Genomic Selection, Quantitative Trait Loci and Genome-Wide Association Mapping for Spring Bread Wheat (Triticum aestivum L.) Improvement

Molecular breeding involves the use of molecular markers to identify and characterize genes that control quantitative traits. Two of the most commonly used methods to dissect complex traits in plants are linkage analysis and association mapping. These methods are used to identify markers associated with quantitative trait loci (QTL) that underlie trait variation, which are used for marker assisted selection (MAS). Marker assisted selection has been successful to improve traits controlled by moderate to large effect QTL; however, it has limited application for traits controlled by many QTL with small effects. Genomic selection (GS) is suggested to overcome the limitation of MAS and improve genetic gain of quantitative traits. GS is a type of MAS that estimates the effects of genome-wide markers to calculate genomic estimated breeding values (GEBVs) for individuals without phenotypic records. In recent years, GS is gaining momentum in crop breeding programs but there is limited empirical evidence for practical application. The objectives of this study were to: i) evaluate the performance of various statistical approaches and models to predict agronomic and end-use quality traits using empirical data in spring bread wheat, ii) determine the effects of training population (TP) size, marker density, and population structure on genomic prediction accuracy, iii) examine GS prediction accuracy when modelling genotype-by-environment interaction (G × E) using different approaches, iv) detect marker-trait associations for agronomic and end-use quality traits in spring bread wheat, v) evaluate the effects of TP composition, cross-validation technique, and genetic relationship between the TP and SC on GS accuracy, and vi) compare genomic and phenotypic prediction accuracy. Six studies were conducted to meet these objectives using two populations of 231 and 304 spring bread wheat lines that were genotyped with the wheat 90K SNP array and phenotyped for nine agronomic and end-use quality traits. The main finding across these studies is that GS can accurately predict GEBVs for wheat traits and can be used to make predictions in different environments; thus, GS should be applied in wheat breeding programs. Each study provides specific insights into some of the advantages and limitations of different GS approaches, and gives recommendations for the application of GS in future breeding programs. Specific recommendations include using the GS model BayesB (especially for large effect QTL) for genomic prediction in a single environment, across-year genomic prediction using the reaction norm model, using a large TP size for making accurate genomic predictions, and not making across-population genomic predictions except for highly related populations