To describe the phylogenetic history of the Fur regulators, and knowing by previous reports that organisms may have more than one Fur, only completely sequenced actinobacterial genomes present in the NCBI (National Centre for Biotechnology Information) database (March/2007) were used. Since regulatory proteins are small and not highly conserved, the Fur homologues included in this study were chosen based not only on sequence similarity/identity values, but also on the presence of specific residues necessary for the in vivo regulatory activity of Fur. Functional studies on Streptomyces reticuli FurS have shown the importance of five key residues: C96 and C99 are involved in reversible S-S bond formation, Y59 is required for DNA binding and C96, H92 and H93 are implicated in zinc coordination 20. These residues were conserved not only in the closely related mycobacterial FurA, but also in the more distantly related Escherichia coli Fur, making them good indicators to validate the occurrence of a Fur homologue.

An initial BlastP screening against each of the 36 actinobacterial genomes yielded 82 putative homologues. These sequences were aligned with S. reticuli FurS to check for the presence of the above mentioned key residues [see Additional File 1]. Two putative homologues were eliminated at this stage, the Mjls_1895 from Mycobacterium sp. JLS, that lacks all the key residues, and the Rxyl_1224 from Rubrobacter xylanophilus DSM 9941, that lacks one of the histidines. Conversely, Lxx02790, from Leifsonia xyli subsp. xyli str. CTCB07, has an H instead of the Y59. Since histidines and tyrosines are both polar and have similar properties, this homologue was nevertheless retained. Also retained were a number of homologues that presented the two histidines corresponding to H92 and H93 in the form HXH. The remaining 80 sequences were distributed as listed in Table 1: five genomes had four Fur homologues, ten genomes had three, eleven genomes had two, eight genomes had one, and finally two genomes did not present any Fur homologue.

Since multiple fur homologues are present in many genomes, it seems logical that duplication and divergence from a common ancestor gene have occurred during their evolution. To test this hypothesis and to analyse the degree of similarity and identity among the different homologues, a multiple alignment was done using ClustalX [see Additional File 2] and a phylogenetic tree was computed using the Neighbor-Joining (NJ) method (Fig. 1). The distribution of the evolutionary distances and the tree topology strongly suggest that a duplication event took place before the divergence of the actinobacterial lineages, implying that two paralogues were already present in the last common ancestor. Duplication of fur after divergence would have yielded a tree more closely resembling the 16S tree (Fig. 2A). However, this pattern was not observed. Instead, homologues within each genome are almost always separated by a node close to the root, while orthologues from different organisms cluster together, with strong bootstrap support.

We propose a duplication-based model to explain the evolutionary history that took place leading to the various fur genes in the Actinobacteria (Fig. 2B). According to this scenario, the ancestral organism possessed a regulator that had affinity for several metals and that could bind to various DNA sequences. This ancestral fur gene underwent a paralogous duplication event, giving rise to homologues designated as A and B in the Figs. 1 and 2. After the lineage leading to Frankia diverged, the initial fur gene underwent a second paralogous duplication event in this specific lineage, giving rise to the C homologue. Finally, in the progenitor of Streptomyces, the A homologue underwent a third paralogous duplication leading to the D homologue. The E homologues (Fig. 1), which are only present in seven copies, apparently had a different origin and will be discussed below.

These duplication events were followed by specific losses and further duplications that likely modulated the response capacity of each organism to the particular set of evolutionary pressures that characterise its ecological niche. Based on the model described above, one would expect to find one homologue of each kind in each taxon. While the B homologue was generally conserved – there are only two losses recognizable in R. xylanophylus and in Tropheryma spp. – the A homologue was the object of several duplications and losses. This pattern suggests that the selective pressures acting on this homologue were more variable than the ones affecting the B homologue. Seven organisms have two A homologues and one organism has three A homologues (Table 1). Based on the principle of parsimony, and looking at the 16S tree (Fig. 2A), one can postulate four gains to explain these nine additional A homologues (Fig. 2): one in the common ancestor of M. smegmatis, Mycobacterium JLS, Mycobacterium KMS, Mycobacterium MCS and M. vanbaaleni; one in the common ancestor of N. farcinica and Rhodococcus RHA1; one M. smegmatis and another in Nocardioides. Given the synteny surrounding some orthologues of each gain (Fig. 3), duplication events are the most parsimonious explanation. This issue will be further discussed below. On the other hand, thirteen organisms do not retain the A homologue, which can be explained by eight independent gene losses (Fig. 2): one in the common ancestor of B. adolescentis and B. longum; one in the common ancestor of C. diphtheriae, C. efficiens, C. glutamicum and C. jeikeium; one in the common ancestor of Tropheryma whipplei TW08/27 and Tropheryma whipplei Twist; and one in each of S. avermitilis, Frankia CcI3, M. leprae, P. acnes and R. xylanophylus.

