Obtaining bacterial strains with varied resistance levels
We experimentally evolved 24 replicate lineages of E. coli to tolerate increasing concentrations of chloramphenicol. By serially passaging bacterial cultures through 14 increasing chloramphenicol levels, we obtained 336 (24 lineages × 14 concentrations) populations of E. coli across a gradient of resistance levels (Fig. 2). The 24 replicate lineages enabled us to study the variability arising from the stochastic nature of mutation acquisition. We refer to these populations as “cultures” rather than “strains” due to the possible coexistence of multiple genotypes.
Resistance incurs costs in both thermal tolerance and maximum growth rate
We measured growth rates of experimentally evolved E. coli cultures at three different temperatures: their historic temperature of 37 °C, and the novel temperatures of 32 °C and 42 °C. We hypothesized that growth rate costs of resistance would be larger in the novel temperatures, consistent with reduced thermal niche breadth.
Overall, we found the growth rates decreased strongly with increasing antibiotic resistance (Fig. 3A). We then calculated relative growth rates for each lineage by dividing the growth rate at each timepoint by the growth rate of the culture at timepoint 1 (T1) at the appropriate temperature (e.g., all cultures at 32 °C were standardized by the ancestral growth rate at 32 °C). Analysis of these relative growth rates showed that there was both a fitness cost in maximum growth rate and a fitness cost in thermal niche breadth; the linear model showed a strong negative effect of increasing resistance on growth rate at 37 °C (F1, 974 = 988.2, p < 0.001), and significant variability in the effect of resistance on growth at the three different temperatures (F1, 974 = 13.8, p < 0.001).
Absolute growth rates (measured as the maximum difference in sequential log-transformed optical densities) varied across the three temperatures, and were generally highest at 42 °C for a given culture (A). Increasing resistance, as measured by timepoint in the evolution experiment (darker colors indicate higher resistance levels), resulted in reduced maximum growth rates (A). These differences are more apparent when analyzing relative growth rates (i.e. the growth rates divided by the respective ancestral growth at T1 at each respective temperature) (B). As resistance increased, the relative growth rate decreased at 37 °C, but showed even larger declines at the novel temperatures of 32 °C and 42 °C (B). Of the three scenarios depicted in Fig. 1, these findings mirror Fig. 1C, where there are costs to both maximum growth rate and thermal niche breadth. Points show the mean and standard error of growth rates from measurements from each of the 24 lineages.
The negative effect of resistance on growth rate was greater at 32 °C and 42 °C than at 37 °C (Fig. 3B). Thus, there were disproportionate fitness costs in both the novel temperatures. This finding is corroborated by counting the number of evolved (T2 or greater) populations that showed an increased growth rate, when compared with their ancestor (T1); at 37 °C, there were 41 evolved cultures with growth rates greater than that of the ancestor at 37 °C, whereas at 32 °C there were 13 evolved cultures that grew faster than the ancestor at 32 °C, while at 42 °C only two did. Therefore, the presence of fitness costs was more consistent in the novel temperatures.
Strong competitive disadvantage of resistance at increased temperature
Next, we evaluated the effects of increasing chloramphenicol resistance on competitive success, measured by the fraction of the community comprised by a resistant strain when grown in mixed culture with a more sensitive strain. We transformed strains from 96 cultures (timepoints 1, 5, 9, and 13 for each of the 24 lineages) with either GFP or mCherry plasmids and quantified population sizes after competition assays using flow cytometry. We found that resistance level was a strong driver of the composition of mixed cultures in the competition experiments (Fig. 4). For example, at 32 °C and 37 °C, the most sensitive strains (T1) grew to a population 1.3x larger when grown against the most resistant strain (T13), as compared with being grown against the same timepoint (T1). The growth differential was much stronger at 42 °C, where the most sensitive strain (T1) grew to 3x their population size when competed against the most resistant strain (T13).
