We recently identified E. coli-like LexA binding sites in the promoter region of intI1 integrase genes from mobile integrons and of the intIA integrase from the V. cholerae superintegron (Figure 2AB). In V. cholerae and some of these mobile integrons, the identified LexA boxes partially overlap the -10 element of the intI promoter in a classic operator organization. We have shown that expression of V. cholerae and E. coli pAT674 integrase genes is indeed controlled by the SOS response, leading to heightened rates of integrase-mediated recombination upon SOS induction 25 .

To gain insight into the general relevance of this observation, we undertook an exhaustive in silico study of integrase regulation by the LexA protein. Using a TBLASTN search (National Center for Biotechnology Information (NCBI); http://blast.ncbi.nlm.nih.gov/), we identified 1,483 homologs of intIA in the non-redundant (NR) (971), environmental (ENV) (381) and Whole Genome Shotgun (WGS) (131) subdivisions of the GenBank database. When sufficient data were available (1,103 sequences), the nucleotide sequences corresponding to the first 50 bp of the coding region plus 100 bp upstream of the translation start site (-100, +50) were systematically searched for LexA binding sites. We conducted independent searches for all the 15 LexA binding motifs described to date in the literature 27 . Putative LexA binding sites were detected in 56.6% (624) of the 1,103 sequences for which the (-100,+ 50) region was available (see Additional file 1), with 40 sequences displaying two LexA binding sites in tandem. All the identified LexA binding sites corresponded exclusively to the motif found in E. coli and most β/γ-Proteobacteria (Figure 2C). Given that we searched for 14 additional LexA binding motifs and that the sample of integrase sequences contained representatives from the respective clades in which these motifs have been reported, including one α-Proteobacteria species, this strongly suggests that the putative LexA regulation of intI genes is restricted to organisms harboring LexA proteins that are able to recognize the β/γ-Proteobacteria. The LexA binding motif of the β/γ-Proteobacteria is markedly divergent from that seen in E. coli and the β/γ-Proteobacteria, and it is known to have arisen after the split of the α- and β/γ-Proteobacteria subclasses 45 46 47 48 49 . Hence, it seems very likely that LexA regulation of integrase genes also arose after this evolutionary branching point. When we examined the core 16 bp sequence of the identified E. coli-like LexA binding sites, we identified 93 distinct sequences (see Additional file 2). These LexA binding sites presented substantial diversity while maintaining a high level of conservation, as reflected in their joint information content logo (Figure 2C). Importantly, E. coli-like LexA sites were detected in almost all Vibrionaceae superintegrons (Figure 2A), and in all but one of the mobile integron classes (Figure 2B), indicating that putative LexA regulation of intI genes is a widespread phenomenon among integrons.

We have previously shown that LexA regulates the expression of intI in V. cholerae, and our in silico search identified LexA binding sites in the promoter region of intI for all sequenced Vibrio species (see Additional file 1). To further assess the overall functionality of the in silico predicted LexA binding sites, we evaluated integrase LexA regulation in Vibrio parahaemolyticus strain ATCC 17802, which harbors a LexA binding site upstream of its intIA gene in a genomic context that is substantially different from that of V. cholerae (Figure 2A). Using quantitative reverse transcriptase (RT)-PCR, we determined the intIA expression level in both the wild-type strain and its lexA(Def) derivative (lacking a functional lexA gene). We found an expression ratio of 9.28, revealing a strong LexA regulation of the intIA gene expression (Figure 3A). Furthermore, electrophoretic mobility-shift assays (EMSA) with purified V. parahaemolyticus LexA protein showed that the observed upregulation of intIA expression was directly mediated by LexA in this organism (Figure 3A).

