Morphological Profiling Using Machine Learning Reveals Emergent Subpopulations of Interferon-γ-stimulated Mesenchymal Stromal Cells That Predict Immunosuppression

Background: Although a preponderance of pre-clinical data demonstrates the immunosuppressive potential of mesenchymal stromal cells (MSCs), significant heterogeneity and lack of critical quality attributes (CQAs) based on immunosuppressive capacity likely have contributed to inconsistent clinical outcomes. This heterogeneity exists not only between MSC lots derived from different donors, tissues and manufacturing conditions, but also within a given MSC lot in the form of functional subpopulations. We therefore explored the potential of functionally relevant morphological profiling (FRMP) to identify morphological subpopulations predictive of the immunosuppressive capacity of MSCs derived from multiple donors, manufacturers and passages.

Methods: We profiled the single-cell morphological response of MSCs from different donors and passages to the functionally relevant inflammatory cytokine interferon (IFN)-γ. We used the machine learning approach visual stochastic neighbor embedding (viSNE) to identify distinct morphological subpopulations that could predict suppression of activated CD4 and CD8 T cells in a multiplexed quantitative assay.

Results: Multiple IFN-γ-stimulated subpopulations significantly correlated with the ability of MSCs to inhibit CD4 and CD8 T-cell activation and served as effective CQAs to predict the immunosuppressive capacity of additional manufactured MSC lots. We further characterized the emergence of morphological heterogeneity following IFN-γ stimulation, which provides a strategy for identifying functional subpopulations for future single-cell characterization and enrichment techniques.

Discussion: This work provides a generalizable analytical platform for assessing functional heterogeneity based on single-cell morphological responses that could be used to identify novel CQAs and inform cell manufacturing decisions.