Bacteria can encode dozens of different immune systems that protect them from infection by mobile genetic elements (MGEs). MGEs themselves may also carry immune systems, such as CRISPR-Cas, to target competitor MGEs. It is unclear when this is favored by natural selection, and whether toxin–antitoxin (TA) systems—common competitive mechanisms carried by plasmids—can alter their efficacy. Here, we develop and test novel theory to analyze the outcome of competition between plasmids when one carries a CRISPR-Cas system that targets the other plasmid. Our mathematical model and experiments using Escherichia coli and competing IncP plasmids reveal that plasmid-borne CRISPR-Cas is beneficial to the plasmid carrying it when the plasmid has not recently transferred to a new host. However, CRISPR-Cas is selected against when the plasmid carrying it transfers horizontally, if a resident competitor plasmid encodes a TA system that elicits post-segregational killing. Consistent with a TA barrier to plasmid-borne CRISPR-Cas, a bioinformatic analysis reveals that naturally occurring CRISPR-Cas-bearing plasmids avoid targeting other plasmids with TA systems across bacterial genera. Our work shows how the benefit of plasmid-borne CRISPR-Cas is severely reduced against TA-encoding competitor plasmids, but only when plasmid-borne CRISPR-Cas is horizontally transferred. These findings have key implications for the distribution of prokaryotic defenses and our understanding of their role in competition between MGEs, and the utility of CRISPR-Cas as a tool to remove plasmids from pathogenic bacteria.

Mathematical modeling predicts that the benefit of CRISPR-Cas depends on toxin–antitoxin (TA) activity of competitor plasmids

Both CRISPR-Cas and TA systems are implicated in plasmid-level competition [8,19]. In a mathematical analysis, we set out to formalize the relative benefits of CRISPR-Cas to a plasmid when competing with a plasmid encoding TA, depending on asymmetries due to distinct offensive versus defensive roles of CRISPR-Cas and TA. A defensive role refers to a plasmid remaining resident in a cell line. In contrast, an offensive role refers to horizontal transfer of a plasmid to a new host and becoming a newly arrived invader in a cell.

To formalize plasmid competition dynamics, we begin with a simple conceptualization of plasmid competition (Fig 1A and 1B). Our approach focuses on cells co-infected with both plasmids (CT and TC cells), with plasmids differentiated by their order of arrival. C represents cells where a CRISPR-Cas plasmid is resident, while T represents cells where a TA plasmid is resident. In the event C cells are infected by a TA-plasmid, we have a CT cell containing both a resident CRISPR-Cas plasmid and an invading TA plasmid. Conversely, TC cells are created when a T cell acquires a CRISPR-Cas plasmid.

Our model defines transition rates from co-infected states (CT or TC) to singly infected states (C or T) (arrows in Fig 1A and 1B). These transition rates are defined by plasmid segregational loss (baseline rate s), modified by C- and T-specific traits of CRISPR-Cas-induced enhanced segregation x (blue in Fig 1A; x > 1 implies multiplicatively increased segregation rate x ∙ s) and TA-induced probability of post-segregational killing (PSK) y (red in Fig 1A). Asymmetries in parameter values between resident and invader roles are captured by subscripts r and i. In the event of PSK, liberated resources (nutrients, space) can enhance the growth of remaining C or T cells, weighted by parameter f [19] where f_r refers to the benefit given to cells containing the previously resident plasmid (T for TC cells and vice versa). Given lower plasmid copy number and lower gene expression for invading versus resident plasmids [22], our default assumptions are s_i > s_r; x_r > x_i ≥ 1; 1 > y_r > y_i ≥ 0. Given spatial structuring so that T cells are enriched in the neighborhood of TC cells (and vice versa for C near CT cells), we further assume 1 > f_r + f_i > f_r > f_i  ≥ 0. The resulting transition rates (Fig 1A and 1B, equations E1–E4) are used to define CRISPR-Cas competition summary metrics Δ_C, which capture whether CRISPR-Cas investment x results in more transitions to C than to T cells (Δ_C > 0), or vice versa (Δ_C < 0). These metrics are calculated for both the context when the CRISPR-Cas plasmid is resident (Fig 1A–1C, equation E5), and when it is an invader (Fig 1B–1D, equation E7).

