A Feature Learning Model Identifies Predictive Attributes of Mesenchymal Stromal Cell Efficacy

Abstract

The therapeutic efficacy of human mesenchymal stromal cells (hMSCs) is highly variable, limiting their clinical translation for musculoskeletal diseases and other regenerative medicine applications. There is a poor understanding of the critical quality attributes correlating to therapeutic efficacy of hMSCs. To address this challenge, we analyzed pre-clinical in vitro secretome profiles and in vivo therapeutic efficacy of hMSCs from multiple human donors. hMSCs from different donors showed significant differences between donors in therapeutic efficacy when assessed in a rat post-traumatic osteoarthritis (OA) model. A partial least squares feature learning model was trained to evaluate differences between more and less therapeutic donor hMSCs by examining cytokine secretion profiles, to predict donor-specific therapeutic outcomes. More therapeutic hMSCs exhibited increased secretion of GM-CSF, GRO, IL-4, and PDGF-AA, whereas less therapeutic donors had higher TNF-, IL-6, and MCP-1 secretion. The cytokine profile was accompanied by evaluation of MAPK pathway, which revealed distinct differences in phospho-protein signaling between more and less therapeutic hMSC secretome profiles. Pharmacological inhibition of JNK signaling in more therapeutic donor cells decreased hMSC secretion of the key therapeutic associated cytokines and shifted hMSC secretome towards a less therapeutic profile. Prospective validation of cells from additional donors demonstrated significant correlations between predicted and observed pre-clinical in vivo efficacy to attenuate OA. This approach identifies critical quality attributes enabling consistent prediction of therapeutic potency, thereby addressing a major barrier to scalable and effective cell therapies. These findings advance precision cell-based therapies and offer a framework for standardized donor screening in clinical applications. SummaryA feature learning model was developed, trained, and validated to identify critical quality attributes of MSCs that predict therapeutic potency.