A central goal in evolutionary biology is to understand the origins of adaptive variation that are linked to ecology. However, the ecological forces that drive natural selection and ultimately shape adaptation are dynamic, and may change across years 123. Consequently, there is growing appreciation that an in depth understanding of adaptation in natural populations requires long-term studies of selection 13.

A long-standing assertion is that variation in limb length and body size of Anolis lizards in the Greater Antilles arose via natural selection in response to differences in ecology 45. Long limbs increase sprinting speed on broad perching surfaces such as tree-trunks, whereas shorter limbs enhance agility on narrow twigs and branches 67. Previous studies 68 have suggested that traits such as running performance in different habitats may be important for predator avoidance and foraging efficiency, and body size variation may be a decisive factor in the strength of competitive interactions 9. Decades of study have repeatedly demonstrated the functional relationships between morphology and habitat use, particularly how suites of traits are favoured under varying ecological conditions 6810. Though phylogenetic studies of anoles 11 propose that these ecological mechanisms of diversification are conserved through time, we still lack a test of this hypothesis.

The hypothesis depends on consistency of selection pressures such that diversification occurring at the population level would ultimately translate into species level diversity 1213. Long-term work on Darwin's finches 1, another adaptive radiation, suggests that environmental fluctuations may make consistency of selection pressures unlikely. However, experimental examinations of environmental variation and its effect on selection have yet to be performed. Indeed, experimental studies of natural selection on combinations of morphological traits over multiple years remain very rare. Fluctuating selection pressures through time make point estimates of selection difficult to interpret, and consequently, a long-term perspective on selection becomes that much more important. Here we provide data from a four-year study to test the hypothesis that changing environmental conditions alter the nature of natural selection on Anolis lizards in the wild. We measured selection on natural and experimental populations, and used remote sensing and field-based measures of precipitation and vegetation changes to link our selection data to ecological change.

We studied viability selection (i.e., survival to the end of the breeding season) acting on wild populations of male A. sagrei (N = 133, 98, 111, and 148 lizards during 2003–2006 respectively) on Kidd cay, a small island (~1600 m²) near Georgetown Exuma, Bahamas, and on an offshore cay (Nightmare cay; ~500 m², N = 98 and 92 during 2005 and 2006 respectively). We studied the natural resident population on Kidd cay during 2003, 2005 and 2006, but during 2004, as part of a different study 14, we experimentally replaced all resident male A. sagrei with males from an immediately adjacent site on Great Exuma. Nearly all lizards (ca. 85%) in our study population mature and die in a single year 79, so each year we studied selection on a different cohort of lizards (the few surviving individuals were excluded from subsequent year's analyses). Though the agent of selection acting on these lizards is difficult to document unambiguously, previous studies suggest that the bulk of mortality arises through predation 15 and competition over territories 9, both of which may depend on variation in locomotor performance in vegetated habitat 716.

We began our study of natural selection during 2003. The following year we experimentally replaced male lizards on Kidd cay, our principal study site, with individuals from an adjacent site on Great Exuma 14. Coincidentally, that same year the Bahamas experienced a severe drought that we quantified using both satellite remote sensing and ground-based rainfall data (Figs. 1, 2A). Satellite data of the entire study area corroborated the rainfall data collected at one specific point on the ground. Thus, we can be reasonably certain that our measures on the ground are representative of adjacent study sites (e.g., Nightmare cay). During 2004, under-story vegetation and canopy foliage on the island were greatly reduced, and this fine scale variation was reflected in a reduction in NDVI (see below).

Changes in vegetative cover over the course of the four-year study impacted habitat use by lizards. Lizards perched more frequently on the ground during the drought relative to wet years when they perched more frequently in the vegetation (χ² = 14.96, df = 3, P < 0.0001; Fig. 2B, C). Drought conditions ended in 2005 with an increase in rainfall (Fig. 1, 2A), which led to increased growth of under-story vegetation and shifts in habitat use by lizards (Fig. 2B, C). Use of vegetated habitat increased significantly from 2004 to 2006 though it remained below pre-drought values. Similarly, the frequency of leaf-use for perching and for cover increased during 2005 and 2006 as lizards moved off of narrow plant stems and onto new vegetative growth (χ² = 13.37, df = 3, P = 0.0003; Fig 2C).

Impacts of the drought on vegetation structure and habitat use by lizards coincided with shifts in natural selection on limb length and body size (snout-vent-length, SVL). During the three wet years of our study, correlational selection gradients acting on size corrected limb lengths (residuals from the regression of limb length against SVL) ranged from significant and positive to significant and negative (Fig. 3; 2003: γ_1,2= 0.29 ± 0.16; 2005: γ_1,2= -0.13 ± 0.18; 2006: γ_1,2 = -0.21 ± 0.09). This indicates a change in the shape of the fitness surface from saddle shaped to dome shaped 3435. An alternative to focusing on the shape of the correlational selection surface (i.e., the sign of the gradients) is to ask simply whether or not selection acted on limb lengths in wet versus dry years. A plot of the absolute values of correlational selection gradients through time (Fig. 2D) shows that correlational selection gradients were significantly different from zero only during wet years. By contrast, selection operated on body size and not on limb lengths during the drought year and immediately following the drought (Fig. 2E and Table 1).

