A chromosomal mutation is superior to a plasmid-encoded mutation for plasmid fitness cost compensation

To understand the dynamics of chrCM and plaCM, we modified a simple, well-understood model of bacteria-plasmid population dynamics to include either chrCM or plaCM. Separate models were preferred because the combined system (i.e., containing both chrCM and plaCM) was too complex to solve analytically. Our goal was to gain a complete understanding of the resulting dynamical system—rather than a time-dependent solution—so that we have knowledge of the ultimate fate of a solution for arbitrary choices of parameters [14,24]. The basic model, without CMs, is detailed in Supplementary Text B in S1 File and recapitulated the key findings of previous studies wherein costly plasmids do not invade unless their conjugation rate, γ, is larger than μ(α−β)/(α−μ), and only competitively displace the plasmid-free population if γ is larger than μ(α−β)/(β−μ), where α is the plasmid free growth rate, β is the plasmid containing growth rate, and μ is the population turnover rate. Positive selection for the plasmid is included via the selection pressure term, η, which was initially set to zero. Spontaneous plasmid loss by segregation was not incorporated, as previous experimental work showed that these events are rare on the timescale of the study such that population change is principally driven by competition with existing plasmid-free cells rather than their de novo generation by segregation [25], and including this term in the models complicates analytical progress.

We then considered how the addition of chrCM affects the outcome of this underlying basic system. Here, plasmid-free wild-type bacteria (f) and wild-type bacteria containing a wild-type plasmid (p) are invaded by a chrCM variant bearing a wild-type plasmid (c). The compensatory effect is assumed to be imperfect so the growth rate of the 3 strains are assumed to be α for f, β_P for p, and β_C for c, respectively, where α>β_C>β_P. The dynamics of the system are then described by the following set of ODEs:
(1)
(2)
(3)

In the case of no selection for the plasmid, η = 0, this set of ODEs can be solved exactly for the fixed points to yield a simple phase plane structure in which chrCM sweeps to fixation (0,0,c*) after invasion by the mutated strain according to the value of the conjugation rate (Fig 1A, green region). When the underlying fixed point is plasmid-free only (f*,0,0) (i.e., ) or mixed (f*,p*,0) (i.e., ), and compensation is imperfect, the fixed points are separated by a saddle at (f_s,p_s,c_s). This means that when the conjugation rate is sufficiently high the chrCM will always invade, but at lower conjugation rates (provided α>β_C) there is a threshold. If the initial proportion of chrCM exceeds the threshold value (given by the saddle), complete replacement occurs and chrCM successfully invades and moves to fixation (Fig 1A, pink region). The reason invasion can occur despite chrCM having lower growth than the plasmid-free is that a plasmid-carrying chrCM can conjugate the costly plasmid into plasmid-free competitors (“weaponisation”), such that chrCM then exceeds the uncompensated competitor’s growth rate (i.e., β_C>β_P) and takes over the system. Selection (η>0) reduces the relative ability of plasmid-free wild-type f to compete against plasmid-bearers, with the result that the dynamics are driven by the competition between p and c and the difference between β_C and β_P, thus favouring chrCM. This result is demonstrated in Supplementary Text B in S1 File.

Fig 1. Phase diagrams describing the fate of compensatory mutations.

(A) Chromosomal CMs. (B) Plasmid CMs. Axes describe relative conjugation rates without (x) and with (y) the corresponding CM, with the black line indicating no difference (γ_P = γ_Q = γ_C, note that plots focus on the region of interest and thus the black line does not intersect the point where the axes meet). Where γ_Q>γ_P (i.e., both growth rate and conjugation rate are increased by the CM, indicated in grey) in the plasmid compensation model the wild-type plasmid is always lost, with the system reverting to the well-known two-member model of a plasmid-free and plasmid-containing type, as described in Supplementary Text B in S1 File. While similar behaviour occurs in the chrCM model, there is not the same mathematical simplification. Plots illustrating temporal dynamics for different regions of parameter space are shown in Fig A in S1 File. The data underlying this figure can be found in https://dx.doi.org/10.5281/zenodo.13963497.

