Twelve populations of E. coli B experienced rapid and parallel losses of d-ribose catabolic function during evolution in glucose minimal medium (Fig. ). We showed that these losses were caused by similar deletions of the rbs operon; in all cases, one end of the deletion was immediately adjacent to an IS150 element located just upstream of the operon, whereas the other endpoint varied (Fig. ). We further showed that mutations from Rbs+ to Rbs− occurred at an unusually high rate, that these mutations were caused by deletions of the rbs operon similar to those during the evolution experiment, and that Rbs− mutations consistently conferred a small but significant competitive advantage in glucose minimal medium (Fig. ).
It may be instructive to examine the logic that led us to identify the evolutionary forces responsible for the loss of ribose function. The rapid and parallel evolution of the Rbs−
phenotype initially suggested to us that this change was adaptive. Repeatable change across lineages is widely taken as evidence of adaptation by natural selection (9
). Furthermore, the fact that this loss of function occurred when adaptation to the new environment was most rapid (23
) reinforced our view that the loss was beneficial for the bacteria in glucose minimal medium. However, our observation that Rbs−
mutants were readily isolated from the ancestor led us to question our preconceptions and focus on the alternative possibility that the losses of ribose catabolic function might be caused by a hypermutable locus of some sort. Indeed, a fluctuation test confirmed this possibility. We then performed competition experiments between the spontaneous Rbs−
mutants and their Rbs+
progenitor, which showed that the mutants were also fitter than their progenitor in the glucose minimal medium. Evidently, both selection for the loss of ribose catabolic function and its hypermutable genetic basis contributed to the rapid and parallel phenotypic evolution that we observed.
The finding that all seven independent spontaneous Rbs−
mutants had a competitive advantage strongly suggested that the beneficial effect was caused by the rbs
deletion, as opposed to some hypothetical secondary mutations that arose during the fluctuation test in which the mutants were generated. It is generally accepted that only a very small fraction of all mutations are beneficial in any particular environment, with the vast majority being either deleterious or neutral (13
). Thus, it is highly unlikely that secondary mutations would produce any improvement in fitness.
Nevertheless, to confirm absolutely that the rbs deletions per se were responsible for the fitness benefits observed in glucose minimal medium, we performed two additional types of strain constructions and corresponding competition experiments. First, we used P1 transduction to restore a functional rbs operon to two of the Rbs− mutants. This restoration reduced their competitive fitness, which indicates that the deletion of the rbs operon itself was beneficial in the glucose environment. Second, we constructed de novo a deletion of most of the rbs operon that did not involve the upstream IS150 element or the rbs promoter region, and we used homologous gene replacement to introduce this into our progenitor. This deletion also produced a small but significant competitive advantage. Thus, three different genetic approaches—spontaneous mutation, transduction, and gene replacement—all demonstrate that the deletion of the rbs operon causes a beneficial effect in minimal glucose medium.
Molecular basis of the high rate of mutation from Rbs+ to Rbs−.
The same class of molecular events led to the loss of ribose function in the evolved lines and spontaneous Rbs−
mutants. A total of 18 Rbs−
genotypes (11 evolved lines and 7 spontaneous mutants) all showed deletions in which one endpoint was located precisely at the end of an IS150
element that was inserted upstream of the rbs
operon. The extent of the deletion varied among the genotypes, but it always encompassed the promoter region and first gene (rbsD
) and in some cases included all six genes in the rbs
operon plus part of an adjacent gene of unknown function (yieO
). The fact that one endpoint of the deletion was always precisely located at the end of this IS150
element suggests that the mechanism of deletion involved first the transposition of an IS150
element (either the one upstream of rbs
or any other) into the site corresponding to the other endpoint and in the same orientation as the one upstream of rbs
. This transposition was presumably then followed by a recombination event between the new IS150
and the one upstream of rbs
, thereby causing deletion of the intervening region. Whether the transposition and recombination events occurred simultaneously or successively is unknown; the fact that none of the spontaneous Rbs−
mutants showed a simple transposition that inactivated the rbs
operon (without the associated deletion) suggests that the two events occurred in the same cell generation. Analysis of the nucleotide sequences at the right endpoints of the different deletions indicated no homology with the end of IS150
corresponding to the left deletion endpoint, which further suggests rearrangements associated with an initial transposition event. Examination of the nucleotide sequence of the presumptive target sites of the IS150
transpositions in the different genotypes failed to reveal any obvious preference for its insertion.
