Genome-assisted prediction of amoebic gill disease resistance in different populations of Atlantic salmon during field outbreak

Abstract

The Atlantic salmon industry in northern Europe is experiencing increasing losses due to the amoeba Paramoeba perurans, which is the causative agent of amoebic gill disease (AGD); a disease that has a debilitating impact on fish's health and welfare. Successful implementation of genomic selection (GS) for AGD can potentially increase selection response and help reduce outbreaks in the commercial farming of Atlantic salmon. However, successful implementation of GS requires the existence of linkage disequilibrium (LD) between markers and quantitative trait loci (QTL). In this study, we evaluated separately the extent of LD present in six Atlantic salmon breeding populations from Mowi. We also investigated the benefit of using genomic information for selection in these populations, comprising 4 year-classes from Mowi's Norwegian population and 2 year-classes from Mowi's Irish population that was recently introgressed into the Norwegian population. The average distance between markers was 43 kb and the average LD (measured by r2) between adjacent markers was approximately 0.3 for each population. As expected, LD decreased as the physical distance between markers increased. In addition, we observed long-range LD (LD extending to several megabases) across all chromosomes and for all the populations studied. Both the heritability and the accuracy of the breeding value estimates for AGD resistance varied considerably among populations, ranging between 0.06 and 0.24, and 0.32 to 0.77, respectively. The GS models studied had overall better performance than the pedigree based best linear unbiased prediction (PBLUP) model with respect to the accuracy of breeding values prediction, whereas no significant difference was found between the linear and nonlinear GS models. We recommend the use of genomic best linear unbiased prediction (GBLUP) model for the genetic evaluation of AGD resistance due to the higher computing requirements of nonlinear GS models.