Machine Learning-Enabled Droplet Microfluidics Reveals Functional Heterogeneity in NK Cell Immunotherapy

Abstract

Natural Killer (NK) cell immunotherapy faces challenges in predicting therapeutic efficacy due to functional heterogeneity within NK populations and tumor microenvironment (TME) suppression. Here, a droplet microfluidic platform enables automated single-cell analysis of NK cell-mediated cytotoxicity against cancer cells. A machine learning-based object detection model identified target cells and death events across image sequences and generated readouts. Distinct NK cells are evaluated to quantify key metrics, including the percentage of cytotoxic NK cells, serial killing capacity, killing time per target and NK-target attachment dynamics. The results demonstrated that expanded NK cells (exNK) exhibited superior cytotoxic activity, serial killing, and rapid killing dynamics, whereas peripheral blood NK cells (pbNK), especially when they were exposed to ascites TME (pbNK-asc), displayed reduced cytotoxic abilities in all parameters. Interestingly, expanded NK cells exposed to ascites TME (exNK-asc) retained partial functionality, indicating that expansion provides resilience against suppressive factors. This single-cell analysis provides novel insights into NK-cancer cell interactions, offering a robust framework for enhancing the efficacy of future immunotherapy applications especially for optimizing off-the-shelf NK cell-based immunotherapies.