Analysis of Δ_C of CRISPR-Cas plasmids provided the following core predictions concerning the impact of CRISPR-Cas and TA investments (x and y) on their relative competitive success (derivations in S2 Text): Firstly, in the absence of PSK (no TA; y = 0), increasing investment in CRISPR-Cas x always increases CRISPR-Cas competitive success, regardless of whether in a defensive (Fig 1C) or offensive role (Fig 1D). Secondly, in the presence of PSK (y > 0), increasing CRISPR-Cas investments is still beneficial in a resident context (Fig 1C), but can become detrimental in an offensive role, given sufficiently strong resident TA y_r (Fig 1D, area above the dashed line). Analytical conditions for this qualitative shift in response are defined in the S2 Text.

In conclusion, our model predicted an asymmetric outcome of plasmid competition, depending on CRISPR-Cas mode of action. When resident, investing in stronger CRISPR-Cas always benefitted its own plasmid. In offense CRISPR-Cas switched from an asset to a liability in plasmid competition, dependent on a sufficiently strong TA response on the competitor plasmid.

Experimentally competing a CRISPR-Cas and TA plasmid showed a universal CRISPR-Cas benefit in defence, but not in offense

Our mathematical modeling predicts that the benefits of a plasmid-borne CRISPR-Cas system are limited during competition with TA plasmids. We sought to test these predictions experimentally in a model system where we could modulate CRISPR-Cas and TA strength in a binary way. To this end, we used competitor plasmids pKJK5::csg (IncP1ε) and RP4 (IncP1α) in E. coli hosts as models to examine how selection acts on plasmid-encoded CRISPR-Cas under defensive and offensive modes of action (Fig 2A and 2B). Due to plasmid incompatibility resulting in segregational loss, these plasmids cannot be co-maintained in the same host indefinitely but can co-exist for several generations [23], giving an ample window of opportunity for plasmid competition within the same bacterial host. Plasmid pKJK5::csg was engineered to carry a minimal Type II CRISPR-Cas9 system [24], and we turned CRISPR-Cas activity on or off by using two isogenic plasmid variants that only differed in their respective fixed targeting specificity (Fig 2A, see Methods). Competitor plasmid RP4 naturally carries the Type II TA system parABCDE, where parABC form a segregational stability system and parDE code for a TA system that inhibits DNA gyrase, leading to cell filamentation and cell death of plasmid-free segregants [25–27]. We turned RP4’s TA activity off by supplying a bystander expression vector engineered to carry RP4’s par TA operon (pCDF1b_par) to all potential plasmid hosts, thus allowing supply of the antitoxin in trans and negating the PSK effect of RP4’s TA system. As a control, empty vector pCDF1b was supplied in all treatments where TA remained turned on (Fig 2A). These vectors were maintained throughout all experiments in the absence of antibiotic selection (S1 Fig).

To assess plasmid competition outcome under different modes of CRISPR-Cas action, we set up mating assays including two differentially tagged E. coli DH5ɑ hosts with different plasmid content (Fig 2B). In this way, we could track plasmid competition when CRISPR-Cas was acting defensively (a pKJK5 host was invaded by RP4) and when CRISPR-Cas was acting offensively (an RP4 host was invaded by pKJK5). After solid-surface filter mating, we calculated the outcome of plasmid competition by measuring the relative abundance of each plasmid in each host by selective plating (competitive ratio; the log odds ratio of the proportion of pKJK5 over RP4 carriage within hosts; positive values indicate that CRISPR-Cas plasmid pKJK5 was more abundant in its host, and negative values indicate that TA plasmid RP4 was more abundant).

