Idiosyncratic variation in the fitness costs of tetracycline-resistance mutations in Escherichia coli

A bacterium’s fitness relative to its competitors, both in the presence and absence of antibiotics, plays a key role in its ecological success and clinical impact. In this study, we examine whether tetracycline-resistant mutants are less fit in the absence of the drug than their sensitive parents, and whether the fitness cost of resistance is constant or variable across independently derived lines. Tetracycline-resistant lines suffered, on average, a reduction in fitness of almost 8%. There was substantial among-line variation in the fitness cost. This variation was not associated with the level of phenotypic resistance conferred by the mutations, nor did it vary significantly across several different genetic backgrounds. The two resistant lines with the most extreme fitness costs involved functionally unrelated mutations on different genetic backgrounds. However, there was also significant variation in the fitness costs for mutations affecting the same pathway and even different alleles of the same gene. Our findings demonstrate that the fitness costs of antibiotic resistance do not always correlate with the phenotypic level of resistance or the underlying genetic changes. Instead, these costs reflect the idiosyncratic effects of particular resistance mutations and the genetic backgrounds in which they occur.


51
Antibiotics are an essential component of modern medicine. Although they have dramatically 52 reduced the morbidity and mortality caused by severe bacterial infections, their benefits have 53 diminished in recent years because of their overuse in the clinic and in agriculture, which has led 54 to the evolution and proliferation of antibiotic-resistant pathogens. As a result, many infections 55 have become more difficult to treat with mainline drug therapies, and in severe cases, some 56 pathogenic strains have become resistant to all available drugs. An understanding of the forces 57 underlying and shaping antibiotic resistance is therefore critical to the future health of the human 58 population. 59 Bacteria can evolve resistance by either spontaneous mutations or horizontal acquisition 60 of resistance genes. Spontaneous mutations commonly confer resistance by altering the cellular 61 target of the antibiotic or increasing its efflux (Blair et al. 2015). Mechanisms associated with 62 horizontal gene transfer include target modification, drug detoxification, and the acquisition of 63 novel efflux pumps (Blair et al. 2015). In either case, resistant variants have a clear advantage 64 over their sensitive counterparts when exposed to the corresponding antibiotic. However, these 65 resistant types often suffer fitness costs because they disrupt the normal functioning of metabolic 66 pathways and physiological processes or increase the energetic burden on the cell ( Resistant types should therefore have lower growth rates than, and be outcompeted by, their 69 sensitive counterparts in the absence of drugs. 70 A resistant bacterium's competitive fitness, both in the presence and absence of a drug, is 71 an important factor that contributes to its ecological success and thus its clinical impact (Lenski 72 1997; Vogwill and MacLean 2015; Hughes and Andersson 2017). For example, the fitness of a 73 resistance mutation determines its likelihood of persisting in a bacterial population prior to drug 74 exposure, its maintenance in a population at a particular drug concentration, and its reversibility 75 when the antibiotic is reduced or removed from the environment ( The expected time required to reduce the frequency of a resistant mutant in a bacterial 78 population following the cessation of antibiotic use is inversely proportional to the fitness cost of 79 the resistance mutation (Lenski 1997). Although mathematical models can predict the rate of 80 these frequency declines (Levin et al. 1997), these theoretical expectations often are not met 81 under real-world scenarios for at least two reasons. First, some resistance mechanisms are 82 inherently cost free, at least in certain environments. Several mutations in the gene rpsL confer 83 resistance to streptomycin, but they have little or no fitness cost in both Escherichia coli and 84 Salmonella typhimurium (Tubulekas and Hughes 1993), and they even confer a competitive 85 advantage over wild-type strains in some animal infection models (Björkman et al. 1998;Enne et 86 al. 2005). These cost-free rpsL mutations are also found in streptomycin-resistant 87 Mycobacterium tuberculosis populations, where they may facilitate the long-term maintenance 88 of this resistant type (Böttger et al. 1998;Andersson and Hughes 2010). Similarly, treatment of 89 Helicobacter pylori infections with clarithromycin has been found to select for highly resistant 90 commensal Enterococcus species that persist for years after drug treatment (Sjölund et al. 2003). 