https://doi.org/10.1371/journal.pbio.3002926.g001

We next considered a system invaded by a variant bearing a plaCM plasmid (q). This leads to a set of 3 differential equations analogous to those above:
(4)
(5)
(6)

Analysis of the system reveals a more complex phase portrait, described in Fig 1B in the case of no selection, η = 0. Consistent with prior work [19], comparison of Fig 1A and 1B shows a wider range of conditions in which plaCM is successful compared with chrCM. Predominantly, this is where the CM enables plasmid invasion or dominance (purple and green regions in Fig 1B): biologically plausible conditions given previous observations of intermediate levels of plasmid carriage by populations [26]. Where there is no trade-off with conjugation rate, plaCM always displaces the uncompensated plasmid, invading the system if and dominating if , and, unlike chrCM, in a manner that does not depend on initial conditions.

Previous experiments have shown that CMs can affect the ability of a plasmid to conjugate [21,22,27]. We therefore investigated how the success of each CM is affected by changes in the conjugation rate. For plaCM, if γ_Q>γ_P, i.e., plaCM confers a higher transfer rate than the wild-type plasmid, the wild-type plasmid is always lost from the system, and the outcome for plaCM collapses into the single-plasmid system described above (Supplementary Text B in S1 File). However, if there is a trade-off such that γ_P>γ_Q, various outcomes are possible depending on the other parameters, including loss of both plasmids (f*,0,0), fixation of plaCM (0,0,q*), coexistence between wild type and plasmid-free (f*,p*,0), and, unexpectedly, a state with a long-term stable coexistence between the f, p, and q populations (Fig 1B, orange region), which would not be possible to find in a linearised adaptive dynamics approach. The stable fixed point is oscillatory in character (a stable spiral) and is driven by Rock-Paper-Scissors (RPS)-like nontransitive dynamics. When f is large, this promotes the conjugative spread of the fastest conjugating population, p. When p is large, f is small, so the opportunities for conjugation are relatively low but the force of infection remains high due to high γ_P. Here, the q outgrows the p. When q is large opportunities for conjugation are also low but as γ_Q is relatively low the f can outgrow the q. This RPS-like dynamic is approximate as the interactions are not perfectly symmetric: in the prototypical model, each type has a direct impact on one other type and is directly impacted by the third. Here, the competition arises through a mixture of competitive growth and competitive infection and is different for each combination of subpopulations. The inequality γ_P>γ_Q can be interpreted as applying on average, e.g., transconjugants could initially have an increased or reduced conjugative rate post conjugation before reverting to the long-term rate, and the ultimate fate of the system would be the same.

Conjugation rates can vary with chromosomal genotype over many orders of magnitude, even for the same plasmid [28], and there is evidence that chromosomally encoded genes can directly regulate conjugation gene expression [29–31], suggesting that chrCM could also result in changes to conjugation rate. We found that for chrCM, the system is generally robust to changes in conjugation rate provided compensation is sufficiently strong (). In cases where chrCM has a more substantial negative effect on γ_C, the CM is lost (Fig 1A, red and blue regions), except where the conjugation rate from uncompensated strains is sufficiently high (Fig 1A, orange area, ). Under these conditions, the force of infection of costly plasmids ensures that the frequency of plasmid-bearers in the system is maintained at a high enough level such that chrCM has sufficient competitive advantage to persist. Like the plaCM system, the chrCM system can also result in an oscillatory stable coexistent solution, driven by a similar form of RPS dynamics when γ_P is large and γ_C small, when p invades f populations, is displaced by c mutants that are resistant to the costly plasmid, which are in turn susceptible to invasion by f which can outgrow re-infection from c. A chrCM that increases conjugation rate is less expected biologically, but the overall long-term patterns would similarly remain dependent on the ability of the plasmid to dominate the system: if it can, then chrCM is always favoured owing to the fitness benefit it enjoys relative to uncompensated plasmid-bearers, if it cannot then the fate of chrCM is dependent on its initial frequency in the manner described above, with lower frequencies required for higher levels of γ_C For both CMs, selection simplifies the dynamics by reducing the contribution of the plasmid-free population (f), concentrating the competition on the growth differences between uncompensated and compensated strains, and when selection is large the plasmid free population is simply eliminated.

