Multiple Fitness Peaks on the Adaptive Landscape Drive Adaptive Radiation in the Wild

See allHide authors and affiliations

Science  11 Jan 2013:
Vol. 339, Issue 6116, pp. 208-211
DOI: 10.1126/science.1227710

Pupfish Speciation

Evolution moves along phenotypic trajectories that can be visualized as a topographic landscape of multiple peaks of relatively high-fitness and low-fitness valleys. Martin and Wainwright (p. 208) examined the adaptive landscape of three species of Cyprinodon pupfishes. These species represent a recent adaptive radiation, each having moved into a difference niche within their specialized environment. Examining replicate hybrid transplants relative to parental types in high- and low-density enclosures, the authors recovered the specialist parental phenotypes and observed higher survival and growth. Thus, high density can drive multiple fitness peaks during the early stages of adaptive radiation.


The relationship between phenotype and fitness can be visualized as a rugged landscape. Multiple fitness peaks on this landscape are predicted to drive early bursts of niche diversification during adaptive radiation. We measured the adaptive landscape in a nascent adaptive radiation of Cyprinodon pupfishes endemic to San Salvador Island, Bahamas, and found multiple coexisting high-fitness regions driven by increased competition at high densities, supporting the early burst model. Hybrids resembling the generalist phenotype were isolated on a local fitness peak separated by a valley from a higher-fitness region corresponding to trophic specialization. This complex landscape could explain both the rarity of specialists across many similar environments due to stabilizing selection on generalists and the rapid morphological diversification rate of specialists due to their higher fitness.

Adaptive radiation, the rapid evolution of ecological and phenotypic diversity within a clade, may account for much of life’s diversity (1, 2). The ecological theory of adaptive radiation is founded on the unifying concept of the adaptive landscape, the topographical relationship between fitness and a continuous phenotypic space (1). This theory predicts early bursts of niche diversification due to rapid invasion of multiple, unoccupied fitness peaks after colonization, the evolution of a key innovation, or mass extinction (15). This “early burst” model is supported by microbial evolution (4), the fossil record (6), and species diversification in some clades (7), but the pattern is rare in comparative data (8). Indeed, observations of disruptive selection suggest that many populations are constrained from ascending fitness peaks (9, 10).

Most studies of selection estimate its local form—directional, stabilizing, or disruptive (1, 9)—but few investigate the broader topography of the fitness surface, particularly among multiple species (1113) or traits (14), or by manipulating phenotypes to measure the fitness of intermediates between species (12, 1517). However, measurement of the broader multivariate fitness landscape is necessary to demonstrate a local maximum and to visualize the selective surface during early bursts of niche diversification. Multipeak fitness landscapes have been demonstrated in laboratory microcosms (4) and inferred from resource availabilities (18), foraging performance (13), and mark-recapture (11) in the field, but not from fitness measurements of manipulated phenotypes. Thus, the key early burst prediction of multiple fitness peaks has never been experimentally tested in the wild.

We measured fitness landscapes directly from growth and survival of F2 hybrids from crosses among the three species in a sympatric adaptive radiation of Cyprinodon pupfishes on San Salvador Island, Bahamas. F2 hybrids spanned the range of morphological diversity represented in the three parental species. This small island radiation contains ecologically novel species and displays rapid morphological diversification rates similar to those of classic adaptive radiations (19), yet is less than 10,000 years old (20). The endemic radiation contains two trophic specialist species, a scale-eater and a hard-shelled–prey specialist (durophage), both novel ecological niches within Cyprinodon, and a third generalist species similar to the wide-ranging species C. variegatus. The three species co-occur in all habitats within the island’s shallow saline lakes containing only two other fish species. Although generalist populations are ubiquitous across similar environments in the Caribbean with identical fish communities, the San Salvador clade is one of only two sympatric radiations of Cyprinodon and exhibits morphological diversification rates up to 51 times faster than those of other young clades for functional trophic traits (19).

We tested the effects of competition in the wild on the topography of fitness landscapes during Cyprinodon adaptive radiation by measuring the fitness of 1865 hybrids placed in high- and low-density field enclosures. Wild-caught breeding colonies of all three species were used to generate outbred F2 hybrid populations for this experiment from F1 hybrid intercrosses and backcrosses (20). Hybrid populations from two isolated lakes, Crescent Pond (CP) and Little Lake (LL), were generated independently. First, these laboratory-reared juvenile F2 hybrids were measured for 16 morphological traits and implanted with coded wire tags. Next, we transported the hybrids to San Salvador Island and introduced them into a low- or high-density field enclosure in the respective lake from which their grandparents originated (CP: high/low-density n = 796/96 hybrids; LL: n = 875/98 hybrids). After 3 months, we recovered all surviving hybrids and assigned a fitness of 1 relative to 0 for unrecovered hybrids. Fitness was also estimated from the growth (the increase in standard length) of survivors in enclosures.

