Constraint to Adaptive Evolution in Response to Global Warming

See allHide authors and affiliations

Science  05 Oct 2001:
Vol. 294, Issue 5540, pp. 151-154
DOI: 10.1126/science.1063656


We characterized the genetic architecture of three populations of a native North American prairie plant in field conditions that simulate the warmer and more arid climates predicted by global climate models. Despite genetic variance for traits under selection, among-trait genetic correlations that are antagonistic to the direction of selection limit adaptive evolution within these populations. Predicted rates of evolutionary response are much slower than the predicted rate of climate change.

Plants have responded to historical climate change by migration and adaptation (1). However, habitat fragmentation is likely to impede migration in the future (2). Furthermore, migration may be slower than during the recession of the glaciers, because migration will depend on seedling establishment in occupied habitats (3). The persistance of populations thus hindered from spread into higher latitudes may depend more heavily on adaptive evolution.

Evolutionary response requires genetically based variation among individuals. However, even given this substrate for natural selection, evolution may be constrained by genetic correlations among traits that are not in accord with the direction of selection (4,5), correlations termed “antagonistic.” For example, if selection favors high values of two traits but these traits are negatively genetically correlated, selection response can be inhibited (Fig. 1A).

Figure 1

Illustration of the influence of genetic correlations among traits on selection response. (A) Hypothetical positive genetic correlation (r A) between two traits (each point represents the breeding value for each of two traits). There are two selection scenarios. For R (reinforcing), selection is in the same direction on traits; the depictedr A is in accord with the direction of selection, enhancing evolutionary response; thus, the genetic correlation is reinforcing. For A (antagonistic), selection is in the opposite direction for both traits; r A is antagonistic to the direction of selection, inhibiting evolutionary response. (B) Scatter plot of MN population reproductive stage and leaf number breeding values (centered on the phenotypic mean), showing signficant negative genetic correlation that is antagonstic to the positive vector of joint selection on these traits. (C) Scatter plot of the MN population leaf thickness and leaf number breeding values (centered on the phenotypic mean), showing signficant positive genetic correlation that is antagonistic to the negative vector of joint selection.

We evaluated the evolutionary potential of three populations of the native annual legume Chamaecrista fasciculata, which were sampled from an aridity gradient in tallgrass prairie fragments in the U.S. Great Plains (Fig. 2A) (6). Natural selection on phenotypic variation inC. fasciculata differs across this geographic range (7). Field and common garden studies of Minnesota (MN), Kansas (KS), and Oklahoma (OK) populations of C. fasciculata demonstrated clinal variation and genetic divergence with respect to physiological and morphological traits (7). Greenhouse drought experiments also demonstrated adaptation of these populations to different water availability conditions; northern plants are less drought-tolerant than southern plants (7).

Figure 2

(A) Three focal populations in Minnesota, Kansas, and Oklahoma, shown with long-term average isoclines of α for evergreen trees [1951–1980 (25)]. α is an integrated measure of seasonal growth-limiting drought stress on plants that accounts for temperature, precipitation, and soil texture. (B) Twenty-five to 35-year prediction of α for Minnesota (8).

We used this spatial gradient in climate as a proxy for the temporal trend in climate predicted for northern populations with global warming. For example, one global climate model predicts that the MN population will experience soil moisture conditions similar to the current climate of KS by 2025–2035 (Fig. 2B) (8). To predict rates of adaptation to climate change, we estimated evolutionary trajectories for three populations reciprocally planted in three environments. The evolutionary trajectory of a northern population reared in progressively more southern sites provides insight into the population's adaptive potential in the face of global warming.

We produced pedigreed seeds for MN, KS, and OK populations by controlled crosses in the greenhouse according to a standard quantitative genetic design (9). Progeny from these crosses were reciprocally planted into field sites in MN, KS, and OK (10). We measured traits subject to differing natural selection under distinct drought regimes (fecundity and leaf number) or varying clinally across the geographic range of this study (leaf thickness and the rate of phenological development) (7). In mid-summer we recorded the leaf number and reproductive stage of each plant (11) and collected the uppermost fully expanded leaf. At the natural end of the growing season, we recorded total pod number and seed counts from three representative pods; from these measures, we estimated total lifetime fecundity (12).

We used restricted maximum likelihood (REML) (13, 14) to conduct multivariate quantitative genetic analyses of all the traits jointly to obtain estimates of the additive genetic covariance (CovAij) between all pairs of traits for each population in each site (15). The predicted change in a trait (Δ) resulting from a single generation of natural selection on phenotypic variation is simply the additive genetic covariance between relative fitness and that trait (16, 17) when other traits under selection are taken into accountEmbedded Image(1)where w is individual relative fitness (absolute fitness divided by mean fitness), and z is the vector of traits. Although these predictions take into account all the traits under consideration, they could be modified by selection on other genetically correlated traits that have not been considered (18, 19).

These evolutionary trajectories are based on the narrow-sense heritability and the strength and direction of selection (Table 1), as well as on the influence of among-trait additive genetic covariance (Table 2). For comparison, we also present univariate predictions from analyses of the traits separately. These univariate predictions involve only the genetic variance and selection on a single trait and indicate expected evolutionary response if traits were genetically independent and, hence, would evolve independently.