91 This last outcome demonstrates a troubling side-effect of antibiotic use, in which the microbiome 92 can act as both a reservoir for resistance genes and as a conduit for their horizontal transfer to 93 pathogens (Sommer et al. 2010). 94 Second, pleiotropic costs associated with chromosomal-or plasmid-mediated resistance 95 can often be reduced or even eliminated through subsequent compensatory evolution (Bouma 96 Barrick et al. 2010). For example, clinically relevant levels of fluoroquinolone 98 resistance occur through the sequential substitution of mutations in several genes (Lindgren et al. 99 2003). Early genetic changes in the mutational pathway exact a cost on bacterial growth in both 100 laboratory media and mouse models, but the cost can be ameliorated through later resistance 101 mutations (Marcusson et al. 2009). Thus, evolution can restore a bacterial population's ancestral 102 growth rate in the absence of drug selection while simultaneously preserving resistance in the 103 event of future exposure to antibiotics. Moreover, compensatory evolution can sometimes drive 104 multidrug resistance; this outcome has been seen when a genetic change simultaneously provides 105 resistance to a newly imposed drug while reducing the fitness cost associated with resistance to a 106 previous antibiotic (Trindade et al. 2009). Compensatory evolution shows how pleiotropic 107 effects of one mutation can set the stage for epistatic interactions with subsequent mutations. 108 In general, a bacterium's genetic background can influence the fitness costs of antibiotic 109 resistance. For example, Vogwill and colleagues (2016) examined the costs of rifampicin-110 resistance mutations in the gene rpoB across several Pseudomonas species. They found that 111 some mutations vary in their fitness effects across backgrounds, and these costs correlate with 112 transcriptional efficiency. Thus, the same rpoB mutation can differentially affect transcriptional 113 efficiency depending on the genetic background, and these idiosyncratic effects in turn lead to 114 heterogeneity in costs. This work evaluated genetic-background effects across a fairly broad 115 phylogenetic scale, while focusing on mutations in a single gene. One can also ask whether 116 genetic background affects the fitness cost of resistance even among recently diverged clones of 117 a single species, and for resistance that has evolved through more diverse mutational pathways. 118 To address these issues, we evaluated the competitive fitness in the absence of drugs of 119 tetracycline-resistant clones that evolved from several different E. coli backgrounds, which 120 previously diverged during a long-term evolution experiment (LTEE). We ask several questions. 121 First, is there a fitness cost to resistance? Second, is the cost greater for mutants that evolved 122 higher levels of resistance (Fig. 1A)? Third, do fitness costs vary in an idiosyncratic manner that 123 does not depend on the level of resistance achieved (Fig. 1B)

EXPERIMENTAL CONDITIONS AND BACTERIAL STRAINS 145
The LTEE has been described in detail elsewhere (Lenski et al. 1991;Lenski 2017). In brief, 12 146 replicate populations of E. coli were founded from a common ancestral strain, called REL606 147 (Daegelen et al. 2009). These populations have been propagated for over 32 years and 73,000 148 generations by daily 100-fold dilutions in Davis Mingioli minimal medium supplemented with 149 25 μ g/mL glucose (DM25). 150 In this study, we examined the competitive fitness of tetracycline-resistant mutants that 151 evolved from the LTEE ancestor and clones sampled from four LTEE populations (denoted Ara-152 5, Ara-6, Ara+4, and Ara+5) after 50,000 generations. Specifically, we analyzed 4 mutants that 153 independently evolved from the ancestral background, and 3 mutants that evolved from each 154 derived background, for a total of 16 mutants (Table S1). We also used three clones as common 155 competitors: REL607, REL10948, and REL11638. REL607 is a spontaneous Ara + mutant of 156 REL606, the LTEE ancestor (Lenski et al. 1991). REL10948 is an Araclone isolated from the 157 Ara-5 population at 40,000 generations, and REL11638 is a spontaneous Ara + mutant of that 158 clone (Wiser et al. 2013;Lenski et al. 2015). The Ara