We identified the roseobacter-specific RepABC plasmid replication operon of Martelella on a linear 105-kb DNA fragment (scaffold_16.17, NZ_AQWH01000017), which has been established in the first phase of the type strain sequencing project One Thousand Microbial Genomes (KMG-I; Kyrpides et al., 2014). The scaffold also contains a complete type IV secretion system (T4SS; Cascales and Christie, 2003) and a characteristic post-segregational killing system (PSK) encoding a stable toxin and an unstable antitoxin (Zielenkiewicz and Cegłowski, 2001) thus indicating that it might still represent a functional plasmid. However, meaningful analyses of extrachromosomal elements and the systematic investigation of HGT via conjugation essentially depend on a complete genome sequence without any gaps and uncertain contig-affiliations. Accordingly, the genome of M. mediterranea DSM 17316^T was sequenced with our PacBio platform. Based on a 130-fold sequence coverage and subsequent Illumina correction, we obtained a finished genome of highest quality with a size of 5.7 Mb, harboring four circular replicons representing the 4.7 Mb chromosome and three large RepABC-type ECRs with sizes of 593-, 259-, and 170-kb (Figure S1; CP020330 to CP020333). Four scaffolds including the 105-kb fragment perfectly match with the 259-kb plasmid pMM259 thus providing independent evidence for the technical reliability of both sequencing approaches, i.e., the initial Illumina assembly and our newly established PacBio genome. Hereby, tandem repeats of transposase and recombinase genes flanking the Illumina scaffolds are limiting de novo genome assembly approaches based on short reads. In return the newly established complete Martelella genome exemplifies the repeat-resolving power of long reads established by PacBio sequencing (Bleidorn, 2016).

The 170-kb replicon pMM170 contains many sugar transport systems and the 593-kb ECR pMM593 holds a striking amount of TRAP- and ABC-transporters (Davidson et al., 2008; Fischer et al., 2010). Accordingly, both extrachromosomal elements might have an important function for Martelella's metabolite exchange with the environment. The composite plasmid pMM259, which harbors the rhodobacteral RepABC module of interest and a rhizobial equivalent (Figure S1), contains two operons for copper export (Cu⁺; Mame_05011-05014, Mame_05047, Mame_05054), two operons for cadmium or zinc export (Mame_04956-04962, Mame_04977-04982) and an arsenate-resistance cassette (Mame_05000-05003), indicating that it represents a resistance plasmid for the detoxification of heavy metals. A specific exposure to toxic metal-ions has not been reported for the natural habitat of M. mediterranea, the subterranean Lake Martel on Mallorca (Rivas et al., 2005), but microorganisms are very sensitive to long time exposures of even moderate concentrations of heavy metals (Gadd and Griffiths, 1978).

The genome of M. mediterranea harbors four RepABC replication modules [pMM593 (Mame_04351-04353, Mame_04880-04882), pMM259 (Mame_04944-04946), pMM169 (Mame_05119-05121)] and a solitary replicase gene [pMM259 (Mame_04960)]. The phylogenetic position of the respective RepC proteins was determined based on a set of roseobacterial and mostly rhizobial reference sequences largely corresponding to those of our former study (Petersen et al., 2009), which allowed us to differentiate between the nine different compatibility groups from Rhodobacteraceae originating from two ancient acquisitions (I: C1 to C8, II: C9). The phylogenetic tree, which was calculated based on 121 RepC replicases, showed that four Martelella sequences are located in the rhizobial part of the tree (blue color; Figure 1A, Figure S2). In contrast, the newly identified RepC protein of interest is placed amidst other rhodobacteral sequences in the distinct subtree C1 statistically supported by a 100% bootstrap proportion (BP; rose color; Figure S2). The internal branching pattern of the distinct subtree C1 is only poorly resolved due to the phylogenetically broad sequence sampling including the extremely divergent subtrees C4–C7, which resulted in only 145 comparable amino acid (aa) alignment positions (Figure S2). Accordingly, further analyses were focused on the Rhodobacteraceae-specific subtrees C1 and C2, whose sister-group relationship is solidly supported (85% BP; Figure 1, Figure S2). We systematically searched the public sequence databases and investigated 81 different complete RepABC operons of the compatibility groups 1 and 2 (Petersen et al., 2009). Comprehensive phylogenetic subanalyses of all three genes were performed in order to detect the closest relative(s) of the roseobacterial RepABC module from M. mediterranea.