This analysis revealed that when CRISPR-Cas was acting defensively, the CRISPR-Cas plasmid pKJK5 won the competition in every case where CRISPR-Cas was switched on (Fig 2C; competitive ratio = 5.04 ± 0.10 and 5.05 ± 0.12, respectively; both significantly higher than 0 with p < 0.001 after T test and Bonferroni adjustment for multiple testing with α = 0.0063). In stark contrast, when CRISPR-Cas was acting offensively, switching on CRISPR-Cas had a differential effect depending on TA status of the competitor plasmid. When TA was absent from the competitor, switching on CRISPR-Cas brought a benefit to pKJK5 (Fig 2D; competitive ratio = 1.14 ± 0.13; p < 0.001). However, when the competitor plasmid’s TA was switched on, switching on CRISPR-Cas was clearly detrimental to pKJK5 (competitive ratio = −2.68 ± 0.10; p < 0.001). In line with these observations, statistical modeling confirmed that TA status was not a significant determinant of plasmid competition outcome in defence (p = 0.40 modeling coefficient of GLM; see Methods for details), but only in offence (p = 0.00053). CRISPR-Cas status contributed significantly to plasmid competition outcome in both plasmid hosts (p < 2*10^-16 in defence; p = 0.0059 in offence).

Next, we tested whether these results were upheld in a more complex model system containing a third mating partner, which was initially plasmid-free (S2 Fig). In this expanded model system, plasmid behavior within the defensive and the offensive CRISPR-Cas host followed the established behavior and CRISPR-Cas was detrimental when competing with a TA-on plasmid in offence, but not in defence (S2C and S2D Fig; S1 Text). In the third, initially plasmid-free host, CRISPR-Cas could act defensively and offensively. Switching on CRISPR-Cas on its own benefitted pKJK5 (S2E Fig; competitive ratio 0.73 ± 0.1; p = 0.0019). When TA was switched on alongside this, there was no clear winner of the competition (competitive ratio = −0.21 ± 0.096; p = 0.1), however, CRISPR-Cas still provided a benefit to pKJK5 relative to the scenario where CRISPR-Cas was switched off, where TA plasmid RP4 won the competition (competitive ratio = −0.94 ± 0.066; p < 0.001).

In addition, to determine whether our observed effects are driven by RP4’s parABCDE TA system rather than other RP4-encoded mechanisms such as entry exclusion or incompatibility, we verified our conclusions by cloning the par operon onto a series of cloning vectors with different backbones of varying incompatibility group and offensively competing pKJK5 against these (S3 Fig, S1 Text). This confirmed that CRISPR-Cas on its own benefits pKJK5, but if a competitor plasmid carries the par TA system, CRISPR-Cas is detrimental to pKJK5 in offense—regardless of competitor plasmid backbone.

Together, our experiments show that a defensive CRISPR-Cas system was always beneficial to its host plasmid—regardless of competitor TA status—in agreement with our core modeling predictions. These results were upheld during a mixed mode of CRISPR-Cas action, where CRISPR-Cas was never detrimental. In contrast, we found that the benefit of an offensive CRISPR-Cas system to its host plasmid depended on payload genes encoded on competitor plasmids; namely, a TA system on a competitor means that offensive CRISPR-Cas was detrimental to its plasmid.

Plasmid-borne CRISPR-Cas systems preferentially target plasmids lacking TA systems

Our model and experiments suggest that the offensive benefits of plasmid-borne CRISPR-Cas systems are limited to conflicts where the competitor plasmid does not encode a TA system. To test whether this constraint is reflected in natural systems, we analyzed genomic data to assess the association between CRISPR-Cas targeting and TA presence. If the fitness benefit of CRISPR-Cas systems in plasmid competition is limited by TA systems, we hypothesized that naturally occurring plasmid-borne CRISPR-Cas systems would preferentially target plasmids without TA systems.

