Diminishing Returns from Mutation Supply Rate in Asexual Populations

See allHide authors and affiliations

Science  15 Jan 1999:
Vol. 283, Issue 5400, pp. 404-406
DOI: 10.1126/science.283.5400.404


Mutator genotypes with increased mutation rates may be especially important in microbial evolution if genetic adaptation is generally limited by the supply of mutations. In experimental populations of the bacterium Escherichia coli, the rate of evolutionary adaptation was proportional to the mutation supply rate only in particular circumstances of small or initially well-adapted populations. These experiments also demonstrate a “speed limit” on adaptive evolution in asexual populations, one that is independent of the mutation supply rate.

Surveys of natural populations of pathogenic (1) and commensal (2) bacteria indicate that more than 1% are dominated by mutator genotypes with increased mutation rates. Such genotypes are even more prevalent among populations of E. coli evolving in the laboratory (3) and in certain tumors (4). Mutators may be favored because they produce rare beneficial mutations more often than do normal genotypes and thereby allow a faster response to selection (5). But the actual relation between mutation rates and adaptive evolution may be more complicated, especially in asexual populations that are subject to strong effects of genetic linkage. Indeed, the logic that drives any empirical association between mutators and rapid adaptive evolution can be reversed: Rapid adaptation to a novel or changing environment provides more frequent opportunities for mutators to “hitchhike” to high frequency along with beneficial mutations to which they are genetically linked, even when mutators themselves have little effect on the rate of adaptation (3).

Moreover, population genetic models predict that the rate of adaptive evolution in asexual populations will increase proportionately with mutation rate only if populations spend most of their time waiting for beneficial mutations (6). Otherwise, two or more beneficial mutations may be simultaneously present in different lineages within a population; they will interfere with one another's spread, and ultimately only the superior mutation prevails while all others are driven extinct (6, 7). Therefore, an increase in the supply rate of beneficial mutations might often be subject to diminishing returns, as the extent of clonal interference increases with the number of beneficial mutations produced in an interval. This effect can be so pronounced that the rate of adaptation reaches an effective “speed limit” (6).

Consider an asexual population that is small and spends most of its time waiting for the next beneficial mutation. An increment in either mutation rate or population size should shorten the waiting time and thereby accelerate adaptive evolution. However, a further increase in the supply of beneficial mutations (because of changes in either parameter) should yield less acceleration as a consequence of more clonal interference (8). Now consider the effect of starting out with a different genotype, one better adapted to the selective environment. This population should spend more time waiting for beneficial mutations than one that is poorly adapted, all else equal, and thus clonal interference should be less important. Well-adapted populations may therefore experience a proportional acceleration of their adaptive evolution over the same range of mutation rates and population sizes that reveal a speed limit in populations founded by an inferior genotype.

We tested these hypotheses by examining how mutation rate, population size, and the level of adaptedness affect the rate of adaptive evolution in asexual populations of the bacterium E. coli. Forty-eight populations evolved for 1000 generations in a simple laboratory environment (9). We manipulated initial adaptedness by using two founding strains: One had not previously experienced the selective environment, whereas the other had already adapted to that environment for 10,000 generations (10). We manipulated mutation rates by moving repair-deficient mutYand mutS alleles into repair-proficient genetic backgrounds (11). We manipulated effective population size by varying the bottleneck during propagation of the evolving populations (12). The rate of adaptive evolution was determined by measuring changes in competitive fitness relative to the corresponding ancestral strain and in the same simple environment (13). Fluctuation tests (14) indicate a moderate increase in mutation rate for the mutY strains and a greater increase for the mutS strains (Table 1). Each combination of mutation rate (wild-type, mutY, or mutS), effective population size (6.6 × 105 or 3.3 × 107), and evolutionary history (nonadapted or adapted) was replicated fourfold.

Table 1

Estimates of relative mutation rates of the six strains used in the evolution experiment, on the basis of eight separate fluctuation tests for each strain (14).

