Computational Epigenomics to Study Pancancer Etiology 

Epigenomic mechanisms such as DNA methylation and histone methylation are important processes known to determine cell type and differentiation state and are often deregulated in cancer. We hypothesize that defining cancer subtypes based on DNA methylation changes in cancer tissues vs. normal tissues is a promising approach to study cancer etiology. We are developing computational epigenomics techniques such as MetylMix that elucidate DNA methylation changes in cancer vs. normal tissues. We will improve MethylMix by integrating a method to correct for cell-type heterogeneity or cancer subtyping and expand subtyping to a wide range of cancers. We will associate these subtypes with clinical outcome data, existing subtypes, and gene expression subtypes.

Final Grant Report (January 2020)

Major new insights:

  • DNA methylation has effects on both the transcriptome and the proteome
  • Deconvolution of DNA methylation data is a complement to RNA sequencing
  • DNA methylation is a stable signal that can be used to diagnose and subtype cancer patients.
  • Methylation signatures can be a complement to mutation signatures, and capture additional etiologies of cancer patients, especially in squamous carcinomas.

Furthermore, we found that DNA methylation signatures are reflected in digital pathology and radiographic images. First, we found that DNA methylation states of individual genes, and groups of genes are associated with cellular architecture as captured by digital pathology. This opens up many opportunities for capturing DNA methylation signals from H&E images, which is a routine data modality that is available for cancer patients, in particular also for large retrospective cohorts (Zheng et al. npj Genomic Medicine 2020). Secondly, we found that DNA methylation signals are also represented in radiographic images (Huang et al. EBioMedicine 2019, Mukherjee et al. Radiology: Cancer Imaging 2020). Taking together we posit that DNA methylation is an important way to define cancer subtypes, it reflects important diagnostic and treatment-related signals. Secondly, some of these signals might be captured from digital pathology and radiology.

This work will be leveraged and advanced as it will be incorporated in new grant applications.

Additional links to papers relating to this project:
Whole slide images reflect DNA methylation patterns of human tumors
The impact of DNA methylation on the cancer proteome