Models with only two predictor variables can accurately predict seed yield in diploid and tetraploid red clover

Red clover ( Trifolium pratense L.) is a valuable protein-rich forage crop, but poor seed yields are often a setback, especially in tetraploid varieties. Published work on factors underlying seed yield report contradictory or (sometimes) inconclusive results. Here we report on the effect of 10 traits related to seed yield and flowering on seed yield in 600 genotypes from 15 diploid and 15 tetraploid cultivars from diverse origins. Multiple linear regression (MLR) models indicated that the variation in seed number per plant was well explained by (1) the number of ripe flower heads per plant and (2) the seed number per ripe flower head in diploids (R 2 = 0.953) and tetraploids (R 2 = 0.919). Flower color was not significantly related to seed yield, and corolla tube dimensions were not (or only borderline) significantly related to seed yield. The applicability of our models to predict seed yield was validated on the same set of plants in the following harvest year, and on a totally different dataset available from Vleugels et al. (Plant Breed 134:56–61, 2015 ). In the 600-plant dataset and the separate Vleugels et al. dataset, the diploid model explained 94.8 and 53.2 %, respectively, and the tetraploid model explained 88.1 and 64.9 %, respectively, of the variation in seed yield. Our models can therefore predict seed yield with good to fair accuracy in red clover plants grown under various conditions. Breeders can increase seed yield by selecting plants that have numerous ripe flower heads and a high seed number per head. Flower color and corolla tube dimensions can be excluded as breeding targets.