Choice of Models for QTL Mapping with Multiple Families and Design of the Training Set for Prediction of Fusarium Resistance Traits in Maize

Sen Han,H. Friedrich Utz,Wenxin Liu,Tobias A. Schrag,Michael Stange,Tobias Würschum,Thomas Miedaner,Eva Bauer,Chris-Carolin Schön,Albrecht E. Melchinger

DOI: https://doi.org/10.1007/s00122-015-2637-3

IF: 5.4

2015-01-01

Theoretical and Applied Genetics

Abstract:Key message QTL analysis for Fusarium resistance traits with multiple connected families detected more QTL than single-family analysis. Prediction accuracy was tightly associated with the kinship of the validation and training set. Abstract QTL mapping has recently shifted from analysis of single families to multiple, connected families and several biometric models have been suggested. Using a high-density consensus map with 2472 marker loci, we performed QTL mapping with five connected bi-parental families with 639 doubled-haploid (DH) lines in maize for ear rot resistance and analyzed traits DON, Gibberella ear rot severity (GER), and days to silking (DS). Five biometric models differing in the assumption about the number and effects of alleles at QTL were compared. Model 2 to 5 performing joint analyses across all families and using linkage and/or linkage disequilibrium (LD) information identified all and even further QTL than Model 1 (single-family analyses) and generally explained a higher proportion p G of the genotypic variance for all three traits. QTL for DON and GER were mostly family specific, but several QTL for DS occurred in multiple families. Many QTL displayed large additive effects and most alleles increasing resistance originated from a resistant parent. Interactions between detected QTL and genetic background (family) occurred rarely and were comparatively small. Detailed analysis of three fully connected families yielded higher p G values for Model 3 or 4 than for Model 2 and 5, irrespective of the size N TS of the training set (TS). In conclusion, Model 3 and 4 can be recommended for QTL-based prediction with larger families. Including a sufficiently large number of full sibs in the TS helped to increase QTL-based prediction accuracy ( r VS ) for various scenarios differing in the composition of the TS.