The Evolution of Fluoroquinolone-Resistance in Mycobacterium tuberculosis is Modulated by the Genetic Background

Fluoroquinolones (FQ) form the backbone in experimental treatment regimens against drug-susceptible tuberculosis. However, little is known on whether the genetic variation present in natural populations of Mycobacterium tuberculosis (Mtb) affects the evolution of FQ-resistance (FQ-R). To investigate this question, we used a set of Mtb strains that included nine genetically distinct drug-susceptible clinical isolates, and measured their frequency of resistance to the FQ ofloxacin (OFX) in vitro. We found that the Mtb genetic background led to differences in the frequency of OFX-resistance (OFX-R) that spanned two orders of magnitude and substantially modulated the observed mutational profiles for OFX-R. Further in vitro assays showed that the genetic background also influenced the minimum inhibitory concentration and the fitness effect conferred by a given OFX-R mutation. To test the clinical relevance of our in vitro work, we surveyed the mutational profile for FQ-R in publicly available genomic sequences from clinical Mtb isolates, and found substantial Mtb lineage-dependent variability. Comparison of the clinical and the in vitro mutational profiles for FQ-R showed that 45% and 19% of the variability in the clinical frequency of FQ-R gyrA mutations in Lineage 2 and Lineage 4 strains, respectively, can be attributed to how Mtb evolves FQ-R in vitro. As the Mtb genetic background strongly influenced the evolution of FQ-R in vitro, we conclude that the genetic background of Mtb also impacts the evolution of FQ-R in the clinic. Significance Newer generations of fluoroquinolones form the backbone in many experimental treatment regimens against M. tuberculosis (Mtb). While the genetic variation in natural populations of Mtb can influence resistance evolution to multiple different antibiotics, it is unclear whether it modulates fluoroquinolone-resistance evolution as well. Using a combination of in vitro assays coupled with genomic analysis of clinical isolates, we provide the first evidence illustrating the Mtb genetic background’s substantial role in fluoroquinolone-resistance evolution, and highlight the importance of bacterial genetics when studying the prevalence of fluoroquinolone-resistance in Mtb. Our work may provide insights into how to maximize the timespan in which fluoroquinolones remain effective in clinical settings, whether as part of current standardized regimens, or in new regimens against Mtb.

encode DNA gyrase (22,38,39). No horizontal gene-transfer (HGT) or plasmid-based resistance 92 to FQ has been documented in Mtb (44,45). Studying FQ-R evolution in Mtb populations thus 93 provides a promising setting for elucidating how the genetic background may affect the 94 emergence and maintenance of clinically relevant chromosomal AMR mutations in bacterial 95 populations. 96 While a great deal of literature exists on the biochemical mechanisms leading to the FQ-R 97 phenotype in Mtb (10, 41-43, 46, 47), little is known on the evolutionary dynamics of FQ-R in 98 different populations of Mtb. Given that antimicrobial regimens against Mtb infections use 99 We first tested for the presence of differential mutation rates between our panel of Mtb 161 strains in Fig. 1A. Mutations in dnaE, which encodes the replicative DNA polymerase and serves 162 as the major replicative exonuclease in Mtb, have been shown to confer a hypermutator 163 phenotype in Mtb in the absence of environmental stress (50,51). While dnaE mutations were 164 present in the genomic data of our panel of drug-susceptible Mtb strains (See SI Appendix, Table  165 S2) (49), none were in the polymerase and histidinol phosphatase domain of DnaE, the region 166 where mutations would impart a hypermutator phenotype (50, 51). We did not test for the 167 presence of dnaE mutations in the resistant colonies following the fluctuation analysis, as we 168 reasoned that the likelihood of gaining both a dnaE and a gyrA double mutation within this 169 relatively short period is extremely low as to be considered negligible. To test for mutation rate 170 variation in vitro, we again conducted a fluctuation analysis on N0157, N1283, and N0145 (the 171 high-, mid-, and low-frequency