| Popis: |
Genomic selection (GS) has high potential interest for improving alfalfa biomass yield and forage quality, to alleviate challenges for phenotypic selection (PS) represented by low narrow-sense heritability, long selection cycles, high evaluation costs, and multi-trait selection. This report discusses various factors that may affect the prediction ability and the cost-efficient exploitation of GS in breeding programs, considering as well specific aspects relative to genotyping-by-sequencing (GBS)-generated markers. We provided an original comparison of six statistical models for GS and four SNP calling procedures for GBS data (based on M. truncatula or M. sativa genomes, the dDocent-mock reference genome, and the UNEAK pipeline) in terms of predictive ability for biomass yield, leaf protein content, and stem NDF digestibility. Current GBS costs and other considerations support the application of GS to predict additive genetic variation effects (as allowed for by phenotyping half-sib progenies of genotyped parent plants) of plants belonging to relatively broad-based reference populations, following a preliminary stage of stratified mass selection. We outlined a procedure for comparing GS versus PS in terms of selection efficiency according to predicted genetic gains per unit time and same selection cost, which suggested predictive accuracy around 0.15 as a threshold value for considering GS more cost-efficient than PS for biomass yield. A similar threshold may apply to alfalfa forage quality traits selected concurrently with crop yield. Pioneer genomic selection studies for biomass yield or forage quality traits of alfalfa and other perennial forages are generally encouraging for GS implementation. However, information on GS prediction accuracy is still lacking or extremely limited for biomass yield in environments featuring different prevailing stresses (e.g., drought, cold, salinity) or specific crop managements (e.g., severe grazing, intercropping). Crucial research issues for alfalfa GS optimization are represented by cost-efficient allele dosage estimation, quality of cross-population predictions (which may affect GS strategies and the definition of genetic bases by breeding programs), the value of parsimonious GS models incorporated into new genotyping tools (e.g., RAD capture ones), and most of all, the comparison of GS versus PS in terms of actual genetic gains per unit time achieved with similar selection costs. |
