We have detected a type 1 association for ISPsy8 in P. syringae DC3000, for ISPa11 in P. aeruginosa PA01, for ISPpu9 and ISPpu10 in P. putida KT2440, and for ISRm22 and ISRm19 in S. meliloti 1021 genome. Figure ​Figure11 shows IS elements flanked by the two fragments of the broken REP sequences, and the alignments of the reconstructed REP sequences corresponding to the insertion sites are shown in Figure ​Figure2.2. In addition, the alignments of the complete sequences of each IS element, including their flanking regions, are in the additional material [see Additional file 1, file 2, file 3, file 4, file 5 and file 6]. Remarkably, in all cases of type 1 association, 100% of the copies of each IS are associated to REP (Tables ​(Tables11 and ​and2),2), proving a high selectivity for their REP sequence target. The results of this sequence analysis allow us to affirm that REP elements are target sequences for transposases.

There are five ISPsy8 elements in the P. syringae DC3000 genome and in all cases, their insertions were into a REP sequence. ISPsy8 always broke the REP element at exactly the same point of the sequence, generating a direct repeat of three bases (Figure ​(Figure1)1) [see Additional file 1]. A conserved arrangement that consists of a fragment of the REP sequence, a direct repeat of three bases, the left end of the ISPsy8, the transposase OrfA, the transposase OrfB, the right end, the other direct repeat, and the remaining fragment of REP sequence is maintained in all ISPsy8 insertion areas (Figure ​(Figure1)1) [see Additional file 1]. In four cases, the broken REP elements are in the minus strand, and in one case, the broken REP element is located in the plus strand (Figure ​(Figure1).1). However, in all cases, the transposase ORFs are in the plus DNA strand [see Additional file 1]. The point of insertion within the REP element is exactly between the bases occupying positions 32 and 33 of the REP sequence. All these REP elements share a consensus sequence (Figure ​(Figure2)2) adjacent to the ISPsy8 insertion point. A direct repeat of three base pairs, corresponding to positions 33, 34 and 35 of each broken REP sequence, is generated and appears at both extremes of the IS element (Figure ​(Figure1)1) [see Additional file 1]. Palindromy can probably induce REP sequences to adopt hairpin secondary structures. Strikingly, the ISPsy8 insertion site is located just at the symmetry axis of one of the two probable hairpin structures predicted for REP sequence of P. syringae [5] (Figure ​(Figure2)2) [see Additional file 1]. The allocation of ISPsy8 into clusters of REP elements was determinant for the detection of REP elements broken at the ISPsy8 insertion point. In four cases, ISPsy8 was inserted into a cluster of REP sequences and in one case, it was inserted into an isolated REP element [see Additional file 7]. When we joined the two fragments located at both sides of ISPsy8, the REP sequence appeared perfectly reconstructed (Figure ​(Figure2)2) [see Additional file 1]. In the cases where the broken REP sequence formed part of a cluster, its reconstructed sequence was very similar to the REP sequences that shared the same orientation in the cluster [see Additional file 8]. This was as expected considering that each cluster used to have two differentiated types of REP sequences [5]. This seems to support the idea that REP sequence fragments flanking ISs do not form part of the IS inverted repeats, but they are the fragments of the target sequences broken by IS transposition.

In P. aeruginosa PA01, there are six copies of ISPa11, and we found the same fragments of REP sequence flanking each ISPa11 copy (Figure ​(Figure1)1) [see Additional file 2]. In this case, we defined the insertion sites that lacked the usual direct repeats (Figure ​(Figure1).1). The arrangement detected for ISPsy8 was also conserved for ISPa11(Figure ​ISPa11(Figure1).1). In this case, all broken REP sequences were in the same DNA strand [see Additional file 2]. For ISPa11, the point of insertion within the REP element is between the sixth and seventh bases of the REP sequence. The shared consensus sequence obtained reconstructing and aligning the six broken REP sequences is displayed in Figure ​Figure22.

In P. putida KT2440, all copies of ISPpu9 and ISPpu10 were inserted into P. putida REP sequences (Figure ​(Figure1)1) [see Additional file 3 and file 4]. Both ISs generate direct repeats of two base pairs at the insertion point (Figure ​(Figure1).1). Our data are in agreement with insertion site data previously reported [15]. It is important to note that, although ISPpu9 and ISPpu10 sequences are not highly similar, both are inserted exactly between the ninth and tenth bases of the REP sequence. In addition, the insertion site consensus sequence is extraordinarily conserved, and the insertion site sequence is practically identical for all copies in both ISs (Figure ​(Figure22).

