The basis for our study is the observation that rifampicin resistance arose in the absence of an antibiotic during an evolutionary experiment. After 2000 generations of thermal stress, 13 of 114 E. coli clones exhibited resistance to rifampicin. Twelve of these 13 clones included a mutation in codon 572 of the rpoB gene, with three different mutations observed in that codon (Table 1). These three mutations have been noted previously to confer rifampicin resistance , a finding we have reconfirmed. Moreover, each of these three mutations occurred independently in more than one population, providing strong evidence by the criterion of evolutionary convergence  that the mutations are beneficial under the experimental conditions. Concerning the 13th and final clone, a mutation in codon 143 has been previously described to confer low resistance to rifampicin (R143W, ), but the mechanistic causes of resistance for this clone remain unclear. However, our analysis of the RNAP 3-D structure suggests that codon 143 folds into the vicinity of the active (i.e., binding) site of the RNAP (Additional file 5). It is possible, then, that mutations in this codon alter rifampicin binding, thus leading to resistance.
We used both direct and indirect evidence to confirm that all three mutations in codon 572 result in a fitness advantage within a thermal stress/low glucose environment. For direct evidence, we introduced single mutations into the ancestral REL1206 background and assessed the relative fitness of mutants to unmutated REL1206. The measured fitness effect varied statistically among the three mutations, with relative fitnesses ranging from 1.18 to 1.25. Perhaps the most notable feature of these measurements is the magnitude of the effect. In the experimental evolution literature, it is rare to find single mutations with relative fitness benefits above ~15% [35, 36]. Thus, with the exception of mutations that compensate the cost of antibiotic resistance , the measured fitness benefits of the single I572N, I572L and I572F mutations are uncommon [6, 37]. We note, however, that these high fitness values still explain only a fraction of the total realized relative fitness benefit of the twelve evolved clones, which have accumulated an average of 8 mutations compared to REL1206 and an average relative fitness increase of ~40% (mean relative fitness 1.396; stdev 0.122) .
Indirect evidence for the benefit of these mutations comes from the assessment of the frequency trajectory of rifampicin resistance over the course of the full 2000-generation experiment. Generally, rifampicin resistance evolved early – within 500 generations - and swept to fixation within a few hundred generations (Figure 2). This steep increase in frequency is consistent with a high selection coefficient for the haplotypes that carry the resistance marker. We have measured the selection coefficient for these haplotypes by estimating s
up, which ranges between 0.015 and 0.077 (Figure 3A, Additional file 3: Table S2). While these are high selection coefficients, they are not directly comparable to our relative fitness estimates, for several reasons (see below). What s
up does, however, is confirm that the capability for antibiotic resistance may be highly beneficial even in the absence of antibiotic.
Fixation dynamics of the resistance mutations
The frequency trajectories also provide crucial insights into the fixation dynamics of beneficial mutations. One interesting observation is that the relative fitnesses of single rifR mutations do not correlate with the estimated selective coefficient (s
up) of the populations that harbor these mutations (Figure 3D). This may reflect a lack statistical power to detect a correlation - since there are only three relative fitness measures – or may reflect the possibility that the resistant individuals observed at intermediate time points harbor different resistant mutations than the one observed at the end of the experiment. Nonetheless, we believe the lack of a relationship is meaningful. For example, the I572F rifampicin-resistant mutation found in lines 56 and 61 has the highest relative fitness as a single mutation (Figure 5), but rifampicin resistance was not fixed rapidly in these two lines. Instead, we find that early-occurring rifR mutations take less time to reach fixation than late-occurring mutations (Figure 3C); this pattern suggests either that epistasis, clonal interference or frequency dependent fitness interactions influences s
For the former (epistatic interactions), diminishing-returns epistasis is expected theoretically  and has been observed empirically as more and more mutations accumulate over the time-course of an experiment [6, 37]. Under diminishing-returns, a relatively late occurring rpoB mutation may have a smaller fitness effect, conditional on the occurrence of previous beneficial mutations. For clonal interference, competition between beneficial haplotypes will slow the process of fixation [38, 39]. Finally, complex dynamics such as those observed in lines 56 and 131 might be due to frequency-dependent selection. In any case, such competition may be more common in the later stages of an experiment when multiple mutations have accrued . In contrast, early rifR mutations likely occurred in a REL1206 background that was fairly devoid of other new mutations, thus minimizing possibilities for either clonal interference or epistatic interactions with other new mutations.