Although four-equation models describing a direct competition between chrCM and plaCM cannot be solved analytically, some insight can be gained by comparing Eqs 3 and 6. Specifically, we can see that as the abundance of plasmid-free recipients decreases, the γ_Qf component that positively affects the success of plaCM correspondingly decreases, such that the relative success of plaCM and chrCM is increasingly determined by the difference between β_C and β_Q, i.e., the relative strengths of amelioration of the 2 CMs. Numerical calculations of the fixed points for experimentally measured parameter values likewise indicate that the extent of compensation is the overriding factor determining the relative success of competing CMs.

Overall, then, our models identify a broader range of parameter space in which plaCM is likely to succeed, relative to chrCM. However, we also predict that trade-offs between compensatory mutations and plasmid conjugation can have complex effects on the overall success of a CM, particularly in communities that contain a mixture of uncompensated plasmid-carrying and plasmid-free competitors. Selection reduces the complexity of the dynamics in both cases by removing potential recipients from the system, thus reducing the contribution of plasmid transmission to the dynamics, and increasing the dependency of the outcome on the relative strength of the CM. We note that these results are likely to represent biologically relevant parameter space, as our numerical models could generate results in all areas of the parameter phase diagrams from experimentally measured parameter values by manipulating conjugation rates over less than 2 orders of magnitude (Fig A and Table A in S1 File).

To experimentally investigate the dynamics of plaCM versus chrCM, we established an experimental system in which fluorescently labelled strains were engineered with plaCM or chrCM, enabling enumeration by flow cytometry. To test that our newly engineered strains exhibited the fitness effects by flow cytometry that we have previously observed by CFU plating, we first performed 24-h competition experiments. As expected, each CM ameliorated the fitness cost of plasmid carriage (Fig 2, linear model [LM] one-sided post hoc comparison against 0: plaCM t₉ = 22.9, p_adj 9 = 24.3, p_adj 9 = 3.41, p_adj = 0.015) and there was only a marginal fitness benefit of combining both CMs in the same cell (coefficient comparing chrCM versus double-compensation at no selection = −0.13, t₉ = −2.47, p_adj = 0.07; coefficient = −0.13, t₉ = −2.86, p_adj = 0.04 with selection). Interestingly, the benefits of CMs relative to non-compensated plasmid-carrying strains were enhanced in the presence of mercury, and more strongly so for chrCM than for plaCM (chrCM difference in coefficients = 1.0, plaCM difference in coefficients = 0.7), suggesting that the costly cellular disruption caused by pQBR57 is further exacerbated by exposure to mercury (as previously described by Carrilero and colleagues [32]). These results cause us to expect acceleration of CM invasion under mercury selection.

Fig 2. Engineered plaCM and chrCM variants both ameliorate plasmid fitness costs, though chrCM is more effective.

Panels indicate test strains. Unfilled circles indicate mean across 10 replicates, each of which is indicated by a semi-transparent filled circle. The asterisk indicates data in the right panel which is also presented in the left panel. The data underlying this figure can be found in https://dx.doi.org/10.5281/zenodo.13963497.

