Horizontal gene transfer (HGT) between bacterial lineages is a fundamental evolutionary process that accelerates adaptation. Sequence analyses show that conjugative plasmids are principal agents of HGT in natural communities. However, we lack understanding of how the ecology of bacterial communities and their environments affect the dynamics of plasmid-mediated gene mobilization and transfer. Here we show, in simple experimental soil bacterial communities containing a conjugative mercury resistance plasmid, the repeated, independent mobilization of transposon-borne genes from chromosome to plasmid, plasmid to chromosome and, in the absence of mercury selection, interspecific gene transfers from the chromosome of one species to the other via the plasmid. By reducing conjugation, positive selection for plasmid-encoded traits, like mercury resistance, can consequently inhibit HGT. Our results suggest that interspecific plasmid-mediated gene mobilization is most likely to occur in environments where plasmids are infectious, parasitic elements rather than those where plasmids are positively selected, beneficial elements.
Conjugative plasmids—semi-autonomous mobile genetic elements that transfer between bacteria—are key agents of horizontal gene transfer (HGT)1,2, facilitating rapid bacterial adaptation by spreading ecologically important traits between lineages3. The physical movement or duplication of genes (gene mobilization) between chromosomes and plasmids, and their subsequent transfer between hosts, can be decisive in microbial evolution, and has facilitated acquisition of antimicrobial resistance4,5 and emergence of virulent pathogens6,7. Such mobilization can be facilitated by transposable elements (TEs), which encode enzymes (transposases) allowing transfer of genes between replicons2. However, the effects of ecological factors and natural selection on gene mobilization and subsequent HGT are unclear. This is particularly the case for natural environments with a high degree of spatial structure8, which is rarely captured by experimental studies. An outstanding question is how positive selection for plasmid-encoded traits, such as resistance genes, affects the ability of that plasmid to spread genes through a community. Although positive selection can favour HGT by benefitting bacteria that have acquired the plasmid (transconjugants), it can also prevent HGT by killing or inhibiting growth of potential plasmid recipients9,10.
Experimental evolution of bacterial communities is a powerful tool for exploring the evolutionary dynamics of plasmid-mediated HGT, bridging the gap between simplified short-term laboratory studies11,12 and comparative genomics of natural populations4,5,13,14. To investigate how positive selection for plasmid-borne genes and the presence of an alternative host species interact to determine plasmid dynamics, we established communities of the common soil bacteria Pseudomonas fluorescens and P. putida, either alone or in co-culture, in sterilized soil microcosms, which offer a complex, spatially structured and experimentally tractable environment15,16,17. Communities began with the 307 kilobase (kb) conjugative plasmid pQBR57 at ~50% frequency; this plasmid was isolated from agricultural soil and carries a 7-kb mercury resistance (HgR) TE, Tn5042. We also established control communities without pQBR57. Microcosms were supplemented with mercuric chloride to 16 µg g−1 Hg(ii), or an equivalent volume of water (0 µg g−1 Hg(ii)). This level of mercury contamination, similar to that found in industrial or post-industrial sites18, selects for specific HgR but does not necessitate it16. Six replicate populations for each combination of treatments were grown for ~440 generations. Analysis of plasmid frequency dynamics showed that pQBR57 was generally maintained by P. fluorescens and lost by P. putida, but persisted in P. putida when co-cultured with P. fluorescens, due to interspecific ‘source–sink’ plasmid transfer17. To investigate consequent effects on gene mobilization and transfer, we sequenced clones from the beginning and the end of the experiment, and used the Bacterial and Archaeal Genome Analyser (BAGA) pipeline to identify structural variations19.
We detected multiple, independent gene mobilization events between plasmid and chromosome in each species. Strikingly and unexpectedly, we also identified numerous interspecific transfers of chromosomal genes in the co-cultured treatments, facilitated by pQBR57 (Fig. 1). Gene transfer from P. fluorescens to P. putida was exemplified by a previously unannotated P. fluorescens TE, Tn6291, in P. putida plasmids from 2/6 co-cultured communities (Fig. 1, replicates b and f). Subsequent polymerase chain reaction (PCR) analyses found Tn6291 in P. putida clones from two further co-cultured communities (replicates d and e). Tn6291, a 22-kb TE carrying 25 predicted open reading frames and located between 2060105 and 2082440 in the P. fluorescens SBW25 reference sequence (European Nucleotide Archive identifier AM181176, part of genomic island SBW_GI-1; ref. 20), carries an array of cargo genes with putative cytochrome c/d oxidase functions. The presence of Tn6291 in P. putida indicates these genes mobilized from the P. fluorescens chromosome onto pQBR57 and subsequently transferred between species via conjugation. All Tn6291+ P. putida clones were also positive for pQBR57, suggesting Tn6291 remains plasmid-borne in these isolates. Indeed, we detected plasmid-borne Tn6291 in 4/6 0 µg g−1 Hg(ii) single-species P. fluorescens populations, and BLAST analysis shows a similar transposon in another soil pseudomonad, P. syringae pv. syringae B301D (ref. 21), suggesting Tn6291 mobilizes readily.
