Skim-sequencing for genomic selection in wheat: a comparison of marker platforms

The promise of genomics-assisted breeding relies on efficient, affordable, and abundant molecular markers. Leveraging modern sequencing technology, commercial laboratory products, and open-source software, we demonstrate how ultra-low whole-genome sequencing coverage (skim-seq, 0.05-0.10x) can be a viable marker platform. The direct generation of sequence data followed by imputation provides an opportunity to implement genomic selection while being robust to future genomic (changes in reference genome) and technological (improvement in sequencing capacity) changes. We genotyped 1709 wheat (Triticum aestivum) lines with genotyping-by-sequencing (GBS), a mid-density DArTAG single nucleotide polymorphism panel, and skim-seq (0.07x). All skim-seq samples were used to identify loci variants using a reference genome without the aid of any high-coverage samples. STITCH software was used for imputation to obtain 121,437 markers. Comparing high-confidence STITCH imputed loci (approximately 65,000 of 14 M imputed loci) to high-coverage samples resulted in the correct imputation for more than 97.5% of the markers. Using phenotypic data, a fivefold cross validation was implemented for each marker platform. No one marker system performed the best in all test cases, with GBS often resulting in the highest correlation between observed and predicted values. The skim-seq correlations were typically within 0.03 of GBS, suggesting skim-seq can be a viable marker strategy for genomic prediction. As technology and computational pipelines advance, skim-seq appears to be a promising method to bridge the gap between targeted genotyping and whole-genome sequencing. The skim-seq method is highly flexible and can be optimized to a variety of program needs, potentially allowing for wide adoption by the plant breeding community.