Taking the two major groups originating from the first duplication, one can speculate why A homologues are subjected to such strong and diverse selective forces when compared to the more conserved B homologues. As mentioned previously in the introduction, A homologues have been extensively studied in several Actinobacteria, and at least in Mycobacterium and Streptomyces spp. they control the transcription of a downstream catalase-peroxidase, having therefore a major role in the oxidative stress response. Given that this kind of stress is inherent to all oxygen-consuming organisms and able to affect many molecules inside a living cell, it is not surprising to find that anti-oxidant mechanisms and their regulators are quite sensitive to evolutionary pressures related to each specific ecological niche. Since orthologues most likely retain their function across different species, it is reasonable to argue that the A homologues are involved in oxidative stress response regulation. On the other hand, B homologues may either be involved in regulating a cellular function more conserved across organisms, or its involvement in cellular metabolism is not as broad as in a situation of oxidative stress, and therefore these homologues are more stable across organisms throughout geological time.

The E group of sequences likely originated from three HGT (horizontal gene transfer) events: one to the Streptomyces ancestor (originating SCO4180 and SAV_4029), and then from that to the Frankia – Acidothermus common ancestor (originating Acel_0061, Franean1_6149, Francci3_2661 and FRAAL2798) and to Nocardioides (originating Noca_4251) in two separate horizontal transfers. These transfers would include the fur and the surrounding genes which, in parallel with the fur-katG case, could be putatively regulated by Fur constituting an operon. In the common ancestor of Frankia CcI3 and ACN14a, a reshuffling of the genome could be responsible for disrupting of the genomic context, which is maintained among Noca_4251, Acel_0061, Franean1_6149, SCO4180 and SAV4029 (Fig. 3). This hypothesis can explain both the presence and the synteny encountered among these genes, and is supported by their %GC values – FRAAL2798, Francci3_2661, SAV4029 and SCO4180 have a %GC value that deviates from the genome (Table 1). Lastly, with the exception of A. cellulolyticus, all these organisms inhabit soil ecological niches which facilitates the occurrence of HGT.

Finally, the R. xylanophilus homologues are the only ones that are not considered in our model – they seem to have been acquired independently from other Actinobacteria by three individual HGTs. In fact, their %GC values are different both among the sequences and comparing with the genome. A single HGT followed by duplication is also an hypothesis to be considered, especially since it is known that laterally transferred genes have higher rates of duplication 33.

In order to validate the NJ results and to evaluate the strength of the proposed model, a Maximum-Likelihood (ML) tree was computed [see Additional File 3]. The outcome of it corroborates the NJ phylogeny. Despite the fact that the ML tree is not completely resolved, having 14.8% unresolved quartets, it is clear that the groups identified by NJ are maintained, as well as the relations between them. In fact, the C group of homologues emerges from a node close to the root, while the D group of homologues is related to the A group, which suggests that C had its origin by duplication of the ancestor gene while D had its origin on a duplication of the A.

