The environmental determinants of natural selection

Background/Question/Methods

Although >6000 estimates of phenotypic selection have been published, the environmental factors that cause this selection have rarely been identified, making it difficult to predict how populations will respond to environmental change.  One approach to identifying the causes of selection is to experimentally manipulate an environmental factor and compare estimates of selection between treatments.  If the strength or direction of selection on a trait differs between treatments, then we can infer that the manipulated environmental factor is a cause of selection on that trait.  But estimates of phenotypic selection from experimental studies, unlike estimates of selection from natural populations, have yet to be compiled and synthesized.  We compiled a database of estimates of selection from >140 experimental studies published between 1991 and 2013.  We used this database — employing a meta-analytical approach — to test two hypotheses as to why different environmental factors will be more or less likely to alter the strength of selection: 1) that the biotic environment exerts stronger selection than the abiotic environment; and 2) that environments with greater effects on mean fitness will cause stronger selection.  Additionally we asked whether the strength of selection in experimental studies is comparable to selection in natural populations. 

 Results/Conclusions

We found that the strength of directional selection did not vary among environmental factors, which is not consistent with the hypothesis that the biotic environment exerts stronger selection than the abiotic environment.  We found however that there was a relationship between the magnitude of change in mean fitness between treatments and the magnitude of change in directional selection between these treatments, supporting the hypothesis that environmental factors that have a large effect on mean fitness will exert stronger selection than factors that have a small effect on mean fitness.  Furthermore, we found that differences in directional selection between experimental treatment levels were larger than differences in selection between temporally or spatially replicated natural populations.  However, the distributions of estimates of directional selection from experimental studies and estimates of selection in natural populations were comparable, supporting the use of experiments to identify the causes of natural selection.