Plasmatic immune extracellular vesicle profiles identify prodromal and early stages of Parkinson's disease

Abstract

Extracellular vesicles (EVs) hold promise as minimally invasive biomarkers for neurodegenerative proteinopathies, but disease- and stage-specific profiles remain unclear. For this study, we enrolled 378 participants across five centers and the MJFF-BioFIND cohort: 100 healthy controls [HC], 64 isolated REM sleep behavior disorder [iRBD], 41 DeNovo Parkinsons Disease [PD], 89 Late PD, 32 other Synucleinopathies, and 52 Tauopathies. All participants underwent clinical evaluation and blood collection. The 77 subjects from the BioFIND cohort also provided CSF samples. EV concentration and size were assessed by nanoparticle tracking analysis; flow cytometry quantified tetraspanins (CD9/CD63/CD81) and 37 surface markers. Multivariable logistic regression, receiver operating characteristic analyses (ROC), and repeated random forest (rRF) classifiers evaluated diagnostic utility. Late PD showed the highest EV concentrations compared to HC and other disease groups. Participants exhibited distinct EV surface immunophenotypes, with the iRBD group displaying the most extensive immune activation signature vs HC, followed by PD patients. Multivariate logistic regression analysis identified diagnostic marker panels: CD3/CD9/CD25/CD56 for iRBD, SSEA4 for Late PD, CD146/CD209 for Synucleinopathies, and CD8/CD45/CD62P for Tauopathies. ROC confirmed good discriminatory performance, with CD56 emerging as the strongest single predictor for iRBD vs HC, SSEA4 showing high sensitivity for Late PD, and marker combinations providing optimal balance for Synucleinopathy/Tauopathy classification vs HC. In the CSF BioFIND subset, Late PD EVs exhibited increased myeloid (CD1c), adhesion (CD29), activation (CD69), and epithelial (CD326) markers compared to HC. Among these, CD326 was independently associated with Late PD diagnosis. Machine learning classifiers using all 37 surface antigens achieved excellent training performance (91.7-94.3% accuracy for iRBD/Synucleinopathies vs HC) and maintained robust validation accuracy, particularly for iRBD (77.8%) and DeNovo PD (76.6%) vs HC. EV immuno-phenotyping reveals distinct signatures across the neurodegenerative proteinopathies spectrum, with the highest diagnostic utility for prodromal iRBD detection. Longitudinal validation and cell-of-origin refinement represent key next steps toward clinical translation.