An epigenetic switch activates bacterial quorum sensing and horizontal transfer of an integrative and conjugative element

Abstract

Horizontal transfer of the integrative and conjugative element ICEMlSym^R7A converts non-symbiotic Mesorhizobium spp. into nitrogen-fixing legume symbionts. Here, we discover subpopulations of Mesorhizobium japonicum R7A become epigenetically primed for quorum-sensing (QS) and QS-activated horizontal transfer. Isolated populations in this state termed R7A* maintained these phenotypes in laboratory culture but did not transfer the R7A* state to recipients of ICEMlSym^R7A following conjugation. We previously demonstrated ICEMlSym^R7A transfer and QS are repressed by the antiactivator QseM in R7A populations and that the adjacently-coded DNA-binding protein QseC represses qseM transcription. Here RNA-sequencing revealed qseM expression was repressed in R7A* cells and that RNA antisense to qseC was abundant in R7A but not R7A*. Deletion of the antisense-qseC promoter converted cells into an R7A*-like state. An adjacently coded QseC2 protein bound two operator sites and repressed antisense-qseC transcription. Plasmid overexpression of QseC2 stimulated the R7A* state, which persisted following curing of this plasmid. The epigenetic maintenance of the R7A* state required ICEMlSym^R7A-encoded copies of both qseC and qseC2. Therefore, QseC and QseC2, together with their DNA-binding sites and overlapping promoters, form a stable epigenetic switch that establishes binary control over qseM transcription and primes a subpopulation of R7A cells for QS and horizontal transfer.

INTRODUCTION

Mobile genetic elements (MGE) contribute to prokaryotic evolution by moving DNA within and between genomes. Gene acquisition events facilitated by MGE are pivotal to the diversification and competitive success of bacteria (1). Conjugative plasmids and integrative-and-conjugative elements (ICEs) disseminate large clusters of genes that endow competitive advantages upon their hosts in particular niches (2). Horizontal transfer can benefit both the recipient bacterium and the MGE, but horizontal transfer may also impart fitness costs on donor cells (1,3,4). The negative impacts of MGE transfer on the bacterial host have likely favored the evolution of MGE that activate transfer infrequently in response to multiple endogenous and environmental cues. This may involve mechanisms that ensure only a small proportion of the cell population act as donors (5–8). In this work, we describe how the Mesorhizobium japonicum R7A (formerly M. loti R7A (9)) symbiosis island ICEMlSym^R7A controls a differentiation event that produces a sub-population of cells that transfer ICEMlSym^R7A at higher frequency. This state termed R7A* is vertically inherited but resets in recipients following horizontal transfer of ICEMlSym^R7A.

Mesorhizobia are soil bacteria that can host ICEs known as ICESyms carrying genetic cargo essential for establishment of a nitrogen-fixing symbiosis with legumes (10–12). Following transfer to non-symbiotic Mesorhizobium recipients, ICESyms integrate into Mesorhizobium spp. chromosomes by recombining with the 3′ ends of highly conserved chromosomal genes and the recipients gain the ability to induce nitrogen-fixing root nodules on specific legume hosts. For example, ICEs in Mesorhizobium spp. have been identified that specify symbioses with species of Lotus, Biserrula and Cicer (12–14). ICESyms carry a complement of conserved core genes required for their excision and conjugation (12,15–18) and orthologs of these genes are identifiable on ICEs and plasmids throughout the proteobacteria (15), suggesting this large family of conjugative elements has evolved from a highly successful common ancestor. The paradigm ICESym, ICEMlSym^R7A, was identified in M. japonicum R7A following in situ transfer of the ICE to indigenous non-symbiotic mesorhizobia in New Zealand soils, where transfer converted recipients into symbionts of the pasture legume Lotus corniculatus (14,19). ICEMlSym^R7A integrates into the 3′ end of the sole phe-tRNA gene in R7A and other mesorhizobia through site-specific recombination facilitated by the recombinase/integrase IntS (14).

ICEMlSym^R7A excision from the chromosome and subsequent conjugative transfer is activated by small diffusible N-acyl-homoserine lactones (AHL), a phenomenon known as quorum sensing (QS) (Figure 1). The LuxR-family QS protein TraR encoded by ICEMlSym^R7A activates transcription of the traI1 gene also present on ICEMlSym^R7A. The AHL synthase TraI1 then produces N-(3-oxohexanoyl)-L-homoserine lactone (3-oxo-C6-HSL) as the major AHL product (16). As well as activating traI1 and AHL production, TraR stimulates ICEMlSym^R7A excision and conjugative transfer by activating transcription of the transcriptional activator FseA encoded in a separate operon downstream of the traR gene. The FseA sequence is coded across two open-reading frames msi172 andmsi171. A low-frequency programmed + 1 ribosomal frameshift occurs within the 3′ end of msi172 fusing the coding sequences of msi172-msi171 culminating in production of a single polypeptide. FseA activates transcription of the rdfS promoter, which stimulates expression of the recombination directionality factor RdfS, the prepilin peptidase TraF and the lytic transglycosylase Msi107 and leads to ICEMlSym^R7A excision and transfer (16,18) (Figure 1A).

Figure 1.

Model of excision and transfer regulation by genes encoded on ICEMlSym^R7A. (A) Gene maps of the traR-qseC2 and rdfS-rlxS regions present on ICEMlSym^R7A are shown with a schematic detailing the molecular activation and repression of QS and ICEMlSym^R7A excision and transfer. Proteins with a positive effect on QS, excision and conjugative transfer are shown in blue, while the QseM antiactivator is shown in red. In cells active for QS, AHLs produced by TraI1 activate TraR which activates transcription from two tra boxes, located upstream of traI1 and the likely pseudogene traI2 (16). traI2 forms an operon with two open-reading frames msi172 and msi171. A single polypeptide product, FseA, is translated from msi172-msi171 following a programmed + 1 ribosomal frameshift encoded within the 3′ end of msi172 (18). FseA activates transcription from the rdfS promoter and RdfS stimulates ICEMlSym^R7A excision and transfer (17,20). rdfS is coded upstream of genes encoding the conjugative prepilin peptidase TraF, the lytic transglycosylase Msi107 and the conjugative relaxase RlxS. In cells repressed for QS, the antiactivator QseM binds TraR and FseA, inhibiting their ability to act as transcriptional activators. The QseC protein controls expression of QseM. QseC both activates its own gene promoter from a leaderless mRNA and represses qseM expression. A schematic illustrating the overlapping -35 regions of the qseC and qseM promoters (PqseC and PqseM) and the QseC binding regions O_L and O_R is shown in the shaded box. Note the role of qseC2 (shown in yellow) is not incorporated in this model yet as it had not been studied prior to this work. (B) Illustration and summary of the features of cells that are either in a transfer-repressed state expressing QseM (shaded red) or those repressed for qseM expression (in blue), which participate in QS and maintain an excised ICE in stationary-phase populations and exhibit elevated rates of conjugation.

