Abstract
Subsampling a reduced number of accessions from ex situ genebank collections, known as core collections, is a widely applied method for the investigation of stored genetic diversity and for an exploitation by breeding and research. Optimizing core collections for genome-wide association studies could potentially maximize opportunities to discover relevant and rare variation. In the present study, eight strategies to sample core collections were implemented separately for two traits, namely susceptibility to yellow rust and stem lodging, on about 6,300 accessions of winter wheat (Triticum aestivum L.). Each strategy maximized different parameters or emphasized another aspect of the collection; the strategies relied on genomic data, phenotypic data or a combination thereof. The resulting trait-customized core collections of eight different sizes, covering the range between 100 and 800 accession samples, were analyzed based on characteristics such as population stratification, number of duplicate genotypes and genetic diversity. Furthermore, the statistical power for an association study was investigated as a key criterion for comparisons. While sampling extreme phenotypes boosts the power especially for smaller core collections of up to 500 accession samples, maximization of genetic diversity within the core collection minimizes population stratification and avoids the accumulation of less informative duplicate genotypes when increasing the size of a core collection. Advantages and limitations of different strategies to create trait-customized core collections are discussed for different scenarios of the availability of resources and data.