To answer this question, we analyzed existing data on the plasmid-targeting spacers of plasmid-borne CRISPR-Cas systems [6], focusing on identifying which of these targeted plasmids encoded TA systems. The resulting targeting network provided a comprehensive picture of plasmids that were targeted by plasmid-borne CRISPR-Cas systems (N = 1,340), and the TA systems these targeted plasmids encoded (Fig 3A). Type II TA, including parDE (N = 565), were most commonly identified. Within this network, 34% of plasmid-borne CRISPR-Cas systems were Type IV systems (S5A Fig), which function by repressing target gene expression, rather than causing nucleolytic cleavage [8,28]. This mode of action may allow targeted plasmids to persist without triggering PSK, potentially obscuring the relationship between targeting and TA presence. Indeed, Type IV-A1 and IV-A3-targeted plasmids have been shown to persist when targeted in non-essential regions [8,29]. Therefore, we repeated the analysis by partitioning the dataset into Type IV and non-Type IV systems. Restricting the data to non-Type IV systems reduced the number of targeted plasmids to 463, of which only 11% (N = 51) encoded TA systems, and none encoded parDE (S5B Fig).

Correlational analyses on the partitioned data (all CRISPR-Cas types, Type IV only, and non-Type IV) yielded contrasting results. In all cases, Pearson’s χ² tests confirmed a significant relationship between CRISPR-Cas targeting and TA presence (χ² = 94.08, 334.48, and 171.83, respectively; p < 2.2 × 10⁻¹⁶), with small effect sizes (φ = 0.073, 0.137, and 0.098, respectively). When considering all plasmid-borne CRISPR-Cas types or Type IV only, TA-encoding plasmids were targeted more often than expected under the null hypothesis of no relationship between CRISPR-Cas targeting and TA carriage (residuals = 7.21 and 13.82; p = 5.71 × 10⁻¹³ and 1.04 × 10⁻⁴³, respectively; Fig 3B). This contradicts our hypothesis that plasmid-borne CRISPR-Cas systems should preferentially target plasmids lacking TA systems. In contrast, when we excluded Type IV CRISPR-Cas systems, the association between CRISPR-Cas targeting and TA carriage reversed: fewer TA-encoding plasmids were targeted than expected (residual = −10; p = 1.49 × 10⁻²³), consistent with our experimental data and model predictions.

Having established that non-Type IV plasmid-borne CRISPR-Cas systems preferentially target plasmids lacking TA systems, we next considered the few cases where they do target TA-encoding plasmids. Based on the expectation that such encounters would trigger PSK, we hypothesized that non-Type IV CRISPR-Cas plasmids would be more likely than Type IV CRISPR-Cas plasmids to encode compatible antitoxin families when targeting TA-encoding plasmids. To test this, we compared the frequency of compatible antitoxin families carried by plasmids with non-Type IV versus Type IV CRISPR-Cas systems when targeting TA+ plasmids. Compatible antitoxin families were present in 22% (15/69) of cases of non-Type IV targeting, but only 6% (352/5914) of Type IV cases (Fig 3C). Fisher’s exact test confirmed enrichment in nucleolytic systems (OR = 4.4, 95% CI = 2.3–8, p = 1.3 × 10⁻⁵). A complementary binomial test, using the Type IV frequency (6%) as the empirical null, also showed significant enrichment of compatible antitoxin families in cases of non-Type IV targeting (observed = 22%, 95% CI = 13%–33%, p = 1.1 × 10⁻⁵).

Overall, these results indicate that nucleolytic plasmid-borne CRISPR-Cas systems preferentially target plasmids lacking TA systems. This was observed with a modest effect size, which may be due to an additional propensity of plasmids with nucleolytic CRISPR-Cas systems to carry their own TA systems of the same family as found on target plasmids, hypothetically ameliorating TA activity. These findings support our model and extend our experimental findings to natural systems across diverse CRISPR-Cas and TA system types. We conclude that nucleolytic, plasmid-borne CRISPR-Cas systems provide significant defensive benefits, even in the face of TA. In contrast, their benefits in offence are constrained by the presence of TA systems on competitor plasmids.