Our subanalyses of proteins from RepABC modules belonging to the compatibility groups -1 and -2 resulted in a major improvement of the phylogenetic resolution with increased bootstrap support (Figures S3–S5). The RepA and RepB trees have comparable branching patterns, which mirrors their concerted evolution in a functional partitioning operon, and the best resolution was obtained in the RepB analysis (Figure S4). Both partitioning proteins of Martelella (Mame_04880-04881) are located in subtrees A2 and B2 (green sequences, 100% BP), which is a priori surprising because they show a deviating localization compared to the replicase RepC that is located in subtree C1 (Figures S2, S5). Our previous study showed a synchronous evolution of all three genes of the RepABC operon (Petersen et al., 2009), which is here exemplified by Dinoroseobacter shibae DFL12^T, Pseudooceanicola batsensis HTCC 2597^T, Roseovarius indicus B108^T and R. atlanticus R12B^T, four roseobacter type-strains that harbor RepABC operons of both compatibility groups (A1B1C1, A2B2C2; Figures S3–S5). Recombination events between a partitioning module of one compatibility group with a replicase of another compatibility group are rare, but they have previously been reported for the A2B2C1 module from Roseovarius sp. 217 (Petersen et al., 2009). The respective partitioning proteins of M. mediterranea (RepA, RepB) group together with Roseovarius sp. 217 and five other Rhodobacteraceae in a distinct branch of subtree -2 (highlighted in green, 100% BP; Figures S3, S4), in contrast to their replicases (RepC) that are all located in subtree -1, which reflects the common origin of the reshuffled A2B2C1 module.

An independent criterion for the classification of RepABC-type plasmid replication modules is the presence of specific palindromes that also allow for differentiating between the nine compatibility groups in Rhodobacteraceae (Petersen et al., 2009). The highly conserved inverted repeats with a length of 14 nucleotides are in RepABC-1 and -2 modules typically located in close proximity of the RepABC operon downstream of RepC (Figure 1). Accordingly, we investigated the sequences of 36 RepABC modules starting 500 base pairs (bp) upstream of the RepA start codon and ending 500 bp downstream of the RepC stop codon. The sampling was focused on the A2B2C1 module of M. mediterranea DSM 17316^T and our model organism D. shibae DFL12^T, which served as a reference due to the presence of two plasmids with characteristic A1B1C1 and A2B2C2 modules (see above; Petersen et al., 2013). All but three of the investigated plasmid replication systems contain two adjacent copies of the specific palindrome separated by only 11 to 42 bp (Table 1). Necessities of partitioning appeared to result in a nearly universal conservation of the palindrome motifs TTAACAG/CTGTTAA for compatibility group -1 and TTCACAG/CTGTGAA for compatibility group -2. A single reciprocal nucleotide exchange in the third and third to last palindrome position (A:C, T:G) is responsible for plasmid compatibility and thus the stable co-existence e.g., of the 86-kb RepABC-1 and the 126-kb RepABC-2 replicons in D. shibae (Wagner-Döbler et al., 2010). Martelella and five additional Rhodobacteraceae with A2B2C1 modules harbor the characteristic doublet of compatibility group -2 palindromes, which is indistinguishable from those of genuine A2B2C2 plasmids. The phylogenetic localization of their partitioning genes in subtrees A2 and B2 reflects, in combination with the presence of compatibility group -2 palindromes, a case example of co-evolution based on functional constraints (Figures S3, S4; Table 1). The palindrome represents the highly specific cis-acting DNA recognition site for the partitioning protein ParB, whose smooth interaction is the prerequisite for successful plasmid distribution during bacterial cell division (Pinto et al., 2012). The lack of conserved palindromes in Oceanicola sp. HL-35, the sixth strain with an A2B2C1 operon (Figures S3–S5), might reflect the inactivation of its RepABC module. This prediction is supported by the early-branching position of strain HL-35 in the RepC tree (Figure S5) that could represent a phylogenetic long-branch attraction artifact (LBA, Philippe et al., 2005). In contrast, our phylogenetic analyses showed neither a conspicuous position nor prolonged branches for M. mediterranea's RepA, RepB and RepC sequences (Figures S3–S5), thus indicating that its A2B2C1 plasmid module should be still functional at least in roseobacters (Rhodobacteraceae). Further phylogenies clearly documented a common origin of the repA2B2 partitioning operon together with Rhodobacter sp. CACIA14H1 and of the repC1 gene together with Paracoccus pantotrophus J40 (Figures 1B,C; Supplemental Material S1). This distribution reflects the frequent reshuffling of the replicase in A2B2C1 modules and moreover showed that the genuine rhodobacteracean donor of Martelella's RepABC cassette has not been detected yet.

