Proteome Profiling of Cerebrospinal Fluid and Machine Learning Reveal Protein Classifiers of Two Forms of Alzheimer’s Disease Characterized by Increased or Not Altered Levels of Tau

Abstract

Alzheimer’s disease (AD) is a multifactorial neurode- generative disorder that presents with heterogeneous clinical and pathological features, necessitating improved biomarkers for accurate diagnosis and patient stratification. In this study, we applied a data- independent acquisition-based proteomics workflow to cerebrospinal fluid (CSF) samples from 138 individuals, including AD patients with high (Aβ+/tau+) or normal (Aβ +/tau−) CSF tau levels, and non-AD controls. Analysis using an Astral mass spectrometer enabled unprece- dented proteome depth, identifying 2661 proteins with high data completeness. Comparative proteomic profiling revealed distinct protein signatures for Aβ+/tau + and Aβ+/tau− subtypes. These findings were validated using an independent internal cohort and further corroborated with publicly available datasets from larger external AD cohorts, demonstrating the robust- ness and reproducibility of our results. Using machine learning, we identified a panel of 15 protein classifiers that accurately distinguished the two AD subtypes and controls across datasets. Notably, several of these proteins were elevated in the preclinical stage, under- scoring their potential utility for early diagnosis and stratification. Together, our results demonstrate the power of data-independent acquisition proteomics on the Astral platform, combined with machine learning, to uncover subtype-specific biomarkers of AD and support the development of personalized diagnostic strategies.