A Billion Reads Assembly: Building a Comprehensive Reference Transcriptome for Bread Wheat

A standard pipeline for RNA-seq analysis (differential gene expression or polymorphism discovery) requires the short reads to be mapped to a reference. De novo transcriptome assembly provides a cost-effective approach in generating a reference unigene set for species without sequenced genomes. In order to capture all transcripts, it is important to sample as many tissues, developmental stages and conditions as possible. Within a variety of projects involving bread wheat, we have sequenced a diverse set of 1.6 billion single-end (90%) and paired-end (10%) reads from different cultivars (Cadenza, Riband, Bobwhite, Hereward, April Bearded, Mercia, Holdfast, Yumai34), tissues (endosperm, grain, leaf, root, rachis and stem), developmental stages (leaf and grain development) and stress conditions (biotic and abiotic). In this study, we pooled these sequences and used digital normalisation (Brown et al., 2012) to discard redundant short reads and to decrease the run time requirements of the assembly. Two state-of-the-art assemblers, Trinity and Oases which are based on single and multiple k-mer methods, respectively, were used on a high memory server (2 TB RAM) and the effect of different parameter settings was studied. The assembly quality (coverage, redundancy and full-length) was assessed using different criteria such as mapping rate of normalised reads to the assembly, number of uniquely mapped reads and alignment coverage to cDNA sequences from wheat, barley, Brachypodium and rice.