During the drought selection on body size was significant and disruptive (γ = 0.60 ± 0.22). Following the drought, natural selection favoured smaller male body sizes (β = -0.23 ± 0.11). In the two years following the drought (Fig. 2A), the island's vegetation gradually recovered (e.g., Fig. 4 increases in NDVI,) and we observed a return to selection on limb length as lizards moved back onto the vegetation (Fig. 2E). The differences in selection on body size and limb traits was significant among years (hindlimb × forelimb × SVL × wet/dry F_1,464 = 4.69, P = 0.03). In sum, we observed selection on limb length only in wet years, selection on body size alone during the drought, and selection for smaller body size the year immediately after the drought. Thus variation in selection was intimately related to variation in ecology.

To test whether changes in habitat structure were sufficient to explain the dynamic patterns of selection observed in our natural study population, we measured selection on a separate island that we experimentally manipulated during the wet year of 2005. Similar to patterns observed during the actual drought, selection acted significantly on body size (β = 0.55 ± 0.23; Fig. 5) but not on limb morphology (γ_1,2= -0.57 ± 0.42) following our experimental vegetation removal.

In 2006, observations in the field suggested re-growth of the under-story and substantial increases in new stem growth in the canopy on the island that we had manipulated the previous year. We again estimated selection on a new cohort of experimentally introduced males (92 juvenile male lizards from the same source-population as used the previous year, all other males were removed from the island) and similar to results from the natural site, in the year after our experimental manipulation of the vegetation, selection on body size was relaxed and non-significant (β = 0.06 ± 0.19; Fig. 5; Table 2). We did not detect significant correlational selection on limb lengths in the population where we experimentally manipulated vegetation (γ_1,2= 0.10 ± 0.23), but for both body size and for limb length, a comparison of slopes revealed significant differences between natural populations and those in which we manipulated vegetation (body size P < 0.001 and limb length P < 0.002).

Together, these results provide strong evidence that ecology can have dramatic impacts on the selective landscape through time 36. We have demonstrated that natural selection operated on different traits in wet and dry years and that this change was associated with changes in habitat use. During wet years, when abundant vegetative habitat was available to lizards for perching, selection pressures were variable among years but operated primarily on limb traits. By contrast, during the drought, when vegetation died off and less habitat was available for perching, natural selection operated primarily on body size. This difference is likely explained by the fact that limb length is linked to variation in locomotor performance on broad versus narrow perches 78, whereas body size may be more important in mediating competitive interactions 37 and surviving periods of physiological stress 3839.

Though demonstrations of natural selection in the wild have become more common in recent years, studies that convincingly illustrate the ecological basis for alternative adaptive optima are rare 34041 and experimental studies such as this one, all the more so. Anolis lizards provide an ideal opportunity to experimentally bridge ecological changes with changes in the adaptive landscape, because the correlations between morphology, performance, and habitat use have been so thoroughly documented in this group 15. Long hind and forelimb lengths increase sprinting speed on broad perching surfaces, but may compromise agility on narrow surfaces, or in complex habitats 68. Body size is likely important in competitive interactions 94243, and this may be especially true when reduced habitat availability places a premium on territoriality 9. Large body size may also be important for surviving periods of physiological stress brought on by lack of food or water.

A limitation of our study, reflecting one of the many constraints of working in natural systems, is that the drought during 2004 coincided with our experimental replacement of male A. sagrei at our main study site 14. It is therefore impossible to conclude with certainty that the observed changes in selection during 2004 arose solely due to drought, rather than our manipulation of the male lizard population. Only a second or third estimate of selection in un-manipulated populations would have allowed us to examine this. However, several lines of evidence suggest that our results are due to natural processes rather than our manipulation. First, we have previously shown that the introduction of novel male A. sagrei did not have a direct impact on selection dynamics 14. Second, behavioral, climatological, and habitat data all support a dramatic effect of the drought. Finally, our experimental manipulation of the habitat during a naturally wet year provides corroborative evidence that habitat use may drive selection.

In addition to having less available habitat, lizards may have spent less time in vegetation during the drought because of changes in prey availability or predation intensity 15 (e.g., a reduction in foliage may have rendered lizards more visible to avian predators). Similarly, changes in precipitation could alter rates of parasitemia as lizards in our study populations are affected by malaria and other blood borne pathogens (Bonneaud et al., in prep). Thus, an additional caveat to consider is that our manipulations of the habitat simulated some of the effects of drought on vegetation, but could not account for potential changes in diet, disease, or predation regime. Consequently, our experimental results may underestimate differences in selection on limb length between wet and dry years.

Making associations between intra- and inter-specific variation has been a central theme of studies attempting to understand the mechanisms giving rise to the adaptive radiation of Anolis lizards. Implicit in this approach is the assumption that processes that gave rise to patterns of species diversity can be revealed by studying processes at the population level 12. Phylogenetic studies of Anolis lizards have suggested that the same ecomorph may evolve repeatedly in response to similar environmental conditions, despite having independent evolutionary histories 11. Our results suggest an important caveat to these findings by demonstrating that natural selection may favour different traits across years depending on environmental conditions. Sustained environmental change, such as that predicted by global climate models, could potentially drive selection pressures to act on different suites of traits 44, thereby fundamentally altering adaptive landscapes 3644. Under these circumstances future processes at the population level might bear little resemblance to the processes that gave rise to present interspecific variation.

RC was responsible for the research planning, field data collection, statistical analysis, and drafting the manuscript. WB gathered and analyzed remote sensing data. TBS contributed to the research planning. All authors were involved in editing the manuscript.