Replication of the Salmonella Genomic Island 1 (SGI1) triggered by helper IncC conjugative plasmids promotes incompatibility and plasmid loss

The mobilizable resistance island Salmonella genomic island 1 (SGI1) is specifically mobilized by IncA and IncC conjugative plasmids. SGI1, its variants and IncC plasmids propagate multidrug resistance in pathogenic enterobacteria such as Salmonella enterica serovars and Proteus mirabilis. SGI1 modifies and uses the conjugation apparatus encoded by the helper IncC plasmid, thus enhancing its own propagation. Remarkably, although SGI1 needs a coresident IncC plasmid to excise from the chromosome and transfer to a new host, these elements have been reported to be incompatible. Here, the stability of SGI1 and its helper IncC plasmid, each expressing a different fluorescent reporter protein, was monitored using fluorescence-activated cell sorting (FACS). Without selective pressure, 95% of the cells segregated into two subpopulations containing either SGI1 or the helper plasmid. Furthermore, FACS analysis revealed a high level of SGI1-specific fluorescence in IncC+ cells, suggesting that SGI1 undergoes active replication in the presence of the helper plasmid. SGI1 replication was confirmed by quantitative PCR assays, and extraction and restriction of its plasmid form. Deletion of genes involved in SGI1 excision from the chromosome allowed a stable coexistence of SGI1 with its helper plasmid without selective pressure. In addition, deletion of S003 (rep) or of a downstream putative iteron-based origin of replication, while allowing SGI1 excision, abolished its replication, alleviated the incompatibility with the helper plasmid and enabled its cotransfer to a new host. Like SGI1 excision functions, rep expression was found to be controlled by AcaCD, the master activator of IncC plasmid transfer. Transient SGI1 replication seems to be a key feature of the life cycle of this family of genomic islands. Sequence database analysis revealed that SGI1 variants encode either a replication initiator protein with a RepA_C domain, or an alternative replication protein with N-terminal replicase and primase C terminal 1 domains.


Introduction
In the vast world of mobile genetic elements, mobilizable genomic islands are increasingly recognized as key players in the global spread of multidrug resistance. To date, more than 30 islands mobilized in trans by helper conjugative plasmids have been described, most carrying drug or heavy metal resistance genes [1]. One of the most atypical mobilizable genomic islands currently known is the 43-kb Salmonella Genomic Island 1 (SGI1) that was first described in 2000 in the multidrug-resistant epidemic strain of Salmonella enterica serovar Typhimurium DT104 [2,3]. SGI1 is integrated in the 3' end of trmE, a gene coding for a GTPase involved in the modification of U34 in tRNA, and carries at its 3' end a complex class 1 integron named In104, which confers resistance to β-lactams, chloramphenicol, tetracycline, streptomycin, and sulfamethoxazole ( Fig 1A) [3]. The discovery of dozens of SGI1 variants with a large variety of resistance profiles revealed the high plasticity of the class 1 integron [4]. SGI1 and its variants have been described in a wide range of S. enterica serovars, in Proteus mirabilis, Morganella morganii, Acinetobacter baumanii, and more recently in avian pathogenic Escherichia coli [5][6][7][8][9]. The identification of SGI1 and variants in a broad range of species is suggestive of the high diffusion capacity of this mobile element.
Although SGI1 is not self-transmissible, it is specifically mobilized by conjugative plasmids of the closely related IncA and IncC incompatibility groups [10,11]. IncC plasmids are widespread in several pathogenic species of Enterobacteriaceae isolated from food products, foodproducing animals and humans. IncC plasmids have been sporadically found in African seventh-pandemic isolates of Vibrio cholerae O1 El Tor, the causative agent of the diarrheal disease cholera [12,13]. Nowadays IncC plasmids are globally distributed and considered to be important contributors to the diffusion of drug resistance genes, including New Delhi metalloβ-lactamase genes (bla NDM ) that confer resistance against most β-lactams including carbapenems [14,15]. Albeit not as epidemiologically successful, a highly conjugative IncA plasmid was recently found to participate in the emergence of carbapenemase-producing Enterobacteriaceae (bla VIM-1 ) that colonized the gut of patients in a university hospital in Italy [16].
Several genes carried by SGI1 have been shown to be essential for its transfer. For instance, int and xis code for the integrase and recombination directionality factor that catalyze the integration of SGI1 into the 3' end of trmE and its excision from the chromosome (Fig 1A) [10]. Furthermore, the recently identified mpsA and mpsB code for key mobilization factors that recognize the upstream cis-acting origin of transfer loci (oriT) [17]. Nevertheless, like all mobilizable genomic islands, SGI1 lacks a complete set of genes coding for a fully functional type IV secretion system (T4SS) to be self-transmissible and relies specifically on the T4SS encoded by IncA or IncC helper plasmids to transfer to a new host [11]. However, unlike any other known mobilizable genomic island, SGI1 alters the mating apparatus encoded by IncC plasmids by replacing the T4SS proteins TraN C , TraG C , and TraH C with its own distant orthologues TraN S , TraG S , and TraH S, respectively [18]. This remodeling not only seems to enhance SGI1 transfer at the expense of the helper plasmids but also disables entry exclusion mediated by IncA and IncC plasmids, thereby enabling the transmission of SGI1 to bacterial cells that bear such plasmids [18,19]. Expression of xis, traN S and traHG S of SGI1 has been shown to be activated by the master activator complex AcaCD encoded by IncA and IncC plasmids [18,20,21]. Therefore, excision of SGI1 from the chromosome occurs only in the presence of a helper plasmid. Together with the presence of the sgiAT toxin-antitoxin system, tight regulation of excision contributes to the high stability of SGI1 in the absence of an IncC plasmid [22,23]. SGI1 and IncC plasmids are never found together in natural isolates [24,25]. In fact, SGI1 was found to destabilize IncA and IncC plasmids after a few generations [23,26]. Conversely, the presence of an IncC plasmid has been shown to enhance the recombination rate within SGI1, leading to the generation of SGI1 deletion variants [22]. Conjugation-dependent rollingcircle replication of SGI1 has been suggested to be a possible explanation for this instability. Here, we characterize the cause of the apparent incompatibility between SGI1 and IncC plasmids using flow cytometry to monitor the segregation of the genomic island and its helper plasmid expressing different fluorescent proteins. While confirming the incompatibility, we AcaCD binding sites are depicted by green angled arrows. Left (attL) and right (attR) junctions with the chromosome are indicated by black bars at the ends. The position of the complex class 1 integron In104 (multidrug resistance region) is indicated by a black arrowhead. B. Fluorescent reporter genes mNeonGreen (green arrow) and mCherry (red arrow) under the control of the P BAD promoter were inserted in pVCR94 between traG C and eexC, and in SGI1 between S009 and S010. C. Representative flow cytometric scatter plots of green and red fluorescence of E. coli KH95 bearing either pVCR94 GreenKn or SGI1 RedKn . D. Schematic representation of the labelled elements in the cells. SGI1 RedKn is integrated into the chromosome at the 3' end of trmE.
https://doi.org/10.1371/journal.pgen.1008965.g001 observed an unexpected high-copy replication of the excised form of SGI1 that only occurred in the presence of the helper plasmid. Replication was confirmed by real-time quantitative PCR as well as DNA extraction and restriction of the replicative form. We identified an iteronbased origin of replication (oriV) downstream of a replication initiator gene whose expression is activated by an AcaCD-dependent promoter. Further investigations showed a clear link between incompatibility and SGI1 replication as replication-deficient mutants allowed stable cohabitation with the helper plasmid, even in the absence of any selection pressure.

