Genomic selection models double the accuracy of predicted breeding values for bacterial cold water disease resistance compared to a traditional pedigree-based model in rainbow trout aquaculture

Authors: Vallejo, Roger; Leeds, Timothy - Tim; Gao, Guangtu; PARSONS, JAMES - Troutlodge, Inc; MARTIN, KYLE - Troutlodge, Inc; Evenhuis, Jason; FRAGOMENI, BRENO - University Of Georgia; Wiens, Gregory - Greg; Palti, Yniv

Submitted to: Genetics Selection Evolution
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/25/2017
Publication Date: 2/1/2017
Publication URL: https://handle.nal.usda.gov/10113/5695429
Citation: Vallejo, R.L., Leeds, T.D., Gao, G., Parsons, J.E., Martin, K.E., Evenhuis, J., Fragomeni, B.O., Wiens, G.D., Palti, Y. 2017. Genomic selection models double the accuracy of predicted breeding values for bacterial cold water disease resistance compared to a traditional pedigree-based model in rainbow trout aquaculture. Genetics Selection Evolution. 49(17):1-33. doi: 10.1186/s12711-017-0293-6.

Interpretive Summary: Using genome-based estimated breeding values for selective breeding for traits that cannot be measured directly in the potential breeders, like disease resistance, holds great promise as it provides individual genetic merit estimates for potential breeders compared to family-average estimates in traditional selective breeding. Previously we have shown that bacterial cold water disease (BCWD) resistance in rainbow trout can be improved using traditional family-based selection, but progress has been limited to exploiting only between-family genetic variation. Therefore, in the current study we used genomic selection (GS) models which effectively exploits within-family genetic variation. We evaluated three GS models to predict genomic-enabled breeding values (GEBVs) for BCWD resistance in a commercial rainbow trout population, and compared the accuracy of the predicted GEBVs to estimated breeding values (EBVs) which were calculated with a traditional pedigree-based model. Using progeny performance data, we have shown that the accuracy of GEBVs from all tested GS models were substantially higher than the traditional pedigree-based EBVs. Overall, we found that using a much smaller sample size than similar studies with terrestrial agricultural animals, genome selection can substantially improve the genetic gains in traits that cannot be measured directly on the potential breeders in rainbow trout aquaculture. Therefore, using genomic selection will reduce the number of generations, time, labor and number of fish that are currently needed for achieving the same level of performance improvement. This approach will have positive impacts on farm productivity, animal welfare, the environment through potentially reducing the use of antibiotics in rainbow trout farming, and the overall sustainability of aquaculture production systems.

Technical Abstract: Previously we have shown that bacterial cold water disease (BCWD) resistance in rainbow trout can be improved using traditional family-based selection, but progress has been limited to exploiting only between-family genetic variation. Genomic selection (GS) is a new alternative enabling exploitation of within-family genetic variation. We compared three GS models (single-step GBLUP (ssGBLUP), weighted ssGBLUP (wssGBLUP), and BayesB) to predict genomic-enabled breeding values (GEBVs) for BCWD resistance in a commercial rainbow trout population, and compared the predictive ability (PA) of GEBVs to traditional breeding values (EBVs) estimated with a pedigree-based BLUP model. We also assessed the impact of sampling design on the accuracy of GEBV predictions. For these comparisons, we used BCWD survival phenotypes recorded on 7893 training fish from 102 families, from which 1473 fish from 50 families had genotypes (57K SNP array). Naïve siblings of the training fish (n = 930 testing fish) were genotyped to predict their GEBVs, and mated to produce 138 progeny testing families. In the following generation, 9968 progeny were phenotyped to empirically assess the accuracy of GEBV predictions made on their non-phenotyped parents. The accuracy of GEBVs from all tested GS models were substantially higher than the BLUP model EBVs. The highest increase in accuracy relative to the BLUP model was achieved with BayesB (97.2 -108.8%), followed by wssGBLUP at iteration 2 (94.4 -97.1%) and 3 (88.9 -91.2%) and ssGBLUP (83.3 -85.3%). Reducing the training sample size to n = ~1000 had no negative impact on the PA (0.67-0.72), but with n = ~500 the PA dropped to 0.53 -0.61 if the training and testing fish were full-sibs, and even substantially lower to 0.22 -0.25 when they were not full-sibs. Using progeny performance data, we have shown that the accuracy of genomic predictions with GS models is substantially higher than the traditional pedigree-based BLUP model for BCWD resistance in rainbow trout. Overall, we found that using a much smaller training sample size compared to similar studies in livestock, GS can substantially improve the selection accuracy and genetic gains in traits that cannot be measured directly on the potential breeders.