In wild-type R7A populations, the activation cascade stimulating ICEMlSym^R7A excision and transfer is strongly repressed. Production of 3-oxo-C6-HSL is barely detectable in stationary-phase cultures and ICEMlSym^R7A is excised in only ∼5% of cells. Such cultures transfer ICEMlSym^R7A at a rate of 10⁻⁶ to 10⁻⁵ per donor cell. In the majority of R7A cells in the population, TraR and FseA are both independently inhibited through protein–protein interactions with the antiactivator QseM (15,18) (Figure 1). A qseM deletion mutant R7AΔqseM exhibits close to 100% ICEMlSym^R7A excision in stationary-phase cells, 1000-fold increased conjugative transfer and increased AHL production. Importantly, ICEMlSym^R7A excision in R7AΔqseM does not result in loss of ICEMlSym^R7A. While our culture medium does not provide any known selection for maintenance of ICEMlSym^R7A, we have never observed spontaneous loss of ICEMlSym^R7A from R7A or R7AΔqseM cells during laboratory culture. Like several other ICEs (21), ICEMlSym^R7A likely maintains itself in the excised state through rolling-circle replication initiated by its conjugative relaxase, as relaxase mutants do not maintain the excised ICE (16). Furthermore, ICEMlSym^R7A excision in R7AΔqseM remains growth-phase and QS-dependent, as ICEMlSym^R7A reintegrates into the chromosome in most cells during exponential growth when AHL concentrations are low and therefore rdfS is not expressed. In this sense, R7AΔqseM cells are primed for transfer, but excision and transfer remain conditional on QS activation. Overexpression of qseM represses excision of ICEMlSym^R7A to levels below that observed in wild-type R7A populations, indicating that qseM expression is repressed in the ∼5% of cells in wild-type populations carrying the excised form of ICEMlSym^R7A and suggesting that these cells are likely primed for ICEMlSym^R7A transfer like R7AΔqseM. R7A populations are therefore likely a mixture of transfer-repressed (QseM expressed) and transfer/QS-enabled (qseM repressed) cells (Figure 1B).

Transcription of qseM is regulated by a helix-turn-helix protein QseC, encoded by the divergently transcribed qseC gene (Figure 1) (15). QseC dimers bind each of two operator sequences O_L and O_R that overlap divergent promoters for qseM and qseC. QseC preferentially binds O_L and appears to only strongly bind O_R in the presence of QseC-bound O_L, indicating cooperativity. The PqseC promoter -35 region is positioned within O_R and QseC activates PqseC transcription (15). The -35 region of the PqseM promoter is also positioned within the O_R site on the opposite DNA strand. Deletion of qseC leads to increased expression of QseM, which in turn inhibits ICEMlSym^R7A excision and conjugative transfer and prevents TraR-mediated activation of AHL production (15). We have postulated that QseM, QseC, O_L and O_R together with the overlapping qseC and qseM promoters comprise a molecular switch that derepresses QS and ICE excision in the ∼5% of cells carrying an excised copy of ICEMlSym^R7A in stationary-phase populations of R7A (15).

QseC and its operator sequences share striking similarities with the controller (C) proteins and DNA operator sites of several type II restriction–modification (RM) systems (15,22). C proteins are dimeric proteins that together with their adjacent DNA operators and overlapping promoters form molecular switches that orchestrate the delay between DNA methyltransferase gene expression and subsequent restriction endonuclease gene expression following entry of an RM module into a naïve cell (22–25). Once RM systems are established in the cell, C proteins stably maintain appropriate levels of endonuclease and methyltransferase expression throughout DNA replication and cell division. This stability is achieved through complex positive and negative autoregulatory mechanisms that, like the qseC-qseM locus, often involve two adjacent operator sites controlling divergent overlapping promoters. While QseC itself is not associated with RM systems, it was one of the proteins found in a systematic search for C proteins in microbial genomes and was classified as C.PvuII family (motif 4)-like (22). Like C.PvuII, QseC binds dual operator sites between two overlapping divergent promoters, shows strong preference for the left operator sequence (O_L) and binds cooperatively with QseC-O_L to the right operator sequence O_R. Also like several C proteins, QseC is translated from a leaderless mRNA (15,26,27). At least six distinct families of QseC/C-protein-like genes are present adjacent to qseM homologues on ICEs identified throughout the alphaproteobacteria. In addition, genes for several members of the same C-protein families are positioned adjacent to the QS antiactivator gene traM on QS-regulated Agrobacterium and Rhizobium spp. plasmids, indicating they may also regulate QS and transfer induction for these plasmids (15,28).

On ICEMlSym^R7A a second C-protein gene, qseC2, is located directly downstream of qseC (Figure 1). QseC2 resembles the C.EcoRV family (motif 7 (22)) of C proteins. C.EcoRV, like C.PvuII and QseC, binds dual operator sites between divergent overlapping promoters (29,30). Two adjacent putative operator sequences (hereafter, O2_L and O2_R) each containing the sequence ATCC-N₇-GGAT are positioned between the 3' end of qseC and the 5' end of qseC2 (15). There are several other examples of tandemly coded C-protein genes like qseC and qseC2 on ICEs (15,31). Previously, we speculated that they may represent remnants of C-protein gene replacements and indeed, several appear to be pseudogenes (15,32); however, others have suggested tandem C-protein genes may comprise higher-order regulatory switches (22).

It is not clear how the balance between molecular components in the ICEMlSym^R7A QS and transfer regulon is tipped to promote entry into a transfer-primed state through repression of qseM. Furthermore, it is unclear if this switch is transient or persistent through cell division once established. In this work, we report that approximately 2% of colonies formed by wild-type R7A contain isolatable cells that have entered a qseM-repressed state we herein name R7A*. R7A* cells are derepressed for QS, ICE excision and conjugative transfer, similar to the phenotypes observed for R7AΔqseM. Isolated R7A* cells vertically inherited their QS and transfer-primed state under standard culture conditions. Extensive genome sequencing did not identify any mutations likely to explain these phenotypes. Using RNA-seq and a variety of genetic and molecular manipulations, we demonstrate that QseC2-mediated repression of antisense-qseC transcription is the first step required for R7A* differentiation, which, once established, is maintained by QseC2 and QseC acting in concert.

MATERIALS AND METHODS

Microbiological techniques

Bacteria were cultured as described previously (16,17,33,34) and further detail is available in the Supporting Information. Strains and plasmids are listed in Supplementary Table S1. Oligonucleotides used in construction of plasmids are listed in Supplementary Table S2. Growth curves, DNA extraction, plasmid construction, mutagenesis, QPCR, CV026 bioassays, β-galactosidase assays and bacterial conjugation assays (Supplementary Table S3) were carried out as previously described (15,18,35).

Competition assays

Plasmids pFAJ1700 and pFAJ1700:neo were introduced into both R7A and R7A* individually. R7A(pFAJ1700) and R7A*(pFAJ1700:neo) were adjusted for density and combined in equal ratio. The same experiment was also carried out with R7A(pFAJ1700:neo) and R7A*(pFAJ1700). Colony-forming units for each strain were calculated after 24, 48 and 72 h. The entire experiment was repeated three times. Further details are provided in Supporting Information and Supplementary Figure S1.