In several class 1 integrons, heightened expression of the cassette genes has been shown to rely on a secondary cassette promoter called P_C2 , located just upstream of the intI1 gene (see Additional file 3). P_C2 is enabled by a GGG insertion (on the top strand) that increases the distance between the -35 box sequence and a sequence resembling the -10 box consensus from 14 to 17 bp, thereby generating a functional σ70 promoter 3 4 50 51 . In all its reported instances, this GGG insertion disrupts a seemingly functional LexA binding site. Therefore, it is likely that the GGG insertion that enables P_C2 should simultaneously abolish integrase regulation by LexA. We tested this hypothesis using the E. coli multi-resistant plasmid pMUR050 52 , which provides an ideal background to test this hypothesis because it harbors two integrons with inactivated copies of the intI1 gene. The promoter regions of both intI genes are almost identical, and differ only in that one (P_intI1 ^-) contains a functional LexA binding site in its promoter, whereas the other (P_intI1 ⁺) presents the aforementioned GGG insertion, disrupting the LexA binding site and enabling the P_C2 promoter (see Additional file 3). As expected, EMSA confirmed that E. coli LexA is able to bind the P_intI1 ^- promoter, but that the GGG insertion effectively prevents LexA binding on P_intI1 ⁺ (Figure 3B). Furthermore, RT-PCR in wild-type and lexA-defective backgrounds confirmed that LexA regulation was only present in the IntI1^- integrase carrying the intact LexA binding site, with a strong deregulation (ratio of 6.55) in the lexA mutant (Figure 3B). Thus, the GGG insert not only enables the secondary cassette promoter P_C2 , but concomitantly disrupts the LexA binding site of the integrase promoter.

To check whether the GGG insert did in fact lead to the activation of P_C2 and increased cassette expression in the pMUR IntI1⁺ integrase, we compared RT-PCR expression profiles for the first cassette gene of both pMUR integrons. We found that cassette-gene expression was enhanced in the integron containing the GGG insertion, and that this increase was independent of LexA (data not shown). In silico searches for disrupted LexA binding sites found 44 instances of similar GGG inserts in integrons from a wide variety of species (see Additional file 4), all corresponding to class 1 mobile integrons harboring multiple antibiotic-resistance cassettes. Together, these results suggest that LexA regulation may eventually be lost under heavy selection to promote higher basal levels of the antibiotic-resistance transcript.

The presence of confirmed LexA regulation in V. cholerae and V. parahaemolyticus superintegrons suggested that SOS control of intI genes probably originated very early in their evolutionary history. Likewise, the complete absence of LexA binding motifs different from that of E. coli in all the intI promoters analyzed in this study indicated that LexA regulation must have been lost in integrons borne by species without LexA, or in which LexA recognizes a divergent motif 27 47 . At the same time, there is ample evidence of extensive (10% to 30%) and independent integrase inactivation across the Bacteria domain, implying that loss of integrase functionality may be an adaptive trait under particular selective pressures 53 . In this respect, the evidence of integrase inactivation in bacterial groups in which it is known that LexA does not recognize the E. coli motif 16 54 55 , such as the Xanthomonadales, suggests that loss of LexA regulation might be linked to mutational inactivation of the integrase gene.

To explore this hypothesis, we developed an automated system to assess integrase functionality based on the detection of generic (nonsense and indels) and integrase-specific missense mutations known to inactivate the protein (see Methods). This method was applied to 1,135 intIA homologs identified in this work for which sufficient coding sequence was available. Consistent with previous results, we found that a substantial fraction of integrase genes (43%, see Additional file 5) seem to be inactivated 53 . For the 755 intIA homologs with sufficient sequence to apply both analyses, the predicted inactivation status for each integrase sequence (active/inactive) was combined with the predicted presence of a LexA binding site in its promoter (-100, +50) region as computed previously. A correlation analysis was carried out to determine the existence of a link between loss of LexA regulation and integrase inactivation. The results of this analysis showed a significant correlation (Pearson r = 0.58, Spearman ρ = 0.53; P < 0.001) between both traits (Figure 4), and give credence to the idea that loss of LexA regulation is associated with integrase inactivation.

