Open Access

You Zhou

• RNA modification is a dynamic and complex process that involves the addition of various chemical groups to RNA molecules, contributing to their diversity and functional complexity. Among all the RNA modifications, N6-methyladenosine (m6A) is the most common post-transcriptional modification found in mRNA molecules, particularly in eukaryotic mRNA. It involves methylation of the adenosine base at the nitrogen-6 position. This modification plays a crucial role in many aspects of RNA metabolism, including splicing, stability, translation, and the cellular response to stress. With the development of m6A sequencing technologies, our knowledge of m6A has evolved rapidly over the past two decades. However, one of the most widely used m6A profiling techniques termed “m6A individual-nucleotide resolution UV cross-linking and immunoprecipitation (miCLIP)” suffers from a high unspecific background signal due to the limited antibody binding specificity. To accurately discriminate m6A sites from the background signal in miCLIP data, in Chapter 4, I first developed different strategies to identify the true miCLIP2 signal changes that are corrected for the underlying transcript abundance changes. I performed this analysis on data that generated with an improved experiment protocol, named miCLIP2. With the best performing strategy, the Bin-based method, I detected more than 10,000 genuine m6A sites. I then used the information embedded in the genuine m6A sites to train a machine learning model - named "m6Aboost" - to enable accurate m6A site detection from the miCLIP2 data without a control dataset from an m6A depletion cell line. To allow an easy access for future users, I packaged the m6Aboost model into an R package that is available on Bioconductor. Although previous studies have reported that m6A is involved in three different RNA decay pathways, it remains unclear how a pathway is selected for a specific transcript or m6A site. In Chapter 5, I reveal that m6A sites in the coding sequence (CDS) induce a stronger and faster RNA decay than the m6A sites in the 3’ untranslated region (3’UTR). Through an in-depth investigation, I found that m6A sites in CDS trigger a novel mRNA decay pathway, which I termed CDS-m6A decay (CMD). Importantly, CMD is distinct from the three previously reported m6A-mediated decay pathways. In terms of its mechanism, CMD relies on translation, where m6A sites in the CDS lead to ribosome pausing and subsequent destabilization of the transcript. The transcripts targeted by CMD are identified by the m6A reader protein YTHDF2, preferentially localized to processing bodies (P-bodies), and undergo degradation facilitated by the decapping factor DCP2. CMD provides a flexible way to control the expression of CDS m6A-containing transcripts which include many developmental regulators and retrogenes. In summary, this PhD thesis introduces a novel workflow for identifying m6A sites in miCLIP data through the implementation of the m6Aboost machine learning model. Using the m6A sites identified by m6Aboost and additional data, a newly uncovered m6A-mediated mRNA decay pathway, CMD, is elucidated, providing valuable insights into m6A-mediated decay processes.