The “chromid” concept of Harrison et al. (2010) introduced an evolutionary dimension into the classification of ECRs based on codon usage (CU) analyses. In brief, so-called chromids are essential ECRs with a CU comparable to that of the chromosome, which mirrors their long-lasting co-evolution, whereas true plasmids are frequently exchanged via conjugation and thus exhibit a largely deviating CU (Petersen et al., 2013). We investigated the affiliation of all replicons from M. mediterranea DSM 17316^T, M. endophytica YC6887^T, and Martelella sp. AD-3, which represent the three completely sequenced genomes of this genus, in a principal component analysis (PCA) of the relative synonymous codon usage (RSCU; Figures 2B,C). The two-dimensional PCA, which explains 90.0% of the CU variance, shows a clear affiliation of pMM593 and pMM170 with Martelella's chromosome, thus justifying their classification as chromids, whereas the composite replicon pMM259 represents a genuine plasmid. The capacity of horizontal exchange of this 259-kb plasmid is indicated by the presence of two T4S systems (see below) and furthermore supported by its RepABC replication system of rhodobacteral origin (A2B2C1 type; Figure 1).

The presence of a composite plasmid with rhizobial and rhodobacteral replication systems in M. mediterranea indicated that this ECR might contain additional horizontally acquired genes from roseobacters. However, a reliable detection of authentic HGTs would need to be based on time-consuming phylogenetic analyses, as documented for the RepABC modules (Figure 1, Figures S2–S7). We used a customized version of HGTector (Zhu et al., 2014) as a rapid discovery tool for the detection of potential HGT-derived genes on the plasmids of Martelella. The program allowed us to identify many putative HGTs (Tables S1–S3), but the two chromids of Martelella indicate that the number of authentic vertically evolving rhizobial genes may be underestimated. HGTector proposed a comparably low number of genes that are vertically transmitted (no HGT; 41% pMM593, 49% pMM170), whereas the best BLAST hits revealed a rhizobial affiliation for a larger part of these genes (55% pMM593, 64% pMM170; Tables S1–S3). Accordingly, we used the more conservative best BLAST hits for the differentiation between vertical inherited and horizontally acquired genes (Figure 2, Figure S1).

