Optimum prediction of three-way crosses from single crosses in forage maize (Zea mays L.)

Abstract
Three-way cross means were predicted with formulae involving linear functions of general (GCA) and specific combining ability (SCA) effects estimated from single-cross factorials between genetically divergent populations. Data from an experiment with 66 single-cross and 66 three-way cross forage maize (Zea mays L.) hybrids was used for comparing the prediction formulae. The genotypic correlation (r) between observed and predicted three-way crosses increased with increasing χ, the weighting factor of SCA effects, for plant height and ear dry matter (DM) content. It displayed slightly convex curves for total and stover DM yield, ear percentage, and metabolizable energy content of stover. For Jenkins' method B, r was considerably less than 1.0 for all traits, indicating the presence of epistasis. The square root of heritability (hĜ) of the predicted means decreased with increasing χ, the reduction being small with a greater number of test environments. Using the product r·hĜ as a criterion of efficiency, none of the prediction methods was consistently superior and the differences among them were rather small (< 7.5%) for all traits, irrespective of the number of test environments. We recommend evaluating the GCA of a greater number of lines from each parent population in testcrosses with a small number of elite lines from the opposite population. All possible three-way or double crosses between both sets of lines should be predicted by Jenkins's method C. This procedure allows one to select with a higher intensity among the predicted hybrids and thus should increase the genetic gain.