Monitoring of SGI1 and pVCR94 incompatibility by flow cytometry
To monitor the fate of SGI1 and its helper IncC plasmid in cell populations by flow cytometry, we constructed SGI1 RedKn and pVCR94 GreenKn by inserting genes expressing red or green fluorescent reporter proteins between two converging genes to avoid interference with key functions ( Fig 1B). Flow cytometry monitoring of E. coli cells bearing either pVCR94 GreenKn or SGI1 RedKn revealed that~97% and~98% of the cells expressed the green and red fluorescent reporters under inducing conditions, respectively ( Fig 1C). These values were the maxima that could be reached for pure populations bearing either element. In this context, we used the green and red fluorescence as a proxy for the presence of the IncC plasmid and SGI1 in cells and assumed that pVCR94 GreenKn was present in a single copy per cell, and that a single copy of SGI1 RedKn was inserted at trmE per chromosome (Fig 1D).
To measure by flow cytometry the previously reported incompatibility between SGI1 and IncC plasmids [23,26], SGI1 RedKn and kanamycin-sensitive pVCR94 Green were introduced together by conjugation in E. coli, and the fluorescence of the resulting cell population was monitored over a 3-day period (~54 generations) in the absence of antibiotics (Fig 2A). While at the beginning of the assays (G0), 97.5% of the cells exhibited both green and red fluorescence, only 51% of the cells produced both signals after 18 generations (G18), and less than 0.5% after 54 generations (G54), thereby suggesting that SGI1 and pVCR94 are incompatible. To confirm that our observation resulted from plasmid or SGI1 loss, not from inactivating mutations in the reporter genes, antibiotics kanamycin and chloramphenicol were added at day 3. Selective pressure for both elements restored dual fluorescence in more than 97% of the cells, and the population of element-free cells nearly vanished (Fig 2A, ATB).
Analysis of the individual scatter plots for the red and green channels revealed additional information regarding the incompatibility phenomenon. While the gradual loss of pVCR94 Green over time yielded two distinct cell populations (IncC + and IncCcells) in the green channel, three distinct cell populations emerged in the red channel (Fig 2C, bottom). At G18,~57% of the cells exhibited high-intensity red fluorescence like 95% of the cells at G0. A second population emerged with a red fluorescent signal weaker than the initial population, but comparable to the intensity emitted by IncC plasmid-free cells bearing a single copy of SGI1 RedKn integrated at trmE (Fig 1C, right). The presence of two cell subpopulations suggests a differentiation into two types of SGI1 + cells, one bearing a single integrated copy of SGI1 R-edKn and the other bearing multiple copies. At G36 and G54, the subpopulation of cells with high-intensity red fluorescence collapsed, while the subpopulations with low or no red signal increased. The collapse of the high-intensity red fluorescent signal subpopulation correlated with the loss of green signal, i.e. the loss of pVCR94 Green (Fig 2C, top), indicating that highintensity red fluorescence was dependent on the presence of the IncC plasmid. Based on these observations, we hypothesized that in IncC + cells, not only does SGI1 excise from the chromosome, but also undergoes a transient replication cycle that increases the production of mCherry protein in cells.

