Mixed model methodology for the identification of genetical factors underlying trait variation in plants

The advent of DNA-marker technology has made the detection of quantitative trait loci (QTLs) a routine activity in plant breeding. While standard procedures are available for QTL mapping, more flexible strategies are required to analyse the complex data typically produced by plant breeding programs. This thesis presents a general and flexible mixed model QTL mapping approach. The philosophy is to include genotypic information derived from molecular markers as explanatory variables to model complex phenotypic responses, while exploiting the flexibility of mixed models to account for complex variance-covariance structures in the data. Attractive generalisations of our QTL models incorporate explicit information on genotypes and environments, for example, as obtained from crop growth models, thereby opening the way to eco-physiological QTL models.The application of the methodology to cases commonly found in plant breeding is illustrated throughout the various chapters of the thesis. In chapter 2, a QTL model is presented for probably the most typical experimental set up in plant breeding, the multi-environment trial. The QTL model describes the genetic basis of genotype by environment interaction for a single trait and tests for environment-specific QTL effects (QTL by environment interaction). After detection of QTLs underlying genotype by environment interaction, the model is extended to make QTL expression dependent on environmental variables. In a reanalysis of yield data stemming from the North American Barley Genome Mapping Project (NABGMP), a QTL for yield was detected on barley chromosome 2H whose effect was proportional to the temperature range at heading time. In chapter 3, the single trait multi-environment model is elaborated to the multi-trait multi-environment situation. We show with another reanalysis of data from the NABGMP how questions related to pleiotropy and genetic linkage as causing genetic correlations between traits can be addressed. Both pleiotropic and linked QTLs were found to explain genetic correlations between heading date and yield. In chapter 4, the linear mixed model is generalised to a non-linear mixed model to describe parameters of a growth curve as a function of QTL effects. The approach is illustrated by an example on potato leaf senescence. QTLs were identified that affect growth trajectories in different ways, thereby contributing to a better understanding of the genetic control of complex traits over time. In chapter 5, we show that not only designed bi-parental cross populations can be naturally handled by mixed model QTL formulations, but also non-designed populations within association mapping approaches. With another example in potato, we analyse historical disease resistance data produced over 25 years of potato variety trial testing in combination with targeted molecular marker techniques to detect interesting markers for breeders. Pedigree information was used to improve the modelling of genetic variances and correlations between genotypes. An association mapping approach using mixed models incorporating pedigree information performed better than commonly used association mapping strategies. Significant associations were detected that proved to be consistent when tested in a confirmatory data set. Finally, chapter 6 consists in a discussion of the mixed model approach as a general framework for QTL mapping in plant breeding.