Accuracy of genomic selection in populations of the EEAOC sugarcane breeding program

J Racedo¹, E Rossi², AN Peña¹, AC Ghio³, DD Henriquez³, JV Díaz³, AS Noguera¹, MF Perera¹ and S Ostengo¹

¹Instituto de Tecnología Agroindustrial del Noroeste Argentino (ITANOA), Estación Experimental Agroindustrial Obispo Colombres (EEAOC) – Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET). CCT CONICET NOA Sur. Las Talitas, Tucumán, T4101XAC, República Argentina; joracedo@gmail.com

²Instituto de Investigaciones Agrobiotecnológicas (INIAB, CONICET-UNRC). Río Cuarto, Córdoba, Argentina

³Estación Experimental Agroindustrial Obispo Colombres (EEAOC). Las Talitas, Tucumán, T4101XAC, República Argentina

Genomic selection (GS) is a promising breeding tool for improving the efficiency of complex trait breeding. The prediction accuracy of genomic breeding values was assessed across populations at different stages of the Sugarcane Breeding Program at the Estación Experimental Agroindustrial Obispo Colombres (SCBP-EEAOC). GS models were tested in three breeding populations phenotyped for early maturation traits and genotyped using DArTseq SNP markers. The breeding populations consisted of: i) 264 accessions from germplasm bank (GB); ii) 160 individuals in the second clonal stage (2CS); and iii) 47 individuals from infield variety trials stage (IVT). The genomic best linear unbiased prediction (G-BLUP) model was used for all genomic predictions. The efficiency of GS was evaluated through intra- and cross-validation, depending on the population. For GB, the training population (TR) consisted of 264 accessions, and accuracies were obtained by correlations within population using an 80:20 split. For 2CS, efficiency was evaluated within the population using an 80:20 split, and the estimated model was applied to estimate genomic estimated breeding values (GEBVs) for population IVT (cross-validation). Additionally, a TR consisting of combined 2CS and IVT populations was analyzed, with accuracy obtained through correlations within the population (80:20). The highest GS efficiencies for GB were observed for pol and sugar-recovered traits (r = 0.43 and 0.40, respectively). For 2CS and IVT, the highest GS efficiencies were observed for sugar content when considering both populations jointly (r = 0.41), and when the TR consisted only of 2CS (r = 0.43). Since the effectiveness of GS in breeding programs depends on the phenotypic and molecular data, the prediction model, and the size and composition of the TR, these results are encouraging for the continued development of strategies to optimize the TR to achieve better accuracies.