Evaluation of the Rate of Evolution in Natural Populations of Guppies (Poecilia reticulata)

See allHide authors and affiliations

Science  28 Mar 1997:
Vol. 275, Issue 5308, pp. 1934-1937
DOI: 10.1126/science.275.5308.1934


Natural populations of guppies were subjected to an episode of directional selection that mimicked natural processes. The resulting rate of evolution of age and size at maturity was similar to rates typically obtained for traits subjected to artificial selection in laboratory settings and up to seven orders of magnitude greater than rates inferred from the paleontological record. Male traits evolved more rapidly than female traits largely because males had more genetic variation upon which natural selection could act. These results are considered in light of the ongoing debate about the importance of natural selection versus other processes in the paleontological record of evolution.

Natural populations of guppies in Trinidad are found in simple communities replicated in a series of drainages (1) that can be characterized as either high or low predation. In high-predation communities, guppies are found with the pike cichlid, Crenicichla alta, and other species of cichlids and characins. Some of these species prey preferentially on large, mature-size classes of guppies (1, 2). Predators are excluded from upstream portions of each drainage by rapids or waterfalls, yielding low-predation communities. Here, guppies co-occur with only the killifish Rivulus hartii. Rivulus is an omnivore that sometimes preys on guppies. Guppies from high-predation sites experience significantly higher mortality rates than those from low-predation sites (3).

When reared in uniform laboratory conditions, guppies from high-predation sites attain maturity at an earlier age and smaller size than their counterparts from low-predation communities. They also devote more resources to each litter, produce more, smaller offspring per litter, and produce litters more frequently than guppies from low-predation localities (4, 5). All of these differences have a genetic basis (5). The differences in mortality rates among high- and low-predation communities provide a potential mechanism driving life history evolution (3, 6).

We evaluated the role of predators in selecting for these patterns by creating two episodes of directional selection (7, 8). In two streams (tributaries to the El Cedro and Aripo rivers), we found waterfalls below which were guppies in a high-predation community. Above the falls, there were no guppies and only the small guppy predator R. hartii. Guppies from below the barrier waterfall were introduced above the waterfall and hence moved from a high- to a low-predation site.

We evaluated the response to selection in the Aripo River study 11 years after the introduction (8). The descendants of the transplanted guppies matured at a later age and larger size than the control population found below the barrier waterfall (Table 1). They also produced fewer, larger offspring per litter and devoted a smaller proportion of their consumed resources to reproduction early in life than did guppies from the control population. Their life histories had thus evolved to be similar to those of guppies naturally occurring in low-predation communities.

Table 1.

Comparisons of guppies from low-predation (experimental) sites with those from high-predation (control) sites, made in controlled laboratory conditions on the second generation of laboratory-reared fish from each locality, following the methods of previous studies (7, 8). Differences between populations that are observed under such conditions are assumed to have a genetic basis and thus provide evidence that evolution has occurred. The means are least square means derived from the statistical comparisons of these values (22) and are adjusted for covariates and unequal sample sizes (7, 8). Sample sizes for the control in the Aripo River were 30 males and 29 females; for the experiment, 24 males and 22 females. In the El Cedro River 4-year assay, the control included 44 males and 43 females; the experimental included 42 males and 43 females. In the El Cedro River 7.5-year assay, the control included 36 males and 40 females; the experimental included 129 males and 110 males. “Response” is the estimated response to selection, or the difference between the control and experimental values. The statistical significance of the response is based on an analysis of variance (7, 8). Significance is based on one-tailed probabilities because the comparative studies provided a basis for predicting how the populations would evolve in response to the introduction. “Rate” is the estimated rate of change in darwins, calculated as (ln X2 − ln X1)/Δt, where X1 and X2 are the values of the trait at the beginning and end of the time interval, respectively, and Δt is the length of the interval in years. We estimated X1 and X2 using the values of the traits for the control and introduction populations, respectively. Gingerich (11) reports that artificial selection experiments attained rates of 12,000 to 200,000 darwins, with a geometric mean of 58,000 darwins. In contrast, the geometric mean rates for the fossil record range from 0.7 to 3.7 darwins, with the estimated rate being inversely proportional to the time interval over which it was evaluated; ns, not significant.