Our comparison of ECRs from M. mediterranea recovered a genuine rhizobial affiliation for the majority of chromid-located genes [291/529 (55%) pMM593, 100/155 (65%) pMM170], but only for 29% of the genes from the composite plasmid pMM259 (69/239), whose largest portion of genes [147/239 (62%)] is of rhodobacteral origin (Figure S1). A 30-kb stretch between 183- and 213-kb on pMM259 exhibits a rather scattered distribution of rhizobial and non-rhizobial genes (Figure 2A), but many of the rhizobial genes represent transposases that might have recently been acquired by intragenomic transposition events. Accordingly, the general composition of pMM259 clearly documents that the 259-kb plasmid harbors a backbone of roseobacter-associated genes and a rhizobial insertion of about 60-kb starting upstream of the blue RepABC module and ending downstream of the Icm/Dot type T4S system [Figure 2A, Table S2 (Mame_04940 to Mame_04999)]. This spatial separation of genes with a vertical and non-vertical history indicates that the present day plasmid still reflects the fusion event of a conjugated roseobacter plasmid with a size of about 200-kb with a 60-kb equivalent from the rhizobial host. Both partners have a different nucleotide composition as illustrated by deviations of the G+C content (Figure 2A). The 60-kb remnant of the rhizobial plasmid has a remarkably low G+C content of just 58% compared to 62% in the rhodobacteral part, a proportion that is, coincidentally, comparable to those of the two chromids (62%) and the chromosome (63%). The observed difference nearly reflects the natural G+C range of rhizobial genomes, thus documenting that Martelella is not the natural host of the rhizobial part of pMM259. This conclusion is supported by a codon-usage subanalysis (RSCU) of the rhizobial and rhodobacteral parts of pMM259 (Figure S8). The clustering clearly documents that (i) the CUs of rhizobial and roseobacter-specific genes largely differ and that (ii) both fusion partners can be classified as plasmids with a CU largely deviating from those of the chromosome. The G+C skew plot in Figure 2A allows one to pinpoint the origin of replication of the rhizobial 60-kb fragment, which is located within the repC gene of the blue RepABC module (Pinto et al., 2012), and it further shows the leading and lagging strand for DNA replication (Lobry, 1996; Grigoriev, 1998). In silico ligation of the 60-kb fragment even allows for predicting the former terminus of replication within the icm/dot operon for pilus formation of the rhizobial T4SS. Our analyses documented that it is still possible to detect specific molecular imprints in the genuine rhizobial plasmid, thus we conclude that the plasmid fusion was, from an evolutionary point of view, a rather recent event.

The HGT analyses showed that between 4 and 11% of the genes encoded on the three ECRs from M. mediterranea have a distinct affiliation that is neither related to rhizobia nor to Rhodobacteraceae (Tables S1–S3; Figure S1). These genes are highlighted in green within the outermost colored circle of Figure 2A and Figure S1. Gene clusters with comparable best BLASTP hits are especially interesting because they indicate that whole DNA modules and not only single genes have been horizontally transferred. One example for the 259-kb plasmid is an operon with two adjacent genes from Rhodospirallales encoding a thiol-disulfide interchange protein precursor and a lipoprotein signal peptidase II involved in protein export (Mame_05059, Mame_05058). However, the most conspicuous finding is the xenologous arsenate-resistance operon arsCHB with the adjacent transcriptional regulator arsR (Figures 2A,D; Mame_05000-05003), which is crucial for the oxidative detoxification of the highly poisonous methylarsenite (III) to methylarsenate (V) by ArsH (Mukhopadhyay et al., 2002; Yang and Rosen, 2016). This operon is, from an evolutionary perspective, of remarkable interest, because it exemplifies that the recombination of distantly related genes from Alpha- and Gammaproteobacteria resulted in the formation of a functional unit. The transcriptional regulator ArsR originate—as indicated by its pink color—from Rhodobacteraceae, whereas the genes marked in green of the resistance operon have a gammaproteobacterial origin with Halomonas zhanjiangensis DSM 21076^T (Oceanospirillales, Halomonadaceae) as closest relative. The chronology of HGT and plasmid fusion is difficult to estimate because the module is located within the transition zone between the rhizobial and the rhodobacteral part of plasmid pMM259 (Figure 2A). The horizontal transfer of the gammaproteobacterial arsCHB operon might thus either reflect a rather recent event in the genus Martelella or it already occurred within roseobacters prior to plasmid conjugation.

We investigated the origin of the two type IV secretion systems (T4SS) located on the composite 259-kb plasmid (Figure 2A) based on the assumption that one of them might have mediated the conjugational transfer of the roseobacter-specific A2B2C1-type RepABC plasmid into M. mediterranea (Figure 1, Table 1). The superoperon marked in blue of rhizobial origin represents an Icm/Dot T4SS with characteristic icm and dot genes (Mame_04961-04992; Juhas et al., 2008). This extremely divergent secretion system harbors a conjugative transfer relaxase gene traA (Mame_04988; Table S2), thus indicating that it is responsible for plasmid conjugation, whereas comparable conserved systems of Legionella pneumophila and Coxiella burnetii are utilized for bacterial pathogenesis (Segal et al., 2005). Syntenous superoperons including the mobilization genes have been identified in other rhizobia, such as Sinorhizobium sp. CCBAU 05631 or Ochrobactrum anthropi OAB. We were surprised that the “pink” T4S system of rhodobacteral origin also contains all essential genes for conjugational plasmid transfer. Its module structure, which comprises the virB secretion apparatus (Mame_04915-04935) and a cluster of mobilization genes including the crucial relaxase and the coupling protein [virD2 (Mame_04917), virD4 (Mame_04918)], is absolutely conserved regarding homologs from other Rhodobacteraceae (Petersen et al., 2013). One example is the duplicated T4SS from the 191-kb and 126-kb sister plasmids of D. shibae DFL12^T, whose conjugation across genus barriers has recently been demonstrated (Patzelt et al., 2016).

