Lessons from Dwarf8 on the Strengths and Weaknesses of Structured Association Mapping

Abstract

The strengths of association mapping lie in its resolution and allelic richness, but spurious associations arising from historical relationships and selection patterns need to be accounted for in statistical analyses. Here we reanalyze one of the first generation structured association mapping studies of the Dwarf8 (d8) locus with flowering time in maize using the full range of new mapping populations, statistical approaches, and haplotype maps. Because this trait was highly correlated with population structure, we found that basic structured association methods overestimate phenotypic effects in the region, while mixed model approaches perform substantially better. Combined with analysis of the maize nested association mapping population (a multi-family crossing design), it is concluded that most, if not all, of the QTL effects at the general location of the d8 locus are from rare extended haplotypes that include other linked QTLs and that d8 is unlikely to be involved in controlling flowering time in maize. Previous independent studies have shown evidence for selection at the d8 locus. Based on the evidence of population bottleneck, selection patterns, and haplotype structure observed in the region, we suggest that multiple traits may be strongly correlated with population structure and that selection on these traits has influenced segregation patterns in the region. Overall, this study provides insight into how modern association and linkage mapping, combined with haplotype analysis, can produce results that are more robust.

Recovering Power in Association Mapping Panels with Variable Levels of Linkage Disequilibrium

“…These LD and F st features led to the observed V-shape analytical power curve along the chromosome, particularly in the CF-Dent panel in which LD was more extended (Figure 1 and Figure 3). This is in good agreement with published manhattan plots of GWAS results, which showed a reduced number of low P-values in the centromeric regions (Bouchet et al 2013;Larsson et al 2013). In our three panels, we observed that this problem also arose with other classical estimators of relatedness (results not shown) such as the IBS estimator or the first estimator provided in Vanraden (2008, p. 4416) As MAF, F st , LD extent, and consequently CorK_Freq were different in the three panels (Table 1), average power was highly variable among the three panels (adjusted for the same population size).…”

“…For example F st and CorK_Freq had a stronger effect on power for markers with higher MAF, and MAF had a stronger effect on power for less differentiated markers. These results show that controlling false positives using the K_Freq model also implies reducing power at differentiated markers (Larsson et al 2013). It is interesting to note that no marker had a CorK_Freq ,0.03 (Figure 2).…”

“…It was used in many association mapping studies and permitted the detection of QTL in humans, animals, and plants (K. Huang et al 2010;Kang et al 2010;Price et al 2010;Zhang et al 2010;Bouchet et al 2013;Romay et al 2013). However, one of the main drawbacks of these structure and relatedness corrections is that it also reduces the number of detectable true positives, particularly if the trait is correlated to the population structure (Larsson et al 2013). Also, including the tested SNP in the computation of K is expected to decrease power at this SNP (Listgarten et al 2012).…”

Genetic Determinants for Enzymatic Digestion of Lignocellulosic Biomass Are Independent of Those for Lignin Abundance in a Maize Recombinant Inbred Population

“…However, GWAS can be compromised when a great many gene products contribute to a complex trait and when kinship relationships are not taken into account. In maize, GWAS was used to find a strong correlation between the Dwarf8 locus and flowering time using a general linear model (Thornsberry et al, 2001), but the association was found to be spurious by association with population structure (Larsson et al, 2013). Nevertheless, models that correct for kinship have allowed the discovery of genes controlling many agronomic traits in rice (Oryza sativa; Huang et al, 2010) and sorghum (Morris et al, 2013).…”

A GWAS platform built on iPlant cyber-infrastructure

“…This is also one of the rationales why we have constructed an automated workflow for simultaneously testing different models. It has also been suggested that the combination of traditional linkage analysis [32] and further haplotype analysis [33] can increase the power of GWA mapping to distinguish true from false associations.…”