We also detected gene transfer from P. putida to P. fluorescens. P. fluorescens clones from 3/6 co-cultured populations had acquired the well-described P. putida TE Tn4652 (ref. 22) in their chromosomes, with plasmid-borne Tn4652 also present. Tn4652 is a 17-kb TE closely related to the Tn4651 toluene degradation transposon and encodes various putative enzymes, including a diacylglycerol kinase and a sulfatase. Tn4652 mobilization to the plasmid occurred readily, with events already detectible in the ancestral P. putida clones used to inoculate the soil and begin the experiment (see Methods). PCR analysis of clones obtained over the course of the evolution experiment detected Tn4652+ plasmids in P. fluorescens as early as transfer 3; however, Tn4652 insertion in the P. fluorescens chromosome was detected only later, after transfer 41 (Supplementary Table 2). In all cases, Tn4652 inserted in a region of the P. fluorescens chromosome with atypical sequence composition, likely to be recently acquired DNA20.
Importantly, interspecific transfer of chromosomal TEs via the plasmid was detected only in populations grown without positive selection for the plasmid (0 µg g−1 Hg(ii)). The amount of plasmid conjugation occurring, and thus opportunities for interspecific gene transfer, is likely to be a function of the densities of plasmid bearers and recipients23,24. By killing potential plasmid recipients, mercury selection reduces encounters between plasmid donors and recipients, and therefore conjugation9. Indeed, short-term experiments examining pQBR57 transfer (Fig. 2) showed reduced effects of conjugation on plasmid dynamics when the plasmid was under selection10, implying limited gene exchange. Together, these data suggest that positive selection for plasmid-borne resistance genes reduced the ability of that plasmid to facilitate HGT of chromosomal genes.
While mercury selection reduced TE transfer between species, we detected frequent mobilization of the HgR TE Tn5042 from pQBR57 to the chromosome. Single-species P. putida populations tended to lose the plasmid17, and sequences show that under mercury selection this was facilitated by acquisition of chromosomal Tn5042. To track the acquisition of chromosomal Tn5042 by P. putida populations, we designed PCR primers targeting ‘focal’ Tn5042 insertions (that is, insertions detected in the end-point genome sequences) in eight different P. putida populations, and applied these to clones collected across the experiment. As with single-species P. putida populations, chromosomal Tn5042 was readily acquired by co-cultured P. putida, although these populations tended also to maintain pQBR57 (Fig. 3). Similarly, chromosomal Tn5042 was detected in the P. fluorescens chromosome, which maintained the plasmid, despite its redundancy17. These findings suggest that long-term plasmid maintenance largely depends on community context and on compatibility between plasmid and host, and, provided there are no restrictions on recombination of plasmid genes into the chromosome, is unlikely to be secured by positive selection for accessory genes alone17,25,26.
Tn5042 also mobilized in the mercury-free treatments: we detected three instances of Tn5042 multiplying on plasmids and one instance of Tn5042 copying to the P. fluorescens chromosome (which occurred by transfer 35, see Fig. 3). Tn5042 insertions sometimes occurred multiple times in a lineage—in co-cultured P. putida with 16 µg g−1 Hg(ii), one clone (from replicate c) ultimately carried six copies. Although Tn5042 copy number increased in some clones from the mercury-free treatment, we detected more copies in clones evolved under mercury selection (Z = −5.4404, P < 0.0001, n = 48, exact general independence test). We did not detect any Tn5042 loss. For P. putida, Tn5042 tended to insert in a ~10-kb region near the origin of replication, while Tn5042 tended to insert in P. fluorescens near or inside Tn6291 (detected in 4/12 populations under mercury selection), in three cases representing the de novo formation of a composite resistance transposon. Here, Tn5042 became part of the cargo of Tn6291, broadening opportunities for spread, because subsequent events favouring Tn6291 mobilization (perhaps different to those of Tn5042) will cause co-mobilization of Tn5042 and its HgR genes. The pervasive mobility of Tn5042 supports a model in which TEs exploit plasmids to spread rapidly in the natural environment5, consistent with sequence analysis suggesting Tn5042 was acquired relatively recently by pQBR57 (ref. 16).
Surprisingly, we found plasmid size generally increased, primarily due to TE accumulation. Plasmid size in one clone evolved in 0 µg g−1 Hg(ii) increased by more than 10% compared with the ancestor (Fig. 1). Increased plasmid size is expected to contribute to increased cost of plasmid carriage27; however, these results suggest that such costs are negligibly small, and may be outweighed by transposition rates and/or general plasmid cost amelioration26.
