Genome-wide association and genomic selection for oil and fatty acid profile in rapeseed (Brassica napus L.)
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Overarching goals of rapeseed (Brassica napus L.) breeding efforts include the improvement of yield- and seed-quality-related traits to meet the evolving demands of a growing population. Genome-wide association studies (GWAS) and genomic selection (GS) are methods that provide the potential to improve selection efficiency, facilitating responses to agronomic and quality challenges while enhancing the sustainability of plant breeding programs. We applied these biotechnologies to B. napus L. in pursuance of three main objectives: 1) GWAS to identify quantitative trait loci (QTL) for five seed quality traits (overall oil content, erucic, oleic, linoleic, and linolenic acids); 2) evaluating GS prediction accuracy for hybrid seed quality traits; and 3) evaluating the GWAS-guided GS method proposed to improve GS prediction accuracy. We analyzed 454 B. napus L. genotypes (92 parents, 362 hybrids) across 48 site-years in the three Canadian prairie provinces. FarmCPU GWAS analyses identified 89 peak QTL, including 14 QTL for oil content. Several QTL coincide with candidate genes identified in previous studies, while novel QTL warrant further candidate gene investigation. GS prediction accuracies were compared across 135 unique GS analyses for each trait, evaluating responses to factors including model choice (nine regression models), population (five training/validation population designs), and marker density (three marker sets containing low, intermediate, and high densities). Prediction accuracies (represented by Pearson’s correlation coefficient (r) for the relationship between predicted and actual phenotypes) ranged from negative values (oil content) to 0.89 (linoleic acid content); however all five traits could be predicted with r > 0.70 depending on the combination of aforementioned factors. Prediction accuracies exhibited negative responses to increasing trait complexity, positive responses to increasing training population size and degree of training/validation population relatedness, and no significant differences among marker densities or parametric models. Machine learning models performed either equivalent to or worse than common parametric models. GWAS-guided GS exhibited slight numeric improvements relative to conventional GS accuracies for the same traits. Although improvements were not statistically significant, the consistency of extremely low-density marker sets is conducive to reducing genotyping density while maintaining or improving genetic gains. The promising accuracy of GS techniques in this study supports their potential implementation in future Brassica breeding programs.