Structures and Dynamics of Tau Assemblies From Solid-State NMR

Conspectus

Aggregation of the microtubule-associated protein tau into β-sheet fibrils is a hallmark of many neurodegenerative diseases. Understanding the molecular mechanism of tau aggregation requires elucidating the structure and dynamics of fibrillar tau as the end product of aggregation, membrane-bound tau involved in nucleation and intercellular transmission of the aggregates, and microtubule-bound tau as the physiological state of the protein. Using solid-state NMR spectroscopy, we have obtained detailed information about these tau assemblies. Full-length tau fibrils formed in the presence of heparin adopt homogeneous structures that depend on the number of microtubule-binding repeats and that differ from ex vivo tau fibril structures. Phospho-mimetic mutations allowed heparin-free fibrillization, and mutation of the PHF1 epitope yielded a three-layered rigid core that resembles the structures of four-repeat (4R) tau in tauopathies. This three-layered structure is also adopted by other posttranslational modification (PTM) mutants, indicating that tau’s PTM code contains redundancy and dominance. In all tau fibrils, the rigid core represents only a fifth of the protein. Truncation of the disordered regions accelerated fibrillization, but the fibril fold is sensitive to temperature, pH and ionic condition of the environment. In a short tau construct, acidic pH stabilized a flat-ribbon dimer structure whereas neutral pH stabilized the twisted C-shaped fold of Alzheimer’s disease (AD) tau. The reproducible in vitro reconstruction of AD-fold tau opens the path for studying small-molecule binding to AD tau for diagnostics and therapeutics. However, seeding experiments indicate that full-length tau adopting the AD fold lacks seeding potency, therefore the rigid core structure alone does not ensure prion-like propagation of pathological tau. An essential property to replicate is likely the fuzzy coat dynamics. Solid-state NMR data show that tau dynamics depends on PTMs, and the most dynamic segments in three-layered tau lie in the proline-rich region, suggesting that separation of this region from the rigid core may be important for prion-like propagation of tau aggregates. Cholesterol-rich high-curvature lipid membranes alone induce tau fibrils and allow their insertion into the membrane, supporting the model that lipid membranes are involved in the nucleation and transmission of tau aggregates. Finally, solid-state NMR data show that the highest-affinity microtubule-anchoring domain of tau is the R’ segment that is N-terminal to the PHF1 epitope, suggesting that stabilizing this segment may inhibit subsequent aggregation. These solid-state NMR data of multiple types of tau assemblies have provided numerous insights into the structures and dynamics of tau in pathology and physiology, advancing our reconstruction of the aggregation mechanism of tau in neurodegeneration.

Conspectus graphic

1. Introduction

The mechanism of protein aggregation in neurodegenerative diseases is a topic of major interest⁵. High-resolution structures of patient-brain extracted amyloid fibrils⁶ showed that some amyloid proteins adopt unique structures in disease⁷. This finding opens the question of how natively disordered proteins transform to unique β-sheet structures in brain, especially since some amyloid peptides have a high propensity for structural polymorphism. One approach to answer this question is to conduct precisely controlled and reproducible assembly of these proteins in vitro to attain the same structures as found in diseased brain. This in vitro reconstruction of the pathological fibril structures should provide insight into the protein sequences and environmental conditions that are required to form disease-specific amyloid folds. Answering this question also requires investigation of the site-specific conformational dynamics of amyloid proteins, because disordered segments that surround the rigid fibril cores commonly contain posttranslational modifications (PTMs) that are implicated in pathogenesis⁸.

We decided to investigate the aggregation mechanism of tau, a central protein in about 20 neurodegenerative diseases. Tau is an abundant microtubule-associated protein that assembles and stabilizes microtubules⁹. In this role, tau is disordered and highly soluble; but in disease, tau aggregates into neurofibrillary tangles and other insoluble deposits, impairing cognitive and motor functions^10–12. Tau tangles are one of the two hallmarks of Alzheimer’s disease (AD), whose stereotypical spread in the brain is the basis of staging of AD^13–15. To date, cryoelectron microscopy (cryoEM) structures of tau aggregates in many tauopathies show that each disease has a single fibril fold, suggesting that there may be a unique combination of protein sequence and environmental factors that leads to the specific aggregate structure. Identifying these conditions is thus crucial for therapeutic intervention.

