Cofactor-Free Tau Filaments Are Dynamic and Undergo Structural Evolution Driven by Thermodynamic Control

Abstract

Tau filaments are a hallmark of neurodegenerative tauopathies, such as Alzheimer’s disease (AD). Structural studies have revealed that patient-derived tau fibrils adopt distinct folds in different tauopathies; however, it is unclear what forces guide this process. To explore this question, we investigated the assembly of a tau fragment containing four disease-associated phospho-mimetics (termed Tau(297–407)-4D) in vitro. Under cofactor-free and quiescent conditions, Tau(297–407)-4D forms fibrils with a core structure that partially resembles the AD fold after about 7 days. Strikingly, we noticed that this filament behaves as a hydrogel and evolves into two new polymorphs as it ages over the next 35 days. Thus, tau fibrils formed under cofactor-free conditions are dynamic, exhibiting substantial nonequilibrium behavior. To probe what types of perturbations might stabilize these structures, we applied mechanical agitation, which drove the filaments toward thermodynamic equilibrium in a mechanism consistent with Ostwald ripening into solid-phase, micrometer-sized particles. Likewise, the addition of polyanionic cofactors to preformed Tau(297–407)-4D fibrils significantly stabilized them, as judged by solubility equilibria and chemical denaturation experiments. A subset of the polyanions also remodeled the fibril structure and tuned the extent of fibril–fibril interactions (i.e., “clumping”). We conclude that environmental factors, such as mechanical stress and/or polyanions, play an important role in promoting the thermodynamic stability of otherwise dynamic tau fibrils. We speculate that, in patients, such factors might contribute to the maturation of disease-specific conformers.

Introduction

Tau fibrils are associated with the symptoms, progression, and spread of various neurodegenerative diseases, including Alzheimer’s disease, chronic traumatic encephalopathy (CTE) and progressive supranuclear palsy (PSP). Extensive research has led to the classification of human tauopathies based on the morphology/polymorphism of these filaments and their immunohistochemical features. For example, high-resolution cryo-electron microscopy (cryo-EM) studies on patient-derived fibrils have shown that some tauopathies, such as PSP, can be defined by specific disease-associated fibril conformations, while other tauopathies seem to share aggregate structures. Now, a key goal of the field is to understand how cellular factors, such as post-translational modifications (PTMs), assembly environments, and molecular cofactors favor the formation of disease-specific conformers.

Tau is a cationic, intrinsically disordered protein, which normally has high solubility, low hydrophobicity, and low aggregation propensity. Purified, full-length tau isoforms, such as the 2N4R splice variant, do not readily self-assemble under physiological conditions due to a significant nucleation energy barrier and/or low filament stability resulting from unfavorable charge repulsion. Accordingly, most in vitro studies of tau assembly have traditionally used polyanionic inducers, such as glycosaminoglycans, nucleic acids, polyphosphates, lipids/micelles, and other anionic polymers, to promote tau filament formation. These polyanions neutralize charge repulsion and favor dehydrative collapse, while also likely bridging tau-tau interactions. In addition, polyanionic cofactors act as templates that guide tau into distinct fibril conformations, as demonstrated by structural studies of heparin- and RNA-promoted tau fibrils. Finally, these cofactors stabilize fibrils, as shown by dissolution of heparin- and RNA-induced tau fibrils when treated with heparinase or RNase. Beyond their utility on the bench, polyanionic cofactors might also be physiologically relevant, as postmortem analysis has shown that biomolecules, such as RNA, phospholipids and glycosaminoglycans, are localized within tau lesions.

Yet not all amyloidogenic sequences need an inducer to undergo fibrilization. For example, it has been shown that the Pronase-resistant “core” of paired helical filaments (PHFs) in AD is composed of residues 297–391 of tau (referred to as Tau(297–391)) and that this fragment spontaneously assembles at supersaturation. In these cases, the assembly environment is important in tau assembly; for example, mechanical stress and heterogeneous surfaces have been shown to overcome the supersaturation barrier. In other examples, the identity/concentration of salts in the reaction buffers have been shown to favor self-assembly through solvent effects (e.g., Hofmeister solvation, macro molecular crowding, and osmotic depletion). ,

Finally, a key, but sometimes under-appreciated, step in the formation of tau pathology in vivo is the coassembly of multiple fibrils into “clumps”. Clumps are defined by extensive fibril–fibril associations, often involving contacts along the long axis. Pathophysiological examples include the neurofibrillary tangles (NFTs) that are a hallmark of AD. NFTs are composed of paired-helical filaments (PHFs), a type of ordered structure involving discrete molecular contacts between protofibrils. In turn, PHFs undergo extensive clustering, becoming intertwined to form insoluble, tightly packed superstructures, as observed by elegant cryo-electron tomography and other landmark studies. , While fibril “clumps” are a striking histopathological feature of AD and other tauopathies, their role in the energetics of fibril maturation are not entirely clear.