To gain insight into the evolutionary history of this correlation, we generated a phylogenetic tree of 44 representative IntI sequences, and applied ancestral-state reconstruction methods for both phenotypic characters (predicted integrase functionality and LexA regulation). The tree (Figure 5) is in overall agreement with previously published IntI phylogenies 9 53 56 . As in previous phylogenies, two major ecological groups can be outlined on the tree: marine and freshwater/soil bacteria. Chromosomal superintegrons and class 5 mobile integrons borne by marine species form a monophyletic clade that sits at the root of the tree. From this early branch, a second radiation of integrons encompassing both chromosomal integrons and all other mobile integron classes splits neatly into integrons borne by, respectively, marine and soil/freshwater bacteria. In the marine species, class 2 and 4 mobile integrons form a monophyletic cluster with Shewanella chromosomal integrons that is also in agreement with previous analyses 57 58 . In the soil/freshwater clade, class 1 and 3 mobile integrases form a distinct group, suggesting an early split from their chromosomal counterparts in the Proteobacteria 59 .

Both parsimony and maximum likelihood (ML) reconstructions of the ancestral state for LexA regulation strongly supported the notion that this feature was present in the common ancestor of bacterial integrons. LexA regulation (Figure 5, filled circles) is pervasive among Vibrio superintegrons and is also widespread within the marine integron radiation. It is also most likely (0.7 likelihood in ML reconstruction, see Additional file 6) that LexA regulation was present in the ancestor of the soil/freshwater radiation, and has been subsequently lost (Figure 5, open circles) in many of its internal clades. A notable exception to this trend are the class 1 and class 3 integrons, in which LexA regulation is still the norm. Our results thus imply that some particular trait in the environment of both chromosomal superintegrons and mobile integrons must be exerting a considerable selective pressure towards preservation of integrase LexA regulation. In the chromosomal integrons of the Vibrionaceae, the most likely source of this pressure is the stabilization of large integrons, which may include essential genes 15 . In mobile integrons, it seems likely that selection might favor integrons that remain largely inactive, but are capable of generating sharp bursts of recombination activity in times of need for evolutionary innovation.

The reconstruction of ancestral states for inferred integrase functionality is relatively congruent with the hypothesis that the loss of LexA regulation might be associated with integrase inactivation (Figure 5; see Additional file 7). Even though there is testimonial evidence of inactivation (Figure 5, dotted lines), integrases from almost all marine species in the tree were found to be active (Figure 5, solid lines). By contrast, integrase inactivation was found to be monophyletic for two soil/freshwater subgroups, hinting at consistent selective pressure towards inactivation.

To further analyze the correlation between integrase LexA regulation and inactivation, we mapped through reciprocal BLAST searches 60 the 755 IntI homologs containing sufficient available sequence to assess both traits against the panel of IntI sequences used to reconstruct the phylogenetic tree. Even though reciprocal BLAST provides only a crude estimate of phylogenetic relationship, this mapping process allowed us to observe the apparent frequencies (Figure 5, pie charts) of both traits in the clusters represented by each tree taxon (Figure 5). Overall, the results of this analysis broadly agree with those of the ancestral-state reconstruction, and give further credence to the idea that loss of LexA regulation is associated with integrase inactivation. Nonetheless, close inspection of these results also reveals a complex pattern of phylogenetic distribution for both traits.

Among marine species, LexA regulation of intI genes is clearly prevalent, and loss of LexA regulation is only present in a few instances. One such instance is the V. cholerae SXT integrative-conjugative element (ICE), which harbors a class 4 integrase and for which SOS-dependent transfer has been reported through an indirect path involving the phage-like SetR repressor 37 38 61 . In spite of this, mapping results show that five out of the twenty sequences clustering with the V. cholerae SXT integrase have predicted LexA binding sites. These sequences belong to mobile integrases in Alteromonadales species that do not have homology with the SetR transcriptional regulator, suggesting that LexA regulation may have been preserved in the absence of SetR-mediated SOS regulation (see Additional file 8). Another exception is Lutiella nitroferrum, but the absence of predicted LexA sites is not surprising in a member of the Neisseriaceae, because all the sequenced members of this family lack a lexA gene 27 . A similar reasoning applies to another exception, Rhodopirellula baltica, because it is known that the LexA of Planctomycetes does not recognize the conventional E. coli LexA binding motif 27 .