Active replication of SGI1
To test whether SGI1 is able to replicate, we first prevented its excision from the chromosome by using a Δint mutant of SGI1 that is unable to excise and remains permanently integrated in the chromosome [10]. pVCR94 Green was then introduced into E. coli bearing SGI1 Red Δint locked in trmE. We did not observe cells producing a high level of red fluorescence (Fig 3A, compare WT and Δint, and S1A Fig), suggesting that SGI1 locked into the chromosome is unable to replicate, even in IncC + cells.
To confirm SGI1 replication, we attempted to lock SGI1 Red Δint out of the chromosome in a circular, replicative form that we called pSGI1. To do this, int was deleted while SGI1 Red was in its excised, circular state in cells that contained spectinomycin-resistant pVCR94 GreenSp with antibiotic selection to counter incompatibility. This can be achieved because the IncC Incompatibility between SGI1 and IncC plasmids is trackable by FC. A. Evolution of the percentage of E. coli KH95 cells bearing SGI1 RedKn (SGI1) and pVCR94 Green (IncC) and grown over 54 generations in the absence of antibiotics as monitored using FC. Plots represent the mean and standard error of the mean obtained from three independent experiments. ATB is a recovery control culture at G54 in LB with selective pressure for both elements. B. Representative FC density plots of the data presented in panel A mapping the green signal (513 nm) as a function of the red signal (610 nm). C. Representative FC density plots of fluorescence intensity over forward scatter corresponding to data presented in panel B. Color keys are identical in all panels. In panel C, pink and red indicate cells producing low-and high-intensity red fluorescence, respectively.
https://doi.org/10.1371/journal.pgen.1008965.g002 plasmid stimulates SGI1 excision by activating the expression of xis through AcaCD [20]. Plasmid DNA was extracted from cells carrying the resulting mutant and digested with EcoRI. A clear restriction pattern was expected only in conditions that would allow replication of a high number of SGI1 molecules. As expected, no restriction pattern was visible for the strain bearing chromosomally locked-in SGI1 Red Δint, (Fig 3C, lane 1). In contrast, EcoRI restriction patterns were easily detected for SGI1 Red or locked-out pSGI1 (Fig 3C, lanes 2 and 3). Only highcopy number replication of SGI1 can explain our ability to extract, recover, and visualize plasmid DNA of the excised form of this genomic island.
To measure the variation of SGI1 copy number due to replication, we quantified SGI1 copy number per chromosome using real-time quantitative PCR (qPCR). In the IncC + strain, SGI1 Red copy number increased 12-fold compared to the IncCstrain and pSGI1 was Ethidium bromide-stained 0.8% agarose gel electrophoresis of EcoRI-digested plasmid DNA preparation of the same IncC + cells bearing SGI1 Red , its locked-in Δint mutant or locked-out Δint mutant (pSGI1). Lane 1, locked-in trmE::(SGI1 Red Δint::aph); Lane 2, SGI1 Red ; Lane 3, pSGI1; Lane L, 2-Log DNA ladder. D. Effect of deletions on SGI1 copy number. Quantification by qPCR of the SGI1 copy number as measured by the sgiA/chromosome ratios at G0. E. Effect of deletions on SGI1 excision. Quantification by qPCR of the percentage of cells at G0 that contain excised SGI1 as a measure of unoccupied attB sites per 100 chromosomes. "x" indicates that excision was below the limit of detection (<10 −5 %). Assays were carried out in E. coli KH95 carrying (+) or lacking (-) the helper IncC plasmid pVCR94 Green or pVCR94 GreenSp (for pSGI1 only). The bars represent the mean and standard error of the mean obtained from at least three independent experiments. Statistical analyses were carried out on the values using the one-way ANOVA with Tukey's multiple comparison test. F. Effect of SGI1 replication on conjugative transfer of SGI1 and its helper plasmid. Conjugation assays were carried out using E. coli KH95 as the donor and E. coli VB113 as the recipient. When indicated (+), Rep was provided in trans from prep. Transfer frequencies are expressed as the number of transconjugant per donor CFUs. Statistical analyses were carried out on the logarithm of the values using the one-way ANOVA with Sidak's post-test to compare each mutant set relatively to SGI1 Red (WT) control. The bars represent the mean and standard deviation values obtained from at least three independent experiments. Statistical significance is indicated as follow: ���� , P<0.0001; ��� , P<0.001; � , P<0.05; ns, not significant.
https://doi.org/10.1371/journal.pgen.1008965.g003 maintained at~6 copies per cell ( Fig 3D). Altogether, these results confirm that the subpopulation of cells that produced high-intensity red fluorescence in IncC + cells corresponds to cells containing an excised replicative form of SGI1.

