| dc.description.abstracteng | RNA-binding proteins (RBPs) impact every aspect of RNA metabolism including RNA transcription, maturation, export, localization, translation, and stability. Specific RNA-protein interactions therefore play a central role in regulating many cellular processes. However, most RBPs preferentially bind short, often degenerate sequence motifs (∼3-5 bases) that alone cannot explain how they target only specific subsets of transcripts in the cell. In this thesis, I report on the analysis and the thermodynamic modeling of RNA-protein interaction datasets, with the aim of cracking the code behind RBP specificity.
In the first part of my dissertation, I examine RBPs involved in the general eukaryotic RNA degradation pathway. We generated transcriptome-wide maps of RNA-protein interactions in yeast for 30 yeast RNA decay factors using photoactivatable ribonucleoside-enhanced cross-linking and immunoprecipitation (PAR-CLIP). In-depth bioinformatic analysis revealed that the decay machineries responsible for degradation of the two RNA ends differ in their substrate specificity. We identified TRAMP4 and exosome as the main complexes involved in Nrd1/Nab3 mediated RNA degradation. Moreover, modeling the dependence of mRNA half-life on degradation factor binding suggested that the recruitment of decapping factors happens only upon RNA degradation, while other decay factors may already associate with mRNAs earlier for their surveillance. Furthermore, global comparison of RNA-binding profiles of decay factors with those of other RNA processing proteins indicated many functional associations with the decay factors.
In the second part of this thesis, I introduce Bipartite Motif Finder (BMF), a computational tool that adopts thermodynamic modeling for the discovery of multivalent RNA-protein interactions. Many RBPs have multiple domains that allow them to target multiple short RNA sequences simultaneously in a cooperative manner, others may achieve cooperativity through oligomerization. This results in specificities and affinities that can be many orders of magnitude higher than those possible by single-domain binding events. Yet, previously available motif discovery approaches have not taken this cooperativity into account. BMF takes full account of the cooperativity and calculates binding probabilities by the weighted sum of all binding configurations determined through thermodynamic modeling. By applying BMF on a high- throughput RNA SELEX (HTR-SELEX) dataset of 78 RBPs, we show that bipartite binding is widespread and that the two motif cores are often similar and low in sequence complexity. We also show that BMF can learn the spatial geometry between the binding sites and predict new RBP binding sites in transcripts with an accuracy competitive with existing motif discovery approaches. We made BMF easily accessible for computationally inexperienced users via the web server (https://bmf.soedinglab.org). BMF source code is also available under a GPL license (https://github.com/soedinglab/bipartite_motif_finder). | de |