Abstract
Genotype × environment interaction (G × E) is the differential response of genotypes in different environments and represents a major challenge for breeders. Genotype × year‐interaction (G × Y) is a relevant component of G × E, and accounting for it is an important strategy for identifying lines with stable and superior performance across years. In this study, we compared the prediction accuracy of modeling G × Y using covariance structures that differ in their ability to accommodate correlation among environments. We present the use of these approaches in two different rice (Oryza sativa L.) breeding populations (indica and tropical japonica) for predicting grain yield, plant height, and three milling quality traits—milling yield, head rice percentage, and grain chalkiness—under different cross‐validation (CV) scenarios. We also compared model performance in the context of global predictions (i.e., predictions across years). Most of the benefits of multienvironment models come from modeling genetic correlations between environments when predicting performance of lines that have been tested in some environments but not others (CV2). For predicting the performance of newly developed lines (CV1), modeling between environment correlations has no effect compared with considering environments independently. Response to selection of multienvironment models when modeling covariance structures that accommodate covariances between environments was always beneficial when predicting the performance of lines across years. We also show that, for some traits, high prediction accuracies can be obtained in untested years, which is important for resource allocation in small breeding programs.
Highlights
Genotypeenvironment interaction (G E) is the differential response of genotypes in different environments and represents a major challenge for breeders
The grain quality traits measured were yield after milling (MY measured in g, as the weight of grain recovered after milling divided by the weight of rough rice before milling, using a 100-g sample of rough rice), percentage of head rice recovery (PHR measured in g, as the weight of whole milled kernels, divided by the weight of rough rice, using a 100-g sample of rough rice), and the percentage of grain chalkiness (GC, measured as % of chalky kernels in a subsample of 50 g of total milled rice, where the area of chalk core—white back or white belly—was larger than half the kernel area based on visual inspection) (Supplemental Table S1)
The trait distributions for each year are similar, we see a slight increase for grain yield (GY) and decrease for plant height (PH) in 2012 (Fig. 1A)
Summary
Genotypeenvironment interaction (G E) is the differential response of genotypes in different environments and represents a major challenge for breeders. We compared the prediction accuracy of modeling G Y using covariance structures that differ in their ability to accommodate correlation among environments. Response to selection of multienvironment models when modeling covariance structures that accommodate covariances between environments was always beneficial when predicting the performance of lines across years. GS methods were focused on single-trait, single-environment analyses, but one major challenge for breeders is the differential response of genotypes in different environments, known as genotypeenvironment interaction (G E). The G E can affect trait heritability and line ranking over environments, frequently affecting decision making For this reason, GS approaches capable of modeling G E have increasingly gained popularity. The use of mixed model approaches was later introduced to improve flexibility and to allow different correlation structures among environments (Piepho, 1998; Piepho and Möring, 2005; Burgueño et al, 2008, 2012)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
