Imputation accuracy from a low density SNP panel in 5 dairy sheep breeds in France

In France, genomic selection (GS) has been implemented in dairy sheep in Lacaune breed (LA) and more recently in Pyrenean breeds: Red-Faced Manech (RFM), Black-Faced Manech (BFM) and Basco-Béarnaise (BB). Genomic estimated breeding values (GEBV) are calculated on the basis of SNP50 Bead-Chip (50K) genotypes. Compared to dairy cattle, the relatively higher cost of genotyping limits the cost-effectiveness of GS in dairy sheep. However, a significant reduction of genotyping cost is forecast with the design of a low density (LD) chip by the International Sheep Genomics Consortium. Use of this chip requires harmonizing genotypes on LD panel and those existing on 50K panel by imputing missing SNPs. The aim of the study was to evaluate quality of imputation in 5 French dairy sheep breeds (with Corsica breed: CO) and its impact on GEBV. Genotypes of rams available in each breed (5,792 in LA, 1,741 in RFM, 470 in BFM, 556 in BB and 609 in CO) were split in two sets: a training set with oldest rams and a validation set with youngest rams (1,144 in LA, 430 in RFM, 80 in BFM, 140 in BB and 117 in CO). The 50K genotypes of validation set, after quality control, were ‘pierced’ in order to mimic LD genotypes. Out of 16,331 SNPs from LD panel, 9,822 were located on 50K panel. These genotypes were then imputed to the 50K panel, from the training population, using the FImpute software. Comparison of 50K imputed and true genotypes permitted to assess imputation accuracy. Concordance rates (CR: percentage of alleles properly imputed) per animal or per SNP were high, ranging from 96.6% to 99.1%, depending on breed. The squared correlations (r2) between true and imputed genotypes ranged from 81.3% to 95%. On the 3 criteria, Lacaune breed had the higher scores followed by RFM, BB, BFM and finally CO breeds, in accordance with training population size. For validation rams, correlations between GEBV computed with true or imputed genotypes exceeded 0.99 for production and type traits whatever the breed. Consequently, the 4 breeds in GS have decided to use LD genotypes for pre-selection of young rams