Electrostatic interaction likely lies at the heart of tau’s aggregation mechanism, because tau is highly charged: the central half of the protein consists of cationic microtubule-binding and proline-rich regions which are flanked by anionic N- and C-terminal domains (Fig. 1a). Numerous phosphorylation, acetylation, and other PTMs, which reduce the positive charges, occur in pathological tau and correlate with disease progression¹⁶. But how individual PTMs induce specific fibril folds are not known. Due to its net positive charge at neutral pH, tau interacts with many negatively charged cellular species such as lipid membranes¹⁷, microtubule surfaces¹⁸ and RNA. The membrane interaction has been implicated in the formation and trans-neuronal propagation of tau aggregates¹⁹, but detailed structural information is lacking.

Figure 1.

Overview of ssNMR studies of structures and dynamics of tau assemblies. (a) Domain nomenclature and electrostatic charges of 0N4R tau. The Pro-rich region (P1 and P2) and microtubule -binding repeats (R1–R4, R’) are positively charged at pH 7.0 whereas the N-terminal domain (NT1 and NT2) and C-terminal domain (CT) are negatively charged. 0N4R tau lacks two N-terminal inserts found in the longest isoform, 2N4R. (b) Using ssNMR, we have investigated wild-type (WT) and phospho-mimetic (P) tau fibrils, tau-membrane complexes, and microtubule-bound tau. “WT Isoform Mixing”, Reproduced from³, available under CC BY 4.0 license. “WT Lipids”, Reproduced from⁴⁴, available under CC BY 4.0 license. “WT Microtubules”, Reproduced from⁴, available under CC BY 4.0 license. “P Lipid”, Reproduced with permission from⁵¹, Copyright 2026 ACS. “P Fibrils”, Reproduced from¹, available under CC BY 4.0 license. “P Dynamics”, Reproduced with permission from²⁹, Copyright 2026 ACS.

In addition to PTM-induced sequence variation, tau exists in six isoforms in the adult human brain, which are chiefly differentiated by whether the protein contains four or three microtubule-binding repeats. Tauopathies are distinguished by the isoform composition of the aggregates. For example, progressive supranuclear palsy (PSP) is a four-repeat (4R) tauopathy whereas AD is a mixed-isoform tauopathy with equal amounts of three-repeat (3R) and 4R tau. Understanding the mechanism of tau aggregation is thus a multidimensional puzzle that must account for protein PTMs, isoform composition, and a variety of environmental factors such as salt, pH, and lipid membranes (Fig. 1b).

NMR spectroscopy has been used to study tau structure and dynamics since the 2000’s^20–22. Early studies employed the polyanionic cofactor heparin to induce fibrillization and truncated constructs such as K18 and K19 that end before the pseudo-repeat R’ to facilitate data analysis^23–24. The discovery that the rigid core of AD paired helical filament (PHF) tau extends into R’⁷ marked a turning point in tau structural biology, calling for studies of the entire rigid cores found in brain tau aggregates. We therefore set out to employ advanced magic-angle-spinning (MAS) solid-state NMR (ssNMR) spectroscopy to investigate long tau constructs, including full-length tau of both 4R and 3R isoforms and PTM-mimetic tau fibrils formed without anionic cofactors. The overarching goal is to reconstruct the pathological structures and dynamics of brain tau aggregates in vitro, in doing so deconstructing the chemical code of tau aggregation.

2. Structures of full-length tau fibrils reveal the impact of PTMs on aggregation

We first studied unmodified full-length 0N4R and 0N3R tau fibrillized with heparin, a polyanionic sulfated glycosaminoglycan commonly used to trigger aggregation of the highly soluble tau²⁵. Reactions were conducted in phosphate buffer saline under shaking and reducing conditions. Low protein concentrations of 0.4–1.0 mg/mL (10–25 μM) were used together with 0.125–0.16 mg/mL heparin. We used thioflavin T fluorescence to follow fibrillization kinetics, trypsin digestion to map the protease-resistant core, and negative-stain transmission electron microscopy (TEM) to observe fibril morphology. MAS ssNMR was used to obtain residue-specific backbone conformation, three-dimensional (3D) fold, and segmental dynamics. 0N4R tau fibrillized with a short lag time of 1 hr²⁶ whereas 0N3R tau had a much longer lag time of 33 hr²⁷, indicating that R2 caused major differences to fibril folds. Both proteins formed straight monomorphic filaments, which are 13 nm wide for 0N4R tau and 23 nm wide for 0N3R tau. Mass spectra after trypsin digestion showed a peak at 8456.58 Da for 0N4R tau, predicting residues I260-K340 to be the rigid core, whereas 0N3R tau showed a peak at 15,106 Da, predicting a massive core for residues K268-L441. The latter was surprising, as no tau fibrils had been known at that time to recruit CT into the rigid core.

