Extended-spectrum beta-lactamase antibiotic resistance plasmids have diverse transfer rates and can be spread in the absence of selection

Horizontal gene transfer, mediated by conjugative plasmids, is one of the main drivers of the global spread of antibiotic resistance. However, the relative contributions of different factors that underlie this plasmid spread are unclear, particularly for clinically relevant plasmids harboring antibiotic resistance genes. Here, we analyze nosocomial outbreak-associated plasmids that reflect the most relevant Extended Spectrum Beta-Lactamase (ESBL) mediated drug resistance plasmids to i) quantify conjugative transfer dynamics, and ii) investigate why some plasmid-strain associations are more successful than others, in terms of bacterial fitness and plasmid spread. We show that, in the absence of antibiotics, clinical Escherichia coli strains natively associated with ESBL-plasmids conjugate efficiently with three distinct E. coli strains and one Salmonella enterica Serovar Typhimurium strain. In more than 40% of the in vitro mating populations, ESBL-plasmids were transferred to recipients, reaching final transconjugant frequencies of up to 1% within 23 hours. Variation of final transconjugant frequencies was better explained by variation in conjugative transfer efficiency than by variable clonal expansion of transconjugants. We also identified plasmid-specific genetic factors, such as the presence/absence of particular transfer genes, that influenced final transconjugant frequencies. Finally, we validated the plasmid spread in a mouse model for gut colonization, demonstrating qualitative correlation between plasmid spread in vitro and in vivo. This suggests a potential for predictive modelling of plasmid spread in the gut of animals and humans, based on in vitro testing. Altogether, this may allow straightforward identification of resistance plasmids with high spreading potential and to implement quarantine or decolonization procedures to restrict their spread. Author summary Antibiotic resistance is a major obstacle to the treatment of bacterial infections in clinics. Plasmids encoding antibiotic resistance genes can spread between bacteria in a density-dependent manner and accelerate the rise of resistant bacterial strains. This is particularly important for densely inhabited ecological niches such as the guts of humans and animals, where many bacteria interact. Understanding the exact contribution plasmids make to the global spread of antibiotic resistance remains an obstacle, because we lack quantitative studies implementing large-scale experimental testing of conjugation rates between clinically relevant bacterial strains. To counteract this knowledge gap, we studied clinical Escherichia coli isolates from human patients that carry extended-spectrum beta-lactamase producing plasmids. We found that these plasmids spread extensively through different bacterial populations and that both bacterial- and plasmid-specific factors determined the extent of plasmid spread. Our study combines detailed bioinformatic analyses, high-throughput in vitro testing and validation in an animal model. It suggests a potential for laboratory testing to understand and predict the spread of clinically relevant plasmids, including in the human gut microbiota, and thereby generates insights into novel treatment strategies to manage antibiotic resistance spread mediated by plasmids.