To provide a biologically relevant frame of reference for hybrid morphology, we laboratory-reared and measured a purebred F1 generation of all three species from each lake. We then plotted the hybrids in a discriminant morphospace separating the F1 purebred species in each lake and visualized fitness landscapes for survival (Fig. 1) and growth (fig. S1). Morphospace coverage was reduced in LL, probably due to missing a backcross to the durophage.

Fig. 1

Survival fitness landscape for F2 hybrids within the high-density field enclosure in each lake (left column: CP; right column: LL). (A and B) Individual F2 hybrid survivors (black dots) and deaths (gray dots) plotted within the discriminant morphospace (LD1 and LD2) estimated from laboratory-reared purebred species. Heat colors indicate survival probabilities estimated from thin-plate splines fit to the data by generalized cross-validation (effective df: CP, 7.6; LL, 19.6). (C and D) Laboratory-reared purebred species are superimposed within 95% confidence ellipses (dashed lines: generalist, blue dots; durophage, green dots; scale-eater, red dots) for reference. (E and F) Relative heights of the fitness landscape.

Fitness landscape topography was complex but largely congruent between high-density enclosures in each lake for both survival and growth (Fig. 1 and fig. S1). A local fitness peak in the high-density CP enclosure, isolated by declining fitness in all directions, corresponded to the phenotype of the generalist species (Fig. 1, A, C, and E). Phenotypic similarity between hybrids on this local fitness peak and the laboratory-reared generalist species cannot be attributed to familial relatedness or shared environments (20). Instead, this correspondence demonstrates strong stabilizing selection on hybrid phenotypes closely resembling generalists. There was a similar trend in LL (Table 1 and Fig. 2). Survival also declined with increasing total phenotypic distance from the generalist in both high-density enclosures (fig. S2).

Table 1

Significance of local regions of stabilizing (γ < 0) and disruptive (γ > 0) selection within survival fitness landscapes for F2 hybrids in high-density field enclosures (Fig. 1). The form of selection on hybrid phenotypes was tested for the quadratic intervals depicted in Fig. 2. Sample sizes within each interval are reported for each test. n, number of hybrids; GLM, generalized linear model.

View this table:
Fig. 2

Probability of F2 hybrid survival in high-density field enclosures within local transects between species (left column, CP; right column, LL). Smoothing splines (black line) and 95% confidence intervals (dotted gray lines) indicate hybrid survival along major phenotypic axes between species ellipses from Fig. 1 (insets at upper left). (A and B) Major axis between hybrids resembling durophage or generalist species. (C and D) Major axis between hybrids resembling generalist or scale-eater species. Interior rug plots show F2 hybrid survivors (upper) and deaths (lower). Exterior rug plots show laboratory-reared F1 purebred species (generalist, blue; durophage, green; scale-eater, red) for reference. Quadratic intervals used for parametric analyses (Table 1) are indicated with significance from logistic regression.

A second region of increased fitness [linear discriminant axis 1 (LD1) < 0 and LD2 < 0 in Fig. 1A: n = 120 hybrids, logistic z-score = –2.33, P = 0.020] corresponded to the phenotype and the diet (inferred from δ15N stable isotopes) of the hard-shelled–prey specialist (durophage; Figs. 1 and 2, figs. S3 and S4, and table S1). In the CP high-density enclosure, increased survival in this region was supported by parametric and permutation tests (20), although less strongly than the generalist peak. This fitness region was also significantly higher (permutation test, n = hybrids, P = 0.044) than the generalist peak (Fig. 1E) and was robust to alternative calculations of the discriminant morphospace (figs. S5 and S6). A similar trend of increased survival of hybrids resembling the durophage specialist was observed in LL (Figs. 1 and 2 and Table 1). Furthermore, CP hybrid survivors in this region occupied a higher trophic position than those on the generalist peak (δ15N stable isotope ratios: n = 64 hybrids, permutation test, P = 0.048; fig. S3 and table S1), reflecting the relative trophic positions of wild-caught durophage and generalist pupfishes (F1,22 = 8.78, P = 0.007; table S1). Thus, trophic divergence between hybrids mirrored trophic divergence in wild-caught species.

