Multiple modes of selection underlie repeated and human-mediated adaptation in a formerly migratory fish.

The extent to which we can predict evolution is crucial in our era of rapid anthropogenic change. Alewives (Alosa pseudoharengus) in the Atlantic coastal USA are a unique model to test for evolutionary predictability in an anthropogenic context, as multiple, formerly anadromous (migratory from ocean to freshwater) populations have been independently restricted to freshwater (landlocked) by dams built in the last 350 years. Landlocked alewives show parallel changes in life history, feeding morphology, and osmoregulatory physiology. To test if recent freshwater adaptations are repeatable and predictable at the genomic level, we compared whole genomes of four landlocked and one anadromous population representing the ancestor. We determined that repeated positive selection is rare, limited to a single region on a single chromosome. Despite this, candidate analysis revealed that regions of repeatability do occur-in some populations but not others-in genes with putative function in freshwater adaptation, most notably in those involved in osmoregulation. Surprisingly, the strongest signal of selection in the genome was not one of positive selection, but one of conserved, balancing selection in a single gene family known as protocadherins, which play an important role in neural circuit formation and neuron recognition. Our results suggest that constrictive demographic histories and/or a polygenic nature of the complex trait architecture limits parallel selection at the genotypic level despite parallelism of phenotype. This highlights the need to understand both demography and trait architecture when determining the degree to which evolution is predictable.