Plasmids link antibiotic resistance genes and phage defense systems in E. coli

10 Phage therapy has been proposed as an alternative to antibiotics to treat resistant 11 infections. However, we have a limited understanding of how antibiotic resistance genes 12 (ARGs) associate with bacterial phage defen se systems (PDSs). Here, we explore the 13 relationship between ARGs and PDSs in a sample of 2,559 plasmids originating from 14 1,044 E. coli isolates, representing a snapshot of clinical and non -clinical diversity in 15 Oxfordshire, UK (2008 -2020). In total, we identify 3,193 ARGs and 14,013 PDSs (180 16 unique types). We demonstrate that E. coli plasmids are enriched for ARGs and PDSs 17 (both p<0.001), with a bias towards toxin -antitoxin, abortive infection , and restriction-18 modification systems (all q<0.025). We proceed to show that ARGs and PDSs are 19 physically linked by plasmids (p<0.001). Together, our results suggest that phage therapy 20 may inadvertently select for antibiotic resistant bacteria , and that antibiotic use may 21 similarly drive resistance to phage. 22 23

Escherichia coli; bacteriophage; anti-bacterial agent 24 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 25 26 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint

27 Bacterial defense systems can protect the cell against invaders, such as plasmids and phage 1. 28 Plasmids are extrachromosomal replicons often able to transfer between bacterial hosts via 29 conjugation2. Phage, and in particular, lytic phage, are bacterial viruses that, like plasmids, rely 30 on the host cell for replication, but unlike plasmids, ultimately kill the host during the infection 31 cycle3. Plasmids are often associated with antimicrobial resistance genes (ARGs), which can 32 confer resistance to the host cell, sometimes at a low fitness cost 4,5. Moreover, plasmids can 33 also encode phage defen se systems (PDSs), such as toxin -antitoxin, restriction modification, 34 and CRISPR -Cas systems, blurring the boundary between selfish genetic elements and 35 mutualistic partners6–9. 36 One unresolved issue is the extent to which ARGs and PDSs co -occur on plasmids. This is 37 important because the physical linkage of different resistance determinants can drive co -38 selection: selective pressure for one trait (phage resistance) may inadvertently maintain another 39 (antibiotic resistance), and vice versa. Co-selection is already well documented for ARGs and 40 other resistance genes, such as those against heavy metals or biocides, where a single stressor 41 can select for multi -resistant bacteria10,11. Some examples have been reported of ARGs and 42 PDSs co-localising on the same mobile genetic element, yet whether this occurs more generally 43 remains to be seen12–14. Understanding these patterns is particularly relevant as phage therapy 44 gains traction as a strategy to combat antibiotic resistant infections15,16. 45 In this study, we investigate d the relationship between ARGs and PDSs in a large sample of 46 2,559 plasmids from 1,044 clinical and non -clinical E. coli isolates17. We first annotate the 47 genomes for ARGs and PDSs, then explore (i) their overall enrichment on plasmids versus 48 chromosomes, (ii) the enrichment of specific PDS types on plasmids versus chromosomes, and 49 (iii) their linkage on plasmids. 50 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 51 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint

52 Dataset curation 53 Isolates and associated genome assemblies were from an existing, pre-collated dataset17. The 54 original dataset contained Enterobacterales species collected from within 60km in Oxfordshire 55 from 2008 -2020, covering multiple clinical and non -clinical niches: human bloodstream 56 infection (BSI), livestock faeces (cows, pigs, poultry, and sheep), and wastewater (influent, 57 effluent, and rivers upstream of effluent) 18,19. All BSI isolat es and a majority of non -BSI 58 isolates were not selectively cultured for antimicrobial resistance . Species were determined 59 with matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF). Authors used a 60 hybrid approach (short- and long-reads) to recover reference-grade assemblies. We selected all 61 the E. coli genomes where the chromosome and all plasmids were fully circularised (n= 1,044). 62 Isolate and assembly metadata was also downloaded. 