Hybrids resembling generalist and durophage phenotypes were separated by a valley of reduced fitness in both lakes (Fig. 2). Transects between species indicated significant disruptive selection on intermediate phenotypes (Table 1, Fig. 2). The form of selection on growth was not always consistent with survival (Table 1); however, multiple regions of increased fitness, corresponding to generalist and durophage phenotypes, were observed in both survival and growth fitness landscapes in the high-density CP enclosure and showed a similar trend in LL (Fig. 1 and fig. S1).

Hybrids resembling the scale-eater phenotype had greatly reduced survival and growth in both lakes at high and low densities (Figs. 1 and 2, fig. S1, and Table 1), demonstrating that low fitness over a large region of morphospace reproductively isolates this species from the others. Hybrid survivors resembling scale-eaters rarely ingested any scales (4 out of 11 relative to 49 out of 53 wild-caught scale-eaters); thus, scale-eating may require an extreme phenotype, not recovered in hybrids, for successful performance. Alternatively, field enclosures may not support the scale-eating niche; however, we consider this unlikely because the frequency of hybrids resembling scale-eaters within high-density enclosures (CP, 0.6%; LL, 0.9%) was similar to wild scale-eater frequencies [1.4% (20)], prey density was higher in enclosures, and fitness did not vary between density treatments. Overall, the reduced fitness of intermediate and transgressive hybrids supports the importance of postzygotic extrinsic reproductive isolation.

High-density enclosures provided a more competitive environment than low-density enclosures: Survival was 4 to 7 times higher and growth was 1.4 to 2 times higher in low-density enclosures (mean survival: CP, 11.4% versus 71.9%; LL, 11.1% versus 43.9%; logistic z = 11.8, P < 10−16; mean growth: CP, 0.196 versus 0.428 cm; LL, 0.164 versus 0.223 cm; t = –2.8, P = 0.005). The curvature of the fitness landscape was significantly greater in high-density enclosures in both lakes (Table 2 and figs. S1 and S7), supporting competition as the driver of multiple fitness peaks on the adaptive landscape. If intrinsic differences in fitness among hybrid phenotypes were responsible, complex fitness landscapes should occur in both field and laboratory environments. However, laboratory survival surfaces, estimated from hybrids that died during laboratory rearing, were flat (table S2 and fig. S8). Combined, these data indicate that multiple high-fitness regions were caused by competition for diverse resources, not intrinsic survival differences.

Table 2

Effect of competition on fitness landscape curvature. Three permutation tests assessed the significance of greater survival surface curvature in high-density enclosures relative to low-density enclosures (fig. S7), controlling for different sample sizes and morphospace coverage. Significance was determined from the number of randomized samples equal to or greater than the observed difference between treatments (20). Effective degrees of freedom (EDF) indicate the smoothness of the surface estimated by generalized cross-validation.

View this table:

This complex fitness landscape paints an intriguing picture of niche diversification driven by competition in Cyprinodon. The generalist species sits atop a local fitness maximum separated by a valley from a higher-fitness region corresponding to specialization on hard-shelled prey. Stabilizing selection on generalist phenotypes could explain the rarity of trophic specialists within Cyprinodon despite their higher fitness: Sympatric adaptive radiations of trophic specialists may have evolved in only two places throughout the Caribbean because most generalist populations are stranded on an isolated local maximum. When subpopulations escape this generalist peak, perhaps through increased competition, ecological opportunity, and large effective population size, the higher fitness of trophic specialists drives a burst of diversification.

The early burst model of adaptive radiation predicts a fitness landscape with multiple peaks at the onset of adaptive radiation (1, 2, 4, 7, 8). We simulated phenotypic diversity within the ancestral population that gave rise to an adaptive radiation of pupfishes in order to measure the initial topography of the fitness landscape. In contrast to theory and examples demonstrating that high-frequency phenotypes cause a fitness minimum due to negative frequency-dependent selection (9, 10), some of the most frequent phenotypes in our field enclosures occurred on a local fitness maximum, suggesting that the broader topography of adaptive landscapes is more strongly determined by stable performance constraints than frequency-dependent dynamics.

Supplementary Materials

Materials and Methods

Figs. S1 to S9

Tables S1 to S9

References (2151)

References and Notes

  1. Acknowledgments: Funded by NSF grant DDIG DEB-1010849, the Gerace Research Centre, and the Center for Population Biology. Permits were approved by the Bahamian government and the University of California, Davis. G. Mount and S. Romero assisted with research; T. Schoener, M. Turelli, A. Hendry, D. Nychka, L. Schmitz, and C. Boettiger provided feedback. C.H.M. designed the study, raised funding, performed research, collected and analyzed data, and wrote the paper. P.C.W. contributed materials, discussed design and results, and commented on the manuscript.
View Abstract

Navigate This Article