Conversely, loss of LexA regulation seems to be the norm among soil and freshwater species harboring chromosomal integrons. In most cases, this loss of regulation has an obvious explanation. Some families, such as the Nitrosomonadaceae and the Chromatiaceae, simply do not possess any LexA homologs, Thus explaining the absence of any LexA binding sites upstream of their intI genes 27 . A similar argument can be made for the Xanthomonadaceae, in which neither of the two identified LexA proteins recognizes the β/γ-Proteobacteria LexA binding motif 54 , and for the Spirochetes, the δ-Proteobacteria and the Cyanobacteria, in which LexA also recognizes divergent motifs 48 62 63 . However, reciprocal BLAST mapping indicates that there is residual LexA regulation persisting within several of these groups. The M. flagellatus cluster, for instance, has six out of thirty-two mapped sequences with predicted LexA binding sites. Careful examination reveals that, in this and all other cases of residual LexA regulation of soil/freshwater bacterial integrons, regulated integrases turn out to be harbored by a β/γ-Proteobacteria species or originate from environmental samples (see Additional file 9). This strongly suggests that, for the most part, LexA regulation is positively maintained when a suitable genomic background (a compatible lexA gene encoding a repressor that recognizes the β/γ-Proteobacteria motif) is available.

Several factors explain partly the absence of LexA binding motifs, other than the β/γ-Proteobacteria motif, regulating integron integrases. An obvious explanation is the lack of evolutionary time to develop such motifs. This is manifestly true for many mobile integrons subject to lateral gene transfer. Indeed, predicted β/γ-Proteobacteria LexA binding sites can still be seen in the mobile integrons harbored by species from distant phyla, such as the Actinobacteria. Integrase inactivation is another mechanism that several groups, such as those of the Xanthomonadales, seem to have evolved to compensate for unregulated integrase expression 16 . Even though this constitutes a general trend (Figure 4) and functionality can be temporarily restored through non-native recombination, the observed correlation is moderate (Pearson r = 0.58***). Moreover, integrase functionality has been assayed experimentally in several soil/freshwater chromosomal integrons in which the integrase is clearly not regulated 64 65 , suggesting that additional mechanisms must be at play.

Class 1 and 3 mobile integron integrases depart sharply from the trend towards loss of LexA regulation that is seen among soil/freshwater integrons. Reciprocal BLAST mapping supports the results of ancestral-state reconstruction methods, providing ample support for the persistence of LexA regulation in these well-sampled mobile integron classes. In addition, the high percentages of LexA regulation seen in both these integron classes (81% and 64%, respectively) are consistent with high percentages of predicted regulation in the marine mobile integrons of classes 2 and 5 (84% and 89%, respectively; Figure 5). Beyond its fundamental relevance to bacterial adaptation, the high prevalence of predicted LexA regulation of mobile integron integrases has serious clinical implications, as it establishes a generic system for genetic interchange under control of a general stress response shared by a large group of human and animal pathogens. Furthermore, bacterial conjugation has been shown recently to induce the SOS response, triggering integrase-mediated cassette recombination, in recipient bacteria 66 . In this setting, it is important to note that integron cassettes encoding resistance to several antibiotics known to induce the SOS response, such as trimethoprim, quinolones and β-lactams, are common today 5 67 . This suggests that the indirect triggering effect of these antibiotics on the capture of resistance cassettes may have resulted in a very efficient selection mechanism.

A less obvious consequence of integrase SOS regulation in clinically relevant mobile integrons is its repercussion on antibiotic-resistance policies. Current policies rely largely on the detrimental effects that most resistance mechanisms inflict on bacteria, which eventually lead to loss of resistance genes in the absence of antibiotic exposure 68 . Because most cassettes are promoterless, the most ancient cassettes (located at the distal part of the integron) are subject to severe polar effects, leading to rare or non-existent protein products (Figure 1) 4 . In this context, the incorporation of SOS regulation in integrons puts forward a mechanism by which antibiotic-resistance genes and other useful adaptations can be silently set aside, while current adaptive traits are maintained. In time of stress, such as exposure to antibiotics, the relevant resistance cassettes can be called upon by integrase-mediated translocation, and thus selected for only when their expression is required. Furthermore, the cassette genes that have been temporarily relegated to distal positions in integrons may also sustain increased evolution rates, generating a substantial pool of variability from which to draw on when the appropriate selective pressures resurface 69 .

