, Desulfovibrio species have an alternative mechanism for protecting against oxidative stress, which includes rubredoxin oxidoreductase (Rbo), which has superoxide reductase activity, rubrerythrin (Rbr) with NADH peroxidase activity, and rubredoxin-like proteins (Rub, Rdl), which are used as common intermediary electron donors 42.

Searching for orthologs of the oxidative stress-related genes in the genomes in this study revealed great variability in content and genomic organization (Figure 9). We also searched for homologs of transcription factors known to be involved in regulation of the peroxide and superoxide stress responses. Lacking orthologs of the E. coli OxyR and SoxR/SoxS regulators, the δ-proteobacteria studied have instead multiple homologs of the peroxide-sensing regulator PerR of B. subtilis 43. The PerR-specific branch on the phylogenetic tree of the FUR/ZUR/PerR family contains at least three distinct sub-branches with representatives from δ-proteobacteria (Figure 10). In all cases except D. psychrophila, the perR genes are co-localized on the chromosome with various peroxide stress-responsive genes (Figure 9). However, the upstream regions of these genes contain no candidate PerR-binding sites conforming to the B. subtilis PerR consensus TTATAATNATTATAA. Applying the SignalX program to various subsets of upstream regions of peroxide stress-responsive genes resulted in identification of candidate PerR operators in δ-proteobacteria (Table 7).

In the Desulfovibrio species, a common palindromic signal was found upstream of the perR and rbr2 genes. In D. vulgaris, perR forms an operon with rbr and rdl genes 42. Searching for genes with the derived profile identified additional candidate members of the PerR regulon, alkyl hydroperoxide reductase ahpC in D. vulgaris (D. desulfuricans has no ortholog of ahpC), and a hypothetical gene of unknown function in both Desulfovibrio species (206199 in D. vulgaris and 395549 in D. desulfuricans).

The perR-rbr-roo operon in both Geobacter species is preceded by a conserved palindromic region (Table 7) which overlaps a candidate -10 promoter element (Figure 11). The second perR paralog in G. sulfurreducens (named perR2), which is followed by a gene cluster containing two cytochrome peroxidase homologs (hsc and ccpA), glutaredoxin (grx) and rubrerythrin (rbr), has a close ortholog in the Desulfuromonas species, where it precedes the rbr gene (Figures 9, 10). For these gene clusters we found a common palindromic signal, which is not similar to other predicted PerR signals in δ-proteobacteria (Table 7). Two other perR paralogs in Desulfuromonas (perR2 and perR3) probably result from a recent gene duplication (Figure 10), and both are co-localized on the chromosome with the peroxide stress-responsive genes katG and rbr2, respectively (Figure 9). A common new signal identified upstream of the katG and perR3 genes is probably recognized by both PerR2 and PerR3 regulators in this organism (Table 7).

The PerR regulons in δ-proteobacteria are predicted to include only a small subset of all peroxide stress-related genes identified in these genomes. In addition to the mainly local character of the predicted regulation, these regulons seem to be highly variable between different species, both in their content and DNA signals.

In bacteria, two major mechanisms regulating expression of heat-shock proteins are positive control by alternative sigma factor σ³², encoded by the rpoH gene, and negative control by binding of the repressor protein HrcA to palindromic operators with a consensus TTAGCACTC-[N₉]-GAGTGCTAA called CIRCE 44. The rpoH gene was identified in the genomes of all δ-proteobacteria studied. Though the HrcA/CIRCE system is conserved in very diverse taxonomic groups of bacteria, it is not universal, as some γ-proteobacteria lack it 45. We detected the hrcA genes and CIRCE sites in all genomes studied except D. psychrophila (Table 8).

We then searched the genomes of δ-proteobacteria with previously constructed profiles for σ³² promoters and CIRCE 45. As was observed previously for other bacteria, the only constant member of the HrcA regulon in δ-proteobacteria is the groESL operon. In addition, CIRCE sites are present upstream of the hrcA-grpE-dnaKJ operons in the Geobacter and Desulfuromonas species and upstream of the rpoH gene in G. sulfurreducens. In contrast to the highly conserved CIRCE signal, the σ³² promoters identified in multiple copies in various proteobacteria are less conserved 4546. Among δ-proteobacteria, we identified σ³²-like promoters upstream of some heat-shock-related genes encoding chaperons (GroE, DnaJ, DnaK, GrpE) and proteases (ClpA, ClpP, ClpX, Lon) (Table 9). Thus, in δ-proteobacteria, as in most proteobacteria, σ³² plays a central part in the regulation of the heat-shock response, although detailed regulatory strategies seem to vary in different species. The alternative HrcA/CIRCE system controls expression of groE and other major chaperons.

Growth using carbon monoxide (CO) as the sole energy source involves two key enzymes in the γ-proteobacterium Rhodospirillum rubrum - CO dehydrogenase (CODH) and an associated hydrogenase - which are encoded in the coo operons and induced by the CO-sensing transcriptional activator CooA 47. Among the sequenced δ-proteobacteria, only Desulfovibrio species have coo operons and the CooA regulator. D. vulgaris has two separate operons encoding CODH and the associated hydrogenase, whereas D. desulfuricans has only one operon encoding CODH (Figure 12). The strongest identified signal, a 16-bp palindrome with consensus TGTCGGCNNGCCGACA, was identified upstream of the coo operons from both Desulfovibrio species and R. rubrum (Table 10a). This consensus conforms to the experimentally known CooA-binding region at the R. rubrum cooFSCTJ operon 48.

