The phenotypic analysis of field experiments includes information about the experimental design, the randomization structure and a number of putative dependencies of environment and design factors on the trait investigated. In QTL studies, the genetic correlation across environments, which arises when the same set of lines is tested in multiple environments, plays an important role. This paper investigates the effect of model choice on the set and magnitude of detected root QTL in rice. Published studies on QTLs for root traits indicate that different results are obtained if varying genetic populations are used and also if different environmental conditions are included. An experiment was conducted with 168 RILs of the Bala x Azucena mapping population plus parents as checks under four environmental conditions (low light, low nitrogen, drought and a control environment). We propose a model that incorporates all relevant experimental information into a composite interval mapping approach based on a mixed model, which especially considers the correlation of genotypes in different environments. An extensive sequential model selection procedure was applied based on the phenotypic model, using the AIC to determine an appropriate random structure and Type 3 Wald F-tests for selection of fixed effects. In a first step we checked whether any of the fixed effects and random (nested) design effects could be dropped. Secondly, an appropriate covariance structure was chosen for genotype x environment interaction. In a third step Box-Cox transformations were applied based on residual analysis. We compared profiles of composite interval mapping scans with and without the inclusion of genotype x environment interaction and the experimental design information. Some distinct differences in profiles indicate that insufficient modeling of the non-QTL part can lead to an overly optimistic interpretation of QTL main effects in interval mapping. It is concluded that mixed model QTL mapping offers a reasonable way to separate environmental and genetic influences in the evaluation of quantitative genes and especially enables a more realistic assessment of QTL and QTL x environment effects than standard approaches by including all relevant effects.
- experimental design
- genotype x environment interaction
- mixed model
- QTL mapping
- rice root traits