Heatmaps show the performance of two strains in competition experiments at the 3 temperatures, with cell values giving the mean ratio of a strain’s population frequency after competition with another strain vs. competition with itself. The axes are the timepoints from which each strain was sampled. A value of 1 indicates no differential performance. Purple cells (values larger than 1) indicate that Strain 1 out-performed in final community composition, as compared with when that same strain was competed against a strain from the same timepoint, but sampled separately (and labeled with a different fluorophore, mCherry for Strain 2 vs. GFP for Strain 1). In contrast, darker pink cells (values smaller than 1) indicate that the first strain performed worse than in competition with the same timepoint. At 42 °C (C), the there was a greater differential in frequencies than at 32 °C or 37 °C (A, B), indicated by the darker colors (note the different scale bars between 42 °C and the other two temperatures).
For each lineage, we evaluated whether the competitive cost of resistance was different at 32 °C or 42 °C when compared with the effect at 37 °C. We found that 12 of the 24 lineages had significantly stronger negative effects of resistance on strain growth at 42 °C, as compared with 37 °C. One lineage (L16) had significantly smaller costs of resistance at 42 °C. There were no lineages where there was a significant difference in the effect of resistance between 37 °C and 32 °C. Full results can be found in Table S1. When removing the interaction between temperature and lineage to evaluate the average effect of temperature across all lineages, we found that the effect of resistance differed across temperatures (F2, 3309 = 51.3, p < 0.001). Specifically, the negative effect was greater at 42 °C than 37 °C (a slope 59% greater at 42 °C), though there was no significant difference between 32 °C and 37 °C (a slope 11% weaker at 32 °C). To visualize these differences in the effect of resistance on strain growth, we show two lineages with contrasting results; lineage 5 had no significant differences in the effect of resistance across temperature, whereas lineage 19 showed a much stronger effect of resistance differences at 42 °C than at the other two temperatures (Fig. 5).
While many lineages demonstrated a greater cost of resistance at 42 °C than the other two temperatures, some lineages showed no increased effect at the warmer temperature. Lineage 5 (A) is an example of an evolutionary trajectory where fitness costs were not substantially different across temperatures. The x-axis is the difference in experimental timepoints at which strains were collected (a proxy for minimum resistance level). The y-axis gives the change in performance of the GFP strain as compared with competition against the same timepoint. Conversely, lineage 19 (B) shows an example of a strong interactive effect of fitness costs and temperature, where strain performance is more impacted by resistance at 42 °C (red circles) than at the other two temperatures.
Variability in routes to resistance among lineages
We then sequenced genomes for the 96 strains used in the competition experiments to evaluate how the genetic mechanism of resistance affected fitness. All samples had coverage of at least 100x, with a minimum number of reads per genome of 2.4 million. Using the genomic data, we identified 220 mutations across these 96 strains; of these mutations, 24 occurred within strains at T1, 44 in strains at T5, 65 in strains at T9, and 87 in strains at T13. We also identified the timepoint at which these mutations were first identified within a lineage. All 24 mutations at T1 were new when comparing against the ancestor, along with 29 mutations present at T5 but not the same lineages at T1, 26 mutations present at T9 but not T5, and 40 mutations present at T13 but not T9. We saw that most mutations fell within genes for known resistance mechanisms, such as efflux pumps, the multiple antibiotic resistance protein (mar), or ion channels. We grouped the mutations into 10 categories for further analysis: those associated with the acr efflux pump (64 mutations), ATP synthase (42 mutations), the mar resistance protein (20 mutations), the mdfA efflux pump (15 mutations), other metabolism (generally related to carbon usage, 16 mutations), the mscK ion channel (12 mutations), outer-membrane-associated proteins (12 mutations), mutations in prophages (15 mutations), slt degragation protein (12 mutations), and mutations affecting transcription/translation (12 mutations). We analyzed the dataset containing the first observation of each mutation to assess whether certain mechanisms appeared earlier or later in evolutionary trajectories (Fig. 6). Using chi-squared tests on the number of mutations in each category across the four timepoints, we found that mutations associated with the acr efflux pump (p < 0.001), ATP synthase (p = 0.041), the mar resistance protein (p = 0.022), outer-membrane proteins (p = 0.018), and the mfdA efflux pump (p = 0.013) were nonrandomly distributed across evolutionary time. Specifically, early mutations at T1 were disproportionately located within the acr efflux pump, while mutations at T5 were often within the acr efflux pump and associated with ATP synthase. At T9, mutations within the outer-membrane proteins became more common. Finally, the most resistant strains frequently had mutations in the mar resistance protein and the mfdA efflux pump. Although conducting 10 chi-squared tests for the 10 categories allows the possibility of finding spurious positive results, the probability of spuriously identifying five categories at a threshold of p < 0.05 is <1 in 16,000.
