RECOMB-CCB

Integrating spatially-resolved multi-omic data to compute the dynamics and drivers of cancer evolution

Cancers evolve. Typically we cannot directly observe the evolutionary dynamics and must rely on mathematical inference to impute the dynamics from single-timepoint data. I will discuss how spatial patterns of genetic heterogeneity across a cancer encode the evolutionary dynamics of cancer growth, how we can interpret these patterns with computational models, and how we can overlay other ‘omics (principally RNAseq, ATACseq) to determine the molecular events that drive cancer evolution. Applying these methodologies to colorectal cancer provides evidence of a causal role for epigenetic changes, prompts reassessment of the role of mutations in some putative driver genes, and shows that phenotypic plasticity is commonplace.

The quest for epigenetic drivers of tumor evolution

My presentation will focus on the role of DNA methylation changes in contributing to carcinogenesis and the use of computational strategies to identify causative driver events. While numerous (point-) mutated driver genes have been identified across various cancer types, the contribution of epigenetic alterations to cancer remains more enigmatic. Certain regions of the cancer genome are more susceptible to dysregulated DNA methylation due to their genomic and epigenomic properties. However, distinguishing between promoter methylation changes that drive carcinogenesis and those that are simply a reflection of genomic location has been challenging. Additionally, identifying regions where epigenetic changes have an impact on transcription is critical in identifying epigenetic driver events. To address these challenges, we have developed methods for identifying regions with transcriptional impact that are under selection for undergoing DNA methylation changes in cancer. In my presentation, I will show how identifying and characterizing these regions can provide insights into phenomena such as colon cancer subtype formation at specific anatomic sites and offer new perspectives into tumors with few known genetic driver events.

Kernel Methods for Discovering and Leveraging Interactions for Cancer 

The talk will focus on the use of some recent methods we developed that rely on kernel methods. The first part will focus on detecting cancer subgroups using graph kernels. I will describe our kernel multiview clustering approach, PAMOGK, that integrates multi-omics patient data with existing biological knowledge on pathways. The method uses a novel graph kernel that evaluates patient similarities based on a single molecular alteration type in the context of a pathway. To corroborate multiple views of patients evaluated by hundreds of pathways and molecular alteration combinations, we use multiview kernel clustering. The second part of the talk will focus on discovering interactions between both coding and noncoding RNAs, where we use kernel statistical interaction tests. One interaction type is sponge-long non-coding RNAs (lncRNAs). LncRNAs can indirectly regulate mRNAs expression levels by sequestering microRNAs (miRNAs), and act as competing endogenous RNAs (ceRNAs) or as sponges. We use statistical interaction tests to discover lncRNAs in breast cancer subtypes. Another type of interaction we focus on is synergistic interactions between miRNAs. Single miRNAs typically induce mild repression on their targets. Cooperativity among miRNAs is reported as one strategy to overcome this constraint. I will describe the method miRCoop that we developed that makes use of kernel interaction tests. We recently applied our approach to patient data of 31 different cancer types, and the pan-cancer analysis report shared synergistic partners across cancers. 

Developing quantitative experimental model systems to study drug resistance in cancer

It is widely known that cancer drug resistance inevitably leads to treatment failure through clonal selection. Therefore, quantifying the clonal evolution using experimental model systems can hold a great promise in designing evolutionarily informed therapies, and thus, in predicting drug response. In this talk, I will present our quantitative experimental model system that contributed to the understanding of collateral drug sensitivity with its direct link to overcome the drug resistance in cancer. More specifically, high-complexity cellular barcoding allowed us the identification of the resistance mechanisms that was ultimately driven by the presence and emergence of multiple pre-existing and de novo resistant barcodes. Overall, our work highlighted evolutionary trade-offs and provided an opportunity to exploit the tumour’s vulnerabilities. Finally, I will present establishment of molecularly characterized Patient-Derived Organoids (PDOs) from metastatic colorectal cancer patients.