Multi-parent multi-environment QTL analysis: an illustration with the EU-NAM Flint population

Abstract

Key message Multi-parent populations multi-environment QTL experiments data should be analysed jointly to estimate the QTL effect variation within the population and between environments. Abstract Commonly, QTL detection in multi-parent populations (MPPs) data measured in multiple environments (ME) is done by analyzing genotypic values ‘averaged’ across environments. This method ignores the environment-specific QTL (QTLxE) effects. Running separate single environment analyses is a possibility to measure QTLxE effects, but those analyses do not model the genetic covariance due to the use of the same genotype in different environments. In this paper, we propose methods to analyse MPP-ME QTL experiments using simultaneously the data from several environments and modelling the genotypic covariance. Using data from the EU-NAM Flint population, we show that these methods estimate the QTLxE effects and that they can improve the quality of the QTL detection. Those methods also have a larger inference power. For example, they can be extended to integrate environmental indices like temperature or precipitation to better understand the mechanisms behind the QTLxE effects. Therefore, our methodology allows the exploitation of the full MPP-ME data potential: to estimate QTL effect variation (a) within the MPP between sub-populations due to different genetic backgrounds and (b) between environments. Electronic supplementary material The online version of this article (10.1007/s00122-020-03621-0) contains supplementary material, which is available to authorized users.

Modeling Allelic Diversity of Multi-parent Mapping Populations Affects Detection of Quantitative Trait Loci

“…Giraud et al (2014) found that their haplotype model outperformed the founder and SNP models in terms of the number of QTL identified using EU-NAM Flint and Dent maize populations. In contrast, Garin et al (2020) found that in the EU-NAM Flint population, the bi-allelic model detected a larger number of unique QTL, compared to parental or ancestral haplotype models. Bardol et al (2013) found that in two multi-parent dent populations, their bi-allelic model and ancestral haplotype model generally outperformed the parental linkage model, although benefits of these models varied by dataset.…”

“…The fact that the QTL F model outperformed the QTL H and GWAS SNP in our population suggests that the MAGIC population contains a relatively more diverse representation of European and North American flint and dent than populations used in previous studies. It is also possible that the structure of multi-parent populations has an effect on the performance of the three models, compared to previous studies which used nested association mapping (Giraud et al 2014;Garin et al 2020) and factorial populations Bardol et al (2013).…”

Developing Novel Rice Genotypes Harboring Specific QTL Alleles Associated with High Grain Yield under Water Shortage Stress

“…However, yet this approach has not been applied to introgress and map these QTLs in the Egyptian rice genetic background. Marker-assisted introgression of QTLs, with major effects for GY under water-deficit stress, are considered a fast-track approach for breeding high-yielding water shortage tolerant rice genotypes [ 20 ]. Because of these major effects, QTLs should be tested across different genetic backgrounds and across environments to check the consistency of their effect.…”

Flexible multivariate linear mixed models for structured multiple traits

“…In humans it is common to study genotype-phenotype associations in population-based samples for genome-wide association studies (GWAS); in model organisms complex breeding designs are often used. Linear mixed models (LMMs) 3–6 are employed for controlling confounding due to genetic relatedness among individuals and are used to identify genetic loci contributing to quantitative traits of interest (QTL) 7–13 . Multivariate LMMs (MLMMs) further enhance statistical power over univariate LMMs 14 because they can accumulate signals across traits 8, 15 with a common genetic locus.…”

“…to provide stable, fast convergence 5, 14 . The computational complexity is O ( n 3 m 3 ) for EM and O ( n 3 m 7 ) for NR, AI ( m traits, n individuals). This suggests that using GCTA, WOM-BAT, and ASReml is not practical for GWAS with a large number of SNPs and a moderate number of individuals 14 .…”