Phenotypic microarrays

TY cultures of R7A, R7A* and R7AΔqseM were grown at 28°C with shaking to stationary phase and diluted to an OD₆₀₀ of 0.05. Cells were washed sequentially in liquid RDM lacking a carbon source four times in 1 ml, 700 μl, 500 μl and 200 μl volumes. Cells were then resuspended in liquid RDM containing Redox Dye Mix MA (Biolog). One hundred microlitres of cell suspension were added to each well of PM1 and PM2A 96-well ‘carbon sources’ microplates (Biolog) and the plates were then incubated in the dark at 28°C with shaking for 72 h. For the PM1 plates measurements of redox dye absorbance at 540 nm and OD₆₀₀ were taken every eight hours throughout growth on an EnSight Multimode Plate Reader (Perkin Elmer). Example OD₆₀₀ measurements for each strain grown in PM1 plates are presented in Supplementary Figure S2. Measurements for PM2A plates were taken only at the 72 h time point (data not shown). Experiments were repeated twice for each strain. No consistent differences were observed for carbon-source utilisation between any of the strains on either the PM1 or PM2A plates.

Genome sequencing and assembly

Expanded methods for genome sequencing, assembly and whole-genome comparisons can be found in Supporting Information. Briefly, DNA for use in Oxford Nanopore Technologies MinION sequencing (R.9.4.1 flowcells and MinION Mk1B) was extracted to maximize read length as described in Supporting Information. Approximately 2.7 and 5.7 gigabases of sequence was generated for R7A and R7A* respectively. Reads were filtered by quality (>q10) and size (>8 kb), producing ∼190-fold read depth for initial assemblies using Flye (36). Assemblies were polished five times with filtered nanopore reads using Racon (https://github.com/isovic/racon), four times using Medaka (https://github.com/nanoporetech/medaka) and five times using Pilon (37) with Illumina sequencing reads (560- and 252-fold read depth for R7A and R7A*, respectively). Both short and long reads were aligned to penultimate assemblies and compared (Dataset S1) with the previous R7A assembly (Genbank accession CP033366.1). Mapped reads and potential variants were manually inspected and assessed as described in Supporting Information and Dataset S1. One nucleotide variant was ambiguous in both Nanopore and Illumina reads and was further assessed using PCR and Sanger sequencing in both R7A and R7A*, which revealed no difference in sequence. Genbank accessions for the final reference sequences for M. japonicum R7A and R7A* are CP051772.1 and CP051773.1, respectively.

Transcriptome sequencing

Detailed description of RNA sequencing methods and differential expression analyses can be found in Supporting Information. Briefly, RNA was extracted from duplicate early-stationary-phase TY cultures (48 h) for R7A and R7A* (see Supplementary Figure S1A for example growth curves). Libraries were sequenced on an Illumina MiSeq using 75-bp paired-end sequencing as described further in Supporting Information. Reads were mapped to the R7A genome (CP051772.1). A summary of differentially regulated ICEMlSym^R7A regulatory genes is presented in Supplementary Table S4 and genome-wide transcriptome data and statistics can be found in Dataset S2.

DNA-binding assays

Protein purification of 6H-QseC and 6H-QseC2 was carried out as described previously for 6H-QseC (15), with adjustments described in Supporting Information. EMSAs were carried out as previously described (38). DNA containing the O2_L and O2_R operators and scrambled derivatives were synthesized and amplified by PCR using IRDye800-labeled DNA oligonucleotides as primers (Supplementary Table S2). The qseC2 operator sites were amplified by PCR from synthesized DNA and then labeled in a second PCR using IRDye800-labeled primers (Supplementary Table S2). Binding reactions were conducted in a similar manner to those carried out previously (38) and detailed EMSA methods are provided in Supporting Information. Reusable DNA Capture Technology (ReDCaT) approach was used for SPR DNA-binding experiments (39) using the Biacore T200 (GE Healthcare) and Biacore SA sensor chip (GE Healthcare). Oligonucleotides used in generation of the ReDCaT chip and ReDCaT assays are listed in Supplementary Table S2. SPR assays were carried out as previously described (38) and as further detailed in Supporting Information and Supplementary Table S5.

RESULTS

Quorum-sensing high-frequency ICEMlSym^R7A donors arise spontaneously in R7A populations

Following passaging of M. japonicum R7A colonies on solid defined medium (Glucose (G)/RDM), a phenotypic variant of R7A was serendipitously isolated, termed R7A*. Unlike wild-type R7A, R7A* induced violacein production in the AHL biosensor strain Chromobacterium violaceum CV026 (Figure 2A). The transfer frequency of ICEMlSym^R7A from R7A* to the recipient R7ANS, a derivative of R7A cured of ICEMlSym^R7A (17), was approximately 400-fold greater than that observed when wild-type R7A was used as donor (Supplementary Table S3), but around 2–3-fold less than when R7AΔqseM was used. Excised ICEMlSym^R7A was detected in 3–4% of log-phase R7A* populations and 75–87% of those stationary-phase R7A* populations using an established quantitative PCR assay (17). In comparison, ICEMlSym^R7A excision in R7A was <0.01% at log phase and 1–2% at stationary phase, while ICEMlSym^R7A excision in R7AΔqseM was ∼20% in log-phase and ∼100% stationary-phase (Figure 2B). Therefore R7A* cells appear to exist in a state similar to that seen for R7AΔqseM except that they exhibit slightly lower frequencies of ICEMlSym^R7A excision and conjugative transfer.

Figure 2.

Quorum sensing and ICEMlSym^R7A excision phenotypes of R7A* and derivatives. (A) Chromobacterium violaceum CV026 bioassays were used to detect AHLs in broth-culture supernatants. Three 25-cm plates are shown. Supernatants from R7A* exconjugants in which ICEMlSym^R7A was reintroduced into the R7ANS* background are labeled as ‘Donor X R7ANS*’. Vector-only (pPR3 and pFAJ1700), positive (10 μM 3-oxo-C6-HSL) and negative controls (DMSO, R7ANS, TY) are included for comparison. (B) Quantitative PCR assays of ICEMlSym^R7A excision. Strains were grown in liquid TY medium and sampled in log phase (24 h) and late stationary phase (64 h). DNA was extracted and assayed by quantitative PCR. PCR primers amplify products that span the circular ICEMlSym^R7A DNA (attP) and vacant chromosomal insertion sites (attB) and these are presented as the percentage of total chromosomes in the sample, measured by amplification of the chromosomal melR gene (17). Bars indicate the average percentage and standard deviation (error bars) of three replicate experiments. The y-axis is in log-scale. A one-way ANOVA followed by Tukey’s post-hoc test was carried out for comparisons between R7A and every other strain and for comparisons between R7A* and every other strain. Green asterisks indicate the mean is statistically different from R7A* at the same time point (* 0.05, **P < 0.01) and orange asterisks indicate the mean is statistically different from R7A at the same time point (*P < 0.05, **P < 0.01). Vector-only controls are presented in Supplementary Figure S3.

To determine how frequently R7A cells differentiated into R7A*, 500 single colonies isolated from plating a stationary-phase liquid R7A culture on G/RDM were tested for the ability to induce violacein production when streaked adjacent to CV026 (Supplementary Figure S4). Ten individual colonies induced violacein production to varying levels. These colonies were subcultured and single colonies were again tested for AHL production, which revealed they contained mixtures of R7A and R7A*-like cells. Further single-colony purification led to isolation of stable R7A*-like cell lines which induced CV026 in a similar manner to the original R7A* (Figure 2A). Examination of one of these isolates, named R7A*2, revealed it exhibited near-identical excision and transfer frequencies to R7A* (Figure 2B and Supplementary Table S3). The R7A* isolation experiment was repeated twice more and, in each case, ∼2% of single colonies contained mixed subpopulations of cells exhibiting R7A and R7A*-like phenotypes. The consistent appearance of R7A* cells within 2% of single colonies that contained subpopulations of both R7A and R7A*-like cells suggested that the R7A* phenotype was arising stochastically within these colonies following plating.

