Unsure

Accurate staging of Alzheimer's disease (AD) pathology is crucial for therapeutic trials and prognosis, yet existing fluid biomarkers lack specificity, particularly for assessing tau deposition in amyloid-beta (Aβ)-positive patients. We analyzed cerebrospinal fluid (CSF) samples from 136 participants in the Alzheimer's Disease Neuroimaging Initiative using approximately 6,000 proteins. We developed machine learning models that accurately predicted AD pathological stages defined by amyloid and tau PET. We identified two distinct protein panels: sixteen proteins, including NEFH and SMOC1, that distinguished Aβ-negative/tau-negative (A–T–) from A+ individuals, and nine proteins, such as HAX1 and GPI, that differentiated A+T+ from A+T– stages. These signatures outperformed established CSF biomarkers (AUC: 0.92 vs. 0.67–0.70) and effectively predicted disease progression over a decade, findings that were validated in both internal and external cohorts. Collectively, these results underscore the potential of proteomic-based signatures to refine AD diagnostic criteria and enhance patient stratification in clinical trials.