frequencies was better explained by variation in conjugative transfer efficiency than by 48 variable clonal expansion of transconjugants. We also identified plasmid-specific genetic 49 factors, such as the presence/absence of particular transfer genes, that influenced final 50 transconjugant frequencies. Finally, we validated the plasmid spread in a mouse model for 51 gut colonization, demonstrating qualitative correlation between plasmid spread in vitro and 52 in vivo. This suggests a potential for predictive modelling of plasmid spread in the gut of 53 animals and humans, based on in vitro testing. Altogether, this may allow straightforward 54 identification of resistance plasmids with high spreading potential and to implement 55 quarantine or decolonization procedures to restrict their spread. 56 Author summary Introduction 7 not associated with a major cost or benefit in the gut, confirming the critical role of 146 conjugation in driving the spread of antibiotic resistance plasmids. We showed that the range 147 of transfer efficiencies in vivo are qualitatively predictable based on our in vitro testing. 148 absence of antibiotics. We investigated the rate at which ESBL-plasmids from eight clinical 151 donor E. coli strains spread through four non-resistant recipient populations. The donor 152 strains stem from hospitalized patients and are representative of clinically relevant ESBL-153 producing E. coli [34]. Of the four recipients, two E. coli strains were isolated from healthy 154 human subjects [35], one E. coli strain was previously isolated from mice housed in our facility 155 [8], and one is a pathogenic wild type strain of Salmonella enterica Serovar Typhimurium. The 156 chosen donor and recipient strains spanned the phylogenetic diversity of E. coli [36,37] (Fig  157   1) and contained many plasmids of various incompatibility groups (S1 Fig, S1 Table). Each 158 donor carried one plasmid containing resistance genes, either of the plasmid family IncI1 or 159 IncF (Table 1,   Amp; Ceftri; Cotri †Genes encoding the ESBL phenotype are highlighted in bold. *Antibiotic resistance is defined as being above the EUCAST defined minimum inhibitory concentration breakpoint (MIC; see S2 Table for all antibiotics tested and MIC information). Donor strains fulfilled criteria for ESBL production based on EUCAST recommendations. ESBL mechanism was phenotypically and genotypically confirmed. Abbreviations: Amp = Ampicillin; Amo/C = Amoxicillin/Clavulanic acid; Pip-T = Piperacillin-Tazobactam; Ceftaz = Ceftazidim; Ceftri = Ceftriaxone; Cefe = Cefepim; Tobra = Tobramycin; Amika = Amikacin; Cotri = Cotrimoxazol; Cipro = Ciprofloxacin.
First, we performed conjugation assays between all clinical donors and non-resistant 166 recipients (approximately 1:1 ratio, referred to below as the "1 st gen in vitro experiment") and 167 measured conjugation as the final fraction of the recipient population that carried the plasmid 168 after 23 hours of growth (hereafter termed "final transconjugant frequency"). Five of the 169 eight ESBL-plasmids were transferred to more than one E. coli recipient (Fig 2) Fig 2B). In S. Typhimurium, variation across donor-plasmid pairs 180 was similar to E. coli recipients, with the exception of p6A_IncI, which did not spread to 181 recipient RS. 182 Qualitatively, phylogenetic relatedness (Fig 1) between donor strains did not explain 184 similarities in their plasmid transfer dynamics in the 1 st gen in vitro experiment. For example, 185 D5, D6 and D7 have effectively the same host background (ST 131), yet transferred their 186 plasmids with very different efficiencies (Fig 2). By contrast, D1 and D8 are relatively 187 phylogenetically distant, yet showed very similar transfer dynamics across recipients (Fig 2). 188 Phylogenetic relatedness between donor strain and recipient was not strongly associated 189 with transfer efficiency either: D5, D6 and D7 are all equally closely related to recipients RE1 190 and RE2, yet we observed very different final transconjugants frequencies, spanning four 191 orders of magnitude, among these 6 donor-recipient pairs (Fig 2) Therefore, the vast majority of transfer events results from donor-to-recipient conjugation 254 rather than transconjugant-to-recipient transfer. Thirdly, regarding assumption (iii), we 255 conducted a control experiment to estimate the number of transconjugants observed solely 256 due to surface mating during the plating step of our conjugation assay (S4 Table). For most 257 donor-recipient pairs we found that the number of transconjugants due to on-plate mating  for which we sequenced three clones). We did not find any mutational changes in ESBL-318 plasmids but a range of chromosomal mutations in transconjugants (S5 Table). However, 319 these mutations were not consistently present across transconjugants and could not be 320 related to adaptation to plasmid carriage. Nevertheless, we expect that host factors play a 321 major role in plasmid spread, and it is simply their interplay and our small sample size that 322 prevent us from detecting one single crucial conjugation rate determinant.

trends. Sequence-based studies have shown that ESBL-plasmids can spread between bacteria 326
in the human gut [5,31], but their actual conjugation dynamics have not yet been studied in 327 vivo (exception using laboratory S. Typhimurium strains [9]). Therefore, we extended our 328 study to a more realistic murine model. We performed conjugation assays in mice with a 329 limited microbiota [42] with three of the clinical donor strains carrying ESBL-plasmids (D4, D8, 330 and D7) and RE3 as a recipient (Fig 6). These mice allow colonization of approximately 10 8 E. 331 coli per gram feces, densities of E. coli that can be found in the guts of humans and animals 332 [37]. Donors and recipients were introduced at equal densities but at different timepoints, 333 the recipient one day prior to the donor, and were allowed to grow and conjugate over 7 days 334 ( Fig 6A). In line with our in vitro conjugation assay, we observed transconjugants for p4A_IncI 335 and p8A_IncF, but not for p7A_IncF/Col156 ( Fig 6B). Similar to the in vitro results in Fig 2,  336 transfer of p4A_IncI resulted in higher transconjugants frequencies than transfer of p8A_IncF 337 (day 1-6, Fig 6B). Plasmid p4A_IncI reached maximal proportions of plasmid carrying 338 recipients of ~1% within only 4-8 hours of colonization ( Fig 6C). Overall, we found that the 339 rank order of final transconjugant frequencies is conserved between in vitro and in vivo 340 conjugation. Moreover, the speed at which these clinical plasmids can establish themselves 341 in the gut in the absence of antibiotic selection is striking. 342

343
As was the case in vitro, the transconjugant population in vivo was relatively minor compared 344 to the size of the donor population (S4-6 Figs). This suggests that most transfer events derived 345 from donor-to-recipient transfer, rather than transconjugant-to-recipient transfer. Given this, 346 it is likely that the conjugation dynamics in vivo can be predicted by the 1 st gen in vitro 347 experiments (Fig 2). 348

