Collateral sensitivity associated with antibiotic resistance plasmids

Collateral sensitivity (CS) is a promising alternative approach to counteract the rising problem of antibiotic resistance (ABR). CS occurs when the acquisition of resistance to one antibiotic produces increased susceptibility to a second antibiotic. Recent studies have focused on CS strategies designed against ABR mediated by chromosomal mutations. However, one of the main drivers of ABR in clinically relevant bacteria is the horizontal transfer of ABR genes mediated by plasmids. Here, we report the first analysis of CS associated with the acquisition of complete ABR plasmids, including the clinically important carbapenem-resistance conjugative plasmid pOXA-48. In addition, we describe the conservation of CS in clinical E. coli isolates and its application to selectively kill plasmid-carrying bacteria. Our results provide new insights that establish the basis for developing CS-informed treatment strategies to combat plasmid-mediated ABR.


Introduction 31
The rapid evolution of antibiotic resistance (ABR) in bacteria reduces the utility of clinically relevant 32 antibiotics, making ABR one of the major challenges facing public health (Jim, 2016;MacLean and 33 San Millan, 2019). There is therefore an urgent need for new treatment strategies to fight against 34 resistant pathogens (Baym et al., 2016). Among the most promising alternatives, much attention 35 has focused on collateral sensitivity (CS). CS occurs when the acquisition of resistance to one 36 antibiotic causes increased susceptibility to a second antibiotic (Szybalski and Bryson, 1952) and 37 can be exploited for the design of multi-drug strategies that select against ABR (Imamovic and 38 Sommer, 2013;Lázár et al., 2013). Moreover, recent evidence suggests that the development of 39 resistance might be prevented by treatments based on the sequential cycling of antibiotics whose 40 resistance mechanisms produce reciprocal CS (Imamovic and Sommer, 2013). Several studies 41 have cataloged CS networks emerging from ABR mutations in chromosomal genes (Imamovic and 42 Sommer, 2013;Lázár et al., 2013;Maltas and Wood, 2019;Podnecky et al., 2018;Roemhild et al., 43 3 2020), but the CS effects produced by the horizontal acquisition of ABR plasmids remain poorly 44 understood. 45 Mobile genetic elements, especially plasmids, play a crucial role in the dissemination of ABR 46 genes between clinical pathogens and are one of the major drivers of the alarming worldwide rise 47 in ABR (Partridge et al., 2018). Despite the evolutionary advantage conferred by plasmids in the 48 presence of antibiotics, plasmids acquisition tends to produce common metabolic alterations in the 49 host bacterium (San Millan et al., 2018), that typically translate into a fitness cost (San Millan and 50 MacLean, 2017). We reasoned that the physiological alterations produced upon acquisition of ABR 51 plasmids might lead to exploitable CS responses. To test this hypothesis, we analyzed the 52 collateral effects associated with the acquisition of 6 natural plasmids carrying clinically relevant 53 ABR genes in Escherichia coli MG1655. Our results reveal that most ABR plasmids, including the 54 clinically important carbapenem-resistance conjugative plasmid pOXA-48, produce CS events of 55 moderate effect. To extend our results, we explored the degree of conservation of CS responses 56 associated with pOXA-48 across phylogenetically diverse wild-type E. coli. Our results 57 demonstrate that these responses can be exploited to preferentially eradicate plasmid-carrying 58 bacteria. 59

Collateral sensitivity induced by ABR plasmids 61
To test whether ABR plasmids elicit exploitable CS to other antibiotics, we selected 6 clinically 62 relevant ABR plasmids (see methods, Table 1 and Supplementary Figure 1). All plasmids carried 63 important ß-lactamase genes, including extended-spectrum ß-lactamases (ESBLs), AmpC-type ß-64 lactamases, or carbapenemases, as well as other relevant ABR genes (Table 1). These plasmids 65 belonged to a broad range of incompatibility groups, covered a range of different sizes, included 66 conjugative and non-conjugative specimens, and produced variable fitness effects, ranging from 67 10% to 27% reductions in relative fitness, when introduced into E. coli MG1655 ( Figure 1A). 68 We performed dose-response experiments with 13 antibiotics from 8 drug families (Supplementary  69   Table 1). For this analysis, we determined the minimal inhibitory concentration (MIC, see methods) 70 4 using the broth microdilution method for plasmid-free MG1655 and each of its 6 plasmid-carrying 71 derivatives (Supplementary Table 1). We analyzed the collateral antibiotic susceptibility effects 72 associated with the presence of each plasmid, measured as the fold-change in the antibiotic MIC 73 between plasmid-carrying and plasmid-free bacteria ( Figure 1B). As expected from their genetic 74 content, all plasmids conferred resistance to ß-lactam antibiotics and most of them conferred 75 additional resistance to other antibiotics from unrelated families ( Figure 1B and Supplementary 76 Table 1). CS was defined as a minimum 2-fold reduction in MIC (calculated as the fold-change of 77 the median MIC value obtained from 4-5 independent MIC determinations). Using this definition, 78 we identified 8 instances of plasmid-induced CS. The plasmids that produced CS effects were 79 pOXA-48 (to colistin and azithromycin), pKAZ3 (to kanamycin and gentamicin), pKA2Q (to 80 tetracycline), and pCF12 (to kanamycin, azithromycin, and tetracycline). 1D). These differences were statistically significant for pCF12-kanamycin, pCF12-azithromycin, 92 pOXA-48-azithromycin, pOXA-48-colistin, pCF12-tetracycline, and pKA2Q-tetracycline (unpaired 93 t-test with Welch's correction; P<0.044 in all cases). Overall, 7 of the 8 CS instances identified by 94 MIC testing were validated with at least one of the additional methods (growth curves or disk-95 diffusion assay), and 6 of them showed full agreement across all 3 methods. The disk-diffusion 96 assays produced notably robust results with minimum hands-on time, suggesting that this 97 technique is appropriate for screening CS responses in large strain collections. Together, these 98 5 results demonstrate that ABR plasmids produce moderate but significant CS to different antibiotics, 99 comparable to that generated by ABR chromosomal mutations in E. coli (Imamovic and Sommer, 100 2013;Lázár et al., 2013;Podnecky et al., 2018). 101