One interesting question regarding evolution of Fur in Actinobacteria is whether the first duplication occurred in the ancestor of the Actinobacteria, or if it was an earlier event that would have appeared in other lineages. To address this question, we randomly selected three different species from each Eubacteria group (except in cases like Acidobacteria where fewer than three genomes are available) and investigated the presence of Fur homologues combining a relaxed BlastP screening against each genome individually with annotation information. The resulting sequences were aligned with the Fur homologues from the Actinobacteria [see Additional File 4] and a NJ tree was computed [see Additional File 5]. The first bifurcation of this tree divides it into two major groups: one of them contains the actinobacterial A homologues and the other contains the actinobacterial B homologues, indicating that the initial duplication occurred in the eubacterial common ancestor. This tree also supports a common origin for the E group of homologues, which are clustered together and separated from the other actinobacterial groups. Multiple origins for this group, either by HGT or duplication, would result in the distribution of these homologues all through the tree. Their clustering shows that either they resulted from a common duplication that was lost in all the other Actinobacteria, or, more parsimoniously, that they resulted from the described HGT. The R. xylanophylus homologues appeared later, which results in a scattered distribution in the tree, supporting an origin in independent HGT events.

Interestingly, previously published work concerning the evolutionary history of Fur and other iron and manganese-responsive transcriptional regulators in alphaproteobacteria 34 indicates that most of bacteria in this lineage have only one Fur, involved in iron homeostasis regulation, which evolved towards a manganese-homeostasis regulator (Mur) in Rhizobiales and Rhodobacteraceae. However, and in the same study, another regulator present in some of the alphaproteobacteria and named Irr was characterised and considered to be part of the Fur superfamily. Putative Irr-binding sites have been found upstream of genes encoding iron-homeostasis and iron-containing proteins, in particular catalase-peroxidases, suggesting that this regulator might functionally correspond to the actinobacterial oxidative stress-related Fur. Although Rodionov et. al (2006) have studied the Fur/Mur phylogeny separately from the Irr one, the fact that they are placed in the same superfamily suggests a common origin, thus supporting the hypothesis that the first Fur duplication occurred in the eubacterial common ancestor, from which a group related with oxidative stress response has emerged.

The proposed evolutionary scenario for fur is consistent with the current view of gene duplication as a major means of microbial genome evolution 35. It has been suggested that broadly functional genes are more easily duplicated than functionally established ones, and that the modifications that follow the duplications should provide the appearance of new functional specificities. In fact, although paralogues and orthologues have the same general function, paralogues usually differ in specific biochemical details such as the primary target or a required co-factor 36. Thus, it is reasonable to conceive that an ancestral fur gene, encoding a Fur protein with a broad range of DNA-binding motifs and ionic co-factors, through duplication and divergence, gave rise to the modern fur genes, now optimised and specialised.

Consistent with the described model are the results of independent biochemical characterization of three of the S. coelicolor homologues: one of them (SCO0561 – group A) is able to bind in vitro several divalent metals (Ni²⁺, Mn²⁺, Zn²⁺ and Fe²⁺) and regulates the downstream catalase-peroxidase in a redox-dependent manner 22; a second (SCO4180 – group E) binds specifically Ni²⁺ and regulates the transcription of a FeSOD and a cluster of genes related to nickel-uptake 9; and the third (SCO5206 – group D) binds metals yet to be identified, responds to the redox changes of the cell and regulates a monofunctional catalase 18. In the same way, the biochemical characterisation of the Fur homologues in M. tuberculosis have shown that the A homologue (Rv1909c) regulates the downstream catalase-peroxidase in the presence of metals and in a redox-dependent manner 2124, while the B homologue (Rv2359) binds Zn²⁺ and is likely involved in the regulation of genes responsible for zinc-uptake 10. Thus, the main function – transcription regulation – is maintained among paralogues and orthologues. However, functional specificities such as the metal coordinated and the genes regulated are only maintained within orthologues (SCO0561 and Rv1909c), and diverge in the paralogues of the same organism (SCO0561/SCO4180/SCO5206 and Rv1909c/Rv2359). As might be expected, paralogues resulting from recent duplications are more similar than more ancient ones. In fact, SCO5206, which according to the present model is predicted to be the result of a duplication of the A homologue, is functionally closer to its correspondent A paralogue (SCO0561) than to the others. Although they regulate different genes, both are functionally related to oxidative stress response and dependent on the redox status of the cell. Paralogues from older duplication events have had more time to evolve and have accumulated more differences than those generated from recent duplication events.

