Grain weight, an essential yield component, is under strong genetic control and at the same time markedly influenced by the environment. Genetic analysis of the thousand grain weight (TGW) by genome-wide association study (GWAS) was performed with a panel of 358 European winter wheat (Triticum aestivum L.) varieties and 14 spring wheat varieties using phenotypic data of field tests in eight environments. Wide phenotypic variations were indicated for the TGW with BLUEs (best linear unbiased estimations) values ranging from 35.9 g to 58.2 g with a mean value of 45.4 g and a heritability of H2=0.89. A total of 12 candidate genes for plant height, photoperiodism and grain weight were genotyped on all varieties. Only three candidates, the photoperiodism gene Ppd-D1, dwarfing gene Rht-B1and the TaGW-6A gene were significant explaining up to 14.4%, 2.3% and 3.4% of phenotypic variation, respectively. For a comprehensive genome-wide analysis of TGW-QTL genotyping data from 732 microsatellite markers and a set of 7769 mapped SNP markers genotyped with the 90k iSELECT array were analyzed. In total, 342 significant (-log10 (P-value) > 3.0) marker trait associations (MTAs) were detected for SSR markers and 1195 MTAs (-log10P-value) > 3.0) for SNP markers in all single environments plus the BLUEs. After Bonferroni correction, 28 MTAs remained significant for SSR markers (-log10 (P-value) > 4.82) and 58 MTAs for SNP markers (-log10 (P value) > 5.89). Apart from chromosomes 4B and 6B for SSR markers and chromosomes 4D and 5D for SNP markers, MTAs were detected on all chromosomes. The highest number of significant SNP markers was found on chromosomes 3B and 1B, while for the SSRs most markers were significant on chromosomes 6D and 3D. Overall, TGW was determined by many markers with small effects. Only three SNP-markers had R2 values above 6%.

Key message: By combining data from different public and private breeding programs for genomic selection, we have increased the size and diversity of the training population, which has led to better predictions of grain yield and plant height in winter wheat compared to using individual training sets. The accuracy of genome-wide prediction is anticipated to improve with an increase in training population size. In our study, we assembled a comprehensive wheat data set consisting of about 18,000 inbred lines and phenotypic data from about 250,000 plots. We evaluated the potential to train genome-wide prediction models using this big data set through data from post-registration trials conducted across a wide range of environments. Our findings demonstrated that using big data can enhance the prediction ability by up to 97% for grain yield and 44% for plant height, outperforming individual training sets. This improvement is primarily attributed to the expansion of the training set size relative to the genetic diversity. In conclusion, big data holds significant potential to accelerate genetic gain in winter wheat predictive breeding, making it a compelling option.