The number of TCS-encoding genes harbored by the strains considered in this study varied between 8, in Lactobacillus helveticus DPC 4571, and 33 in Lactobacillus casei BL23 and L. casei ATCC 334 (Table 1). Taking the Bacteria domain as a whole, a correlation between genome size and the number of encoded TCS was observed 17 . The genomes of the Lactobacillaceae strains considered here have very similar genome sizes with an average of about 2 Mb, except L. casei and Lactobacillus plantarum (Table 1). Hence, this correlation cannot be observed although the strains with the largest genomes encode the highest numbers of TCS genes (Figure 1A). Additionally, no correlation was observed between the main habitat of the strains and the number of TCS genes in their genomes (Figure 1B). Several authors have observed that species with complex lifestyles, colonizing varied environments or possessing numerous alternative metabolic pathways tend to encode larger complements of signal-transducing proteins 10 21 . The lack of differences between Lactobacillaceae isolated from distinct environments likely reflects the low metabolic diversity within this group and their similar lifestyles and it also suggests that they do not have to cope with significantly different levels of environmental challenges.

No hybrid HKs were encoded by any strain included in this study. The genes encoding HKs and their corresponding RR partners were organized in operons (not shown). In a few cases, one of the partners was a pseudogene (Table 2 and additional file 1). In addition, some true orphan genes were also detected although they accounted for a very small fraction of the total (10 genes out of 173 TCS; Table 2 and additional file 1).

The TCS present in Lactobacillaceae were classified according to the schemes of Fabret et al. 15 for HKs and Galperin 17 for RRs. The classification of HKs is based on the comparison of the amino-acid sequence of the region around the phosphorylatable histidine 15 . This analysis divided the HKs present in B. subtilis into five classes (I, II, IIIA, IIIB and IV). The classification of RRs is based primarily on their domain architectures and structures of the constituent domains 17 . Most HKs and RRs could be accommodated within these classification schemes. The only exceptions corresponded to a group of HKs associated to LytR RRs, which correspond to the HPK₁₀ family of the classification of Grebe and Stock 14 , and a group of RRs homologous to the E. coli CitB not included in Galperin's classification 17 . A strong correlation in the association of families of HKs and RR was observed in Lactobacillaceae, for example, IIIA HKs are invariably associated to OmpR RRs. This correlation has been previously pointed out as a common feature of TCS 4 14 15 and led to Grebe and Stock to propose that many HKs and their cognate RRs have evolved as integral units 14 , a view in agreement with the coevolution model 4 .

A summary of the types of TCS found in each strain is shown in Table 2 and detailed lists of TCS identified in each strain are provided in the additional file 1. By far, the OmpR/IIIA family was the most prevalent in Lactobacillaceae, accounting for 71% of the TCS present in this group (Table 2). Furthermore, this is the only family present in all the strains included in this study. For these reasons, we focused our attention in this family for subsequent analyses.

Preliminary identification of clusters of orthologs of RR and HK sequences was performed by creating an orthology table of the 19 genomes used in this study using the clustering algorithm implemented in MBGD 43 and manually checking the clusters of orthologs thus obtained for each previously identified TCS gene. The clusters were named according to the following criteria: when a putative ortholog with characterized function was identified, the cluster was named after this ortholog; if no functionally characterized ortholog was found, the group was named after the locus tag of a representative sequence of the cluster. The clusters of orthologs are listed in Table 3.

A phylogenetic reconstruction was performed in order to investigate the evolutionary relationships of the clusters identified in MBGD. Lactobacillaceae sequences and selected outgroup sequences (see Methods) were aligned with Muscle and the alignments subsequently refined with Gblocks. The resulting datasets consisted in 147 sequences with 96 conserved positions for the HK alignment and 149 sequences and 158 conserved positions for the RR alignment (additional file 2).