Sulfate-reducing bacteria are characterized by their ability to utilize sulfate as a terminal electron acceptor. To try to identify the regulatory signals responsible for this metabolism, we applied the signal detection procedure SignalX to a set of upstream regions of genes involved in the sulfate-reduction pathway in Desulfovibrio species. A conserved palindromic signal with consensus sequence TTGTGANNNNNNTCACAA was detected upstream of the sat and apsAB operons, which encode ATP sulfurylase and APS reductase, respectively. This novel signal is identical to the E. coli CRP consensus, and we hypothesized that a CRP-like regulator might control the sulfate-reduction regulon in Desulfovibrio. Scanning the Desulfovibrio genomes resulted in identification of similar sites upstream of many hypothetical genes encoding various enzymes and regulatory systems (Table 10b and Figure 12). One of them, the hcp gene in D. vulgaris, encodes a hybrid-cluster protein (previously called the prismane-containing protein) of unknown function 49, which is coexpressed with a hypothetical ferredoxin gene, named frdX*: new gene names introduced in this study are marked by asterisk. In both Desulfovibrio species, the hcp-frdX* genes are co-localized with a hypothetical regulatory gene from the CRP/FNR family of transcriptional regulators, named HcpR* for the Hcp regulator (Figure 12).

Close HcpR* orthologs were detected in two other δ-proteobacteria, D. psychrophila and Desulfuromonas; however, the same CRP-like signals were not present in their genomes. Examination of a multiple alignment of the CRP/FNR-like proteins revealed one specific amino acid (Arg 180) in the helix-turn-helix motif involved in DNA recognition, which is changed from arginine (for example, in E. coli CRP and Desulfovibrio HcpR*) to serine and proline in these two δ-proteobacteria (data not shown). As both these species have multiple hcp and frdX paralogs, we applied SignalX to a set of corresponding upstream regions and obtained another FNR-like palindromic signal with consensus at ATTTGACCNNGGTCAAAT, which is notably distinct from the CRP-like signal in the third position (which has T instead of G). Such candidate sites were observed upstream of all hcp and frdX paralogs identified in D. psychrophila and Desulfuromonas, as well as upstream of some additional genes in Desulfuromonas, for example those encoding polyferredoxin and cytochrome c heme-binding protein (Table 10 and Figure 12).

The HcpR regulon was also identified in other taxonomic groups, including Clostridium, Thermotoga, Bacteroides, Treponema and Acidothiobacillus species, and in all cases candidate HcpR sites precede hcp orthologs (data not shown). Moreover, the hcpR gene is often co-localized with hcp on the chromosome. In clostridia, frdX orthologs are also preceded by candidate HcpR sites. These data indicate that the main role of HcpR is control of expression of two hypothetical proteins - hybrid-cluster protein and ferredoxin - which are most probably involved in electron transport. However, the HcpR regulon is significantly extended in some organisms. Additional members of this regulon that are conserved between the two Desulfovibrio species include two operons involved in sulfate reduction (apsAB and sat), a hypothetical cluster of genes (206515-206516) with similarity to dissimilative sulfite and nitrite reductases, polyferredoxin, a hypothetical gene conserved in Archaea (209119), and the putative thiosulfate reductase operon phcAB (209106-209105). Notably, both CooA and HcpR candidate sites precede the cooMKLXUHF operon for CODH-associated hydrogenase, which is present only in D. vulgaris.

Because regulators from the CRP/FNR family are able to both repress and activate gene expression, it was interesting to predict the mode of regulation of the HcpR regulon members. To this end, we investigated the positions of candidate HcpR sites in pairwise alignments of orthologous regulatory regions from the two Desulfovibrio species. These two closely related genomes are diverse enough to identify regulatory elements as conserved islands in alignments of intergenic regions. For the sat and apsAB operons, the HcpR sites were found within highly conserved parts of alignments and in both cases the site overlaps the -10 box of a site strongly resembling a promoter (Figure 13a,b), suggesting repression of the genes by HcpR. In contrast, positive regulation by HcpR could be proposed for the hcp-frdX, 206515-206516 and 209119 operons, which have HcpR sites upstream or slightly overlapping the -35 box of predicted promoters (Figure 13c). In the case of the cooMKLXUHF operon in D. vulgaris, the HcpR site is located upstream of the candidate site of the known positive regulator CooA; thus it is also predicted to be an activator site.

By analysis of the functions of genes co-regulated by HcpR, it is difficult to predict the effector for this novel regulon. The physiological role of the hybrid iron-sulfur cluster protein Hcp, the most conserved member of the HcpR regulon, is not yet characterized despite its known three-dimensional structure and expression profiling in various organisms. In two facultative anaerobic bacteria, E. coli and Shewanella oneidensis, the hcp gene is expressed only under anaerobic conditions in the presence of either nitrate or nitrite as terminal electron acceptors 5051. More recent expression data obtained for anaerobic D. vulgaris have showed strong upregulation of the hcp-frdX* and 206515-206516 operons by nitrite stress (J. Zhou, personal communication). While HcpR is predicted to activate these two hypothetical operons, as well as the CODH-associated hydrogenase operon, it most probably represses two enzymes from the sulfate reduction pathway, APS reductase and ATP sulfurylase. We hypothesize that HcpR is a key regulator of the energy metabolism in anaerobic bacteria, possibly controlling the transition between utilization of alternative electron acceptors, such as sulfate and nitrate. The absence of the dissimilatory sulfite reductase DsrAB in the predicted HcpR regulon of Desulfovibrio could be explained by its experimentally defined ability to reduce both sulfite and nitrite 52.