Extracellular vesicle-derived FTIR spectroscopy enables sensitive detection of colorectal cancer compared to plasma

Fourier transform infrared (FTIR) spectroscopy combined with multivariate analysis was used to investigate biochemical differences between plasma and extracellular vesicles (EVs) derived from patients with colorectal cancer (CRC) and healthy controls, and to evaluate their diagnostic potential. Both plasma- and EVs-derived spectra exhibited characteristic FTIR bands of biological materials, including carbohydrate and phosphate vibrations (900-1200 cm^-1), protein-related amide I and II bands (∼1650 and 1540-1560 cm^-1), lipid-associated CH stretching modes (2800-3000 cm^-1), and broad O-H/N-H stretching bands (3200-3400 cm^-1). While plasma spectra from cancer and control subjects showed substantial overlap with only subtle spectral differences, EVs-derived spectra revealed more pronounced alterations, particularly in protein- and lipid-associated regions. Unsupervised analyses confirmed these observations. Hierarchical cluster analysis in the 800-1800 cm^-1 range showed limited separation between plasma cancer and control samples, reflecting the biochemical heterogeneity of plasma. In contrast, EVs spectra exhibited clearer clustering according to disease status. Principal component analysis and uniform manifold approximation and projection further demonstrated stronger and more consistent separation for EVs-derived samples, whereas plasma samples remained largely intermixed. Feature selection stability was assessed using a leave-one-patient-out (LOPO) framework with Boruta and Monte Carlo feature selection. Stable discriminatory wavenumbers distinguishing CRC patients' plasma from control plasma were sparse, indicating subtle disease-related changes. In contrast, comparisons of CRC patients' EVs versus control EVs showed highly stable features associated with protein-, lipid-, and phosphate-related regions. Supervised classification using LDA, PLS-DA, random forest and support vector machines under LOPO cross-validation yielded moderate accuracies for CRC patients' plasma versus control plasma (60.4-68.8%) but consistently high accuracies for CRC patients' EVs versus control EVs (95.6-97.8%). Presented results demonstrate that EVs-derived FTIR spectra provide more robust and diagnostically informative signatures of CRC than bulk plasma, highlighting EVs as a promising source for non-invasive cancer detection.