The dataset of the 124 complete chromosomes from Streptococcus species available at the start of the present study was taken from Genbank¹. This initial database of reference proteins contains signature proteins from ICEs and the few IMEs reported for Firmicutes in the literature. It includes protein sequences from 50 tyrosine integrases, 13 serine integrases, 2 DDE transposases, 50 relaxases, 37 CPs, and 26 VirB4 proteins (a T4SS ATPase). This last protein has never been found in any mobilizable elements and is used here as the main criterion for differentiating between ICEs and IMEs at the detection step.

The overall workflow of our search strategy to detect and characterize IMEs in streptococcal chromosomes was described previously (Ambroset et al., 2016). This workflow allows (i) detecting ICEs from the presence of signature CDSs grouped on the genome, (ii) identifying ICE insertion sites and (iii) delineating ICEs. The workflow was adapted to IME detection by modifying the signature CDSs in step (i): a putative IME is detected when no VirB4 CDS is present and when an integrase CDS is found in the vicinity of a relaxase CDS. Steps (ii) and (iii) were conducted in the same way as for ICEs.

When an IME signature CDS was missing or incomplete (pseudogene), the corresponding complete CDS encoded by the closest known IME was taken and compared to the putative defective one by tBlastN in order to detect possible genome annotation errors (e.g., identification of an authentic gene as a pseudogene most frequently due to the presence of a type II intron within the gene or mis-identification of the “start” codon). Because before this work, the known IMEs did not encode a CP, those elements carrying an integrase gene, a relaxase gene, and a CP pseudogene were considered as IMEs. Moreover, our workflow detects elements containing only an integrase and a CP CDS but apparently no relaxase gene or pseudogene. In such cases, an exhaustive manual analysis was performed to search for new relaxase genes, in particular for genes encoding proteins related to RCR initiators. Newly found relaxases were added to the database of reference proteins and step (i) was reiterated on all genomes. In summary, we considered as IMEs all elements delimited by direct repeats and containing at least one CDS for one complete integrase as well as one complete relaxase, but no CDS for VirB4 or other proteins of the T4SS.

Each IME name indicates by letters and numbers the species and strain of the host bacteria. When an IME encodes a site-specific integrase, its denomination also specifies the name of the target gene. IMEs marked with an asterisk are not integrated in their primary site but in a secondary one as previously observed for ICEs (Ambroset et al., 2016).

The retrieval of the domain composition of all IME signature proteins from Uniprot annotations was done in batch using the BioMart Central Portal². De novo conserved domain search (CD-search³) and/or PSI-Blast analyses were performed when no data was available through BioMart. The correspondence with Mob families was established using the CONJscan-T4SSscan program⁴ (Guglielmini et al., 2011). This tool is no longer accessible but should soon be available on the Pasteur Galaxy server.

The signature proteins were aligned using Clustal omega with default parameters (Sievers et al., 2011). The trees of signature proteins were built with MEGA (Tamura et al., 2013) using (i) maximum likelihood (ML) based on the JTT (Jones–Taylor–Thornton) model including amino acid empirical frequencies (partial deletion of gaps and missing data at 80% cutoff, Gamma distribution in five categories, allowance for invariant sites), and (ii) BioNJ methods with the Poisson model (Gouy et al., 2010). The branch support of the groupings was estimated using bootstrap (100 replicates).

Protein clustering at 90 and 40% sequence identity was performed using BLASTclust⁵ (Alva et al., 2016) with default parameters. Circos⁶ was used to visualize the signature protein associations (Krzywinski et al., 2009). The functional annotation of IMEs was performed using Agmial as described previously (Ambroset et al., 2016).