Reciprocal BLAST mapping also shows that predicted integrase inactivation is very common among soil/freshwater bacteria, coinciding with a prevailing loss of putative LexA regulation. Nonetheless, predicted integrase inactivation is also relatively common in marine species. Even though the predicted integrase inactivation correlates well with reduced LexA regulation (Figure 4), there are notable outliers to this trend in both radiations. For instance, among mobile class 5 integrons, only 44% of mapped integrases seem to be functional, despite predicted LexA regulation in 89% of them. The opposite is also true; many mobile integrons with putatively functional integrases have disrupted, absent or non-native LexA binding sites. This suggests that lack of LexA regulation can be tolerated or selected for when it provides adaptive benefit. We have shown here that in some mobile class 1 integrons, the LexA binding site has been disrupted by a GGG insertion that drastically increases the expression of antibiotic-resistance cassettes (Figure 3). In a similar vein, it seems likely that sustained integrase activity (with its associated shuffling of gene cassettes 21 25 ) must be preferable to permanent inactivation under the shifting selective environments associated with clinical environments and mobile integrons. This would explain why integrase inactivation is not seen as frequently in mobile integron classes associated with clinical settings, in spite of their dissemination into bacterial species that do not harbor a lexA gene capable of regulating the preset LexA binding site.

Overall, however, the pattern of integrase inactivation is broadly in agreement with that reported previously 53 . In fact, we found a higher proportion (46%) of inactivated IntI proteins than that reported previously 53 , indicating that integrase inactivation is a pervasive phenomenon and typically correlated with loss of LexA regulation. Hence, our findings suggest that putative integrase inactivation is the main mechanism evolved to deal with lack of LexA regulation, but it seems likely that other factors must provide heightened tolerance to unregulated integrase activity in soil/freshwater bacteria. Smaller integron sizes and lessened integrase activity may both contribute to make unregulated integrase expression more tolerable, but regulation by an alternative transcription factor is an obvious possibility that needs to be carefully explored. This is particularly true because most integrase functionality assays have been carried out in a non-native context 64 65 and may thus have missed regulatory effects. The quest to define precisely the multiple mechanisms behind this adaption is an important goal, because the lack of a mechanism to mitigate the effects of integrase activity upon loss of LexA regulation may well lie at the root of the intriguing absence of chromosomal integrons from many bacterial phyla 53 .

A custom set of scripts was developed in BioPhyton to search for intI homologs on NCBI GenBank databases (NR, ENV and WGS). The scripts retrieved and re-annotated both the intI coding sequences and their corresponding upstream sequences. The scripts used the whole VchIntIA protein sequence (AAC38424) and its IntI specific domain 70 (positions 186 to 245 in VchIntIA) as a query for a TBLASTN search. To limit the number of false positives, a cut-off e-value of 10^-5 was set, and only sequences matching both queries were retrieved.

TBLASTN results were used to identify frameshift and deletion events of up to 100 bp. Larger events where not considered. The nucleotide sequences spanning the full length of the processed hits and 1 kb upstream of the hit start were recovered. Conceptual translations of these sequences (corrected for frameshift when necessary) were then used to search a curated reference panel using BLASTP. The reference panel comprised 43 phylogenetically diverse IntI proteins, phage integrases and XerCD recombinases. The reference sequence of the best reciprocal hit was used to consistently re-annotate the start and stop points of all retrieved sequences, thereby allowing homogenization of the dataset and efficient detection of in-frame premature stop codons. Sequences with a best reciprocal hit not belonging to the IntI family (that is, phage integrases and XerCD recombinases) were removed from further analysis. Similarly, all IntI homologs lacking a significant amount of coding sequence at both ends of the predicted coding region (+30 bp downstream of the start codon and -90 bp upstream of stop codon) were also removed from further analysis. Duplicates resulting from the use of partially redundant databases were removed, defining duplicates as two sequences having the same sequences, coordinates and NCBI taxonomical assignment, and the same strain or plasmid number when applicable. The final annotated dataset comprised 1,483 sequences, and is available online as supplementary material in both GenBank (.GBK) and spreadsheet-compatible (.XLS) format (see Additional file 10, see Additional file 11).