ProtTest was used to determine the best fit model of amino acid substitution. Model LG 44 with a discrete gamma distribution to account for heterogeneity in evolutionary rates among sites, an estimation of the proportion of invariant sites and the empirical frequencies of amino acids (LG+G+I+F) was identified as the best fit model for both datasets. The phylogenetic information content of the datasets was then evaluated by using likelihood mapping. Briefly, this analysis enables to estimate the suitability for phylogenetic reconstruction of a data set from the proportion of unresolved quartets in a maximum likelihood analysis. The analysis was carried out using TreePuzzle with the WAG 45 model of substitution (the second best model selected by ProtTest) since the LG model is not implemented in this program. On the basis of ProtTest results, the datasets were analysed with a discrete gamma distribution and the empirical amino acids frequencies (WAG+G+F). The likelihood mapping showed that both datasets contained relatively low phylogenetic information, with only 68.2% and 77.7% fully resolved quartets in HKs and RRs, respectively (Fig. S1 in additional file 3).

The phylogenetic reconstructions were performed with PhyML using the LG+G+I+F model (Figure 2 and Fig. S2 in additional file 3). In accordance with the results of the likelihood mapping, very few nodes had bootstrap support values higher than 75%. Most clusters of orthologs identified in MBGD could be distinguished in the RR tree, although some of them were not supported (clusters 950, Bce, Cia and Ycl2), and in other groups some outgroup sequences did not cluster with their corresponding Lactobacillaceae counterparts (clusters 1209, Kin and Ycl1; see Figure 2 and Fig. S2 in additional file 3). Furthermore, the orphan RRs Lreu_1569 and LAF_1230 encoded by Lactobacillus reuteri and Lactobacillus fermentum, respectively, constituted a separate cluster (Figure 2 and Fig. S2 in additional file 3). However, these genes were located next to a gene cluster encoding a putative phosphate uptake system homologous to those located next to Pho TCS (Fig. S3 in additional file 3).

The HK tree was less resolved, as expected from the likelihood mapping result, and in many cases outgroup sequences did not cluster with their corresponding Lactobacillaceae counterparts. Furthermore, some clusters were not observed in the HK phylogenetic reconstruction. HKs belonging to clusters Pho and 872 constituted one cluster (although with low support in their basal nodes; Figure 2). HKs belonging to clusters Ycl1 and Ycl2 were identified by MBGD as belonging to the same cluster of orthologs and the phylogenetic analysis also suggested a relationship between these two clusters. However, the phylogenetic reconstruction and MBGD clustering indicated that Ycl1 and Ycl2 RRs constituted separate clusters of orthologs.

In order to determine whether the above mentioned incongruent cases were due to the low resolution of the trees or they indicated wrong assignments of clusters of orthologs, detailed analyses of Ycl1 and Ycl2 HKs, Pho and 872 RRs and HKs, and Eta and Kin RRs and HKs were carried out.

HK sequences belonging to groups Ycl1 and Ycl2 were aligned, resulting in a dataset of 233 sites after trimming the initial alignment with Gblocks (additional file 2). The best fit model for this dataset was LG+G+I+F. The likelihood mapping (using again WAG+G+F) showed an increase in phylogenetic signal compared to the complete HK dataset (89% resolved quartets; Fig. S4 in additional file 3). The phylogenetic analysis of Ycl1 and Ycl2 HKs showed that Ycl1 and Ycl2 formed separate clusters with strong support that included their corresponding outgroup sequences (Figure 3) with the exception of the putative Ycl1 sequences of Clostridium botulinum and Thermoanaerobacter tengcongensis. This result confirms that they constitute two different clusters of orthologs.

Pho and 872 RRs and HKs were aligned and trimmed, resulting in 193 and 239 site datasets, respectively (additional file 2). ProtTest analysis also selected LG+G+I+F as the best fit model for both datasets. Likelihood mapping analysis also showed an increase in phylogenetic signal in the HK dataset (85.5% resolved quartets; Fig. S4 in additional file 3) but the phylogenetic signal in the RR dataset was slightly lower than in the complete OmpR dataset (73.3% resolved quartets for Pho and 872 vs. 77.7% for the OmpR dataset; Fig. S4 in additional file 3). The phylogenetic reconstruction of Pho and 872 HKs (Figure 4) separated both groups, thus confirming that they constitute separate clusters of orthologs. The phylogenetic reconstruction of Pho RR also showed the separation between Pho and 872 clusters. Furthermore, the orphan genes Lreu_1569 and LAF_1230 appeared in a long branch within the other Pho sequences (Figure 4). Although the basal nodes were not supported in the maximum likelihood reconstruction, the position of these two sequences in the phylogenetic tree and the analysis of their genomic context (Fig. S3 in additional file 3) strongly suggest that they belong to the Pho cluster of orthologs.