R7A* cells do not exhibit reduced fitness or readily revert to R7A under laboratory conditions

Despite the differences in ICEMlSym^R7A excision, no differences in growth rate were observed between R7A, R7A* or R7AΔqseM cells in either complex or defined medium. Moreover, examinations of mixed liquid cultures of R7A and R7A* did not reveal any competitive differences in growth (Supplementary Figure S1). Phenotypic microarrays also failed to reveal any differences in carbon-source utilization (see Materials and Methods and Supplementary Figure S2). In previous work, we found overexpression of rdfS is lethal for R7A aside from a small proportion of cells that survive through curing of ICEMlSym^R7A (17); similarly, overexpression of msi172-msi171 (fseA) (16), which activates rdfS expression, results in lethality. Interestingly, attempts to introduce pJRtraR (expressing traR downstream of its native promoter) into R7A* and R7AΔqseM resulted in the formation of small colonies that ceased growth after 5 days. Since introduction of pJRtraR into wild-type R7A induces AHL production, conjugation and ICEMlSym^R7A excision (16,17), this result suggested that R7A* and R7AΔqseM were sensitive to further QS induction through increased expression of traR. Deletion of msi172, which abolishes FseA production, in R7A* restored the ability to introduce pJRtraR, indicating that the growth-inhibitory effects of pJRtraR in R7A* were mediated by FseA. In summary, growth inhibition effects were not manifested in R7A* or R7AΔqseM under the conditions tested; however, fseA-dependent growth inhibition was induced in R7A* and R7AΔqseM cells as a result of additional traR expression from pJRtraR.

Throughout this work we did not observe any reversion of R7A* back to R7A under standard laboratory conditions. We attempted to identify spontaneous revertants of R7A* and R7A*2 using a CV026 screening method similar to that used in the isolation of R7A*2, where we screened 500 colonies for loss of AHL production. While single colonies that exhibited both reduced AHL production and conjugative transfer rates were identified (Supplementary Figure S5 and Table S3), all returned to an R7A* state following passaging, suggesting factors indirectly related to activation of QS predisposed these derivatives to re-establish R7A* phenotypes (see Supporting Information for further details).

The R7A* phenotype is unlikely to be caused by mutation and is reset in recipients of ICEMlSym^R7A

It seemed possible that structural rearrangements or mutations in the R7A* genome might be responsible for the R7A* phenotype. Reference-quality genome sequences were assembled using both long and short-read sequencing of R7A and R7A* DNA (Genbank accessions CP051772 and CP051773). Both assemblies produced 6530403-bp genomes and whole-genome alignments did not reveal any structural differences. The R7A and R7A* sequences differed by three single-nucleotide substitutions in chromosomal regions with no obvious link to transfer or QS. No differences were observed between the ICEMlSym^R7A sequences. R7A*2 DNA was sequenced (CP052769.1) using short-read sequencing and when aligned to the R7A sequence, two synonymous coding-sequence changes were identified, along with a 7-bp deletion in one of five direct repeats in an intergenic region. All three variations in R7A*2 were at distinct loci from those observed in R7A* and again no differences were present within ICEMlSym^R7A (see Supporting Information and Dataset S1 for detailed descriptions of genome assemblies and variant calling).

We next investigated if the R7A* state was maintained following conjugative transfer of ICEMlSym^R7A. Eight exconjugants derived from each of three matings using R7A, R7A* or R7AΔqseM as donors were screened for AHL production. None of the exconjugants receiving ICEMlSym^R7A from R7A or R7A* induced CV026, whereas exconjugants receiving ICEMlSym^R7A from R7AΔqseM induced CV026 as expected (Supplementary Figure S6). Therefore, the R7A* state was not transferred with ICEMlSym^R7A to R7ANS. It seemed possible that the chromosomal background of the R7ANS recipients might somehow suppress the R7A* phenotype following transfer. To discount this possibility, ICEMlSym^R7A was cured from R7A* using plasmid pJR204 that overexpresses RdfS, as described previously (17). The resulting strain R7ANS* was sequenced (CP052770.1) and alignment with the R7A* sequence confirmed loss of ICEMlSym^R7A and identified a single synonymous nucleotide change compared to R7A* (again, this nucleotide change was distinct from others identified, Supporting Information and Dataset S1). R7ANS* was then used as a recipient in conjugation experiments using R7A, R7A* and R7AΔqseM as donors. Conjugation frequencies for all matings were near-identical to those carried out using R7ANS as the recipient (Supplementary Table S3). Eight R7ANS*-derived exconjugants for each mating were examined for AHL production using CV026. Again, none of the exconjugants receiving ICEMlSym^R7A from R7A or R7A* induced CV026 while all those receiving ICEMlSym^R7A from R7AΔqseM did (Supplementary Figure S6). Therefore, the R7A* phenotypes were not transferred with ICEMlSym^R7A, even when ICEMlSym^R7A was transferred into the isogenic R7ANS* chromosomal background obtained by curing ICEMlSym^R7A from R7A*. In summary, these experiments indicate an epigenetic factor was responsible for the R7A* phenotype.

Transcription of the antiactivator gene qseM is repressed in R7A*

Given the similarities between R7A* cells and R7AΔqseM, we suspected that qseM expression might not be expressed in R7A* cells. pNqseM, which constitutively expresses qseM from the nptII promoter, was introduced into R7A* and R7A*2. pNqseM abolished AHL production (Figure 2A) and reduced excision to levels below that of R7A (Figure 2B and Supplementary Figure S4). Deletion of traR in R7A* and R7AΔqseM had a similar effect (Figure 2), consistent with QS being essential for full activation of ICEMlSym^R7A excision in stationary-phase cells. Interestingly, while R7A*ΔtraR exhibited a reduced transfer frequency (Supplementary Table S3), it remained around 10-fold higher than that of R7AΔtraR for which transfer was barely perceptible. This suggested that even without traR, R7A* cells remain partially upregulated for transfer. This is consistent with qseM expression being repressed in R7A* and R7A*ΔtraR, since QseM also independently binds and inhibits FseA (Figure 1A) (18).

RNA-seq experiments were carried out to compare RNA transcript abundance in R7A and R7A*. Abundance of traI1, traI2, rdfS-msi107 and RNA encoding conjugation pore proteins was increased in R7A* (Supplementary Table S4 and Dataset S2), consistent with the increased rate of excision and conjugative transfer observed (Figure 2B and Supplementary Table S3). Transcripts mapping to qseM were vastly reduced in R7A* compared to R7A (Figure 3). Notably, some of the other pronounced differences observed were in sequencing reads that mapped in the antisense direction to the msi172-msi171 (fseA) and qseC regions. The positive-sense msi172-msi171 (fseA) transcript initiating from the traI2 promoter was marginally increased in R7A*; however, antisense msi172-msi171 transcripts extending from the qseM promoter were 11-fold reduced in R7A* compared to R7A. Similarly, strong antisense transcription of qseC observed in R7A was almost completely absent from R7A* (Figure 3 and Supplementary Table S4). Antisense-qseC transcription initiated from a previously unrecognized promoter region herein named PasqseC, which was positioned downstream of qseC and upstream of the putative qseC2-associated operator sites O2_L and O2_R (Figure 4A). Both qseC2 and qseC transcripts were present at higher levels in R7A* but only the 1.7-fold increase for qseC2 R7A* was statistically significant (Figure 3 and Supplementary Table S4).