The prevalence of IMEs was studied in a large set of species (27 Streptococcus species) and strains (n = 124). This exhaustive examination led to the identification of 144 IMEs. Their sizes ranged from 5 to 18 kb (Supplementary Table S1) except for IME_Sparas15912_rpsI which was 53 kb. The larger size of IME_Sparas15912_rpsI has likely resulted from a tandem ICE-IME insertion in the element. Within streptococcal genomes, this study showed that IME sizes (mean = 10 kb) were generally smaller than those of ICEs (mean = 41 kb). This difference is reminiscent of that observed between mobilizable and conjugative plasmids (Smillie et al., 2010).

Ten IMEs were found to be integrated in tandem accretion with other IMEs (IME1-IME2) or with ICEs (ICE-IME or ICE-IME1-IME2) (Supplementary Table S1) and many with decayed elements (data not shown). Nine tandems were found to be integrated in the 3′ end of rpsI, rpmG, or rplL. The last one was integrated in a secondary site inside IME_Sparas15912_rpsI. All IMEs in accretion in the same site had a tyrosine integrase but some decayed IMEs in accretion were found to encode serine integrases (data not shown). These integrases had from 21 to 69% sequence identity, showing that there was no particular relatedness between integrases encoded by elements in accretion.

More than half (n = 78) of the streptococcal chromosomes contained at least one IME. As seen in Supplementary Table S1, the occurrence of IMEs varied within species and strains. For instance, most of the 20 S. pyogenes chromosomes were deprived of IMEs, whereas the 3 S. anginosus chromosomes each contained from 4 to 6 IMEs. Interestingly, the genomes that contained the lowest numbers of IMEs (S. pyogenes with an average of 0.1 IME per genome) also contain few ICEs (0.1 ICE per genome). The opposite was also true; the genomes from S. anginosus were among those containing the highest numbers of IMEs (mean = 4.7) as well as ICEs (mean = 4.3).

Almost all of the integrase genes of streptococcal IMEs were located at one end of the element and were outward facing. These integrases belonged to two unrelated superfamilies: tyrosine integrases and serine integrases.

A total of 154 relaxase genes were detected within IMEs (including three pseudogenes in IMEs harboring two relaxases). Based on their domain composition, the relaxases were classified in nine distinct superfamilies (Table 1). Among the four most prevalent relaxase superfamilies, only the Rel_PF02486 superfamily was recognized by the CONJscan-T4SSscan analysis as belonging to a known type of relaxase (MobT). The three others were novel superfamilies of relaxases characterized by the following domains: PF01719 (Rel_PF01719), PHA00330 (Rel_PHA00330) and by the combination of PF001719 and PF00910 (Rel_PF001719-PF00910). The Rel_PF02407 superfamily was the fourth novel superfamily identified in this study. All these new relaxase superfamilies discovered in IMEs harbored domains (PF01719, PHA00330, or PF02407) that were previously found exclusively in RCR initiators from viruses or plasmids (Ebisu et al., 1995; Bachrach et al., 2004; Gibbs et al., 2006; Lorenzo-Diaz et al., 2014). They were assumed to correspond to novel relaxase superfamilies since all these proteins were associated with an integrase and a large majority of these non-canonical relaxases (or all for PF02407) were associated with a CP in our IME collection. Four other relaxase superfamilies described in Table 1 (Rel_PF03389, Rel_PF01076, Rel_PF13814, and Rel_PF03432) were recognized by the CONJscan-T4SSscan server as the MobQ, MobV, MobC, and MobP superfamilies, respectively.

On the basis of their relaxase content, the IMEs were grouped in nine classes (IME_Class_1 to 9) encoding a unique relaxase, one for each superfamily of relaxase (Table 2). An additional class (IME_Class_10) contained 10 IMEs that carry 2 relaxases: one belonging to the Rel_PF02486/MobT and the other to the Rel_PF01076/MobV. Each of the two relaxase superfamilies present in this class also exists as standalone relaxases in IME_Class_1 and IME_Class_6.