Eta and Kin sequences were also identified as separate clusters of orthologs; however, the phylogenetic reconstructions of RR and HKs suggested that they might constitute a cluster of orthologs. In order to ascertain this point a detailed analysis of these groups was also carried out. The trimmed alignments of the corresponding HK and RR sequences consisted of 262 and 203 conserved sites, respectively (additional file 2). ProtTest selected LG+G+I+F for the HK dataset and LG+G for the RR dataset. The likelihood mapping analysis (using WAG+G+F) showed an increase in phylogenetic signal for both datasets (85% and 89.1% resolved quartets for HK and RRs, respectively; Fig. S4 in additional file 3). The ML reconstruction showed that Eta and Kin sequences were clearly separated with strong support, thus demonstrating that they constitute separate clusters of orthologs (Figure 5).

In summary, the phylogenetic reconstructions of OmpR RRs and IIIA HKs showed the clustering of the Lactobacillaceae orthologous sequences with their corresponding outgroup sequences thus indicating that the TCS systems present in Lactobacillaceae have not resulted from duplications (lineage-specific gene expansion) after the differentiation of this taxonomical group. This result suggests that these systems either were present in the last common ancestor of the group or that they were acquired by HGT during the evolution of this group.

In order to gain insight on the origin of the OmprR/IIIA TCS present in Lactobacillaceae, we compared their distribution with a concatenated reference species tree (Figure 6). The reference tree was derived from a 139204 sites dataset obtained from the Gblocks-trimmed concatenated alignments of 141 genes (see Methods). The tree was obtained by maximum likelihood using the (GTR+G+I+F) nucleotide substitution model 46 selected with jModelTest. The topology of the tree was essentially the same as that obtained by Claesson et al. 28 and the four groups identified by these authors were also identified in this phylogenetic reconstruction (Figure 6).

Clusters of orthologs with only one Lactobacillaceae sequence were not considered, as this analysis cannot provide clues about their origin. The widespread distribution of clusters Cro, Eta (only absent in Oenococcus oeni), and Yyc strongly suggests that they were present in the last common ancestor of Lactobacillaceae. Similarly, the distribution of Pho can be explained by lineage-specific gene losses in the last common ancestor of group A and in O. oeni. Alternative scenarios would require three independent HGT events in the last common ancestor of group B, the last common ancestor of group C, and the last common ancestor of Leuconostoc mesenteroides and Leuconostoc citreum or two HGT events in the last common ancestors of group C and groups B and D and a subsequent lineage-specific gene loss in O. oeni. The distribution of the Ycl1 cluster also points to its presence in the last common ancestor of Lactobacillaceae, with a subsequent lineage-specific gene loss in group D. The origin of other clusters is more controversial: the distribution of Kin sequences could be explained by five HGT events or seven lineage-specific gene losses; the distribution of Cia by three HGT events or six lineage-specific gene losses; the distribution of Bce by four HGT events or five lineage-specific gene losses, and, the distribution of Bil by one HGT or two lineage-specific gene losses. Although future analyses with more sequences may shed light on the phylogenetic history of these clusters, it is worth mentioning that if they had resulted from HGT events these must have occurred long ago, because clearly orthologous genes are shared by distantly related strains within the Lactobacillaceae.

