Use of genomic selection in breeding rice (Oryza sativa L.) for resistance to rice blast (Magnaporthe oryzae)

• Published in: Molecular breeding Vol. 39; no. 8; pp. 1 - 16
• Main Authors: Huang, Mao, Balimponya, Elias G., Mgonja, Emmanuel M., McHale, Leah K., Luzi-Kihupi, Ashura, Wang, Guo-Liang, Sneller, Clay H.
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01-08-2019; Springer Nature B.V
• Subjects: Accuracy; Bayesian analysis; Biomedical and Life Sciences; Biotechnology; Disease control; Genes; Genomics; Life Sciences; Magnaporthe oryzae; Markers; Mathematical models; Molecular biology; Nucleotides; Oryza sativa; Phenotypes; Plant biology; Plant breeding; Plant Genetics and Genomics; Plant Pathology; Plant Physiology; Plant reproduction; Plant Sciences; Polymorphism; Population; Population structure; Regression analysis; Regression models; Rice; Rice blast; Single-nucleotide polymorphism; Genotype by sequencing; Susceptible; Accuracy; Resistant; Rice blast; Genomic selection
• ISSN: 1380-3743; 1572-9788
• DOI: 10.1007/s11032-019-1023-2

Rice blast (RB), caused by the fungal pathogen
Magnaporthe oryzae
, is a major disease in rice (
Oryzae sativa
L.) with resistance controlled by major and minor genes. Genomic selection (GS) is a breeding technology applicable for selecting traits controlled by many genes. Our objective was to assess the utility of GS in improving RB resistance. A population of 161 accessions from Africa and another population of 162 accessions from the USA were evaluated for resistance to six and eight RB isolates, respectively. Each rice population was genotyped with single nucleotide polymorphism (SNP) markers. The accuracy of GS was determined using seven models: genomic best linear unbiased prediction (gBLUP), gBLUP with some markers as fixed effects (fgBLUP), gBLUP model with population structure as a covariate (sgBLUP), multitrait gBLUP (mgBLUP), Bayesian (BayesA and BayesC) models, and a multiple linear regression model using significant markers (MLR). Each set of population had accessions with good resistance to multiple isolates. Using cross-validation, the accuracy of gBLUP ranged from 0.15 to 0.72; the gBLUP, sgBLUP, mgBLUP, and Bayesian methods had similar accuracy, while fgBLUP gave the greatest accuracy. Without cross-validation, gBLUP, sgBLUP, fgBLUP, and Bayesian methods were similar and were superior to mgBLUP and MLR. In general, a GS model built on data from one isolate was able to predict the phenotypes generated from other isolates, suggesting common genes controlling resistance across isolates. Our results demonstrate that GS may be a very useful method to improve RB resistance. The fgBLUP model could be used to effectively select for both durable and resistance traits conferred by major genes.