Abstract P22: Decoding ecDNA in Breast Cancer: From Patient-Derived Models to AI-Enhanced Clinical Detection

Abstract

Extrachromosomal DNA (ecDNA) drives oncogene amplification and therapy resistance across cancers, yet its role in breast cancer requires more comprehensive mechanistic understanding. We developed comprehensive ecDNA breast cancer models including patient-derived cell lines, xenografts (PDX), and organoids (PDO) from tumor specimens, characterizing them through whole-genome sequencing to map ecDNA profiles and copy number variation/structural variant landscapes, revealing distinct ecDNA patterns within our cohort. Analysis of TCGA public datasets comparing expression profiles between ecDNA-positive and ecDNA-negative groups demonstrated that ecDNA-positive breast cancers significantly downregulate immune-activating pathways while upregulating mitosis and cell-cycle pathways, indicating ecDNA's dual impact on immune evasion and enhanced replicative capacity. These findings establish ecDNA as a critical driver of breast cancer progression through coordinated regulation of immune suppression and cellular proliferation mechanisms. To democratise ecDNA detection in breast cancer, we validate the use of AI-based image processing to predict ecDNA status from routine H&E histopathology slides. Our pilot study achieved 82.4% accuracy and 60% precision. The promising results obtained highlights the need for enhanced model fine-tuning specifically for breast cancer and local population characteristics. This integrated experimental-computational framework establishes ecDNA as a therapeutically targetable driver in breast cancer while pointing to possible implementations of ecDNA testing in clinical practice.