As we have just seen, most TCS of the OmpR/IIIA family have a limited distribution in Lactobacillaceae (Table 3) making it difficult to obtain reliable information about their evolutionary history. Only two systems, Cro and Yyc are present in all the strains used in this study. In addition, Eta TCS is also present in all the strains except O. oeni. Hence, we selected these three systems to further analyze two points. Firstly, we were interested on the relative roles of coevolution and gene recruitment in the evolution of the OmpR/IIIA family in Lactobacillaceae. Secondly, we wanted to determine whether vertical inheritance could explain the phylogenetic relationships of the OmpR/IIIA TCS.

For this purpose, the nucleotide sequences of the genes encoding the RR and HK of the Cro, Eta and Yyc clusters were aligned resulting in datasets of 684 and 1011 (RR and HK, respectively) sites for Cro, 678 and 1041 for Eta, and, 693 and 1752 for Yyc. The GTR+G+I+F was identified as the best substitution model by jModelTest. Likelihood mapping showed limited phylogenetic signal, especially in the RR datasets (70.4%, 76.5% and 72.9% resolved quartets for Cro, Eta and Yyc RR datasets, respectively; 83%, 79.9% and 83.3% for the HK datasets; Fig. S5 in additional file 3). The phylogenetic reconstructions of HKs and RRs (Figure 7) showed, in accordance with the likelihood mapping results, that only a few nodes of the phylogenetic tree had support values higher than 75%. Comparisons between both trees and the reference tree were evaluated with the Shimodaira-Hasegawa test (SH; see Methods) to determine whether the likelihood of the data associated to each tree was significantly different at an alpha level of 0.05 (a value above the threshold indicating a non-significant difference).

The analysis of Cro sequences showed that the HK dataset rejected the topologies of the reference and the RR tree (p = 0.047 and p = 0.026, respectively) whereas the RR dataset did not reject any of the two other topologies (p = 0.317 and p = 0.18 for the reference tree and the HK tree, respectively). This discrepancy could be partly due to the low resolution of the trees. Therefore, a concatenated alignment of the HK and RR datasets was built in order to increase the phylogenetic signal. The likelihood mapping of the concatenated alignment (Fig. S5 in additional file 3) showed an increase in the phylogenetic signal of the dataset (86.9% resolved quartets) compared to the HK and RR cognate datasets. The phylogenetic reconstruction obtained with the concatenated dataset was similar to that obtained with the HK dataset (although the positions of Lactobacillus brevis, Lactobacillus delbrueckii subsp. bulgaricus and Pediococcus pentosaceus changed; see Figure 7). The Shimodaira-Hasegawa test of the concatenated dataset showed that this dataset did not reject the reference, HK or RR topologies (p = 0.089, p = 0.663 and p = 0.297, respectively). Considering that the concatenated alignment included the phylogenetic signal of the HK and RR datasets and that both topologies were not rejected by the SH test, we concluded that both genes share the same evolutionary history in Lactobacillaceae and, given that the reference topology was not rejected either, that vertical inheritance can explain the evolution of this TCS within this group.

The analyses of the Eta datasets showed that the HK dataset rejected the RR topology but not the reference topology (p = 0.041 and p = 0.386, respectively). On the contrary the RR dataset rejected both the reference topology and the HK (p = 0.014 and p = 0.008). A more detailed examination of the two topologies revealed that group A in the reference tree (Figure 6) was also found in the HK and RR trees for the Eta datasets, where it was recovered with 100% bootstrap support (Figure 7). However, the relationships among the other three groups changed quite dramatically. Group D still appeared in the two trees, but it was no longer a sister group to group B for the HK sequences and it clustered within them. This makes group B to be paraphyletic for HK. Furthermore, group C sequences did not group in the HK tree and appeared at the base of a B/D clade. A similar case occurred for the RR tree, in which group B was paraphyletic due to the inclusion of group C sequences. Since the RR dataset rejected both the HK and the reference topologies, it can be hypothesized that some evolutionary events, apart from vertical inheritance, occurred during the evolutionary history of this cluster. However, the possibility that these sequences do not hold enough phylogenetic signal for deriving their true relationships cannot be ruled out and in order to derive reliable conclusions more sequences will be necessary.

