For decades the two proteins at the center of Alzheimer's disease have been narrated as parallel plots: amyloid-beta collecting in plaques outside neurons, tau twisting into tangles inside them, the two meeting mostly in statistics and scan overlays. A group working between the Mario Negri Institute in Milan, Cambridge and Lund put the plots in the same tube and watched. They took a fragment of tau spanning the amyloid core, residues 297 to 391, coaxed it into two of the folds that cryo-electron microscopy has found in human brains (the paired helical filament of Alzheimer's disease and the looser fold of chronic traumatic encephalopathy), then dropped tiny amounts of each into solutions of amyloid-beta 42 and tracked the aggregation with thioflavin T fluorescence. The tau did not simply sit there. It sped the amyloid up.
The work, led by Michele Mosconi with Luisa Diomede and Tuomas Knowles among the senior authors, went up on bioRxiv on October 22, 2025 and was published, peer reviewed, in the Proceedings of the National Academy of Sciences on March 25, 2026, under a slightly altered title: tau catalyzes amyloid-beta aggregation and toxicity in a polymorph-dependent manner. The central claim is mechanistic and unusually specific. Left alone, amyloid-beta 42 in these conditions grows mostly by secondary nucleation, new nuclei budding off existing fibril surfaces; the authors confirmed this with a scaling exponent of minus 1.39, plus or minus 0.05. Add tau fibrils and a different step lights up: primary nucleation, the birth of new amyloid nuclei, now happening on the tau. Using 0.1 micromolar tau seeds against 2 to 7 micromolar amyloid-beta (roughly one tau for every thirty to seventy amyloid molecules), the reaction followed enzyme-like kinetics, fitted to a Hill equation. Tau, in other words, was not being consumed. It was acting like a surface that hosts a reaction.
Then the fold mattered. The CTE-type fibrils were consistently the more efficient catalyst. Docking the hydrophobic core of amyloid-beta, the KLVFFA stretch, onto published tau structures (PDB entries 8Q8R for the paired helical filament and 6NWP for the CTE fold) pointed to a single catalytic surface spanning tau residues 360 to 369. In the CTE fold that pocket is broadly exposed; in the paired helical filament it is hindered. That is the paper's explanation for why two fibrils made of the same protein sequence behave differently toward a second protein entirely. The CTE fibrils also appeared to shape what came out: electron microscopy of the resulting amyloid-beta fibrils showed their cross-over distances constrained relative to amyloid grown alone, which the authors read as a templating effect rather than a simple speed-up.
The biology arm is small and should be read as such. In human SH-SY5Y neuroblastoma cells, 10 micromolar amyloid-beta 42 alone cut viability to about 81 percent of vehicle-treated cells after 24 hours. Add 1.25 micromolar tau seeds and viability fell to about 53 percent with the paired helical filament fold and about 37 percent with the CTE fold, a combined harm the authors calculate as roughly 24 percent stronger than independent additive effects would predict. In transgenic Caenorhabditis elegans of the CL2120 strain, which makes amyloid-beta 42 in body-wall muscle, tau seeds reduced movement in a concentration-dependent way: body bends per minute down 12 percent at 20 micromolar and 22 percent at 100 micromolar for the Alzheimer's fold, and 14 percent and 24 percent for the CTE fold. Cells in a plate and a millimetre-long nematode are not a brain, and the authors do not pretend otherwise.
For a family living with a diagnosis, nothing here changes a medication, a scan, or a prognosis. What it offers is a candidate answer to a question that has nagged at the field: why amyloid and tau pathology so often escalate together, and why the pairing looks worse in some conditions than others. If tau fibril surfaces really do seed amyloid, and if the strength of that seeding depends on which fold is present, then the shape of a tangle becomes a variable worth measuring, not just a classification label. The authors frame their pocket as a blueprint for structure-based design of agents that block specific amyloid-to-amyloid interactions. That is a research programme, not a drug, and the distance from a docking model to a tested compound in people is measured in years.