Search

Background: Cell-free DNA (cfDNA) is released from dying cells, including tumor cells, and can be isolated from peripheral blood for studying cancer. In the bloodstream, cfDNA is protected from degradation by nucleosomes and other DNA binding proteins, leading to a coverage pattern that reflects the genomic organization in the cells-of-origin. Recent work has shown that it is possible to use this pattern to predict gene and transcription factor activity in cancer cells. This is known as nucleosome profiling. Breast cancer is among the most common causes of cancer, accounting for 23% of cancer diagnoses and 14% of cancer-related deaths among women worldwide. Targeted therapy is guided by tumor subtype, including the expression of three key receptors: ER, PR and HER2. Typically, subtyping involves a tumor biopsy and immunohistochemistry. However, in late-stage cancer, surgical biopsies for disease monitoring are difficult to obtain. Accurate subtype determination is critical to address hormone subtype switches during metastasis or treatment resistance. cfDNA offers an alternative, non-invasive method for identifying tumor subtypes through nucleosome profiling and, to the best of our knowledge, has not been shown for breast cancer. Methods: We developed a method, called Griffin, to examine nucleosome protection and genome accessibility by quantifying cfDNA fragments around accessible sites. Unlike previous methods, Griffin uses fragment length-based GC correction to remove GC biases that obscure signals. We used ATAC-seq data from TCGA to identify differentially accessible sites between ER positive and negative breast cancers. We developed a machine learning classifier that predicts ER subtype based upon the signals at these differentially accessible sites. Results: We then tested Griffin by examining differentially accessible sites in ultra-low pass sequencing (ULP-WGS, 0.1X) of several hundred cfDNA samples from patients with ER positive or negative breast cancer. We found that overall, differential sites were more accessible in the cfDNA of their respective subtypes. Additionally, we found that site accessibility within patient cfDNA samples was correlated to the cfDNA tumor fraction. We built and tested a prediction model with cross-validation, which revealed an accuracy of >80% for correctly classifying tumor status as ER positive or negative from this ULP-WGS dataset. Conclusion: This study has several novel aspects compared to prior nucleosome profiling approaches. First, we use fragment-based GC correction which reduces sample variability and allows us to observe previously obscured signals. Second, we demonstrated that signals are correlated to tumor fraction. And finally, we applied this method to cost-effective and scalable ULP-WGS of breast cancer and demonstrated the ability to predict breast cancer ER subtype in these samples. Citation Format: Anna-Lisa Doebley, Hanna Liao, Caroline Kikawa, Eden Cruikshank, Minjeong Ko, Anna Hoge, Joseph Hiatt, Navonil De Sarkar, Viktor A. Adalsteinsson, Paz Polak, David MacPherson, Peter S. Nelson, Heather A. Parsons, Daniel Stover, Gavin Ha. Griffin: A method for nucleosome profiling and breast cancer subtype prediction from ultra-low pass whole genome sequencing of cell-free DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr LB022.

Tumours evolve through time and space. Computational techniques have been developed to infer their evolutionary dynamics from DNA sequencing data. A growing number of studies have used these approaches to link molecular cancer evolution to clinical progression and response to therapy. There has not yet been a systematic evaluation of methods for reconstructing tumour subclonality, in part due to the underlying mathematical and biological complexity and to difficulties in creating gold-standards. To fill this gap, we systematically elucidated the key algorithmic problems in subclonal reconstruction and developed mathematically valid quantitative metrics for evaluating them. We then created approaches to simulate realistic tumour genomes, harbouring all known mutation types and processes both clonally and subclonally. We then simulated 580 tumour genomes for reconstruction, varying tumour read-depth and benchmarking somatic variant detection and subclonal reconstruction strategies. The inference of tumour phylogenies is rapidly becoming standard practice in cancer genome analysis; this study creates a baseline for its evaluation.