Hypothesis

Integrating tissue‑specific epigenetic clocks with microbiome‑derived metabolite gradients via a hierarchical variational auto‑encoder (hVAE) will reveal a latent aging axis that is causally driven by vagal afferent signaling, and that disrupting this signal decouples tissue‑specific omics aging trajectories.

Mechanistic Basis

Recent multi‑omics fusion shows that metabolomic age acceleration links to renal function while lipidomic signals map to diet and immune pathways Phenome-Wide Multi-Omics Integration. The microbiome contributes the largest unique aging signal and pleiotropy across organ systems [same source]. Tree‑based models capture nonlinear dynamics and SHAP provides interpretability, yet causality remains unclear [same source]. Vagal afferents transmit gut‑derived metabolites and inflammatory cues to the nucleus tractus solitarius, influencing systemic physiology and epigenetic regulation.

We propose that the hVAE learns a shared latent space where the first principal component corresponds to vagal tone‑dependent variation across tissues.

This aligns with manifold alignment approaches that use PCA‑based reduction for cross‑dataset harmonization A Novel Aging Clock Built on Seven Clinical Biomarkers A mathematical model that predicts human biological age.

Testable Predictions

1. In a longitudinal cohort (n ≥ 2000) with multi‑omics blood, fecal, and tissue‑specific biopsies, the hVAE‑derived latent axis will predict mortality and multimorbidity better than existing multi‑omics clocks (ΔR² > 0.03, p < 0.01) Phenome-Wide Multi-Omics Integration.

2. Experimental vagotomy in aged mice will attenuate the covariance between gut microbiome metabolite shifts and hepatic epigenetic age acceleration, reducing the latent axis score by ≥ 25 % compared with sham controls.

3. Mendelian randomization using genetic variants associated with vagal activity (e.g., CHRNA3 polymorphisms) will show a causal effect on the latent axis but not on omics layers when the axis is conditioned out, indicating mediation.

Methods Outline

• Collect paired blood, stool, and liver biopsies from participants at baseline and 2‑year follow‑up.
• Generate transcriptome, lipidome, metabolome, microbiome, and DNAm profiles.
• Train a hVAE with separate encoders for each modality and a shared decoder; enforce disentanglement via β‑VAE loss.
• Extract the first latent dimension and test associations with phenotypical age, mortality, and multimorbidity using Cox models.
• Perform longitudinal SHAP on the hVAE to track modality contributions over time.
• Validate findings in an independent cohort (UK Biobank subset) and in mouse vagotomy experiments.

Potential Implications

If confirmed, this framework would shift aging clocks from correlative biomarkers to mechanistic probes of neuro‑gut signaling, opening interventions that target vagal tone to modulate multi‑omic aging trajectories.