Mathematical model

Mathematical modeling was carried out using Wolfram Mathematica v14.1. Equations E5 and E7 (see Fig 1; S2 Text) were imported into R version 4.3.2. Output plots were generated using RStudio version 2023.09.1 + 494 with the following packages: dplyr, ggplot2, and ggnewscale.

In order to visualize our mathematical results in Fig 1, we set parameters as detailed in Table 1.

Plasmids and strains

Plasmids and strains used in this study are summarized in Table 2. E. coli DH5α strains DH5α::CmR, DH5α::GmR, and DH5α::SmR were constructed by electroporating DH5α with pBAM1-Gm, pBAM1-Sm, and pBAM1-Cm [48,49], respectively, and selecting for resistant recombinants; resistance gene insertion was confirmed by PCR and by verifying their respective resistant phenotypes after transfer in the absence of antibiotics.

RP4-targeting pKJK5::csg[aphA99] (‘CRISPR-Cas on’ version of pKJK5) was constructed using λ-red mediated recombineering and pKJK5 as a template, as described in [24] with the exception that cas9, sgRNA (programmed to target aphA with specificity AAATGGGCGCGTGATAAGGT) and gfp were inserted into pKJK5’s intI1 gene. Non-targeting control pKJK5::csg[nt2] (‘CRISPR-Cas off’ version of pKJK5) was previously published as pKJK5::gfp^PL [50] and carries a random non-functional sgRNA.

Vector pOGG99_par was constructed by deleting genes from pOPS0378 [51] to generate a minimal vector: mCherry was removed by digestion with EcoRI, extraction, and re-ligation of the 6,043 bp band. Next, site-directed mutagenesis of aphA was carried out using primers pOGG0_mut_fw and pOGG0_mut_rv and Thermo Scientific’s site-directed mutagenesis kit according to manufacturer’s instructions. In this way, the nucleotide at position 96 within aphA was silently mutated from C to G to bring the gene sequence in line with that found on our laboratory’s version of RP4 and allow targeting by pKJK5::csg[aphA99]. Successful mutation was confirmed by Sanger sequencing of the finished plasmid using primer pOGG0_sequence. pOGG99 was constructed by excising parABCDE from pOGG99_par: pOGG99_par was digested with BlpI and BglII, the 3,664 bp fragment was extracted and religated with annealed and phosphorylated linker oligos (pOGG99_parRemoval_top & btm). This yielded pOGG99, a version of pOGG99_par where parABCDE are replaced with a multiple cloning site module.

Vector pHERD99 was constructed by annealing oligos aphA99PAM_top and aphA99_btm and inserting them into pHERD30T’s HindIII and KpnI restriction sites following standard molecular cloning protocols. Vector pSEVA251-99 was constructed by site-directed mutagenesis of pSEVA251’s aphA gene as described for pOGG99_par above.

Vectors pCDF1b_par, pHERD99_par, and pSEVA251–99_par were constructed by parABCDE amplification using pOGG99_par as a template and primers par_fw and par_rv with high-fidelity Phusion polymerase (NEB) and insertion into pCDF1b, pHERD99’s, and pSEVA251-99’s KpnI restriction site, respectively. This parABCDE operon is 99.96% identical to RP4’s with a single nucleotide mismatch in the non-coding area between parA and parB.

All cloned plasmids were subjected to long-read sequencing by Plasmidsaurus using Oxford Nanopore Technology with custom analysis and annotation. Annotated sequences are deposited on Genbank with IDs listed in Table 2.

Culture conditions

Unless otherwise stated, all strains were cultured shaking 180 rpm in LB broth or statically on LB plates at 37˚C. Where appropriate, cultures or plates were supplemented with the following antibiotics at the following concentrations: Ap—100 µg/mL ampicillin; Cm—25 µg/mL chloramphenicol; Gm—50 µg/mL gentamicin; Km—50 µg/mL kanamycin; Sm—50 µg/mL streptomycin; Tc—12 µg/mL tetracycline; Tmp—10 µg/mL trimethoprim.