Here, inspired by pioneering reports on how thermodynamics shape the evolution of filaments formed from the islet amyloid precursor protein (IAPP), , we asked whether the structure of tau fibrils formed under cofactor free conditions without shaking (e.g., quiescent) is under thermodynamic control. Using cryo-EM, we found that Tau(297–407)-4D fibrils, formed under these conditions, have a protein fold that partially resembles those found in AD and CTE after 7 days. But we noticed that they undergo a striking structural evolution after ∼40 days, producing two additional polymorphs. This transition occurred despite the fact that the thioflavin T (ThT) signal, a widely used measure of fibril content, had reached a plateau. Thus, tau fibrils formed under quiescent and cofactor free conditions are structurally dynamic over these time scales. Then, using this system, we tested whether common environmental perturbations, mechanical stress (e.g., agitation) and treatment with natural polyanions, impact this process. Indeed, we found that these factors accelerate fibril assembly, alter the extent of fibril–fibril interactions (“clumping”) and thermodynamically stabilizing the fibrils, as measured by solubility equilibria and chemical denaturation assays. Moreover, some of these perturbations also caused tau fibrils to adopt different polymorphs. Together, these findings provide mechanistic insight into how physical and cellular factors shape tau fibril conformations, with repercussions on our understanding of how tau adopts distinct structural states in tauopathies.

Results

Tau­(297–407)-4D Filaments Exhibit Time-Dependent Morphological Changes

Recent work has shown that a tau fragment, composed of residues 297–407 with four phosphorylation-mimicking changes (S396D, S400D, T403D, and S404D), hereby referred to as Tau(297–407)-4D, self-assembles without a cofactor (that is, without addition of a polyanionic inducer) in the presence of mechanical stress. ,,, To test if this process would occur under quiescent conditions, we turned to alternatives to mechanical stress, specifically by salting-out in citrate buffer. Indeed, Tau(297–407)-4D assembled into filaments under the optimized conditions of 200 mM citrate and no shaking for 1 week (Figure a). Thioflavin T (ThT) fluorescence experiments showed that these samples had reached the apparent end point (see Figure a). Based on negative stain-transmission electron microscopy (NS-TEM) and manual counting, these filaments had a predominant (68%) morphology of dimeric twisted ribbons with a long crossover length (289 ± 58 nm), alongside two relatively minor populations with PHF-like morphologies (crossover distances of 75 ± 6 and 111 ± 4 nm) (Figure b; Supplemental S1). These structures were broadly maintained across several staining conditions (Supplemental S2). By cryo-EM, we observed different fibril morphology compared to those observed in NS-TEM, transitioning to a single, predominant polymorph, which was resolved to 3.3 Å from the 7-day samples. This structure contained two protofibrils (PFs) that somewhat resemble those found in AD and CTE (Figure c,d).

1.

Tau fibrils structurally evolve over time. a) Assembly conditions: 200 μM Tau(297–407)-4D, 200 mM potassium citrate tribasic, 10 mM DTT, 3 μM ThT, 10 mM KPB-KOH, pH 7.4, 37 °C, without mixing. The polymorph totals determined by 2D classification. The table shows the polymorph percentages, as determined from the 2D class-average particle counts. The totals include unresolvable classes. b) NS-TEM images of the corresponding fibril morphologies. The average crossover lengths and polymorph percentages were determined from 30 NS-TEM micrographs using ImageJ/Fiji. The scale bars are 50 nm. c) Cryo-EM grids were prepared after incubation for 7 days at 37 °C, followed by an additional 35 days at 23 °C (42 days total), yielding three polymorphs amenable to structural determination. d) 3D reconstruction of each polymorph. Note the presence of disorder in polymorph 1 after 42 days. We speculate that conformational instability may explain the differences in crossover-length morphology between the same samples visualized by NS-TEM or cryo-EM. Also, the filament counts differ between NS-TEM and cryo-EM, so it is difficult to determine which method more faithfully represents the ensemble.

3.

