Nucleic Acid Turnover and Lipid Remodeling Distinguish T Cell Activation and Exhaustion States via Label-Free Raman Spectroscopy.

T cell exhaustion has widespread implications for the progression and treatment of chronic diseases including tuberculosis, HIV, malaria, and cancer, yet current detection methods require expensive and tedious antibody labeling, destructive workflows, or days-long functional assays that limit dynamic monitoring capabilities. Here, we introduce Raman spectroscopy as a label-free assay for distinguishing T cell states directly from culture while preserving viability for downstream use. We leverage a 1-D convolutional neural network with sharpness aware minimization for machine learning-based spectral analysis, allowing us to identify critical Raman features for distinguishing exhausted T cells. We achieve >97% accuracy in discriminating unstimulated, activated, and exhausted T cells across three donors and multiple hardware setups, with >92% accuracy in identifying an intermediate activation-exhaustion transition state. We identify vibrational modes associated with alterations in nucleic acids and lipids as key features that distinguish T cell activation and exhaustion. In heterogeneous populations, we quantify exhaustion percentage with R2 = 1 and strong correlation to adenine (r = -0.91) and amide II protein (r = 0.94) vibrational modes. This work establishes vibrational fingerprinting as a direct measure of T cell exhaustion beyond surface marker expression toward scalable immune diagnostics, in-line monitoring, and selective immunopheresis.