The possibility that frequency trajectories have been shaped in part by epistasis (whether as diminishing-returns or one of several other possible forms ) is not surprising given the study of Tenaillon et al. . This study detected statistical associations among mutations that were consistent with extensive and varied epistatic effects. These associations shaped the adaptive response to thermal stress into one of two distinct genetic solutions typified by mutations either in rpoB or in the termination factor rho, but rarely in both genes. To investigate the potential relationship of s
up to these statistical associations, we examined genetic data from Tenaillon et al. . Clones from lines 43, 61 and 131, all of which had high τfix values (> 400 generations; Figure 2) carried mutations in both rho and rpoB, a combination statistically highly disfavored among the full dataset of 114 clones. This observation suggests that the long fixation time in these lines could be due in part to negative epistatic interactions between rho and rpoB mutations that reduces beneficial effects of both mutations. The strength and mechanism of these interactions need to be characterized more fully, however.
Previous studies have identified potential epistatic interactions with mutations in codon 572 , and epistasis must contribute to varying fitness effects among our genetic backgrounds (Figure 4). In the high temperature and low glucose condition, our three codon 572 mutations conferred a slightly (but not significantly) higher relative fitnesses in the REL606 background than in the ancestral REL1206 background. The similar effects in these two backgrounds may not be surprising, however, given that REL1206 and REL606 differ by only a handful of mutations: REL1206 differs from REL606 in 3 SNPs, an IS element and a large deletion [25, 42]. In contrast, the rifR mutations are detrimental in the K12 MG1655 background (Table 2), even though K12 and B are genetically similar (> 99% sequence identity over ~92% of their genomes ).
The specificity of adaptation
The effects of the rifampicin resistance mutations also vary as a function of environment. In our study, the only environment in which the mutations are demonstrably beneficial is that of the original evolution experiment (high temperature and low glucose). In contrast, the effects of rifR mutations are indistinguishable from neutrality in a high temperature and rich glucose environment and demonstrably detrimental at 37°C in poor and rich glucose environment [19, 20, 31] (Figure 5, Additional file 4: Table S3).
With the exception of lethal selection or niche creation  experiments, most other studies have demonstrated that the fitness advantage conferred by a mutation is maintained across environments and conditions [9, 10]. In other words, they have found that beneficial mutations are generally not severely compromised in other environments . The logical extension of these observations is that a single beneficial mutation is unlikely to result in niche specialization, because it will not lead to drastic fitness differences across environments.
In stark contrast to these studies, we do observe the potential for the evolution of ecological specialization in a single mutational step, because all three mutations in codon 572 of rpoB confer a selective advantage in the conditions of the original evolution experiment but significant disadvantages in other environments (Table 3) and genetic backgrounds (Table 2). In this context, it is important to repeat that this potential for niche specialization is not a function of antibiotic resistance, for which niche specialization is well known, but rather due to fitness effects across antibiotic-free environments.
The question remains as to whether our single rpoB mutations are rare or instead cast doubt on previous conclusions that niche specialization is “… unlikely to occur through the substitution of a single mutation” . The degree of ecological specialization for our single mutations could be due in part to the drastic selection pressure (high temperature) in the original experiment or to rpoB itself. Because mutations within rpoB can be highly pleiotropic, they can affect a series of downstream traits like gene expression [44, 45] that may be fine-tuned for specific selective regimes. We note that highly pleiotropic (but non-rpoB) mutations have been observed in early stages of adaptation to ethanol stress  and glycerol minimal media , suggesting that early mutations in adaptation are commonly involved in transcriptional regulation with large fitness and pleiotropic effects [48, 49]. As such, our rpoB mutations may not be uncommon, either in their effects or in their potential for ecological specialization. Thus, in our opinion, the frequency and occurrence of niche-specialization by single beneficial mutations is still an open question worthy of further study.
Several experimental evolution studies have shown the fixation of mutations in RNAP during stress adaptation [44, 45]. This suggests that modifications in the RNAP could be a general mechanism for adaptation to new environments. Yet, the mechanistic basis for the beneficial effect of rpoB mutations at high temperature remains unclear. Since temperature affects the stability and activity of proteins [50–52], rpoB mutations may modify the stability and/or activity of RNAP at high temperatures. For example, previous studies have shown that mutation I572F increase transcription termination , and mutation I572L reduces transcription efficiency at 37°C . Another (but not mutually-exclusive) hypothesis is that rpoB mutations cause changes in gene expression through the redistribution of RNAP in manner that favors adaptation to new environments . The unique challenge here is explaining how these mechanistic effects can be advantageous in REL1206 but (for example) disadvantageous in K12 (Figure 4). Fortunately, questions of mechanism are amenable to future experimental investigation.