Quiescent conditions generate hydrogel-mesh fibrils, whereas agitation drives a transition into solid-phase clumps. a) Assembly conditions: 200 μM Tau(297–407)-4D, 200 mM potassium citrate tribasic, 10 mM DTT, 3 μM ThT, 10 mM KPB-KOH, pH 7.4, 37 °C, with and without mixing. The ThT curves represent an average of 3–6 experiments, and the error bars represent standard deviation. b) NS-TEM of quiescent and agitated gel-phase fibrils. The scale bars are 2 μm. c–d) Determination of the soluble monomer (C_sat) of tau gradients that were seeded with 5 μM fibrils under quiescent and agitated conditions. The soluble monomer in the supernatant was measured with HPLC, and ultracentrifugation was performed at 100 k x g for 60 min (Supplemental S6). e) The solubility of the quiescent fibrils slightly decreases upon aging, and it decreases significantly upon agitation, suggesting that agitation drives it to the end state. For the agitated sample, the quiescent fibrils were mixed at 400 rpm for 4 days before determining the monomer concentration. f) Chemical denaturation assays reveal clumped fibrils are more stable than quiescent gel fibrils. Conditions: 2 μM filaments were incubated at the indicated denaturant concentration at 23 °C for 16 h without mixing. The curves were measured by ThT fluorescence, and the IC₅₀ values were obtained by sigmoidal fitting (Supplemental S7).

Then, to explain why the apparent fibril morphology might be distinct when observed by NS-TEM and cryo-EM, we hypothesized that the tau fibrils are not at thermodynamic equilibrium under these conditions. To test this idea, we placed the sample in the dark at room temperature for an additional 35 days. By NS-TEM, we found that the fibrils underwent an apparent morphological change. Specifically, the tau fibrils had a new major filament with glide symmetry (crossover distance = 159 ± 21 nm) (Figure b). In addition, a secondary fibril class had a PHF-like morphology (crossover distance = 75 ± 9 nm) and a minor product with a unique symmetry was observed (Figure b). By cryo-EM, the 45-day sample contained a structure that resembled the original polymorph, plus two additional polymorphs in which the PFs were arranged in distinct orientations (Figure c,d).

Together, these observations show that Tau(297–407)-4D fibrils formed under cofactor free, quiescent conditions are dynamic and continue to sample new conformers over time.

Local Structural Differences in Tau Protofibrils Underlie Distinct Fibril Folds

Given that the Tau(297–407)-4D fibrils seemed to change over time, we wanted to examine the protomer structure carefully to understand the structural underpinnings of the changes. We considered two types of mechanisms: (i) the PFs were identical between the polymorphs and only changed their relative position (e.g., quaternary structure) over time or (ii) the PFs were not equivalent, such that individual protomers must be dynamic (e.g., in exchange with fibrils). In the young sample (7 days), the major product was termed polymorph 1, and it contains two identical “J-shaped” PFs. These PFs have a well-defined “head” position and an elongated “tail”, roughly similar to those observed in AD and CTE. For a comparison between these PFs and disease examples, see Supplemental S3. In the aged sample (45 days), polymorph 1 accounted for less than a quarter of the particles, and it was refined to a lower-resolution density map, as evidenced by disorder in one of the PF tails (Figure c–d, Supplemental S3). Polymorph 2 (30%), the major product of the aged sample, has an asymmetric protofilament interface with a highly curved cross-section and contains two different PFs. The minor polymorph 3 (9% of the aged sample) has pseudo-C2₁ symmetry and, like polymorph 1, contains empty space between the PFs. All three polymorphs have distinct protofilament interfaces that resemble filaments generated from Tau(297–391) (Figure c–d, Supplemental S3). ,

Atomic modeling revealed that the ordered cores comprise residues 305–379, and that all three polymorphs have nonidentical PFs (Figure a). The tail regions are largely superimposable between these structures, but the head regions (residues 324–361) are shifted relative to the tails and have different degrees of swiveling. The root-mean-square deviations (RMSDs) of the tail regions of each filament compared to polymorph 1 range from 1.3 to 1.5 Å, while the head region RMSDs range from 3.9 to 5.9 Å. Each of the polymorphs involves a unique protofilament interface (Figure b).

2.

Model building and analysis of the filaments. a) Atomic models of each filament that have been refined to the density. The terminal residues of the ordered region are R379 and S305, while the nonresolved residues remain disordered. Polymorph 2 consists of two nonidentical protofilaments (PF-2A and PF-2B). b) Close up of the protofilament interfaces. c) The protofilaments and the AD/CTE folds have alignment in the tail region (residues 305–323 and 362–379), and different levels of curvature in the head region. Several residues localized in the head region (residues 324–361) differ in solvent exposure and cause the conformational differences.