View this table:

We evaluated the response to selection by fish in the El Cedro River 4 (7) and 7.5 years after the introduction. Males from the experimental site evolved delayed maturity and a larger size at maturity within 4 years (Table 1). No female life history trait showed a significant change by 4 years. By 7.5 years, females from the experimental site also matured at a later age and larger size than those from the control site (Table 1). No other aspect of the female life history had changed significantly by 7.5 years. Therefore, for this replicate, only some components of the life history have evolved thus far; all changes are in the same direction as in the Aripo experiment.

We analyzed the response to selection for just the age and size at maturity in males and the age and size at first parturition in females because these traits responded to the manipulation in both experiments. For each trait, we estimated the response to selection (R) as the difference between the mean of the trait for lab-reared guppies from the control and experimental populations (9) (Table 1). We then used R to quantify the relative rate of evolution in terms of darwins (10, 11). The estimated rates of phenotypic evolution range from 3700 to 45,000 darwins (Table 1). They are similar in magnitude to rates that have been obtained by artificial selection and four to seven orders of magnitude greater than those observed in the fossil record (11).

One notable result in the 4-year assay for the El Cedro River is the 10-fold difference between the sexes for the rate of change in the age at maturity and a 3- to 5-fold difference for the size at maturity (Table 1). When the same comparisons are made in the 7.5-year assay, the relative rate of change (“Rate” in Table 1) for male age was still higher than that for female age, but the difference was far less pronounced; the absolute amount of change (R) was approximately equal. Males appear to have evolved to an end point within 4 years, so the 7.5-year result averages the same amount of change over a longer interval of time. On the other hand, females continued to evolve over the entire 7.5-year period. Differences in the estimated rates of change can also be attributed to changes in the experimental conditions between the 4- and 7.5-year assays (12).

The results for the Aripo River comparison are similar to those for the 7.5-year assay on the El Cedro; the relative rate of change in male age at maturity was twice the rate of female age, and the absolute rate of change was 23% greater (Table 1). The rate of change in size was slightly less for males than for females. If the pattern of evolution was similar to that in the El Cedro River, with males rapidly evolving to a new value then leveling off before the 11-year assay whereas the females continued to change, then this comparison would also tend to underestimate the rate of evolution in males relative to females.

Inferences about the strength of natural selection require estimates of generation time (13) and the genetic variance-covariance matrix (Table 2) (14), which quantifies the genetic variation available for a response to natural selection. The heritabilities of male age and size at maturity were high and significantly greater than zero in all four estimates (Table 2). These values are exceptionally high for life history traits: a more typical value is between 0.1 and 0.2 (15). The corresponding values for age and size at first parturition in females were lower, with the only significant value being for the age at first parturition in females from the El Cedro River (Table 2). One possible explanation for the difference in the heritabilities of male and female traits is that some of the genetic variation for these traits in males is associated with the Y chromosome, as reported for other species of Poeciliid fishes (16).

Table 2.

Heritabilities and genetic variance-covariance matrices G. We used restricted maximum likelihood estimation (23) to estimate the genetic variance and covariance for the age and size at maturity. Data were derived from the same experiments that were used to evaluate the response to selection (Table 1). Heritabilities and genetic correlations are reported, with genetic variances and covariances in parentheses (24); ns, not significant.

View this table:

The genetic correlations between age and size at maturity (males) and age and size at first parturition (females) were very high for the Aripo River study and moderately high for the El Cedro River study. They were significant for both male comparisons (Table 2). The absence of significance for equally high correlations in the female data is attributable to the lower heritabilities for these traits in females. The genetic basis of age and size is thus broadly overlapping. Selection on one trait strongly influences the evolution of the other (9, 14).