Our analyses suggest that the “pink” T4S system mediated plasmid conjugation from a still unknown roseobacter (donor) into the rhizobial recipient Martelella, thus explaining the large portion the rhodobacteracean genes on pMM259 (Figure 2A). Moreover, the structural integrity of the investigated T4SS systems indicates that the 259-kb plasmid from Martelella is still conjugative. Accordingly, horizontally transferred syntenous plasmids are waiting to be discovered in other marine bacteria. This aim seems to be like searching for a needle in the haystack, but horizontal plasmid transfer in the ocean has—concomitant with the exponential increase of whole genome sequences—very recently been reported for two roseobacters i.e., D. shibae DFL12^T and Confluentimicrobium naphthalenivorans NS6^T (Petersen and Wagner-Döbler, 2017).

The conspicuous separation of rhodobacteral and rhizobial genes on the composite plasmid is indicative of a rather recent fusion event (see above; Figure 2). Accordingly, we tried to identify close syntenous relatives of pMM259 with BLASTN searches in the non-redundant (nr) and whole-genome shotgun (wgs) nucleotide databases of the NCBI. This approach allows for the identification of conserved genetic modules and is based on the detection of silent mutations more sensitive than a standard BLASTP search. However, our analyses revealed no highly specific Rhodobacteraceae hits with more than 95% sequence identity. This outcome documents that the donor for plasmid fusion is still undetected, which is in agreement with the phylogenetic analyses of the A2B2C1 plasmid-replication module (see above). In contrast, the composite M. mediterranea plasmid pMM259 specifically matches with the 167-kb plasmid “p2” from Martelella sp. AD-3 (CP014277.1), and the four syntenic regions with a total size of 47-kb exhibit an average sequence identity between 95 and 99% (highlighted in yellow, Figure 3A). Their close affiliation is independently shown by the RSCU comparison of plasmid p2 with the rhizobial part from pMM259 (Figures 3B,C). Three of the conserved areas including region “three,” which contains the arsenate-resistance operon (Figure 2D), are matching the rhizobial part of M. mediterranea's 259-kb plasmid, but region “four” shares 97% sequence identity with the rhodobacteral part of pMM259. This distribution indicates that the 167-kb plasmid from Martelella sp. AD-3 might also originate from the composite rhizobial/rhodobacteral fusion plasmid and secondarily lost the majority of roseobacter-specfic genes including the A2B2C1-type RepABC replication module and the T4SS.

The presence of two complete RepABC-type replication systems of rhizobial and rhodobacteral origin on the composite M. mediterranea plasmid is of particular interest because it suggests that pMM259 might represent a natural replicon mediating HGT between two alphaproteobacterial orders. Accordingly, we established a transformation assay that allowed us to monitor the replication of RepABC-type plasmid modules based on antibiotic selection and chose the model organisms P. inhibens DSM 17395 and A. tumefaciens C58 DSM 5172 (synonyms Agrobacterium radiobacter, Rhizobium radiobacter; Tindall, 2014) as test strains.