Conservation of pOXA-48-mediated CS 102
The success and applicability of CS-informed therapeutic strategies are crucially dependent on the 103 conservation of CS across diverse genetic contexts (Podnecky et al., 2018). To address the 104 phylogenetic preservation of plasmid-induced CS, we focused on the plasmid pOXA-48, which 105 confers resistance to last-resort carbapenem antibiotics, and whose prevalence in clinical settings 106 is rising alarmingly (David et al., 2019). We tested the degree of conservation of pOXA-48-107 mediated CS patterns (to azithromycin and colistin) across phylogenetically diverse E. coli strains 108 using disk-diffusion assays. We determined antibiotic susceptibility in 9 diverse E. coli clinical 109 isolates ( Figure 2A) and their pOXA-48-carrying derivatives. CS to azithromycin showed a striking 110 degree of conservation across the strains, with 8 out of 9 plasmid-carriers showing significantly 111 higher susceptibility than their plasmid-free counterparts. In contrast, CS to colistin was conserved 112 in only 2 of the tested strains ( Figure 2B). Nevertheless, all strains showed CS responses to at 113 least one antibiotic, suggesting that pOXA-48 elicits CS responses that could potentially be 114 exploited therapeutically. 115

Exploitation of CS for the preferential extinction of pOXA-48-carrying E. coli 116
We next tested the potential of exploiting CS effects to eradicate plasmid-carrying subpopulations. 117 We rationally selected 4 E. coli isolates and their pOXA-48 carrying derivatives with defined CS 118 profiles: to colistin only (Ec10), to azithromycin only (Ec21 and Ec25), or to both antibiotics (Ec18). 119 MG1655 was included as a control. Each bacterial population was propagated in the presence of a 120 single antibiotic or with sequential exposure to both antibiotics over 2 days. On the first day of the 121 experiment, 24 bacterial populations of each strain were inoculated into media containing either 122 colistin (4 mg/L) or azithromycin (8 mg/L). After growth for 22 hours, surviving populations were 123 transferred to fresh media containing colistin or azithromycin in a full factorial experimental design. 124 This approach gave rise to 4 antibiotic treatments: 2 single-drug treatments (denoted Azi→ Azi and 125 Col→ Col) and 2 treatments in which the antibiotic changed (Azi→ Col and Col→ Azi). In 11 of the 126 6 20 possible strain-treatment combinations, survival patterns showed significantly higher mortality 127 rates for the pOXA-48-carrying strains than for their plasmid-free counterparts (Figure 3; log-rank 128 test P<0.001 in all cases). Crucially, the mortality patterns observed were consistent with the CS 129 patterns obtained in the disk-diffusion technique ( Figure 2B). For instance, plasmid-carrying 130 MG1655 and Ec18 strains, which show CS to both antibiotics, exhibited significant reductions in 131 survival in all treatments. In contrast, for strain Ec10, which shows CS exclusively to colistin, 132 significant extinction was observed only when the treatment included colistin as the first antibiotic 133 (Col→ Col or Col→ Azi). 134

Concluding remarks 135
Our results reveal that the acquisition of clinically-relevant ABR plasmids induces CS to unrelated 136 antibiotics. These findings thus serve as a stepping stone toward the development of new 137 approaches aimed at blocking plasmid-mediated horizontal spread of ABR genes. These anti-138 plasmid strategies could help to tackle the alarming clinical and community spread of ABR. 139 However, the molecular basis of plasmid-associated CS is currently unknown. The feasibility of 140 rational broad-spectrum anti-plasmid strategies will depend on gaining comprehensive knowledge 141 about the molecular mechanisms that increase antibiotic susceptibility upon plasmid acquisition. 142 Until these therapies are available, CS-informed empirical treatment of plasmid-carrying bacteria 143 has the potential to help resolve antibiotic resistant infections. In this regard, our results suggest 144 that colistin and azithromycin may be valuable antibiotics for the treatment of pOXA-48-carrying 145 enterobacteria. Both antibiotics are already used to treat enterobacterial infections (Lübbert, 2016;146 Morrill et al., 2015), and colistin in particular is currently used as a last resort antibiotic against  Table 1) and by 176 sequencing the complete genomes of transconjugant clones, which also revealed that no 177 significant chromosomal mutations accumulated during the process of plasmid acquisition (0-2 178 8 mutations between plasmid-carrying and plasmid-free complete genomes, see Supplementary 179 Table 3). 180 The clinical E. coli strains used in this study were isolated during the R-GNOSIS project, which 181 containing the antibiotic were placed on the plates. Plates were incubated at 37 °C for 22 hours 202 and pictures were taken using an in-house photographic system and a cell phone (Huawei Mate 203 20). Inhibition halos were measured using ImageJ software. Antibiotic disk content used in 204 9 diffusion assays were as follows (all from Bio-Rad): gentamicin 10 µg, azithromycin 15 µg, 205 kanamycin 30 µg, colistin 10 µg, and tetracycline 30 µg. 206