https://doi.org/10.1371/journal.pbio.3002926.g002

We previously did not observe any significant difference in the intraspecific conjugation rates of pQBR57 with or without chrCM or plaCM [12]. However, these experiments were conducted over a relatively long time window (24 h), which could allow conjugation from transconjugants or fitness differences between transconjugants, donors, and recipients to mask differences in transfer rate [33,34]. We therefore re-measured conjugation rate of pQBR57 and pQBR57::plaCM using the Approximate Extended Simonsen approach [34]—an extension to the popular “Simonsen’s gamma” [35] that accommodates the potential for variation in growth rate and secondary conjugation events. Again, we did not detect any significant difference in conjugation rate between the wild-type and plaCM variants of pQBR57 (Fig B in S1 File; t_9.86 = 0.75, p = 0.94, TOST equivalence test with log₁₀-transformed bounds ±0.5, p = 0.02), and the similarity with previously measured values provided confidence that fitness differences and onward conjugation did not substantially affect our previous conjugation rate measurements [12]. Together, these measurements indicate that both CMs exhibit a fitness advantage over uncompensated pQBR57, with chrCM providing the greater benefit, and that neither CM exhibits a detectable trade-off with per capita conjugation rate.

To test our model predictions, we used our validated strains to test the relative success of chrCM versus plaCM under varying ecological conditions. First, we varied the strength of selection for the plasmid-encoded mercury resistance trait. Informed by our models, we predicted that selection would increase the success of chrCM relative to plaCM. This is because selection would remove plasmid-free recipients from the system, and as plasmids prevent superinfection by similar plasmids through surface exclusion and/or entry exclusion, plaCM would gain no benefit from its ability to transfer by conjugation. For these experiments, our strains were labelled to track the relative success of each mode of compensation, and so the label was inserted into the replicon encoding the compensatory mutation, i.e., the SBW25ΔPFLU4242 chromosome for chrCM and pQBR57ΔPQBR57_0059 for plaCM. SBW25::chrCM carrying wild-type pQBR57 was competed against pQBR57::plaCM in a wild-type chromosomal background in populations initially containing wild-type plasmid-free recipients at 50% frequency (1:1:2 chrCM:plaCM:recipients), and populations were transferred for 8 transfers (approximately 50 generations). Consistent with our predictions, mercury selection indeed favoured chrCM (Fig 3, generalised linear mixed effects model [GLMM] interaction effect of selection:transfer χ² = 102.9, p Fig 3). However, chrCM was also fitter than plaCM without selection (coefficient for transfer = −0.32, z = −15.7, p Fig 3).

In parallel, we investigated whether the benefits of chrCM could be increased in microbial communities hosting multiple costly plasmids. Our previous work showed that pQBR57 and pQBR103 could be harboured in the same cell. We also showed pQBR103 could be compensated by the chrCM, ΔPFLU4242, both by itself and together with pQBR57, whereas evidence suggested that plaCM exacerbated the cost of pQBR103 [12,32]. We therefore predicted that chrCM would be favoured over plaCM both with and without selection in pQBR103-harbouring communities, since chrCM would ameliorate both plasmids. We established experiments as with Fig 3 except replacing the plasmid-free subpopulation with a subpopulation carrying pQBR103. Indeed, in communities harbouring pQBR103, chrCM again outcompeted plaCM (Fig 4, GLMM coefficient for transfer = −0.30, z = −14.5, p Fig 4), but there was no detectable additional effect of pQBR103 on the relative success of chrCM versus plaCM.

We reasoned that increasing the pool of potential recipients may tip the balance towards plaCM, since conjugation (and thus transmission of the CM) could then play a bigger role in the population dynamics. We again established populations beginning with equal proportions of plaCM and chrCM but varied the proportions of recipients: (i) 10-fold excess of plasmid-free; (ii) equal abundance of plasmid-free; (iii) 10% plasmid-free; and (iv) without added plasmid-free. Given the lack of detected marker effect in Figs B and C (in S1 File), these experiments were conducted with SBW25::dTomato::chrCM(pQBR57) and SBW25(pQBR57::GFP::plaCM), and populations were propagated without selection. Contrary to expectations, plaCM performed relatively poorly against chrCM across all frequencies of plasmid-free recipients (Fig 5, GLMM transfer:ratio interaction χ² = 5.00, p = 0.17; main effect of transfer χ² = 88.6, p p

Fig 5. ChrCM was more successful than plaCM regardless of recipient availability.