The trypsin digestion data turned out to be in remarkable agreement with ssNMR data. Assignment of rigid-selective 3D correlation spectra showed that the 0N4R tau fibril core spans R2, R3 and R4 (S262-T361), whereas the 0N3R core spans R3 to the C-terminus of the protein, as well as a small portion of R1 (S262-L441) (Fig. 3a, 3b). Definitive proof for the inclusion of CT in the 0N3R core was provided by mobile-selective 2D spectra, which lacked CT signals, in contrast to 0N4R tau. For both proteins, rigid-selective NMR spectra show a single set of chemical shifts with narrow linewidths, indicating that there is a single molecular conformation, consistent with the uniform fibril morphology. However, when proteolysis occasionally occurred during fibrillization due to insufficient protease inhibition, polymorphic fibrils similar to those reported in the literature were observed²⁸. This serendipitous finding showed that intact proteins must be maintained throughout fibril growth, and heparin-fibrillized tau is not intrinsically polymorphic.

Figure 3.

Ser/Thr and Cys regions of 2D ¹³C-¹³C correlation spectra of tau fibrils, illustrating the sensitivity of ¹³C chemical shifts to fibril structure. (a) Heparin-fibrillized full-length 0N4R tau²⁶. (b) Heparin-fibrillized 0N3R tau²⁷. (c) Full-length AT8–3E tau. Reproduced from¹, available under CC BY 4.0 license. (d) Full-length PHF1–4E tau¹. (e) ΔD421 tau (1–421). Reproduced from⁵³, available under CC BY 4.0 license. (f) Heparin-fibrillized P2R tau (198–399) at 12°C⁵². (g) Flat-ribbon tau (297–391)². (h) AD-fold 4E tau (297–407)³⁴. The rigid-core structures are color-coded in the same way as in Fig. 2.

We determined the 3D folds of 0N4R and 0N3R tau by analyzing long-range correlation peaks in 2D and 3D NMR spectra, which constrain distances between residues well separated in the amino acid sequence. The 0N4R tau fibril core consists of a β-arch formed by R2–R3, whereas 0N3R tau folds into a multi-layered structure containing an R3–R4 β-arch and a long CT strand that stacks against R1–R3 (Fig. 2b). The CT stacking with R1–R3 is reminiscent of transient CT-repeat interactions detected in soluble tau by fluorescence and electron paramagnetic resonance experiments. The fact that this contact is trapped in a rigid fibril core likely explains the long lag time of 0N3R tau. Because CT is chemically heterogeneous in brain due to PTMs, its absence from all brain tau fibril cores known to date can be understood. The CT-including 0N3R tau fibril core thus suggests that CT may play a protective role by folding back over R3 and R4, preventing side-on nucleation and pathological aggregation.

Figure 2.

Three-dimensional folds of tau assemblies determined using ssNMR and cryoEM. (a) Ex vivo AD and PSP tau folds^6–7. AD is a mixed isoform tauopathy whereas PSP is a 4R tauopathy. (b) Rigid core folds of in vitro fibrillized tau determined by ssNMR and cryoEM^{1–2, 26–27, 34, 52–53}. Reproduced from¹, available under CC BY 4.0 license. (c) Microtubule -bound tau domain⁴. Right column shows the immobilized domains whereas left column illustrates the folds with matching colors for the repeats.

Like all brain tau fibrils, the rigid core of 0N4R tau accounts for less than a quarter of the entire protein. While the rigid core is structurally homogeneous, the disordered remainder of the protein is dynamically heterogeneous²⁶. Measurement of peak intensities and order parameters showed that R1 and R4 contain semi-rigid β-sheet residues, the Pro-rich region is anisotropically mobile with intermediate order parameters of 0.2–0.6, whereas the N- and C-terminal domains are nearly isotropically dynamic. These experiments laid the groundwork for subsequent investigation of the fuzzy coat dynamics in phospho-mimetic tau fibrils²⁹.

