A multilocus association analysis method integrating phenotype and expression data reveals multiple novel associations to flowering time variation in wild‐collected Arabidopsis thaliana

Abstract

The adaptation to a new habitat often results in a confounding between genomewide genotype and beneficial alleles. When the confounding is strong, or the allelic effects is weak, it is a major statistical challenge to detect the adaptive polymorphisms. We describe a novel approach to dissect polygenic traits in natural populations. First, candidate adaptive loci are identified by screening for loci directly associated with the adaptive trait or the expression of genes known to affect it. Then, a multilocus genetic architecture is inferred using a backward elimination association analysis across all candidate loci with an adaptive false discovery rate-based threshold. Effects of population stratification are controlled by accounting for genomic kinship in both steps of the analysis and also by simultaneously testing all candidate loci in the multilocus model. We illustrate the method by exploring the polygenic basis of an important adaptive trait, flowering time in Arabidopsis thaliana, using public data from the 1,001 genomes project. We revealed associations between 33 (29) loci and flowering time at 10 (16)°C in this collection of natural accessions, where standard genomewide association analysis methods detected five (3) loci. The 33 (29) loci explained approximately 55.1 (48.7)% of the total phenotypic variance of the respective traits. Our work illustrates how the genetic basis of highly polygenic adaptive traits in natural populations can be explored in much greater detail using new multilocus mapping approaches taking advantage of prior biological information, genome and transcriptome data.

Polygenic adaptation: From sweeps to subtle frequency shifts

“…A homogeneous pattern of subtle frequency shifts at many loci is only observed for high mutation rates. This contrasts with experience gained from breeding and modern findings from genome-wide association studies, which are strongly suggestive of an important role for small shifts with contributions from very many loci (reviewed in [1, 15, 47–49], see [12, 50, 51] for recent empirical examples). For traits such as human height, there has even been a case made for omnigenic adaptation [8], setting up a “mechanistic narrative” for Fisher’s (conceptual) infinitesimal model.…”

Genome-wide Association Studies of Agronomic Traits Consisting of Field- and Molecular-based Phenotypes

“…Candidate gene-based association analysis revealed that major genes controlling the flowering time, FLOWERING LOCUS C (FLC), FRIGIDA (FRI), VERNALIZATION-INSENSITIVE 3 (VIN3) and CRYPTOCHROME 2 (CRY2), are associated with natural variation (Shindo et al, 2005;Balasubramanian et al, 2006;Wollenberg and Amasino, 2012). After developing a number of accessions with the SNP information, the genome-wide association of genes regulating flowering time was identified by GWAS (Atwell et al, 2010;Li et al, 2014;Sasaki et al, 2015;The 1001Genomes Consortium, 2016Zan and Carlborg, 2018). GWAS on flowering time have been conducted in various crops other than Arabidopsis, leading to the discovery of new loci associated with flowering time (maize: Bouchet et al, 2013, rice: Huang et al, 2012, barley: Muñoz-Amatriaín et al, 2014, soybean: Zhang et al, 2015b, rapeseed: Xu et al, 2015, common bean: Moghaddam et al, 2016.…”

Wild GWAS—association mapping in natural populations

“…()), it appears that loci of large effect size in general contribute relatively little to variation, as their allele frequencies tend to be low (Flint & Mackay, ; Yang et al., ). Zan and Carlborg () present a framework for dissecting the genomic basis of a polygenic trait by integrating information from previously identified candidate loci, for example, loci significantly associated with the phenotype or with gene expression changes correlated with the phenotype. Candidate loci are then assessed for inclusion in a multilocus model using backward elimination with a relatively lenient false discovery rate threshold, given that all such loci have already been significantly associated with the trait.…”

“…Candidate loci are then assessed for inclusion in a multilocus model using backward elimination with a relatively lenient false discovery rate threshold, given that all such loci have already been significantly associated with the trait. Zan and Carlborg () demonstrate the application of the new approach to understand the polygenic basis of flowering time variation in wild‐collected A. thaliana . Their analysis demonstrates the power of integrating previous evidence—for example, from previous association scans—into a dissection of the number and effect size distribution of loci contributing to a polygenic trait.…”

“…Further, the increasing focus on models that jointly infer the contribution of small‐effect loci to the phenotype, rather than SNP‐by‐SNP association scans, appears to offer more robust inference of genetic architectures. A contribution in this issue also makes it clear that chromosome partitioning, which has been a widely used approach to test for polygenicity, is likely to lack power to infer a polygenic basis if sample sizes and/or heritabilities are low, which is common in samples from wild populations (Kemppainen & Husby, ). Loci of large effect: Given that some loci do have significant and substantial effects on the trait of interest, it is clear that a more powerful approach to detect regions of association is to fit these known loci as fixed effects and rescan the genome (see Zan & Carlborg, ). This may improve power to detect independent signals of associated loci in close proximity to an identified large‐effect locus.…”