Bacterial growth curves 207
Starter cultures were prepared and incubated as described above. Each culture was diluted 208 1:2,000 in LB medium and 200 µl were added to a 96-well microtiter plate containing the 209 appropriate antibiotic concentration. Plates were incubated 24 hours at 37 °C with strong orbital 210 shaking before reading OD 600 every 10 minutes in a Synergy HTX (BioTek) plate reader. 6 211 biological replicates were performed for each growth curve. The area under the growth curve was 212 obtained using the 'auc' function from the 'flux' R package. Data was represented using a R 213 custom script and the 'ggplot2' package. 214

Competition assays 215
Competition assays were performed to measure the relative fitness of plasmid-carrying strains 216 using flow cytometry as previously described (Rodriguez-Beltran et al., 2018). Briefly,competitions 217 were performed between MG1655 clones carrying each of the 6 plasmids and plasmid-free 218 estimated again by flow cytometry as described above. The fitness of each strain relative to 227 MG1655::gfp was calculated using the formula: w=ln(N final,gfp-/N initial,gfp-)/ln(N final,gfp+ /N initial,gfp+ ) where w 228 is the relative fitness of the non GFP-tagged strain, N initial,gfp-and N final,gfp-are the numbers of non 229 GFP-tagged cells before and after the competition and N initial,gfp+ and N final,gfp+ are the numbers of 230 MG1655::gfp cells before and after the competition. Data was normalized by dividing the relative 231 10 fitness of plasmid-carrying derivative strains by the fitness calculated for plasmid-free MG1655. Six 232 biological replicates were performed for each competition. 233

CS-informed antibiotic treatments 234
Overnight bacterial cultures were diluted and seeded into 24 independent wells of a 96-well plate 235 filled with 200 µl of LB. After 16 hours of incubation at 37 ºC, these populations were diluted 236 1:2,000 into fresh medium containing either azithromycin (8 mg/L) or colistin (4 mg/L) and grown at 237 37 ºC for 22 hours. The following day, bacterial populations were diluted (1:2,000) and inoculated 238 into new plates containing, again, either azithromycin or colistin and allowed to grow as above. 239 This approach led to four antibiotic treatments, two in which the antibiotic remains constant (Azi→ 240 Azi and Col→ Col) and two in which the antibiotic treatment alternates (Azi→ Col and Col→ Azi). 241 The survival of bacterial populations was assessed by measuring OD 600 every day. Populations 242 that did not reach an OD 600 of 0.2 were declared extinct. 243

Whole genome sequencing 244
Genomic DNA was isolated using the Wizard genomic DNA purification kit (Promega) and 245 quantified using the QuantiFluor dsDNA system (Promega) following manufacturer's instructions. 246 Whole genome sequencing was conducted at the Wellcome Trust Centre for Human Genetics 247 (Oxford, UK), using the Illumina HiSeq4000 platform with 125 base pair (bp) paired-end reads. For 248 plasmid pCF12, first described in this study, we performed additional long-read sequencing using 249 PacBio technologies. PacBio sequencing was performed at The Norwegian Sequencing Centre 250 PacBio RSII platform using P5-C3 chemistry. Illumina and PacBio sequence reads were trimmed 251 using the Trimmomatic v0.33 tool (Bolger et al., 2014). SPAdes v3.9.0 was used to generate de 252 novo assemblies from the trimmed sequence reads with the -cov-cutoff flag set to 'auto' 253 (Bankevich et al., 2012). Unicycler was used to generate hybrid assemblies from Illumina and 254 PacBio data (Wick et al., 2017). QUAST v4.6.0 was used to generate assembly statistics (Gurevich 255 et al., 2013). All the de novo assemblies reached enough quality including total size of ~4.6 Mb, 256 and a total number of contigs over 1 kb lower than 200. Prokka was used to annotate the de novo 257 assemblies (Seemann, 2014). The plasmid content of each genome was characterised using 258 PlasmidFinder 2.1 (Carattoli et al., 2014), and the antibiotic resistance gene content was 259 11 characterised with ResFinder 3.2 (Zankari et al., 2012). MOB families were characterised using 260 MOB-typer tool included in MOB-suite (Robertson and Nash, 2018