Hidden antibiotic resistance fitness costs revealed by GWAS-based epistasis analysis

Understanding how multi-drug resistant pathogens evolve is key to identifying means of curtailing their further emergence and dissemination. Fitness costs imposed on bacteria by resistance mechanisms are believed to hamper their dissemination in an antibiotic free environment, however, some have been reported to have little or no cost, which suggests there are few barriers preventing their global spread. One such apparently cost-free resistance mechanism acquired by the major human pathogen Staphylococcus aureus is to the clinically important antibiotic mupirocin, which is mediated by mutation of the highly-conserved and essential isoleucyl-tRNA synthethase (ileS) gene. In Genome Wide Association Studies (GWAS) on two genetically and geographically distinct MRSA lineages we have found this mutation to be associated with changes in bacterial virulence, driven through epistatic interactions with other loci. Given the potential dual effect of this mutation on both antibiotic resistance and virulence we adopted a proteomic approach and observed pleiotropic effects. This analysis revealed that the activity of the secretory apparatus of the PSM family of cytolytic toxins, the Pmt system, is affected in the mupirocin resistant mutant, which explains why it is less toxic. As an energetically costly activity, this reduction in toxicity masks the fitness costs associated with this resistance mutation, a cost that becomes apparent when toxin production is required. Given the widespread use of this antibiotic, and that this resistance often results from a single nucleotide substitution in the ileS gene, these hidden fitness costs provide an explanation for why this resistance mechanism is not more prevalent. This work also demonstrates how population-based genomic analysis of virulence and antibiotic resistance can contribute to uncovering hidden features of the biology of microbial pathogens.


INTRODUCTION 55
Antibiotic resistance can evolve in many ways, and frequently incur a fitness cost to 56 the organism 1 which has to either mutate the target site of the antibiotic, acquire 57 and express a gene encoding an alternative non-susceptible version of the target 58 protein, or acquire and produce an efflux pump that removes the antibiotic before it 59 can attack its target 2 . As antibiotics are most commonly used for short and defined 60 periods of time, resistant bacteria are under selection to reduce these costs to avoid 61 displacement once treatment has finished. In many cases this is achieved through 62 compensatory mutations that allow many resistance mechanisms to be maintained 63 stably populations for long periods of time 3 . However, some antibiotic resistance 64 mechanisms have been reported to incur no detectable fitness costs 4,5,6 , which 65 suggests there are no barriers to their widespread dissemination. 66 Staphylococcus aureus is an example of a major human pathogen 7 that has 67 become more challenging to treat due to the emergence of antibiotic resistance, 68 with Methicillin-Resistant S. aureus (MRSA) being the most notable example 8 . S. 69 aureus resides asymptomatically as part of the normal nasal flora of up to 50% of 70 humans 9 , however, this is a significant risk factor for infection 10 , to the extent that 71 carriers are often decolonised using antibiotics such as mupirocin prior to invasive 72 procedures such as surgery or dialysis 11 . Mupirocin is a polyketide antibiotic that is 73 applied as an ointment to eradicate nasal carriage of MRSA in patients at risk of 74 infection 12 . Such decolonisation has been reported to reduce S. aureus infections of 75 post-surgical wounds by 58%, of haemodialysis patients by 80% and of peritoneal 76 dialysis patients by 63% 13 . 77 The molecular target for mupirocin is the bacterial isoleucyl-tRNA synthetase 78 (IleRS), which charges tRNAs with the amino acid isoleucine (Ile) 12 . By binding to this 79 enzyme the antibiotic halts protein synthesis, so inhibiting bacterial growth 14 . As a 80 consequence of the widespread use of mupirocin, resistance has emerged where the 81 bacteria have mutated the gene encoding IleRS, ileS, resulting in an amino acid 82 substitution (e.g. V588F, encoded by a G to T single nucleotide polymorphism (SNP) 83 at position 1,762 in the ileS gene (G1762T)) which alters the structure of the 84 protein's active site, but retains functionality and renders mupirocin less effective 15 . 85 This confers a low to intermediate level of resistance to the antibiotic 16 . 