Comparing genomic selection and marker-assisted selection for Fusarium head blight resistance in wheat (Triticum aestivum L.)

• Published in: Molecular breeding Vol. 36; no. 7; pp. 1 - 11
• Main Authors: Arruda, M. P., Lipka, A. E., Brown, P. J., Krill, A. M., Thurber, C., Brown-Guedira, G., Dong, Y., Foresman, B. J., Kolb, F. L.
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01-07-2016; Springer Nature B.V
• Subjects: Biomedical and Life Sciences; Biotechnology; Blight; Breeding; Fusarium; Fusarium head blight; Gene mapping; Genomes; Genomics; Life Sciences; Linear prediction; Marker-assisted selection; Mathematical models; Model accuracy; Molecular biology; Plant biology; Plant Genetics and Genomics; Plant Pathology; Plant Physiology; Plant Sciences; Quantitative trait loci; Statistical analysis; Statistical models; Triticum aestivum; Wheat; Winter wheat; Molecular breeding; Marker-assisted selection; Deoxynivalenol; Genomic selection; Genotyping-by-sequencing
• ISSN: 1380-3743; 1572-9788
• DOI: 10.1007/s11032-016-0508-5

Genomic selection (GS) and marker-assisted selection (MAS) rely on marker–trait associations and are both routinely used for breeding purposes. Although similar, these two approaches differ in their applications and how markers are used to estimate breeding values. In this study, GS and MAS were compared in their ability to predict six traits associated with resistance to a destructive wheat disease, Fusarium head blight (FHB). A panel consisting in 273 soft red winter wheat lines from the US Midwestern and Eastern regions was used in this study. The statistical models for MAS were built using
Fhb
-
1
, the best-studied quantitative trait loci (QTL) for FHB resistance, and two sets of QTL: one independently identified by other groups and a newer set identified “
in house
”. In contrast, genomic selection models relied on 19,992 SNPs distributed throughout the genome. For the MAS and GS models, marker effects were estimated with ordinary least square and ridge regression best unbiased linear prediction, respectively. Intermediate to high values of prediction accuracy (0.4–0.9) were observed for most GS models, with lower values (<0.3) found for MAS models. Treating QTL as fixed effects in GS models resulted in higher prediction accuracy when compared with a GS model with only random effects, but overestimated accuracies were obtained with
in house
QTL. For the same selection intensity, GS resulted in higher selection differentials than MAS for all traits. Our results indicate that GS is a more appropriate strategy than MAS for FHB resistance.