Conjugation experiments

In our two-strain competition between pKJK5 and RP4 (Fig 2), we used E. coli DH5α::GmR containing pKJK5::csg and E. coli DH5α::CmR containing RP4 as mating partners. In three-strain competitions (S2 Fig), bacterial hosts used were E. coli DH5α::SmR containing pKJK5::csg, E. coli DH5α::CmR containing RP4, and E. coli DH5α::GmR. In both experiments, all strains also contained empty vector pCDF1b, or pCDF1b_par for TA-off treatments. A control experiment to assess the differential effects of using Ampicillin or Kanamycin to select for RP4 (S4 Fig) was set up in the same way, with the omission of bystander vectors pCDF1b or pCDF1b_par in all strains.

In additional two-strain competitions using vectors (S3 Fig), E. coli DH5α::SmR containing pKJK5::csg was used as donor, and the recipient was E. coli DH5α::CmR containing either RP4, pHERD99, pSEVA251-99, pOGG99, or their par-encoding derivatives.

To turn CRISPR-Cas activity on or off, we used two different pKJK5::csg variants with fixed targeting specificity: pKJK5::csg[aphA99] was programmed to target RP4’s aphA gene, and pKJK5::csg[nt2] encoded a non-functional sgRNA with a random 69-nt target sequence that is not present in the model system (plasmid pKJK5::gfp^PL; [50]).

Single colonies from streak plates of mating partners were suspended in 15–25 mL LB with antibiotics appropriate to maintain their plasmid content (see Table 2) and grown overnight. These T0 cultures were then washed twice with 0.9% (w/v) NaCl and adjusted to OD600 = 0.5. Washed, OD-adjusted cultures were filter-mated in a 1:1(:1) ratio:

Filter matings were carried out using a 12-stream Millipore vacuum pump, sterilized with 70% Ethanol and UV light before and after each batch of filter mating and assembled in a Cat2 biosafety cabinet. For each mating, a 0.22 µm glass microfibre filter (Whatman) was placed onto a vacuum pump position, dampened with 200 µL sterile 0.9% NaCl, and topped with a 0.22 µm cyclopore membrane (Whatman). Fully assembled filter positions were equilibrated by applying a vacuum until 2 mL of 0.9% NaCl were pumped through. Next, 1 mL of OD-adjusted donors and 1 mL of OD-adjusted recipients (or 600 µL of each mating partner in 3-strain competitions) were added to each filter position together with 1 mL 0.9% NaCl, and a vacuum applied until the strain mix was pumped through (N = 5). Control matings always included donor-only, recipient-only (and naïve-strain-only where appropriate), and buffer-only sterility controls and yielded results as expected. Cyclopore membranes were placed cell-side-up onto a 10% LB plate (supplemented with 0.9% NaCl) and incubated at 37 °C for 36 hours. Then, cells were recovered by placing each filter into 3 mL of 0.9% NaCl and vortexed for 15 s. This cell suspension was frozen in 20% (w/v) glycerol at −70 °C and plated onto selective plates with various antibiotic regimes to assess strain prevalence and strains’ plasmid content.

Selective plating

Samples were plated in triplicate in 5 µL droplets in dilutions ranging from 10⁰ to 10⁻⁷ onto LB medium containing different combinations of antibiotics. Cm, and Gm were used to select for DH5α::CmR and DH5α::GmR, respectively, and combining these antibiotics with Tmp and Km allowed growth of only DH5α hosts carrying pKJK5 and RP4, respectively. Colonies were counted at the most appropriate dilution, and GFP fluorescence of pKJK5::csg-containing colonies confirmed stringency of selective regimes.