86 this was lineage specific. We analysed toxicity and sequence data for a collection of 119 130 USA300 MRSA isolates 22 , where the SNP conferring mup R resistance emerged for 120 the USA300 isolates as the most dominant toxicity-affecting epistatically-interacting 121 locus ( fig. 1), demonstrating the widespread nature of this effect across diverse 122 clonal lineages. See Supp. Table 1 for the list of loci associated with the mup R 123 conferring SNP for both the ST239 and USA300 collections. 124 To illustrate the epistatic effect on toxicity we selected at random one of the 125 interacting loci and present the mean toxicity of the four combinations of each allele 126 of the two genes ( Table 1). The SNP we selected was at position 480640 (relative to 127 the origin of replication) and confers a non-synonymous change in the protein 128 encoded by the open-reading frame with the locus tag SAUSA300_0426, which is 129 described as a conserved hypothetical protein. When in combination with the mup R 130 encoding SNP in ileS, strains with the allele of SAUSA300_0426 containing a T at 131 position 480640 are significantly more toxic than those with a C at this site. Whereas 132 in the mupirocin sensitive strains, those with the C at position 480640 are more toxic 133 than those with a T at this site (Table 1). This example illustrates the toxicity 134 affecting epistatic signal detected by our analysis. 135 136 Mupirocin resistance exerts a pleiotropic effect on the S. aureus proteome. 137 Across both the ST239 and USA300 collections of isolates, 59 loci associated with the 138 toxicity of S. aureus through epistatic interactions with the mup R mutation (Supp. 139 Table 1). To understand how such potential interactions could affect toxicity we 140 examined this list of loci, however, no known toxicity affecting genes were 141 identified, and no loci were common between the two clones. As IleRS is involved in 142 the translation of proteins, and our GWAS data suggests this antibiotic resistance 143 conferring mutation also affects the ability of S. aureus to secrete toxins, we 144 hypothesised that this mutation may have pleiotropic effects on protein production, 145 which could explain the observed epistasis. To examine this, we isolated a mup R 146 version of the S. aureus laboratory strain SH1000 by plating overnight cultures on 147 agar containing 4µg/ml mupirocin. Mupirocin resistant colonies of the SH1000 strain 148 were recovered, and to confirm that the V588F conferring SNP was present the 149 colonies were sequenced and compared to SH1000. We selected a colony which we 150 have designated MY40, where the only non-synonymous SNP found in this strain 151 was that conferring the V588F change in IleRS, although a small number of non-152 synonomous SNP differences were detected (Supp. open reading frames predicted for the NCTC8325 chromosome (which is the closest 158 reference genome to SH1000), this proteomic approach was able to detect and 159 quantify 1284 proteins, where we used a cut-off of a minimum of a two-fold 160 difference in protein abundance to identify differentially produced proteins. When 161 we compared the proteome of SH1000 and MY40, there were 140 proteins that 162 were differentially produced (Supp. Table 3), which verified our hypothesis that this 163 resistance mutation has pleiotropic effects. However, when we compared this list of 164 differentially produced proteins with our list of loci associated with toxicity through 165 epistasis with the mup R mutation (Supp. Table 1) there was no overlap, suggesting 166 any toxicity affecting interactions between these loci must be indirect. 167 168 Exposure to mupirocin and mupirocin resistance have common effects on protein 169 production. To examine whether the mup R mutation has a similar effect on IleRS 170 activity as the presence of mupirocin has, alongside our proteomic analysis of the 171 the mup R mutant we also analysed lysate of SH1000 exposed to the highest 172 concentrations of mupirocin for which we found no inhibition of growth (10ng/ml). 