The most striking differences between the PFs are found in two regions: 324–330 and 356–357, with several residues that are alternatively solvent-exposed (“out”) or buried (“in”) (Figure c, Supplemental S3). Notably, H329 and H330 are exposed and buried, respectively, in polymorph 1, but the exposure of these residues is reversed in polymorphs 2 and 3. Within polymorph 2, S356 is buried in one of the protofilaments (PF-2A) but it is exposed in the other (PF-2B). It is interesting to note that residues S324–I328 of the AD and CTE folds are reversed in exposure (in–out-in–out for the disease folds) when compared to polymorphs 2 and 3, but H329 and H330 of the AD and CTE folds are similar to polymorph 1. We speculate that this region in recombinant PFs might be less stable than the patient-derived folds, in part, because the polar S324 and N327 side chains are buried and the hydrophobic I325 and L328 side chains are exposed. It is not clear if the fibrils studied here would eventually form the AD or CTE folds at longer time points (see Discussion). Overall, this close examination of the cryo-EM structures reveals that conformational differences at several residues occur alongside more dramatic quaternary rearrangements of the individual PFs.

Agitation Favors “Clumps” That Stabilize Tau Fibrils

Given that cofactor free tau fibrils appear to be dynamic over these timeframes, we wanted to understand what perturbations might thermodynamically stabilize them. Quiescent assembly conditions are known to improve polymorph selectivity (i.e., homogeneity) and provide better control of the assembly. , In contrast, shear forces have been shown to promote (heterogeneous) nucleation and fragmentation. Consistent with these observations, we found that, while our cofactor free, quiescent conditions slowly produce fibrils, mixing accelerates assembly, as measured by ThT (Figure a). In our hands, the quiescent samples contained long, well-resolved fibrils by NS-TEM, while shaking favored clumping, likely by increasing (heterogeneous) primary nucleation and fragmentation, as well as lateral association between fibrils (Figure b). , Moreover, agitation of quiescent fibrils results in decreased binding to ThT (Supplemental S4), presumably due to restricted access to ThT-binding sites within the tightly packed tau fibril clumps. We could also favor association of preformed tau fibrils into clumps by mixing or centrifugation. Finally, it is worth noting that quiescent assembly is sensitive to the reaction vessel and volume (Supplemental S5), further supporting a role for agitation and collision effects.

It would be interesting to understand whether the fibril structures formed under quiescent conditions (at 42 d) are similar to those formed under mechanical agitation. However, it is challenging to determine the structure of fibrils within the clumps and we estimate that only 10–20% of agitated-assembly reactions yielded any noticeable, isolated fibrils that would be amenable to structural analysis. Moreover, these occasional fibrils are typically short and with inconsistent morphology. While we are hesitant to make firm conclusions, we note that the agitated fibrils did sometimes resemble the 42 d aged fibrils by NS-TEM (Supplemental S10).

The free monomer left at the end of the assembly reaction, or equilibrium solubility (C_sat), represents thermodynamic equilibrium because the chemical potential of free monomers and monomers in the fibrils is equal. Thus, if the fibrils are solids, the concentration of soluble monomer in the supernatant upon ultracentrifugation will remain constant and independent of the total tau concentration. In contrast, we found that the free monomer concentration under seeded quiescent conditions is dependent on the total tau concentration, even though the ThT analysis indicates completion (Figure c, Supplemental S6). Thus, these quiescent fibrils are not a solid phase and behave as semisoluble species within a hydrogel phase, which we envision may result from noncovalent interactions between the fibrils (Figure b). Then, we wondered whether agitation drives these assembly reactions to equilibrium, as shown in other systems. Indeed, upon agitation, we found that the equilibrium solubility values are consistent with theoretical independence from total tau concentration (C_sat ∼ 1.5 μM; Figure d, Supplemental S6). Thus, agitation seems to favor a gel-to-solid transition, in addition to the clumping observed by NS-TEM.

