109 out of 503 COG0789 members were selected as metal-sensing based on the presence of at least two out of three cysteine residues required for the cation binding 13. The selected regulators were re-aligned and a phylogenetic tree was constructed (Fig. 1). The branches containing known regulators CadR, ZntR, CueR, HmrR, and MerR can be clearly identified on the tree. Several branches contain no regulators with known specificity. The CueR regulators cluster with HmrR and the PbrR protein (YP_145623) clusters with CadR. There are two MerR branches containing proteins from firmicutes and from proteobacteria. To predict the specificity of regulators that have not been studied in experiment and do not belong to the main branches, we analyzed potential regulatory sites.

There are seven experimentally confirmed binding sites of CueR, HmrR, CadR and ZntR (two sites per regulator except the last one, for which one site is known) 1324.

Despite a small number of sites in the training sets, the derived recognition profiles turned out to be rather selective (data not shown). Further requirement of co-localization with candidate promoters allowed us to make reliable predictions. For example, in Nitrosamonas europaea there are only three high-scoring candidate sites, and only one site upstream of the merT gene is accompanied by an appropriate promoter. We have observed no cases when a candidate site and a promoter occurred upstream of a gene with clearly irrelevant function.

Almost all analyzed loci of COG0789 metal-dependent regulators contained candidate binding sites (Additional file 1).

COG0789-family proteins are widely distributed in proteobacteria and the mercury resistance itself has been also described in Gram-positive species 34. At least 47 merR loci dependent on COG0789 regulators were found in α-, β-, and γ-proteobacterial genomes, whereas the Gram-positive members of this group are the merR operons in Bacillus, Clostridium, Staphylococcus and Streptococcus genomes (Additional file 1, sheet 3 "MerG+"). The merR loci of Proteobacteria are found mainly on transposons and plasmids (28 entries out of 47). In firmicutes, about half of the loci were on the chromosomes (Additional file 1, sheet 3 "MerG+"). Other members of COG0789 have variable localization preferences. For example, zntR-dependent zinc resistance systems and the cueR system have been found only on chromosomes (Additional file 1, sheet 8 "zntR" and sheet 7 "cueR" respectively), whereas hmrR, the α-proteobacterial ortholog of cueR, is also present in the Sinorhizobium meliloti plasmids pSymA and pSymB. While the cadR loci could be seen both on chromosomes and plasmids, a closely related system, pbrR, has been found only on plasmids.

Altogether, GenBank contains about 500 COG0789-related entries, but only a fraction of them are candidate metal-dependent transcriptional regulators. These regulators can be selected by considering specific cysteine residues known to be for ion binding crucial based on experimental data (see Data and Methods). The bioinformatic implementation of this criterion in combination with phylogenetic analysis and analysis of conservation of regulatory sites seems to be sufficient for predicting metal specificity of the studied genomic loci.

To identify new candidate sites when only several examples form the training sample is usually impossible without additional data about the regulatory system. In the case of metal-dependent regulators from COG0789, the specific structure of the regulatory signal which is a combination of a candidate transcription factor-binding site and a promoter, combined with conservation of sites in related genomes, provides for reliable recognition of candidate regulatory signal. A combination of methods allows for non-trivial predictions like the chimerical signal structure in P. rettregeri, Pseudomonas sp. (Q9F3U8) and A. haloplanktis (see Results, MerR), and completely new members of the CueR regulon (Results, CueR).

The computational analysis resulted in gradual improvement of our understanding of the heavy metal resistance systems. The main result of this study is selection of the metal-binding regulators from the general set of COG0789 proteins and assigning several loci with unknown specificity to particular metal exporting system.