Two RepABC operons from the 259-kb M. mediterranea plasmid, i.e., the roseobacter-specific RepABC module (4415 bp; A2B2C1) and the genuine rhizobial RepABC module (4986 bp), were cloned into the commercial vector pCR2.1 (see Experimental Procedures). Furthermore, we searched for a rhodobacteral positive control for our stability tests and analogously cloned the RepABC-8 operon from P. inhibens T5^T (DSM 16374^T; 3909 bp, A8B8C8). The respective module is specific for the type strain and located on an 88-kb plasmid, which is missing in other isolates, such as P. inhibens DSM 17395 (Thole et al., 2012; Dogs et al., 2014). The resulting plasmids pPI88-Roseo, pMM259-Rhizo and pMM259-Roseo that are shown in Figure 4A represent artificial shuttle vectors with a host-specific copy number. In Escherichia coli (Enterobacteriaceae, Gammaproteobacteria) they replicate based on the modified pUC origin derived from a ColE1/pMB1 vector as high copy number plasmids [500–700 copies per chromosome (HCNP); Gelfand et al., 1978; Lee et al., 2006], in contrast to the alphaproteobacterial host(s) where the respective RepABC system ensures a stable maintenance as a low copy number plasmid [1 copy per chromosome (LCNP); Pappas, 2008].

Phaeobacter inhibens DSM 17395 and A. tumefaciens C58 are both sensitive to the antibiotic kanamycin, and we accordingly used the respective resistance gene of pCR2.1 as a selection marker for our experiments (Figure 4A). Transformation of a circular pCR2.1 plasmid without an insert was used as negative control and confirmed that the E. coli cloning vector does not replicate in Alphaproteobacteria. A potential pitfall of the functionality test is a stable integration of the pCR2.1 construct into the chromosome, which would also result in kanamycin-resistant transformants. Accordingly, and as an ultimate proof of functional plasmid replication, we isolated the LCNPs from the alphaprotebacterial host, retransformed them into E. coli and showed that the EcoRI restriction patterns of isolated plasmid DNA are identical to those of the original digests (lane 1&3, lane 1&5; Figure 4C). The absence of DNA fragments in lanes 2 and 4 mirrors the low copy number of RepABC-type plasmids in Alphaproteobacteria but control PCRs showed that the respective LCNPs are present in all three (five) samples (Figure 4D).

Based on this experimental setup, we were able to document the functionality of our assay including the selected test strains. The positive control pPI88-Roseo mediated—based on its RepABC-8 module—stable plasmid replication in P. inhibens DSM 17395, but it does not replicate in A. tumefaciens C58 (Figure 4). The rhizobial RepABC module from Martelella (pMM259-Rhizo) showed a reciprocal pattern and is only replicated in Agrobacterium. This finding is in agreement with the strict phylogenetic separation of rhizobial and rhodobacteral RepABC replication systems (Figure 1), which led to the in silico prediction of functional incompatibility (Petersen et al., 2009). Furthermore, our assay did not only validate the functionality of Martelella's xenologous A2B2C1 plasmid replication system in Phaeobacter, it surprisingly also showed that pMM259-Roseo is—at least under kanamycin selection—replicated and stably maintained in A. tumefaciens (Figures 4C,D). The outcome is contradictory to the experiments with the RepABC-8 operon of P. inhibens T5^T and indicates that some RepABC-type plasmids of Rhodobacteraceae might have a broader host range than previously assumed. This prediction is supported by the presence of a rhodobacteral operon on the composite 322-kb plasmid from Rhizobium sp. NT-26 (Andres et al., 2013) but especially by former host-range tests with the RepABC-1 operon of the composite 107-kb plasmid pTAV1 from P. versutus UW1, which documented stable replication in Rhizobium etli CE3 and Rhizobium leguminosarum 1062 (Bartosik et al., 1998). In contrast, the rhizobial RepABC module from Martelella (pMM259-Rhizo; Figure 4) showed the expected host range limited to rhizobia. This outcome was independently validated by the respective operon from the Ti-plasmid (A. tumefaciens C58) that does also not replicate in P. inhibens DSM 17395 (data not shown), thus indicating that functional constraints prevent the replication of rhizobial RepABC plasmids in Rhodobacteraceae.