For Yyc sequences, the comparison of the HK dataset with the RR and the reference tree showed that whereas the topology of the RR tree was rejected (p = 0.000) the topology of the reference tree was not significantly different (p = 0.466). On the other hand, the RR dataset did not reject the HK topology (p = 0.064) nor that of the reference tree (p = 0.111). Taking into account the low resolution of the RR tree the results of these tests indicate that there are no significant differences between the topologies obtained with the two datasets and that these topologies are not significantly different to that obtained with the reference tree. We conclude therefore that both genes share the same evolutionary history and that vertical inheritance explains the phylogenetic relationships between the different sequences.

In summary, the analyses of the evolutionary history of these three TCS in this bacterial group do not provide evidence against a parallel evolution of the two genes, with no signs of gene recruitment and a vertical signal explaining their evolution. Therefore, and taking into account the results obtained from the analysis of the distribution of these systems, our results indicate that Cro and Yyc systems (and possibly also Eta) were present in the last common ancestor of Lactobacillaceae and have been conserved during the evolution of this group.

TCS-encoding genes corresponding to 19 completely sequence genomes of Lactobacillaceae/Leuconostocaceae (Table 1) were identified by using the tools provided by the Microbial Genome Database for Comparative Analysis (MBGD; http://mbgd.genome.ad.jp/) 43 . Briefly, an orthology table of all genes present in the 19 genomes was obtained using the clustering algorithm implemented in MBGD. The orthology table was queried for response regulators and histidine kinases in order to retrieve the corresponding genes. The genes were confirmed as RRs or HKs by checking the presence of typical conserved domains. Due to the low similarity at the nucleotide level observed in both datasets, amino acid sequences were used for subsequent analyses. In order to obtain additional sequences that might have been bypassed in the first search, similarity searches were performed with BLASTP 48 with the genomic BLAST service provided by the National Center for Biotechnology Information (NCBI; http://www.ncbi.nlm.nih.gov/sutils/genom_table.cgi) against the 19 genomes using a representative sequence of each cluster of orthologs previously identified. In order to obtain putative outgroup sequences for each cluster of orthologs identified, a representative sequence of each cluster was used to query the non-redundant protein sequence database at the NCBI using BLASTP. Sequences not belonging to Lactobacillaceae that scored the lowest E-values were selected and checked to belong to the same orthology group than the corresponding query sequence in MBGD. At least two sequences were used as putative outgroup sequences for each cluster of orthologs. Detailed information on the sequences used in these analyses is provided in additional file 1. Multiple alignments were obtained with Muscle 49 . Gaps and positions of doubtful homology were removed using Gblocks 50 . The final multiple alignments used for the analyses are available in additional file 3.

In order to obtain accurate phylogenies, the best fit model of amino acid substitution was selected using ProtTest 51 . The AIC, which allows for a comparison of likelihoods from non-nested models, was adopted to select the best models 52 . The phylogenetic signal contained in the different data sets was assessed by likelihood mapping 53 using Tree-Puzzle 5.2 54 . The models selected by ProtTest were implemented in PhyML 55 to obtain maximum likelihood trees for the different alignments. Bootstrap support values were obtained from 1,000 pseudorandom replicates. Congruence among topologies for TCS genes and/or the reference species tree (see below) was evaluated using Shimodaira-Hasegawa's test 56 implemented in TreePuzzle 5.2 54 and, when necessary, represented graphically using TreeMap 57 .

The 141 core proteins identified by Claesson et al. 28 were used to obtain a reference phylogenetic tree for the 19 strains considered in the analysis. The nucleotide sequences were retrieved from MBGD. The sequences were translated into amino acids, aligned with ClustalW and the corresponding nucleotide sequences realigned on the basis of the amino acid alignment using MEGA 4 58 . Gaps and positions of doubtful homology were removed using Gblocks 50 with default parameters. The resulting multiple alignments were concatenated using the tool available in the Phylemon suite 59 . The best fit model of nucleotide substitution was selected using jModelTest ver. 0.1.1 60 with the AIC criterion. The phylogenetic reconstruction by maximum likelihood was obtained with PhyML using the previously selected evolutionary model.