63 Genome annotation 64 We annotated for PDSs using the tool DefenseFinder (v. 2.0.0) with default paramaters, which 65 queries against a database of known PDSs20–22. DefenseFinder also denotes when a system is 66 “complete” (“systems” output), meaning all component genes have been annotated. We only 67 considered complete systems in this study. For ARGs, we used pre-existing annotations from 68 AMRFinderPlus (v. 3.10.18) with default parameters 23,24. Plasmid mobility predictions 69 (conjugative, mobilisable, or non-mobilisable) also used pre-existing annotations from MOB-70 typer from MOB -suite (v. 3.03) with default parameters. We annotated for integrons using 71 IntegronFinder 2.0 with default parameters 25. For gene annotations we used Prodigal (v. 72 2.6.3)26 with default parameters. 73 74 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint Testing for the enrichment of ARGs and PDS genes on plasmids 75 We first calculated the total number of plasmid genes as a fraction of the total number of genes 76 in the genome. Similarly, we calculated the fraction of just ARGs and PDSs genes on plasmids. 77 These proportions were compared using a Pearson’s chi -squared test with Yates’ continuity 78 correction. We conducted this across the whole dataset (see Results), but also separately for 79 BSI-associated and non-BSI-associated plasmids (Figure S5). 80 Testing for the over-representation of PDS families on plasmids 81 First, for each PDS type, we calculated a binary presence/absence score for each replicon. We 82 next wanted to generate a null distribution that accounted for some genomes containing more 83 plasmids than others. To do this, we performed b=10,000 label permutations of replicon type 84 (plasmid or chromosome) within each genome. After each permutation, we recorded the 85 number of plasmid -labelled contigs carrying each PDS. This generated a one -sided p-86 value=(n+1)/(b+1), where n was the number of permutations with a plasmid count at least as 87 extreme as observed (≥ for the plasmid‑enrichment test; ≤ for the chromosome‑enrichment test) 88 and b=10,000. Separate p‑values were obtained for plasmid (H₁: observed > null) and 89 chromosome (H₁: observed < null) bias. To mitigate against multiple tests increasing the rate 90 of Type I errors, we applied both the Benjamini–Hochberg (BH) and the Benjamini–Yekutieli 91 (BY) false‑discovery‑rate procedures. Families with BH‑adjusted q < 0.025 for the relevant 92 one‑sided test (plasmid or chromosome) were called significant; BY‑adjusted q‑values are also 93 reported when q < 0.025 (providing a stricter, dependence‑robust correction). 94 Testing the association between ARGs and PDSs on plasmids 95 If two plasmids are vertical descendants, or share large portions of their sequence from a 96 recombination event, they might contain many of the same genes. This can be problematic 97 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint when testing for the association of ARGs and PDSs, as it introduces pseudo -replicates to the 98 model. We minimised the risk of pseudo -replication by genetically dereplicating plasmids in 99 the dataset. We first took the k-mer plasmid clusters generated by the original authors 17. We 100 then randomly selected a representative from each cluster. In total, we recovered n=712 101 genetically representative plasmids. 102 We next predicted the count of ARGs per plasmid with a generalised additive mixed -model 103 (GAMM) using the mgcv library27. Model choice was guided by Akaike Information Criterion 104 (AIC), which balances goodness-of-fit and parsimony. We found that a Tweedie family (log-105 link) GAMM with a log-link family outperformed analogous GAMM/GLMMs with Poisson 106 and negative binomial families. 107 The model used the fixed effects (i) plasmid PDS count, (ii) plasmid GC content (mean 108 centred), and (iii) sampling niche (BSI -associated, livestock -associated, or wastewater -109 associated). We used plasmid replicon type as a random intercept to control for plasmid 110 lineage. We also included an offset of log 10-scaled plasmid length (bp). Originally, we also 111 included plasmid integron count and plasmid mobility (non -mobilisable, mobilisable, 112 conjugative) as fixed-effects, but these were dropped since they were insignificant (likelihood-113 ratio test p-value>0.05). Plasmid PDS count and GC content initially used thin‑plate regression 114 splines (k = 4), but by assessing the effective degrees of freedom and plots of the smooth terms, 115 we concluded they were consistent with linearity, so the smooth terms were dropped. 