The Analysis and Use of Genotype by Environment Interactions in Genetic Evaluations of Livestock and Plants

Abstract

<p>This thesis explores methods for estimating genotype by environment (G×E) interactions in livestock and plants. Genotype by environment interactions occur when the genetic architecture of a trait changes depending on the environment it exists in. They are particularly interesting as a source of genetic variation that could be utilised in breeding programs to select genotypes who have genetic merit that is more robust to environmental variation. This thesis aims to estimate genotype by environment interactions in livestock and plant populations using different methods and improve our understanding of how these interactions could be used in breeding programs to increase the robustness of agricultural populations to environmental variation.</p> <p>The first experiment of this thesis investigated genotype by environment interactions in the bodyweight of Australian sheep using reaction norm and multi-trait models in combination with genomic data. It found significant variation in the slope of the reaction norm model that could be used to increase robustness of sheep, and that this variation was highly polygenic. It highlighted that both heterogenous genetic variance (scale-type G×E) and heterogenous genetic correlations (rank-type G×E) contributed to the variation in the reaction norm slope and found it could be important to separate these sources to better understand the genetic variation in robustness. </p> <p>The second experiment of this thesis utilised a multi-environment trial of a Barley population to examine the effectiveness of two methods to partition the different types of G×E interactions when estimating the robustness of genotypes in reaction norm models. It found that genetic regression, which made breeding values for the slope independent of the intercept, was very effective in removing the impact of G×E interactions due to scale and isolating the variability across environments due to heterogenous genetic correlations. This enables the change in genetic architecture of traits across environments to be studied more clearly in reaction norm models. We also showed that factor analytic models, which are an alternative to reaction norms, are better equipped to capture complex G×E interactions because of their flexibility.</p> <p>The third experiment of this thesis examined the use of factor analytic models to capture genotype by environment interactions in a multi-environment trial of sheep. The factor analytic models were able to approximate the unstructured genetic co(variance) matrix between 31 discrete environments using 85% fewer parameters than what would have been required with a multi-trait model. The model enabled the flock-years to be clustered by their similarity and showed that G×E interactions were large both between flocks and across years within flocks. It was unclear whether factor analytic models were preferrable to reaction norms based on the goodness-of-fit tests, and the estimates of heritability, genetic variance and genetic correlations between environments were inconsistent between the models. </p> <p>The final experimental chapter assessed the capacity of reaction norm models to predict the robustness of a sire’s progeny performance across different growth environments. Using data collected in a research flock, the reaction norm models were predictive of the ability of a sire’s progeny to reliably gain weight across different growth environments at an accuracy that was consistent with the power of the data. It then found that these breeding values for robustness were consistent with breeding values estimated using data from the wider industry population recorded by commercial stud breeders. Selection based on reaction norm breeding values could be used to increase the robustness of body weight gain in Australian sheep to variation in growth environments. </p>