rep is required for SGI1 replication
Although its role in replication has never been demonstrated, the putative protein encoded by S003, also known as rep, contains a RepA_C domain (Pfam PF04796). The predicted translation product of rep shares 60% identity (78% similarity) with the putative replication initiator protein RepA encoded by plasmid pXAC64 of Xanthomonas axonopodis pv. citri [27]. To determine whether rep plays a role in SGI1 replication, we constructed a rep deletion mutant of SGI1 Red , and tested its ability to replicate in cells bearing pVCR94 Green using flow cytometry and qPCR to quantify its rate of excision and copy number. The data were compared with wild-type SGI1 Red and its ΔtraN S mutant as positive controls, and its Δint and Δxis mutants as negative controls. Excision of wild-type SGI1 Red was barely detectable in IncCcells (0.0000838%) whereas 81% of the cells contained excised SGI1 Red in IncC + cells (~7-log increase) ( Fig 3E). SGI1 Red and its ΔtraN S mutant behaved alike in IncC + cells, with virtually all cells bearing the excised, replicative form (>80% excision and >10 copies per chromosome) ( Fig 3D and 3E). As expected, deletion of int or xis abolished excision, and led to a low copy number of SGI1 Red that was comparable to the wild-type in IncCcells. In contrast, while the Δrep mutant retained the ability to excise from the chromosome (>52% excision), it remained under 2 copies per cell (Fig 3D and 3E), thereby supporting a key role of rep in SGI1 replication.
Flow cytometric data confirmed the inability of the Δint, Δxis and Δrep mutants to replicate. Only low-intensity red fluorescence was detected despite the presence of the helper IncC plasmid, unlike the ΔtraN S mutant ( Fig 3A, and S1A, S2, S3A and S4A Figs). Complementation of Δint, Δxis and Δrep mutations by expressing the corresponding gene under the control of the P BAD promoter restored the high-intensity red fluorescence phenotype of the cells, indicating a restoration of SGI1 replication (S1B, S3B and S4B Figs).

Identification of the origin of replication (oriV) of SGI1
Based on the presence of a rep gene and the inability of SGI1 Δrep to replicate, we searched for nearby intergenic regions that could act as a potential oriV locus. Sequence analyses revealed a putative oriV immediately downstream of rep ( Fig 4A). This locus contains four GC-rich 17-bp direct repeats (iterons) and a putative DnaA box flanked by two AT-rich regions. The longest AT-rich region contains three direct and inverted copies of an imperfect 8-bp repeat ( Fig 4B). This configuration resembles the origin structure of typical iteron-containing replicons [28]. To confirm that this putative oriV is essential for SGI1 replication, it was replaced with a Kn R cassette without any alteration of the promoter region of xis. The resulting 255-bp deletion abolished SGI1 Red replication as shown by the lack of highly red fluorescent cells despite the presence of pVCR94 Green (Fig 3A). Moreover, SGI1 Red ΔoriV retained the ability to excise at the same level as wild-type SGI1 Red ; however, its copy number per cell was considerably reduced and remained comparable to SGI1 Red in IncCcells or to its Δrep mutant in IncC + cells (Fig 3D and 3E). Together, these results indicate that the sequence downstream of rep contains oriV.

Expression of SGI1 rep is activated by IncC plasmid encoded AcaCD
To test the role of the IncC helper plasmid in SGI1 replication, the locked-out replicative form pSGI1 was subjected to the cohabitation test with pVCR94 GreenSp for 54 generations in the absence of selective pressure and analyzed by flow cytometry (Fig 5A and S5 Fig). As observed for SGI1 Red , incompatibility led to the loss of both elements. However, while pVCR94 GreenSp  persisted in 50% of the cells, pSGI1 was lost in virtually all cells after 54 generations. The other half of the cell population consisted of cells lacking both elements. This behavior is in stark contrast with SGI1 Red which was retained chromosomally integrated after 54 generations in more than 45% of the cells while the IncC plasmid was retained in less than 11% of the cells, with less than 0.5% of the cell population containing both elements (Fig 2). Strict correlation of the plots of SGI1 + cells and SGI1 + IncC + cells suggests that pSGI1 replication and stability are directly linked to the IncC plasmid. As pSGI1 did not integrate into the chromosome due to the missing int gene, we hypothesize that the IncC plasmid codes for an essential factor that sustains pSGI1 replication.
Our previous identification of an AcaCD binding site upstream of the S004-rep gene cluster suggests that rep could be expressed under the control of the IncC plasmid-encoded transcriptional activator AcaCD (Fig 1) [20]. To test this hypothesis, we first constructed a translational fusion between the fifth codon of rep and the eighth codon of lacZ in SGI1 Kn (Fig 4C and S6A  Fig). This fusion also removes xis and rep, thereby preventing SGI1 excision and replication. β-galactosidase activity of the rep'-'lacZ fusion was then used as a proxy to measure the level of rep expression (transcription and translation) in the presence or absence of AcaCD. β-galactosidase activity was undetectable in IncCcells or in cells lacking pacaCD ( Fig 4D). In contrast, when AcaCD was provided by pacaCD with arabinose induction, we observed a 172±21-fold increase in β-galactosidase activity, and the presence of pVCR94 Sp resulted in a 11.9±0.3-fold increase compared to strain containing its ΔacaCD mutant. Comparable results were observed using a fusion that retained xis, allowing SGI1 excision but not its replication (S6B and S6C  Fig).
Furthermore, to confirm that S004 and rep are part of the same mRNA transcript initiated at P S004 , we conducted a PCR assay aimed at amplifying both genes from the same cDNA. To do this, RNA was extracted from cells bearing SGI1 Red alone, or in the presence of pacaCD with or without arabinose. rep-specific mRNA was reverse transcribed using a primer located at the 3' end of rep and used in a PCR assay to amplify fragments located within S004 and rep ( Fig 4C). Our results confirmed that these two genes are cotranscribed and show unambiguously that S004 and rep are part of the same operon that is activated by AcaCD (Fig 4E).

