Background

Multi-omics integration improves risk prediction and patient stratification across cancers and neurodegenerative diseases [1][2]. However, most approaches rely on static tissue or blood snapshots, missing dynamic intercellular communication that precedes clinical onset.

Hypothesis

We propose that longitudinal profiling of extracellular vesicle (EV) cargo—specifically astrocyte‑derived metabolites and neuron‑enriched proteins—combined with plasma proteomics and metabolomics yields a pre‑symptomatic signature that predicts conversion to Alzheimer’s disease ≥2 years before clinical diagnosis.

Mechanistic Rationale

Astrocytes regulate neuronal metabolism via lactate shuttle and release of lipid‑bound signaling molecules packaged into EVs [3]. Early astrocytic reactivity alters EV metabolite composition (e.g., increased myo‑inositol, decreased glutamate) while neurons package synaptic proteins (e.g., neurogranin, SNAP‑25) into EVs as synaptic stress mounts. Simultaneously, systemic inflammatory shifts alter plasma protein levels. Integrating these layers captures the bidirectional astrocyte‑neuron metabolic coupling that drives early network dysfunction.

Testable Predictions

1. Individuals who later develop AD will show a baseline EV metabolite shift (↑myo‑inositol/↓glutamate) and neuronal protein enrichment (↑neurogranin) in astrocytes‑derived EVs compared with stable controls.
2. The multi‑omic EV‑plasma model will achieve ≥75% classification accuracy for future converters, outperforming polygenic risk scores or plasma proteomics alone.
3. Perturbing astrocytic lactate transport in vitro will normalize the EV metabolite shift, providing a mechanistic link.

Experimental Design

• Enroll 300 cognitively normal adults aged 60‑75, genotype for APOE ε4, collect blood every 6 months for 3 years.
• Isolate astrocyte‑enriched EVs using LCAM1‑based immunocapture; perform untargeted metabolomics and targeted proteomics (neuronal markers).
• Parallel plasma proteomics (SomaScan) and metabolomics (LC‑MS).
• Build a graph convolutional network that fuses EV metabolite, EV protein, plasma protein, and plasma metabolite nodes, trained on converters vs non‑converters.
• Validate in an independent cohort of 150 participants.

Potential Implications

If validated, this EV‑centric multi‑omic biomarker could enable enrollment of pre‑symptomatic individuals into prevention trials, guide astrocyte‑targeted interventions, and demonstrate a scalable route for multi‑omics integration into clinical workflows.