116 Lastly, we checked for multicollinearity and correlations among predictors. For numeric-117 numeric fixed -effect pairs, we computed Spearman’s rank correlation , and for numeric-118 categorical fixed-effect pairs, we calculated the Kruskal -Wallis effect size . No issues were 119 identified. 120 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint Of the n=712 genetically representative plasmids, 23% (167/712) had inconclusive plasmid 121 replicon typing by PlasmidFinder (recorded as “NA”). Nonetheless, re -running the GAMM 122 with these plasmids excluded did not meaningfully change the coefficient estimates (95% CIs 123 overlapped). 124 Code and data availability 125 The data and metadata used in this study will be stably archived on Zenodo. The code to 126 reproduce the analysis will be made available on Git Hub, and also archived on Zenodo. 127 Supplementary File 1 contains the permutation test output. Supplementary File 2 contains the 128 GAMM diagnostics. 129 130

131 Plasmids are enriched for antimicrobial resistance genes and phage-defense system genes. 132 We first curated a sample of n=1,044 E. coli isolates, a common human commensal and 133 pathogen, with n=2,559 plasmids (see Methods). Isolates were sampled within a geographically 134 (<60km in Oxfordshire, UK) and temporally (2008 -2020) restricted frame , and represented 135 both clinical (human bloodstream infection [BSI]) and non -clinical (livestock-associated and 136 wastewater-associated) niches, totaling 52% (547/1,044) BSI, 42% (433/1,044) livestock -137 associated, and 6% (64/1,044) wastewater -associated. Livestock-associated isolates were 138 further divided into 32.8% (142/433) cattle-associated, 28.9% (125/433) pig-associated, 11.8% 139 (51/433) poultry-associated, and 26.5% (115/433) sheep-associated. In total, 89% (938/1,044) 140 of isolates carried a plasmid. Isolates carried a median of 2 plasmids (range : 0-16), with the 141 most common replicon types being small colicinogenic-type plasmids (see Methods). 142 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint We next annotated our genomes for ARGs and PDSs (see Methods). In total, we identified 143 n=14,013 PDS hits with 180 distinct system types. Every chromosome carried at least one PDS, 144 contrasted with only 41% (1,051/2,559) of plasmids. Comparatively, ARGs were less common, 145 with 3,193 identified in total. Overall, 27% (286/1,044) of chromosomes and 20% (521/2,559) 146 of plasmids carried at least one ARG, respectively. Of plasmids that carried at least one ARG 147 or PDS, the three top replicon types were IncF -types: IncFIB, Col156 -IncFIB-IncFII, and 148 IncFIB-IncFIC, which together accounted for 27% (325/1,225) of that plasmid subset. 149 Distributions of ARGs and PDSs were variable across E. coli niches (Figures S1 –S4). 150 Livestock-associated plasmids (cattle, pig, poultry, and sheep) carried defense systems in 44% 151 (475/1,080) of cases but carried ARGs in only 13% (143/1,080). Within this group, pig -152 associated plasmids were distinct: they encoded ARGs roughly four-fold more frequently: 25% 153 (101/403) of pig plasmids versus 6% (42/677) of plasmids from the other three livestock 154 groups, while their PDS prevalence remained comparable (39% [158/403] versus 47% 155 [317/677]). Wastewater-associated plasmids showed intermediate prevalence: 35% (53/150) 156 encoded PDSs and 19% (29/150) carried ARGs. BSI-associated plasmids exhibited the greatest 157 combined load outside pigs, with 39% (523/1,329) containing PDSs and 26% (349/1,329) 158 carrying ARGs. 159 Next, we tested whether ARGs and PDSs were enriched on plasmids versus chromosomes 160 (Figure 1; see Methods). Of the n=4,999,278 annotated coding sequences in our dataset, only 161 3.42% (170,778/4,999,278) were found on plasmids. By contrast, 8.66% of all PDS genes 162 (2,474/28,581) and 75.76% of all ARGs (2,419/3,193) were plasmid-encoded. Pearson’s χ² 163 tests (Yates -corrected) confirmed that both categories were highly over -represented on 164 plasmids relative to the genomic baseline (ARGs: χ² = 49,943.4, df = 1, p < 0.001; PDSs: χ² = 165 2,343.8, df = 1, p < 0.001). The magnitude of enrichment was therefore ~2.5-fold for PDS 166 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint genes and ~22-fold for ARGs. Surprisingly, this trend carried for both clinical and non-clinical 167 isolates (Figure S5). 