(502c) DEEP Phenotyping of Immune CELLS to Distinguish Disease States in EARLY-STAGE Breast Cancer

Background: Liquid biopsy, a non-invasive prognostic tool for monitoring disease states, holds significant promise for personalized and precision medicine in breast cancer management. Conventionally, it has concentrated on investigating various cancer-associated cells using specific biomarkers, with current research predominantly centered around the study of rare circulating tumor cells (CTCs). Here, we propose redirecting focus towards analyzing immune cell phenotypes, which are abundant and express several biomarkers. Our aim is to build a deep phenotyping framework capable of distinguishing disease states using immune cell biomarkers from individuals enabling personalized disease monitoring.

Methodology: To achieve our goal, we developed a computer vision framework to detect immune cells expressing the DAPI, CD45, and Vimentin biomarkers from immunostained images of peripheral blood of early-stage breast cancer patients. We extracted phenotypic features comprising immune cell morphology and biomarker expression, and used these multi-dimensional datasets to distinguish healthy and diseased patients. Datasets comprising 19 blood samples, 800,000 immune cells and 9 million phenotypic data points were used to conduct this study.

Results: Using a lower-dimensional representation of the extracted phenotypic features, we find clear distinction in data clusters associated with healthy, stage II and stage III breast cancer patients. To understand the phenotypic features that cause this distinction, we conducted statistical analysis between patient cohorts by comparing the variations observed with each biomarker. Our investigation highlighted the most notable differences in the nucleus of immune cells from diseased patients compared to those from healthy individuals, with morphological and biomarker expression features showing significant disparities [50% decrease & 100% increase in size & biomarker expression, respectively, p < 0.05]. Leveraging these insights, we quantitively predicted the disease states, with an accuracy > 80%, based on the phenotypic nature of the cells using logistic regression.

Conclusion: Our study reports a phenotypic framework to characterize immune cell phenotypes present in the liquid biopsy of early-stage breast cancer. We believe this innovative approach broadens the scope for clinical utility of liquid biopsy for precision medicine, paving the way for effective targeted therapeutic interventions in non-invasive, blood-based diagnostics.