Figure 3.

Transcriptome sequencing of R7A and R7A* in the traR-qseC2 region. Gene maps of the traR-qseC2 region are shown between coverage maps (y-axis represents moving 20-bp average of reads per nucleotide) for strand-specific RNA-seq reads from (A) R7A and (B) R7A* cells, together with a schematic highlighting proteins involved in activation (blue) and repression (red) of quorum sensing and ICEMlSym^R7A excision and transfer. Genes are indicated as gray blocks and RNA-seq read-depth averages from two experiments are shaded on the y-axis in blue for the forward strand and orange for the reverse strand. Positions of QseC and QseC2 operator sites are indicated as vertical rectangles. Numbers underneath genes in (A) represent average mRNA abundance fold-change relative to R7A* and numbers above genes in (B) represent abundance fold-change relative to R7A.

Figure 4.

QseC2 binds O2_L and O2_R and represses transcription from PqseC2 and PasqseC. (A) A gene map of the qseM-qseC2 region (to scale). DNA operator sequences are shown below the map. The promoters PqseM and PqseC were mapped previously (15). Approximate transcription start positions for asqseC and qseC2 were manually estimated from transcriptome alignments (gray highlight). The positions of the constructed ΔqseC2 and ΔPasqseCΔqseC2 deletions are indicated with dashed lines. (B) Electrophoretic Mobility Shift Assays (EMSAs) using purified 6H-QseC2 with fluorescently labeled dsDNA oligonucleotides containing the O2_L-O2_R sequence. The top EMSA labeled O2_L-O2_R was carried out using dsDNA containing the wild-type O2_L-O2_R sequence; the O2_L EMSA used DNA carrying a scrambled O2_R sequence; the O2_R EMSA used DNA carrying a scrambled O2_L sequence and both operators were scrambled in DNA for the RDM2 EMSA. All assays were carried out using a final concentration of 5 nM labeled dsDNA. (C) SPR responses for purified 6H-QseC2 (i) and 6H-QseC (ii) with their cognate operator sequences and mutated derivatives as described in (B). R_max is the theoretical maximum binding response assuming 6H-QseC and 6H-QseC2 each bind as dimers to an individual operator (i.e. dimers bound to both operators would have a theoretical R_max of 200%). Proteins were added at either 0.1 or 1 μM concentration as indicated. (D) Partial gene map (not to scale) of pSDz containing cloned PasqseC or PqseC2 upstream of lacZ, and qseC2 (when present) under control of Plac, which exhibits leaky expression in the absence of IPTG (dashed arrow) or elevated expression in the presence of IPTG. (E) β-Galactosidase activity of stationary-phase R7ANS cultures containing pSDz plasmid constructs with PqseC2 (i) or PasqseC (ii) oriented to drive transcription of the lacZ gene. Modified versions of these constructs harboring qseC2 under control of the lac promoter were also assayed, with or without 1 mM IPTG as indicated. β-Galactosidase activity was measured using the fluorescent substrate 4-methylumbelliferyl-β-D-galactopyranoside (40). Bars represent the mean of four biological replicates and error bars represent standard error of the mean. Statistical significance values from Student's t-test are represented by asterisks (< 0.05 * or < 0.01 **). The mean expression and standard error values for pSDz lacking any cloned promoter (indicated approximately by a dashed line in both (i) and (ii)) were 3,374 ± 127 and 3,542 ± 1,128 with IPTG.

QseC2 and PasqseC control entry into the R7A* state

The differences observed between R7A and R7A* in the RNA-seq data for the PasqseC-qseC2 region suggested these genes and transcripts had a role in the R7A* phenotypes. To investigate the role of qseC2, qseC2 deletions were constructed in both R7A and R7A* (Figure 4A). In the R7A* background qseC2 deletion abolished AHL production (Figure 2A) and reduced ICEMlSym^R7A excision (Figure 2B) and conjugative transfer frequencies (Supplementary Table S3) to levels equivalent to those observed for R7A. Introduction of plasmid pNqseC2, expressing qseC2 from the strong constitutive nptII promoter, activated AHL production, excision and transfer in both R7A and R7AΔqseC2 strains to levels phenotypically indistinguishable from R7A* (Figure 2, Supplementary Figure S3 and Table S3). To examine the role of PasqseC, a deletion beginning downstream of qseC and extending through to the end of qseC2 (ΔPasqseCΔqseC2), was constructed in R7A and R7A* (Figure 4A). In the R7A background this deletion stimulated excision and conjugative transfer to levels similar to those of R7A* (R7AΔPasqseCΔqseC2 and R7A*ΔPasqseCΔqseC2, Figure 2 and Supplementary Table S3). This result suggested PasqseC was indeed responsible for preventing R7A cells from entering the R7A* state, possibly through RNA-mediated repression or interference with qseC transcription or translation.

To test if anti-qseC RNA was able to repress the R7A* phenotype in trans, we cloned a region spanning from the likely PasqseC transcriptional start site (Figure 4A) through to the start codon of qseC, downstream of the strong nptII promoter in pPR3G. Introduction of this plasmid into the PasqseC-deletion strains described above did not repress AHL production (data not shown), suggesting that the mechanism of antisense regulation might only function in cis (41). To test this, we constructed mutagenesis vectors to replace PasqseC-qseC2 in the R7A/R7A* chromosome with the nptII promoter. Two mutagenesis vectors were constructed, one in which PnptII was orientated to drive expression anti-sense to qseC analogous to PasqseC and another to drive expression away from qseC. Each mutagenesis plasmid was used to replace the PasqseC-qseC2 region in both R7A and R7A*. R7A* carrying the anti-qseC nptII promoter ceased producing AHLs, confirming that transcription antisense to qseC converted R7A* cells back to an R7A-like state. Conversely, R7A carrying the nptII promoter in the forward orientation became activated for AHL production, while there was no change in AHL production observed for R7A* (Supplementary Figure S7).

Introduction of pNqseC2 into the ΔPasqseCΔqseC2-deletion mutants had no discernible effect on their already-elevated levels of AHL production or excision (Figure 2). Likewise, no changes in AHL-production phenotypes were observed following introduction of pNqseC2 into any of the constructed strains carrying the nptII promoter (Supplementary Figure S7). Taken together, these results suggested PasqseC expression determined the R7A* AHL-production phenotype and that QseC2 controlled PasqseC expression. We further hypothesized that elevated levels of QseC2 in R7A* cells were responsible for the repression of PasqseC and derepression of qseC transcription and/or translation and that this in turn led to the QseC-mediated repression of qseM, the activation of TraR and FseA and, ultimately, TraR-mediated activation of QS and FseA-mediated activation of ICEMlSym^R7A excision_.