Within each superfamily of relaxases, the diversity was estimated by phylogenetic analyses and 40% sequence identity clustering. The most abundant superfamily, Rel_PF02486/MobT (Supplementary Figure S2), includes six families (Rel_PF02486_1 to 6), among which the family PF02486_3 was associated with a Rel_PF01076/MobV in IME_Class_10. In contrast, the 35 relaxases from the Rel_PF01719 superfamily are closely related and were therefore clustered in a unique family (Supplementary Figure S3). The same was true for the 12 relaxases of the Rel_PF03389/MobQ superfamily (data not shown). The 21 members of the Rel_PHA00330 superfamily were grouped into three families (Supplementary Figure S4), and the Rel_PF01719-PF00910 superfamily (15 members) analysis yielded four families (Supplementary Figure S5). Finally, the superfamily Rel_PF01076 was clustered in two families, with the Rel_PF01076_1 being always found in IME_Class_10. Overall, the IME relaxases showed a great diversity: they were classified in nine distinct superfamilies subdivided in 20 families (Table 1). It should be noted that the diversity of IME relaxases largely exceeds that of ICEs, since ICEs are classified in three superfamilies/eight families according to the same criteria (Ambroset et al., 2016). Only two superfamilies of relaxases corresponding to Rel_PF02486/MobT and Rel_PF03432/MobP are encoded by both IMEs and ICEs. However, our analysis showed that within these superfamilies, the relaxases of IMEs always belong to clearly distinct families from those including ICE relaxases. Moreover, whereas the number of Rel_PF02486/MobT relaxases found for ICEs and IMEs are similar, it can be stressed that Rel_PF03432/MobP relaxases constitute the most abundant superfamily in ICEs (62/105 relaxases) and the least abundant in IMEs (only 1/154 relaxases).

Prior to the present study, none of the previously known or predicted IMEs encode a CP (Bellanger et al., 2014), and the few CPs that are known to be encoded by mobilizable plasmids belong to the VirD4 superfamily (Garcillan-Barcia et al., 2009). Surprisingly, our results show that more than half of the streptococcal IMEs encode a CP (85/144 including 11 pseudogenes of CP) and that almost all these CPs do not belong to the canonical VirD4 family. Indeed, only two IMEs encode a VirD4 CP characterized by a C-terminal VirD4 domain (COG3505 in the NCBI CDD classification). All others (72 proteins excluding pseudogenes) were found to display a unique PF01580 “FtsK-SpoIIIE” catalytic domain and to be more closely related to FtsK (a DNA translocase involved in DNA segregation during cell division) than to the canonical VirD4. According to the CONJscan-T4SSscan analysis, these proteins belong to a particular superfamily of CP named TcpA, found only in Firmicutes (Guglielmini et al., 2013). Reconstruction of their phylogeny and 40% identity clustering allowed their classification in 12 distinct families designated TcpA_1 to TcpA_12 (Supplementary Figure S6). The three most abundant families, TcpA_4, TcpA_7, and TcpA_12, contained 11, 11, and 30 proteins, respectively. The nine others were found from only 1 to 6 IMEs.

The two superfamilies (VirD4 and TcpA) of CPs were found in both ICEs (Ambroset et al., 2016) and IMEs with different prevalence. Within Streptococcus genomes, VirD4 CPs are encoded by 61/105 ICEs and by only the 2 IMEs belonging to IME_Class_8. In contrast, TcpA CPs are encoded by 44/105 ICEs (belonging to the Tn916 superfamily) and 83/144 IMEs (belonging to class_1 to 4 and to class_7). Moreover, the diversity of TcpA CPs in IMEs is much larger than in ICEs (12 vs. 3 distinct families) which is reminiscent of relaxases (see above).