Since tau is heavily modified in brain, we next investigated whether installing PTM mimetics that reduce the positive charges can cause full-length tau aggregation without heparin. Based on proteomics studies of the frequency and occupancy of tau PTMs at different stages of AD¹⁶, we chose three prominent modifications that occur early in AD: AT8 phosphorylation, PHF1 phosphorylation, and ΔD421 truncation. We mimicked AT8 phosphorylation by mutating three Ser/Thr residues in the Pro-rich region to Glu (S202E, T205E, S208E), and PHF1 phosphorylation by installing four Glu mutations at the R’-CT junction (S396E, S400E, T403E, S404E). These 0N4R tau mutants fibrillized without cofactors in two weeks under shaking and reducing conditions. Interestingly, the two proteins showed distinct NMR fingerprints (Fig. 3c, 3d), indicating that they have distinct 3D folds. CryoEM helical reconstruction yielded 2.5 Å density maps for both proteins, showing that AT8–3E tau adopted a large multi-layered triangular core whereas PHF1–4E tau formed a three-layered structure equally divided among R2, R3 and R4 (Fig. 2b). Importantly, this three-layered fold closely resembles PSP and CBD tau: the only deviations occur at residues near unidentified densities in cryoEM maps, suggesting that these unknown cofactors may be responsible for the departure of brain tau structures from the ideal tripartite structure formed by these pseudo-repeats.

Joint analysis by ssNMR and cryoEM provided information that is difficult to obtain by either method alone. Chemical shift assignment yielded the β-strand positions in the amino acid sequence, which are in excellent agreement with cryoEM reconstructions for both AT8–3E and PHF1–4E tau. A single set of chemical shifts was found, indicating that both proteins contain a single polymorph. This removes the concern that cryoEM reconstructions based on ~3% of all particles may not represent the entire structural ensemble of these fibrils. Conversely, cryoEM analysis obviated the need for measuring numerous long-range correlations by NMR, dramatically accelerating the determination of the 3D fold.

Interestingly, when we combined the AT8 and PHF1 phospho-mimetic mutations to create a 7E tau, the protein showed identical 2D NMR fingerprints as PHF1–4E tau, indicating that 7E tau adopts the same three-layered fold. This result reveals an intriguing dominance and redundancy in the PTM code of tau, with PHF1 phosphorylation overriding AT8-driven aggregation. Given the large number of phosphorylation sites in tau, we hypothesize that this PTM dominance and redundancy might contribute to the small number of pathological tau folds in tauopathies.

The redundancy of tau’s PTM code was also manifested in the truncation mutant ΔD421, a PTM that occurs early in AD. We purified ΔD421 tau(1–421) without (WT) or with AT8–3E mutations. Both proteins formed cofactor-free fibrils, but their structures diverged sharply. For ΔD421-WT tau, 2D NMR fingerprints matched the PHF1–4E spectra (Fig. 3e), and cryoEM reconstruction confirmed the three-layered structure, superimposable with the PHF1–4E structure (Fig. 2b). In contrast, ΔD421–3E tau has a distinct rigid core from R4 to CT (N359-I417). Again, because all known brain tau fibrils exclude CT, this result implies that AT8 phosphorylation and ΔD421 truncation do not co-occur in pathology or are overridden by other PTMs.

Because ΔD421 truncation increased the positive charge while PHF1–4E mutation decreased it, the fact that both PTM mimics adopt the same structure implies that tau fibril folds are not solely determined by electrostatics but are more broadly responsive to the region of the protein altered by PTMs. The fact that three C-terminal modified tau all aggregated into the same three-layered structure that resembles the PSP, GPT, CBD, and AGD tau folds⁷ indicates that a PTM-driven chemical code with built-in redundancy exists in tau aggregation. In contrast, charge modification in the Pro-rich region directs tau assembly to distinct folds centered around R’-CT, which do not match any disease tau structures. These results raise the possibility that many tau fibril folds may emerge in cells but only a few polymorphs could spread effectively in the brain in a prion-like manner.