168 169 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 170 171 Figure 1. Plasmids are hotspots for ARGs and PDSs in E. coli. Bars show the percentage of genes in that category residing on plasmids. Both comparisons were significant at the p<0.001 level (Pearson χ² tests with Yates correction). .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint Plasmids and chromosomes are associated with distinct repertoires of phage-defense 172 systems. 173 We next explored whether the types of PDSs varied between plasmids and chromosomes. We 174 conducted a permutation test for each PDS (see Methods) and found 8% (15/180) of the PDS 175 families preferentially reside d on plasmids (Figure 2) . These 15 families made up 42.46% 176 (1,343/3,163) of defense-system occurrences on plasmids . In contrast, 20% ( 36/180) PDSs 177 were associated with chromosomes (Figure S6). Tests for the remaining 72% (129/180) PDSs 178 were inconclusive. 179 We found that TIR-III and Pif systems and two small uncharacterised families (HEC -09, DS-180 28) were exclusively associated with plasmids in this dataset . Conversely, toxin-antitoxin 181 module MazEF and the cyclic-oligonucleotide producer CBASS were enriched more modestly 182 (~0.25, Figure 2A). When enrichment was quantified as absolute surplus (Figure 2B), the toxin-183 antitoxin complex Mok-Hok-Sok was an outlier, occurring on plasmids 661 times versus the 184 approximately 100 occurrences expected by chance. Progressively smaller and significant 185 surpluses were observed for the remaining systems. 186 187 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 188 189 Figure 2. Defense systems associated with plasmids. (A) The x-axis shows the proportion of contigs that carried the PDS type. Each point represents a PDS type whose frequency on plasmids was greater than expected when contig labels (plasmid/chromosome) were randomly shuffled within isolates under two one-tailed permutation tests (see Methods). Open grey circles indicate p<0.05 significance under Benjamini -Hochberg (BH) correction only. Filled blue circles indicate significance under both BH and Benjamini -Yekutieli (BY) correction, which does not assume independence. The dashed line marks the genome -wide plasmid share of defence systems ( π̄ ≈ 0.16). (B) Bars count the excess number of plasmid-labelled contigs carried by each type (“observed – expected”, where the expectation equals the type’s total contig count × π̄ ). Bars are coloured the same as points in (A). Only PDS families that were significantly over-represented on plasmids in the within-isolate permutation test are shown. .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint Phage-defense system count predicts ARG gene count on plasmids independent of plasmid 190 size. 191 Lastly, we explored the distribution of ARGs and PDSs within the genomes. We tested whether 192 ARGs and PDSs co-occurred on the same pl asmids by constructing a generalised additive 193 mixed-model (GAMM) with a Tweedie family and log-link (see Methods). Briefly, we used 194 plasmid ARG count as a response with fixed effects (i) plasmid PDSs count, (ii) plasmid GC 195 content (mean centred), and (iii) sampling niche (BSI -associated, livestock -associated, or 196 wastewater-associated). We used plasmid replicon type as a random intercept to control for 197 plasmid lineage. To avoid pseudo -replication of plasmid sequences, we used single 198 representatives from clusters generated by genetic similarity. We also included an offset of 199 log10-scaled plasmid length (bp). Our total sample size was n=712 genetically representative 200 plasmids. The model output is detailed in Table 1 and visualised in Figure 3a. Further, Figure 201 3b visualises the ARG-PDS co-occurrence heatmap for all plasmids (n=2,559) in the dataset. 202 We found that the count of PDSs was a strong positive predictor of ARG count: each additional 203 phage-defense system increased the chance of an ARG being present on a plasmid by 48% 204 (95% CI=[18%, 86%]; p<0.001). Similarly, each 1% increase in plasmid GC content above 205 average increased ARG count by 37% (95% CI=[28%, 48%]; p<0.001). We found no signal 206 from isolate sampling niche. The random‐effect smooth term for plasmid replicon type was 207 consistent with substantial non -linear variation (edf = 2 5.7, F = 1.21, p<0.001). Overall, the 208 model demonstrated good fit with adjusted R2=0.40. 