Since its introduction over seventy years ago, Wright's metaphor of the adaptive landscape has become one of the most influential concepts in evolutionary biology, yet empirical understanding of the structures of actual landscapes remains elusive. In this study, we monitored the evolution of 42 experimental populations of E. coli under 7 resource regimes for 2000 generations in order to investigate the effects of environmental complexity on their dynamics. We were interested, in particular, in whether heterogeneous resource environments would influence the repeatability of evolution by impacting the ruggedness of the adaptive landscape. Our results can be summarized as follows. (1) Populations evolved under all experimental regimes exhibited significant increases in fitness. (2) The magnitude of their fitness gains varied across the regimes, as did their correlated fitness responses to other regimes. (3) Among-population genetic variation for fitness was highest, and most sustained over time, in those groups that had evolved under fluctuating resource regimes. Below we discuss these findings in more detail.
Effect of environment on patterns of divergence
We only found evidence for highly significant divergence among the replicate populations that evolved in the fluctuating resource environments, glu/mal and glu/lac (Table 3). To address whether this divergence was sustained over time, we also quantified the change in the variance for fitness between 1,000 and 2,000 generations (Table 5). Divergence among the replicate populations in the glu/mal group increased over the second half of the experiment, whereas divergence among the populations in the glu/lac group decreased over this period. However, the variation in the glu/lac group at 1,000 generations was strongly influenced by a single population that showed no fitness gain to that point. When this atypical population was excluded from the analysis, divergence trended higher in this group as well. On the whole, these results indicate that divergence of replicate populations was both significant and increasing over time only in those two groups that evolved in the temporally fluctuating environments. This finding is consistent with the hypothesis that the replicate populations in these treatment groups were evolving toward different adaptive peaks.
In principle, replicate populations could diverge from one another not only by selection acting on different beneficial mutations but also by drift and hitchhiking. Divergence by drift could occur through the accumulation of mutations that are neutral in the selective environment, but which might have some fitness effects in other environments. Deleterious mutations might hitchhike to high frequency if they become linked with a beneficial mutation [45, 46], which could occur since the bacteria in our experiments are strictly asexual (i.e., they lack any mechanism for horizontal gene transfer). However, the E. coli strain we used has a very low total genomic mutation rate , which should limit the rates of substitution by drift and hitchhiking. Indeed, high-coverage whole-genome sequencing of another population founded from the same strain found that only three synonymous mutations achieved detectable frequencies in 20,000 generations . Moreover, the patterns of correlated responses in 12 populations, again founded from the same ancestral strain, indicate that pleiotropic effects of beneficial mutations have been more important than mutation accumulation by drift or hitchhiking in explaining patterns of phenotypic evolution over 20,000 generations [10–12]. Therefore, it appears unlikely that drift or hitchhiking have contributed much, if at all, to the among-population divergence in our 2000-generation experiment, nor is it evident why any such effects would be stronger in the fluctuating environment treatments than in other treatments that experienced the same resources alone or in combination.
Selection can also lead to among-population divergence due to stochastic differences in the timing and order of beneficial mutations . If only a single local adaptive peak is available, then this divergence should be transient. However, if mutations interact such that some mutations are beneficial only in combination with other mutations, then this 'sign epistasis' may generate multiple peaks separated by valleys of maladapted intermediates [33, 49]. In the latter case, differences in the order in which the mutations arise could cause initially identical populations to evolve to different peaks, resulting in sustained divergence [33, 49, 50]. While stochastic differences in the timing and order of mutations might well contribute to population divergence, they cannot alone explain why the extent of variation and its persistence over time should differ between groups. Although we cannot rule out the possibility that the divergence in those groups that evolved with fluctuating resources might eventually prove to be transient (say, over tens of thousands of generations), this possibility would nevertheless imply differences in the structure of fitness landscapes between the various regimes that contributed to their different early dynamics of divergence. In future work, we intend to identify the genetic bases of the adaptation of these populations and then combine mutations to test directly for the presence of sign-epistatic interactions.
A possible caveat to this interpretation is that, because the benefit of adaptive mutations must be averaged across two different environments, those populations evolved in fluctuating environments may have experienced weaker effective selection than the populations that evolved in constant environments . If so, then adaptive mutations would take longer to fix and the period of transient divergence would be extended. However, we did not observe substantially smaller fitness gains for the groups that evolved in the fluctuating environments (Table 1).