The considerable amount of free tau remaining (∼20%) at the end of quiescent assembly raises questions about the nature of the time-dependent morphological differences. Do the new polymorphs slowly nucleate from the free monomer pool, or do the existing filaments undergo rearrangements? We found that the soluble monomer concentration at 42 d is very similar to the concentration measured at 7 d (Figure e, Supplemental S6). This result suggests that the new polymorphs arise from dynamic transitions in the gel-like state, such as monomer release and subsequent population of the new polymorph(s), rather than depletion of the existing monomer pool to form new fibrils. Moreover, we found that agitation significantly decreases solubility (Figure e), suggesting that the solid-phase clumps likely represent the system’s final state. Taken together, we speculate that the slope of the soluble vs total tau line will decrease to a flat line as time increases, and the filaments will slowly evolve into more stable conformations. We also speculate that upon prolonged aging, the gel filaments will slowly transition into solid-phase clumps; however, this idea remains untested.

Another way to explore the impact of agitation on the stability of tau fibrils is through chemical denaturation. Moreover, such experiments can alternatively employ guanidinium hydrochloride (Gnd·HCl), which screens both electrostatic and hydrophobic interactions, or urea, which screens hydrophobic interactions, to estimate the relative contributions of these forces to stability. , To ask this question, we first created cofactor free tau fibrils, denatured them with chaotropic agents and then calculated the half-maximal denaturant concentrations (IC₅₀) for each condition from ThT fluorescence curves. These experiments revealed that cofactor free fibrils are stabilized by electrostatic interactions, while hydrophobics seem to provide a relatively modest effect, as evidenced by the relative potencies of Gnd·HCl (IC₅₀ = 0.32 ± 0.05 M) and urea (IC₅₀ = 0.81 ± 0.05 M). Then, we performed the same denaturation experiments on the agitated sample, which primarily contains “clumps”. These measurements indicated that hydrophobic interactions play a role in stabilizing these structures, because they have greater urea resistance (IC₅₀ = 1.00 ± 0.07 M) and nearly equivalent Gnd·HCl resistance (IC₅₀ = 0.29 ± 0.05 M) (Figure f, Supplemental S7).

In summary, we conclude that quiescent fibrils have not reached thermodynamic equilibrium under these conditions and that they exhibit solution-phase (or gel-like) behavior with a pool of exchangeable monomer. In contrast, consistent with Ostwald ripening, the large filament clumps formed after agitation represent a thermodynamic end point because they act as solids, have low solubility and relatively higher denaturant stability. Thus, we speculate that fibril–fibril assembly steps, such as those needed to form PHF bundles and NFTs in vivo, might represent a thermodynamic stabilization action that could “lock-in” fibril conformers. Traditionally, these “clumped” structures have been difficult to study by structural methods and are often ignored because of their close-packed density. Often, experimental conditions are optimized to purposefully avoid their formation. Yet, our results, and the predominance of PHFs and other higher-order structures in patient brains, suggest that fibril clumps could meaningfully contribute to the landscape of tau fibrillization, such that additional effort to characterize them could be informative of disease mechanisms.

Polyanions Remodel and Trap Fibril Morphology

Next, we explored the impact of polyanions on the kinetics and thermodynamics of tau fibril formation. It has been proposed that polyanions bind cationic residues across the filament rungs, favoring aggregation and guiding the conformation(s) of the resulting fibrils (Figure a). To test this idea in our hands, we measured the assembly kinetics for Tau (297–407)-4D in the presence of four natural polyanions (heparin, chondroitin sulfate, poly­(A) RNA, and polyphosphate). Indeed, we confirmed that these polyanions accelerated the kinetics of fibril formation (Figure b). Moreover, we found that these samples had strikingly low solubility, with undetectable amounts of soluble tau remaining (Figure b, Supplemental S8). Finally, chemical denaturation experiments on these fibrils revealed that heparin, polyphosphate, and chondroitin sulfate stabilize the filaments through both ionic and nonionic interactions (Figure c). For example, heparin treatment increased resistance to Gnd-HCl from 0.32 ± 0.05 M to 0.60 ± 0.05 M, while this sample was nearly completely resistant to urea. In contrast, the poly­(A) RNA samples seem to be electrostatically stabilized and have slightly lower nonionic stability, as evidenced by increased urea resistance (1.91 ± 0.21 M) and decreased Gnd·HCl resistance (0.24 ± 0.1 M). Thus, these polyanions accelerate assembly kinetics, while also producing more thermodynamically stable, insoluble fibrils.

4.