Table 1

The multivariate prediction of evolutionary response after one generation of selection, CovA[w, z], for three traits measured on three populations of C. fasciculata reared in three environments. Univariate predictions and narrow-sense heritabilities, h 2 (6), from separate analyses are given below evolutionary trajectories. Significance levels are not corrected for multiple testing. Of the 108 tests conducted (Tables 1 and 2), one would expect to erroneously assign significance in 5.4 cases by chance alone, assuming α0.05.

View this table:
Table 2

Additive genetic correlations,r A, among traits. The concordance ofr A with the direction of the vector of selection on pairs of traits is given in parentheses (R = reinforcing; A = antagonistic).

View this table:

Three cases are relevant to global warming (the MN population in KS and OK and the KS population in OK). Seed production was dramatically reduced in the nonnative populations as compared with the local population (for the MN population, 84% in KS and 94% in OK; for the KS population, 42% in OK) (Fig. 3A). In each of these cases, we predict adaptive evolution in response to climate warming, because the multivariate prediction is in a direction consistent with that of the univariate prediction. Overall, 14 of 18 evolutionary predictions of nonnative populations are toward the mean of the native population, which further supports the interpretation that the direction of evolutionary response is adaptive (none of the four exceptions are statistically significant). Yet with only one exception, the multivariate prediction of evolutionary response is less in absolute magnitude than the univariate prediction; in many cases, half or less. Considering the MN population grown in KS and OK, selection favoring plants bearing more and thicker leaves is expected to result in evolutionary change consistent with the direction of selection but less than if the traits evolve independently. The prediction of evolutionary response for reproductive stage of MN plants in KS (slower) is opposite that of OK (faster), which may reflect selection for different mechanisms of drought tolerance in the intermediate site versus drought avoidance in the more southern site. For transplants of southern populations to northern sites (the KS population in MN and the OK population in KS and MN), most of the multivariate predictions of evolutionary response are also less than the univariate ones.

Figure 3

Least-squares means and standard errors (very small) of (A) fecundity, (B) reproductive stage, (C) log (leaf number), and (D) log (leaf thickness) for MN (circles), KS (triangles), and OK (squares) populations reciprocally planted in each of MN, KS, and OK sites (26). The direction of the evolutionary trajectory is indicated with an arrow, and the number of generations required to achieve the phenotypic mean of the local population is shown in parentheses for the MN population reared in KS and OK (only leaf number in KS is statstically significant).

Why is evolutionary change predicted to be slow, given the significant heritabilities of most of the traits? Numerous additive genetic correlations are antagonistic to the direction of selection jointly on pairs of traits, as shown in Fig. 1, B and C (Table 2). Among-trait correlations that oppose the direction of selection can alter evolutionary response from expectation by (i) retarding the evolutionary response of heritable traits under selection, (ii) reversing the direction of selection response from expectation, and (iii) promoting the evolutionary response of traits not under direct selection. The first case is most evident here; the second is also illustrated by the case of the KS population at the KS site, for which the multivariate prediction of reduced leaf number conflicts with the univariate prediction for leaf number increase. These findings demonstrate that genetic relationships among traits can substantially influence evolutionary change. In each case where the univariate analysis would indicate substantial evolutionary change but the multiple trait analysis predicts a smaller change, at least one among-trait additive genetic correlation is opposite in sign to the vector of selection (antagonistic; Table 2 and Fig. 1, B and C). Although few among-trait correlations are individually significant (5 of 27), they constrain the genetic architecture of these populations and alter predicted selection response from expectation.

According to the climate model cited herein, the MN population is predicted to experience climate similar to the current climate of KS in only 25 to 35 years. Making the simplistic assumptions of constant genetic variation and selection coefficients, the number of generations required before the trait means of the MN population are expected to match those of the native KS population generally exceeds the time predicted for this climate change (reproductive stage, 21; leaf number, 42; leaf thickness, 79) (Fig. 3). The MN population is predicted to achieve the local population means in OK in fewer generations because of stronger selection and greater expression of additive genetic variance. However, these are probably underestimates of the number of generations required, because strong selection over as few as 10 generations can substantially deplete genetic variation (20); moreover, selection coefficients would not remain constant (21, 22). Furthermore, the extreme fitness costs in terms of seed production incurred by the MN population when reared in the KS or OK climate would influence the genetic variance, inbreeding, and demography of subsequent generations and hence population persistence. Thus, even though there is significant genetic variation for all but one of these traits, the rate of multivariate evolution is expected to be slower than the rate of climate change.

When the MN population is reared in the warmer and drier climates of KS and OK, slow evolutionary response is predicted even though this population harbors significant additive genetic variance for vegetative and phenological traits under selection. Similarly, little evolutionary response is predicted for the KS population at the OK site. We do not rule out the possibility that predicted selection responses that are in the direction of the local population mean but not statistically significant will, nevertheless, be biologically significant as populations experience an incrementally changing climate. This study demonstrates, however, that adverse additive genetic correlations among traits may severely retard evolutionary change.

It could be argued that species will persist in the face of global warming, because fossil evidence indicates that many taxa have survived through numerous episodes of climate change in the past. However, historical climate changes were generally much slower (by one or more orders of magnitude) than those predicted for the future (23,24). Slower changes may have provided opportunities for taxa to adapt to climate change while persisting in refuges or shifting ranges to new latitudes despite genetic constraints on adaptive evolution.

  • * To whom correspondence should be addressed.

  • Present address: University of Virginia, Department of Biology, Glimer Hall, Charlottesville, VA 22903, USA. E-mail: jre7e{at}


View Abstract

Stay Connected to Science

Navigate This Article