Tau is a highly soluble protein, yet it aggregates abnormally in Alzheimer's disease. Here, we address the question of proteolytic processing of tau and the nucleation of aggregates by tau fragments. We show in neuronal cell models that fragments of the repeat domain of tau containing mutations of FTDP17 (frontotemporal dementia with parkinsonism linked to chromosome 17), produced by endogenous proteases, can induce the aggregation of full-length tau. Fragments are generated by successive cleavages, first N-terminally between K257 and S258, then C-terminally around residues 353-364; conversely, when the N-terminal cleavage is inhibited, no fragmentation and aggregation takes place. The C-terminal truncation and the coaggregation of fragments with full-length tau depends on the propensity for -structure. The aggregation is modulated by phosphorylation but does not depend on it. Aggregation but not fragmentation as such is toxic to cells; conversely, toxicity can be prevented by inhibiting either aggregation or proteolysis. The results reveal a novel pathway of abnormal tau aggregation in neuronal cells.Alzheimer's disease ͉ paired helical filaments ͉ Tau protein ͉ frontotemporal dementia T he neurofibrillary pathology, based on the aggregation of tau protein into paired helical filaments (PHFs), is a hallmark of Alzheimer's disease (AD) and other tauopathies (1), and the distribution of tau deposits correlates with the loss of neurons (2). The discovery that mutations in the tau gene cause frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP17) (3) and tau aggregation provided evidence that abnormalities in tau were sufficient to cause neurodegeneration. Despite the development of transgenic mice that recapitulate some of the hallmarks of AD (4, 5), the mechanism of tau aggregation remains enigmatic.Tau is a natively unfolded protein that shows little tendency to aggregate in physiological buffers (6). However, aggregation can be dramatically accelerated by polyanionic cofactors such as sulfated glycosaminoglycans, RNA, acidic peptides, or fatty acid micelles (7-10). The aggregation is thought to follow a nucleation-elongation pathway (11), whereby an oligomeric nucleus is first formed by self-association of protein subunits, followed by addition of subunits to filament ends. However, several issues remain open in this scheme: (i) Although tau fragments aggregate spontaneously (12), full-length tau is difficult to fibrillize even at high concentration unless aided by polyanions. (ii) Seeding of recombinant tau with Alzheimer PHFs is inefficient without fibrillization inducers. (iii) The nature of the nucleating species of tau fibrils is not known.A further unsolved question is whether tau aggregation is cytotoxic. Because the occurrence of PHFs correlates with the loss of cognitive functions in AD, it is widely assumed that PHFs are one of the causes of neurodegeneration. On the other hand, neurons bearing PHFs can survive for a long time, and memory deficits in transgenic mice can occur indep...