The structure of a metal resistance regulon is more diverse than a simple transporter-plus-regulator model. In addition to well described mercury and lead detoxification regulons, the copper regulon also contains more than two genes. Beside known genes encoding cation transporters, possible copper regulon member are glutaminase ybaS and candidate amino acid transporter ybaT 33. The S. typhimurium situation is not clear because its genome contains a CueR paralog (GenBank protein ID AAL19308.1) which has a CueR-type potential binding site, but is regulated by extensive gold concentrations (F. Sonchini, private communication). Conserved candidate CueR sites were found also upstream of predicted cytochrome genes in V. cholerae, V. vulnificus and V. parahaemolyticus. A correct promoter is present upstream of the V. vulnificus and V. parahaemolyticus gene, but not in V. cholerae, making it likely that these genes could be only repressed or derepressed, but not activated, by CueR. The existing annotation of these genes is based on the database similarity search. Their closest relatives characterized in experiment are cytochrome c552 from Marinobacter hydrocarbonoclasticus and cytochrome c544 from Paracoccus sp. 3536. Since there is experimental evidence that cytochrome biosynthesis genes are involved in copper resistance in Pseudomonas fluorescens 37, it is likely that the cytochrome genes in Vibrio spp. are indeed regulated by CueR and have a role in copper resistance, despite the fact that no candidate CueR-binding sites were found in the Ps. fluorescents locus.

Another interesting observation is candidate CueR-binding sites upstream of glutathione-S-transferase genes. In V. vulnificus and Ps. syringiae we have found CueR-like palindromes operons upstream of operons containing the putative gst gene (VV12767-VV12766-VV12765(GST) and secA-argJ-PSPTO4398(GST) respectively), whereas in V. parahaemolyticus and Ps. putida, a candidate CueR-like site has been found upstream of the GST gene itself (VP2086 and Pp3742 respectively). There are experimental data about participation of GST proteins in heavy metal resistance 38 and stability of glutathione-Cu (I) complexes 39. However, the prediction that GST genes are regulated by CueR should be considered as preliminary, especially as other genes in the operons from V. vulnificus and Ps. syringiae do not seem to be involved in heavy metal resistance. One more tentative prediction is the candidate ZntR-binding site upstream of PLU4679 from Ph. luminescens, encoding a homolog of multidrug efflux transporters.

Some more specific non-trivial observations are identification of MerR-binding sites of mixed structure and description of non-orthologous substitutions of cadmium transporters in P. putida and A. lwoffii plasmids.

On the technical side, this study demonstrates that comparative genomic analysis is applicable even to relatively small regulons subject to frequent horizontal transfer.

COG0789 proteins were retrieved from the SMART database (domain accession number SM00422) 52. Multiple sequence alignments were done using the ClustalX program 40. Phylogenetic trees were constructed using the PROML program from the PHYLIP package (implementing the maximum likelihood algorithm) 41. The GenomeExplorer package 42 was used to construct recognition profiles and to identify candidate regulatory sites in genomic sequences.

Positional nucleotide weights in the recognition profiles were defined by:

W(b, k) = log [N(b,k) + 0.5 ] ~ 0.25 ∑_{i = A, C, G, T} log [N(i,k) + 0.5 ]

where N(b,k) denoted the count of nucleotide b at position k. The score of an L-mer candidate site was calculated as the sum of the respective positional nucleotide weights:

Z(b₁...b_L) = ∑_k=1...L W(b_k,k)

42. The promoter profile was constructed using the sample from 13.

We collected all known binding sites of metal-dependent regulators from COG0789 and constructed recognition profiles for several groups of orthologous factors in order to search for suitably arranged candidate regulatory sites and promoters. Candidate regulon member genes were initially identified by similarity search. We selected metal-dependent regulators by the analysis of conserved cysteine residues and tentatively assigned them to specificity groups by the analysis of protein phylogenetic trees. Recognition profiles were constructed for each branch of the tree and used to identify candidate sites. We retained only those sites that co-occurred with candidate promoters having the correct spacer length between the -35 and -10 boxes. Since it has been shown that the length of the spacer is crucial for promoter activation 1, we strictly fixed this parameter during the search for COG0789-type promoters. Dependent on the studied system, the spacer length was either 19 or 20 bp (for recognition profiles see Additional Files, Additional file 2, Promoter recognition matrix).