Genomewide selection with biallelic versus triallelic models in three-way maize populations

While single nucleotide polymorphism (SNP) markers are typically biallelic, quantitative trait loci (QTL) may have three alleles per locus in three-way populations. Our objective in this study was to determine if multiallelic markers or haplotypes improve the prediction accuracy of genomewide selection in three-way breeding populations. Simulated and empirical maize (Zea mays L.) doubled haploid populations were used to compare a biallelic model, marker interval model (which used adjacent markers to create haplotypes), and allele phasing model (which inferred triallelic markers from parental SNP data). The simulation experiments differed in the number of QTL (10, 40, or 100), heritability (0.30, 0.50, or 0.80), and sizes of allelic effects. Four empirical three-way populations were pheno-typed at four to seven locations between 2012 and 2015 and were genotyped with 356 to 960 polymorphic SNP markers. Genomewide marker effects were obtained by ridge regression-best linear unbiased prediction. In the simulation experiments, differences in prediction accuracy were <0.01 among the biallelic, marker interval, and allele phasing models. For grain yield, moisture, and test weight in the four maize populations, the differences in predictive ability among the three models were nonsignificant (P = 0.05). Further simulations showed that the small or nonsignificant differences in prediction accuracy were caused by large linkage blocks found among inbreds, particularly doubled haploids. Overall, we recommend the marker interval model in three-way populations because of its simplicity, similar prediction accuracy, and theoretical advantage over the two other models.