Enhancing genomic prediction with genome-wide association studies in multiparental maize populations

Authors: BIAN, YANG - North Carolina State University; Holland, Jim - Jim

Submitted to: Heredity
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/12/2017
Publication Date: 6/1/2017
Citation: Bian, Y., Holland, J.B. 2017. Enhancing genomic prediction with genome-wide association studies in multiparental maize populations. Heredity. 118:585–593.

Interpretive Summary: Genomic selection methods that have been widely adopted by commercial animal and plant breeding programs.have optimal prediction accuracy when traits are controlled by many genes each with small effects (‘polygenic genetic architecture’). Genome-wide association analyses attempt to identify specific genes with larger effects to understand what genes might be most important for controlling variation in traits. These two approaches are complementary, as they try to model different aspects of genetic control of traits. We analyzed simulated and real data from maize to compare standard polygenic genomic prediction models to models that combine prediction of the polygenic effect plus effects of a few specific genetic variants discovered with genome-wide association analyses. Results indicate that including ‘discoveries’ from association analysis in prediction models improves accuracy only when there are larger effect genes controlling the trait in addition to polygenic effects.

Technical Abstract: Genome-wide association mapping using dense marker sets has identified some nucleotide variants affecting complex traits which have been validated with fine-mapping and functional analysis. Many sequence variants associated with complex traits in maize have small effects and low repeatability, however. In contrast to genome-wide association study, genomic prediction is typically based on models incorporating information from all available markers, rather than modeling effects of individual loci. We considered methods to integrate results of genome-wide association studies into genomic prediction models in the context of multiple interconnected families. We compared association tests based on a biallelic additive model constraining the effect of a SNP to be equal across all families in which it segregates to a model in which the effect of a SNP can vary across families. Association SNPs were then included as fixed effects into a genomic prediction model that also included the random effects of the whole genome background. Simulation studies revealed that the effectiveness of this joint approach depends on the extent of polygenicity of the traits. Congruent with this finding, cross-validation studies indicated that genomic prediction including the fixed effects of the most significantly associated SNPs along with the polygenic background was more accurate than the polygenic background model alone for moderately complex but not highly polygenic traits measured in the maize nested association mapping population. Individual SNPs with strong and robust association signals can effectively improve genomic prediction. Our approach provides a new integrative modelling approach for both reliable gene discovery and robust genomic prediction.