Predicting cancer subtypes from nucleosome profiling of cell-free DNA

Abstract

Modern cancer treatments take advantage of genomic differences between tumors to kill cancer cells using targeted approaches. Typically, this requires a tumor biopsy in order to get tissue for phenotypic and genotypic analysis. However, in late-stage cancer, surgical biopsies of metastases may not be part of the standard of care and repeated biopsies cannot be performed for monitoring. However, late-stage cancer patients can benefit from targeted therapies that address specific tumor phenotypes and resistance mechanisms. Because of this, non-invasive diagnostic approaches are needed. Cell-free DNA (cfDNA) provides a promising approach for non-invasive tumor characterization. In cancer patients, a fraction of cfDNA is derived from tumor cells which have died and released their DNA into the bloodstream. This DNA remains wrapped around nucleosomes which protect it from degradation by apoptotic and plasma nucleases. cfDNA can be extracted from the blood and sequenced, revealing which regions of the genome were protected by nucleosomes and which were more accessible in the tumor cells. This type of analysis, known as nucleosome profiling, has the potential to be used for phenotypic characterization of tumors, such as determining the activity of key transcription factors. In this thesis, we develop methods to perform nucleosome profiling in cfDNA from cancer patients. First, we quantify the effects of GC bias on nucleosome profiles and implement a GC correction procedure to reduce these impacts. Next, we develop a nucleosome profiling method called Griffin that uses this GC bias correction procedure and generates composite nucleosome profiles around transcription factor binding sites (TFBSs) and other accessible sites. We then apply this method to detect cancer in early-stage cancer patients. Finally, we identify assay for transposase-accessible chromatin using sequencing (ATAC-seq) sites that are differentially accessible in estrogen receptor (ER) positive or ER negative metastatic breast tumors and use Griffin nucleosome profiling around these sites to predict the ER status in cfDNA samples from breast cancer patients. Additionally, in a separate study, we design a targeted sequencing panel to predict the activity of lineage defining transcription factors in small-cell lung cancer (SCLC) and use this approach to predict SCLC subtypes from cfDNA.