Glio-SERS: Label-Free Molecular Profiling of Plasma Extracellular Vesicles in Brain Tumors Using SERS and Artificial Intelligence

Abstract

Extracellular vesicles are increasingly recognized as important carriers of disease-associated molecular information, yet robust methods for their isolation and molecular characterization from limited clinical samples remain challenging. Here, we present an integrated approach combining standardized EV isolation, label-free Surface-Enhanced Raman Spectroscopy (SERS), and artificial intelligence (AI) for comprehensive molecular profiling of small extracellular vesicles (sEVs) from human plasma. Here, we show systematically isolated and characterized plasma sEVs using ExoTIC in accordance with MISEV2023 guidelines, with SERS analysis revealing quantifiable spectral differences across samples from patients with glioblastoma (n=20) and meningioma (n=23) compared to healthy controls (n=30). Among the evaluated AI models, the convolutional neural network most effectively captured group-level spectral differences in sEVs, achieving accuracies up to 88% in this pilot cohort. Further, an EGFR-based spectral regression model was explored to examine molecular variability across sEV samples. Parallel proteomic analysis presented statistically significant differences in several proteins elevated in glioblastoma or meningioma. This label-free, rapid approach provides a proof-of-concept framework for sEV molecular profiling establishing the basis for broad validation studies across diverse diseases.