Integrase functionality was assessed systematically using a custom rule-based system operating on aligned IntI sequences. To generate functional rules to detect inactivation, we analyzed published structural and mutational studies of both the chromosomal V. cholerae IntI4 and the mobile IntI1 integron integrases 22 70 71 72 73 74 . From this analysis, we identified a list of five essential residues in the catalytic site that cannot be mutated (R135, K160, H267, R270, H293, Y302; positions relative to the V. cholerae IntIA sequence), and eight residues essential for binding, for which only a limited range of substitutions is likely to be tolerated (L202 (→LIVM), P203 (→PST), K209, Y210 (→YFWH), P211 (→PRQ), R239 (→KRH), H240 (→KRH), H241 (→KRH); positions again relative to the V. cholerae IntIA sequence).

A multiple alignment of all IntI sequences in the reference panel was generated using MUSCLE software http://www.drive5.com/muscle/ with an opening gap penalty of -20, and otherwise standard parameters 75 . This alignment was used to propagate the functional rules defined on the VchIntIA sequence towards the reference panel IntI sequences. The consistency of this propagation was reviewed manually. Pairwise alignments of all the TBLASTN identified homologs with their corresponding best hits were used to further propagate the functional model and allow a decision on whether each particular protein should be considered functional. IntI sequences containing an internal stop, a frameshift and/or any number of inactivating mutations were tagged as 'non-functional'. If either the start or stop of sequence was unavailable (see above), the functionality of the corresponding protein was tagged as 'unknown'. Otherwise, the protein was considered functional by default.

The automated rule-based system was evaluated against a reference set of integron integrase sequences for which activity has been experimentally assessed 64 65 76 77 78 79 80 . This reference set encompasses active and inactive integrases from both marine and soil/freshwater chromosomal integrons, and class 1, 2 and 3 mobile integrons. The rule-based system was able to correctly predict integrase activity in all these cases. In addition, it also detected all indels, frameshift and nonsense mutations that have been reported previously in independent studies as leading to integrase inactivation 16 53 .

The presence of LexA binding sites on all the retrieved intI homolog sequences was assessed by scanning them using xFITOM http://compbio.umbc.edu/2280/, a generic program for binding site search in genomic sequences 81 82 . Searches were conducted using the R_i index 83 and a motif-normalized threshold as reported previously 84 . Identified sites were considered 'w/functional box' if located within -100 or +50 bp of the re-annotated intI start codon. When the sequence in the specified range was not fully available, this feature was tagged as 'unknown'. Searches were conducted using the 15 different LexA binding motifs reported to date 27 , which include those of largely sampled phylogenetic groups, such as the Firmicutes, the Actinobacteria, the Cyanobacteira or the Alpha Proteobacteria 62 85 86 87 . We also identified, and specifically searched for, a particular motif consisting of a LexA binding site inactivated by the insertion of a GGG triplet. These sites are referred to as 'broken', and were categorized as 'without functional box'. The results of integrase functionality and LexA binding site searches are fully annotated on the main dataset files (see Additional file 10, see Additional file 11).

Alignments of the reference-panel protein sequences were carried out using a combined procedure to improve alignment quality as described previously 88 . Protein sequences were first aligned with CLUSTALW (version1.83; http://www.ebi.ac.uk/Tools/msa/clustalw2/ 89 using Gonnet matrices and default 10 , twenty-five and five gap-opening penalties for the multiple alignment stage, thus generating three different alignments. These three different alignments, together with a local alignment generated by the T-COFFEE Lalign method, were integrated as libraries into T-COFFEE (version 1.37; http://www.ebi.ac.uk/Tools/msa/tcoffee/ 90 for optimization. The optimized alignment was then processed with Gblocks (version 0.91b; http://molevol.cmima.csic.es/castresana/Gblocks.html 91 with the half-gaps setting and otherwise default parameters to select conserved positions and discard poorly aligned and phylogenetically unreliable information. Phylogenetic analyses were then carried out using MrBayes (version 3.1.1; http://mrbayes.csit.fsu.edu/ and PHYML version 2.4.1; http://code.google.com/p/phyml/ 92 for Bayesian inference of tree topologies as reported previously 88 . A mixed four-category γ distributed rate plus proportion of invariable sites model [invgamma] was applied and its parameters were estimated independently by the program. Eight independent MrBayes Metropolis-Coupled Markov Chain Monte Carlo runs were carried out with four independent chains for 10⁶ generations. The resulting phylogenetic trees were plotted with TreeView (version 1.6.6; http://taxonomy.zoology.gla.ac.uk/rod/treeview.html 93 and edited for presentation using CorelDraw Graphic Suite (version 12; Corel Corp., Fremont, CA, USA).