marker is selectively neutral in the 159 glucose-limited medium; it serves to differentiate competitors during fitness assays because the 160 Araand Ara + cells form red and white colonies, respectively, on tetrazolium-arabinose (TA) 161 agar. We used REL607 as the common competitor for REL606 and the four tetracycline-resistant 162 clones derived from it. The 40,000-generation clones served as common competitors for the four 163 50,000-generation parental clones and twelve resistant mutants that evolved from them; using 164 these common competitors ensured that the differences in fitness were not so large that their 165 densities would fall below the detection limit during the fitness assays. 166

FITNESS ASSAYS 167
Assays were performed in the absence of antibiotics to assess the relative fitness of drug-resistant 168 mutants and their susceptible counterparts. Fitness was measured in an environment identical to 169 that of the LTEE, except the medium contained 250 μ g/mL glucose (DM250). Resistant mutants 170 and their sensitive parents each competed, in paired assays, against the same common competitor 171 with the opposite Ara-marker state (Fig. 2). To set up each competition assay, the competitors 172 were revived from frozen stocks, and they were separately acclimated to the culture medium and 173 other conditions over two days. The competitors were then each diluted 1:200 into fresh medium, 174 and a sample was immediately plated on TA agar to assess their initial densities based on colony 175 counts. The competition cultures were then propagated for 3 days, with 1:100 dilutions each day 176 in fresh medium. At the end of day 3, a sample was plated on TA agar to assess the competitors' 177 final densities. We quantified the realized growth rate of each competitor based on its initial and 178 final density and the net dilutions imposed (Lenski et al. 1991). We then calculated relative 179 fitness as the ratio of the realized growth rate of the clone of interest (either a resistant clone or 180 its sensitive parent) to that of the common competitor. Lastly, the fitness of a resistant mutant in 181 each assay was normalized by dividing it by the relative fitness of the paired assay obtained for 182 its parental strain. We performed a total of 80 pairs of fitness assays (160 competitions in total) 183 to produce 5 replicate estimates of the fitness of each of the 16 tetracycline-resistant mutants 184 relative to its sensitive parent. The relative fitness values were log e -transformed before the 185 statistical analyses reported in the Results below. competitors were mixed at an equal volumetric ratio in a common medium. These cultures were 199 propagated for three days in the absence of tetracycline by serial 1:100 transfers. We quantified 200 each competitor's realized growth rate from the initial and final densities after plating on TA 201 agar, taking into account the net dilution over the three days. These realized growth rates were 202 then used to calculate the fitness of a resistant line relative to its sensitive parent (see Materials  203 and Methods

OF THE ANTIBIOTIC 207
We ask first whether tetracycline resistance is costly, on average, in the absence of the drug. The 208 grand mean of the log e -transformed fitness of the 16 resistant lines relative to their paired 209 parental strains is -0.0771, indicating that the resistant mutants grow ~7.7% more slowly than 210 their sensitive counterparts during head-to-head competitions with a common competitor. This 211 value differs significantly from the null hypothesis that the resistant lines and their sensitive 212 parents are equally fit (t s = 2.9973, 15 d.f., one-tailed p = 0.0045). 213

COST OF RESISTANCE VARIES AMONG RESISTANT MUTANTS 215
We measured the relative fitness of each resistant line with 5-fold replication. This replication 216 allows us to test whether the variation in fitness among the 16 tetracycline-resistant lines is 217 simply measurement noise or, alternatively, reflects genetic variation in the cost of resistance. 218 Table 1 shows the analysis of variance (ANOVA). The variation among the 16 lines is about 10-219 fold greater than expected from the variation between replicate assays performed on the same 220 line (F 15,64 = 10.34, p << 0.0001). 221  Without a much larger number of resistant lines, it is not possible to rigorously disentangle these 277 various sources of idiosyncratic fitness costs. However, by examining and contrasting specific 278 cases, we are able to shed light on some of the sources of these differences. 