Polyanion remodeling and stabilization of the quiescent fibrils. a) Graphical overview of polyanion-induced remodeling. b) Polyanions promote seeding and increase stability, as indicated by the decrease in the t _1/2 and C_sat upon assembly in the presence of a heparin, chondroitin sulfate, poly­(A) RNA, and polyphosphate (p700) concentration gradient. Conditions: 200 μM Tau(297–407)-4D, 10 mM DTT, 3 μM ThT, 10 mM KPB-KOH, pH 7.4, 37 °C, 4 d. The t _1/2 values were obtained from kinetic traces, and the tau concentration in the supernatant was measured by HPLC (Supplemental S8). c–e) Conditions: the quiescent fibrils were diluted to 160 μM, and were treated with the polyanions (1 mg/mL heparin; 1 mg/mL chondroitin sulfate; 2 mg/mL poly­(A) RNA; 1 mg/mL polyphosphate (p700)) and incubated for 5 h at 37 °C. c) Fibrils treated with polyanions have significantly higher denaturant resistance. Conditions: The polyanion-treated filaments were diluted to 2 μM and incubated at the indicated denaturant concentration in 10 mM DTT, 10 μM ThT, 10 mM KPB-KOH, pH 7.4, 23 °C, 16 h (see Supplemental S8 for detailed experimental conditions). d) Detailed NS-TEM of reveals a high specificity for polyanion stabilization and remodeling (Supplemental S9). The morphology of the polyanion-treated preformed fibrils appears different from that of the nontreated control (see Figure b). The average crossover lengths and polymorph percentages were determined from 15 NS-TEM micrographs using ImageJ/Fiji. The scale bars are 50 nm. e) NS-TEM of polyanion-treated tau fibrils reveals changes in lateral association (Supplemental S9). The scale bars are 500 nm (top) or 50 nm (bottom).

From this starting point, we sought to explore how these same polyanions might affect the morphology of preformed fibrils made under quiescent conditions. Because these samples behave as soluble hydrogels, we considered it possible that treatment with polyanions might convert the bulk of the tau into new fibril polymorphs. In this model, the polyanion would be expected to convert the soluble pool of tau monomers, while also “capturing” the tau that exchanges with dynamic fibrils. Indeed, we found that incubation of preformed, quiescent tau fibrils with polyanions at concentrations that maximize rate acceleration (Supplemental S9), depleted the previously observed fibril polymorphs and produced a distinct set of new structures by NS-TEM. For example, heparin treatment remodeled both major fibril conformations, yielding a greater proportion of helical filaments (crossover distance = 108 ± 3 nm) and reducing the dimensions of the long filaments (crossover distance = 241 ± 17 nm vs the 305 ± 9 nm measured for the untreated samples). The major polymorph of chondroitin sulfate-treated fibrils appeared largely unchanged (crossover distance = 309 ± 11 nm), with a helicity reminiscent of the quiescent fibrils. Yet, this polyanion selectively remodeled the less abundant polymorph (crossover distance = 74 ± 4 nm vs 105 ± 5 nm in the untreated). The other two polyanions also had a dramatic effect on cofactor free tau fibrils: Poly­(A) RNA converted the filaments into amorphous aggregates (see below), while polyphosphate remodeled the filament into kinked helical architectures. Thus, we conclude that polyanions are generally able to remodel a subset of cofactor-free fibrils, producing different polymorphs. Importantly, we did not observe the original polymorphs in the sensitive samples, suggesting that the polyanion eventually captures tau monomers released from the dynamic, quiescent fibrils. Consistent with this idea, the bulk of the fibrils seemed too abundant to be ascribed purely to the relatively small amount of soluble tau in the quiescent samples (∼20%).

During these experiments, we noted in NS-TEM images that chondroitin sulfate and heparin treatment reduced lateral associations between tau filaments (e.g., the fibrils seemed more spread out on the grid) (Figure e). This observation suggests that these polyanions might introduce electrostatic repulsion between tau polymers or, alternatively, directly compete for fibril–fibril binding sites. In contrast, polyphosphate treatment favored self-association (e.g., the fibrils tended to be laterally aligned and more intertwined) and, as noted above, the poly­(A)­RNA treatment strikingly produced amorphous aggregates.

Together, these results show that dynamic, cofactor free tau fibrils can sometimes be guided into alternative conformers by polyanionic cofactors. The effects of polyanions include both changes in fibril conformation and, sometimes, fibril clumping, as judged by NS-TEM. We envision that similar processes might contribute to the selectivity of tau conformers observed in tauopathy patients.