In S. meliloti 1021, all copies of ISRm22 were inserted into S. meliloti REP elements (Figure ​(Figure1)1) [see Additional file 5]. The analysis of ISRm22 flanking regions allowed us to characterize the ISRm22 insertion sites and to describe direct repeats of six base pairs generated at the insertion points. There are 9 copies of ISRm22 in the S. meliloti 1021 genome but, curiously, only six copies have perfectly conserved direct repeats at both extremes of the IS. The three direct repeats marked with light blue background in Figure ​Figure11 are not perfect direct repeats. From observing their sequences, it could be the result of recombination events between the copies. Homologous inter- or intramolecular recombination between two IS elements, each with a different direct repeat, would result in a hybrid element carrying one direct repeat of each parent, and is the case for these three ISRm22 instances [16]. Formation of adjacent deletions resulting from duplicative intramolecular transposition could also result in a single copy of the direct repeat located on each of the reciprocal deletion products [16]. We have reconstructed the REP sequences corresponding to the six ISRm22 copies with identical direct repeats, thus obtaining a clearly defined consensus sequence at the point of insertion. This consensus is highly palindromic and the insertion point is located just at the palindromy axis (Figure ​(Figure2).2). The other case of type 1 association detected in S. meliloti 1021 genome was ISRm19. The four instances of ISRm19 were also inserted into REP sequences (Figures ​(Figures11 and ​and2)2) [see Additional file 6]

While many IS elements display little obvious target site selectivity, some IS elements display considerable selectivity [18]. We have detected a set of elements displaying high target selectivity by REP sequences (elements with type 1 association, Table ​Table2).2). There is experimental evidence for IS1397 and ISKpn1, suggesting that transposases themselves appear to be responsible for target specificity [12]. The results of our study show that REP-targeting is not restricted to only one IS family, but it extends to five different IS families: IS3, IS110, IS4, IS256 and IS5. This is not surprising since the features of the DNA-target and of the transposase domain responsible for target choice, are not included in the criteria to define IS families [16].

There are two families that include elements with experimental evidence of target selectivity by REP sequences. One family is the IS3 family which includes IS1397 and ISKpn1 [11,12,17]. The other one is the IS110 family, to which belongs IS621 [10]. We have detected additional IS elements belonging to these two families that specifically transpose into REP elements. The element representing the IS3 family that we have detected with a strong type 1 association with P. syringae REP sequences is ISPsy8 (Tables ​(Tables11 and ​and2).2). Members of the IS3 family are similar in many aspects, and form an extremely coherent and highly related family. Usually, the transposase is encoded by two ORFs that are sometimes overlapping. The OrfB products, similar to retroviral integrases [19,20], carry a DD(35)E motif and are responsible for catalytic activity. The target recognition capability is usually located in the OrfA protein, which in various members of the family exhibits a relatively strong helix-turn-helix motif that could provide sequence-specific binding to DNA [21,22]. Many members also carry a putative leucine zipper located at the end of OrfA that could be involved in multimerization [23]. The N-terminal domain of the OrfA protein of ISPsy8 is positive for the Pfam Hidden Markov Model profile PF01527, named Transposase_8 (Table ​(Table2).2). The region identified by this profile includes a helix-turn-helix motif at the N terminus followed by a leucine zipper motif that is also present in other IS3 family elements. Probably, this HTH motif is involved in DNA target choice. There are experiments proving that IS30 needs an H-HTH motif [24], similar to the H-HTH motif involved in the DNA binding of the response regulator FixJ, in order to bind specific DNA target sequences [24]. This data suggests that some transposases could recognize palindromic REP sequences in a similar way that some transcriptional regulators recognize their palindromic binding sites.

The IS110 family is the other family with one element, the IS621, with an experimentally proved REP sequence target [10]. In our genomic analysis results, this family is the family most represented in the set of IS elements associated with REP sequences (Table ​(Table2).2). IS110 is a very special family of IS elements that has characteristics very different from the other families. The majority of their members have not inverted repeats flanking the transposase gene and little overall similarity can be detected between the ends. The mechanism of transposition of these elements is not well determined. However, the target sequences of the members IS117 from Streptomyces coelicolor and IS900 from Mycobacterium paratuberculosis exhibit similarities to the circle junction, suggesting an insertion mechanism by site specific recombination [16,25,26]. The site-specific invertase Piv from Moraxella lacunata also belongs to the IS110 family. This protein is included in the IS110 family, because it exhibits amino acid homology with the transposases of this family. The tertiary structure of amino-terminal domain of Piv invertase has been modelled [27], based on crystal structures of catalytic domains of HIV-1 integrase [28], avian sarcoma virus integrase (ASV) [29], and Tn5 transposase-related inhibitor protein [30], and the predicted structure matched with mutagenesis studies [27]. These results led Tobiasson and colleagues to propose that Piv invertase and the IS110 transposases could mediate DNA recombination by a common mechanism involving a catalytic DED or DDD motif [27]. Our study adds data that relates the IS110 family with site specific recombination processes. ISPa11, ISPpu9 and ISPpu10 exhibit a high selectivity in their target choice and could share mechanisms of target recognition and/or catalytic activity with some site-specific recombinases and viral integrases.