SGI1 replication promotes instability of the IncC plasmid and inhibits cotransfer
We observed that SGI1 replication and IncC instability are strongly correlated. SGI1 replication resulted in a strong incompatibility phenotype as shown by low counts of SGI1 + IncC + cells that remained at G36 and G54 (Figs 2A and 5B and S2 Fig). In contrast, mutants impaired for excision (Δint and Δxis) or replication (Δrep and ΔoriV) exhibited a drastically reduced incompatibility as over 85% of the cells retained both SGI1 and the IncC plasmid at G54 (Fig  5C-5F). Therefore, replication of SGI1 is responsible for the incompatibility phenotype observed with IncC plasmids.
SGI1 and several of its variants were previously shown to inhibit cotransfer of IncC plasmids such as pRMH760 and pVCR94 [20,26]. To test whether SGI1 replication could play a role in this inhibition, we carried out mating assays using the Δrep and ΔoriV mutants of SGI1. These mating assays revealed that the frequency of cotransfer was comparable to the transfer of each single element (Fig 3F), indicating that both mutations alleviated cotransfer inhibition. Furthermore, transfer of pVCR94 increased by 3 logs in both mutants suggesting that SGI1 replication strongly impairs the transfer of the IncC plasmids. We also observed that inhibition of cotransfer was restored as the frequency of cotransfer dropped below the detection limit when the Δrep mutation was complemented by providing Rep from prep. In this context, we observed a 27-fold reduction of pVCR94 transfer compared to the Δrep mutant (Fig 3C), while transfer of SGI1 increased 6-fold, thereby suggesting that replication provides a significant advantage to SGI1 compared to the IncC plasmid.

SGI1-like elements can encode an alternative replication initiator protein
A search for divergent SGI1 homologues sharing a genomic structure similar to SGI1 and GI-15 of V. cholerae O1 B33 (sulfamethoxazole and streptomycin resistance) revealed two SGI1like elements with an alternative rep gene located at the same position as SGI1 rep. The first one, named GIVchO27-1, was identified in the unique chromosome of V. cholerae O27 strain 10432-62, a strain isolated in the feces of a patient with diarrhea in 1962 in the Philippines [29]. The second one, named GISen-26, was found in the genome of S. enterica serovar Muenster 26, a multidrug resistant strain isolated from horse feces at a Texan equine referral hospital. GISen-26 carries a class I integron (aadA1, sul1, qacEΔ1), a Tn3-like transposon (strAB, tetA), and a mercury resistance transposon (merRTPFADE). Both GIs are integrated at trmE and encode a replication protein unrelated to SGI1 and GI-15 Rep (Fig 6). This alternative replication initiator protein contains an N-terminal replicase (PF03090) domain and an adjacent primase C terminal 1 (PriCT-1, PF08708) domain instead of the RepA_C domain. As expected in this context, SGI1 oriV is missing, and likely replaced by an alternative oriV that remains to be characterized.