The central role of HGT in adaptation is increasingly apparent, as ever-wider sequencing of isolates reveals the dynamic nature of microbial genomes4,5,28. Between-species transfer of chromosomal genes occurred only where plasmid-encoded mercury resistance was not under positive selection and the plasmid persisted instead as an infectious element. Bacterial genome evolution is determined by the interaction between selection and recombination29—here, we observed that recombination indeed makes an increased contribution to genome evolution when selection is relaxed. The transferred genes were part of the ‘accessory’ genome, which can vary even between closely related strains and is often more strongly associated with ecological niche than phylogenetic lineage30. In this case, the transferred genes were located on TEs and putative transposases could be identified. This is relevant because TEs can transfer between replicons at a high rate31, providing an efficient platform for the movement of genes between chromosomes and conjugative elements2,3. Plasmids and TEs have a close—even symbiotic—relationship. TEs can comprise a substantial fraction of a plasmid genome32,33 and, where their genes are under positive selection, they can boost the fitness of the plasmids that carry them due to genetic linkage. Similarly, unless they encode their own conjugative machinery, TEs must collaborate with elements such as conjugative plasmids to access new hosts34; indeed, models suggest that conjugative plasmids are required for TE survival and spread35. Transposase activity can be affected by stress, for example nutritional deprivation or oxidative damage36, and one intriguing possibility is that stresses caused by plasmid acquisition37 could signal to a TE that a vehicle had arrived, triggering transposition and thus increasing rates of exchange from the chromosome to that plasmid. At least one pseudomonad TE has been shown to increase activity following conjugation38.
Our results provide rare direct experimental evidence of pervasive plasmid-mediated gene mobilization, transfer and acquisition in a simple soil microcosm community. This has profound implications for the spread of accessory genes in natural communities. Consistent with our findings, two recent studies of resistance plasmids in hospital outbreaks5,39 indicate that TE mobilization dominates plasmid evolution. Furthermore, both studies suggest that plasmids may have acquired TEs outside of patients, that is, in the environment, where they are less likely to experience direct antibiotic selection. HGT vastly expands the evolutionary opportunities available to bacteria, allowing species to draw upon a collective mobile gene pool: our data indicate that environmental and ecological factors will be key modulators of the rate and extent of HGT in natural communities. HGT, particularly of antibiotic resistance and virulence genes, poses a major health concern40, and understanding the ecology of HGT-mediated bacterial evolution will be crucial to predicting and designing interventions to prevent and mitigate such threats.
The evolution experiment, described previously17, was designed to understand the effect of an alternative host species on plasmid population dynamics and evolution. The experiment used P. fluorescens SBW25 and P. putida KT2440—representative soil Pseudomonas species, a widespread and naturally co-occurring genus41 — and the 307-kb HgR plasmid pQBR57, which was isolated from the same geographic site as P. fluorescens SBW25 (ref. 16). Cultures were grown at 28 °C and 80% relative humidity in soil microcosms consisting of 10 g twice-autoclaved John Innes No. 2 potting soil, supplemented with 900 µl sterile H2O or 900 µl HgCl2 solution. We used a fully factoral design with two levels of mercury treatment (0 µg g−1, or 16 µg g−1); two levels of plasmid treatment (pQBR57+ starting with pQBR57-bearers at 50% frequency, or plasmid-free starting without plasmid); and three levels of culture treatment (single-species P. fluorescens, single-species P. putida, or co-culture with each species starting at 50% frequency). Six independent biological replicates (‘populations’) were initiated for each treatment, consistent with previous evolution experiments42,43 and sufficient to detect differences in population dynamics between the treatments17. Each replicate was initiated from independent single colonies and populations were blocked by replicate to minimize confounding effects. The experiment was not blinded. To control for marker effects, replicates a–c used gentamicin-labelled (GmR) P. fluorescens and streptomycin-labelled (SmR) P. putida, whereas replicates d–f used SmRP. fluorescens and GmRP. putida. Samples of culture (100 µl soil wash) were serially transferred into fresh soil microcosms containing either H2O or HgCl2 every four days for 65 transfers (estimated as ~440 generations17); this was decided before the experiment to be broadly consistent with other plasmid experimental evolution studies42,43. At 16 points during and at the end of the experiment, samples were spread on selective media to isolate clones, which were archived for subsequent analysis. After 65 transfers, a random number generator was used to select one plasmid-bearing and one plasmid-free clone (where present) from each pQBR57+ population for DNA sequencing. If plasmid-free or plasmid-bearing clones were present throughout the experiment but not at transfer 65, a clone from transfer 59 were used (this was the case for plasmid-bearing P. putida from single-species 16 µg g−1 replicate f, and plasmid-bearing P. putida from co-cultured 16 µg g−1 replicates a and c). We also sequenced ancestral clones and three (single-species) or two (co-culture) clones from plasmid-free treatments to test for mutations occurring in the absence of plasmid. No Tn6291, Tn4652, or Tn6290 activity was detected in the plasmid-free treatments.