The greater divergence among populations in the fluctuating environments is consistent with the hypothesis that environmental heterogeneity can increase the ruggedness of the adaptive landscape and thereby increase the likelihood that replicate populations will find distinct fitness peaks [13–15, 35, 36]. However, it is less obvious why divergence was more pronounced in regimes with two resources that fluctuated temporally, as opposed to regimes that presented the same two resources simultaneously. Temporal fluctuations would presumably have favoured generalists that are well adapted to each of the alternating resources, whereas coexisting specialists might evolve when the resources are simultaneously present. That distinction, while important, still begs the question of whether independently evolved populations would achieve similar or diverse generalist types or admixtures of specialists under those respective scenarios. Although the specific adaptations that occurred in the populations are not yet known, we can imagine a scenario where one might well expect greater among-population divergence under the fluctuating resource regimes. Consider a mutation that increases fitness in the glucose component of a fluctuating environment, and which becomes fixed in one population. Despite conferring a net fitness advantage, this mutation might nevertheless reduce fitness in the lactose component of the environment. (In fact, such trade-offs are evident in most or all the populations that evolved in glucose, as seen in Figure 1). This deleterious side-effect will select for compensatory mutations that alleviate the fitness trade-off in the lactose environment [37, 52–55]. Such compensatory mutations are beneficial only in combination with the previous mutation, and they would not evolve without the adaptation to glucose. As a consequence, the spectrum of potential future adaptations will differ from those available to a population that by chance fixed an alternative mutation that did not engender a trade-off, and therefore would not select for a compensatory mutation. By contrast, in those environments where two resources were presented simultaneously, an evolving lineage might, at least in principle, split into two subpopulations that each adapt to one resource while forgoing adaptation to the other. The two subpopulations might then be on different peaks, yet if all of the replicate lineages split into subpopulations in the same way, then the replicate populations per se might not diverge from one another.
A caveat to this conceptual distinction between simultaneous and fluctuating environments is the fact that E. coli typically uses multiple resources (including glucose and either lactose or maltose) sequentially rather than simultaneously [56–58]. Because only one resource is used at a time, the effect is to partition the environment physiologically, thus giving rise to two sequential adaptive landscapes just as in the fluctuating environment. However, in contrast to the externally imposed fluctuating regimes, the physiological separation between these successive landscapes is under genetic control. Preliminary results (T.F.C., unpublished data) indicate that four of the six glu+mal populations and all six of the glu+lac populations evolved some change in the activity of catabolite repression (whereby the presence of glucose represses the use of other sugars), in which case the resources may have been used simultaneously. These changes, if confirmed, might have reduced the among-population variation in those regimes where resources were presented simultaneously rather than sequentially.
Finally, a limitation of our analyses of the evolved lines thus far is that they only address variation in fitness. Thus, we cannot exclude the possibility that replicate populations might have diverged genetically, and even physiologically, in ways that nevertheless yield similar fitness levels in the assayed environments. Some of the potential targets of selection in this experiment include the length of the lag phase following transfer to fresh medium, maximum growth rate and, in mixed resource environments, regulation of substrate utilization preferences. Different combinations of changes in these traits might produce similar fitness gains [59, 60]. Although measurements of additional characters can only increase the likelihood of detecting variation among populations, we see no obvious reason why adding traits would substantially affect the relative among-population variation across the different treatment groups.
By taking advantage of the short generations of bacteria and the ease of manipulating their environments, we have assessed the effect of environmental variation on the reproducibility of evolution, with implications for understanding the structure of adaptive landscapes. This study extends previous work on the effects of environmental heterogeneity [13–15, 19, 26, 28, 29, 61] by focusing specifically on the effect of temporal fluctuations in resource availability on population divergence. We observed that divergence among replicate populations was greater in environments with alternating resources than in environments with the same resources presented either singly or simultaneously, suggesting that epistatic interactions among mutations were stronger or more influential under the temporally fluctuating regimes. Future work will take advantage of the potential for genetic analysis and manipulation of E. coli to examine the molecular and physiological changes that underlie the parallel and divergent trends observed in this study.