Tau AD fragment aggregates proliferate through autocatalytic secondary nucleation

The self-assembly of the protein tau into neurofibrillary tangles is one of the hallmarks of Alzheimer’s disease and related tauopathies. Still, the molecular mechanism of tau aggregation is largely unknown. This problem may be addressed by systematically obtaining reproducible in vitro kinetic measurements under quiescent conditions in the absence of triggering substances. Here, we implement this strategy by developing protocols for obtaining an ultra-pure tau fragment (residues 304-380 of tau441, tau AD fragment) and for performing spontaneous aggregation assays with reproducible kinetics under quiescent conditions. We are thus able to identify the mechanism of fibril formation of the tau AD fragment at physiological pH using fluorescence spectroscopy and mass spectrometry. We find that primary nucleation is slow, and that secondary processes dominate the aggregation process once the initial aggregates are formed. Moreover, our results further show that secondary nucleation of monomers on fibril surfaces dominate over fragmentation of fibrils. Using separate isotopes in monomers and fibrils, through mass spectroscopy measurements, we verify the isotope composition of the intermediate oligomeric species, which reveals that these small aggregates are generated from monomer through secondary nucleation. Our results provide a framework for understanding the processes leading to tau aggregation in disease, and for selecting possible tau forms as targets in the development of therapeutic interventions in Alzheimer’s


Introduction
Alzheimer's disease (AD) is a devastating neurodegenerative disease of increasing prevalence (1)(2)(3). The pathology of AD is linked to the self-assembly of the amyloid β peptide (Aβ) into extracellular plaques and of the tubulin-associated unit (tau) protein into intracellular tangles in the brains of AD patients. According to the amyloid cascade hypothesis, extracellular Aβ aggregation leads to intracellular hyper-phosphorylation and aggregation of tau (4,5).
Significant progress has been made over recent years in understanding the mechanism of Aβ aggregation and its associated toxicity (6)(7)(8)(9)(10). However, obtaining a comparable understanding of the microscopic mechanism underlying tau self-assembly remains a key challenge.
Tau was isolated and identified in 1975 (11), and its concentration in neuronal cells is found to be around 2 μM (12,13). The protein is essential for microtubule assembly, but it may misfold and aggregate into neurofibrillary tangles. Such misfolded forms of tau have been identified in several human neurodegenerative diseases, collectively known as tauopathies, including AD, Danish and British dementias, white matter tauopathy with globular glial inclusions, frontotemporal dementia, Pick's disease, progressive supranuclear palsy (PSP), corticobasal degeneration (CBD), argyrophilic grain disease (AGD), Guam Parkinsonism-dementia complex, tangle-only dementia, Parkinsonism linked to chromosome 17 (FTDP-17T), and Gerstmann-Strӓussler-Scheinker disease (14)(15)(16)(17). Several reports indicate that posttranslational modifications may differentiate between these tauopathies (18)(19)(20)(21). Tau can also be secreted from neurons (22) and has been observed in blood and cerebrospinal fluid (CSF, 23), in which the levels of phosphorylated tau isoforms serve as biomarkers for AD (24). Since aggregated forms of tau have been identified in the extracellular space (25), both intracellular and extracellular tau may play a role in the disease. Several reports indicate that misfolded and aggregated forms of tau may be taken up by cells, and tauopathies have been reported to spread by cell-to-cell transmission (26,27).
Tau aggregates are composed of β-sheet fibrils that may contain a variety of hyperphosphorylated isoforms of tau, and several polymorphs have been reported (19,20,28,29). In a healthy brain, at least six isoforms of tau, containing between 352 and 441 residues, are produced by alternative splicing of the mRNA from the MAPT gene on chromosome 17. Three isoforms contain four microtubule binding repeats (4R) of 31-33 amino acids, and three isoforms contain three such repeats (3R). The sequence of one major isoform (referred to as tau441, 4R2N, or full-length tau) with four repeats is shown in Fig. 1A with the residue numbering used in this study. Cryo-EM structures of the so-called paired helical filaments (PHF) extracted from the brains of AD patients revealed that each plane of these PHF fibrils contains two monomers with curved β-strands perpendicular to the fibril axis, and with extended β-sheets formed between identical segments of adjacent planes along the fibril axis (28) (Fig. 1B). The ordered part of each monomer in the fibril comprises residues 306-378, corresponding closely to the shorter of two stable tau peptide fragments (9.5 and 12 kDa) identified in PHFs extracted from AD patient brains (30). Tau proteolysis, resulting in a variety of aggregation-prone tau fragments is known to occur in tauopathies (31,32), and recently a recombinant fragment (residues 297-391) was reported to form highly ordered PHFs, although a detailed mechanistic understanding of its aggregation was not described (33,34).
Since full length tau is highly soluble, finding its mechanism of aggregation has remained a major challenge, with previous work using various additives or foreign surfaces to trigger heterogeneous primary nucleation, or vigorous agitation to speed up primary and secondary processes. The additives include anionic substances and polymers such as heparin, polyglutamic acids, nucleic acids, negatively charged phospholipids and surfactants that accelerate fibril formation (35)(36)(37)(38)(39)(40)(41)(42). Although tau contains two cysteines at positions 291 and 322, the rate of assembly has been found to be inhibited (33) or independent of cysteine crosslinking (43,44); however, similar assembly rate was found for wild-type in the presence of reducing agent as for a variant with the Cys322 side chain replaced though mutation (33).
Here we studied the spontaneous self-assembly of the tau AD fragment into amyloid fibrils under quiescent conditions in pure buffer in sample containers selected for low surface interaction of proteins. The tau AD fragment comprises residues 304-380, which covers the ordered part of the fibril, i.e. residues 306-378 (Fig. 1B, 28) plus two extra residues at each terminus, and a Ser residue replaces Cys322. Analogous protocols have previously been successfully applied in the cases of Aβ peptide from AD (6-10) and α-synuclein from Parkinson's disease (45,46) to identify the microscopic steps underlying their self-assembly reaction, and are now translated to the study of the tau AD fragment. The formation of amyloid fibrils in supersaturated monomer solutions, without and with the addition of pre-formed seed fibrils, was followed by the fluorescence of the amyloid-specific dye 4-bis(3-carboxy-4hydroxyphenylethenyl)-benzene (X34), and validated using cryo-EM. Global kinetics analysis (47) of data was used to find the minimal combination of microscopic processes that produces a good fit to the experimental data. To distinguish between models, and to pinpoint the origin of the small oligomeric intermediates in the aggregation process, which are of relevance for neurotoxicity, we used separate nitrogen isotopes in monomers and fibril seeds (6). The results reveal that the aggregation mechanism of the tau AD fragment is governed by secondary processes and that secondary nucleation on fibril surfaces is the dominant source of new oligomers.