Our analysis of the co-occurrence of relaxase superfamilies, CPs, and integrases in IMEs reveals some mandatory associations, consistent with the classification defined in section “New Relaxase Families Related to RCR Initiator Proteins in IMEs” (Table 2). Tyrosine integrases were found in all IMEs encoding non-canonical relaxases related to RCR initiators, i.e., (i) a Rel_PF02486/MobT relaxase or (ii) relaxases from the four new superfamilies identified in this study. On the other hand, serine integrases were found in all IMEs encoding relaxase from canonical Rel_PF03389/MobQ, Rel_PF13814/MobC, Rel_PF01076/MobV (in the absence of Rel_PF02486/MobT relaxase) and Rel_PF03432/MobP superfamilies. Furthermore, IMEs encoding a single Rel_PF02486/MobT or a relaxase belonging to one of the four new superfamilies also encode either TcpA or no CP. TcpA CPs were never found in IMEs containing one of the canonical relaxases (Rel_PF03389/MobQ, Rel_PF13814/MobC, Rel_PF01076/MobV (alone or with Rel_PF02486/MobT) and Rel_PF03432/MobP). These IMEs contain either no CPs (MobQ, MobV, and MobP) or CPs from the VirD4 superfamily (MobC).

An illustration of all possible ternary Integrase–Relaxase–CP associations in our collection of 144 IMEs is shown in Figure 3. For each type of signature protein, the families identified by phylogenetic analyses and clustering at 40% identity are represented by arcs on the circle. The numbers of members in each family are reported in black. The numbers of clusters of sequences sharing more than 90% identity are reported in gray on each arc in order to estimate the sequence diversity in each family of signature protein. Apart from signature protein families with few representatives (n < 3) or with low diversity (90% identity clusters <3), this analysis of the co-occurrence of the different families of integrases, CPs, and relaxases (Figure 3) reveals no exclusive associations. Altogether, 39 different ternary associations were observed suggesting a high frequency of shuffling between signature proteins.

For a better characterization of the IMEs, a search for conserved CDSs in IMEs encoding the same superfamily of relaxases was undertaken. The various conserved CDS architectures are schematized in Figure 4. Apart from IME_Class_10, the IMEs with a tyrosine integrase encode a unique non-canonical relaxase (Rel_PF02486/MobT, Rel_PF01719, Rel_PHA00330, Rel_PF01719-PF00910, and Rel_PF02407). These IMEs display a compact structure composed of successive genes (relaxase, excisionase, and integrase), generally preceded by a TcpA gene (83/118). Interestingly, the same compact structure was found in IME_Class_8, even if the encoded proteins are not related (a VirD4 CP instead of a TcpA CP, a canonical Rel_PF13814/MobC relaxase instead of relaxase related to RCR initiators and a serine integrase instead of tyrosine integrase).

All other classes of IMEs encode other types of canonical relaxases and do not encode a CP. In IME_Class_10, the two genes encoding relaxases (a canonical Rel_PF01076/MobV and a non-canonical Rel_PF02486/MobT) are located far upstream of the tyrosine integrase gene in the opposite orientation. As well as the two relaxases and a tyrosine integrase, these IMEs encode a protein with a repA_N domain that is also found in proteins that initiates the theta replication of plasmids and of ICEs belonging to the Tn5252 superfamily from firmicutes (Weaver et al., 2009; Guerillot et al., 2013; Ambroset et al., 2016). However, repA_N proteins from IMEs are shorter (∼100 amino acids) than those found in plasmids and ICEs (∼340 amino acids), suggesting that they probably serve a different function.

Apart from IME_Class_8, all IMEs encoding a serine integrase and relaxases from canonical superfamilies (Rel_PF03389/MobQ, Rel_PF01076/MobV, and Rel_PF03432/MobP) also encode one or several proteins that could be involved in the maintenance of the excised elements. These proteins include homologs to: (i) ‘replisome organizers’ or ‘DnaC-related’ proteins involved in the initiation of theta replication of various phages from Firmicutes such as phi5218, phi4268, or phi9871 (Trotter et al., 2006; Tang et al., 2013; McDonnell et al., 2016), (ii) ‘ParB’ proteins involved in chromosome and plasmid partitioning (domain TIGR00180) (Figure 4). Such proteins are not encoded by any other class of IMEs analyzed in this study.

