Abstract only

Traditional marker-assisted selection (MAS) strategies relying on QTL detection have not greatly accelerated improvement of the highly polygenic quantitative traits such as yield. Genomic selection (GS) is a new approach that couples the phenotypic data generated routinely by plant breeders and high density marker information. It relies on recently developed statistical approaches that use the marker information as a whole to improve prediction of phenotypic performance, rather than focusing on detecting the most significant genetic effects like in QTL mapping. CIMMYT has tested diverse GS models on grain yield, rust and heading date data generated on 306 elite lines. Grain yield data from two environments in 2009 and five in 2010 in C.E.N.E.B (Campo Experimental Norman Ernest Borlaug, Cd Obregón, Sonora, North of Mexico), stem rust Ug99 notes from three seasons of scoring in Kenya, and heading date data from five environments scored in 2009 and 2010 in Mexico were included in the models. Lines were genotyped with 1667 DArT polymorphic markers. Correlations between the predictive and the observed values were computed using a 10-fold cross-validation scheme that predicts 10% of the missing lines. Models tested were the standard infinitesimal model relying on pedigree information (P), the Bayesian lasso (BL) and the Bayesian reproducing Kernel Hilbert Spaces (RKHS) models with markers alone, and the two latter combined with pedigree information (PBL and PRKHS, respectively). Considering averages of all environments, the best correlations were obtained with the PBL models, with values of 0.78, 0.71 and 0.85 for heading date, stem rust and yield respectively. Correlations using only marker information were also high for the three traits, but slightly lower than when marker and pedigree information were combined in the model. Considering average of all environments, predictions with BL were 0.76, 0.67 and 0.82 for heading date, stem rust and yield. Examining individual environments, correlation increases with heritability in the case of grain yield. This underscores the importance of good quality data for the implementation of GS in plant breeding programs. These preliminary results are promising, especially considering the relatively low number of lines and markers in this data set. Next step will be the application of GS as early as F2 in wheat breeding schemes, selecting and intercrossing F2 plants purely on the basis of genetic information. This should accelerate the accumulation of favorable alleles and therefore increase genetic gains per time.

Genetic Resources Program|Global Wheat Program

English

Lucia Segura

INT2692|CCJL01

CIMMYT Staff Publications Collection