Discussion
Mobile genetic elements such as conjugative plasmids and mobilizable genomic islands are potent drivers of antibiotic resistance gene dissemination. A better understanding of their lifecycle and interactions is essential to devise novel strategies aimed at curbing their propagation. In this study, we unraveled a yet unforeseen aspect of the complex interactions between IncC plasmids and SGI1, two types of mobile genetic elements that are both frequently found in several species of Enterobacteriaceae including Salmonella enterica, Morganellaceae and Vibrionaceae [4,[32][33][34]. Previous reports of incompatibility between IncC (and IncA) plasmids and SGI1 emphasize the complexity of these interactions, especially when considering that SGI1 needs IncC or IncA plasmids to propagate [20,23,26]. We undertook to investigate the underlying cause of incompatibility by labelling SGI1 and the helper IncC plasmid pVCR94 with fluorescent reporter genes to follow by flow cytometry their respective fates in a cell population. This strategy allowed us to confirm that SGI1 and IncC plasmids are incompatible as cells segregated into two subpopulations containing one or the other element after only a few generations. Furthermore, different cells contained different configurations of SGI1. Besides IncC plasmid-free cells that bear quiescent SGI1 integrated at trmE, we found that cells bearing an IncC plasmid contain SGI1 in an excised, high-copy replicative state. In IncC plasmid-free cells, virtually no excision of SGI1 was detected, whereas in IncC + cells, up to 12 copies of SGI1 could be found in more than 80% of the cells. DNA of the replicative form of SGI1 (pSGI1) was easily extracted and profiled by EcoRI restriction, confirming the high copy number per cell.
We found that replication requires prior excision of SGI1 from the chromosome which occurs only in the presence of an IncA or IncC plasmid [10,11]. Hence, replication of SGI1 mutants lacking xis or int could not be detected, even if initiation of replication of integrated SGI1 cannot be ruled out. Our analysis of SGI1 sequence revealed an iteron-based replicon with a rep gene coding for an uncharacterized replication initiator protein. rep is preceded by S004, a gene coding for a protein of unknown function that contains a predicted helix-turnhelix domain (HTH_17, Pfam PF12728). Murányi et al. [35] presented evidence supporting an expression of S004 driven from the AcaCD-responsive promoter P S004 , while expression of the downstream gene rep (S003) would be very low and driven from a constitutive weak promoter located within S004 regardless of AcaCD. Their analysis of rep expression relied on β-galactosidase assays with a transcriptional lacZ fusion that substituted lacZ for rep at the start codon in a multi-copy reporter plasmid containing most of S004 and its promoter region (position 3,261-3,639 of SGI1) seemingly displacing S004 stop codon. In contrast, we devised a translational rep'-'lacZ fusion constructed in SGI1. In this more natural context, our results show that S004-rep is an operon driven from P S004 , directly under the control of AcaCD. Hence excision and replication of SGI1 are coordinated events controlled by the same transcriptional activator produced by IncA and IncC plasmids. Consistent with this observation, we found that replication of pSGI1, the locked-out form of SGI1 Δint, cannot persist in the absence of the helper IncC plasmid. Furthermore, consistent with AcaCD-induced expression of rep, destabilization of the IncC plasmid triggered by pSGI1, ultimately resulted in the concomitant loss of pSGI1 (Fig 5A).
We also found that although S004-rep expression was off in IncCcells, the operon was weakly expressed in the presence of a ΔacaCD mutant of the helper plasmid (Fig 4D). This observation suggests that another IncC plasmid-encoded factor triggers rep expression, directly or indirectly, perhaps through the activation of sgaCD (aka flhDC SGI1 ) expression. sgaCD is carried by SGI1 and encodes a close homologue of AcaCD (89% and 67% identity for the C and D subunits, respectively) [21,35,36]. SgaCD has been shown to activate AcaCDdependent promoters of SGI1, though at a lower level than AcaCD [35]. While the exact role of SgaCD in SGI1 biology is currently unknown, a natural variant of SGI1 that lacks sgaCD, SGI1-K, has been shown to be unable to destabilize the IncC plasmid pRMH760 [26], suggesting a possible link between SgaCD, SGI1 replication and IncC plasmid destabilization. Identification of the IncC plasmid-encoded factor and pathway that control rep expression in the absence of acaCD is ongoing.
Little is known about SGI1's replicon as the product of rep has not been characterized. To the best of our knowledge, the closest relative of SGI1 Rep is RepA of plasmid pXAC64 of X. axonopodis pv. citri that was identified by sequencing only and not studied further [27]. Therefore, additional studies are needed to better understand the molecular mechanism by which SGI1 replicates, the control mechanisms that affect the copy number of its excised form as well as the mechanism, whether direct or indirect, by which it affects the stability of IncC plasmids. We showed that suppression of SGI1 excision and/or replication functions (xis, int, rep, oriV) enables stable coexistence of SGI1 and its helper plasmid. Surprisingly, overexpression of rep in complementation assays led to a destabilization of SGI1 over the IncC plasmid (Fig 5E). High intracellular Rep concentration could inhibit replication through "handcuffing", a mechanism that couples replication origins via iteron-bound Rep proteins, turning off origin function as reported for the π replicase of plasmid R6K [37]. Replication of iteron-based plasmids involves the recruitment of the chromosomal replication initiator protein DnaA. We identified in the oriV of SGI1 a putative DnaA box and four 17-bp iterons (consensus RKGGGGGHRATTATGCG) (Fig 4A). IncA and IncC replicons also contain a single DnaA box and 11 to 14 copies of 19-bp iterons that differ in sequence (yaTRTGGGDNHgcTGCACG and yaTRTGGGNNcgcTGCACG, respectively) with SGI1 iterons [38,39]. Destabilization of IncA and IncC plasmids by SGI1 replication could result from the titration of host-encoded replication proteins by the oriV of replicating SGI1. DnaA titration by the DnaA box of multicopy SGI1 could be a mechanism preventing proper initiation of IncA and IncC plasmid replication. Alternatively, SGI1 could produce a protein that interferes with key maintenance functions that are conserved in IncA and IncC plasmids such as replication (repA), partition (parAB, 053) or post-segregational killing (tad-ata) [20,40]. This factor would be produced when SGI1 is excised and replicating, and not produced or produced in insufficient quantity when in single copy in the cell, that is in cells devoid of helper plasmid.
Integrative and conjugative elements (ICEs), such as ICEBs1 of Bacillus subtilis, ICESt3 of Streptococcus thermophilus, Tn916 of Enterococcus faecalis, R391 of Providencia rettgeri or ICEclc of Pseudomonas putida, were shown to undergo transient replication after excision from the chromosome [41][42][43][44][45]. For these mobile elements, increased copy number results from an intercellular rolling-circle replication mechanism initiated at the oriT locus by the conjugative relaxase that is used as a replication initiator protein. In contrast, we found that replication of SGI1 relies on a dedicated rep gene and oriV that are distinct from the oriT and mpsAB mobilization genes [17]. In this respect, SGI1 resembles integrative and conjugative elements found in Actinomycetes such as Streptomyces pSAM2 [46]. These self-transmissible elements rely on a single FtsK/SpoIIIE-like protein channel that translocates double-stranded molecules into adjacent cells within the hyphae of these filamentous bacteria. Hence to prevent loss in the donor cell, they replicate using a dedicated oriV and rolling-circle replication initiator proteins, such as RepSA, RepAM or Rep2, that seem to be expressed only after excision from their host chromosome [47,48]. Staphylococcus aureus pathogenicity islands (SaPIs) and recently discovered phage-inducible chromosomal islands (PICIs) in Enterobacteriaceae and Pasteurellaceae are phage satellites that parasite temperate helper bacteriophages for their own dissemination [49]. Like SGI1, SaPIs and PICIs excise from the chromosome of their host and undergo active replication in the presence of their helper element. Their replication, which is usually mediated by a primase and replicase that they encode, is necessary to allow packaging of the phage satellite genome into viral particles. Whereas in all these instances, the biological role of replication is clear, the function ensured by SGI1 replication remains uncertain. The need for a dedicated replicon on SGI1 for the sole purpose of retaining a copy of SGI1 in donor cells after conjugative transfer seems to be overkill as this function is expected to be ensured by the template strand during intracellular rolling-replication initiated at oriT, unless SGI1 lacks a functional single-strand origin (sso) that could be required to resynthesize the transferred strand from the template strand as shown for Tn916 or ICEBs1 [43,50]. Nevertheless, the discovery of SGI1-like elements with alternative rep genes (Fig 6) suggests an important role of replication in the life cycle of genomic islands of the SGI1 family at large.
Unambiguously, replication of SGI1 has a destabilizing effect on IncC plasmid maintenance. In addition, replication functions seem to play a role in the reduction of cotransfer of SGI1 and its helper plasmid, which in return could help stabilize SGI1 in its new host by preventing futile excision and eventual loss. Furthermore, as shown previously, SGI1 reshapes the mating pore encoded by the IncC plasmid to promote its self-propagation. SGI1 encodes alternative TraN, TraG, and TraH subunits, which results in the replacement of the cognate subunits encoded by the IncC plasmid in the mating pore [18]. This substitution was shown to enhance the transmissibility of SGI1. High SG1 copy number could increase the production of TraN S , TraG S , and TraH S , facilitating the alteration of the mating pore to enhance SGI1 propagation at the expense of the helper plasmid. Consistent with this hypothesis, deletion of rep or oriV suppressed the inhibition of IncC plasmid transfer phenotype (Fig 3F). SGI1 replication is linked to its incompatibility with IncC plasmids, and one may wonder what evolutionary benefit SGI1 can get from destabilizing its helper element. IncC plasmids mobilize other genomic islands, such as the multidrug resistance island MGIVchHai6 found in clinical non-O1/non-O139 V. cholerae isolates [20,[51][52][53]. Yet, unlike SGI1, MGIVchHai6 lacks a rep homologue and transfers at a much lower rate [51]. Furthermore, MGIVchHai6 is not as prevalent in Gammaproteobacteria. One reason could be that by allowing the co-residence of its helper plasmid in the cell, MGIVchHai6 is subjected to a high rate of excision that undermines its stability during cell division. On the contrary, SGI1 prevents long-term co-residence of its helper plasmid, as it has been shown to inhibit co-transfer and promote plasmid loss [26]. A key element of SGI1 epidemiological success could be the association between replication and entry exclusion evasion. SGI1 evades entry exclusion of IncC plasmids, allowing its transfer to recipient cells containing an IncC plasmid [18,19]. Upon entry into such cells, SGI1 cannot stably integrate into the chromosome. Instead, SGI1 actively replicates which could favor its epidemic spread into neighboring IncC + cells, thanks to entry exclusion evasion. Ultimately, by triggering population-wide helper plasmid loss, SGI1 prevents xis and rep expression due to AcaCD depletion, thereby promoting its stabilization through integration into the host chromosome.
SGI1 and its siblings seem to have evolved a fiery love/hate relationship with their helper plasmids, relying exclusively on them to disseminate while preventing their cotransfer and mid-and long-term coexistence within the same cell. Based on our results, SGI1 replication likely plays an important role in this complex interaction. Even so, how exactly replication affects plasmid stability remains to be unraveled. One key aspect to investigate is the influence of copy number shift and how it affects gene expression and production of SGI1-specific proteins that may interfere with the replication, partition in daughter cells and transfer of IncA and IncC plasmids.

