Methods for Transcriptome Assembly in the Allopolyploid Brassica napus

Canada is the world’s largest producer of canola and the trend of production is ever increasing with an annual growth rate of 9.38% according to FAOSTAT [1]. In 2017, canola acreage surpassed wheat in Saskatchewan, the highest producer of both crops in Canada. Country-wide, the total farming area of canola increased by 9.9% to 22.4 million acres while wheat area saw a slight decline to 23.3 million acres [2]. While Canada is the highest producer of the crop, yields are lower than other countries [1]. To maximize the benefit of this market, canola cultivation could be made more efficient with further characterization of the organism’s genes and their involvement in plant robustness. Such studies using transcriptome analysis have been successful in organisms with relatively small and simple genomes. However, such analyses in B. napus are complicated by the allopolyploid genome structure resulting from ancestral whole genome duplications in the species’ evolutionary history. Homeologous gene pairs originating from the orthology between the two B. napus progenitor species complicate the process of transcriptome assembly. Modern assemblers: Trinity [3], Oases [4] and SOAPdenovo-Trans [5] were used to generate several de novo transcriptome assemblies for B. napus. A variety of metrics were used to determine the impact that the complex genome structure has on transcriptome studies. In particular, the most important questions for transcriptome assembly in B. napus were how does varying the k-mer parameter effect assembly quality, and to what extent do similar genes resulting from homeology within B. napus complicate the process of assembly. These metrics used for evaluating the assemblies include basic assembly statistics such as the number of contigs and contig lengths (via N25, N50 and N75 statistics); as well as more involved investigation via comparison to annotated coding DNA sequences; evaluation softwares scores for de novo transcriptome assemblies and finally; quantification of homeolog differentiation by alignment to previously identified pairs of homeologous genes. These metrics provided a picture of the trade-offs between assembly softwares and the k-parameter determining the length of subsequences used to build de Bruijn graphs for de novo transcriptome assembly. It was shown that shorter k-mer lengths produce fewer, and more complete contigs due to the shorter required overlap between read sequences; while longer k-mer lengths increase the sensitivity of an assembler to sequence variation between similar gene sequences. The Trinity assembler outperformed Oases and SOAPdenovo-Trans when considering the total breadth of evaluation metrics, generating longer transcripts with fewer chimers between homeologous gene pairs.