The presence of two replication systems on Martelella's 259-kb plasmid is surprising, because the rhizobial module should be sufficient for replication. Accordingly, we proposed that the stability of pMM259-Roseo is reduced in rhizobia and tested this hypothesis experimentally based on the four previously established transformants (Figure 4). The tests of pPI88-Roseo and pMM259-Roseo in Phaeobacter, which served as a reference, showed that about 5% of the cells, i.e., two of 40 tested colonies, lost their RepABC-type plasmid over night during exponential growth under non-selective growth conditions (Figure S9). The comparable stability of both constructs documented that the roseobacter-specific module from Martelella is not only functional in Rhodobacteraceae, it is even unaffected in its viability. The outcome of analogous tests with the two Martelella constructs pMM259-Roseo and pMM259-Rhizo in Agrobacterium was completely unexpected, because it showed that 90% of the host cells lost the genuine rhizobial RepABC construct spontaneously (36/40 colonies), whereas the xenologous rhodobacteral module was lost in just one of the 40 tested colonies (2.5%; Figure S9). We validated the presence of the respective construct for two resistant colonies to exclude any sample mix up and repeated the experiment, which resulted in comparable rates of spontaneous plasmid loss (pMM259-Roseo: 0/40; pMM259-Rhizo: 32/40). Yet, pMM259-Rhizo is still functional and maintained in Agrobacterium under selective pressure, but the high frequency of loss under non-selective growth conditions might reflect an ongoing degeneration of the RepABC-system into a “pseudogene module.” Accordingly, the most probable evolutionary scenario predicts that the selective pressure exclusively remains on the functional rhodobacteral RepABC cassette. The inactivated rhizobial module will get lost soon thus erasing the plasmid-specific molecular footprint of one fusion partner in the composite plasmid pMM259.

The rate of plasmid loss observed in the current study correlates with an exponential growth of the host cell in extremely nutrient-rich medium and is thus not representative for the natural habitat. Stable maintenance of natural plasmids is promoted by beneficial and sometimes even essential genes and furthermore ensured by toxin/antitoxin systems (Zielenkiewicz and Cegłowski, 2001), which is exemplified by three respective modules on pMM259 (Figure 2). Taken together, the replication module pMM259-Roseo has a broader host range than its equivalent pMM259-Rhizo (Figure 4) and it moreover showed an unexpected stability in both tested host strains [Phaeobacter (Rhodobacteraceae), Agrobacterium (Rhizobiaceae); Figure S9]. Accordingly, this A2B2C1-cassette represents the “functional heart” of a natural plasmid that should mediate stable genetic exchange between alphaproteobacterial orders essentially based on the presence of two different T4SSs (Figure 2).

In the current study we established the complete genome sequence of the rhizobium M. mediterranea DSM 17316^T. Its composite 259-kb replicon, which originated from a plasmid fusion, still harbors two functional RepABC modules that ensure plasmid replication in Rhizobiaceae and Rhodobacteraceae (Figures 1, 2, 4). M. mediterranea has been isolated from the subterreanean Lake Martel in the Dragon Cave on the Spanish island Mallorca (Rivas et al., 2005). This saline karst lake is located in very close proximity of the Mediterranean Sea and represents an ideal location for the intimate contact of a halotolerant rhizobium with roseobacters. Accordingly, there are no ecological boundaries preventing trans-order conjugation. The presence of two conserved T4S systems strongly indicates that pMM259 is still mobilizable (Figure 2), and it is thus likely that it mediates HGT from the globally occurring marine genus Martelella (Figure 3) into new rhizobial as well as roseobacter recipients. This plasmid is to our knowledge the first example of a natural replicon bridging the phylogenetic gap between these alphaproteobacterial orders. The presence of two replication systems on pMM259 overcomes the problem of the narrow-host-range of the rhizobial RepABC-type plasmids, and we thus propose that analogous plasmid fusions facilitate the genetic exchange even between bacterial classes. Previously, an outsourcing of the complete photosynthesis gene cluster for aerobic anoxygenic photosynthesis (AAnP) from the chromosome to a plasmid has been documented within the genus Roseobacter (Petersen et al., 2012). According to the “Think Pink” scenario (Petersen et al., 2013), plasmid conjugation could explain the presence of a homologous superoperon for AAnP in the marine gammaproteobacterium Congregibacter litoralis KT71^T (Fuchs et al., 2007). Natural shuttle vectors would hence connect distantly related bacterial lineages from the same habitat thereby providing access to the metabolic potential of the marine pan-genome.