Studying differential isomiRs in high-throughput sequencing data identifies miRNA end-reads as a novel, putative miRNA degradation product

With the advancement of deep-sequencing technologies, it has become clear that individual miRNAs show varying degrees of heterogeneity in the expressed mature sequence. Recent evidence has suggested isomiR variants may be biologically relevant, and that small sequence variations can affect e.g. miRNA stability and RISC loading. The goal of this work was to investigate and characterize such isomiRs to determine whether isomiRs of the same miRNA show biologically relevant differ- ences. The work can further be divided into two sub-studies. I first present a support-vector machine classifier for predicting expression changes for isomiRs in Ago2-knockout cells. The classifier is constructed from miRNA sequence features, and is trained and tested using the observed read data for individual isomiRs, thus handling the possibility that highly similar isomiRs can experience different expression changes. The classifier is not successful, sug- gesting expression changes in Ago2-knockout is determined by other factors. I also look more closely at whether there are isomiR variants that experience sig- nificantly different expression changes, but find no strong evidence of this. Working with the Ago2-knockout sequencing data, a second line of work was set off from the discovery of a group of short ∼ 10 nt reads that align to the 3 end of mature miRNAs. I initially searched for such reads with the goal of finding products of Ago2 cleavage, but interestingly the reads are found in both wildtype and Ago2-knockout samples. I present an analysis of these reads and their corresponding miRNAs, and suggest they are produced by a previously undescribed regulated miRNA degradation process. In light of this, I also study the project data for isomiRs with non-templated 3 A/U-tails, which previously has been reported to affect miRNA degradation. I find both A- and U-tailing is common in the studied samples, but find neither a strong correlation nor a lack of overlap between miRNAs targeted by 3 tailing and miRNAs with corresponding short 3 -aligning reads