209 All isolates (1044/1044) carried a chromosomal PDS, meaning that all isolates with plasmid -210 borne ARG (429/1,044) had the potential for dual resistance. Additionally, of the 41% 211 (429/1,044) of isolates carrying an ARG-positive plasmid, 88% (378/429) also carried at least 212 one plasmid-borne PDS. In 75% (324/429) of these cases, they were both on the same plasmid 213 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint (Figure 4). Further, Figure S7 shows the relative frequencies of ARGs, PDSs, and their co -214 localisation on plasmids, and Figure S8 shows how these frequencies vary with plasmid size in 215 clinical versus non-clinical E. coli, consistent with clinical isolates, unlike non-clinical isolates, 216 often carrying small plasmids (<13kbp) that link ARGs and PDSs. 217 218 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 219 220 Figure 3. PDSs and ARGs co-occur on plasmids. (A) Forest plot of GAMM predictor coefficients as fold- changes with 95% confidence intervals (log10-scaled axis). The vertical dashed line at x=1 marks the reference. (B) Heatmap of frequency of co-localisation of ARGs and PDSs on same plasmid (n= 347). Cell colour indicates count. .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 221 222 Covariate Coefficient (95% CI) Odds ratio (95% CI) p-value (Intercept) -6.58 (-7.25, -5.91) 0.00 (0.00, 0.00) <0.001 PDS count 0.39 (0.16, 0.62) 1.48 (1.18. 1.86) <0.001 Plasmid GC content (mean-centred) 0.32 (0.24, 0.39) 1.37 (1.28, 1.48) <0.001 Niche BSI-associated 0.72 (0.19, 1.25) 2.05 (1.21, 3.48) 0.008 Wastewater-associated 0.48 (-0.43, 1.39) 1.61 (0.65, 4.00) 0.3 Table 1. Parameter estimates for the GAMM. For niche, livestock-associated was the reference level. Odds ratios were calculated by taking the exponent. .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 223 224 Figure 4. Overlap of ARGs and PDSs on plasmids within E. coli isolates. Euler diagram illustrating the intersection between sets of E. coli isolates that carried (i) at least one plasmid with a PDS, (ii) at least one plasmid with an ARG, and (iii) at least one plasmid with both an ARG and a PDS. The size of each ellipse is proportional to the set size. The total sample size was n=884 isolates. .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint

225 We have analysed the distribution of ARGs and PDSs in a diverse sample of E. coli genomes. 226 We demonstrate that plasmids are enriched for both types of features, disproportionately carry 227 specific PDS families compared to the chromosome, and physically link ARGs and PDSs in 228 the genome. Overall, our findings suggest a fundamental role for plasmids in the dissemination 229 of PDSs on top of known associations with ARGs. 230 Our study has limitations. Whilst E. coli represents a predominant clinical pathogen, plasmids 231 are widespread across bacterial taxa and frequently cross species boundaries 28. Future work 232 should expand these modelling approaches to incorporate a broader range of bacterial 233 hosts. We also did not model the co -occurrence of specific ARGs and PDSs. Capturing such 234 associations require s modelling f rameworks that account for both vertical inheritance and 235 horizontal transfer events (such as those mediated by transposons) as well as methods capable 236 of resolving multivariate networks of association. Without such controls, pairwise tests risk 237 conflating indirect correlations with direct genetic linkage, as well as inflating Type I errors29. 238 The co-localisation of ARGs and PDSs on plasmids can facilitate joint horizontal transfer and 239 coordinated regulation. We hypothesise that plasmids are an ideal platform for co -localising 240 ARGs and PDSs due to their variable copy number and potential for rapid evolution. In clinical 241 isolates, multicopy plasmids carrying blaTEM-1 can amplify phenotypic resistance and 242 accelerate adaptive evolution of the gene 30,31. In our dataset, blaTEM-1 was often co -localised 243 with toxin -antitoxin systems such as RnlAB, PARIS, and AbiQ 32–34on small (>13kb) 244 colicinogenic-type plasmids in the BSI-associated isolates . Analogously, plasmids carrying 245 PDSs may experience rapid and fluctuating selection in response to episodic lytic phage 246 attacks, favouring those that can quickly up or downregulate their PDSs35,36. 