Background: Eight diverse sorghum (Sorghum bicolor L. Moench) accessions were subjected to short-read genome sequencing to characterize the distribution of single-nucleotide polymorphisms (SNPs). Two strategies were used for DNA library preparation. Missing SNP genotype data were imputed by local haplotype comparison. The effect of library type and genomic diversity on SNP discovery and imputation are evaluated.
Results: Alignment of eight genome equivalents (6 Gb) to the public reference genome revealed 283,000 SNPs at ≥82% confirmation probability. Sequencing from libraries constructed to limit sequencing to start at defined restriction sites led to genotyping 10-fold more SNPs in all 8 accessions, and correctly imputing 11% more missing data, than from semirandom libraries. The SNP yield advantage of the reduced-representation method was less than expected, since up to one fifth of reads started at noncanonical restriction sites and up to one third of restriction sites predicted in silico to yield unique alignments were not sampled at near-saturation. For imputation accuracy, the availability of a genomically similar accession in the germplasm panel was more important than panel size or sequencing coverage.
Conclusions: A sequence quantity of 3 million 50-base reads per accession using a BsrFI library would conservatively provide satisfactory genotyping of 96,000 sorghum SNPs. For most reliable SNP-genotype imputation in shallowly sequenced genomes, germplasm panels should consist of pairs or groups of genomically similar entries. These results may help in designing strategies for economical genotyping-by-sequencing of large numbers of plant accessions.