All IS-associated mutations have been studied (30
) in 2 of the 12 evolving populations, in part to identify candidate mutations that may have contributed to their adaptation. In both of these focal populations, the rbs
deletions were already present at generation 500, when adaptation was most rapid (24
), and they remained present in all clones sampled in later generations. By performing restriction fragment length polymorphism analyses using IS elements as probes with Eco
RV-digested genomic DNA extracted from clones sampled over time, Papadopoulos et al. (30
) found several other mutations that had been fixed in these focal populations. By characterizing these mutations at the sequence level, Schneider et al. (33
) showed that they also involved IS-mediated events, including transpositions, inversions, and deletions; they further showed that none of the mutations in the two focal populations involved the same genes. However, Papadopoulos et al. (30
) did not detect the parallel rbs
deletions that we have described here, for the following reason: Eco
RV cuts at the right end of IS150
, whereas the probe they used corresponded to the left end of IS150
, so that the type of rearrangement involved in the rbs
operon (deletions at the right end of IS150
) went undetected. Based on our analysis of Hin
cII-digested genomic DNA of the evolved clones, the rbs
deletions are the only IS-associated mutations that were fixed in the two focal populations but not detected by Papadopoulos et al. (30
Some other studies have reported that certain strains of E. coli
B are Rbs−
), whereas our founding strain was clearly Rbs+
but predisposed genetically to become Rbs−
. One plausible explanation for the reports that some other B strains are Rbs−
is that they lost the ribose catabolic function during propagation in the laboratory, much as we have observed in our experimental lines. Under this hypothesis, the ancestor of all B strains would have had two rbs
operons, a nonfunctional one at 2 min and a functional one at 83 min, with deletions independently arising in different B lineages at different points in time. However, other sources of genetic instability may also contribute to such differences. For example, a transposable element containing the rbs
operon was found in one experiment, and the operon at 2 min appears to have been generated by a duplication of the operon at 83 min, suggesting that the operon has been mobile in E. coli
IS and other transposable elements generate a substantial fraction of the mutations in bacteria, and they are therefore important evolutionary factors (5
). Nonetheless, there are conflicting views about their costs and benefits and the balance of forces that maintain these elements in populations. One view emphasizes that a much higher proportion of mutations are deleterious than are beneficial, infers that transposable elements impose a burden on adaptation by substantially increasing the overall mutation rate, and concludes that active elements can thus be maintained only if horizontal gene transfer allows them to exist as genomic parasites. Another view emphasizes that transposable elements may, on balance, be adaptive to an evolving population just as mutator alleles are under certain conditions. Like mutators, transposable elements may spread in asexual populations by “hitchhiking” along with the occassional beneficial mutations that are produced by their activities (5
). Our findings indicate the high mutation rate that can result locally from the presence of one IS element in a particular gene region. If these deletions occurred near any essential gene, then the load created by the IS element would be equal to the mutation rate, implying a weak but nontrivial selection coefficient of about 5 × 10−5
against that one element alone. Our results also show that some IS-mediated mutations are beneficial and promote the adaptation of an evolving population. Interestingly, and in contrast to another evolution experiment in which point mutations and IS transpositions generated functionally equivalent beneficial mutations (39
), in our study all of the many spontaneous beneficial disruptions of the ribose function were associated with IS activity. This difference may arise because the mutations in our study caused the loss of gene function, which occurs readily by IS-mediated mutations, whereas the mutations in the earlier study involved a more subtle change in gene regulation.
