Label-free detection of glioma radioresistance using exosomal SERS spectra and machine learning.

Glioma ranks among the most intractable malignancies. Given the intricate anatomy of the brain, complete surgical resection is seldom achievable, making radiotherapy a necessary adjunct. Nevertheless, radioresistance, which is closely linked to recurrence, still represents a critical clinical barrier. Effective tumor biomarkers and novel detection methods are urgently required to predict treatment resistance and monitor therapeutic response. Here, we employed surface-enhanced Raman spectroscopy (SERS) coupled with proteomics to profile, for the first time, the characteristic spectral patterns of exosomes secreted by our established radioresistant glioma cells. We further uncovered specific shifts in protein expression during the development of radioresistance, including glycolysis-related proteins (ALDOA and GAPDH) and ribosomal proteins (RPS5 and RPLP0). These proteins are correlated with glioma prognosis. Moreover, bioinformatic analysis revealed that expression levels of all four genes positively correlate with tumor malignancy grade. Furthermore, we established a machine learning-based diagnostic model, Principal component analysis and convolutional neural network (PCA-CNN), for the accurate identification of exosomes derived from radioresistant glioma cells. These findings validate exosomes as a strong candidate biomarker for predicting radioresistance. This approach enables rapid and reliable assessment of radiotherapy resistance in glioma, paving the way for personalized and precise clinical management.