Single-Cell Immune Profiling and Machine Learning Reveal a Predictive Immune Signature for Immunotherapy Response in NSCLC

Background Metastatic non-small cell lung cancer (NSCLC) exhibits high heterogeneity in response to immune checkpoint inhibitor (ICI) therapy. The identification of predictive biomarkers that are easily applicable and minimally invasive is therefore of great interest. Methods Baseline peripheral blood samples from real-world stage IV NSCLC patients treated with first-line pembrolizumab were analyzed using high-dimensional single-cell mass cytometry, clustering, and machine learning according to their response to ICI therapy. Functional immune characterization was additionally performed for validation. Results We observed significantly higher frequencies of circulating CD4⁺ and CD8⁺ EMRA T cells, defined as terminally differentiated effector memory cells re-expressing CD45RA (CD45RA⁺CCR7⁻), in non-responder patients to programmed cell death protein 1 (PD-1) inhibitors. Upon in vitro stimulation, EMRA T cell subsets from non-responders showed reduced expression of activation markers and effector molecules, including CD25, CD69, IFN-γ, and GZMB. Based on these findings, we developed the CD4⁺ EMRA ImmunoPredict Score (CEIPS), a simplified predictive model that integrates activation and regulatory markers of CD4⁺ EMRA T cells. CEIPS stratified patients into Low- and High-Risk groups, with the latter showing significantly poorer progression-free and overall survival (p = 0.005 and p < 0.001, respectively). Conclusions Peripheral CD4⁺ and CD8⁺ EMRA T cells are associated with anti-PD-1 immunotherapy response in metastatic NSCLC patients and represent clinically feasible blood-based biomarkers to improve patient stratification. Building on these findings, we developed the CEIPS score, which integrates these biomarkers and demonstrates predictive value for immunotherapy outcomes in NSCLC.