173 This concentration was selected to avoid any confounding effects differences in 174 growth rates might have on the proteome. Exposure of the wild type SH1000 strain 175 to mupirocin resulted in differential production of 67 proteins when compared to 176 the untreated SH1000 (Supp. Table 4), 18 of which were also different in the mup R 177 mutant, suggesting many common down-steam effects of interfering with the 178 activity of IleRS. These include an increase in the production of several of the 179 proteins involved in Ile biosynthesis (i.e. LeuA, LeuC and IlvA), which suggests that 180 the bacteria are compensating for the interference in IleRS activity due to the 181 mutation and exposure to the antibiotic. Another common expression difference 182 was that several of the iron-regulated surface determinant (Isd) proteins were 183 expressed at lower levels in both the mutant and the mupirocin-exposed wild type 184 when compared with the untreated wild type. As yet we have no explanation for this 185 observation. 186 187 Effect of mupirocin resistance on virulence regulating proteins. The first virulence 188 related protein abundance difference between the SH1000 and its mup R mutant we 189 noted was the AgrA protein 24 . This is the cytoplasmic response regulator of the Agr 190 toxicity regulating system which was produced at significantly higher levels in the 191 mutant (7-fold, Supp. Other known virulence regulators were also identified as being differentially 203 produced in the mupirocin resistant strain. Both the SrrB and SarR proteins 24 were 204 expressed at higher levels in the mutant, whereas the Rot protein 24 was expressed at 205 lower levels compared to the wild type strains. Three other known regulatory 206 proteins were also differentially expressed (i.e. a GntR family transcriptional 207 regulator, the lytic regulatory protein SAOUHSC_02390, and the Pur operon 208 regulator, PurR), which when considered alongside the effect on the three virulence 209 regulators suggests that mupirocin resistance results in a significant re-wiring of S. 210 aureus regulatory processes. 211 212 Effect of mupirocin resistance on toxin production. Of the toxins encoded on the 213 SH1000 genome, there was significantly more of both delta toxin and PSMa1 in the 214 lysate of mupirocin resistant mutant (32-and 21-fold respectively, Supp. Table 3). 215 These are two of the most abundantly produced members of the Phenol Soluble 216 Modulin (PSM) family of cytolytic toxins 25 . As our original GWAS analysis was 217 focused on toxicity, we hypothesised that these differences may explain the 218 association we observed, and it suggests that the mutant should be more toxic than 219 the wild type strain. To test this we quantified the cytolytic activity of the wild type 220 and mutant by incubating bacterial supernatant with cultured THP-1 cells, which are 221 sensitive to the majority of cytolytic toxins produced by S. aureus, including the 222 PSMs 22 . While we found that mupirocin resistance did influenced toxicity, the effect 223 was the opposite to what our proteomic analysis suggested it should be, in that the 224 mupirocin resistant MY40 strain killed only 71% of the cells compared to the wild 225 type mupirocin sensitive SH1000 strain which killed 87% of the cells (n=6, two tailed 226 t-test; p=0.023). As our toxicity assays uses bacterial supernatant, whereas our 227 proteomic analysis was on whole cell lysates, it is possible that the difference in suggests it is critical to the exporters activity. In our mup R mutant we found that 246 there was more than twice as much of one of the two ATP binding proteins 247 (SAOUHSC_02152, gene name pmtC) compared with the wild type strain. We 248 hypothesised that interference with the stoichiometry of the proteins in this export 249 system may explain why we have more PSM inside the cell but less outside. To test 250 this we cloned and expressed the pmtC gene, from an inducible promoter in the 251 SH1000 wild type background, and found that there was 2.8-fold less PSMs in the 252 extracellular environment when this ATPase was overexpressed (n=6, two tailed t-253 test; p=0.003, a representative image of is presented in fig. 2b). This suggests that 254 the effect mupirocin resistance has on S. aureus toxicity is mediated by interfering 255 with the activity of the PSM secretory apparatus. It is interesting to consider that the 256 complete inactivation of the Pmt system has been shown to be lethal to the bacteria, 257 presumably as a result of the damage the PSMs can cause to internal membrane 258 structures 26 . While we demonstrate a partial blocking of the Pmt system by 259 mupirocin resistance, and some accumulation of PSMs internally, it must be a sub-260 toxic level, as we see no in vitro growth defects associated with this. 261 262 Reducing the production of toxins alleviates the fitness cost of mup R . With one 263 consequence of the significant changes mup R causes to the proteome of S. aureus 264 