Abstract 3247: Liquid based methylation profiling for quantification of breast cancer subtypes

Background: Hormone receptor (HR) status, typically determined through immunohistochemistry (IHC), and human epidermal growth factor receptor 2 (HER2) status, assessed by IHC and FISH, are critical factors in guiding breast cancer treatment. However, in metastatic breast cancer, resistance to endocrine or anti-HER2 therapies often develops, and receptor status may change over time due to tumor evolution under selective treatment pressures. Both IHC and FISH require tissue biopsies, which can become increasingly difficult to obtain in later treatment stages. Moreover, HR and HER2 status may evolve as the tumor adapts to therapy, highlighting the need for ongoing monitoring. A non-invasive, blood-based method could provide a practical alternative, enabling real-time assessment of receptor status. DNA methylation offers a potential avenue for identifying subtype changes in breast cancer. This could serve as an early indicator of resistance to treatment. In this study, we investigated whether methylation-based signatures in cell-free DNA (cfDNA) could act as a surrogate to predict breast cancer subtypes. Methods: A novel cfDNA methylation-based deconvolution model was developed to quantify the proportional contributions of each breast cancer subtype in patient blood samples. The model leverages methylation signals across 4,500 cancer-associated genomic regions to estimate the contribution of each subtype while simultaneously predicting patient-specific subtype methylation signatures. A novel probabilistic model was developed to calculate the posterior probability of methylation for each region, based on hypermethylated molecule counts. To improve reliability, regions with high noise were filtered out, and penalized logistic regression models were created for each subtype using these posterior probabilities. Using a training set of 1,078 patients with breast cancer subtype derived from real-world data (HER2+ N=114, HR+ N=277, and TNBC N=71, with tumor fraction > 0.5%), logistic regression models were trained and optimized to predict HR, TNBC, and HER2 statuses. Results: Model performance was evaluated by comparing the predicted histological subtypes to the known histological subtype annotations of 366 breast cancer samples. The overall accuracy on the test set was 80.1%, with the highest accuracy for TNBC (90.5%), followed by HR+/HER2+ (88%), HR+/HER2- (80%) and HR-/HER2+ (80%). Importantly, the model demonstrated the ability to detect sub-type in samples with tumor fraction as low as 0.5%. Conclusions: Our methylation-based subtype deconvolution model provides a highly accurate method for predicting breast cancer subtypes from patient blood samples. This approach may enable the early non-invasive detection of growing treatment-resistance amongst breast cancer subtypes, with the potential to allow for more informed clinical decision-making and personalized treatment selection. Citation Format: Denis Tolkunov, Caroline Weipert, Sara Wienke, Derek Dustin, Michael Weindel, Catalin Barbacioru, Justin Odegaard, Darya Chudova. Liquid based methylation profiling for quantification of breast cancer subtypes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3247.