How Many Paternal Lines Are Enough? Genomic Prediction of Combining Ability in a Sunflower Breeding Panel

Background/Objectives: Sunflower (Helianthus annuus L.) paternal lines are evaluated by testcrossing over several seasons. We asked how well parental genotypes predict the general combining ability (GCA) of an untested line. Methods: Fifty-four paternal lines and three cytoplasmic male-sterile testers were genotyped by sequencing. Their 162 hybrids were phenotyped for two years at one site and genotyped in silico. Genomic best linear unbiased prediction (GBLUP) was evaluated on the 52 oilseed lines by holding out each line with its hybrids. Results: Predictive ability for paternal GCA was 0.66 for 1000-seed weight, 0.40 for oil content, 0.47 for seed yield, and 0.48 for oil yield. Gains beyond about 30 training lines were unclear within this panel, with small test sets at the largest sizes. Parental distance correlated weakly with hybrid yield (r = 0.31), and additive scans found no significant associations. A chromosome 14 candidate for 1000-seed weight had a dominance deviation of +3.38 g, but five-trait-adjusted p-values were 0.047 (bootstrap) and 0.052 (within-tester permutation). Inferred hybrid heterozygosity follows parental combinations, preventing separation of locus-specific dominance from parental-combination effects. Conclusions: These within-panel results support further evaluation of genomic preselection, especially for 1000-seed weight. Neither training-set size nor predictive performance is established for other populations or environments; the chromosome 14 candidate requires independent validation.