It is clear that deletions of part or all of the rbs operon are beneficial to E. coli B in glucose minimal medium and that the IS150 element located immediately upstream of the operon plays a role in generating those deletions. However, the physiological basis for the benefit that accrues is unclear. In all 18 spontaneous deletions we examined, the promoter region and first gene (rbsD) of the operon were eliminated, suggesting that it was silencing of the operon that provided the selective advantage. The constructed rbs deletion in strain GBE127 retains the promoter region and rbsD, and this strain obtains a similar benefit. Also, the fact that similar benefits accrued whether the deleted region was 2 or 7 kb implies that neither energetic savings associated with chromosomal replication nor conformational changes in the chromosome can account for the beneficial effect. Taking all these considerations together, it appears likely that the beneficial effect involves the elimination of the ribose-catabolic function per se.
Quantitative analysis of the contributions of mutation and selection to the evolution of the Rbs− phenotype.
Both positive selection for loss of the rbs operon and its underlying mutability contributed to the evolutionary losses of ribose-catabolic function in the 12 experimental populations. Here, we examined mathematically their contributions as well as their interplay with one another and with selection at other loci.
We considered first the expected time course if selection had favored the loss of this function, but without any hypermutable basis. For simplicity, we assumed the mutation is just common enough that there is no waiting time for the mutation to appear, and then we calculated the time for the mutant genotype to increase from one cell to 50% of the total population. In our calculation, we used the 1.4% selective advantage (s
), which we measured as the average of seven spontaneous Rbs−
mutants, and the effective population size (Ne
= 3 × 107
) that prevailed during the evolution experiment (24
). The ratio (R
) of genotypes changes in a log-linear fashion under constant selection (14
). We expressed bacterial generations using a log2
transformation of the daily dilution and regrowth. Thus, the time in generations (g
) for a single mutant to increase to 50% (a final ratio of 1) of the population was calculated as follows:
In fact, however, the actual frequency of mutants reached 50% much sooner than this in most of the experimental populations (Fig. ).
Next we calculated the time required for the Rbs−
mutants to achieve a 50% frequency, assuming they increased by recurring mutation only, without the benefit of any selection. In that case, the frequency (p
) of the Rbs+
type should have decayed exponentially from an initial frequency of 1. Given the estimated mutation rate (μ), of 5.4 × 10−5
per cell per generation, the time for the progenitor to decline to 50% (and the mutant to reach 50%) was calculated as follows:
The actual spread of Rbs−
mutants was much faster than this in all populations (Fig. ).
Finally, we calculated the approximate time required for the mutant to reach a 50% frequency, given both its observed selective advantage and its high mutation rate. To do so, we noted that mutants should have been present after the first day of the evolution experiment in the same average frequency as in the fluctuation test, which was about 0.000512. We then used the fact that s
μ to deduce that selection will drive the subsequent increase in the frequency of mutants, so that we could apply the first equation above to calculate the time to reach 50% (a final ratio of 1), given an initial ratio of 0.000512. We obtained the estimated time as follows:
This last calculation gives better agreement with the observed spread of Rbs−
mutants than does either calculation that ignores the contribution of mutation or selection. To summarize, the expected times required for Rbs−
mutants to reach a frequency of 50% in a population were found to be 18,519 generations under the mutation accumulation process (μ = 5.4 × 10−5
per generation), 1,774 generations under the selection process (s
= 0.014 per generation), and 781 generations under the selection-plus-mutation process.
Two features of the dynamics of the Rbs−
mutants that are not explained by these simple models are (i) the pronounced variability in the frequency of Rbs−
mutants among the replicate populations at generations 500 and 1000 and (ii) the temporary reversals in the frequency of Rbs−
mutants in a few populations during that interval (Fig. ). Both features are understood by realizing that selection was simultaneously acting on mutations at other loci (3
), including some mutations that were much more beneficial than were the rbs
deletions. Each population experienced several selective sweeps by beneficial mutations during these 2,000 generations, and the underlying mutations conferred fitness advantages, on average, of about 10% (23
). Given the asexual nature of the evolving populations, the short-term fate of Rbs−
mutants in any population would depend on how long it took for one of the highly beneficial mutants to appear in a Rbs−
clone and whether an even more beneficial mutation appeared in a Rbs+
clone, which could cause a reversal owing to clonal interference (16
). Thus, variation among populations in the dynamics of the Rbs−
mutants is expected from the stochastic appearance of beneficial mutations at other loci, which leads to divergence in the genetic linkage among beneficial alleles across the replicate populations.