Molecular biology methods
Genomic and plasmid DNA were prepared using the QIAmp DNA Mini Kit (Qiagen) and EZ-10 Spin Column Plasmid DNA Minipreps Kit (Biobasic), respectively, according to the manufacturer's instructions. Oligonucleotide primers were purchased from Integrated DNA Technologies. When necessary, PCR products were purified using the QIAquick PCR Purification Kit (Qiagen) according to the manufacturer's instructions. Ligations were performed using T4 DNA ligase (New England Biolabs) with cloning vectors that were dephosphorylated using Antarctic Phosphatase (New England Biolabs) according to the manufacturer's instructions. E. coli was transformed by electroporation according to Dower et al. [58]. Electroporation was carried out in a BioRad GenePulser Xcell apparatus set at 25 μF, 200 V and 1.8 kV using 1-mm gap electroporation cuvettes. Sequencing reactions were performed by the Plateforme de Séquençage et de Génotypage du Centre de Recherche du CHUL (Québec, QC, Canada).

Construction of insertion and deletion mutants
All insertion and deletion mutants in pVCR94 and SGI1 were constructed in E. coli BW25113 using the one-step chromosomal gene inactivation technique [54]. The λRed recombination system was expressed using pMS1 as described by Datta et al. [59]. All mutations were designed to be non-polar. SGI1 Cm , a Cm-resistant derivative of SGI1 devoid of In104 was obtained using primer pair SGI1In104cm2.f/SGI1In104cm2.r and pKD3 as the template. SGI1 RedKn was constructed by introducing araC-P BAD ::mCherry-FRT-aph-FRT into the intergenic region between S009 and S010 in SGI1 Cm using primer pair FwpBADInsMCSGI1/ RvpBADInsMCSGI1 and pRedKnFRT as the template. pVCR94 GreenKn and pVCR94 GreenSp were constructed by introducing araC-P BAD ::mNeonGreen-FRT-aph-FRT between traG and eexC in pVCR94 and pVCR94 Sp using primer pair FwpBADInsNGpVCR/RvpBA-DInsNGpVCR, and pGreenKnFRT as the template. The FRT-flanked Kn R cassette was removed by Flp-catalyzed excision using pCP20 [54]