Using pairwise whole genome alignments, it is possible to segment bacterial genomes into a common conserved backbone and strain-specific sequences called loops [31]. These strain-specific loops include mobile elements, genes adapted to specific ecological environments, genes involved in pathogenicity, and other genes acquired by horizontal gene transfer. Strikingly, whole genome comparative analysis in Escherichia coli strains showed that strain-specific loops are associated with BIMEs (composed by different types of E. coli REP elements) [31]. In parallel, the mapping of the IS elements in different E. coli strains revealed that ISs are associated with deletion of genome fragments and incorporation of horizontally acquired genes [32]. In addition, some phenotypic features of E. coli are explained by the inactivation of genes by IS elements. This is the case for the absence of expression of the OmpC porin with the correspondingly elevated expression of the OmpF porin reported for E. coli B [32]. Thus, REP elements and IS elements are related with similar genome evolution events. Our detection of REP elements as frequent targets for transposases could explain the involvement of both in common genome plasticity phenomena. All these facts suggest that REP-recognizer transposases could be contributing to the repertoire of bacterial adaptive mechanisms.

The IS4 family had not been previously related to REP sequence target selectivity, but our genome analysis has detected that ISRm22, a member of this family, has its nine copies inserted into REP elements along the S. meliloti 1021 genome (Figure ​(Figure11 and ​and22 and Table ​Table2).2). There is data about the Tn5 transposon that helps to understand this IS4 family. The Tn5 transposon is comprised of a cluster of antibiotic resistance genes bordered by two IS50 Insertion Sequences. IS50 belongs to the IS4 family and a truncated version of the IS50 transposase that contains the catalytic active site, termedTn5 transposase-related inhibitor protein, has been crystallized [30]. The structure of its catalytic domain is probably similar to the Piv invertase member of the IS110 family of transposases (See above), connecting both families with detected REP-recognizer members. One of the characteristics frequently found for Tn5 transposition target sites is the palindromic structure of the insertion site, and also, there is a frequent occurrence of GC pairs at each end of the Direct Repeats [33,34]. The insertion sites that we have detected for ISRm22 fulfill both requirements (Figure ​(Figure2).2). Another proposed characteristic of Tn5 transposition is the preferable integration in actively transcribing or highly super-coiled DNA regions [33]. In this sense, REP sequences are frequently located in regions between convergent genes. These DNA fragments are especially prone to be highly supercoiled since simultaneous transcription of both convergent genes can generate increased positive supercoiling at the end of the genes [3]. Through testing the frequency of Tn5 insertion into specifically designed synthetic target sequences, it has been found that IS50 recognizes a preferred 9-bp sequence as its target. Moreover, sequences resembling this consensus target function optimally when embedded in a cluster of overlapping similar sequences [33]. In accordance with these Tn5 data, we have found that the majority of ISRm22 copies are inserted into a cluster of REP sequences.

In the type 1 association cases (ISPa11, ISPpu9, ISPpu10, ISRm22 and ISRm19) the conserved sequence encompasses almost the complete REP sequence (Figure ​(Figure2).2). All consensus sequences share a high percentage of GCs, a greater conservation in GCs than in ATs, a palindromic structure, and a similar length (with the exception of ISPsy8 which displays a shorter consensus). In spite of the differences in their corresponding transposase sequences, ISPpu9 and ISPpu10 show the same point of insertion within the consensus sequence. REP-recognizer ISs could share some features in their target recognition domains. The determination of the transposases belonging to this subset could provide new clues to search for a common mechanism of recognizing the DNA target.

Target selectivity differs significantly between different ISs. While some ISs display high target specificity, other elements exhibit regional preferences that could reflect more global parameters such as local DNA structure [16]. Thus, regional specificity has been related with GC or AT abundance, degree of supercoiling, DNA bending, replication related factors, and transcription related factors [16]. Transposition activity is frequently modulated by various host factors. The list of such factors includes the histone-like protein IHF, which has been experimentally proved to bind REP sequences. Another two REP-binder proteins, DNA polymerase I [35,36] and DNA gyrase [37-39] have also been implicated in transposition activity. Clusters of REP sequences could provide an appropriate context to recruit all the elements playing a role in transposition. The detected type 2 associations could reflect a favourable context for transposition provided by REP sequence clusters in combination with a minor stringency for the DNA target.

Multiple alignments of the sequences have been obtained using the program Multalign [48,49] and CLUSTAL W [50,51]. In some cases, the extremes of the ISs were not clearly defined and we have defined them by comparing and aligning the sequences of all copies. Manual correction of some alignments and alignment of fragments was necessary in cases of partial sequences and copies with internal insertions.

We have defined the boundaries of each IS copy based on data obtained from the annotations of the genomes and from ISfinder database [52], but in many cases, there were not available data and we have studied the IS ends by managing multiple alignment data and the meticulous analysis of the sequences.

To facilitate the detection of association between the genome positions corresponding to REP elements and IS genome positions, we have used C++ programming.

To analyze the domain structure of ISs we have used Pfam database [53,54].