Genomic predictions for Fusarium head blight resistance in a diverse durum wheat panel: an effective incorporation of plant height and heading date as covariates

Abstract

Selection for multiple traits is a highly challenging task for breeders due to potential unfavorable associations between characters. Fusarium head blight FHB, being one of the most relevant diseases affecting durum wheat frequently shows in this respect an unfavorable correlation with morpho-agronomical traits like plant height (PH) and heading date (HD). In this study, we used a cross-validation scheme to assess the prediction ability of the genomic predictions (GP) for FHB severity relying on genomic best linear unbiased prediction models in a diverse panel of 178 durum wheat lines evaluated across five environments. Additionally, we compared three types of approaches to include HD and PH as covariates into the analysis: (1) correcting FHB severity values before training GP models, (2) tuning the GP model parameters that included multi-trait alternatives, and (3) adjusting the genomic-based predictions by restriction indexes. Models that weighted genomic estimated breeding values (GEBV) by restriction indexes as well as models that predicted FHBms values corrected by regression-based methods were efficient alternatives in diminishing the HD trade-off, nonetheless they were also associated with large reductions in prediction ability for FHB severity. After a simulated round of genomic selection, considering HD as fixed effect in the GP model were the most suitable alternative to select a higher proportion of genotypes moderately resistant with lower-than-average HD and PH estimations. Hence, an appropriate GP model given unfavorable association between characters should combine high predictabilities and adequate reduction of undesired trade-offs.