For a decade and a half, the protein blamed for a small, cruel corner of frontotemporal dementia was the wrong one. Cases with FET-protein inclusions were called FTLD-FUS, after fused in sarcoma, and the field moved on. Then in December 2023, a cryo-electron microscopy study in Nature, led from the MRC Laboratory of Molecular Biology, pulled amyloid filaments out of the prefrontal and temporal cortices of four people who had died with that diagnosis and read off their identity: not FUS, but TAF15, a TATA-box binding protein associated factor, with roughly the first hundred residues of its low-complexity domain forming the ordered core. It was a structural portrait of a culprit nobody had suspected. What it could not say was whether the filaments did anything, or simply sat there.

A paper published in Nature Communications on 14 July 2026 by Katerina Konstantoulea, Nikolaos Louros, Marc Diamond and colleagues at UT Southwestern Medical Center, working with brain tissue from collaborators at Mayo Clinic and the University of California, San Francisco, takes the next step. Purified TAF15 assembled into amyloid fibrils under physiological conditions. The team then built a cellular biosensor, a single-fluorophore reporter based on mEos3.2, that lights up when TAF15 inside a living cell converts from soluble protein into aggregate. Fed lab-made TAF15 fibrils, the sensor fired. Fed extracts from the brains of people who had died with atypical FTLD with ubiquitin-positive inclusions (aFTLD-U), the predominant TAF15 subtype, it also fired, in a concentration-dependent way, and the induced aggregates passed on serially to fresh cells. Those are the textbook signatures of prion-like propagation, meaning self-templating protein assembly, not contagion between people.

The specificity is as interesting as the seeding. Tau, alpha-synuclein and amyloid beta 42 seeds left the TAF15 sensor unmoved. So did fibrils of FUS, the very protein the disease was named for, revealing what the authors describe as a cross-seeding barrier. Yet when FUS was co-expressed in cells during TAF15 seeding, it was drawn into the inclusions. That neatly reframes twenty years of neuropathology: FUS may be a passenger recruited to the wreck rather than the driver, which is exactly why the field has begun arguing for the name FTLD-TAF15.

Then the sequence work. Using two computational amyloid predictors and a thermodynamic analysis of the patient-derived fibril structure (PDB entry 8ONS), the team identified four stabilising hotspots in the fibril core, then tested them empirically with a 27-member library of overlapping 15-residue peptides marching in three-residue steps across residues 7 to 99. Six windows formed amyloid on their own, confirmed by electron microscopy, and each of the six accelerated the aggregation of the larger TAF15 fragment and seeded inclusions in the biosensor cells at rising doses. In other words, the propagating behaviour is not diffuse chemistry spread over a floppy domain; it is written into a few short, addressable stretches. That matters practically, because short motifs are the kind of thing antibodies, designed peptides and small molecules can be aimed at. Two of the authors disclose that they co-founded a biotechnology company, Handshake Bio, which the paper states did not fund or influence the experiments.

For families, the honest translation is narrow and still worth having. aFTLD-U typically presents as early-onset, rapidly progressive behavioural variant frontotemporal dementia, in people often in their forties and fifties, and FET-type FTLD accounts for something like 10 to 20 percent of frontotemporal lobar degeneration overall. Until 2023 there was no molecule to aim at. Now there is a protein, a fibril structure, a cell-based assay sensitive enough to detect seeding activity in brain tissue, and six specific motifs. Cell-based seeding assays of this kind have become the workhorses of tau and alpha-synuclein research, feeding into biomarker development and drug screening; TAF15 now has one. A separate January 2026 preprint from the Cambridge group, not yet peer reviewed, found TAF15 filaments in all 17 FTLD-FET brains examined across three neuropathological subtypes, with several distinct folds, which suggests this proteinopathy is larger and more varied than the four original cases implied.

What nobody should read into it: a treatment, a test they can ask for, or a reason to think dementia is catching. The seeding here happened in engineered cells in a dish, with brain extract delivered directly to them. Whether TAF15 aggregates travel between neurons in a living human brain, and whether blocking these six motifs slows anything, are the next questions, and they are unanswered.