The selection gradient β (Table 3) estimates the change in relative fitness as a function of the change in a trait as other traits are held constant. A positive coefficient implies that fitness increases as the value of the trait increases, whereas a negative coefficient implies that fitness increases as the value of the trait decreases (17). The bivariate β coefficients for males on the Aripo River indicate strong direct selection for increased age at maturity but weaker selection for decreased size at maturity (Table 3). The values for age and size increased together over time because of the positive genetic correlation between them (Table 2). The β coefficients for Aripo River females and El Cedro River males imply strong direct selection for increased age at maturity but weaker direct selection for increased size at maturity. Finally, the 7.5-year assay of El Cedro females indicates that direct selection on size was somewhat stronger than direct selection on age.

Table 3.

Estimates of the linear selection gradients and standardized coefficients of selection for the two introduction experiments. The selection gradients β = G−1R, where G−1 is the inverse of the additive genetic variance-covariance matrix and R is the vector of responses to selection (Table 1). These values were divided by the estimated number of generations during the experiment and multiplied by the phenotypic standard deviation estimated from the control population to yield the standardized coefficients reported here. Values are not reported for the El Cedro females at the end of the 4-year assay because there were no significant changes in the traits in question at that point. Coefficients of selection S were estimated as S = Pβ, where P is the phenotypic variance-covariance matrix. All computations were based on log-transformed data. Confidence intervals for β and S, given in parentheses, were calculated by parametric bootstrap. If the lower confidence limit is greater than zero, then the coefficient is significantly greater than zero (25).

View this table:

Coefficients of selection S are an alternative way of characterizing the process of natural selection (Table 3): S estimates the covariance between a trait and fitness. It incorporates the effects of direct selection on that character and the indirect effects of selection on correlated characters. When rates of evolution are compared, a given trait may evolve slowly because the intensity of selection on it is low, its heritability is low, or its rate of change is affected by selection on a correlated trait. We can use these coefficients, in combination with the other classes of information, to evaluate the causes of the apparent differences in the rate of evolution of males and females on the El Cedro River (Table 1). The rate of evolution tended to be higher in males (Table 1), possibly because males had higher heritabilities for these traits (Table 2). The most instructive comparison is between the 4-year values for males and the 7.5-year values for females because this avoids the bias caused by males reaching a plateau at 4 years. The coefficients of selection for age are higher for males than for females, but the magnitude of the difference (0.310 for males versus 0.091 for females) is smaller than the difference in the rate of evolution or the response (Table 3 versus Table 1). Similarly, in the Aripo River, the rate of change in male age at maturity is greater than that for female age, but the coefficient of selection is much higher for females. It thus appears that the differences in rate between the sexes are attributable to the differences in the heritabilities of the traits, with females evolving more slowly because of their lower genetic variance for age and size at maturity, rather than because of weaker selection on female traits.

If evolution can be so fast, why does it appear to be so slow in the fossil record? First, evolution is only sustained in response to a changing environment (18); when a new optimum is attained, no more evolution is expected [El Cedro River males in this study (19)]. Second, if environmental conditions vary erratically, so will patterns of evolution, as seen in Galápagos finches (20). Evaluating evolution with the fossil record averages across intervals of no change, intervals of rapid change, and possibly includes reversals in the direction of change, yielding an estimate of rate averaged over the entire interval (11). The net effect could well be no measurable change in morphology, or “stasis.” On the other hand, sustained directional selection can support far more rapid directional change than seen in the fossil record.

The evidence from studies of microevolution (19, 20) bears on the current debate over micro- versus macroevolution and the patterns of change recorded in the fossil record. In the fossil record, there is a well-established pattern of periods of little or no change (stasis) punctuated by brief intervals of rapid change associated with the origins of new taxa. Some have argued that selection among individuals within populations (natural selection) cannot account for these large-scale trends in evolution (21). Specifically, Gould and Eldredge argue for the necessity of bursts of speciation followed by species selection to sustain the rapid change associated with punctuations in the fossil record (21). Our work cannot address the efficacy of mechanisms other than natural selection, but it extends our understanding of what is attainable through this process. It is part of a growing body of evidence that the rate and patterns of change attainable through natural selection are sufficient to account for the patterns observed in the fossil record (18).


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
View Abstract

Stay Connected to Science

Navigate This Article