The types of mutations acquired at each timesteps changed as antibiotic levels increased. The most common pathways to resistance at the four timepoints were mutations in the acr efflux pump (T1), ATP synthase genes (T5), outer-membrane proteins (T9), and the mdfA efflux pump (T13).
Fitness effects of resistance mutations are more extreme at higher temperature
We calculated the estimated effect of mutations in 37 genes on the outcomes of the pairwise competition experiments, as to compare the distribution of mutational fitness effects across the three temperatures. Many of these genes were part of the same cassettes (e.g., acrA, acrB, and acrR). Initially, we identified mutations in 43 genes, though the distribution of mutations in 6 of these genes (yejA, yfaQ, xdhB, rplD, rpoC, and uvrA) were equivalent to the mutation occurrence of other genes, and thus could not be included in the analyses due to the fact that the predictor variables were identical. Of these remaining 37 genes, we saw significant differences in the estimated effects of mutations in 24 genes between 37 °C and 42 °C. Conversely, there were zero significant differences in effect size between the temperatures of 32 °C and 37 °C. Effect sizes in this analysis signify the difference in the success of a strain as a result of carrying one extra mutation in the indicated gene. Then, we evaluated whether the range of gene effects was larger at 42 °C or 32 °C, as compared with the range of gene effects at 37 °C (Fig. 7). We conducted F-tests on the distribution of gene effect sizes at 37 °C vs. 42 °C and at 37 °C vs. 32 °C to find whether the variances of these distributions were unequal. We found that the range of gene effects was greater at 42 °C than 37 °C (F36, 36 = 5.68, p < 0.001) but that the range of gene effects was not different between 32 °C and 37 °C (F36, 36 = 0.91, p = 0.79).
Results of the linear regression quantifying the effects of mutations within 37 genes show that the estimated impact of mutations is greatest at 42 °C, as compared with 32 °C or 37 °C. Values in histograms are the coefficients from the linear model, where each coefficient is the estimated effect of a gene-specific mutation on the ratio of that strain in competition. Effect sizes near zero indicate no effect, whereas an effect size of 1 indicates a one-log-fold increase in the competitive success of a strain that has acquired a mutation in the given gene. The variance of effect sizes at 42 °C is significantly greater than at the other two temperatures.
The genes with the most significant additional fitness cost at 42 °C were: mscK, opgH, dgcF, atpD, uvrA/ssb, and ssb. Of these, only mutations in atpD and mscK were common, appearing in more than 1 lineage. These genes are involved in ATP synthesis and in ion transport via the mechanosensitive channel MscK, respectively.
Finally, we combined the competition data with the genomic data to examine whether specific categories of mutations were more likely to result in decreased thermal tolerance. For this analysis, we used Fisher’s Exact Tests to determine whether lineages with a greater cost of resistance at 42 °C (12 lineages) were more likely to have any specific type of mutation (using the 10 mutation categories shown in Fig. 5). Surprisingly, of these 10 tests, we found no significant (p < 0.05) associations between mutation type and the presence of fitness costs to thermal niche breadth, suggesting that the tradeoff is not tied to a specific mechanism of resistance.
Source: Ecology - nature.com