In summary, whereas the analysis of IME signature proteins shows a remarkable diversity, preventing their classification on the basis of their relationships, the analysis of their CDS organization shows that most of them (IME_Class_1, _2, _3, _4, _7, and _8, representing 120/144 IMEs) harbor a similar compact organization. This conserved compact organization was observed for all the IMEs encoding a single relaxase related to RCR initiators (Rel_PF02486/MobT type and four other new superfamilies discovered in this study) and for all IMEs encoding a CP (TcpA or VirD4).

Sequence comparison of IMEs suggests that most of the exchanges of modules or CP-encoding genes occur between IMEs with similar structures, especially between elements harboring the compact structure described above. Figure 5 illustrates two such situations involving IME_Class_2 and IME_Class_7 (Figure 5A), and IME_Class_3 and IME_Class1 (Figure 5B). In the first example, two members of IME_Class_2 (named here IME_A1 and IME_A2 for simplicity), exhibit a closely related mobilization module but their integration/excision modules are very different. Moreover, the integration module of IME_A2 is related to that of IME_A3 (from IME_Class_7) suggesting a probable exchange of integration modules between IMEs. In the second example, IME_B2 and IME_B3 (from IME_Class_1) have related integration and mobilization modules with the exception of their CP. However, the TcpA encoded by IME_B2 is related to the one encoded by IME_B1 (from IME_Class3). These data point to an exchange of genes encoding TcpA CPs between IMEs encoding different types of CPs and integrases.

Integrative and mobilizable elements are by far the least known elements that transfer by conjugation. Indeed, until now, very little information on their prevalence has been available. Considering their diversity, very few IMEs with different mobilization and/or recombination modules have been reported (only 15 in 2013, see Bellanger et al., 2014 for a review). Here, we identified all IMEs carried by 124 genomes from 27 species of Streptococcus and compared their abundance and diversity to those of ICEs previously identified in the same set of strains (Ambroset et al., 2016). We demonstrated that IMEs have a very high prevalence and are about 40% more abundant than ICEs. We also found that the mobilization modules of IMEs display a larger diversity than the conjugation modules of ICEs.

Only 20% of our collection of streptococcal IMEs encode canonical relaxases (i.e., belong to the Rel_PF03389/MobQ, Rel_PF01076/MobV, Rel_PF13814/MobC, or Rel_PF03432/MobP superfamilies), already found to be encoded by conjugative and mobilizable elements from G- bacteria (Garcillan-Barcia et al., 2009). Among them, none encode any CP (IME_Class_5, _6, and _9) except the two IMEs with a Rel_PF13814/MobC relaxase (IME_Class_8) that encode a canonical VirD4 CP. This is reminiscent of mobilizable plasmids of G- bacteria and firmicutes (Garcillan-Barcia et al., 2009).

Unexpectedly, the large majority (90%, n = 129/144) of streptococcal IMEs in our collection encode non-canonical relaxases (Rel_PF02486/MobT, Rel_PF01719, Rel_PHA00330, Rel_PF01719-PF00910, and Rel_PF02407) related to proteins responsible for RCR initiation involved in the maintenance of plasmids from firmicutes or viruses. MobT was previously identified in a few predicted IMEs (Bellanger et al., 2014) and in ICEs belonging to the Tn916 superfamily (Ambroset et al., 2016). The four other superfamilies of putative relaxases are unrelated to any known relaxase of mobilizable or conjugative elements. It should be emphasized that the MobT “relaxase” of the integrative and conjugative element ICEBs1 from the firmicute Bacillus subtilis is involved in not only the initiation of the conjugative transfer of ICEBs1 but also the initiation of RCR needed for the maintenance of ICEBs1 after excision (Lee et al., 2010). In the same way, another family of “RCR initiators” exhibiting a Rep_1/PF01446 domain is involved not only in the maintenance of three plasmids from firmicutes but also in their mobilization by ICEBs1 (Lee et al., 2012). Therefore, the classical distinction between RCR initiators and relaxases could lose its relevance. In IMEs encoding a CP (n = 85/144), RCR initiator-related relaxases are always associated with a non-canonical TcpA CP. The strict association of RCR initiator and TcpA is also observed in conjugative elements encoding a RCR relaxase (i.e., the ICEs belonging to the Tn916/ICEBs1/ICESt3 from firmicutes) (Guglielmini et al., 2013; Ambroset et al., 2016). In these IMEs and ICEs, the tcpA gene is located upstream from the RCR “relaxase” gene. Taken together, these findings reveal that non-canonical relaxases related to four types of RCR initiators and non-canonical CPs related to FtsK DNA translocase are involved in the mobilization of most streptococcal IMEs.

