Repository logo

A Canadian oat genomic selection study incorporating genetic and environmental information



Journal Title

Journal ISSN

Volume Title





Degree Level



Oat (Avena sativa L.) is an important crop in Canada that has been seeded on an average of 3.3 million acres over the past five years. It is considered a healthy cereal due to the presence of beta-glucan in the grain, which been shown to reduce the risk of heart disease, as well as being a good source of protein that is rich in globulins. Identifying new breeding strategies that can improve breeding efficiency in oat is important for future progress in this crop. To this end, genomic and environmental factors, along with their interactions, were examined to determine what contributed to variation in important oat traits. This information was then used to develop genomic selection (GS) models that can be used in oat breeding programs. In the first study, 305 elite oat breeding lines grown in the Western Cooperative Oat Registration Trial (WCORT) from 2002 to 2014 were used to investigate important factors for genomic selection model building. The influence of phenotypic data, genotyping platforms, statistical model, marker density, population structure, training population size and trait heritability were assessed. It was determined that the machine learning model Support Vector Machine and the additive linear model rr-BLUP offered the best overall prediction accuracies. Prediction accuracy increased when using the iSelect Oat 6K SNP chip, as the marker number increased, with larger training population size and with traits that were more heritable. In the second study, environmental and correlated agronomic variables, along with their inter-relationships, that contributed to variation in yield and grain β-glucan content in oat lines was investigated. A hypothesized structural equation model (SEM) that included variables related to environmental and phenotypic traits was created and tested against observed yield data. Significant paths were identified to explain yield variation (59%-76%) among the three oat varieties. A similar approach was taken for β-glucan in which significant paths were found which explained 16%-41% of the variation in β-glucan. Results from this study suggest that a longer period to heading and maturity, and a taller stature were the three phenotypic traits that most positively influence yield. Limited precipitation before maturity, high temperatures during heading and grain filling were the three environmental variables that contributed to decreased yield. Precipitation and July temperature were the two most important environmental variables that influenced β-glucan, while maturity was the most important trait affecting β-glucan, although the direction of effect for maturity varied by oat variety. In the third study, additional information was added into the previous GS models to determine if prediction could be improved. Genotype, environment and their interaction were used to conduct genomic selection for yield. Four mega-environments were identified from Ward’s hierarchical clustering using the significant environmental variables identified in the second study. It was found that using individual locations to represent environment provided more accuracy compared to using mega-environments. The reaction norm model was also tested which allowed significant environmental variables to be incorporated as a covariance matrix in the model. Including an environmental covariance matrix and interaction terms increased prediction accuracy compared to models with only genotype main effects. Multiple trait GS did not provide better prediction accuracy for most the traits. In the final study, GS was used to predict the GEBVs of two populations, a biparental derived population and a population consisting of elite breeding lines from several different breeding programs. Higher predication accuracy was found in the elite breeding line population which was likely due to the closer genetic relationship between it and the training population. Finally, random selection and genomic selection were compared in the two populations. Genomic selection out-performed random selection in the elite breeding population, but not in the bi-parental population. Again, the poor performance of GS in the bi-parental population was best explained by the unrelatedness between it and the training population. Taken together, these studies provided deeper insight into how GS could be applied in oat breeding programs.



Genomic selection, structural equation model, mega-environments, correlated traits, multiple trait genomic selection, empirical testing



Doctor of Philosophy (Ph.D.)


Plant Sciences


Plant Sciences


Part Of