being a reduction in toxin production, and with toxin production and secretion being 265 an energetically costly activity, we hypothesized that the down regulatory effect of 266 mupirocin resistance on toxin production may mask or alleviate the resistance 267 related fitness costs that are incurred. To test this we quantified the relative fitness 27 268 of the mup S and mup R strains by competition in two genetic backgrounds; one in the 269 SH1000 wild-type background (i.e. SH1000 and MY40) where the bacteria can 270 express toxins. For the other, we inactivated the Agr quorum sensing system in both 271 SH1000 and MY40 by transducing in the inactivated Agr system (an erythromycin 272 resistance has been inserted into it) from the strain ROJ48 28 , resulting in strains 273 MY18 and MY41. As reported previously, in the wild type background with a 274 functional Agr system, there was no difference in fitness between the mup S SH1000 275 and the mup R MY40 strains ( fig. 3A; n=10; two-tailed t-test p=0.6). However, when 276 we quantified the relative fitness in the Agr defective background where neither 277 strain could produce toxins, such that any alleviation of fitness costs that result from 278 the relative reduction in toxicity of the mup R mutant was nullified, we found the 279 fitness of the mup R MY41 strain to be significantly lower than the mup S MY18 strain 280 ( fig. 3A; n=10; two-tailed t-test p=0.04). This demonstrates that this mupirocin 281 resistance conferring mutation does incur a fitness cost, but this cost can be masked 282 by reducing the costly production of toxins. 283

284
As the ability to produce toxins is selected for in environments such as the nose 29 , 285 we hypothesised that in such an environment, the costs associated with this 286 resistance mechanism should become apparent, without having to genetically 287 manipulate the Agr system. To test this we developed a growth media with very low 288 levels of nutrients (0.1X TSB) but added 5% horse blood such that strains that can 289 produce toxins can release and utilise further nutrients from these blood cells for 290 growth. To avoid the less toxic cells benefiting from nutrients the more toxic cells 291 release, we quantified the growth rate of SH1000 and its mup R mutant separately, 292 quantifying their Malthusians parameters at 3, 6 and 24hours in this medium. After 293 3hr of incubation in the medium there was no detectable growth of either strain, 294 however, by 6hrs both bacterial strains had grown, although the mup R strain was 295 significantly more fit than the SH1000 strain ( fig. 3B; n=6; two-tailed t-test p=0.007). 296 This suggests the bacteria enter a long lag phase in this medium, but once adapted 297 (after 3hrs) there was sufficient nutrients available to sustain some growth, with the 298 SH1000 strain at an apparent disadvantage, presumably by expending energy on 299 producing more toxins than its mup R mutant. After the 6hr time-point there was a 300 distinct shift in the relative fitness of the strains, with the SH1000 strain becoming 301 relatively more fit than the mup R strain ( fig. 3B; n=6; two-tailed t-test p=0.04). The 302 increased relative growth rate of the wild type SH1000 strain is presumably as a 303 result of its increased capability to lyse cells and release nutrients necessary for its 304 growth. While the relative fitness of any organism is dependent upon its 305 environment, here we demonstrate how readily this can fluctuate within an 306 environment. That the fitness consequences of mupirocin resistance became 307 apparent when cell lysis became necessary may provide an explanation for why, 308 despite the ease at which this mutation can occur, it is not more prevalent and is 309 readily lost within healthy communities 20 . 310

311
The mup R mutation has differing effects on toxicity in different S. aureus 312 backgrounds. Our proteomic analysis did not provide any evidence in support of a 313 direct interaction occurring between the mup R ileS gene and the loci our GWAS 314 analysis has identified as interacting with this locus to affect toxicity ( fig. 1 and Supp. 315 Table 1). As GWAS in bacteria can be confounded by population structures and 316 linkage disequilibrium, it is therefore possible that these polymorphic loci are instead 317 reflective of the genetic backgrounds of the bacteria that are sensitive to the effect 318 of the mup R mutation to varying degrees. If true, then the introduction of the mup R 319 mutation into different S. aureus strains should have differing