Subpanels indicate, from left to right, the starting fraction of wild-type/plasmid-free cells relative to a 50:50 mix of chromosomal CM (with wild-type plasmid) and plasmid CM in wild-type cells. Plots indicate the mean of 6 independent experiments; replicate-level plots are provided in Fig E in S1 File. The data underlying this figure can be found in https://dx.doi.org/10.5281/zenodo.13963497.

https://doi.org/10.1371/journal.pbio.3002926.g005

Our model predicted that transmission of wild-type plasmids from chrCM cells could play an important role in the success of chrCM. Furthermore, the theoretical advantage gained by plaCM through horizontal transfer may be reduced when in competition against wild-type plasmids carried by cells with chrCM, because wild-type plasmid transfer from chrCM could remove potential recipients for plaCM. However, our initial experiments used fluorescent labels to track the fates of the different compensatory alleles, i.e., the chromosomes of chrCM (SBW25::chrCM), and the plasmids of plaCM (pQBR57::plaCM), rather than the plasmids which began in each background. To understand how the plasmids themselves were affected by the different CMs, we established a complementary experiment to that in Fig 5, except, rather than tracking chrCM, we tracked the wild-type pQBR57 that began in the chrCM background (Fig 6), i.e., using SBW25::chrCM(pQBR57::tdTomato) and SBW25(pQBR57::plaCM::GFP). Compared with pQBR57::plaCM, wild-type uncompensated pQBR57 from chrCM was significantly more successful, with a considerable proportion of the plasmids at the end of the experiment being wild-type uncompensated plasmids that began in the chrCM population, and patterns of invasion depending on the abundance of potential recipients (GLMM third-order polynomial transfer:ratio interaction χ² = 287.4, p Fig 5 (which were performed in parallel) revealed that invasion of the wild-type uncompensated plasmid from chrCM preempted the invasion of chrCM, indicating that plasmid transmission likely contributed to the overall success of chrCM. Essentially, as plasmid-free competitors are removed from the system by infection with the wild-type uncompensated plasmid, the space is filled by competition between CMs, which favours the more effective mechanism of compensation: chrCM. Indeed, when plasmid-free recipients were initially rare, the dynamics essentially come down to a competition between the compensatory mutations, with the most effective mechanism, chrCM, becoming dominant. To understand our experimental results further, we ran numerical simulations based on Eqs 1–6 and the equations in Supplementary Text A (in S1 File) parameterised with biologically plausible values (Table A in S1 File), which resulted in dynamics resembling those observed in the experiments (Fig G in S1 File). By contrast, inactivating the ability for plasmid “weaponisation” by chrCM (setting γ_C = 0), fundamentally altered the dynamics, preventing chrCM fixation. The experimental results were therefore consistent with our model prediction that transmission of costly plasmids, and the consequent removal of plasmid-free recipients, facilitates the success of chrCM.

The relative inability of plaCM to invade the recipient population, when compared with wild-type pQBR57 from chrCM cells, suggested that either plaCM exerted a negative pleiotropic effect on the ability of pQBR57 to invade a recipient population, or, conversely, that chrCM enhanced the ability of wild-type pQBR57 to invade. To distinguish between these possibilities, we conducted a similar experiment to Fig 6, but here, plaCM was competed against pQBR57 harboured by an uncompensated competitor, i.e., SBW25(pQBR57::tdTomato) and SBW25(pQBR57::plaCM::GFP). As with Fig 6, fluorescent labels were designed to track the relative success of the 2 plasmids. Unexpectedly, plaCM was only able to outcompete wild-type pQBR57 in the situation where plasmid-free recipients started at low frequency, i.e., effectively a head-to-head competition between compensated and uncompensated plasmid-bearers. We detected a significant effect of plasmid-free recipient ratio on the competition dynamics (GLMM fourth-order polynomial transfer:ratio interaction χ² = 414.4, p Fig 2), an observation that suggests that plaCM inhibits plasmid transmission or establishment in transconjugants relative to the wild type. Overall, plaCM did not appear to gain a substantive fitness benefit from horizontal transmission, only reaching dominance under conditions that prioritise vertical replication and only in populations without chrCMs.

