Detection and monitoring of treatment resistance in metastatic cancer through non-invasive DNA methylation-based biomarkers

This work focuses on empowering plasma circulating tumor DNA (ctDNA) analysis, as a non-invasive approach for precision oncology, with applications ranging from early detection and minimal residual disease (MRD) assessment to dynamic monitoring of treatment response and molecular profiling in advanced disease. Low ctDNA plasma concentrations, relevant to many clinical scenarios, as early-stage disease and disease monitoring during initial lines of metastatic treatment, represent a major challenge in ctDNA applications. Despite several approaches for ctDNA analysis have been developed in the last decade, tumor tissue agnostic detection and quantification of ctDNA in low tumor content (TC) scenarios remain a significant challenge. The molecular, biological, and clinical heterogeneity of breast cancer (BC), the main tumor type investigated throughout this research, is examined, along with the complexities of BC subtype classification and contingent therapeutic strategies. Particular attention is given to the emergence of resistance mechanisms and tumor subtype switching, particularly hormone receptors (HR) expression loss, as additional challenges in managing metastatic BC (mBC), highlighting the potential of ctDNA analysis for monitoring tumor phenotypic changes, especially in cases with multiple or inaccessible metastatic sites. The main contribution of this thesis is the introduction and rigorous validation of a robust and reproducible strategy combining low-pass (0.5-1X) whole genome bisulfite sequencing (lpWGBS) of plasma DNA with METER, a novel computational tool, designed for low TC scenarios in metastatic setting. Exploiting tumor-type specific DNA-methylation patterns informed from independent tumor tissue datasets, METER performs a comprehensive analysis of ctDNA, enabling detection and quantification of TC and molecular cancer subtyping. Validation in two independent mBC cohorts, comprising 338 longitudinal plasma methylomes from 124 patients, demonstrated accurate TC quantification and sensitive ctDNA detection, even at low TC levels (<3%), undetectable by ichorCNA, the state-of-the-art tool for TC analysis in low-pass sequencing data. Importantly, ctDNA detection through METER demonstrated clear clinical relevance also in low-TC cases, as reflected by its strong correlation with matched circulating tumor cells (CTC) and its ability to anticipate patient outcomes earlier than imaging, independently of other prognostic indicators. Additionally, METER enabled accurate, non-invasive inference of estrogen receptor (ER) status, providing a proof-of-concept for ctDNA-based phenotypic profiling through lpWGBS. To evaluate its generalizability and clinical potential in multiple tumor types, METER was then applied to a metastatic colorectal cancer (mCRC) cohort (VALENTINO trial), including 302 longitudinal plasma methylomes from 158 patients. METER demonstrated technical robustness and confirmed sensitive ctDNA detection, and early clinical outcome prediction. While METER offers a methodological framework for non-invasive ER status monitoring in mBC, a deeper understanding of the molecular mechanisms underlying HR shifts, particularly HR loss, remains essential. Given the relevance of HR loss in mBC, involving ER and/or progesterone receptor (PR), this thesis integrates a complementary genomic analysis, exploiting HER2– mBC patients’ samples from the MSK-2018 dataset, to explore the genomic features associated with this phenotypic shift. ER loss was associated with a genomic context intermediate between ER+ and ER– tumors. In addition, HR loss emerged as a distinct mechanism of metastatic progression, divergent from the acquisition of ESR1 mutations. These findings provide a foundation for future investigations aimed at better characterizing the molecular drivers of HR loss and identifying new targets to personalize treatment for patients with these tumors. In conclusion, this work offers a robust tumor tissue-agnostic alternative to targeted or sophisticated machine learning approaches for comprehensive ctDNA analysis in precision oncology, particularly in low TC metastatic settings, with significant potential for future optimization in early disease contexts. Applicability of the approach to multiple cancer types (i.e. mBC and mCRC) is also proposed. Complementary findings on genomic features linked to HR loss, while suggesting potential insights to inform ctDNA-based HR status monitoring, further underscore the biological complexity of metastatic progression and the need for deeper characterization of this phenomenon and its therapeutic implications.