In support of the proposed evolutionary model, it can be observed that several fur orthologues are located in equivalent regions of their genomes (Fig. 3). Syntenic genes reveal the core chromosomal segments present in a common ancestor, encoding a high proportion of essential gene functions and presenting a significantly lower HGT rate 37. These syntenic regions are evidence for a group with a common origin, and were considered significant whenever at least two genes remained contiguous across different chromosomes 37. The high degree of synteny observed for the fur orthologues points toward an early origin. The identification of the regulators that were lost or gained in each specific case may provide clues concerning the metabolic properties and pathways that are common to the lineage and those which specificity is more species-related.

Assuming that genome size is somehow related to the selective pressures acting upon an organism, and that Fur is a global transcriptional regulator able to sense different metals and to regulate the expression of different genes, we argue that the larger the genome size and the underlying selective pressures, the higher is the need for Fur regulators and the lower the pressure toward gene loss. Indeed, the number of Fur homologues tends to increase with increasing genome size. As seen in Table 1, organisms with genomes smaller than 2 Mbps have no Fur homologues, organisms with genomes between 2 and 5 Mbps have 1 or 2 Fur homologues, while organisms with genomes between 5 and 7 Mbps have 3 Fur homologues and organisms with genomes between 7 and 9 Mbps have 4 Fur homologues, with seven exceptions:A. cellulolyticus, M. avium 104, M. ulcerans, Nocardioides JS614, Rhodococcus RHA1, R. xylanophilus and S. avermitilis. This relation finds support in statistical analysis: the Pearson Product Moment Correlation was calculated using genome size and number of Furs as variables, and the result was 0.823, statistically different from 0 with α = 0.05.

T. whipplei spp. is an interesting case. Adaptation to a strictly host-adapted lifestyle has led to gene loss and several metabolic pathways, namely those related to amino acid biosynthesis and energy production, have been lost 38. In this scenario, the need for regulators is reduced, leading to the loss of Fur proteins. On the other hand, the ecological niche occupied by each organism may explain the seven exceptions noted above. A. cellulolyticus, Nocardioides JS614, and R. xylanophilus have one more Fur than predicted by their chromosome size. While the first organism is a thermophile and the second one inhabits the soil, the third organism exhibits high tolerance to radiation. These considerable stressful situations may have imposed selective pressures that maintained an extra fur homologue in the chromosome, despite its size. On the other hand, M. avium 104 and M. ulcerans have one less homologue than predicted by their chromosome size. This may be explained by the fact that these organisms are host associated: it is known that host-associated bacteria tend to undergo a genome reducing process. Finally, Rhodococcus RHA1 and S. avermitilis also have one less homologue than predicted by their chromosome size. There is no obvious explanation for these cases.

Frankiae, a group of plant microsymbionts able to fix atmospheric nitrogen, illustrates what is stated above. Recently, three Frankia strains have been sequenced, ACN14a, EAN1pec and CcI3, and the presence of a high number of transcriptional regulators has been noted 39. Interestingly, these genomes present highly divergent sizes: ACN14a has a genome of 7.5 Mbps, encoding 6711 proteins; EAN1pec has a genome of 9.0 Mbps, encoding 7976 proteins; and finally, CcI3 has a genome of 5.4 Mbps, encoding 4499 proteins. One possible explanation proposed to account for the differences in genome size is related to the variation in their lifestyles. While ACN14a and EAN1pec survive well in the soil, CcI3 appears to be undergoing a genome reducing process, becoming more and more dependent on the plant symbiont and less able to survive by itself 39. The increased gene contents of ACN14a and EAN1pec provide a variety of "extra" genes that allow these strains to survive in a variety of soils and in symbiosis. CcI3 is apparently losing its ability to survive in the free-living state so selective pressures are lower and fewer regulators are needed. This is clearly in agreement with the number of Fur homologues that were identified for each strain:Frankia ACN14a and EAN1pec have 4 homologues each, while CcI3 has only 3.

