Genomic limitations to RNA sequencing expression profiling

Abstract

SUMMARYThe field of genomics has grown rapidly with the advent of massively parallel sequencing technologies, allowing for novel biological insights with regards to genomic, transcriptomic, and epigenomic variation. One widely utilized application of high-throughput sequencing is transcriptional profiling using RNA sequencing (RNAseq). Understanding the limitations of a technology is critical for accurate biological interpretations, and clear interpretation of RNAseq data can be difficult in species with complex genomes. To understand the limitations of accurate profiling of expression levels we simulated RNAseq reads from annotated gene models in several plant species including Arabidopsis, brachypodium, maize, potato, rice, soybean, and tomato. The simulated reads were aligned using various parameters such as unique versus multiple read alignments. This allowed the identification of genes recalcitrant to RNAseq analyses by having over-and/or under-estimated expression levels. In maize, over 25% of genes deviated by more than 20% from the expected count values, suggesting the need for cautious interpretation of RNAseq data for certain genes. The reasons identified for deviation from expected expression varied between species due to differences in genome structure including, but not limited to, genes encoding short transcripts, overlapping gene models, and gene family size. Utilizing existing empirical datasets we demonstrate the potential for biological misinterpretation resulting from inclusion of 'flagged genes' in analyses. While RNAseq is a powerful tool for understanding biology, there are limitations to this technology that need to be understood in order to improve our biological interpretations.

Draft Assembly of Elite Inbred Line PH207 Provides Insights into Genomic and Transcriptome Diversity in Maize

“…This observation is further supported using the B73 gene atlas, which includes transcript abundance estimates for 79 B73 tissues throughout development (Supplemental Figure 5; Stelpflug et al, 2016). Small transcripts (<300 bp) are undersampled and therefore have lower estimated transcript abundance due to technical biases resulting from size selection during library preparation (Hirsch et al, 2015). However, based on the distribution of transcript size for genotype-specific and shared genes, the lower observed expression of genotypic-specific genes was not a product of this bias ( Figure 4C).…”

“…There are many biases that exist with RNA-seq for transcript abundance estimates when using a single reference genotype (Hirsch et al, 2015); these biases are further confounded outside of the reference genotype by the alternative gene model structures and partial gene deletions that exist between individuals within a species. For example, transcript abundance estimates for 15.9% of genes were highly influenced by deletion of more than 25% of the gene model in the opposite genome, and 20.5% were falsely identified as differentially expressed in one or more tissues based on this bias.…”

“…Within the genotype that had the expanded family, on average 39.7% of the genes within the family were expressed in at least one of six tissues tested (leaf blade, root cortical parenchyma, root stele, germinating kernel, root tip, and whole seedling) when requiring reads to align uniquely. This number is likely an underestimation of the true frequency of expression within these families due to the limitations of unique alignments with short reads in large gene families (Hirsch et al, 2015). Across the families that showed large changes in gene family size, an enrichment of Gene Ontology terms related to stress response was observed, among other functions (Supplemental Data Set 2).…”

The ‘TranSeq’ 3′‐end sequencing method for high‐throughput transcriptomics and gene space refinement in plant genomes

“…One noteworthy weakness of ‘standard’ RNA‐seq methods in which sequence reads cover the entire transcript is the erroneous assignment of reads in‐between highly related sequences such as members of the same gene family (Hirsch et al ., ). Plant genomes are exceedingly enriched in large gene families that often include very similar genes in tandem gene clusters, that could be part of a duplicated genomic region or spread out across the genome (Soltis et al ., ).…”

“…Yet, in terms of throughput, experimental setups reaching the hundreds and thousands of samples scale are laborious and expensive for most research laboratories. Furthermore, established RNA‐seq methods covering the entire length of transcripts frequently fail to accurately assign reads to gene family members that are the products of recent duplication events and share high sequence similarity (Hirsch et al ., ).…”