Our analytical models for plaCM predicted cyclical RPS-like dynamics for some combinations of parameters (Fig 1B, orange region). The experimental results in Fig 7 were consistent with such nontransitive dynamics. Specifically, plasmid-free populations were invaded by the wild-type plasmid (left panel, and mid-left panel after transfer 4), the wild-type plasmid was outcompeted by plaCM (right panel, mid-right panel, and mid-left panel before transfer 4), and plaCM was outcompeted by plasmid-free (mid-right panel). To explore the dynamics in further detail, we compared experimentally determined parameters (Table A in S1 File) with the analytical model (Fig 1). With our system, and with experimentally measured parameter values, the uncompensated plasmid exceeds the threshold for plasmid invasion and domination, which is indeed what we have described previously [36]. Both chrCM and plaCM likewise exceed the threshold for domination by 18- and 10-fold, respectively. However, numerical simulations based on Eqs 1–6 and the equations in Supplementary Text A (in S1 File), and parameterised with biologically plausible values, required a considerable reduction of γ_Q relative to γ_P (10–100×) for the system to exist in the (f*,p*,q*) region of parameter space that yields oscillatory RPS-like dynamics resembling the experimental results of Fig 7 (Fig I in S1 File)—a reduction inconsistent with the measurements of γ_Q and γ_P described above (Fig B in S1 File) or in our previous work. An interactive (Shiny) app enabling readers to explore these patterns is provided by following the link at www.jpjhall.net/plasmid-dynamics-model.

We therefore considered 2 other processes that could affect plaCM transmissibility beyond per capita conjugation rate: acquisition cost, and conjugation derepression. Each process has a transient effect on newly formed transconjugants. Acquisition costs are imposed by plasmids following arrival in a recipient, are often expressed through increased lag time, and are usually resolved in a matter of hours, in contrast to fitness costs which are genetically based, and inherited following cell division [37,38]. Conjugation derepression occurs when transconjugants exhibit a transient increase in onward conjugation rate, owing to a delay in the expression of the machinery that regulates conjugation genes [34,39]. If plaCM imposed greater acquisition costs, or a reduced degree of conjugation derepression, this might impede invasion of plaCM into a naïve population, relative to the wild-type pQBR57. To assess whether such effects could be measured experimentally in our system, we performed conjugation experiments with a 100-fold excess of recipients, such that conjugation from novel transconjugants would make an increased contribution to the calculated conjugation rate. However, no significant difference in conjugation rate was observed when comparing pQBR57 and pQBR57::plaCM (Fig J in S1 File, t_4.89 = 0.33, p = 0.775). We also measured the lag time of de novo transconjugants upon gaining pQBR57 or pQBR57::plaCM, but again, we could not detect a trade-off (Fig K in S1 File), with de novo plaCM transconjugants in fact displaying a significantly reduced lag time compared with wild-type pQBR57 transconjugants (t_4.71 = 3.81, p = 0.014). We therefore conclude that the differences in pQBR57 transmissibility between wild-type and plaCM variants predicted by our models and observed experimentally in Fig 7 emerge from yet-to-be-determined processes that will likely require further detailed examination of the molecular mechanisms underlying plaCM to elucidate.