QseC2 binds two operator sites and represses PasqseC and PqseC2

qseC2 is located downstream of a pair of inverted repeat sequences O2_L and O2_R that seemed likely QseC2 binding sites. We purified hexahistidine-tagged QseC2 (6H-QseC2) and used this in EMSAs with a 106-bp DNA region spanning from the predicted transcriptional start site of PasqseC through to the start of the qseC2 gene, containing the putative QseC2 operators O2_L and O2_R (Figure 4B). DNA containing wild-type operators exhibited two distinct mobility shifts in EMSAs with increasing concentrations of 6H-QseC2, consistent with 6H-QseC2 binding one site and then a second site in a concentration-dependent manner. EMSAs performed using DNA with scrambled O2_L or O2_R sequences only exhibited single mobility shifts and DNA containing completely scrambled sequence was not shifted. QseC2 shifted DNA containing individual O2_L or O2_R operators with similar affinity, but with a slight preference for O2_L. QseC2 exhibited comparatively weaker affinity for its operator sites than QseC does with O_L and O_R (15). In previous work, a 17-fold excess of 6H-QseC (monomeric) shifted both O_L and O_R (15), whereas here a 64-fold molar excess of 6H-QseC2 was needed to observe a full shift of both O2_L and O2_R. There was also less evidence for QseC2 cooperativity in the binding of O2_L and O2_R than for QseC with its binding sites. In our previous EMSAs (15), 6H-QseC was unable to fully shift the O_R site when present in isolation even when present in ∼220-fold molar excess, whereas 6H-QseC2 here fully shifted O2_L and O2_R at comparable concentrations.

Surface plasmon resonance (SPR)-based DNA-binding assays (42) were also used to compare binding of each protein with each of the identified operator sites (Figure 4C and Supplementary Table S5). At concentrations of 1 μM, 6H-QseC2 induced a SPR response 183–210% of the theoretical maximum (R_max) expected for binding of a single DNA site, as would be expected for 6H-QseC2 dimers simultaneously binding both O2_L and O2_R. 6H-QseC2 exhibited a slightly stronger response for the left operator sequence O2_L (115%) than to O2_R (80–86%) (Figure 4C). Purified 6H-QseC produced 165% of R_max when assayed with wild-type O_L and O_R together. Consistent with our previous work, 6H-QseC produced a much stronger response to O_L (106% R_max) than to O_R (34% R_max) (Figure 4C). Using a lower concentration of 0.1 μM, 6H-QseC produced 85% R_max for O_L and O_R, 91% for O_L alone, but only 13% for O_R alone. In contrast, 6H-QseC2 produced <14% R_max for any of the oligonucleotides tested when used at the lower 0.1 μM concentration, consistent with 6H-QseC2 exhibiting a weaker binding affinity for its operator sites. Neither 6H-QseC nor 6H-QseC2 induced a SPR response with the other protein’s cognate operator sequences, confirming regulatory independence of QseC2 and QseC with respect to DNA binding. In summary, like QseC, QseC2 binds two adjacent operator sequences between divergent overlapping promoters; however, in contrast with QseC, QseC2 exhibits overall weaker affinity for its operator sites and only weakly favors the O2_L site.

To test if QseC2 regulated PasqseC and/or PqseC2, we used the pSDz vector, which carries two cloning cassettes. One cassette is adjacent to a promoterless lacZ gene, while a second physically separated and divergently orientated cassette is positioned downstream of an IPTG-inducible lac promoter, which itself is located downstream of a lacI^q gene (Figure 4D) (18). We previously noted that the lac promoter on pSDz is expressed at a low level in TY medium in the absence of IPTG, and expression is further induced with IPTG (35). We exploited this inducibility to examine how varying qseC2 expression levels might impact expression from PasqseC and PqseC2 through concentration-dependent occupancy of O2_L and O2_R. A fragment containing PasqseC and PqseC2 was cloned in both orientations upstream of the lacZ gene in pSDz. Constructs additionally carrying the qseC2 gene downstream of the lac promoter were derived from the two reporter plasmids and all plasmids were introduced in R7ANS to avoid interference from ICEMlSym^R7A-derived QseC2. β-Galactosidase assays performed using the PqseC2 promoter construct (Figure 4E(i)) revealed that, unlike PqseC which requires activation by QseC (15), PqseC2 was strongly expressed in the absence of qseC2. Assays with PqseC2 additionally carrying qseC2 revealed PqseC2 expression was slightly reduced by uninduced/leaky qseC2 expression but was repressed ∼9-fold in the presence of IPTG-induced qseC2 expression. Assays of the PasqseC reporter fusion (Figure 4E(ii)) revealed it was expressed at a similar level to PqseC2 in the absence of qseC2, but was repressed to the level of the empty pSDz control with either uninduced or IPTG-induced qseC2 expression. Thus, while a low level of QseC2 was sufficient to repress PasqseC expression, IPTG induction of qseC2 and thus a higher level of QseC2 was required to repress transcription from PqseC2. These observations are consistent with a model in which QseC2 is a transcriptional repressor of PasqseC through binding O2_L but only represses its own expression from PqseC2 when higher concentrations of QseC2 enable occupancy of both O2_L and O2_R.

Transient overexpression of QseC2 triggers epigenetic maintenance of the R7A* state

Since the R7A* state appears to be epigenetically maintained, QseC2 and/or QseC concentrations must remain at an elevated level once the R7A* state is established. To test this hypothesis directly, pNqseC2 was modified to carry the sacB gene to enable curing from cells by growing on sucrose medium (S/RDM). pNqseC2-sacB and the control plasmid pPR3G-sacB were each introduced into R7A via conjugation in five separate mating experiments. As expected, AHL production was activated in each of the R7A strains carrying pNqseC2-sacB but not in strains carrying plasmid pPR3G-sacB (Supplementary Figure S8A). Stationary-phase broth cultures were inoculated from the five R7A/pNqseC2-sacB strains and grown without selection. Dilutions were spread on sucrose medium to select for loss of pNqseC2-sacB or pPR3G-sacB. Individual colonies isolated from these plates were further single-colony-purified on S/RDM and once more on G/RDM. Each strain was also checked for lack of growth on Gm-containing medium to confirm loss of the plasmids. Passaged single colonies from the G/RDM plates were then used to inoculate TY broths. CV026 bioassays of the supernatants of these broths revealed that each of the strains previously carrying pNqseC2-sacB remained activated for AHL production (Supplementary Figure S8A) and QPCR confirmed these strains exhibited excision frequencies indistinguishable from R7A* (Supplementary Figure S8B). In summary, through introduction of pNqseC2-sacB, R7A cells were artificially induced to enter the R7A* state by qseC2 overexpression from a plasmid and they remained in this state following curing of this plasmid.

Both QseC and QseC2 are required for maintenance of the R7A* state

Since the transient introduction of pNqseC2-sacB stimulated cells to enter the R7A* state, this suggested chromosomally-encoded QseC and/or QseC2 maintained the R7A* state through autoregulation following the curing of pNqseC2-sacB. To confirm if QseC2 was required for R7A* maintenance, we introduced pNqseC2-sacB into R7AΔqseC2 (in 5 separate mating experiments) and then cured the plasmid as described above. pNqseC2-sacB induced AHL production in R7AΔqseC2 as expected; however, R7AΔqseC2 ceased producing AHLs following curing of pNqseC2-sacB, confirming that the chromosomal qseC2 gene was essential for R7A* maintenance (Supplementary Figure S8A).