effects on toxicity. To 320 test this we isolated mup R colonies of two temporally and geographically diverse 321 isolates from the same clone as SH1000 (ST8 as determined by MLST), in the 322 laboratory strain RN6390B and in the USA300 clinical isolate USFL34. The presence 323 of the V588F mutation was confirmed by sequencing and the effect of the mutation 324 on toxicity quantified. Despite reducing the toxicity of the SH1000 background strain, 325 the mup R mutation had no effect on toxicity in either RN6390B or USFL34 (n=6; two-326 tailed t-test p=0.59 and 0.12 respectively), supporting our hypothesis that strains 327 respond differently to this antibiotic resistant mechanism, and providing an 328 explanation for our epistasis findings. is perhaps surprising that they have not become more prevalent. Here, by adopting a 337 population-based functional genomics approach, we uncover hidden fitness costs 338 associated with an apparently cost-free, antibiotic resistance mechanism. We 339 demonstrate that the mupirocin-resistance conferring mutation of the gene 340 encoding the IleRS enzyme has pleiotropic effects on bacteria, which in hindsight is 341 perhaps unsurprising given the highly conserved and essential nature of this gene. In 342 this instance, while reducing the energetic expense associated with toxin production 343 provides a relief of the resistance-associated fitness costs, this appears to be 344 unsustainable when the ability to produce toxins is required. Given that the nose of 345 healthy carriers has been shown to be such a toxicity-dependent environment for S. 346 aureus, this toxicity related fitness off-setting may explain why this resistance 347 mechanism is not more prevalent, and is readily displaced, once this antibiotic has 348 been removed from their environment. An effect we would have been unable to 349 explain had we not adopted a population-based approach to studying this major 350 bacterial pathogen. 351 352 353

Strains and growth conditions 355
All strains used in this study are listed in Supp. Table 5 As an essential gene ileS cannot be inactivated, and as a consequence, a homologous 365 recombination based mutational approach were unsuccessful. Therefore, to 366 generate isogenic mupirocin resistant and sensitive strains we utilised a selection 367 based method where an overnight culture of a mupirocin sensitive strain (e.g 368 SH1000 was plated onto agar plates with 4µg/ml mupirocin. This was incubated at 369 The strains were cultured individually overnight and diluted to 10 4 cfu/ml. For the 396 direct competitions 25µl of each diluted culture was added into 5ml fresh TSB, and 397 grown at 37°C with shaking for 24h. The mixed culture was diluted and plated onto 398 agar plates with and without 4µg/ml mupirocin and incubated at 37°C. The resulting 399 colonies were counted, and the number of colonies from the mupirocin plate was 400 subtracted from the count from no antibiotic plate. The Malthusian parameter was 401 calculated using the following formula: 402

Ln (final density (colony forming units (CFU)/ml) / starting density (CFU/ml)) 403
The Malthusian parameters of the mup S and mup R strains were compared using a 404 two tailed t-test. For the individual fitness assays the strains were cultured 405 individually overnight and diluted and added to 0.1X TSB made with phosphate 406 buffered saline rather than water to maintain the integrity of the horse blood cells to 407 which and 5% horse blood (Oxoid) was added. Bacterial growth was determined by 408 plating the cultures on TSA plates after 3, 6 and 24hrs of incubation at 37 o C in air. 409 410 Deletion of the agr locus from SH1000 and MY40. 411 Phage transduction was used to construct the mup R and mup S Agr mutants. In the S. 412 aureus strain ROJ48, the entire Agr locus has been replaced with an erythromycin 413 resistance cassette 28 . This was moved from ROJ48 into SH1000 by phage 414 transduction as follows: ROJ48 ϕ11 lysates were prepared from 200µl of overnight 415 ROJ48 culture in LK (1% Tryptone, 0.5% yeast extract, 1.6% KCl) which was added to 416 3ml of fresh LK and 3ml of phage buffer (10mM MgSO 4 , 4mM CaCl 2 , 50mM Tris-HCl 417 pH 7.8, 100mM NaCl and 0.1% gelatine powder in molecular/MiliQ water), and to 418 this 500µl ϕ11-RN6390B lysate was added. This was incubated at 30 o C shaking until 419 the media became clear, which indicated bacterial lysis. The lysates were then filter 420 sterilised, and a second round of lysis was carried out on ROJ48 with this first round 421 lysate. After these two lysis steps, transduction into SH1000 and MY40 strains was 422 performed by adding 200µl of overnight culture to 1.8ml LK with 10µl 1M CaCl 2 , and 423 500µl the ϕ11-ROJ48 lysate. This was incubated at 37 o C with shaking for 45 min, 424 then 1ml ice cold 20mM trisodium citrate was added and the transducing mixture High pH RP fractions were further fractionated using an Ultimate 3000 nanoHPLC 485 system in line with an Orbitrap Fusion Tribrid mass spectrometer (Thermo Scientific). 