247 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint Phage pressure alone can maintain mobile genetic elements carrying ARGs even in the absence 248 of antibiotics. For example, in Vibrio cholerae, episodic phage challenge increased conjugation 249 frequencies of integrative conjugative elements carrying both ARGs and PDSs 14. In another 250 study, E. coli IncF plasmids carrying antibiotic resistance reduced susceptibility to coliphage 251 infection and persisted for around 10 days without antibiotic selection 37. This suggests that 252 simply curbing antibiotic use may be insufficient to reverse the selection for antibiotic and/or 253 phage resistance. Future work should monitor plasmid gene frequencies, copy‑number 254 dynamics, and host‑fitness effects under isolated antibiotic, isolated phage, and combined 255 selection regimes, ideally incorporating varied dosing schedules and temporal patterns. 256 257 Funding 258 This work was supported by the UKRI Frontiers grant (EP/Y031067/1) to RCM . WS is 259 supported by the Wellcome Trust grant (218514/Z/19/Z), Merck Sharp and Dohme Corp., and 260 Janssen Pharmaceutica NV. PJ is supported by funding from the Biotechnology and Biological 261 Sciences Research Council UKRI-BBSRC grant (BB/T008784/1). 262 263

264 The authors thank Rachel Wheatley and Liam P. Shaw for the valuable help and discussions. 265 266 Conflicting interests 267 The authors declare no conflicting interests. 268 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 269 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint

270 1. Koonin, E. V., Makarova, K. S. & Wolf, Y. I. Evolutionary Genomics of Defense 271 Systems in Archaea and Bacteria. Annu Rev Microbiol 71, (2017). 272 2. Coluzzi, C., Garcillán -Barcia, M. P., De La Cruz, F. & Rocha, E. P. C. Evolution of 273 Plasmid Mobility: Origin and Fate of Conjugative and Nonconjugative Plasmids. Mol 274 Biol Evol 39, (2022). 275 3. Egido, J. E., Costa, A. R., Aparicio -Maldonado, C., Haas, P. -J. & Brouns, S. J. J. 276 Mechanisms and clinical importance of bacteriophage resistance. FEMS Microbiol Rev 277 46, (2022). 278 4. Partridge, S. R., Kwong, S. M., Firth, N. & Jensen, S. O. Mobile genetic elements 279 associated with antimicrobial resistance. Clinical Microbiology Reviews vol. 31 Preprint 280 at https://doi.org/10.1128/CMR.00088-17 (2018). 281 5. Vogwill, T. & Maclean, R. C. The genetic basis of the fitness costs of antimicrobial 282 resistance: A meta-analysis approach. Evol Appl 8, (2015). 283 6. Siedentop, B., Rüegg, D., Bonhoeffer, S. & Chabas, H. My host’s enemy is my enemy: 284 plasmids carrying CRISPR-Cas as a defence against phages. Proceedings of the Royal 285 Society B: Biological Sciences 291, (2024). 286 7. Picton, D. M. et al. The phage defence island of a multidrug resistant plasmid uses both 287 BREX and type IV restriction for complementary protection from viruses. Nucleic Acids 288 Res 49, (2021). 289 8. Vassallo, C. N., Doering, C. R., Littlehale, M. L., Teodoro, G. I. C. & Laub, M. T. A 290 functional selection reveals previously undetected anti-phage defence systems in the E. 291 coli pangenome. Nat Microbiol 7, (2022). 292 9. Leroux, M. & Laub, M. T. Toxin-Antitoxin Systems as Phage Defense Elements. Annual 293 Review of Microbiology vol. 76 Preprint at https://doi.org/10.1146/annurev -micro-294 020722-013730 (2022). 295 10. Wales, A. D. & Davies, R. H. Co -selection of resistance to antibiotics, biocides and 296 heavy metals, and its relevance to foodborne pathogens. Antibiotics vol. 4 Preprint at 297 https://doi.org/10.3390/antibiotics4040567 (2015). 298 11. Pal, C., Bengtsson -Palme, J., Kristiansson, E. & Larsson, D. G. J. Co -occurrence of 299 resistance genes to antibiotics, biocides and metals reveals novel insights into their co -300 selection potential. BMC Genomics 16, (2015). 301 12. Mahata, T. et al. Gamma-Mobile-Trio systems define a new class of mobile elements 302 rich in bacterial defensive and offensive tools. bioRxiv (2024). 303 13. Botelho, J. Defense systems are pervasive across chromosomally integrated mobile 304 genetic elements and are inversely correlated to virulence and antimicrobial resistance. 305 Nucleic Acids Res 51, (2023). 