OFX-R strains, respectively), but used streptomycin (STR; 100 172 µg/mL) instead of OFX. We hypothesized that if the frequency of OFX-R is driven by 173 differential mutation rates, then we should expect similar differences in the frequency of STR-174 resistance (STR-R). However, we observed no evidence for differences in the frequency of STR-175 99.7% of the QRDR mutations observed (645 gyrA mutations, 2 gyrB mutations), and no QRDR 184 double-mutants were present (See SI Appendix, Tables S4-S5). The mutational profiles for OFX-185 R were also highly strain-specific ( Fig. 3A, P = 5.00 × 10 -4 , Fisher's exact test). Specifically, the 186 GyrA A90V mutation was most prevalent in the high-frequency OFX-R strains, while GyrA 187 D94G was most prevalent in all other strains. There was also a slight trend showing that strains 188 with a greater number of unique gyrA mutations present also had higher rates of OFX-R ( Fig. 1A; 189 The strain-dependent variation the mutational profile for OFX-R may be due to gyrA 191 mutations conferring different resistance levels depending on the Mtb strain they are present in. 192 To test this hypothesis, we first isolated OFX-R mutants carrying one of four possible GyrA 193 mutations (G88C,A90V,D94G,or D94N) in the three strains used in Fig. 2: N0157, N1283, and 194 N0145. The OFX MIC was determined for each of the twelve OFX-R mutant strains, along with 195 their respective wild-type ancestors. We found that each parental wild-type strain had different 196 susceptibilities to OFX, with N0157, N1283, and N0145 having OFX MICs of 2 µg/mL, 0.6 197 µg/mL, and 0.5 µg/mL, respectively ( Fig. 4A; See SI Appendix, Table S6). This was consistent 198 with the fluctuation analysis results shown in Fig. 1B. Furthermore, we observed that the OFX 199 MIC conferred by a given gyrA mutation varied depending on the strain it was present in (Fig. 200 4B; See SI Appendix, Table S6). For example, mutants in the N0157 strain generally had higher 201 OFX MICs than mutants in either the N0145 or N1283 strains. The only mutation that deviated 202 from this trend was GyrA G88C, which conferred a higher OFX MIC when in the N0145 strain. 203 Notably, the GyrA A90V mutation conferred a resistance level equal to or greater than 4 µg/mL 204 fluctuation analysis using 4 µg/mL OFX ( Fig. 1A; Fig. 3). In summary, the differences in OFX 207 MIC reflected the strain-dependent mutational profiles for OFX-R in Mtb, as expected. 208 209 Fitness of ofloxacin-resistance mutations are associated with their relative frequency in vitro 210 While the OFX MICs may determine which mutations may be observed in a fluctuation 211 analysis, it is not the sole parameter to influence the OFX-R mutational profile for a given strain. 212 We found that while the same gyrA mutation can be observed in two different Mtb strains, their 213 relative frequencies may vary (Fig. 3). This variation may be due to the fitness of a given gyrA 214 mutant being different across genetic backgrounds. To test this hypothesis, we used cell growth 215 assays in antibiotic-free conditions to measure the in vitro fitness of our panel of OFX-R mutants 216 relative to their respective parental wild-type ancestors. We observed that the relative fitness of 217 the OFX-R mutants was modulated by both the gyrA mutation and the Mtb strain they were 218 present in (Fig. 5A; See SI Appendix, Fig. S2-S3, Table S7). Furthermore, there was a positive 219 association between the fitness of a given gyrA mutation with its relative frequency in the 220 fluctuation analysis for the N0157 and N1283 strains ( Fig. 5B, P = 0.03 for N0157, P = 0.05 for 221 N1283). There was no evidence of an association in the N0145 background due to the lack of 222 GyrA G88C and A90V mutants in its fluctuation analysis. 223 The results from Fig. 4 and Fig. 5, as well as the apparent lack of mutation rate 224 differences between our strains (Fig. 1C), suggested that differential mutational profiles was an 225 important contributor in the variation in OFX-R frequency in Mtb. These mutational profile 226 differences appear to be driven by the Mtb genetic background's effect on both the MIC and the 227 relative fitness cost of OFX-R mutations. We next explored whether these in vitro results would 228 be relevant in clinical settings.