FtsK-related proteins were previously found to be encoded by various small plasmids mainly from firmicutes (see for examples, NCBI reference sequences NP_203541, NP_613077, YP_251910, WP_011669127, YP_001967631, YP_006939188) but they were never proposed to be mobilization CPs (see for example Bachrach et al., 2004; Bjorland et al., 2007; Shkoporov et al., 2008). However, according to our analysis using CONJscan-T4SSscan, all these proteins belong to the TcpA family. Interestingly, none of these plasmids encode a canonical relaxase. Rather, they carry a “RCR initiator” gene (located next to the tcpA) encoding either one of the domains found in IME relaxases or a PF01446 domain (i.e., the domain found in RCR initiator/relaxases from the small plasmids mobilized by ICEBs1). Therefore, it seems highly probable that these small plasmids are mobilizable. As previously proposed by Lee et al. (2012), it is probable that at least some of the small plasmids from firmicutes encoding a “RCR initiator” and lacking any canonical relaxase gene, CP, or T4SS protein, could be also mobilizable. Finally, it should also be noted that many plasmids devoid of relaxase could also carry an oriT related to those of conjugative element and therefore could be mobilizable in trans, as recently found for most plasmids from staphylococci (O’Brien et al., 2015; Pollet et al., 2016). Taken together, all these data point to a previous underestimation of the number of mobilizable plasmids and instead support their very high prevalence.

Besides IMEs and ICEs, our analysis of streptococcal genomes reveals many elements that (i) are flanked by DRs and (ii) encode an integrase but (iii) are devoid of CP and of canonical or RCR initiator-related relaxase (data not shown). At least some of these could correspond to IMEs. This hypothesis is supported first by the discovery of a novel type of relaxase related to tyrosine recombinases that is encoded by the pCW3 conjugative plasmid from the firmicute C. perfringens (Wisniewski et al., 2016). Second, in proteobacteria, IMEs devoid of relaxase but carrying an oriT are found to be mobilizable (Daccord et al., 2010) and one such IME (IME MTnSag1) have previously been described in S. agalactiae (Achard and Leclercq, 2007). Thus, the prevalence and diversity of IMEs within analyzed streptococcal genomes could be even greater than that described here.

The comparison of phylogenetic analyses of integrases with those of relaxases and CPs reveals many inconsistencies, probably due to multiple replacements of integration/excision or mobilization modules between IMEs. Such replacements were previously observed in streptococcal ICEs (Ambroset et al., 2016). The phylogenetic analysis of integrase families (Figure 1 and Supplementary Figure S1) clearly shows that replacement of integration/excision modules can occur not only between IMEs or between ICEs but also between ICEs and IMEs.

