Gene-Based Genome-Wide Association Study Identified Genes for Agronomic Traits in Maize

SIMPLE SUMMARY: Genome-wide association studies (GWAS) have successfully detected many SNPs related to complex quantitative traits. However, SNPs significantly associated with quantitative traits usually have only mild effects. Quantitative traits are usually caused by the combined effects of multiple loci in a gene. Maize is one of the world’s most important foods and feed crops. Earlier silking, kernel oil concentration, and fatty acid composition are all important agronomic traits in maize. To further explore the gene-level variations affecting maize economic traits, we propose an efficient gene-based GWAS method. We applied this method to the economic traits of maize and identified many candidate genes. Many of the same candidate genes were found in the analysis of related maize traits, which proved the reliability of our method. These findings will provide a theoretical basis for maize breeding with the targeted earlier silking and kernel oil concentration traits. ABSTRACT: A gene integrates the effects of all SNPs in its sequence span, which benefits the genome-wide association study. To explore gene-level variations affecting economic traits in maize, we extended the SNP-based GWAS analysis software Single-RunKing developed by our team to gene-based GWAS, which used the FaST-LMM algorithm to convert the linear mixed model into simple linear model association analysis. An F-test statistic was formulated to test and identify candidate genes. We compared the statistical efficiency of using 80% principal components (EPC), the first principal component (FPC), and all SNP markers (ALLSNP) as independent variables, which predecessors commonly used to integrate SNPs and represent genes. With a Huazhong Agricultural University (HAU) genomic dataset of 2.65M SNPs from 540 maize plants, 34,774 genes were annotated across the whole genome. Genome-wide association studies with 20 agronomic traits were performed using the software developed here. Another maize dataset from the Ames panel (AP) was also analyzed. The EPC method fits the model well and has good statistical efficiency. It not only overcomes the false negative problem when using all SNP markers for analysis (ALLSNP) but also solves the false positive problem of its corresponding simple linear model method EPCLM. Compared with FPC, the EPC method has higher statistical efficiency. A total of 132 quantitative trait genes (QTG) were identified for the 20 traits from HAU maize dataset and one trait of AP maize.