306 14. LeGault, K. N. et al. Temporal shifts in antibiotic resistance elements govern phage -307 pathogen conflicts. Science (1979) 373, (2021). 308 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 15. BiomX, Inc. BiomX Announces Positive Topline Results from Phase 2 Trial Evaluating 309 BX211 for the Treatment of Diabetic Foot Osteomyelitis (DFO). 310 https://ir.biomx.com/news-events/press-releases/detail/130/biomx-announces-positive-311 topline-results-from-phase-2-trial. 312 16. Chan, B. K. et al. Personalized inhaled bacteriophage therapy for treatment of 313 multidrug-resistant Pseudomonas aeruginosa in cystic fibrosis. Nat Med 31, 1494–1501 314 (2025). 315 17. Matlock, W. et al. Enterobacterales plasmid sharing amongst human bloodstream 316 infections, livestock, wastewater, and waterway niches in Oxfordshire, UK. Elife 12, 317 (2023). 318 18. Shaw, L. P. et al. Niche and local geography shape the pangenome of wastewater -and 319 livestock-associated Enterobacteriaceae. Sci Adv 7, (2021). 320 19. Lipworth, S. et al. Ten-year longitudinal molecular epidemiology study of Escherichia 321 coli and Klebsiella species bloodstream infections in Oxfordshire, UK. Genome Med 13, 322 (2021). 323 20. Tesson, F. et al. Systematic and quantitative view of the antiviral arsenal of prokaryotes. 324 Nat Commun 13, (2022). 325 21. Néron, B. et al. MacSyFinder v2: Improved modelling and search engine to identify 326 molecular systems in genomes. Peer Community Journal 3, (2023). 327 22. Tesson, F. et al. A Comprehensive Resource for Exploring Antiphage Defense: 328 DefenseFinder Webservice,Wiki and Databases. Peer Community Journal 4, (2024). 329 23. Feldgarden, M. et al. AMRFinderPlus and the Reference Gene Catalog facilitate 330 examination of the genomic links among antimicrobial resistance, stress response, and 331 virulence. Sci Rep 11, (2021). 332 24. Robertson, J. & Nash, J. H. E. MOB-suite: software tools for clustering, reconstruction 333 and typing of plasmids from draft assemblies. Microb Genom 4, (2018). 334 25. Néron, B. et al. IntegronFinder 2.0: Identification and Analysis of Integrons across 335 Bacteria, with a Focus on Antibiotic Resistance in Klebsiella. Microorganisms 10, 336 (2022). 337 26. Hyatt, D. et al. Prodigal: Prokaryotic gene recognition and translation initiation site 338 identification. BMC Bioinformatics 11, (2010). 339 27. Pedersen, E. J., Miller, D. L., Simpson, G. L. & Ross, N. Hierarchical generalized 340 additive models in ecology: An introduction with mgcv. PeerJ 2019, (2019). 341 28. Redondo-Salvo, S. et al. Pathways for horizontal gene transfer in bacteria revealed by a 342 global map of their plasmids. Nat Commun 11, (2020). 343 29. Kim, P. J. & Price, N. D. Genetic co -occurrence network across sequenced microbes. 344 PLoS Comput Biol 7, (2011). 345 30. San Millan, A., Escudero, J. A., Gifford, D. R., Mazel, D. & MacLean, R. C. Multicopy 346 plasmids potentiate the evolution of antibiotic resistance in bacteria. Nat Ecol Evol 1, 347 0010 (2016). 348 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint 31. Matlock, W. et al. E. coli phylogeny drives co -amoxiclav resistance through variable 349 expression of blaTEM-1. Preprint at https://doi.org/10.1101/2024.08.12.607562 (2024). 350 32. Deep, A., Liang, Q., Enustun, E., Pogliano, J. & Corbett, K. D. Architecture and 351 infection-sensing mechanism of the bacterial PARIS defense system. (2024) 352 doi:10.1101/2024.01.02.573835. 353 33. Koga, M., Otsuka, Y., Lemire, S. & Yonesaki, T. Escherichia coli rnlA and rnlB 354 compose a novel toxin-antitoxin system. Genetics 187, (2011). 355 34. Pecota, D. C. & Wood, T. K. Exclusion of T4 phage by the hok/sok killer locus from 356 plasmid R1. J Bacteriol 178, (1996). 357 35. Hall, A. R., Scanlan, P. D., Morgan, A. D. & Buckling, A. Host-parasite coevolutionary 358 arms races give way to fluctuating selection. Ecol Lett 14, 635–642 (2011). 359 36. Betts, A., Gifford, D. R., MacLean, R. C. & King, K. C. Parasite diversity drives rapid 360 host dynamics and evolution of resistance in a bacteria -phage system. Evolution (N Y) 361 70, 969–978 (2016). 362 37. Montelongo Hernandez, C., Putonti, C. & Wolfe, A. J. Urinary Plasmids Reduce 363 Permissivity to Coliphage Infection. Microbiol Spectr 11, (2023). 364 365 .CC-BY-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666796doi: bioRxiv preprint