Fluorescently labelled strains of Pseudomonas fluorescens SBW25 and P. fluorescens SBW25ΔPFLU4242 [15] were generated using the mini-Tn7 system and plasmid pUC18T-mini-Tn7T-Gm-eyfp, gifts from Herbert Schweizer via Addgene plasmids #65031 and #64968 [53] or a derivative in which dTomato was cloned to replace eyfp. The pQBR57ΔPQBR57_0059 knockout and fluorescently labelled variants of megaplasmid pQBR57 [18,25] were generated using homologous recombination with plasmid pTS-1 [54]. Briefly, for the knockout, 1 kb flanking regions of pQBR57 immediately upstream and downstream of the PQBR57_0059 CDS (nucleotides 47925 and 48589 in the reference sequence LN713926) were amplified and cloned into the MCS of Xba-KpnI-digested pTS-1 using NEB HiFi assembly. For the fluorescently labelled plasmids, 1 kb fragments of pQBR57 targeting the site between PQBR57_0027–0028 at nucleotide 16050 in the reference sequence LN713926 and an expression cassette consisting of a Ptac promoter driving a fluorescence protein gene (either eGFP or tdTomato), followed by lambda t0 and rrnB T1 terminators were amplified and cloned into the MCS of XhoI/KpnI-digested pTS-1 using NEB HiFi assembly. Constructs were transformed into SBW25(pQBR57) by electroporation [53], and merodiploids selected on KB supplemented with 100 μg/ml tetracycline. Double-crossovers were selected on LB supplemented with 10% w/v sucrose and 20 μm HgCl₂, and candidates screened by PCR and tetracycline sensitivity before sending for whole genome sequencing (2 × 250 bp, >30× coverage, MicrobesNG) to test for second-site mutations. Using breseq [55], no second-site mutations were detected for the strains used in these experiments. For experiments, each plasmid-containing replicate was established with an independent transconjugant from a genome-sequenced donor strain. Conjugation experiments were performed with non-fluorescent antibiotically labelled strains described previously [12].

To determine the short-term fitness effects of compensation, direct competitions were performed between YFP- versus dTomato-labelled strains. For these experiments, all strains were chromosomally labelled. Overnight cultures were mixed at 1:1 ratio (test:reference) before inoculation at 1:100 dilution into 6 ml King’s B media in a 30 ml glass universal with loose-fitting lid (“microcosm”), with or without mercury (Hg (II), 40 μm) and incubated at 28°C, 180 rpm for 24 h. Each competition was repeated with 10 biological replicates. Flow cytometry was used to estimate bacterial counts: starting mixtures and endpoint competition cultures were diluted 1:100 into M9 buffer and run on a Beckman Coulter CytoflexS machine at 17 μl.min^-1 for either 5,000 counts (determined as events with signal in SSC-H channel >10³) or 90 s maximum. Between samples, M9 buffer was sampled for 5 s to minimise cross-over between samples. Strain counts were determined by gating in the following channels: FITC-H (gate = 10^3.4 for YFP) and PE-H (gate = 10^3.4 for dTomato). Relative fitness was calculated as the difference in Malthusian parameters, . Control experiments to assess potential marker effects, performed by competing isogenic, differently marked SBW25(pQBR57) strains, showed a small, nonsignificant fitness cost of the marker (coefficient = −0.03, t₉ = −0.77, p = 0.46), and all values were corrected by subtracting the marker effect from the calculated relative fitness value. Other control experiments, performed to assess the effect of plasmid-encoded markers, produced results consistent with expectations from unmarked plasmids (Fig L in S1 File).

Raw data from flow cytometry experiments (relative fitness and dynamics experiments) are provided at https://doi.org/10.5285/51046841-deaa-422f-a303-2c0759f014b4, and the processed data used for figures and analysis are provided as part of the Zenodo repository associated with this article at https://dx.doi.org/10.5281/zenodo.13963497.

