AUTHOR=Li Jinlong , Cheng Dehe , Guo Shuwei , Chen Chen , Wang Yuwen , Zhong Yu , Qi Xiaolong , Liu Zongkai , Wang Dong , Wang Yuandong , Liu Wenxin , Liu Chenxu , Chen Shaojiang TITLE=Genome-wide association and genomic prediction for resistance to southern corn rust in DH and testcross populations JOURNAL=Frontiers in Plant Science VOLUME=14 YEAR=2023 URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2023.1109116 DOI=10.3389/fpls.2023.1109116 ISSN=1664-462X ABSTRACT=

Southern corn rust (SCR), caused by Puccinia polysora Underw, is a destructive disease that can severely reduce grain yield in maize (Zea mays L.). Owing to P. polysora being multi-racial, it is very important to explore more resistance genes and develop more efficient selection approaches in maize breeding programs. Here, four Doubled Haploid (DH) populations with 384 accessions originated from selected parents and their 903 testcross hybrids were used to perform genome-wide association (GWAS). Three GWAS processes included the additive model in the DH panel, additive and dominant models in the hybrid panel. As a result, five loci were detected on chromosomes 1, 7, 8, 8, and 10, with P-values ranging from 4.83×10-7 to 2.46×10-41. In all association analyses, a highly significant locus on chromosome 10 was detected, which was tight chained with the known SCR resistance gene RPPC and RPPK. Genomic prediction (GP), has been proven to be effective in plant breeding. In our study, several models were performed to explore predictive ability in hybrid populations for SCR resistance, including extended GBLUP with different genetic matrices, maker based prediction models, and mixed models with QTL as fixed factors. For GBLUP models, the prediction accuracies ranged from 0.56-0.60. Compared with traditional prediction only with additive effect, prediction ability was significantly improved by adding additive-by-additive effect (P-value< 0.05). For maker based models, the accuracy of BayesA and BayesB was 0.65, 8% higher than other models (i.e., RRBLUP, BRR, BL, BayesC). Finally, by adding QTL into the mixed linear prediction model, the accuracy can be further improved to 0.67, especially for the G_A model, the prediction performance can be increased by 11.67%. The prediction accuracy of the BayesB model can be further improved significantly by adding QTL information (P-value< 0.05). This study will provide important valuable information for understanding the genetic architecture and the application of GP for SCR in maize breeding.