PREDICTION OF DH LINES TESTCROSS VALUE FOR GRAIN YIELD AND HARVEST MOISTURE IN MAIZE: EFFICACY OF QTL MAPPING AND GENOME-WIDE APPROACHES IN MULTI-PARENTAL ELITE POPULATIONS

Abstract Grain Yield (GY) and Grain Moisture content at harvest (GM) are complex quantitative traits and by far the two performance traits of major interest for maize breeder. Prediction accuracy of the testcross performance of untested Double Haploid (DH) maize lines for these two traits is of tremendous importance in order to increase genetic gain. We analysed genomic and phenotypic data of testcross progenies of 1066 DH lines genotyped with 3072 SNP markers and derived from three large half-sib populations phenotyped for GY and GM in eight locations and applied cross-validation method to compare the accuracy of whole genome prediction (GP) and QTL based prediction approaches. GP showed higher accuracy for both traits with mean predictive ability of 0.58 and 0.73 for GY and GM respectively in comparison of 0.15 and 0.30 as mean predictive ability obtained using the QTL-based approach. Both methods showed higher accuracy in predicting GM in comparison to GY. The lack of accuracy for QTL based prediction confirmed the major issues traditionally faced in playing with QTL for polygenic and complex traits. For GP the simultaneous use of the three half-sib population did not increase the accuracy prediction obtained working within the same parental population while the predictive ability dropped when predicting a population with a training set formed by other population (0.39 and 0.58 versus 0.53 and 0.73 for GY and GM respectively) confirming the strong influence of genetic relationship between estimation set and test set on the predictive ability. Our experimental results confirm that Whole Genome prediction accuracy is surpassing QTL prediction accuracy for the two maize-breeding target-traits and raised the concern on the way grain yield and grain moisture QTL had been implemented in breeding program so far. One could be interested in exploring possibility to combine relevant QTL and whole genome prediction together to advance towards performance prediction accuracy increase.