Discussion

The Structure of Cofactor Free Tau Fibrils Evolves over Time

Tau is a normally soluble, intrinsically disordered protein that forms fibrils in AD, CTE and other tauopathies. While this process has been intensively studied for many decades, the more recent observation that tau fibrils adopt distinct folds in different tauopathies has focused interest in understanding the kinetics and thermodynamics of this process and how this landscape can be shaped by cellular and environmental factors. It has recently been shown that tau and IAPP fibrils evolve over time. , This observation suggests that some fibrils can be more metastable than previously appreciated. For IAPP, cryo-EM structural snapshots have shown that the fibrils are dynamic and that they approach thermodynamic equilibrium over time. Also, for tau, it has been proposed that salts shape this process through subtle effects on the protein–solvent interface (e.g., Hofmeister salting out and direct anion binding). ,, Here, we have used structural and biophysical studies to further explore how agitation and polyanion cofactors impact the thermodynamics of tau filament formation.

Using the Tau(297–407)-4D model under cofactor free and quiescent conditions, we find that the quaternary and tertiary structure of the fibrils evolves over time, as judged by NS-TEM and cryo-EM, even though the ThT signal is unchanged. We find that the major polymorphs (1–3) are related to each other, but that they each have distinct dimeric interfaces and unique PFs with local differences in side chain positions. Although we originally selected the Tau(297–407)-4D model because it has been reported to form the PHF-like AD fold in the presence of MgCl₂, but we were not able to achieve that result (Supplemental S10). ,, Rather, in the presence of citrate, we obtained fibril polymorphs from this tau construct that are similar to those previously observed in the proposed AD/CTE intermediates, with low RMSD values (see Supplemental S3). These gel-phase structures have micromolar equilibrium solubility values, allowing them to change conformation by some combination of partially unfolding, rotating, and/or refolding into new protofilaments. Our cryo-EM studies suggest that the protofilament interfaces also change during this process, possibly through equilibrium with partially unfolded intermediate strands or complete dissociation/reassociation of monomers. For example, polymorph 2 consists of two nonidentical protofilaments, which could represent an intermediate in this type of quaternary structure evolution.

Over time, thermodynamic control, or Ostwald’s rule of stages, drives rearrangements that create more energetically favorable conformations. This process likely occurs by a combination of reducing strain and steric encumbrances and/or increasing entropic/enthalpic interactions, such as salt bridges, van der Waals interactions, hydrogen bonds, and dehydrative forces. , We propose that the initial polymorph 1 has the lowest energy barrier (under the conditions tested), making it the most kinetically accessible product, while subsequent evolution to polymorphs 2 and 3 overcomes more difficult structural transitions to increase stability over time. In support of this idea, several known amyloid rearrangements are consistent with thermodynamic control. ,,

Assembly of Mature Fibrils Contributes to Insolubility and Thermodynamic Stability

The mechanisms described here could be relevant to disease. For example, patient-derived fibrils are collected postmortem, so we assume that they represent a thermodynamic end state. Our data supports a general idea, previously articulated, that alternative structures (e.g., intermediates) might appear earlier in disease and could be pathologically meaningful. ,

Yet, different tauopathies can still feature tau fibrils with distinct and often dramatically different folds. If tau fibril evolution toward thermodynamic equilibrium were the only feature of the self-assembly process, then one would expect to find similar folds in all late-stage diseases. Thus, there must be other aspects of the environment, such as metabolites and cofactors, which further shape filament formation. One uniform feature of tau pathology in vivo is that the fibrils are densely packed. Recent cryo-electron tomography images of fibrils nicely illustrate the fibril side-by-side stacking that occurs in cells. , Inspired by these observations, we used mechanical stress (i.e., agitation, centrifugation) to promote the lateral association of tau fibrils. We found that this perturbation induces a gel-to-solid transition and produces larger, less soluble, and more denaturant-stable filaments that are seen as “clumps” in NS-TEM (see Figure ). In material science, this process is known as Ostwald ripening, wherein large assemblies are more thermodynamically stable, driven by a decrease in solvent-exposed surface area. This process also features favorable dehydration entropy and van der Waals interactions. In patient brains, tau pathology appears as laterally bridged filaments and PHF bundles, rather than micrometer-sized particles predicted by Ostwald ripening. We speculate that tau does not reach this advanced stage of ripening due to the lack of shear force and, in addition, that the growth area is physically restricted. It is also worth noting that (partial) proteolysis of the fuzzy coat in vivo might remove regions that normally limit fibril-to-fibril associations, ,, and that this enzymatic process could contribute to clumping. Consistent with this idea, lateral associations between individual filaments in vitro and in vivo tend to occur at sterically accessible locally proteolyzed regions of the filament surface. , – Together, we suggest that, in patients, mechanical forces might be one factor that “traps” tau fibril conformers within large assemblies (e.g., neurofibrillary tangles, extracellular ghost tangles). Once in these assemblies, it remains unclear if further conformational transitions are plausible, or if the fibril conformer is effectively frozen into place.