Within the mobilization modules encoding non-canonical relaxases, the comparison of phylogenetic analyses and/or co-occurrence of relaxases and TcpA CPs reveals inconsistencies. Thus, 10 out of 25 IMEs encoding a MobT relaxase that belong the PF02486_6 family do not carry a tcpA gene or pseudogene, whereas the 15 others have one. The distribution of these latter suggests that an ancestral IME devoid of TcpA has recently acquired a tcpA gene. On the contrary, among IMEs encoding relaxases belonging to the PF01719, PF1719-PF00910, and PHA00330 superfamilies, only 13 do not encode a TcpA and probably lost their tcpA by deletion. In some cases, almost identical IME relaxases (see for example the Rel_PF02486_6 cluster in Supplementary Figure S2) are associated with TcpA from different families suggesting that intramodular gene replacements occurred. In most cases, the data does not allow us to determine precisely what happened. However, the phylogenetic tree of the PF02486_6 relaxases suggests a replacement of a TcpA_5 CP by a TcpA_6 CP. Interestingly, although two superfamilies of relaxase (Rel_PF02486/MobT and Rel_PF03432/MobP) and the two superfamilies of CPs (VirD4 and TcpA) are shared by ICEs and IMEs from streptococci, there is no evidence of exchange of these genes between ICEs and IMEs.

Various IMEs integrated in rpsI, rpmG, and rplL were found to be integrated in tandem with other IMEs or ICEs encoding either related, distantly related, or unrelated integrases, relaxases, and/or CPs. We have also found many decayed elements or genomic islands in accretion with streptococcal IMEs (data not shown). These accretions result from the integration of an incoming IME or ICE by site-specific recombination in the attL or attR site of related or unrelated resident element that may not be followed by a deletion (Pavlovic et al., 2004; Bellanger et al., 2011). An accretion between elements targeting the same insertion gene and subsequent deletion of one of the transfer modules and one of the recombination modules is probably responsible for a large part of the replacement of modules or TcpA genes.

In fact, the non-canonical TcpA superfamily of CPs was previously reported to be encoded by conjugation modules from firmicutes, but TcpA has not yet been found in a mobilization module. In conjugative elements, TcpA proteins are associated with non-canonical relaxases: (i) the relaxase of the conjugative pCW3 plasmid from C. perfringens that is related to tyrosine recombinases (Wisniewski et al., 2016) or (ii) the MobT relaxases from the ICEs belonging to ICEBs1/Tn916/ICESt3 superfamily that are related to RCR initiators. In this work, we identified many IMEs encoding a TcpA CP: all of them encode a non-canonical relaxase. Although all IMEs encoding a non-canonical relaxase have a similar organization, their mobilization modules are highly versatile. First, closely related relaxases can be associated or not with a TcpA CP, suggesting that IME-encoded CP might not be needed for mobilization. If so, it can be hypothesized that the non-canonical relaxase might interact with the T4SS of the mobilizing conjugative element, either via the CP encoded by the conjugative element or via its cognate CP. In this hypothesis, the IME-encoded TcpA might enhance the mobilization efficiency and/or enlarge the mobilization range. Second, closely related relaxases can be associated with different distantly related TcpA CPs, indicating that the IME relaxase can interact with distantly related CPs. We hypothesize that the change of CP might have an impact on the mobilization efficiency and/or range.

