Comprehensive investigation of epigenetic regulation during tumorigenesis and discovery of therapeutic applications

Abstract

Cancer, the second leading cause of death globally, is no longer seen as only a genetic disease. The landscape of cancer genome is shaped by both genetic and epigenetic alterations. Epigenetics includes chromatin regulators, microRNAs, and DNA methylation. Abnormal DNA methylation is a common epigenetic mechanism mediating tumor suppressor inactivation and genomic instability. And microRNAs play critical roles in oncogenesis by targeting key regulators or a large cohort of genes impinging on downstream signaling pathways. Compared to enetic changes, epigenetic alterations are naturally reversible, and miRNA activity can be titrated by competitive endogenous RNA such as lncRNA with sponge effect. These features make DNA methylation and miRNAs to be suitable targets for disease diagnosis and therapeutics. The study in this thesis focuses on the analysis of cancer gene regulatory network modulated by both DNA methylation and microRNAs. First, we developed a set of tools to identify genome-wide DNA methylation in regulatory elements with effects on tumorigenesis called MICMIC, "Methylation Regulation Network Inference by Conditional Mutual Information Based PC-algorithm". This method applied information theory to infer the direct regulation between epigenome and transcriptome. Many predictions were directly validated by dCas9-based epigenetic editing to support the accuracy and efficiency of our tool. Oncogenic and lineage-specific transcription factors were shown to aberrantly shape the methylation landscape by modifying tumor-subtype core regulatory circuitry. Secondly, we developed a web-based server, miRNACancerMap, aiming to unravel lncRNA-miRNA-mRNA tripartite complexity to predict the function and clinical relevance of miRNA with a network perspective. In conjunction with large-scale data and information integration, miRNACancerMap implemented various algorithms and pipelines to construct a dynamic miRNA-centered network with rigorous Systems Biology approaches and the state-of-the-art visualization tool. The capability of the server to generate testable hypotheses was exemplified with cases to identify hub miRNAs regulating most of the differentially expressed genes involved in cancer stage transition, miRNA-TF pairs shared by pan-cancers and lncRNA sponges cross-validated by independent datasets. Lastly, we developed a mathematical model, DEComb, which was based on the biological hypothesis without relying on any training data to predict the drug combination efficacy. We conducted multiple analyses to evaluate the prediction performance of our model. The results showed that DEComb achieved an AUC of 0.92 to predict synergistic drug pairs and outperformed other algorithms in literature. Integrated with the genetic algorithm, DEComb was enabled to identify the most effective drug pairs from billions of potential combinations within a short time period. These features empowered DEComb to analyze the large scale datasets from high-throughput drug perturbation profiling. Taken together, our study reveals the importance of regulatory network derived from DNA methylation and miRNAs in cancer development and provides multiple valuable bioinformatics software for cancer research community.