Unsupervised clustering identifies a CD161+ MAIT-like T cell population associated with anti—PD-1 response and immune-related toxicity in melanoma 2309523

Abstract

Abstract Introduction Manual gating has been the standard for cytometry analysis but does not scale well to high-parameter panels. Unsupervised clustering enables unbiased profiling of immune cells without biaxial constraints. We developed an unsupervised clustering pipeline and applied it to 29 melanoma patients treated with anti—PD-1 therapy to identify immune populations associated with response and toxicity. Methods Peripheral Blood Mononuclear Cell (PBMC) samples from 29 melanoma patients (70 samples) treated with anti—PD-1 therapy provided by the Huntsman Cancer Institute were analyzed using a 43-marker mass cytometry panel. Cells were clustered using a self-organizing map to iteratively group cells with similar marker expression. Cluster identities were assigned using cosine similarity to manually gated reference populations, followed by expert quality control using cluster-versus-marker heatmaps and marker-specific UMAP expression overlays. Results Unsupervised clustering identified a distinct CD161+ memory mucosal-associated invariant T (MAIT)-like T cell population that was not captured by standard T cell subset and functional marker gating. This subset showed heterogeneous CD8 expression, low CD45RA, variable CD27, and high CD161, consistent with a memory-like MAIT phenotype. The frequency of this population was significantly higher in responders to anti—PD-1 therapy and increased during on-treatment timepoints in patients who developed immune-related adverse events, linking this subset to both therapeutic response and immune-mediated toxicity. Conclusion Unsupervised clustering of high-dimensional cytometry enables identification of immune populations that manual gating misses. Discovery of a CD161+ memory MAIT-like T cell population associated with both anti—PD-1 response and immune-related adverse events suggests overlapping immune mechanisms linking efficacy and toxicity. Unsupervised analysis reduces the risk of missing outcome-associated biomarkers in immunotherapy. Funding Source Teiko Topic Categories Computational and Systems Immunology (COMP)