A key challenge in understanding how natural selection operates is to identify the mutations and genes that make it possible. Positive selection on beneficial mutations distorts linked variation by altering the site frequency spectrum, the configuration of haplotypes, and population differentiation. By comparing patterns of sequence variation to neutral predictions across genomes, the targets of positive selection can be located. We applied this logic to an unusual population of house mice that shows phenotypic and ecological hallmarks of selection. Mice from Gough Island are twice the body size of mainland mice, eat live seabirds, maintain a very high population density, and inhabit an environment without predators or humans. We used massively parallel short-read sequencing to survey the genomes of 14 Gough Island mice. We computed a set of summary statistics to capture diverse aspects of variation across these genome sequences, used approximate Bayesian computation to reconstruct a null demographic model, and then applied machine learning to estimate the posterior probability of positive selection in each region of the genome. We conducted parallel analyses on genome sequences from 8 mice from Germany, treating them as representatives of a mainland reference population. A few thousand 5kb windows show strong evidence for positive selection in Gough Island mice but not in German mice. Genic regions and the X chromosome contain disproportionate shares of these selection windows. Over-represented gene ontologies in selection windows emphasize neurological themes. Inspection of genomic regions harboring many selection windows with high posterior probabilities pointed to genes with known effects on exploratory behavior and body size as potential targets. Some genes in these regions have missense mutations and/or putative regulatory mutations with large differences between Gough Island mice and German/French mice in the frequency of the derived allele; these are candidates for adaptive variants. Our results provide a genomic portrait of adaptation to island conditions and position Gough Island mice as a powerful system for understanding the genetic component of natural selection.

SNP vcf files.

Summary statistics used for ABC inference and selection scan.