Environment-specific genomic prediction ability in maize using environmental covariates depends on environmental similarity to training data.

Technology advances have made possible the collection of a wealth of genomic, environmental, and phenotypic data for use in plant breeding. Incorporation of environmental data into environment-specific genomic prediction is hindered in part because of inherently high data dimensionality. Computationally efficient approaches to combining genomic and environmental information may facilitate extension of genomic prediction models to new environments and germplasm, and better understanding of genotype-by-environment (G × E) interactions. Using genomic, yield trial, and environmental data on 1,918 unique hybrids evaluated in 59 environments from the maize Genomes to Fields project, we determined that a set of 10,153 SNP dominance coefficients and a 5-day temporal window size for summarizing environmental variables were optimal for genomic prediction using only genetic and environmental main effects. Adding marker-by-environment variable interactions required dimension reduction, and we found that reducing dimensionality of the genetic data while keeping the full set of environmental covariates was best for environment-specific genomic prediction of grain yield, leading to an increase in prediction ability of 2.7% to achieve a prediction ability of 80% across environments when data were masked at random. We then measured how prediction ability within environments was affected under stratified training-testing sets to approximate scenarios commonly encountered by plant breeders, finding that incorporation of marker-by-environment effects improved prediction ability in cases where training and test sets shared environments, but did not improve prediction in new untested environments. The environmental similarity between training and testing sets had a greater impact on the efficacy of prediction than genetic similarity between training and test sets.

Genomic prediction for the Germplasm Enhancement of Maize project.

The Germplasm Enhancement of Maize (GEM) project was initiated in 1993 as a cooperative effort of public- and private-sector maize (Zea mays L.) breeders to enhance the genetic diversity of the U.S. maize crop. The GEM project selects progeny lines with high topcross yield potential from crosses between elite temperate lines and exotic parents. The GEM project has released hundreds of useful breeding lines based on phenotypic selection within selfing generations and multienvironment yield evaluations of GEM line topcrosses to elite adapted testers. Developing genomic selection (GS) models for the GEM project may contribute to increases in the rate of genetic gain. Here we evaluated the prediction ability of GS models trained on 6 yr of topcross evaluations from the two GEM programs in Raleigh, NC, and Ames, IA, documenting prediction abilities ranging from 0.36 to 0.75 for grain yield and from 0.78 to 0.96 for grain moisture when models were cross-validated within program and heterotic group. Predicted genetic gain from GS ranged from 0.95 to 2.58 times the gain from phenotypic selection. Prediction ability across program and heterotic group was generally poorer than within groups. Based on observed genomic relationships between GEM breeding lines and their tropical ancestors, GS for either yield or moisture would reduce recovery of exotic germplasm only slightly. Using GS models trained within program, the GEM programs should be able to more effectively deliver on its mission to broaden the genetic base of U.S. germplasm.

The importance of dominance and genotype-by-environment interactions on grain yield variation in a large-scale public cooperative maize experiment.

High-dimensional and high-throughput genomic, field performance, and environmental data are becoming increasingly available to crop breeding programs, and their integration can facilitate genomic prediction within and across environments and provide insights into the genetic architecture of complex traits and the nature of genotype-by-environment interactions. To partition trait variation into additive and dominance (main effect) genetic and corresponding genetic-by-environment variances, and to identify specific environmental factors that influence genotype-by-environment interactions, we curated and analyzed genotypic and phenotypic data on 1918 maize (Zea mays L.) hybrids and environmental data from 65 testing environments. For grain yield, dominance variance was similar in magnitude to additive variance, and genetic-by-environment variances were more important than genetic main effect variances. Models involving both additive and dominance relationships best fit the data and modeling unique genetic covariances among all environments provided the best characterization of the genotype-by-environment interaction patterns. Similarity of relative hybrid performance among environments was modeled as a function of underlying weather variables, permitting identification of weather covariates driving correlations of genetic effects across environments. The resulting models can be used for genomic prediction of mean hybrid performance across populations of environments tested or for environment-specific predictions. These results can also guide efforts to incorporate high-throughput environmental data into genomic prediction models and predict values in new environments characterized with the same environmental characteristics.