As mentioned above, there are nine gains of the A homologue that cannot be explained by the proposed model. Although four independent duplications appear as the most parsimonious explanation for their origin, the hypothesis of HGT should also be considered, especially since six out of the nine homologues have a %G+C content different from the genome (MSMEG_6383, Mjls_5253, Mkms_4974, Mmcs_3447, nfa3250 and Noca_0874). The presence of synteny within at least one of the groups of homologues (MSMEG_6253, Mjls_5253, Mmcs_4885, Mkms_4974) favours the hypothesis of duplication but it does not exclude HGT, since the transfer of entire chromosomal fragments (instead of single genes) is possible.

On the other hand, one of the factors that limits HGT is that the transferred genes must outcompete indigenous ones, which are already part of a complex and adapted network, in order to be fixed in the genomes 4041. One expects essential regulatory genes to be stable during evolution. In fact, it is not simple for a regulator like fur to be horizontally transferred, enter a complex network, and establish itself as a major regulator. Furthermore, recent work has suggested that HGTs seldom affects orthologues 42. Therefore, due to the number of the fur homologues in the genomes, their synteny and their nature as global regulators, explaining the presence of these genes by HGT should be used with care, and consequently %G+C value alone should not be the exclusive argument to sustain a HGT situation.

Regarding the E sequences, the situation is inverted. Besides %G+C values, other factors seem to favour HGT. One could hypothesise three independent duplication events as the origin of these genes: in the Streptomyces ancestor, in the Frankia – Acidothermus common ancestor and in Nocardioides. However, the presence of synteny across these different groups – and not only within them, as happens in the duplicated A homologues – indicates a common origin for all of the genes that constitute them, excluding the three independent duplications as well as three independent HGTs. Another explanatory hypothesis would include a single genomic duplication in the last common ancestor of the organisms involved. However, this last common ancestor is actually the last common ancestor to most of all the other Actinobacteria considered, and so the possibility of duplication would imply a high rate of gene loss. In fact, and in terms of number of evolutionary steps, the 3 in-tandem described HGT is the most parsimonious explanation for this group of genes.

In order to identify all the fur homologues in the completely sequenced actinobacterial genomes present in NCBI database (March/2007), a two-step approach was used. Initially, BlastP analyses were performed against each genome individually, using Frankia alni ACN14a FRAAL3168 as the query sequence: only hits with an e-value below or equal to e-05 were retained for further analyses. Afterwards, the retrieved sequences were aligned with Streptomyces reticuli FurS (CAA74697), in order to check for the presence of five key residues shown by D. Ortiz de Orué Lucana et al. (2003) to be essential for Fur functionality: cysteines 96 and 99, histidines 92 and 93 and tyrosine 59.

Multiple alignments were performed using ClustalX 1.81 43 with all the default parameters. The data set included the Fur homologues' amino-acid sequences, with or without Archaeoglobus fulgidus DSM 4304 Fur (NP_071057) as the outgroup, and 16S ribosomal RNA nucleotide sequences [see Additional File 6] retrieved from each genome page on NCBI. The resulting alignments were used to generate phylogenetic trees by the Neighbor-Joining (NJ) method 44 using the same software, and by the Maximum Likelihood (ML) method 45 using Tree-Puzzle 5.2 46. Bootstrap values were calculated for the NJ trees using 10000 replicates to evaluate the robustness of the nodes 47.

For the ML analysis, the evolution model used was the WAG model 48, selected by Tree-Puzzle as being the one that best described our data. The parameter estimation was exact and used quartet sampling (for substitution process) and NJ data (for rate variation). The chosen tree search procedure was Quartet Puzzling and 50000 puzzling steps were computed in order to obtain the consensus tree.

Trees were visualised by NJPLOT 49 and PhyloDraw 0.8 50.

For each fur homologue, the adjacent regions were visually inspected in order to detect the presence or absence of synteny. The inferred amino acid sequences for the 4 genes found upstream and downstream of each fur were used as query in a BlastP search against all the sequenced actinobacterial genomes, and those with e-values below e-05 were analysed to determine if any hit corresponded to a gene-encoding protein occupying a similar position relative to a fur orthologue in another organism.

The Pearson Product Moment Correlation was computed using XLSTAT 2008.2.02.