3. Truncated tau fibril structures demonstrate the influence of salt and pH on aggregation

Because WT 0N4R and 0N3R tau fibril structures do not match brain tau structures, we asked whether the large disordered N- and C-terminal domains and the Pro-rich region might inhibit fibrilization in distinct ways from those of brain tau, which is subject to numerous cellular interactions that are not replicated in vitro. We hypothesized that shortening the disordered region might better coax the aggregation-prone region of tau to fold into structures similar to those found in brain. To test this hypothesis, we progressively shortened the flanking regions while maintaining the full repeats found in brain tau aggregates. A second motivation for studying truncated tau is to find a robust protocol for large-scale production of AD-fold tau fibrils in vitro, to enable drug discovery and mechanistic studies without relying on scarce patient brain material.

We first removed half of the disordered region of tau while retaining P2, R1–R4, and the pseudo-repeat R’. This P2R tau (residues 199–399) still required heparin to fibrillize, but fibrilization was much faster than full-length tau, confirming the inhibitory effect of the flanking regions. At 37°C, P2R tau formed amorphous aggregates, unlike 0N4R tau, and the reaction required lower temperatures to yield ordered filaments. At 24°C the P2R tau filaments were well dispersed, while at 12°C the fibrils became bundled. Unexpectedly, ssNMR spectra showed large chemical shift perturbation for residues around L284 between the two temperatures, indicating a turn conformation at 24°C but a β-strand conformation at 12°C (Fig. 3f). Structure calculation using distance and torsion angle restraints found that R2 folded into a β-arch at 24°C but a straight β-strand at 12°C, which stacks in antiparallel with R2 of a second protofilament to form a dimeric core (Fig. 2b). Thus, the R2 repeat has larger conformational plasticity than R3, which is structurally invariant between the two temperatures. This result suggests a previously unrecognized reason for the variable folds among brain 4R tau aggregates and highlights the region around L284 as a dynamic hotspot for modulating fibril folds.

Soon after the P2R tau study, Scheres and coworkers reported in vitro assembly of AD-fold tau³⁰ using a construct from the middle of R2 to the middle of R’ (residues 297–391)³¹. This result opened the possibility to obtain NMR fingerprints of AD-fold tau for drug discovery. With only 2 mg of protein, a 2D ¹⁵N-¹³Cα (NCA) correlation spectrum can be measured in less than 12 hours to assess the fibril fold. We thus set out to fibrillize ¹³C, ¹⁵N-labeled tau (297–391) using the reported protocol, which involves shaking a pH 7.4 phosphate buffer containing high concentrations of MgCl₂. To our surprise, tau (297–391) predominantly formed flat ribbons instead of twisted filaments under these conditions, which cannot be studied by cryoEM. We thus turned to ssNMR, using mixed ¹³C and ¹⁵N-labeled samples that distinguish intramolecular and intermolecular contacts. The 2.1 Å structure shows a β-arch from S305-L357 in R3–R4, with a turn at ³²²CGS³²⁴. Two protofilaments stack at the long β-strand from G323 to I354, creating a four-layered structure (Fig. 4a, c)². Importantly, the center of the dimer interface involves a pair of E338, whose Cδ chemical shifts indicate a protonated and neutral carboxyl group, stabilizing the dimer.

Figure 4.

AD-fold tau and flat-ribbon dimer tau have distinct NMR fingerprints. (a) 2D NCA spectrum of tau (297–391), which adopts a four-layered dimer fold. Reproduced from², available under CC BY 4.0 license. (b) 2D NCA spectrum of 4E tau (297–407), which adopts the AD fold. Reproduced with permission from³⁴. Copyright 2026 JBC. (c) Secondary Cα and Cβ chemical shifts of the tau (297–391) dimer. Reproduced from², available under CC BY 4.0 license. (d) Cα and Cβ secondary chemical shifts of AD-fold 4E tau (297–407). Reproduced with permission from³⁴. Copyright 2026 JBC.

The reason for the failure to reproduce the twisted PHF-like filaments became clear upon analysis of the pH and salt equilibria of the reaction mixture. The high concentrations of MgCl₂ precipitated the phosphate ions to form insoluble magnesium phosphates, acidifying the solution. Indeed, all reactions drifted to acidic pH from the neutral pH initial condition. This acidic pH protonated Glu residues in R4, stabilizing the tightly packed four-layered dimer fold. Therefore, nominally identical fibrillization conditions can have significantly different chemical compositions, leading to distinct fibril polymorphs^32–33.

