Human disturbance and the magnitude and predictability of phenotypic divergence

Santiago Ortega , Eduardo S. A. Santos , Kiyoko Gotanda, Sarah Sanderson, Lucas Damian Gorne, Andrew Hendry, Shinichi Nakagawa 

Environmental change can drive phenotypic change within species, but it remains unclear whether such responses produce a general and predictable degree of phenotypic divergence across populations. We conducted a preregistered reanalysis of PROCEED v6.2, comprising 7,186 contrasts from 254 references and 253 species, to test whether the magnitude and predictability of phenotypic divergence differ among disturbance contexts and across temporal, biological, and methodological characteristics of studies. We used ln M to quantify phenotypic differences between paired population samples relative to the variation retained within them and fitted phylogenetic location--scale meta-regressions to model average divergence and residual heterogeneity jointly. Across the dataset, average phenotypic divergence was substantial but did not consistently exceed within-population variation, while total heterogeneity reached 96.6%, providing little support for a universal magnitude of divergence. Disturbance contexts differed more strongly in predictability than in average divergence. Landscape change and introduced populations showed greater divergence than in situ natural variation, while landscape change was also comparatively predictable. By contrast, responses to introduced species consistently showed greater variation among contrasts than most other contexts, whereas the lower predictability of hunting or harvesting and in situ natural variation depended partly on a single large Pacific salmon dataset. Moderator associations should be interpreted cautiously because study characteristics were unevenly represented and partly confounded, and some results were sensitive to robustness analyses. Overall, environmental change can produce substantial phenotypic divergence, but neither its magnitude nor its predictability is uniform across systems. Accounting for heterogeneity therefore reveals when broad comparative responses are likely to generalize and when population-specific evidence is required.