Single-Cell Transcriptomic Attributes and Unbiased Computational Modeling for the Prediction of Immunomodulatory Potency of Mesenchymal Stromal Cells

Abstract

Mesenchymal stromal cells (MSCs) are currently being tested in numerous clinical trials as potential cell therapies for the treatment of various diseases and due to their potential immunomodulatory, pro-angiogenic, and regenerative properties. However, variabilities in tissue sources, donors, and manufacturing processes and the lack of defined critical quality attributes (CQAs) and clinically relevant mechanism of action (MoA) pose significant challenges to identify MSC cell therapy products with a predictable therapeutic outcome. This also hinders regulatory considerations and broad clinical translation of MSCs. MSC products are often administered to the patient immediately after thawing from cryopreserved vials (out-of-thaw). However, the qualifying quality-control assays are either performed before cryopreservation, or after culturing the post-thaw cells for 24-48 hours (culture-rescued), none of which represent the out-of-thaw product administered to patients. In this study, we performed a broad functional characterization of out-of-thaw and culture-rescue MSCs from bone marrow (BM-MSCs) and cord tissue (CT-MSCs) using macrophage activation and T cell proliferation-based in vitro potency assays and deep phenotypic characterization using single-cell RNA-sequencing. Using this data, we developed unbiased computational models, specifically symbolic regression (SR) and canonical correlation analysis (CCA) models to predict the immunomodulatory potency of MSCs. Overall, our results suggest that manufacturing conditions (OOT vs. CR) have a strong effect on MSC-function on MSC interactions with macrophages and T cells. Furthermore, single-cell RNA-seq analyses of out-of-thaw BM and CT-MSCs indicate a tissue of origin-dependent variability and heterogeneity in the transcriptome profile. Using symbolic regression modeling we identified specific single-cell transcriptomic attributes of MSCs that predict their immunomodulatory potency. In addition, CCA modeling predicted MSC donors with high or low immunomodulatory potency from their transcriptome profiles. Taken together, our results provide a broad framework for identifying predictive CQAs of MSCs that could ultimately help in better understanding of their MOAs and improved reproducibility and manufacturing control of MSCs.