Multimapping confounds ribosome profiling analysis: A case‐study of the Hsp90 molecular chaperone

Abstract

Ribosome profiling (Ribo‐seq) can potentially provide detailed information about ribosome position on transcripts and estimates of protein translation levels in vivo. Hsp90 chaperones, which play a critical role in stress tolerance, have characteristic patterns of differential expression under nonstressed and heat shock conditions. By analyzing published Ribo‐seq data for the Hsp90 chaperones in S. cerevisiae, we find wide‐ranging artifacts originating from “multimapping” reads (reads that cannot be uniquely assigned to one position), which constitute ~25% of typical S. cerevisiae Ribo‐seq datasets and ~80% of the reads from HEK293 cells. Estimates of Hsp90 protein production as determined by Ribo‐seq are reproducible but not robust, with inferred expression levels that can change 10‐fold depending on how multimapping reads are processed. The differential expression of Hsp90 chaperones under nonstressed and heat shock conditions creates artificial peaks and valleys in their ribosome profiles that give a false impression of regulated translational pausing. Indeed, we find that multimapping can even create an appearance of reproducibility to the shape of the Hsp90 ribosome profiles from biological replicates. Adding further complexity, this artificial reproducibility is dependent on the computational method used to construct the ribosome profile. Given the ubiquity of multimapping reads in Ribo‐seq experiments and the complexity of artifacts associated with multimapping, we developed a publicly available computational tool to identify transcripts most at risk for multimapping artifacts. In doing so, we identify biological pathways that are enriched in multimapping transcripts, meaning that particular biological pathways will be highly susceptible to multimapping artifacts.

Translation initiation downstream from annotated start codons in human mRNAs coevolves with the Kozak context

“…In addition, the expression levels of reporter constructs differ from that of endogenous mRNAs, and this may also lead to differences. Although ribosome profiling has no such limitations, it is subject to several artifacts (Hussmann et al 2015;O'Connor et al 2016;Gerashchenko and Gladyshev 2017;Halpin et al 2020) that may cause distortions between the real distribution of the ribosomes and the observed distribution of their footprints obtained with this technique. We also cannot exclude the possibility that the differences between the two approaches reflect the real biological difference between cell lines in which the experiments were executed.…”

Inferring translational heterogeneity from Saccharomyces cerevisiae ribosome profiling

“…RD profiles of unambiguously mapped reads were computed with Scikit‐ribo [44]. Genes with an average coverage of less than one read per position across datasets, or with more than 30% multimapping (MM), that is, the fraction of mRNA sequences that yields ambiguous read mapping [111], were omitted. This yielded a final dataset of 2443 yeast genes.…”

Discovering microproteins: making the most of ribosome profiling data

“…This typically results in as low as only ~ 10% [ 35 ] of the sequenced reads to be usable. From the remaining reads about ~ 20–80% are reported as uniquely mapped to the transcriptome in various studies, considering the short read length and varying data quality [ 36 ]. Apart from these pre-processing steps, QC such as length distribution and 3-nt periodicity is also tested.…”