To reproducibly assemble tau into the AD fold, we next tested a longer construct, 4E tau (297–407), which contains the same PHF1 phospho-mimetic mutations as PHF1–4E 0N4R tau³⁴. We systematically varied buffer, salt concentration, protein concentration, pH, and order of mixing of reagents. Interestingly, while tau (297–391) fibrilization is sensitive to small environmental differences and can form flat or twisted fibrils, 4E tau (297–407) reproducibly formed twisted filaments under a variety of buffer and mixing conditions, and the solution pH remained neutral during the reaction. 2D NCA fingerprints showed a boomerang-like pattern for the Gly region, which differs from the rectangular pattern for the flat-ribbon dimer fold. CryoEM reconstruction of the twisted 4E tau (297–407) filaments confirmed that the boomerang-Gly fibrils adopted the C-shaped structure, which associate into paired, triple and quadruple filaments. The reproducible milligram-scale assembly of AD-fold tau and the discovery of its NMR chemical shifts allowed us to subsequently determine the binding site of a PET ligand³⁵.

4. Seeded tau structures test the model of prion-like propagation

Seeding recombinant tau with brain-derived aggregates is a powerful approach for investigating the mechanism of prion-like propagation of pathological aggregates in brain. Faithful amplification of amyloid structures requires preformed fibrils to convert soluble monomers into the same structure as the seed. Extensive biophysical studies have probed the mechanisms of templated growth of amyloid fibrils, including secondary nucleation, elongation and fragmentation^36–38. Given the many mechanisms for templated growth, the sequence heterogeneity of tau, and the complexity of the cellular environment, how structural fidelity is achieved in vivo to result in a single tau aggregate structure in each tauopathy remains poorly understood.

We first used seeding to investigate the isoform mixing pattern of AD PHF tau³. AD tau aggregates are a mixture of 3R and 4R tau. Whether these two isoforms are phase separated, obligatorily alternating, or randomly mixed in the filament was not known structurally³⁹. To answer this question, we seeded 1:1 mixtures of ¹⁵N-labeled monomer of one isoform and ¹³C-labeled monomer of the other isoform with 10% sarkosyl-insoluble AD tau and measured intermolecular ¹³C-¹⁵N dipolar couplings. With ten-fold excess monomer to seeds, half of the recombinant tau fibrillized, indicating a five-fold increase of fibril mass. These seeded fibrils aggregated WT mouse tau, indicating that they are pathologically active. Electron microscopy data showed that the templated fibrils have the same twist and C-shaped structure as AD PHF tau⁴⁰, and ssNMR spectra confirmed that the ¹³C, ¹⁵N-labeled recombinant tau indeed converted to β-sheet structures.

Using these structurally accurate and biochemically active AD-PHF seeded tau, we measured intermolecular ¹⁵N-¹³C dipolar coupling between ¹³C-labeled tau of one isoform and ¹⁵N-labeled tau of the other isoform. To account for the multi-spin environment of each protein, we measured dipolar dephasing of unseeded 0N4R tau fibrils with varying ratios of ¹³C and ¹⁵N labels. Higher concentrations of ¹³C-labeled 0N4R tau caused larger dephasing of ¹⁵N-labeled tau. Calibrated by these data, we found that AD PHF-seeded fibrils showed 37% probability for 4R tau to be next to 3R tau and 56% probability for 3R tau to be next to 4R tau (Fig. 5b). Therefore, the two isoforms are neither phase-separated nor obligatorily alternating in AD tau filaments but are fluently mixed, and there is a small preference for 4R tau in the filament. It’s tempting to speculate that this fluent mixing may facilitate the propagation of AD tau in brain and contribute to the prevalence of AD among neurodegenerative disorders.

Figure 5.

