Abstract
Caligid sea lice represent a significant threat to salmonid aquaculture worldwide. Population genetic analyses have consistently shown minimal population genetic structure in North Atlantic Lepeophtheirus salmonis, frustrating efforts to track louse populations and improve targeted control measures. The aim of this study was to test the power of reduced representation library sequencing (IIb-RAD sequencing) coupled with random forest machine learning algorithms to define markers for fine-scale discrimination of louse populations. We identified 1286 robustly supported SNPs among four L. salmonis populations from Ireland, Scotland and Northern Norway. Only weak global structure was observed based on the full SNP dataset. The application of a random forest machine-learning algorithm identified 98 discriminatory SNPs that dramatically improved population assignment, increased global genetic structure and resulted in significant genetic population differentiation. A large proportion of SNPs found to be under directional selection were also identified to be highly discriminatory. Our data suggest that it is possible to discriminate between nearby L. salmonis populations given suitable marker selection approaches, and that such differences might have an adaptive basis. We discuss these data in light of sea lice adaption to anthropogenic and environmental pressures as well as novel approaches to track and predict sea louse dispersal.