Expression and purification of the tau AD fragment, tau304-380_C322S
As a first step towards a mechanistic investigation, we developed methods for the preparation of the tau AD fragment in a highly pure monomeric form. The tau AD fragment, tau304-380_C322S, has the human sequence with a Cys322 to Ser mutation, with residues numbered as in the human tau 441 isoform (Fig. 1A). We expressed this fragment in E. coli with no tag except the starting Met residue. This strategy affords a simple and scalable purification protocol based on boiling and two ion exchange steps with intervening switch in pH to alter the elution of the fragment relative to impurities ( Fig. 2A). The protocol ends with monomer isolation using size exclusion chromatography (SEC) (Fig. 2B), which also exchanges the buffer to the one used in the kinetics experiments. This protocol leads to the isolation of highly pure and sequence homogeneous tau304-380_C322S peptide.

Aggregation kinetics by X34 fluorescence
As a second step towards a mechanistic investigation, we optimised the conditions for reproducible spontaneous self-assembly of the tau AD fragment, tau304-380_C322S, into amyloid fibrils. We measured aggregation kinetics starting from supersaturated solutions of and 125 µM tau AD fragment in Fig. 2C. All data have a sigmoidal-like appearance with a lag phase, a steep transition, and a final plateau. The fragment displays a long lag time, indicative of low nucleation and/or growth rates, however, there is no evidence of nucleation being slow enough to be stochastic under the studied conditions in bulk solution. Moreover, the kinetics show a clear concentration dependence with faster overall aggregation for higher concentrations of monomer, but with saturation observed above 10 µM (Fig. S1). The data in the 1-8 µM range were therefore used in the kinetic analyses, covering the physiological range of tau concentrations (12,13,48). The time to reach the reaction midpoint, t1/2, was extracted from each curve as the point in time at which the X34 fluorescence reached the mid-point between the initial baseline and plateau values. To connect the timescale for each reaction with the corresponding total monomer concentration, the half-time data were fitted by a power function, t1/2 = A c γ yielding a scaling exponent γ=-0.65 (Fig. 3A).