Mutational profile for fluoroquinolone-resistance in vitro reflects clinical observations 231
To explore the clinical relevance of our in vitro work, we surveyed the FQ-R mutational 232 profile from publicly available Mtb genomes obtained from clinical isolates. FQs are generally 233 used for treatment against MDR-TB (29). While it is unclear whether resistance mutations for 234 isoniazid (INH) and/or rifampicin (RIF) predispose a strain to become FQ-R, the prevalence of 235 FQ-R is heavily biased towards MDR-TB strains due to treatment practices. We therefore based 236 our analyses on a collated dataset of 3,452 publicly available MDR-TB genomes (See SI 237 Appendix, Table S8), which we confirmed to be MDR-TB based on the presence of known INH-238 and RIF-resistance mutations. This dataset provided a reasonable sampling of the overall genetic 239 diversity of Mtb, as six of the seven known phylogenetic Mtb lineages were represented 240 (Lineages 1 -6) (17, 30). We catalogued their FQ-R mutational profiles, and found 950 FQ-R 241 mutations in 854 genomes (See SI Appendix, Tables S9-S10), showing that multiple FQ-R 242 mutations may be present in the genome of a single Mtb clinical isolate. The frequency of FQ-R 243 differed between lineages, with the highest frequencies present in L2 and L4 strains (P < 2.2 × 244 10 -16 , Chi-square Goodness of Fit Test). Moreover, we noticed a lineage-dependent mutational 245 profile for FQ-R ( Fig. 6, P = 3.00 × 10 -5 , Fisher's exact test; See SI Appendix, Fig. S4, Tables 246 S10-S11). For example, while the GyrA D94G mutation was most prevalent in strains belonging 247 to L1, L2, and Lineage 3 (L3), the GyrA A90V mutation was most prevalent in L4 and Lineage 6 248 (L6).  Table S10). The relative frequencies of 253 gyrA mutations for each genetic background in vitro were also similar to their relative 254 frequencies in the clinic. We compared the frequency of gyrA mutations from the OFX-R 255 mutational profile assay in Fig. 3 to our genomic data survey in Fig. 6, but limited it to L2 and L4 256 strains (the two lineages with the highest clinical frequencies of FQ-R). We observed a positive 257 association between the frequency of a given gyrA mutation in our fluctuation analysis compared 258 to the frequency in the clinic, with the association being significant for L2 strains ( with the study conducted here, highlight the importance of the genetic background when testing 288 for the frequency of AMR in Mtb. Furthermore, these results show that differential DNA 289 mutation rate is not the only parameter relevant in determining the frequency of FQ-R in Mtb. 290 If DNA mutation rates do not contribute to the variation in OFX-R frequency, we 291 hypothesized that differences in the phenotypic effects of OFX-R mutations, and their consequent 292 effect on the mutational profiles for OFX-R, may be important contributors. By sequencing the 293 QRDR from resistant colonies in our OFX fluctuation analysis, we observed strain-specific 294 patterns in the mutational profiles for OFX-R (Fig. 3). This suggested that the mutational profile 295 for FQ-R is not only a function of the FQ type and concentration (10,14,47,53), but that 296 epistatic interactions between a given FQ-R mutation and the genetic background may also play a spp. (16,27), M. smegmatis (54), and Mtb (24, 28, 31), where a given RIF-R rpoB mutation 299 conferred differential MIC and fitness costs depending on the genetic background it occurred in, 300 or on the presence of other AMR mutations. In line with these previous studies, we found that the 301 OFX MIC and the fitness effect conferred by a given gyrA mutation varied significantly 302 depending on the Mtb genetic background they occur in (Fig. 4; Fig. 5A; See SI Appendix, Table  303 S6). These results support the hypothesis that epistasis plays a role in determining the strain-304 dependent OFX-R frequencies and mutational profiles observed during our fluctuation analyses 305 ( Fig. 3; Fig. 5B). 