279 Two resistant clones, Ara+4-3 and Ara+5-2, have fitness costs that are very similar to one 280 another, but more than double the cost of any of the other 12 resistant mutants (Fig. 3). Yet these 281 two cases occurred on different genetic backgrounds and have different mutations. Ara+4-3 has 282 mutations in hns, which encodes a histone-like global regulator, and lpcA, which encodes a 283 phosphoheptose isomerase; Ara+5-2 has a single mutation in ompF, which encodes an outer-284 membrane porin (Card et al. 2020). We asked whether these two extreme cases are solely 285 responsible for the heterogeneity in fitness costs by performing an ANOVA that excludes them. 286 The variation in fitness costs among the other 12 clones is reduced, but it nonetheless remains 287 highly significant (F 11,48 = 4.44, p = 0.0001). 288 289 Figure 3. Fitnesses of 14 tetracycline-resistant mutants relative to their parental strains. The 290 mutants are arranged from lowest to highest fitness. Each symbol shows the mean log e -291 transformed fitness, based on 5-fold replication of paired fitness assays. Error bars show 95% 292 confidence limits calculated using the t-distribution with 4 d.f. and the pooled standard deviation 293 estimated from the ANOVA (Table 1). Letters above the error bars identify mutants with relative 294 fitnesses that are not significantly different, based on Tukey's "honestly significant difference" 295 test for multiple comparisons. 296 Nine of the 14 resistant clones have a single mutation each, while four of them (Ara-5-2, 297 Ara-6-2, Ara+4-3, Ara+5-3) have two mutations, and another (Ancestor-2) has three mutations 298 (Card et al. 2020). It is reasonable to imagine that in each clone one mutation confers the drug 299 resistance, while the others merely hitchhiked with the resistance mutation. Such hitchhikers 300 might include deleterious mutations that reduce fitness. Therefore, we compared the fitness costs 301 for the resistant clones with and without secondary mutations. The average fitness cost of the 302 clones with multiple mutations is higher (13.8%) than the average of those with single mutations 303 In previous work, we examined the role that genetic background plays in both the phenotypic 322 and genotypic evolution of antibiotic resistance. First, we examined the potential of several 323 different LTEE backgrounds to evolve increased resistance to several antibiotics. We found that 324 evolvability was idiosyncratic with respect to the parental genotype, such that resistance was 325 more constrained in some backgrounds than in others (Card et al., 2019). Genetic differences 326 will accumulate between populations, even if they evolve in the same permissive environment. 327 These differences can unpredictably alter their ability to respond evolutionarily when challenged 328 with antibiotics. Second, we sequenced the complete genomes of some of these resistant mutants 329 and assessed whether the different initial genotypes took similar or divergent mutational paths to 330 increased resistance (Card et al. 2020). Again, we found that the initial genetic background is 331 important. On average, the replicate lines that evolved from the same founding genotypes had 332 more gene-level mutations in common than lines derived from different founding genotypes. 333 The aim of this study was to examine whether and how genetic background influences 334 the fitness effects of resistance mutations in the absence of antibiotic. In particular, we examined 335 the fitness costs of tetracycline resistance in 16 lines that evolved from five sensitive parental 336 backgrounds. We found that the resistant lines are, on average, less fit than their sensitive 337 counterparts in the absence of the antibiotic. This result is not surprising, given that resistance 338 mutations often disrupt the normal function of metabolic or physiological processes, or impose 339 energetic demands that reduce growth and competitiveness (Andersson and Hughes 2010). We 340 also observed highly significant variation among the resistant lines in their fitness costs (Table  341 1). This variation remained substantial (Fig. 3) even after we excluded two strains without 342 identified mutations (Card et al. 2020). These two strains exhibited phenotypic resistance in our 343 earlier work (Card et al. 2019), but that resistance might have been conferred by unstable 344 genomic changes, such as gene amplifications or frameshift mutations in homopolymeric tracts 345 that can cause "phase variation" (Moxon et al. 1994). If so, these unstable changes could have 346 reverted prior to the genomic analysis and the competition assays that we performed. 