Ancestral-state reconstruction was conducted with the Mesquite ancestral-state reconstruction package (Mesquite Software Inc., Austin, TX, USA) 94 using the majority-rule consensus tree generated by MrBayes. The results of in silico searches for LexA binding sites were mapped into a discrete (1/0/?) character for each taxon of the tree. Reconstruction of LexA binding site presence was first carried out using the ML reconstruction method 95 96 and the AsymmMk model (Asymmetrical Markov k-state two-parameter model), estimating asymmetric rates of change between characters. The estimated rates (0.145 forward, 0.813 backward) were then converted into parsimony steps by direct inversion (6.89, 1.23), and used to generate the step matrix for parsimony reconstruction 97 . The results from in silico integrase functionality assessment were also mapped into a discrete (1/0/?) character for each taxon. Ancestral-state reconstruction for this character was carried out using both an ordered parsimony model and AsymmMk-based maximum-likelihood model. The results of both reconstruction methods were broadly in agreement (see Additional file 7), but for clarity, only parsimony results are superimposed on Figure 5.

The V. parahaemolyticus and E. coli lexA genes were amplified using suitable primers (see Additional file 12) and cloned into a pET15b vector (see Additional file 13). Overexpression and purification of the corresponding LexA protein was performed as described previously for other LexA proteins 84 . Each DNA probe was constructed using two complementary 100 bp synthetic oligonucleotides (see Additional file 12). EMSA experiments were performed as described previously 84 , using 80 nmol/l V. parahaemolyticus LexA or 200 nmol/l of E.coli LexA protein and 20 ng of each DIG-marked DNA probe in the binding mixture. For EMSA competitive assays, 200 fold of either specific or non-specific non-labeled DNA was added to the binding mixture. In all cases, samples were loaded onto 6% non-denaturing Tris-glycine polyacrylamide gels. Digoxigenin-labeled DNA-protein complexes were detected using the manufacturer's protocol (Roche Applied Science, Indianapolis, IN, USA).

RT-PCR experiments were performed (Titan One Tube RT-PCR System; Roche) with suitable oligonucleotides (see Additional file 12 for list), following the manufacturer's instructions. Real-time quantitative RT-PCR analysis of total RNA was carried out in a PCR system, (LightCycler; Roche), using a commercial kit (LCRNA Master SYBR Green I Kitl Roche) according to the manufacturer's instructions. Transcription of pMUR050 intID1 and intID2 genes (under control of P_intI1 ^- and P_intI1 ⁺, respectively) was determined in wild-type E. coli K12 and in a lexA-defective strain (UA6189). Both strains contained either the pUA1105 (intA1) or the pUA1106 (intA2) plasmid. Expression of the V. parahaemolyticus intI gene was tested in the ATCC17802 wild-type strain and in a lexA-defective strain (UA10001) (see Additional file 13). In both cases, expression of the recA gene was used as the positive control, and the mRNA concentration for each gene was normalized to that of the housekeeping dxs gene. The expression factor was calculated as the ratio of the relative mRNA concentration for each gene in the corresponding lexA mutant strain with respect to that in the wild type. In each case, the mean value from three independent experiments (each in triplicate) was calculated. Strains UA6189 and UA10001 were constructed, respectively, using the Lambda-Red recombinase system 98 or the marker exchange procedure, as described previously 99 .