Polyanions Interrupt the Folding Trajectory and Thermodynamically Trap Specific Filament Conformations

Polyanions have emerged as another key factor in guiding tau filament conformation. For example, primary nucleation with total RNA from mouse liver is able to drive tau into a fibril conformation that has not been observed previously, and a screen of ∼30 natural polyanions showed that many of them template tau into distinct conformations. ,

To better understand the impact of these natural cofactors on tau thermodynamics, we treated Tau (297–407)-4D fibrils, preformed under cofactor-free conditions, and found that heparin, poly­(A) RNA and polyphosphate all cause an alteration in apparent fibril structure. These changes included both alterations in fibril morphology and changes in lateral association (see Figure ). Moreover, all four of the polyanions caused a dramatic stabilization of the fibrils. Thus, one can imagine that differences in polyanion availability (e.g., cell types and disease state) and polyanion accessible sites will contribute to the selectivity observed in patients with distinct tauopathy folds (e.g., PSP vs AD). Moreover, the polyanions appear to thermodynamically “trap” specific conformers (see chondroitin sulfate in Figure d) and may explain the role of the of unknown stabilizing cofactors bound to the detergent-resistant brain-purified tau aggregates. Given that the cofactor-free aggregates have not yet reached thermodynamic equilibrium, we propose that polyanions may interrupt the innate folding trajectory and redirect it toward alternative end state conformations.

Limitations of the Study

We set out to probe whether tau fibrils are dynamic and, if so, how environmental factors might promote their stability and tune their conformations. The results suggest that tau fibrils formed under cofactor free conditions are, indeed, dynamic for many weeks and that agitation and polyanions stabilize these structures. While these findings are valuable for understanding tau assembly, we caution that these studies were performed using a single tau proteoform Tau(297–407)-4D and it seems possible that different relationships and/or conformers could be uncovered for other proteoforms, such as full-length tau with an intact fuzzy coat or tau variants with alternative PTMs. Thus, the relationship between the in vitro maturation pathway and disease-associated tau structures remains incompletely understood. It is possible that much longer timeframes are required to resolve this relationship. For example, future studies may help confirm whether extended incubation shifts the S324 to I328 side-chain orientations toward closer agreement with AD or CTE structures. Ultimately, the mechanisms proposed here will require validation in relevant cellular and animal models of tauopathy.

Conclusion

This study demonstrates that cofactor-free tau filaments are malleable and dynamic, and that they slowly evolve into new polymorphs, as anticipated by Ostwald’s rule of stages. The findings suggest a mechanism of thermodynamically controlled rearrangements and demonstrate the importance of agitation and stabilizing polyanionic cofactors on tau pathology.

Glossary

Abbreviations

AD

Alzheimer’s Disease

CTE

chronic traumatic encephalopathy

cryo-EM

cryo-electron microscopy

C_sat

equilibrium solubility

EMDB

electron-microscopy data bank

Gnd·HCl

guanidinium hydrochloride

HPLC

high-performance liquid chromatography

IAPP

islet amyloid polypeptide

IC₅₀

half-maximal denaturant concentration

NFT

neurofibrillary tangle

NS-TEM

negative stain transmission electron microscopy

PDB

protein-data bank

PF

protofilament

PHF

paired helical filament

PSP

progressive supranuclear palsy

PrP

prion protein

RMSD

root-mean-square deviation

ThT

thioflavin T.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.6c04928.

• Detailed experimental procedures; NS-TEM; cryo-EM biophysical characterization; supplemental figures; cryo-EM data were deposited into the Protein Data Bank (PDB) under accession codes 11LU (7 d polymorph 1); 11LV (42 d polymorph 2); 11LW (42 d polymorph 3) and into the Electron Microscopy Data Bank (EMDB) under accession codes EMD-75820 (7 d polymorph 1); EMD-75822 (42 d polymorph 1); EMD-75823 (42 d polymorph 2); EMD-75824 (42 d polymorph 3) (PDF)

#.

W.C.P. and N.Y. contributed equally.

The authors declare no competing financial interest.