Implications of pentose phosphate metabolism and astrocyte co-expression patterns in the pathogenesis of Alzheimer's disease: evidence from artificial intelligence-driven omics and clinical validation.

Background: Metabolic reprogramming in glial cells is increasingly recognized as a pivotal factor in the pathogenesis of AD. While both the pentose phosphate pathway (PPP), a crucial metabolic route for redox balance and biosynthesis, and astrocyte reactivity are implicated in AD, their integrated molecular interaction between processes remains largely uncharted. Methods: We employed an integrated multi-omics and artificial intelligence (AI) framework. First, we applied the Limma, WGCNA, and xCell algorithms to bulk RNA-seq profiles from the hippocampus of AD patients to identify a gene signature linking the PPP and astrocyte reactivity (PA). This signature was then used to construct a diagnostic model via an explainable machine learning pipeline and to stratify patients into molecular subtypes through consensus clustering. The central hub gene of this PA-associated signature was subsequently validated using spatially and temporally resolved single-cell data from the AD hippocampus. An AI-based drug-repurposing screen (Drugreflector) and molecular docking simulations were used to identify natural compounds targeting this hub gene for the treatment of AD. Finally, the dysregulation of this target was confirmed by quantifying its expression in peripheral blood samples from a clinical AD cohort. Results: We identified eight PA-associated gene signatures in AD patients that can help elucidate the pathogenesis and molecular stratification. Conclusion: Our study reveals a novel PA-associated molecular axis in AD, with HDAC3 serving as a central epigenetic-metabolic regulator within astrocytes. This co-expression axis provides a framework for patient subtyping, offers a promising diagnostic biomarker, and identifies a potential therapeutic target for AD.