147 A novel single-cell method using optical barcodes and machine learning to identify correlative biomarkers predictive of T cell function

Le résumé fourni par la source

Background Cell therapy has been successful for treating some hematological malignancies, largely through autologous chimeric antigen receptor (CAR) T-cell therapies. However, the complexity of CAR T-cell therapies makes them costly and time-consuming to develop. One of the major challenges is heterogeneity in CAR-T products due to underlying variability across the entire development process and the lack of correlative biomarkers in existing in vitro models that can predict treatment efficacy and performance. In this study, we describe a novel flow cytometric approach using optical barcodes, called laser particles (LPs), to identify, track and characterize activated T-cells over several time points (Time-lapse flow cytometry). This method enables kinetic phenotyping, identification of cells defined by dynamic changes in one or more markers over time. We also demonstrate how connecting these kinetic phenotypes to single-cell function data can be used in a supervised machine learning (ML) approach to identify biomarkers predictive of future cell function. These biomarkers can then be leveraged by cell therapy developers to better understand the heterogeneity of their samples and improve the correlation of in vitro models to in vivo performance. Methods T cells from healthy donors were barcoded and incubated with an anti-CD3/CD28 activator analyzed by flow cytometry at 24, 32, and 48 hours of culture. The flow cytometry panel consisted of memory (CD45RA, CCR7, CD4, CD8) and activation markers (CD25, PD-1, CD69). After 48h, cells were stimulated with PMA/Ionomycin for 4 hours and stained intracellularly for IL-2, TNFa, and IFNy and acquired. Data for each cell at each time point were integrated using optical barcodes and analyzed with manual gating and supervised ML (decision tree algorithm). Results Using time-lapse flow cytometry, we identified novel kinetic phenotypes of T cells based on marker expression kinetics during activation (figure 1). With our ML approach, we predicted cells that secrete cytokine with an AUC of >0.85. We found that the main contributors to predicting this functional outcome were CCR7 and CD25 at 24 hrs as well as PD1 at 48 hr. This is likely the first ever demonstration of kinetic measurements of individual immune cells over time at scale. Conclusions In the future, researchers may be able to enrich for specific cells of interest based on their initial or kinetic phenotypes and their corresponding likelihood to exert specific functions at a later time point.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.