Genomic prediction and GWAS of Gibberella ear rot resistance traits in dent and flint lines of a public maize breeding program

Abstract

Gibberella ear rot (GER) is a serious threat to maize cultivation, causing grain yield losses and contamination with mycotoxins. Genomic prediction (GP) has great potential to accelerate resistance breeding against GER. However, small training sets (TS) consisting of both phenotyped and genotyped individuals are a challenge for obtaining high prediction accuracy (ρ) in GP. A potential solution would be combining small-size populations across heterotic pools. However, genetic heterogeneity between populations in terms of segregating QTL, linkage disequilibrium (LD) pattern and genomic relationships can impair ρ of GP. In this study, we investigated the genetic architecture of GER severity, deoxynivalenol concentration (DON) and days to silking with genome-wide association studies within two elite panels of 130 dent and 114 flint lines from the maize breeding program of the University of Hohenheim tested in four environments. We also assessed the consistency of LD pattern and genomic relationships between the two heterotic pools. Furthermore, we compared four GP approaches differing in the composition of the TS with lines from a single or combined pool(s) and statistical models with marker effects identical or different but correlated between pools. We detected two and six QTL for DON within the dent and flint pool, respectively, but no common QTL. The LD pattern was consistent between pools for marker pairs ≤ 10 kb apart. GP across pools yielded low or even negative ρ. Combined-pool GP had no higher ρ than within-pool GP, regardless of the statistical model. Our findings underline the importance of investigating the genetic heterogeneity between populations prior to implementing GP using combined TS.