DNA sequencing and analysis
DNA was extracted using the QIAGEN DNeasy Kit, prepared using the TruSeq Nano DNA Library Preparation Kit (350-bp insert size) and sequenced on the Illumina HiSeq platform. Reads trimmed using Cutadapt (version 1.2.1) and Sickle (1.200) were analysed using the BAGA pipeline19, which uses the Burrows–Wheeler short read aligner44 and calls variants using the Genome Analysis Toolkit HaplotypeCaller45. To identify structural variation (deletions, duplications and TE insertions) in the re-sequenced clones, we used the BAGA module Structure, which uses a threshold ratio of non-proper to proper paired reads to identify putative genome disruptions. Reads mapping to putative disruptions were re-assembled using SPAdes46 and contigs were aligned with the reference to identify structural variants. We also used two complementary approaches to identify structural variation: BreakDancer47 and custom scripts that examined coverage for characteristic direct repeats introduced by TE insertion (increase in coverage of ≥25% over a <30-bp region, compared with neighbouring positions). These different approaches were broadly consistent and all putative structural variants were examined using the Integrated Genome Viewer (IGV)48. Owing to differences between ancestral clones and the sequenced reference genome, variation appearing in all samples (including the ancestor) was removed from the analysis. In addition, apparent variation in hard-to-map regions (identified in an examination of parallel mutations in the IGV) was considered unreliable and excluded (Supplementary Table 3). We also examined putative single-nucleotide variants called near TE insertions and removed these manually if miscalled. Representative TE insertions were tested by PCR on clones and in all cases yielded products of the anticipated size.
Sequence analysis of ancestral clones revealed that in three cases, pQBR57 had acquired a TE before the experiment was initiated, indicated by dotted lines in Fig. 1. In the SmRP. fluorescens ancestor, pQBR57 had acquired Tn6290 at position 164349–164354. This event probably occurred in P. putida UWC1 during preparation as a donor for transfer of pQBR57 into P. fluorescens16, because Tn6290 is present in P. putida UWC1 and not in P. fluorescens SBW25. In SmRP. putida, pQBR57 had acquired Tn4652 at 152552–152558, while in GmRP. putida pQBR57 had acquired Tn4652 at 162797–162802. These events may have either occurred in the donor P. putida UWC1 strain or in the recipient P. putida KT2440 strain, as both contain identical copies of Tn4652. In any case, TE insertion must have occurred rapidly, as our stocks were all prepared from single colonies and ancestral pQBR57 contains neither Tn6290 nor Tn4652 (ref. 16). Tn4652 insertion into resident plasmids is consistent with previous work that found Tn4652 in pQBR plasmids pQBR55 and pQBR44 (ref. 16), presumably after acquisition by P. putida UWC1 (ref. 49).
PCR analysis of clones
We tested archived clones for TE insertions by PCR. Standard reactions were performed using GoTaq Green Master Mix (Promega), 0.4 µM each primer (Supplementary Table 4), and 0.2 µl archived culture on a programme of 95 °C for 5 min, followed by 30 cycles of 95 °C for 30 s, 58 °C for 30 s and 72 °C for 1 min, followed by a final extension of 72 °C for 5 min. Tn6291 was detected in reisolated P. putida clones and parallel reactions using primers targeting the P. fluorescens 16 S rDNA locus were performed to rule out the presence of contaminating P. fluorescens.
To analyse the number of Tn5042 insertions in 0 µg g−1 and 16 µg g−1 Hg(ii), we used the R package ‘coin’ to perform an exact general independence test. To avoid pseudoreplication with populations from which >1 sample was sequenced, we analysed the mean number of Tn5042 insertions per species per population. To analyse the effect of mercury on conjugation dynamics, we performed an exact general independence test on plasmid distribution between donor and recipient after five transfers.
The Bacterial and Archaeal Genome Analyser (BAGA) is available online at https://github.com/daveuu/baga. Representative scripts used to analyse our data are on Dryad Digital Archive https://doi.org/10.5061/dryad.6gf28.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank P. Koldkjaer and others at the Liverpool Centre for Genomic Research for assistance with sample preparation and sequencing. This work was supported by ERC Grant Agreement no. 311490-COEVOCON to M.A.B. and a Philip Leverhulme Prize from Leverhulme Trust to M.A.B.