Hypothesis

Core idea: Tau propagation across the brain is not merely a passive diffusion of seeds but is actively gated by activity‑dependent liquid‑liquid phase separation (LLPS) of tau into viscoelastic condensates that form preferentially along low‑dimensional neural manifolds revealed by UMAP/t‑SNE. Amyloid‑β (Aβ) accelerates this process by altering intracellular ion homeostasis (e.g., raising Ca²⁺) and promoting phosphorylation of tau’s low‑complexity domain, thereby lowering the critical concentration for tau LLPS. Consequently, regions where the intrinsic manifold curvature concentrates tau (high local density in reduced‑dimensional space) become nucleation hotspots, and exosomal release ([4]) exports these condensates to connected nodes, seeding further LLPS.

Testable predictions:

1. Manifold‑dependent threshold: In vitro, recombinant tau subjected to crowding agents that mimic the effective volume fraction derived from UMAP‑based manifold density maps will show a lower critical concentration for LLPS compared with homogeneous crowding. This threshold will shift downward when treated with Aβ‑induced Ca²⁺ flux or kinase cocktails that mimic AD‑like phosphorylation.
2. Activity dependence: Optogenetic enhancement of neuronal firing in entorhinal cortex slices will increase local tau LLPS (measured by fluorescent tau‑FRET biosensors) and accelerate the spread of tau‑positive puncta to connected hippocampal regions; silencing will suppress LLPS and spread despite identical seed load.
3. Pharmacological disruption: Application of 1,6‑hexanediol or a tau‑specific LLPS inhibitor will reduce exosomal tau load ([4]) and slow the epidemic‑style tau‑PET spread predicted by connectome models ([2]), without altering total Aβ burden.
4. Cross‑seeding synergy: Co‑expression of islet amyloid polypeptide (IAPP) will further lower the LLPS threshold for tau via heterotypic interactions ([5]), producing a supra‑additive increase in manifold‑restricted tau condensates and accelerating spread in diabetic mouse models.

Falsifiable outcomes: If manipulating neuronal activity fails to alter tau LLPS or spread, or if LLPS inhibitors do not diminish exosomal tau propagation, the hypothesis is refuted. Similarly, if the manifold‑derived density map does not predict regional tau‑PET uptake better than pure connectome‑based models, the proposed geometric gating is unsupported.

Mechanistic insight: The hypothesis bridges polymer physics (LLPS and sol‑gel transition), neural manifold geometry, and prion‑like seeding by positing that the brain’s intrinsic low‑dimensional activity patterns create microenvironments where tau’s phase behavior is tuned. Aβ acts as a modulator of the solvent quality (ions, phosphorylation) rather than merely a seed accelerator, providing a mechanistic link between the amyloid cascade and the observed acceleration of tau spread beyond connectivity predictions ([2]). This framework generates concrete, multi‑scale experiments—from biomolecular condensate assays to whole‑brain imaging—that can validate or reject a unifying mechanism of AD progression.

References (inline): [1] Adaptive Modelling of Anti-tau Treatments for Neurodegenerative Disorders, [2] Spread of pathological tau proteins through communicating neurons in human Alzheimer's disease, [3] Neural manifold analysis of brain circuit dynamics in health and disease, [4] Alzheimer's disease pathology propagation by exosomes containing toxic amyloid-beta oligomers, [5] Molecular interaction between type 2 diabetes and Alzheimer's disease through cross-seeding of protein misfolding