Genome-Wide Association Mapping Using Single-Step GBLUP

The purpose of this study was to extend single-step GBLUP (ssGBLUP) to genome-wide association analysis (GWAS). ssGBLUP is a procedure that calculates GEBVs based on combined pedigree, genomic and phenotypic information. In this study, GEBVs were converted to marker (SNP) effects. Unequal variances for markers were incorporated by deriving weights from SNP solutions and including the calculated weights into a new genomic relationship matrix.  Improvements on the SNP weights were obtained iteratively either by recomputing the SNP effects only or also by recomputing the GEBV.  Efficiency of the method was examined using simulations for 10 replications with 15,800 subjects across 6 generations, of which 1500 were genotyped with 3000 SNP markers evenly distributed on 2 chromosomes. Heritability was assumed 0.5 all due to 30 QTLs. Comparisons included accuracy of breeding values and cluster of SNP effects of ssGBLUP and BayesB with several options for each procedure. For genomic evaluation, an accuracy of prediction of 0.89 (0.01) was obtained by ssGBLUP after only one iteration, which was slightly higher than BayesB of 0.88 (0.02), but required only a small fraction of time. Power and precision for GWAS applications was evaluated by correlation between true QTL effects and the sum of m adjacent SNP effects, where m varied from 1 to 40. The highest correlations were achieved with m=8, and were 0.82 (0.02) for ssGBLUP; and 0.83 (0.07) for BayesB with m=16 according to marker density and extent of linkage disequilibrium in simulated population. Computing time for ssGBLUP took about 2 min, while BayesB took about 5 hrs. GWAS by ssGBLUP was applied to 6-week body weight of 245,000 chicken, of which about 4500 were genotyped. Comparisons also included the classical GWAS (CGWAS), where each SNP was sequentially fit as a fixed effect. The Manhattan plots of CGWAS and ssGBLUP were very similar, while the plot by BayesB was dominated by a single region. ssGBLUP with marker weights is fast, accurate and easy to implement for GWAS applications. In particular, ssGBLUP is applicable to GWAS with complex models including multitrait, maternal and random regression.