486 In brief, peptides in 1% (vol/vol) formic acid were injected onto an Acclaim PepMap 487 C18 nano-trap column (Thermo Scientific). After washing with 0.5% (vol/vol) 488 acetonitrile 0.1% (vol/vol) formic acid peptides were resolved on a 250 mm × 75 μm 489 Acclaim PepMap C18 reverse phase analytical column (Thermo Scientific) over a 150 490 min organic gradient, using 7 gradient segments (1-6% solvent B over 1min., 6-15% 491 B over 58min., 15-32%B over 58min., 32-40%B over 5min., 40-90%B over 1min., held 492 at 90%B for 6min and then reduced to 1%B over 1min.) with a flow rate of 300 nl 493 min −1 . Solvent A was 0.1% formic acid and Solvent B was aqueous 80% acetonitrile 494 in 0.1% formic acid. Peptides were ionized by nano-electrospray ionization at 2.0kV 495 using a stainless steel emitter with an internal diameter of 30 μm (Thermo Scientific) 496 and a capillary temperature of 275°C. 497 All spectra were acquired using an Orbitrap Fusion Tribrid mass spectrometer 498 controlled by Xcalibur 2.0 software (Thermo Scientific) and operated in data-499 dependent acquisition mode using an SPS-MS3 workflow. FTMS1 spectra were 500 collected at a resolution of 120 000, with an automatic gain control (AGC) target of 501 200 000 and a max injection time of 50ms. Precursors were filtered with an intensity 502 threshold of 5000, according to charge state (to include charge states 2-7) and with 503 monoisotopic precursor selection. Previously interrogated precursors were excluded 504 using a dynamic window (60s +/-10ppm). The MS2 precursors were isolated with a 505 quadrupole mass filter set to a width of 1.2m/z. ITMS2 spectra were collected with 506 an AGC target of 10 000, max injection time of 70ms and CID collision energy of 35%. 507 For FTMS3 analysis, the Orbitrap was operated at 50 000 resolution with an AGC 508 target of 50 000 and a max injection time of 105ms. Precursors were fragmented by 509 high energy collision dissociation (HCD) at a normalised collision energy of 60% to 510 ensure maximal TMT reporter ion yield. Synchronous Precursor Selection (SPS) was 511 enabled to include up to 5 MS2 fragment ions in the FTMS3 scan. 512

Proteomic data analysis 514
The raw data files were processed and quantified using Proteome Discoverer 515 software v2.1 (Thermo Scientific) and searched against the UniProt Staphylococcus 516 aureus strain NCTC 8325 database using the SEQUEST algorithm. Peptide precursor 517 mass tolerance was set at 10ppm, and MS/MS tolerance was set at 0.6Da. Search 518 criteria included oxidation of methionine (+15.9949) as a variable modification and 519 carbamidomethylation of cysteine (+57.0214) and the addition of the TMT mass tag 520 (+229.163) to peptide N-termini and lysine as fixed modifications. Searches were 521 performed with full tryptic digestion and a maximum of 2 missed cleavages were 522 allowed. The reverse database search option was enabled and all peptide data was 523 filtered to satisfy false discovery rate (FDR) of 5%. one of the ATP binding proteins of the PSM secretory system, Pmt, in the wild type 635 SH1000 strain causes a reduction in the abundance of the PSM in the S. aureus 636 supernatant. An Agr mutant has been provide as a control. A full length SDS-PAGE 637 gel has been provided in Supplementary material (Supp. Fig. 1), to illustrate why we 638 only provide a 'letter-box' snap-shot of the gels here. with and without a functioning Agr system by direct competition in TSB. There was 647 no difference in fitness in the Agr +ve background, but in the absence of Agr, the 648 mupirocin resistant strain was less fit that the mupirocin sensitive strain. b: The 649 effect of mupirocin resistance on relative fitness was determined by individual 650 culture in a nutrient poor environment (0.1X TSB) supplemented with 5% horse 651 blood. At the early stages of growth (0-6hrs) the mupirocin sensitive strain was more 652 fit, whereas between 6 and 24hr when the nutrients in the TSB were depleted and 653 cell lysis became necessary, the mupirocin sensitive strains was relatively more fit.