306 These epistatic interactions may have clinical consequences. A recent study has shown 307 that drug-susceptible Mtb strains with higher MICs to INH and RIF were associated with 308 increased risk of relapse following first-line treatment (20). Specific FQ-R gyrA mutations have 309 also been associated with poorer treatment outcomes in MDR-TB patients (40, 55). Considering 310 our observation that the Mtb genetic background affected both the OFX MICs and OFX-R 311 mutational profiles ( Fig. 3; Fig. 4; See SI Appendix, Tables S4-S6), the genetic background may 312 therefore contribute to differences in patient treatment outcomes when using FQs as first-line 313 drugs. 314 Using publicly available genomic data from Mtb clinical isolates, we observed significant 315 lineage-dependent variation in the frequency of and mutational profiles for FQ-R (Fig. 6). As 316 expected, the vast majority of FQ-R mutations were observed in gyrA (10,22,38,39,(41)(42)(43). 317 FQ-R was also most frequent in L2 and L4. This was also as expected, as strains from the L2 R frequencies observed in these two lineages. Furthermore, we observed that almost half of the 322 variability in the clinical frequency of gyrA mutations of L2 strains can be explained by how Mtb 323 evolves in vitro (Fig. 7). However, the in vitro FQ-R evolution could only account for 19% of the 324 variability for gyrA mutation frequencies in clinical L4 strains. This suggested that while the Mtb 325 genetic background can influence the evolution of FQ-R in the clinic, other factors (which may 326 be independent of the Mtb genetic background) likely played strong roles as well. 327 Epidemiological factors including socioeconomic disruptions, health system inefficiencies, and 328 human behaviour are well known risk factors for the emergence and transmission of AMR in Mtb 329 (3-7). Meanwhile, biological factors not explored in this study, such as antibiotic type and 330 concentration (10-13, 46, 47), pharmacodynamic and pharmacokinetic features (58, 59), and the 331 selective pressure of the host immune system (60), may also influence the evolution of FQ-R. 332 Our study is limited by the fact that our survey of clinical FQ-R frequencies involved a 333 genomic dataset that was sampled by convenience. This dataset was used due to its public 334 availability, and may not be fully representative of FQ-R frequencies in Mtb populations. We 335 noted that lineage-specific frequencies of FQ-R were likely biased due to the overrepresentation 336 Another limitation of our study is that fluctuation analyses only model AMR emergence. 355 Long-term population dynamics also play an important role in AMR evolution (8,12,14). For 356 example, population bottleneck events modulate AMR evolution during serial transfer 357 experiments (14,27,63,64), and have also been hypothesized to strongly influence Mtb 358 evolution in the clinic (65). Thus, modeling FQ-R evolution in Mtb in epidemiological settings 359 would benefit from the use of some measure of long-term population dynamics and between-host 360 transmission. Nevertheless, the fitness of AMR mutants is an important factor in determining its 361 evolutionary fate (8,9,12,14,26,54,64) and its potential for between-host transmission (63, 66, 362 67). Considering that the Mtb genetic background modulated the fitness effect of FQ-R mutations 363 ( Fig. 5; See SI Appendix, Table S7), the genetic background may modulate how likely FQ-R 364 mutants transmit between patients. 365 In conclusion, we illustrate how the genetic variation present in natural populations of Mtb genetic variants (17, 30), our work suggests that there may be regional differences in the rate 368 of FQ-R emergence and FQ-R prevalence when using FQs as a first-line drug. We therefore 369 highlight the importance of bacterial genetics in determining how FQ-R evolves in Mtb and, in 370 general, how AMR evolves in pathogens. 371 372 373

Collection of drug-susceptible clinical isolates of M. tuberculosis strains for in vitro studies 375