To assess overall sample composition and plasmid content of the DH5α::SmR host, we plated 50 µL of a 10⁻⁵ dilution of each replicate onto LB plates without selection and suspended 96 colonies of each sample in 250 µL LB each following overnight incubation. Previous assessment of colonies from selective plates sub-cultured on non-selective media indicated that RP4 and pKJK5::csg did not become lost over this timeframe in the absence of selection. The 96 clones picked from each replicate were then replica-stamped onto plates containing no selection, Cm, Gm, Km, and/or, Tmp. Clones growing on LB, but not LB supplemented with Cm or with Gm were scored as DH5α::SmR clones, and the Km/Tmp selective plates were used to score their plasmid content. Resistance to Km is conferred by aphA, which is the target of the CRISPR spacer on pKJK5::csg. Plasmid of escape of CRISPR-Cas by target mutation is unusual [56], and in a control experiment, we accordingly found selection using Km to be robust to assess RP4 presence (S4 Fig).

Plasmid competitive ratio

We calculated pKJK5 and RP4’s competitive ratio (log odds ratio) as follows:

This gives a measure of which plasmid is more frequent in a certain host and therefore won the competition for the host at the end of the experiment. Positive competitive ratios indicate pKJK5 winning the competition, negative ratios indicate RP4 winning, and ratios ≈ 0 indicate equal carriage of pKJK5 and RP4 in the host.

Statistical analyses

Colony counts on all selective plates from all experiments, together with plasmid competitive ratios used to produce figures are available in S1 Data.

Data processing, data visualization, and statistical analyses were carried out using R version 4.3.2 and RStudio version 2023.09.1 + 494 with the following packages: flextable, dplyr, readr, tidyr, ggplot2.

Bioinformatic analysis of plasmid TA carriage

The sequences of 17,828 de-replicated plasmid found across prokaryotic genomes [6] were retrieved from NCBI using rentrez [57]. CRISPR-Cas presence and targeting information were adopted from [6]. Plasmid open reading frames were predicted with prodigal v2.6.3 [58]. TA nucleotide or protein sequences were retrieved from TADB v3.0 [14]. parDE and other well-studied PSK systems were identified manually based on a recent review [13]. To identify complete TA systems, plasmids were searched with BLAST v2.16.0 [59] and hits were filtered for cases where both a toxin and antitoxin component were directly adjacent. For the analysis of TA systems carried by CRISPR-Cas bearing plasmids, full TA systems were considered alongside orphan antitoxin genes, and a >70% alignment coverage in comparison with TADB was considered as a match. Counts of CRISPR loci, Cas loci, TA systems, and the network of TA families on plasmids carrying or targeted by CRISPR-Cas systems are available in S1 Data. Statistical analyses were run using R v4.4.2 base functions.

S1 Fig. Mean ± standard error of the proportion of Streptomycin resistant Colony Forming Units (CFU) after mating was never significantly different from 1 (T test followed by Bonferroni adjustment for multiple testing; Table 7).

Additionally, when stamp plating, only two individual clones out of 2,574 were found to be streptomycin sensitive, one of which was from a single-host control treatment. Vector presence was confirmed by PCR in all tested DH5α::SmR colonies. The data underlying this Figure can be found in S1 Data; the competitive outcome for this experiment is presented in S2 Fig.

https://doi.org/10.1371/journal.pbio.3003658.s001

(TIF)

S2 Fig. Plasmid competition results were upheld in a three-strain model system.

