Modelling β-glucan Content in Western Canadian Elite Oat Lines Using Environmental and Genomic Variables

This study attempted to explain the variation in oat β-glucan content in a set of historical Western Canadian elite oat lines using environmental variables, SNP markers and their interactions. In the first part of the study, the effect of environmental variables was modelled using correlation analysis on 3 check lines over 45-62 site-year combinations (environments). Only relatively highly correlated environmental variables (r > 0.25) were included in the model. This environment model gave prediction accuracy between 0.47-0.57 among three checks. In general, hot temperatures in June, July and August result in higher β-glucan content. In the second part of the study, phenotypic data of β-glucan content in 306 lines, along with 2,584 SNP markers obtained from the Oat iSelect 6K SNP chip were used to calculate marker effects in a SNP-BLUP analysis using the Ridge Regression-Best Linear Unbiased Prediction model. Use of all markers (2,584) gave the highest prediction accuracy (0.73) after cross validation in comparison to the most significant 1,000 SNPs, 500 SNPs, 100 SNPs, or 20 SNPs. Including the 3 environmental variables along with all markers increased the prediction accuracy to 0.84. In the final part of the study, 62 environments were clustered into 2 mega-environments (warm vs cool) based on the 3 environmental variables in order to create a balanced dataset so that marker by environment interactions could be evaluated using a  Bayesian Generalized Linear Regression model. Including marker-by-environment interactions as a factor in the model did not improve accuracy in comparison to the among-environment model. Both models provided higher prediction accuracy than the single-environment model.