Streptomycin-resistant lacZ-carrying donors (either SBW25::Sm^R(pQBR57) or SBW25::Sm^R(pQBR57ΔpQBR57_0059)), Gm^R recipients (SBW25), and Gm^R transconjugants (either SBW25(pQBR57) or SBW25(pQBR57ΔpQBR57_0059)), cultured overnight in 150 μl KB broth in an untreated CytoOne 96-well microtitre plate at 28°C, were subcultured 1:30 into 150 μl fresh media and placed in a Tecan Nano plate reader for incubation at 28°C with shaking. When exponential phase was reached (assessed by examination of growth curves; OD600 ~ 0.4), cultures were again diluted 30-fold into KB. Mixed cultures containing donors and recipients, or single-strain donor, recipient, or transconjugant cultures were sampled and cultured in the plate reader and spread on KB agar plates supplemented with 50 μg/ml X-gal for enumeration. After approximately 4 h growth, cultures were again sampled and spread on KB agar plates, some of which were supplemented with antibiotics (250 μg/ml streptomycin or 30 μg/ml gentamicin) and 20 μm mercury to enumerate transconjugants. Conjugation rates were calculated with the Approximate Extended Simonsen Method [34]. Excess-recipient conjugation experiments were performed as above, except using gentamicin-resistant donors (either SBW25(pQBR57) or SBW25(pQBR57ΔpQBR57_0059)) and streptomycin-resistant lacZ-carrying recipients (SBW25), and donor and recipient cultures were spread for enumeration before mixing equal volumes of recipient and a 1:100 dilution of donor and incubating in a CytoOne 96-well microtitre plate at 28°C on a Stuart SSL-5 shaker for 18 h. Endpoint samples were spread for enumeration as above. While conjugation rate was therefore measured under slightly different conditions from those used in the serial passage experiment (glass vial versus 96-well polystyrene plate), consistency with previously measured rates in glass vials and the similarity between the 2 plasmid variants (indicated in Fig B in S1 File), suggest that such modifications do not significantly affect our conclusions. Data from these experiments are provided as part of the Zenodo repository associated with this article (https://dx.doi.org/10.5281/zenodo.13963497).

Conjugations were performed as described above, except using gentamicin-resistant donors (either SBW25(pQBR57) or SBW25(pQBR57ΔpQBR57_0059)) and streptomycin-resistant lacZ-carrying recipients (SBW25). Cultures were also set up using established transconjugants (SBW25::Sm^R(pQBR57) or SBW25::Sm^R(pQBR57ΔpQBR57_0059)) and grown into the exponential phase. After approximately 4 h growth of the cultures/co-cultures, strains were diluted 1:10 into KB in a 96-well plate supplemented with 40 μm HgCl₂ and 200 μg/ml streptomycin and incubated shaking for 1 h. Samples were then pelleted at 1,000 g for 10 min, washed with 1 ml KB, and resuspended in 200 μl KB. Samples were diluted and spread on KB agar for enumeration, and growth curves were set up in KB supplemented with 200 μg/ml streptomycin and 20 μm HgCl₂ in the Tecan Nano plate reader for incubation at 28°C with shaking and measurements (OD600) every 15 min. Data are provided as part of the Zenodo repository associated with this article (https://dx.doi.org/10.5281/zenodo.13963497).

Relative fitness was analysed using linear models, with post hoc pairwise comparisons performed using the package emmeans [56] and Bonferroni-corrected t tests. Dynamics were analysed with GLMMs using the R package glmmTMB [57], with a beta-binomial response distribution, a logit link function, and the counts of each competitor as the response variables. Preliminary analyses identified overdispersion in the data, justifying the use of a beta-binomial rather than a binomial response distribution [58]. Non-independence of measurements arising from repeated sampling of populations was accommodated with random effects of “population” on intercept and slope, except for the experiment presented in Fig 7 which included only the random effect on intercept due to extremely low variance and high correlation between random effects preventing model convergence. For experiments presented in Figs 6 and 7, polynomial terms were added to accommodate potentially non-monotonic relationships between competitors over time. Significance of fixed effects was determined by comparison of nested models using likelihood ratio tests. Conjugation rate data were analysed by t test and 2 one-sided tests using the package TOSTER [59]. Growth curve data were analysed using gcplyr [60]. Analysis scripts are provided on GitHub as part of the Zenodo repository associated with this article (https://dx.doi.org/10.5281/zenodo.13963497).