(A) CRISPR-Cas and TA activity were tweaked on competitor plasmids in a binary manner by switching each system on or off. pKJK5::csg carries a gene cassette encompassing cas9, sgRNA, GFP with either an RP4-targeting or non-targeting guide, RP4 naturally carries TA system parABCDE. (B) Plasmids were competed using three Escherichia coli DH5α hosts with different plasmid content, allowing us to assess competitive outcome under defensive, offensive, or no clear mode of CRISPR-Cas action. After filter mating, plasmid content of each host was assessed by selective plating. (C–E) Mean ± standard error of the competitive ratio (log odds ratio of pKJK5-carrying hosts/RP4-carrying hosts) describes outcome of plasmid competition (N = 5). Values >0 indicate CRISPR-Cas plasmid pKJK5 winning the competition, and values <0 indicate TA plasmid RP4 winning the competition for a certain host. The dashed line indicates a neutral outcome with competitive ratio = 0. Data are presented for treatments in which CRISPR-Cas and TA activity were toggled on or off in all combinations. Opacity of datapoints indicate number of replicates used to calculate means, this affects panel C TA-off data (N = 1) and panel C TA-on and CRISPR-Cas-on data (N = 4). Stars indicate significant differences from 0 as assessed by T test and Bonferroni adjustment for multiple testing; *p < 0.005 with α = 0.005; see Table 4 for all p values. na2014not assessed due to N = 1 after removal of missing values (no carriage of TA plasmid RP4 recorded in 4 out of 5 replicates). The data underlying this Figure can be found in S1 Data. See S1 Text for additional information.

https://doi.org/10.1371/journal.pbio.3003658.s002

(TIF)

S3 Fig. Outcome of competition with cloning vectors.

Mean ± standard error of the competitive ratio (log10 of pKJK5-carrying hosts/competitor-carrying hosts) describes the outcome of plasmid competition (N = 5). Values <0 indicate CRISPR-Cas plasmid pKJK5 winning the competition, and values >0 indicate the TA competitor plasmid winning the competition for a DH5α host where CRISPR-Cas was acting offensively. Data are presented for treatments in which CRISPR-Cas and TA activity were toggled on or off in all combinations. RP4 competition outcome (A, B) in this model system matched the outcome observed earlier (Fig 2B). Relative differences between treatments are indicated with gray lines and were assessed by Tukey’s HSD after fitting individual Generalized Linear Models; see Methods and Tables 6 and 7 for model details and p values. *p < 0.05; ***p < 0.001; n.s. not significant p > 0.63. The data underlying this Figure can be found in S1 Data. See S1 Text for additional information.

https://doi.org/10.1371/journal.pbio.3003658.s003

(TIF)

S4 Fig. Assessed RP4 proportion is largely robust to antibiotic used to select for RP4.

Mean and standard error of RP4 content of hosts in pilot mating experiment lacking a bystander plasmid assessed by ampicillin or by kanamycin, N = 5. Data are presented for CRISPR-Cas switched on or off. The data underlying this Figure can be found in S1 Data. See S1 Text for additional information.

https://doi.org/10.1371/journal.pbio.3003658.s004

(TIF)

S5 Fig. Plasmid-borne CRISPR-Cas systems and their targets.

(A) CRISPR-Cas systems encoded on plasmids; split into distribution of Cas operon and CRISPR array types. Type IV systems are highlighted to emphasize their predominance. (B) The total number of plasmids encoding each TA family, highlighting parDE and other well-studied post-segregational killing (PSK) systems [13]. The other bar charts show the proportion of plasmids encoding each TA family targeted by plasmid borne Type IV and non-Type IV CRISPR-Cas systems. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003658.s005

(TIF)

S6 Fig. Contingency tables of expected versus observed values for the association between plasmid-borne CRISPR-Cas targeting and TA carriage.

Contingency tables showing the relationship between plasmid targeting and TA presence. The tables display the observed counts of plasmids with or without TA systems and whether they are targeted or not by CRISPR spacers in another plasmid, either for all CRISPR-Cas types (A), Type IV CRISPR-Cas only (B), or for non-Type IV CRISPR-Cas (C). The bar plots show the sum of counts for each vertical or horizontal group. Facets in the table are coloured by standardized residual after Pearson’s χ² test, indicating magnitude and directionality of association (red—less than expected, blue—more than expected). The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003658.s006

(TIF)

S2 Text. Supplementary mathematical analysis.

Development of the mathematical model (Fig 1) including derivation of equations used, and mathematical analysis of the impact of changing competitive traits x and y on plasmid transition rates and the outcome of plasmid competition.

https://doi.org/10.1371/journal.pbio.3003658.s008

(DOCX)