Kinetic analysis
Global chemical kinetic analysis of the data was carried out using integrated rate laws describing protein aggregation in terms of the component microscopic steps (47). The analysis revealed that models including only primary nucleation and elongation, but not secondary nucleation, do not fit the data (Fig. 3B). Strikingly, however, the inclusion of multi-step secondary nucleation of monomers on the fibril surface, together with primary nucleation and elongation, resulted in a good global fit (Fig. 3C). In this model, secondary nucleation is treated in analogy with the Michaelis-Menten model for enzyme kinetics, with an arrival step for binding of substrate (monomers) to the fibril and a second step involving product formation and release. Accordingly, secondary nucleation saturates at high monomer concentration and, where the reaction order is n2=2, the square root of the Michaelis constant, √KM, corresponds to the monomer concentration at which secondary nucleation is half saturated. From the fits to the data describing de novo aggregation in the absence of pre-formed seed material, it is possible to estimate the values of the products of the rate constants for primary nucleation and elongation, knk+, and for secondary nucleation and elongation, k2k+, as listed in Table 1.
Interestingly, while the data are not well described by models that lack secondary processes, i.e. processes which generate new aggregates at a rate dependent on the concentration of existing aggregates, the data are also reasonably well fitted by a model including primary nucleation, elongation and the alternative secondary process of fibril fragmentation (Fig. 3D).
Thus, additional experiments are required to discriminate with certainty the relative roles of monomer-dependent secondary nucleation and fragmentation.

Seeded aggregation kinetics
Monitoring the aggregation process of proteins in the presence of preformed fibril seeds makes it possible to distinguish secondary processes (secondary nucleation and fragmentation) from primary processes (primary nucleation). Indeed, an observed acceleration of the aggregation process in the presence of added seeds as compared to aggregation in their absence indicates definitively the formation of more aggregates triggered by the fibril seeds (49). In the case of tau304-380_C322S in the absence of seeds, the macroscopic aggregation curves as observed using X34 fluorescence are sigmoidal with a lag phase before aggregates can be detected using macroscopic bulk methods. As little as 0.03% seeds (in monomer equivalents) cause a significant shortening of the lag phase, but the sigmoidal shape of the curve is preserved (Fig.   4). This observation indicates unambiguously the existence of a secondary, i.e. fibrildependent, process, the rate of which increases when the amount of fibril increases. As the concentration of seeds is increased, the length of the lag phase is gradually decreased and at 10% seed it is completely abolished. Above this concentration, the number of seed ends is high and elongation occurs at a high rate from time zero. Again, the data can be well fitted both by a model dominated by monomer-dependent secondary nucleation, and by a fragmentation model. Aggregation as a function of monomer concentration was repeated for a series of samples containing 0.1% seeds (of the respective monomer concentration) at time zero making the low rate of primary nucleation insignificant. These data can be globally fitted over the entire range from 1 to 200 µM monomer (Fig. S2). We show for comparison the data in 1-8 µM concentration range in Fig. 5. These data are well fitted by the model including monomerdependent secondary nucleation and elongation ( Fig. 5A) but also reasonably well by the fragmentation model (Fig. 5B).

Origin of oligomers
The origin of the small oligomeric aggregates appearing as intermediate species during the fibril formation process of the tau AD fragment was evaluated using mass spectrometry with different isotopes in monomers and seeds in analogy with earlier studies of Aβ42 (6,10,50).
Seeded samples of 10 µM 14 N-monomers with 0.1 or 1.0 µM 15 N-seeds were monitored by X34 fluorescence until t1/2, centrifuged to remove fibrils and subjected to SEC to remove monomers and collect oligomer fractions. By following this procedure, we found only tau peptides with 14 N isotope in the oligomer fractions (Fig. 6, Fig. S3). Given the high sensitivity of this assay, and by measuring the baseline noise relative to the 14 N signal we can conclude that the lack of observable 15 N signal means that less than 0.035% of the peptides in the oligomeric fraction originate from the original seeds. The initial 15 N-seeds make up 100% of the total aggregate mass at time zero, but at the half time only ~17% of the total aggregate mass due to the formation of new fibrils from 14 N monomer. Thus, even in a pure fragmentation system the proportion of 15 N peptide from initial seeds is expected to be less than 17% in small aggregates. The failure to observe any 15 N above the 0.035% detection limit in the oligomer fraction thus strongly suggests that monomers are the main source of formation of small aggregates in a seeded sample, ruling out the possibility that they are generated through fragmentation of existing fibrils. We thus conclude that the predominant source of oligomers is secondary nucleation of monomers in a reaction catalysed by fibrils.

