GBS-SNP-CROP: a reference-optional pipeline for SNP discovery and plant germplasm characterization using variable length, paired-end genotyping-by-sequencing data

Abstract

Background: With its simple library preparation and robust approach to genome reduction, genotyping-by-sequencing (GBS) is a flexible and cost-effective strategy for SNP discovery and genotyping, provided an appropriate reference genome is available. For resource-limited curation, research, and breeding programs of underutilized plant genetic resources, however, even low-depth references may not be within reach, despite declining sequencing costs. Such programs would find value in an open-source bioinformatics pipeline that can maximize GBS data usage and perform high-density SNP genotyping in the absence of a reference.

Results: The GBS SNP-Calling Reference Optional Pipeline (GBS-SNP-CROP) developed and presented here adopts a clustering strategy to build a population-tailored "Mock Reference" from the same GBS data used for downstream SNP calling and genotyping. Designed for libraries of paired-end (PE) reads, GBS-SNP-CROP maximizes data usage by eliminating unnecessary data culling due to imposed read-length uniformity requirements. Using 150 bp PE reads from a GBS library of 48 accessions of tetraploid kiwiberry (Actinidia arguta), GBS-SNP-CROP yielded on average three times as many SNPs as TASSEL-GBS analyses (32 and 64 bp tag lengths) and over 18 times as many as TASSEL-UNEAK, with fewer genotyping errors in all cases, as evidenced by comparing the genotypic characterizations of biological replicates. Using the published reference genome of a related diploid species (A. chinensis), the reference-based version of GBS-SNP-CROP behaved similarly to TASSEL-GBS in terms of the number of SNPs called but had an improved read depth distribution and fewer genotyping errors. Our results also indicate that the sets of SNPs detected by the different pipelines above are largely orthogonal to one another; thus GBS-SNP-CROP may be used to augment the results of alternative analyses, whether or not a reference is available.

Conclusions: By achieving high-density SNP genotyping in populations for which no reference genome is available, GBS-SNP-CROP is worth consideration by curators, researchers, and breeders of under-researched plant genetic resources. In cases where a reference is available, especially if from a related species or when the target population is particularly diverse, GBS-SNP-CROP may complement other reference-based pipelines by extracting more information per sequencing dollar spent. The current version of GBS-SNP-CROP is available at https://github.com/halelab/GBS-SNP-CROP.git.

Fig. 1

Schematic of the four stages of the SNP-GBS-CROP workflow

Fig. 2

Structure of the final SNP genotyping matrix. As shown here, the GBS-SNP-CROP final genotyping matrix contains summary statistics as well as complete genotype-specific alignment data for each SNP called. The cells in red represent instances in which a genotypic state could not be assigned, either due to insufficient read depth (-|0/4) or a read depth ratio outside of the user-specified acceptable range (-|132/5)

Fig. 3

Bar plot showing the extent of marker overlap among the five evaluated pipelines. The sets of SNPs called by the five pipelines are largely orthogonal to one another, as shown by the fact that both the reference-based and reference-independent pipelines call high proportions of SNPs called by no other pipeline (grey bars). Shared SNPs among pipelines are indicated by color-coordinated bars. Whereas only 0.6 and 0.4 % of the 8,907 and 5,593 SNPs called by TASSEL-GBS-64 and TASSEL-32, respectively, were identified by TASSEL-UNEAK, 33.7 % of the SNPs called by GBS-SNP-CROP-RG were called by GBS-SNP-CROP-MR01