Abstract
Banana breeding is hampered by the very low fertility of domesticated bananas and the lack of knowledge about the genetic determinism of agronomic traits. We analysed a breeding population of 2723 triploid hybrids resulting from crosses between diploid and tetraploid Musa acuminata parents, which was evaluated over three successive crop cycles for 24 traits relating to yield components and plant, bunch, and fruit architectures. A subset of 1129 individuals was genotyped by sequencing, revealing 205 612 single-nucleotide polymorphisms (SNPs). Most parents were heterozygous for one or several large reciprocal chromosomal translocations, which are known to impact recombination and chromosomal segregation. We applied two linear mixed models to detect associations between markers and traits: (i) a standard model with a kinship calculated using all SNPs and (ii) a model with chromosome-specific kinships that aims at recovering statistical power at alleles carried by long non-recombined haplotypic segments. For 23 of the 24 traits, we identified one to five significant quantitative trait loci (QTLs) for which the origin of favourable alleles could often be determined amongst the main ancestral contributors to banana cultivars. Several QTLs, located in the rearranged regions, were only detected using the second model. The resulting QTL landscape represents an important resource to support breeding programmes. The proposed strategy for recovering power at SNPs carried by long non-recombined rearranged haplotypic segments is an important methodological advance for future association studies in banana and other species affected by chromosomal rearrangements.