We used nine genetically-distinct Mtb strains, with three strains from each of the 376 following Mtb lineages: Lineage 1 (L1; also known as the East-Africa and India Lineage), 377 Lineage 2 (L2; the East Asian Lineage), and Lineage 4 (L4; the Euro-American Lineage) (17, 378 68). All strains were previously isolated from patients, fully drug-susceptible, and previously 379 characterized by Borrell et al. (49) (See SI Appendix, Table S1). 95% confidence intervals for each F were calculated as previously described by Rosche & Foster 409 (69). Hypotheses testing for significant differences between the r dist between strains for the 410 fluctuation analyses at 4 µg/mL of OFX (Fig. 1A) and at 100 µg/mL of STR (Fig. 1C) were 411 performed using the Kruskal-Wallis test; significant differences in the r dist between strains in the 412 fluctuation analyses at 2 and 8 µg/mL (Fig. 1B) were tested for using the Wilcoxon rank-sum 413 test. Statistical analyses were performed using the R statistical software (70). 414 415

Determining the mutational profile for ofloxacin-resistance in vitro 416
From the parallel cultures plated on 4 µg/mL of OFX (Fig. 1A), up to 120 resistant 417 colonies per strain (at least 1 colony per plated parallel culture if colonies were present, to a 418 maximum of 6) were transferred into 100 μ L of sterile deionized H 2 O placed in Falcon® 96-well Clear Microplate (Corning Inc.). The bacterial suspensions were then heat-inactivated at 95ºC for 420 1 h, and used as PCR templates to amplify the QRDR in gyrA and gyrB using primers designed 421 by Feuerriegel et al. (71). PCR products were sent to Macrogen, Inc. or Microsynth AG for 422 Sanger sequencing, and QRDR mutations were determined by aligning the PCR product 423 sequences against the H37Rv reference sequence (72). Sequence alignments were performed 424 using the Staden Package, while the amino acid substitutions identification were performed using 425 the Molecular Evolutionary Genetics Analysis Version 6.0 package. Fisher's exact test was used 426 to test for significant differences between the strains' mutational profiles for OFX-R. Data 427 analyses were performed using the R statistical software (70). (approximately 14 to 21 days). Resistant colonies were picked and re-suspended in fresh 10 mL 439 7H9 ADC, and incubated at 37ºC. Once the culture reached early stationary phase, two aliquots 440 were prepared. The first aliquot was heat-inactivated at 95ºC for 1 h, and the gyrA mutation 441 identified by PCR and Sanger sequencing, as described in the mutational profile for OFX-R assay. If the first aliquot harboured one of four OFX-r gyrA mutations (GyrA D94G , GyrA D94N , 443 GyrA A90V , or GyrA G88C ), the second aliquot was stored in -80ºC for future use. 444 Prior to further experimentation with the spontaneously OFX-R mutant strains, starter 445 cultures were prepared in the same manner as for the drug-susceptible strains. We set up three or four 1,000 mL roller bottles with 90 mL of 7H9 ADC and 10 mL 469 borosilicate beads. Each bottle was inoculated with a volume of starter cultures so that the initial 470 bacterial density was at an OD 600 of 5 × 10 -7 . The inoculated bottles were then placed in a roller 471 incubator set to 37ºC, and incubated for 12 to 18 days with continuous rolling. OD 600 472 measurements were taken once or twice every 24 hours. Two independent experiments in either 473 triplicates or quadruplicates were performed per strain. 474 We defined the exponential phase as the bacterial growth phase where we observed a 475 log 2 -linear relationship between OD 600 and time; specifically, we used a Pearson's R 2 value on the presence of both isoniazid (INH)-and rifampicin (RIF)-resistance mutations. This 488 provided a dataset of 3,452 genomes with confirmed MDR-TB; their accession numbers are 489 reported in Table S8 (See SI Appendix). These MDR-TB genomes were then screened for the 490 presence of FQ-resistance mutations, and we identified 854 genomes that were classified as FQ-491

R. 492
The INH-, RIF-, and FQ-resistance mutations used for screening are the same mutations 493 used by Payne,Menardo et al. (75), and are listed in Takiff HE, et al. (1994) Cloning and nucleotide sequence of Mycobacterium tuberculosis 597 gyrA and gyrB genes and detection of quinolone resistance mutations. Antimicrob Agents