New Look at an Old Problem: Breeding for Multiple Traits.

Understanding plant trait interactions may allow breeders to select for multiple traits at the same time more effectively than evaluating traits on net effects. Plant trait relationships can be evaluated in multiple ways: multiple regression, ANOVA, and mixed models. However, all of these methods describe or quantify net effects and use one general error function. Structural equation modeling (SEM), differs in three main ways 1) the pathway or process is modeled, 2) each element in the model is assigned an error, and 3) variables that are not directly measured can be modeled using observed measurements. These three elements allow for SEM to have the ability to model highly complex networks of collinear and causally linked variables, such as plant trait relationships. We were interested in determining if SEM could model breeding changes within a populations and increase breeding accuracy and decrease time to genetic improvement when selecting for multiple traits within a breeding program. SEM was applied to elite oat and barley breeding populations to determine network relationships between common agronomic traits and a historic oat population to verify if network changes to traits occurred over time due to breeding pressure. Breeding for single traits (i.e. yield) did change the relationship between traits substantially between historic and modern varieties (R²=0.29 and 0.74, respectively). Multiple common multi-trait selection indices were applied to one breeding population in barley. Economic value and agronomic trait relationships were then used for determining selection weights.