349
To determine whether the increase in transconjugant population size in the gut over the 350 seven-day experiment was driven by clonal expansion or by conjugation, we investigated the 351 competitive advantage of transconjugants versus recipients in vivo. For two transconjugants 352 from the 1 st gen in vitro experiment, we deleted the ESBL-plasmid origin of transfer (oriT) and 353 created "locked", non-conjugative plasmids. These were competed 1:1 with RE3, the recipient 354 used in the in vivo conjugation assay. A change in fitness conferred by plasmid carriage would 355 be reflected in the competitive index, which is calculated as the ratio between the population 356 sizes of recipient and "locked" transconjugant. After seven days of competition, only 357 transconjugants carrying p8A_IncF outcompeted RE3, with a competitive index of ~0.1 (Fig  358   6D). Over the course of the experiment, this plasmid-borne fitness advantage led the initial 359 transconjugant frequency of 0.5 to increase to a final transconjugant frequency of 0.9. 360 Because this two-fold increase in transconjugant frequency is small compared to the 361 transconjugant frequencies observed in the conjugation experiment (increasing from the 362 detection limit of 10 -6 to a frequency of 0.01; Fig 6B), we conclude that final transconjugant 363 frequencies in the conjugation assay were mainly driven by conjugative transfer, rather than 364 by clonal expansion of transconjugants. This is seen to a larger extent for p4A_IncI, where the 365 transconjugant frequency increased substantially faster than with p8A_IncF (Fig 6B), despite 366 a total lack of a growth advantage over recipient RE3 (Fig 6D). Altogether, we show that in 367 the murine gut, ESBL-plasmids can be transferred rapidly between E. coli in the absence of 368 antibiotic selection. 369 plasmid transfer. It has been proposed that the transfer rate of plasmids can be modified by 372 either the presence of other plasmids in the same bacterial host, by the co-transfer of other 373 plasmids or by the presence of other plasmids in recipients [19,43]. All donor and recipient 374 strains used in this study carry at least one plasmid (S1 Table). We investigated whether 375 interactions with these plasmids might affect transfer of the ESBL-plasmids, using sequenced  The cause and effect we can not trace back: loss of resident plasmids could have allowed subsequent acquisition of ESBL-plasmids, or the acquisition of ESBL-plasmids could have been 395 followed by the loss of resident plasmids. Either way, despite this incompatibility of resident 396 and newly-acquired ESBL-plasmids comparably high transconjugant frequencies could be 397 reached (Figs 2 and 5). growth, we found that the increase in plasmid frequency was strongly correlated with plasmid 423 transfer rate (Fig 4). This might hold true even for costly plasmids, as others have shown that 424 transfer rates can surpass a fitness cost and allow plasmids to spread [26,47]. Foremost, we 425 found that all plasmids carrying the required tra genes spread from various donors through a 426  with appropriates amounts of antibiotics (none, 100µg/mL Ampicillin (Amp) , 50µg/mL Cm 523 and 50µg/mL Kan). We stored isolates in 25% glycerol at -80°C. 524 525 Antibiotic resistance profiling. We used microdilution assays with a VITEK2 system 526 (bioMérieux, France) to determine the minimum inhibitory concentrations (MIC). MIC 527 breakpoints for ESBL were interpreted according to EUCAST guidelines (v8.1). In addition, we 528 confirmed the resistance mechanism of suspected isolates phenotypically using ROSCO disk 529 assays (ROSCO diagnostica, Denmark) and/or genotypically with detection of CTX-M1 and -9 530 groups using the eazyplex Superbug assay (amplex, Germany). transconjugants having adapted to the LB growth conditions. Thus, we exposed recipients (n 629 = 4) to the same culture handling as a conjugation assay would and found their growth rates not to be different from the naïve recipient strains used in the plasmid cost assay (S9 Fig). We Resequencing. We re-sequenced isolates from the 2 nd generation in vitro experiment as well 669 as the in vivo transfer experiment, to study the genetic contribution to the observed transfer 670 rates. Resequencing was performed on an Illumina MiSeq (paired end, 2x150 bp). Reads were 671 mapped to the closed assemblies of respective donor and recipient strains using the breseq 672 pipeline (v 0.32.0). Mutations or indels shared by all re-sequenced strains were treated as 673 ancestral (S5 Table). and S. Typhimurium recipient RS. When single replicates for a given donor-recipient 682 combination lacked transconjugants (D5 and D6), we assigned these replicates a final 683 transconjugant frequency at the detection limit of 10 -8 . The same data was used to perform 684 the correlation of transconjugant frequency and transfer rate. Data of the 2 nd gen in vitro 685 experiment was not fully factorial. To enable testing of higher-order interactions here, we 686 therefore performed two 3-way ANOVAs: one excluding plasmid p1B_IncI and one excluding 687 donor RE2, for which we had to take the two replica blocks into account: P < 0.001). For two 688 replica populations of donor RE2 transferring p4A_IncI to RE1, we were not able to detect 689 donors on plates and assigned them a donor population size at the detection limit of 1 colony 690 per plate (10 8 cells/mL). For two replica populations (RS self-self transfer with p1B_IncI), we 691 had higher counts on plates selecting for transconjugants than on plates selecting for 692 recipients+transconjugants and replaced the resulting negative CFU/mL for recipients with 1 693 CFU/mL.