It remained unclear if the QseC protein was critically involved in maintaining the R7A* state as it seemed possible QseC2 alone might stimulate and maintain the phenotypes by directly repressing read-through transcription of qseM from the PasqseC promoter. To abolish translation of QseC without disrupting surrounding regulatory elements, we introduced an amber mutation into the qseC gene. A 3-bp change was introduced to convert the 10th codon to TAG; this change also removed a downstream ATG codon and introduced an overlapping XbaI restriction-endonuclease site for mutant screening. This mutation was constructed in R7A and R7A*, producing strains R7AqseC(Am) and R7A*qseC(Am). AHL production was abolished in R7A*qseC(Am). Introduction of pNqseC2-sacB did not induce AHL production in R7AqseC(Am) or R7A*qseC(Am) (Supplementary Figure S9). Translation of qseC was therefore essential for both R7A* establishment and maintenance. In summary, QseC and QseC2 together, coupled through anti-qseC transcription from PasqseC, form the epigenetic switch controlling entry into the transfer-primed R7A* state.

DISCUSSION

In this work we showed that a subpopulation of R7A cells can transition into a stable state we termed R7A*, which exhibits upregulated AHL production, ICEMlSym^R7A excision and horizontal transfer. RNA sequencing of R7A* cells revealed they were nearly abolished for qseM mRNA, explaining the derepression of phenotypes normally inhibited through QseM interactions with TraR and FseA. The R7A* state appeared spontaneously in cells within ∼2% of colonies derived from stationary-phase cultures. Extensive genome sequencing failed to identify structural or genetic changes that could likely explain the R7A* phenotypes. R7A* cells heritably maintained their phenotypes under standard laboratory culture conditions. The R7A* state was not transferred along with ICEMlSym^R7A from R7A* to the non-symbiotic isogenic recipient R7ANS*, indicating that inheritance of the R7A* state depended on factors present in the donor cell that were not transferred with DNA to the recipient during conjugation and that the R7A* state did not involve a heritable factor associated with the R7A* chromosome. In summary, R7A* cells are epigenetically maintained in a state primed for QS and horizontal transfer.

In previous work we proposed that a molecular switch comprising QseC and its operator sequences facilitated bimodal repression of qseM and that individual cells in R7A populations were either in an ‘on-state’ or ‘off-state’ for QS, ICEMlSym^R7A excision and conjugative transfer (Figure 1) (15). This was based on the findings that the -35 regions of PqseC and PqseM are located within O_R, qseM RNA is more abundant in a qseC mutant and expression from PqseC is activated by QseC (15). Therefore, in conditions where QseC is absent or at low concentration, e.g. following entry of ICEMlSym^R7A into a new host, expression of qseC is weak while qseM expression is strong, thus keeping cells repressed for QS and excision. Once QseC concentrations reach some critical threshold, QseC binds to O_L, further activating its own expression. QseC also represses the PqseM promoter, either by binding O_L alone, or as previously speculated through cooperative binding of O_L and O_R (Figure 5). This results in the derepression of TraR and FseA and the subsequent activation of QS and horizontal transfer. In this work we discovered that QseC-dependent repression of PqseM is itself inhibited by transcription from the PasqseC promoter, which produces RNA antisense to qseC. Deletion of the PasqseC promoter stimulated cells to enter an R7A*-like state. We also discovered that a second adjacently encoded C-protein, QseC2, bound a distinct set of operator sites and repressed transcription from PasqseC. Overexpression of QseC2 also stimulated cells to enter an R7A*-like state. In summary, QseC2 represses PasqseC, which likely enables QseC concentrations to increase and repress PqseM. The resultant absence of QseM allows TraR to activate FseA, which in turn activates PrdfS (Figure 5).

Figure 5.

Model of the molecular steps leading to the R7A* state. Abridged gene maps (not to scale) illustrate regulatory steps involved in transition into the R7A* state. In the absence of any of the protein regulators (for example in a recipient upon receiving ICEMlSym^R7A immediately post-transfer), PqseM, PasqseC and PqseC2 promoters are active. Once QseC2 concentrations increase above a certain threshold, QseC2 binds O2_L and represses PasqseC. Repression of PasqseC derepresses the translation of QseC through an unknown antisense mechanism. QseC activates expression from its own promoter through binding O_L. QseC also represses PqseM and, based on the position of the PqseM -35 region, this repression may involve QseC-dimer occupancy of both O_L and O_R (indicated by a question mark) (15). Once QseM concentrations decrease, TraR is free to activate QS through activation of the PtraI1. TraR activation of PtraI2 leads to expression of FseA, which in turn leads to expression of PrdfS.

Introduction of a stop codon into qseC prevented establishment of the R7A* state, confirming the R7A* state depends on active QseC protein. Deletion of the qseC-antisense promoter PasqseC caused R7A cells to enter an R7A*-like state, suggesting PasqseC somehow reduces QseC levels and prevents QseC-mediated repression of PqseM. Several mechanisms of antisense-RNA regulation have been characterized in prokaryotes (41). One possibility for regulation by PasqseC is that transcription from this promoter results in antisense RNA that anneals with qseC transcripts and the resultant dsRNA is targeted for RNase degradation. However, plasmid-based overexpression of antisense-qseC RNA from the nptII promoter did not complement a strain carrying a PasqseC deletion. This could suggest transcription from PasqseC is only effective in repressing QseC levels when it occurs in cis with qseC (43). Supporting this hypothesis, replacement of PasqseC with the nptII promoter to drive antisense-qseC transcription switched off AHL production in R7A*. The PasqseC promoter could cause a reduction in qseC mRNA through cis-acting transcriptional interference, whereby RNA polymerases on opposite DNA strands occlude or terminate each other's transcription initiation or elongation (44). However, the RNA-seq data did not indicate any reduction in qseC mRNA abundance to support either a dsRNA degradation or a cis-acting transcriptional interference mechanism. The abundance of full-length qseC mRNA in R7A cells was not significantly different from that observed in R7A* and there was no obvious evidence for degradation of qseC transcripts in R7A compared to R7A* (Figure 3B). In prokaryotes, translation inhibition via antisense RNA can occur through antisense binding to the sense mRNA in such a way that the Shine-Dalgarno sequence is occluded from the ribosome. Again, this mechanism seems unlikely since qseC mRNA is a leaderless transcript lacking a canonical ribosome-binding site. As far as we are aware, antisense regulation of leaderless mRNA transcripts remains an unexamined phenomenon. While we were unable to resolve the exact mechanism of PasqseC-mediated repression of QseC, the observations nevertheless provide the first insights into how tandemly-coded C-protein genes can be genetically combined to form higher-order switches. Tandemly-coded C-protein genes with distinct operator sequences are present on other ICEs (15,22,32), suggesting that the genetic and molecular coupling of tandemly coded pairs of C-protein genes could be a common evolutionary solution for the assembly of highly stable genetic switches.

