Understanding microRNA targeting with high-throughput biochemistry
Author(s)
McGeary, Sean E.(Sean Edward)![Thumbnail](/bitstream/handle/1721.1/130661/1251767216-MIT.pdf.jpg?sequence=4&isAllowed=y)
Download1251767216-MIT.pdf (65.00Mb)
Other Contributors
Massachusetts Institute of Technology. Department of Biology.
Advisor
David P. Bartel.
Terms of use
Metadata
Show full item recordAbstract
MicroRNAs (miRNAs) are short RNAs that, in complex with Argonaute (AGO) proteins, guide repression of mRNA targets. miRNAs negatively regulate most mammalian mRNAs, and disruption of this regulation often results in severe defects at the cellular and organismal level. miRNA repression occurs primarily through base-pairing between the miRNA seed region (nucleotides 2-8) and mRNA 3'-UTR sites, leading to transient recruitment of mRNA-destabilizing factors. However, only a small fraction of the gene-expression changes caused by a miRNA can currently be predicted, which precludes a deeper understanding of how miRNA regulation impacts the animal transcriptome. miRNA targeting efficacy should in principle be a function of the affinity between AGO-miRNA complexes and their targets. However, only a few such measurements had been reported, with measured values differing from those predicted for RNA-RNA pairing in solution. We therefore adapted a high-throughput biochemical platform utilizing random-sequence RNA libraries to obtain the vast quantity of affinity values required to predict miRNA targeting efficacy. Through a novel analytical approach, we assigned relative dissociation (K[subscript D]) constants to all binding sites </-12 nt in length, for six miRNAs. These analyses revealed unanticipated miRNA-specific differences in the affinity of similar sites, unique sites for different miRNAs, and a 100-fold influence of flanking dinucleotide context surrounding a site. These measurements informed a biochemical model of miRNA targeting that outperformed all existing models of miRNA targeting, which was extended to all miRNAs using a convolutional neural network (CNN) trained on both affinity and repression data. We also applied this high-throughput biochemical approach to understand the role of the miRNA 3' region using partially random RNA libraries. We found unique 3'-pairing preferences for each miRNA, and evidence for two distinct binding modes. The miRNA-specific differences and two binding modes depended on G nucleotides in the miRNA 3' region, thus providing a heuristic by which to extend these findings to target prediction in vivo. This work establishes high-throughput biochemistry combined with mathematical modeling and deep learning as a powerful paradigm for building quantitative models of gene regulation, which might aid in eventually building a complete model of the cell.
Description
Thesis: Ph. D., Massachusetts Institute of Technology, Department of Biology, February, 2021 Cataloged from the official PDF of thesis. Vita. Includes bibliographical references.
Date issued
2021Department
Massachusetts Institute of Technology. Department of BiologyPublisher
Massachusetts Institute of Technology
Keywords
Biology.