Fibril morphology by Cryo-EM
Cryo-EM was used to provide information on the morphology of the aggregates formed by the tau AD fragment. Samples of 25 or 200 µM tau304-380_C322S were monitored by X34 fluorescence, collected directly after reaching the plateau, and found to contain long fibrils with a tendency to associate laterally (Fig. 7A). Another sample of 200 µM tau304-380_C322S was taken for imaging at t1/2, showing laterally associated long fibrils and in their vicinity a number of shorter filaments (Fig. 7B), which may be the results of the monomer-dependent secondary nucleation process detected above.

Discussion
Previous studies of tau aggregation have included various additives to trigger heterogeneous primary nucleation, or have used agitation to accelerate primary and secondary processes. Such approaches have been based on observations that tau does not readily form fibrils spontaneously under quiescent and physiological conditions in terms of pH, ionic strength, temperature and concentration (35)(36)(37)(38)(39)(40)(41)(42)(43)(44). Here we have investigated whether a highly pure fragment of tau is able to spontaneously aggregate in the absence of triggers.
A key challenge in elucidating the microscopic mechanisms underlying tau aggregation has been the generation of reproducible experimental data. Here, we have used macroscopic samples at concentrations expected to be outside the stochastic regime (a 100 µl sample of 5 µM tau fragment corresponds to 3 . 10 14 molecules per sample), and thus behave in a reproducible manner, given high sample purity and careful control of other experimental variables. Indeed, we have first established a reliable protocol for isolation of highly pure monomeric tau fragment comprising residues 304-380 (Fig. 2). Thereafter, we optimised the experimental conditions to allow for reproducible observation of the spontaneous selfassembly of the tau fragment into amyloid fibrils under quiescent conditions. Supersaturated solutions of initially monomeric tau fragments at a series of concentrations were created by a temperature jump and the formation of amyloid fibrils was followed by X34 fluorescence, with the formation of fibrillar aggregates validated using cryo-EM.
Global kinetics analysis (47) of the de novo aggregation data reveals that a model including primary nucleation and elongation alone is not sufficient to reproduce the observed data.
However, two minimal models including secondary processes produce a good fit to the experimental data. The model including secondary nucleation of monomers on the fibril surfaces is only marginally better (5% lower error square sum) than the model including fragmentation of fibrils. Thus, these data provide no discrimination between these two models.
The data from experiments performed for solutions of tau AD fragment monomers supplemented with known concentrations of pre-formed fibrils (Fig. 3, 4) provide only slightly better discrimination in favour of monomer-dependent secondary nucleation.
To further discriminate the two potential classes secondary processes, we used separate isotopes in monomers ( 14 N) and seeds ( 15 N) to pinpoint the origin of the oligomers. This approach identifies monomers from solution as the main component of the oligomers, therefore indicating that the dominant process responsible for the generation of oligomers is the secondary nucleation of monomers on fibril surfaces, rather than the fragmentation of existing fibrils.
We note that approaches similar to the one described here may be suitable for establishing the mechanism of aggregation of other tau forms, including the six full-length isoforms and their post-translationally modified variants. However, one may expect the solubility to be considerably higher and the lag time considerably longer for these other species. A recent combined Monte Carlo simulation and experimental study showed that extending an amyloidogenic peptide with a non-amyloidogenic segment of equal length leads to about 20fold longer lag time (51). Based on these results, the 441-residue full-length isoform could be expected to aggregate on three orders of magnitude longer time-scale than the 77-residue amyloid core fragment studied here. We anticipate that seeded aggregation kinetics will be instrumental for bringing the reaction of the longer isoforms into a reasonable experimental time frame.
We also note that fibrils formed from truncated forms of tau may act in vivo as seeds for the aggregation of full-length tau, and that it is also possible that cellular factors may induce heterogeneous primary nucleation of tau in the cellular environment. In any case, our results indicate that the proliferation of tau aggregates can take place by fibril-dependent secondary processes, which, being autocatalytic, may proceed very rapidly (Fig. 8) This aspect is particularly relevant in the spreading of tau aggregation across different regions of the brain (52), as the ability of the traveling seeds to proliferate may influence the rate of the overall process.
It might also be of relevance to compare the overall rate as well as the rate constants for the microscopic processes to other amyloid systems. Such comparison is most straightforward under identical experimental conditions. In Table 1 we show the estimated rate constant products for the tau AD fragment alongside values for Aβ42 under identical solution conditions (20 mM sodium phosphate, 0.2 mM EDTA, pH 8.0, 37 ºC), mechanical (quiescent) and surface conditions (PEGylated half-area 96-well plates). The rate constant products for secondary pathways are much more similar (within a factor of 15) than for primary pathways (differing by at least five orders of magnitude, although precise determination is challenging due to the low rate for the tau AD fragment). These results indicate that a key difference between the two peptides lies in the primary nucleation rate, which is much slower for the tau AD fragment than for Aβ42.
In summary, we have found that the aggregation mechanism of the tau AD fragment is governed by a secondary process, and that the generation of oligomers is dominated by the secondary nucleation of monomers on fibril surfaces.
Tetraethyl p-xylylenediphosphonate was first dissolved in anhydrous dimethylformamide (DMF). Dry potassium butoxide was then added. 5-Formylsalicylic acid dissolved in DMF was carefully added to the resulting mixture and stirred overnight. The reaction mixture was then added to distilled water to precipitate the desired product and filtered. Finally, the powder was washed with ethyl acetate to remove any remaining organic reagents and the powder was dried overnight under vacuum, yielding a bright yellow-green X34.

