Landscape of the Dark Transcriptome Revealed Through Re-mining Massive RNA-Seq Data

Abstract

The “dark transcriptome” can be considered the multitude of sequences that are transcribed but not annotated as genes. We evaluated expression of 6,692 annotated genes and 29,354 unannotated open reading frames (ORFs) in the Saccharomyces cerevisiae genome across diverse environmental, genetic and developmental conditions (3,457 RNA-Seq samples). Over 30% of the highly transcribed ORFs have translation evidence. Phylostratigraphic analysis infers most of these transcribed ORFs would encode species-specific proteins (“orphan-ORFs”); hundreds have mean expression comparable to annotated genes. These data reveal unannotated ORFs most likely to be protein-coding genes. We partitioned a co-expression matrix by Markov Chain Clustering; the resultant clusters contain 2,468 orphan-ORFs. We provide the aggregated RNA-Seq yeast data with extensive metadata as a project in MetaOmGraph (MOG), a tool designed for interactive analysis and visualization. This approach enables reuse of public RNA-Seq data for exploratory discovery, providing a rich context for experimentalists to make novel, experimentally testable hypotheses about candidate genes.

Accuracy of functional gene community detection inSaccharomyces cerevisiaeby maximizing Generalized Modularity Density

“…Here, we compare commonly-used methods for inferring functional gene communities from RNA-Seq expression data [32] from the model eukaryote, S. cerevisiae . First, using the expression profiles of S. cerevisiae genes, we obtained three different weighted co-expression networks, each by employing a distinct method for assigning link-weights between pairs for genes.…”

Origins, evolution, and physiological implications of de novo genes in yeast

“…These models postulated that some of the hundreds of and microscopy studies have detected the protein products of some of these noncanonical translation events in yeast cells (He et al, 2018;Lu et al, 2017;Yagoub et al, 2015), but the vast majority remain undetected. The de novo-translated ORFs include upstream ORFs and downstream ORFs (translated ORFs located upstream and downstream of annotated coding sequences, respectively) as well as ORFs translated from transcripts containing no annotated gene (Blevins et al, 2021;Carvunis et al, 2012;Durand et al, 2019;Li et al, 2021;Smith et al, 2014;Wacholder et al, 2021;Wilson & Masel, 2011). To what extent the noncanonical translatome generates an entirely novel proteome or yields rapidly degraded products that serve to regulate translation and transcript stability remains an open research question.…”

“…These characteristics are thought to derive directly from their de novo emergence and to be associated with possible physiological corollaries. For example, condition-specific expression of yeast de novo ORFs has been reported in the context of various stresses (Blevins et al, 2021;Carvunis et al, 2012;Doughty et al, 2020;Li et al, 2021;Wacholder et al, 2021). Could these species-specific translated elements represent a rapidly evolving part of the cell's response to stress?…”

“…Genome‐wide deletion and overexpression libraries have been developed for multiple yeast strains, particularly in S. cerevisiae (Alberti et al, 2007; Brachmann et al, 1998; Douglas et al, 2012; Fasanello et al, 2020; Gelperin et al, 2005; Giaever et al, 2002; McIsaac et al, 2013; Sopko et al, 2006), enabling advanced, high‐throughput approaches that can be expanded to characterize phenotypes for de novo candidates (Costanzo et al, 2010, 2016; Douglas et al, 2012; Parsons et al, 2006; Piotrowski et al, 2017; Vizeacoumar et al, 2010). Once a phenotype is detected with confidence, mechanisms can be inferred with many tools, for example, with deep mutational scanning (Fowler & Fields, 2014) or network‐based computational approaches (Li et al, 2021). As de novo ORFs often overlap with noncoding sequences that may function as regulatory elements or noncoding RNAs, it can be important to experimentally dissect which aspects of null mutant phenotypes are truly caused by loss of translation or loss of the protein product.…”

“…Yeast de novo ORFs, whether annotated or not, tend to share general characteristics: Their primary sequences tend to be very similar to those of intergenic ORFs, and they tend to be short, rapidly evolving, and often expressed in both lineage-specific and condition-specific manners (Basile et al, 2017;Blevins et al, 2021;Carvunis et al, 2012;Durand et al, 2019;Ekman & Elofsson, 2010;Li et al, 2021;Papadopoulos et al, 2021;Vakirlis et al, 2018;Wu & Knudson, 2018). These characteristics are thought to derive directly from their de novo emergence and to be associated with possible physiological corollaries.…”