Pubblicazioni

Wheat contamination with mycotoxins caused by Fusarium is of great economic importance. The current routine method of disease evaluation (single-location trial based on visual symptoms) does not yield the comprehensive data needed to accelerate breeding for resistance to Fusarium. We intend to lay the foundation for more efficient resistance breeding against Fusarium, by targeting deoxynivalenol (DON), the most important mycotoxin, and improve resistance to DON accumulation by genomic selection (GS). A reference set (RS) of 300 cultivars and breeding lines was phenotyped in a three-location, artificially inoculated trial over two years. We evaluated resistance to Fusarium by visual scoring of symptoms, expressed as Area Under the Disease Progress Curve (AUDPC), and by measuring DON content of harvested grain by HPLC. We additionally scored plant height, heading date and anther extrusion. In the third year, a validation set (VS) of 225 new lines was integrated in the same experimental setting. The purpose of the VS was, on the one hand, to evaluate the potential of GS by comparing the observed phenotypes with corresponding genomic predictions obtained from models trained using the RS. On the other hand, the new lines would ultimately extend the number of genotypes available for GS model training to a total of 525. We obtained satisfactory phenotypic results for all years and most locations, with generally high and homogeneous Fusarium pressure and subsequent DON accumulation, resulting in high heritability for both AUDPC and DON. The correlation between Fusarium symptoms and DON was moderate (r = 0.54), confirming that measurement of DON is a valuable asset for resistance breeding. Correlations with other traits were low to moderate, but significant, and confirmed that both height and earliness influence Fusarium infection and should be considered into account when implementing GS. For both Fusarium resistance traits, the first independent validation of GS through the VS yielded moderate prediction abilities (0.3 - 0.4), while prediction abilities from cross-validation schemes with the full set of available data (525 genotypes, 3 years) reached values of 0.5 to 0.6. Realistic estimates of prediction abilities for future GS-use should lie in the 0.4 – 0.5 range, and are sufficient for significant breeding progress through GS. By the end of the project, we are thus able to employ predictions for 5’000 already genotyped breeding lines, which will help selection across all stages of the breeding program, from the choice of crossing parents up to selection of advanced lines.