Expression of tau304-380_C322S
A synthetic gene with E. coli optimized codons for the fragment tau304-380_C322S (residue numbering referring to human tau 441) was cloned between NdeI and BamHI in a Pet3a plasmid (purchased from Genscript, Piscataway, New Jersey) to code for the fragment "as is"

Purification of tau304-380_C322S
The purification protocol is based on sonication, boiling, ion exchange and size exclusion chromatography (SEC). Pellet from 2 L culture was sonicated in 50 mL of ice-cold 10 mM Tris/HCl, 1 mM EDTA, pH 8,0 (buffer A) for two minutes (50% duty cycle -1 s on 1 s off) and centrifuged at 18,000 g for 10 minutes. The supernatant (4 °C) was poured into 100 mL boiling buffer A (100 °C) under stirring in a glass beaker, thus immediately raising the temperature to 69 °C, followed by rapid heating to 95 °C. The solution was cooled by placing the beaker in ice and swirling the solution, followed by centrifugation at 18,000 g for 10 minutes to pellet the precipitated E. coli proteins. The supernatant was passed through 20 mL using Oxford CT3500 cryoholder and its workstation into the electron microscope, Philips CM120 Biotwin Cryo, equipped with a post-column energy filter, Gatan GIF100. The acceleration voltage was 120 kV and images were recorded digitally with a CCD camera under low electron dose conditions.

Origin of oligomers
Mass spectrometry with isotope discrimination was used to identify the origin of oligomers. 15 N-labeled monomers were isolated and allowed to form fibril seeds. These seeds were added to final concentrations of 0.1 (1%) and 1.0 µM (10%), respectively, to freshly purified 10 µM        (average and standard deviation over three repeats) for reactions starting from pure monomer are shown in black, and data for reactions starting from monomer supplemented with 0.1% seed in monomer units is shown in orange. The data obtained without seeds seem to fall into three regimes, and because of the saturation effects seen above 10 µM, data from the regime 1-8 µM was used in the kinetic analyses (Fig. 3). Data from the same or full range were used in kinetic analysis of the data for seeded samples.     Figure S3. LC-MALDI-TOF-TOF spectra of all spots for oligomer fractions eluting between 8 and 9 ml (fraction 8), between 9 and 0 mL (fraction 9) and between 10 and 11 mL (fraction 10).