View this table:

Fitness increased under each combination of mutation rate and population size in the nonadapted strains (Fig. 1). For each population, we calculated its time-averaged rate of adaptation from the regression of fitness against time. We then regressed these rates against the product of relative mutation rate and effective population size, which equals the mutation supply rate, using two different models (Fig. 2A). The first is a linear model, in which the rate of adaptation is directly proportional to the mutation supply rate. The second is a rectangular hyperbola, such that the rate of adaptation is subject to an upper limit because of clonal interference. The y intercept is fixed at 0 in both models, because an asexual population with no mutations cannot adapt genetically. The linear model indicates that the rate of adaptation increases with the mutation supply rate (F 1,23 = 7.47, P = 0.0118). But the additional degree of freedom required for the hyperbolic model provides a substantial improvement over the linear model (F 1,22 = 397.53,P < 0.0001) (15). Our experiment thus demonstrates a speed limit on adaptive evolution in asexual populations, which was predicted by a population genetic model with clonal interference (6).

Figure 1

Fitness trajectories of wild-type and mutator populations, founded by the nonadapted strain, during experimental evolution. Points indicate the average fitness of four populations in each treatment relative to the corresponding ancestor; error bars show standard errors. Circles are wild type, squares mutY, triangles mutS. (A) Small effective population size (= 6.6 × 105). (B) Large effective population size (= 3.3 × 107).

Figure 2

Effect of mutation supply rate on rate of adaptive evolution. Mutation supply rate is the product of mutation rate and effective population size; values are expressed relative to the treatment with the lowest rate and are shown on a log-transformed scale to spread the treatments along the x axis (statistical regressions use untransformed values). Open symbols are small populations, filled symbols large populations; circles are wild type, squares mutY, triangles mutS. (A) Populations founded by nonadapted strains. The curve is a hyperbolic regression, which fits the data better than a linear regression (P < 0.0001). (B) Populations founded by previously adapted strains. The curve is a linear regression (P = 0.0234), which appears exponential because of the logarithmic scale. A hyperbolic regression provides no significant improvement (P = 0.4565).

This model also predicts that the speed limit should be shifted to much higher values of the mutation supply rate for a founding strain that is so well-adapted that it spends most of its time waiting for further beneficial mutations. The rate of adaptation was much slower for the 24 populations founded by the adapted strain (Fig. 2B) than for those founded by the nonadapted strain (Fig. 2A). Even so, the rate of adaptation in these well-adapted populations increased significantly with mutation supply rate, on the basis of linear regression with the y intercept held at 0 (F 1,23 = 5.90, P = 0.0234). But the hyperbolic model provides no statistical improvement for the well-adapted populations (F 1,22 = 0.57,P = 0.4565), in contrast to the strong improvement for the nonadapted populations. Thus, the contrasting predictions for the effects of the mutation supply rate on the rate of adaptation in well-adapted versus poorly adapted populations are supported.

Our findings have three evolutionary implications. First, higher mutation rates need not accelerate the pace of evolutionary adaptation in asexual populations. Acceleration proportional to the mutation rate occurs only in limited circumstances, such as small effective population size, where an evolving population spends most of the time waiting for beneficial mutations (16). This is particularly relevant to bacterial pathogens that may experience severe population bottlenecks during colonization of the host (17). Second, mutators are common in asexual populations under conditions of rapid adaptive evolution because such conditions provide numerous opportunities for mutators to hitchhike to high frequency with beneficial mutations to which they are linked (3, 5). Mutators need not—and often will not—substantially accelerate adaptive evolution. Third, clonal interference imposes a speed limit on adaptive evolution in asexual populations, because two or more beneficial mutations that arise in different lineages cannot be combined into the same lineage. An advantage of sex is that it allows beneficial mutations to be combined into the same lineage, minimizing clonal interference and removing the speed limit on adaptive evolution that constrains asexual populations (7, 18, 19).

  • * Present address: Laboratory of Microbiology, Wageningen Agricultural University, Wageningen, The Netherlands.

  • To whom correspondence should be addressed. E-mail: arjan.devisser{at}


Stay Connected to Science

Navigate This Article