Cohabitation assays
All cohabitation experiments were done as follows. The strains containing both elements were inoculated in LB broth with selective pressure (Kn for SGI1 RedKn or SGI1 Red deletion mutants, and Tc for pVCR94 Green ) and arabinose for induction of the reporter genes, and grown overnight at 37˚C. On the next day, the cultures were used to inoculate (1:2,000 dilution) fresh LB broth with arabinose without antibiotics and incubated at 37˚C. Furthermore, the overnight grown culture (stationary phase cells) was diluted 1:1,000 in PBS and analyzed by flow cytometry. This quantification was defined as the initial population composition (G0). This population was considered as pure and suitable for the experiment if more than 95% of the cells were positive for green and red fluorescence. Subsequently, cultures were passaged twice a day in fresh medium, which equals approximately 9 generations per passage. Flow cytometry analyses were done on stationary phase cells at days 1, 2 and 3, corresponding approximately to generations 18, 36 and 54. At the end of the experiment, the last passage was performed with selective pressure to restore the initial population. This control confirmed that loss of fluorescence resulted from element instability, not from mutations in the reporter genes. For mutant complementation assays, the same conditions were used except that pVCR94 Green was replaced with pVCR94 GreenSp due the Ap R phenotype conferred by both pVCR94 Green and pBAD30.

Flow cytometry
Culture samples were diluted 1:1,000 in 500 μl of PBS. Fluorescence intensity of NeonGreen and mCherry in cells was monitored by flow cytometry analysis on a BD FACSJazz (BD Biosciences), and data were acquired with the BD FACS Sortware. mNeonGreen and mCherry were excited with 488 and 561 nm solid-state lasers, and their emission was detected using 513/17 and 610/20 nm emission filters, respectively. For each sample, fluorescence of 20,000 cells was captured, and the data was analyzed using FCS Express 7 (De Novo Software).

RNA extraction and cDNA synthesis
Briefly, RNA extractions were performed as follows. E. coli KH95 containing SGI1 Red with or without pacaCD was grown at 37˚C for 16 h in LB broth containing the appropriate antibiotics. The cultures were diluted 1:100 in fresh medium containing the appropriate antibiotics and grown to an OD 600 of 0.2 before being diluted 1:10 again in fresh medium containing the appropriate antibiotics and supplemented with 0.02% arabinose when needed. After a 2h incubation period, 1 ml of the culture was used for total RNA extraction using Direct-zol RNA extraction kit (Zymo Research) and TRI Reagent (Sigma-Aldrich) according to the manufacturer's instructions. Once purified, the RNA samples were treated using 2 units of DNase I (New England Biolabs) according to the manufacturer's instructions to eliminate any residual gDNA. cDNA was synthesized from 0.2 μg of RNA and 2 pmol of gene-specific primer rep_RT (Integrated DNA Technologies), using the reverse transcriptase SuperScript III (Invitrogen), according to the manufacturer's instructions. Control reactions without reverse transcriptase treatment ('noRT') were performed for each sample. PCR reactions aiming at amplifying rep, S004 and traN S were carried out using cDNAs as described in Garriss et al. [60].

β-galactosidase assays
Qualitative assays were performed by depositing 10μl aliquots of overnight cultures with appropriate antibiotics on solid agar supplemented with 5-bromo-4-chloro-3-indolyl-β-Dgalactopyranoside (X-gal) with or without 0.02% arabinose. The plates were observed after an overnight incubation at 37˚C. Quantitative assays were performed with 2-nitrophenyl-β-Dgalactopyranoside (ONPG) according to a protocol adapted from Miller [61]. After an overnight incubation at 37˚C with appropriate antibiotics with or without 0.2% arabinose, cultures were diluted 1:100 in 50ml LB broth with appropriate antibiotics with or without 0.2% arabinose and grown until an OD 600 of 0.2 was reached. Two series of 1:10 dilutions were then prepared in total volumes of 5ml LB broth supplemented with appropriate antibiotics with or without 0.2% arabinose and incubated for 2 hours at 37˚C. Values are expressed as the mean and standard error of the mean values calculated from three biological replicates.

qPCR
Quantification of the copy number of pVCR94 as well as excision rate and copy number of SGI1 forms (circular or integrated) were assessed by real-time quantitative qPCR. Primers used in this quantification are listed in S1 Table. The data were analyzed using the threshold cycle (ΔΔC T ) method using the three target genes dnaB, hicB and trmE as chromosomal references. Copy number of pVCR94 and SGI1 was assessed using repA and sgiA as the targets, respectively. Excision of SGI1 was assessed using the chromosomal free attB site as the target. qPCR experiments were performed in triplicate on the RNomics Platform of the Laboratoire de génomique fonctionnelle de l'Université de Sherbrooke (https://rnomics.med.usherbrooke. ca) (Sherbrooke, QC, Canada).

Identification of oriV
Search for direct and inverted sequence repeats in the region downstream of S003 was carried out using the genomic similarity search tool Yass [62] available at https://bioinfo.lifl.fr/yass/ index.php and the MEME motif discovery algorithm [30] available at http://meme-suite.org/ index.html. Identification of putative DnaA boxes was carried out using MAST [63] with the DnaA motif matrix (accession MX000098) obtained from PRODORIC Release 8.9. Prediction of Rho-independent transcription terminator was carried out using ARNold [64] available at http://rssf.i2bc.paris-saclay.fr/toolbox/arnold/index.php.