347 We then addressed two broad possibilities regarding the variation in fitness cost between 348 the 14 lines with known, stable mutations. First, we asked whether there is a relation between a 349 line's phenotypic resistance and its fitness cost, such that mutations that confer greater resistance 350 are more costly (Fig. 1A). A meta-analysis of fitness costs across several species and drug 351 classes by Melnyk and colleagues (2015) supported this association, and the authors suggested it 352 could be understood from evolutionary and mechanistic perspectives. Imagine a population that 353 is well-adapted to one environment and hence near a local fitness optimum. If the environment 354 changes, such as with the addition of an antibiotic, then the population may evolve toward a 355 different optimum through the substitution of new mutations. Mutations of large effect will bring 356 the population closer to this new optimum than mutations of small effect. However, if the 357 environment later reverts to its previous state, then populations that substituted the large-effect 358 mutations will be further from their previous optimum than those populations that acquired 359 small-effect mutations. From a mechanistic standpoint, the increased expression of efflux pumps 360 or drug targets diverts resources from other cellular processes. Also, resistance mutations that 361 change evolutionarily conserved proteins are more likely to disrupt their functions than improve 362 them. In our study, however, there was no significant association between fitness costs and the 363 level of resistance conferred by mutations, whether on an absolute basis or relative to the parent 364 strain. 365 The second broad possibility is that the fitness costs of resistance can vary for reasons 366 unrelated to the level of resistance conferred (Fig. 1B). There are several potential reasons for 367 such idiosyncratic variation. One possibility is that the same resistance mutation may have 368 different fitness costs in different genetic backgrounds. In Campylobacter jejuni, for example, a 369 C257T mutation in the gene gyrA confers fluroquinolone resistance. When fluroquinolone-370 resistant and -susceptible strains were inoculated separately into chickens, they colonized equally 371 well and each persisted even in the absence of drug exposure (Luo et al. 2005). However, when 372 resistant and sensitive strains were co-inoculated, the resistant variants often prevailed. Further 373 work indicated that this particular gyrA mutation was beneficial in some genetic backgrounds, 374 even in the absence of antibiotic, and costly in others (Luo et al. 2005). In our study, by contrast, 375 the variation in fitness costs among strains was not explained by genetic-background effects, but 376 instead involved several other factors. 377 One such factor is that resistance mutations can occur in different genes, which can lead 378 to different fitness costs. In this study, the relative fitnesses of clones Ara+4-3 and Ara+5-2 were 379 significantly lower than the other 12 strains. Another factor is that mutations in different genes that are part of the same physiological 393 pathway may confer similar resistance levels but have different fitness costs. In our study, four 394 tetracycline-resistant lines derive from the same LTEE ancestor: one had a mutation in envZ, 395 while the other three had mutations in ompR. These genes encode proteins that comprise a two-396 component regulatory system that regulates cellular responses to osmotic stress, and which 397 affects antibiotic resistance through altered expression of the major porin OmpF (Chakraborty 398 and Kenney 2018; Choi and Lee 2019). We observed significant heterogeneity in fitness even 399 among these lines, implying that different changes within this one pathway can impose unique 400 burdens. The evolution of carbapenem resistance in E. coli K12 can also occur by mutations in 401 this same two-component system, again with variable fitness costs (Adler et al. 2013). In their 402 study, Adler and colleagues (2013) found that envZ mutants had no measurable loss of fitness in 403 the absence of antibiotic, whereas ompR mutations suffered a large cost. By contrast, in our study 404 the envZ mutation was more costly, which may reflect differences between the E. coli fitness costs are lower. Given that the cost may depend on the particular mutation and its genetic 441 background, the time to treatment failure is harder to predict. We think that these issues and their 442 relevance for treatment options are important avenues for future research. 443