The IMEs from streptococci carry diverse recombination modules and have a large array of integration specificity. Almost all serine integrases from IMEs catalyze site-specific integration within genes leading to their disruption. Interestingly, the majority of IME serine integrases (12/16) specifically target several conserved genes from the Tn5252 superfamily, a group of ICEs that is widespread in streptococci. As previously discussed for ICEs (Ambroset et al., 2016), we would expect that target specificity should be selected to have the least effect on host fitness. Here, the disruption of a conjugation gene would have little or no effect on bacterial host but would be deleterious or lethal for the host ICE. For instance, insertion in traG, that encodes a VirD4 CP, would abolish the ICE transfer and therefore the mobilization of the IME by the host ICE. The consequences of the integration/excision balance of ICE or IME encoding serine recombinases have never been studied but are documented for some prophages encoding serine recombinase. For these prophages, excision occurs not only during the activation of lytic phase but also when expression of the host target gene is needed (Kunkel et al., 1990; Rabinovich et al., 2012). By analogy, we can hypothesize that excision of IMEs integrated within specific conjugation genes of ICEs would be caused by the induction of conjugation. After IME excision, the conjugation module would be functional and could be expressed, thus allowing ICE transfer. The IME could use the CP and T4SS of this ICE to transfer and then could integrate in the ICE in the transconjugant. Furthermore, if the ICE that primarily hosts the IME does not transfer or integrate in the recipient cell, the incoming IME could integrate in another resident element (related ICE or decayed ICE as long as it carries the IME integration site). This would explain the presence of such IMEs in many decayed ICEs from streptococci. ICESp2905 from S. pyogenes, an ICE integrated in rumA, was demonstrated to transfer (Giovanetti et al., 2012) although it carries two IMEs: one integrated in snf2 and the other in maff2 (Bellanger et al., 2014). Although we cannot exclude that the disrupted genes are not required for ICE transfer, we can also hypothesize that the transfer could be divided into successive stages including excisions of the ICE and IMEs, independent transfers of the IMEs and of the ICE devoid of IMEs, insertions of the transferred ICE in rumA site and insertion of the IMEs within the transferred ICE. Such a mobilization mechanism of an IME integrated in an ICE can also be proposed for IMEs encoding tyrosine integrases that are site-specifically integrated in the putative oriT from ICEs belonging to Tn916 and ICESt3 families that were found in S. agalactiae and in S. mutans (Puymege et al., 2015).

Integrative and mobilizable elements and ICEs are integrated in the chromosome and are transmitted to the daughter cell as part of the chromosome. However, an excised IME would be lost in one of the daughter cell during cell division in the absence of replication. Several recent studies indicated that extrachromosomal replication is involved in the maintenance of various ICEs transferring as single-strand DNA (Ramsay et al., 2006; Lee et al., 2012; Guerillot et al., 2013; Carraro et al., 2015), but data suggesting replication of excised IMEs has not been reported so far. All streptococcal IMEs encoding a Rel_PF03432/MobP relaxase and most of those encoding a Rel_PF03389/MobQ carry one or two genes encoding proteins downstream from the relaxase gene that are distantly related to primosome proteins. Such proteins are known to be responsible for the initiation of the theta replication of prophages. Altogether, these data suggest that these IMEs are able to replicate with a theta mechanism after excision. Ten other IMEs (IME_Class_10 family) sharing a similar organization encode two putative relaxases. One belongs to Rel_PF02486/MobT superfamily, i.e., a family including both RCR initiators and relaxases. The other belongs to Rel_PF01076/MobV family. The structure of the region carrying these genes is reminiscent of the one of numerous mobilizable plasmids from firmicutes where the Rel_PF01076/MobV protein was found to be involved in mobilization whereas the Rel_PF02486/MobT protein is involved in RCR (Lorenzo-Diaz et al., 2014). Accordingly, the Rel_PF01076/MobV protein of an IME of the IME_Class_10 family was very recently shown to be a mobilization relaxase (Lorenzo-Diaz et al., 2016). Therefore, it is likely that the Rel_PF01076/MobV protein is the IME relaxase, whereas the Rel_PF02486/MobT protein would be involved in RCR replication of the excised IME. Alternatively, as previously found for the MobT relaxase of ICEBs1 (Lee et al., 2012), the Rel_PF02486/MobT protein could ensure both functions. Finally, all IMEs encoding a single Rel_PF01076/MobV relaxase encode a protein related to the ParB proteins that could be involved in the partition of the excised IME. Taken together, these data suggest that almost all streptococcal IMEs encoding canonical relaxases (except the two elements encoding a MobC relaxase) encode proteins involved in the maintenance of their excised forms. Such proteins were not found for any IME encoding single non-canonical relaxases such as MobT. Since the MobT protein from ICEBs1 is both the relaxase and the initiator of RCR involved in the maintenance of excised ICEBs1 (Lee et al., 2012), we cannot exclude the possibility that the five superfamilies of non-canonical relaxases might have both functions. Overall, many if not all IMEs and ICEs might exist in two states, namely a main dormant integrated state and an activated excised state which would be maintained by replication.