QTL mapping in outbred tetraploid (and diploid) diallel populations

Abstract

Over the last decade, multiparental populations have become a mainstay of genetics research in diploid species. Our goal was to extend this paradigm to autotetraploids by developing software for quantitative trait locus (QTL) mapping in connected F1 populations derived from a set of shared parents. For QTL discovery, phenotypes are regressed on the dosage of parental haplotypes to estimate additive effects. Statistical properties of the model were explored by simulating half-diallel diploid and tetraploid populations with different population sizes and numbers of parents. Across scenarios, the number of progeny per parental haplotype (pph) largely determined the statistical power for QTL detection and accuracy of the estimated haplotype effects. Multiallelic QTL with heritability 0.2 were detected with 90% probability at 25 pph and genome-wide significance level 0.05, and the additive haplotype effects were estimated with over 90% accuracy. Following QTL discovery, the software enables a comparison of models with multiple QTL and nonadditive effects. To illustrate, we analyzed potato tuber shape in a half-diallel population with three tetraploid parents. A well-known QTL on chromosome 10 was detected, for which the inclusion of digenic dominance lowered the Deviance Information Criterion (DIC) by 17 points compared to the additive model. The final model also contained a minor QTL on chromosome 1, but higher-order dominance and epistatic effects were excluded based on the DIC. In terms of practical impacts, the software is already being used to select offspring based on the effect and dosage of particular haplotypes in breeding programs.

Keywords: MPP; Multiparental Populations; dominance; haplotypes; multiparental; polyploidy.

Figure 1

Examples of diallel populations with four parents (labeled a, b, c, d); each square represents a family of full-sibs. (A) full diallel; (B) half diallel with selfing; (C) half diallel without selfing; (D) circular; (E) linear; (F) factorial; (G) testcross or nested design; and (H) arbitrary partial diallel. Figure adapted from Verhoeven et al. (2006).

Figure 2

Threshold to control the genome-wide Type 1 error rate at $\alpha =$ 0.05, for a half-diallel design without selfing. The statistic –$\mathrm{\Delta }$DIC is the Deviance Information Criterion for the null hypothesis (no QTL) relative to the alternative hypothesis of an additive QTL. Each point is based on 1000 simulations, and the best-fit linear regression is shown as a solid line.

Figure 3

Statistical power to detect a multiallelic QTL at significance level $\alpha$ = 0.05, as a function of the number of progeny per parental haplotype, number of parents, ploidy, and QTL heritability (h²). Each point is the average of 1000 simulations with an additive model. The dashed line is a monotone increasing, concave spline for h² = 0.1, and the solid line is the spline for h² = 0.2.

Figure 4

Accuracy of the estimated haplotype effects, as measured by Pearson's correlation with the simulated values. Each point is the average of 1000 simulations with an additive model. The dashed line is a monotone increasing, concave spline for h² = 0.1, and the solid line is the spline for h² = 0.2.

Figure 5

R/diaQTL results for potato tuber shape using a half-diallel population with three parents: VillettaRose, W6511-1R, and W9914-1R. (A) Genome scan with scan1. The dashed horizontal lines correspond to $\alpha$ = 0.05 (gold) and $\alpha$ = 0.1 (red). The most significant marker was solcap_snp_c2_25522 on potato chromosome 10. (B) Additive effect estimates for the 12 parental haplotypes using fitQTL. Error bars are the 90% CI.