Overexpression of qseC2 from plasmid pNqseC2-sacB induced R7A cells to enter the R7A* state and cells remained in the R7A* state even after curing of pNqseC2-sacB. The phenotypes were not maintained in strains carrying mutations in either qseC or qseC2, confirming both proteins are required to epigenetically maintain the R7A* state. This implies that once the state is established, concentrations of QseC and QseC2 proteins remain elevated compared to those in R7A cells. For QseC2, the ∼1.7-fold increase in qseC2 mRNA in R7A* compared to R7A may be enough to account for this (Figure 3). As demonstrated in our promoter fusion assays using IPTG-inducible qseC2, even leaky uninduced expression of qseC2 from the lac promoter strongly repressed transcription from PasqseC. QseC2 only strongly repressed its own promoter following IPTG induction. The elevated levels of qseC2 mRNA in R7A* suggest in most R7A* cells QseC2 levels are not high enough to trigger such negative autoregulation or, more likely, any negative autoregulation by QseC2 is short-lived due to the resulting decline in QseC2 concentration and derepression of PqseC2. The negative autoregulation observed for QseC2 likely serves to maintain a moderately-elevated equilibrium concentration of QseC2 in R7A* that is high enough to strongly repress PasqseC but not PqseC2. While the levels of qseC sense mRNA were not significantly different in R7A* compared to R7A, antisense-qseC RNA was almost completely absent from R7A. As discussed above, repression of PasqseC derepresses QseC-mediated repression of PqseM, strongly suggesting PasqseC represses QseC translation or activity either directly or indirectly. It remains unclear what triggers the initial increases in QseC2 and/or QseC during entry into the R7A* state but it is plausible this is a stochastic process resulting from uneven initial concentrations of QseC and/or QseC2 following cell division. Future experiments examining the cellular concentrations of QseC and QseC2 in R7A and R7A* may resolve these questions.

While not interrogated here, our RNA-seq experiments also revealed a global downregulation of ribosomal protein genes in R7A* with a 4- to 14-fold reduction in mRNA abundance for 46 ribosomal-subunit genes (Dataset S2). Downregulation of ribosome synthesis is commonly observed in cells experiencing nutritional deficiencies or other forms of stress. In diverse bacteria repression of ribosome synthesis can be stimulated by an increase in concentration of the alarmone molecule ppGpp, a phenomenon termed the stringent response. ppGpp together with the DksA protein binds RNA polymerase and decreases transcription from rRNA promoters, while activating pathways in amino-acid biosynthesis (45,46). The transcriptome of R7A* was not entirely consistent with the stringent response as several of the most upregulated genes were involved in arginine catabolism (including 8-fold increases in arginase and ornithine deaminase genes, R7A2020_27410 and R7A2020_27415, Dataset S2). Regardless, changes in ribosomal protein abundance might provide a mechanism to trigger and/or maintain the R7A* state. Various stress conditions can alter ribosome composition and selectively enhance translation of leaderless mRNAs (47,48). Increased leaderless mRNA translation is at least partly stimulated through downregulation of the redox-regulated ATPase YchF, which normally represses translation from leaderless mRNA through interactions with translation initiation-factor-3 (48). A 3.5-fold reduction in mRNA abundance was observed for the ychF gene in R7A* (R7A2020_10900, Dataset S2), suggesting translation from leaderless mRNAs like qseC (and possibly qseC2) may be enhanced in R7A* cells.

The observation that R7A* exhibits slightly lower excision frequencies and a ∼2- to 3-fold lower conjugation frequency than R7AΔqseM could indicate the presence of a subpopulation of cells in the R7A* population that have reverted to a QS and transfer ‘off’ state like most R7A cells. Several attempts were made to isolate cells from R7A* populations that had reverted to a state resembling R7A; however, it appears that the reversion rate is at least lower than the R7A > R7A* conversion rate (Supporting Information). In our attempts to isolate R7A* revertants, we isolated several colonies that had seemingly lost the ability to make AHL, but they all regained the ability to make AHL following further subculture. We suspect these R7A-like cells derived from R7A* had likely inherited lower QseC2/QseC protein concentrations than most R7A* cells and therefore exhibited higher expression of qseM (repressing QS) but that the levels of QseC2/QseC were still above the concentration required to re-establish the R7A* state through repression of PqsqseC and positive autoregulation of qseC expression. We only observed a complete reset of the R7A* switch state in recipients of ICEMlSym^R7A following conjugative transfer. By analogy with RM systems, conversion to the R7A* state resembles the delayed activation of restriction endonuclease expression following entry of an RM system into a naïve cell, where neither the C protein nor the endonuclease are expressed highly until such time as C-protein concentrations increase above a certain threshold. However, once the RM system becomes established in the cell, C-protein concentrations do not revert back to a naïve state until the RM system is transferred to a naïve host. Like establishment of an RM system, it seems possible that induction of the R7A* state in donor populations may be an epigenetically permanent switch.

The generation of two distinct phenotypic states within a genetically identical population, often termed bistability, effectively enables bacterial populations to ‘hedge their bets’ (49) on future natural selection events. Phenotypes regulated by bistability include induction of bacterial sporulation (50), DNA competence (51), antimicrobial-resistant persister cell formation (52), motility (53) and transfer of other ICEs (54). We and others have proposed bistable regulation of ICE transfer (and likely transfer of other mobile elements) allows ICEs to benefit from stable vertical descent with minimal impact to their current host organism, whilst simultaneously partitioning a small proportion of cells from each generation to become primed for ICE transfer and bear any associated impacts on host fitness (15,54,55). Regulation of transfer of ICEclc from Pseudomonas knackmussii B13 seems to illustrate this model well. ICEclc, induces a proportion of the P. knackmussii B13 population to differentiate into cells derepressed for transfer functions (56,57). These so-called ‘transfer-competent’ cells exhibit severely reduced growth and frequent lysis, highlighting the host fitness costs associated with ICE transfer. While our RNA-seq experiments here revealed a global downregulation of ribosomal protein genes and suggested some disruptions to metabolism (Dataset S2), R7A* did not exhibit any obvious fitness disadvantage (Supplementary Figure S1) or display any overt phenotypes suggestive of cells suffering stress. However, the observation that introduction of pJRtraR into R7A* inhibited cell growth in an FseA-dependent manner suggests there may indeed be host fitness impacts of ICEMlSym^R7A transfer but that the stoichiometry of TraR and FseA in transfer-primed R7A* cells is balanced finely enough to avoid these impacts under the conditions tested. It is possible that while R7A* cells are primed for transfer, negative impacts on host fitness may only surface once horizontal transfer has commenced. From this perspective, the study of R7A and R7A* cells represents a unique platform from which to precisely examine potential fitness costs associated with horizontal transfer in isogenic cell populations and expose the intricate molecular mechanisms by which mobile genetic elements mitigate their impact on the host.

DATA AVAILABILITY

All genome assemblies and sequencing read files are available from NCBI BioProject accession PRJNA627051. RNA-seq data are also available through the Gene Expression Omnibus (GEO) series accession GSE189468. Comparisons of R7A and R7A* genome assemblies are presented in Dataset S1 and complete RNA-seq data are presented in Dataset S2.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

FUNDING

Royal Society of New Zealand Marsden Fund [16-UOO-207 to C.W.R., J.P.R., C.S.B.]; Australian Research Council [FT170100235 to J.P.R]; Natural Sciences and Engineering Research Council (NSERC) (to M.F.H., C.K.Y.). Funding for open access charge: Australian Research Council [FT170100235].

Conflict of interest statement. None declared.