Genomic Selection for Disease and Morphological Traits in Diverse Wheat Landraces

Genomic selection, where selection decisions are based on estimated breeding values derived from genome-wide DNA markers, is set to accelerate genetic progress in plant breeding.  We assessed the accuracy of genomic breeding values (GEBVs) in bread wheat using genotype and phenotype data for 250 phenotypically diverse landrace accessions from 32 countries (Watkins collection).  Single observations (S) were available for several morphological traits (awns presence-absence, seedling chlorosis) and trait-linked PCR markers for Lr34 and Sr2.  Repeated observations (R) were recorded for field-based leaf rust (Lr), stem rust (Sr) and stripe rust (Yr) disease resistance.  All lines were genotyped using an Illumina Infinium 9K wheat single nucleotide polymorphism (SNP) chip and quality control reduced the number of SNPs and accessions to 5568 and 222, respectively.  Principle component analysis and genomic relatedness revealed moderate to high genotype relationships within and across countries. Genomic Best Linear Unbiased Prediction (GBLUP) and a Bayesian Regression Method (BayesR) were used and the accuracy of the GEBVs was calculated as their correlation with phenotypes adjusted for fixed effects.   For S and R observations, the accuracy ranged from 0.31 to 0.87 and from 0.11 to 0.54, respectively.   In Lr and Sr, BayesR showed a small advantage over GBLUP (<5%), while for Yr there was no difference.  Two cross-validation methods were used (leave-1-out and random-5-fold) and produced similar accuracies.  There was a strong trend for higher accuracy with closer relatedness between validation and reference sets.  The accuracies achieved confirm the feasibility of genomic selection in wheat.