Investigating tau isoform mixing (a-b) and fuzzy coat dynamics (c-e). (a) 3R (cyan) and 4R (orange) tau are fluently mixed in AD PHF tau. Top view shows the AD tau structure. Reproduced from³, available under CC BY 4.0 license. (b) ¹⁵N-¹³C dipolar dephasing of AD PHF seeded recombinant tau fibrils shows that 4R tau is followed by 3R tau with 37% probability whereas 3R tau is followed by 4R tau with 56% probability. Reproduced from³, available under CC BY 4.0 license. (c) ¹⁵N INEPT intensities of PHF1–4E 0N4R tau fibrils measured under MAS. Reproduced with permission from²⁹. Copyright 2026 ACS. The highest intensity signals are found in NT2 and the Pro-rich region, indicating these are the most dynamic regions of the fuzzy coat. (d) Representative 2D INEPT spectrum. Reproduced with permission from²⁹. Copyright 2026 ACS. (e) Fuzzy coat topology of the three-layered PHF1–4E tau fibril. Side view illustrates the asymmetric dynamics of the fuzzy coat, with P1 and P2 being the most dynamic. Reproduced with permission from²⁹. Copyright 2026 ACS.

Since the first-generation seeded tau have the same structure and pathological activity as AD PHF tau, we next asked whether the templated fibrils can seed a second generation of monomers. To our surprise, despite their correct structure, the gen-1 fibrils were unable to aggregate new monomers⁴⁰, indicating that these gen-1 fibrils miss one or more essential properties of AD PHF tau. The missing property could be specific PTMs in the fuzzy coat or unidentified cofactors. Because gen-1 fibrils aggregated mouse tau, we must conclude that mouse neurons either modified these fibrils or provided the missing cofactors to enable further templating. This conclusion is partially supported by cryoEM data that the cellular environment affected the structures of seeded fibrils⁴¹. The fact that replicating the rigid-core structure is insufficient for reproducing all the pathological properties of AD PHF tau has profound implications, as it challenges the notion that tauopathies are defined by structure alone⁷. These results have stimulated recent efforts to investigate site-specific dynamics of the fuzzy coat of phospho-mimetic tau fibrils²⁹. By measuring intensities of dynamic residues, order parameters and polarization transfer rates, we found that the most dynamic portion of the fuzzy coat for the three-layered PHF1–4E tau lies in the Pro-rich region, whereas the N- and C-terminal segments have intermediate mobilities (Fig. 5c, d). These results led to a new fuzzy coat model that revises the previous two-layered polyelectrolyte brush model⁴²: instead of segregating the intrinsically disordered region by charges into two layers, the anionic N- and C-terminal regions stack together whereas the highly mobile cationic Pro-rich region coats the rigid core (Fig. 5e). This fuzzy coat model suggests that propagation of full-length tau fibrils may proceed significantly by elongation of the fibril ends, promoted by detachment of the cationic Pro-rich domains from the rigid core and attachment of the anionic N- and C-termini to the core.

5. Structure and dynamics of membrane-bound and microtubule-bound tau

While studying mature fibrils provides insights into the tau aggregation mechanism, investigating membrane-bound tau offers clues to tau nucleation and intercellular transmission. The negatively charged membrane surface attracts and concentrates positively charged tau, which may initiate aggregation. Lipid membrane is also a barrier through which tau species, whether oligomers or mature aggregates, may need to cross in order to spread between cells. Evidence that tau has strong affinity for lipid membranes has been long reported in electron micrographs showing AD PHF tau stemming from ER membranes¹⁷ and HPLC data showing tau-bound phosphatidylcholine, cholesterol and other lipids⁴³.

We first investigated tau binding to membranes with different curvatures and lipid compositions, to identify whether membrane alone can induce tau fibrillization⁴⁴. We prepared high-curvature small unilamellar vesicles (SUVs) and low-curvature large unilamellar vesicles (LUVs) and multilamellar vesicles (MLVs). A mixture of POPC, POPE, POPS and cholesterol was used to mimic eukaryotic membranes. After incubating lipid vesicles with soluble P2R tau (198–399) at 37°C overnight, the protein co-sedimented with lipids quantitatively, indicating high affinity to the membrane. The sedimented mixtures were followed over several weeks by NMR. The freshly prepared samples were highly dynamic, with weak intensities in rigid-selective ¹³C NMR spectra. Over time, however, β-sheet signals grew. Interestingly, the equilibrated spectra depended on membrane curvature: SUV-bound tau developed β-sheet chemical shifts for residues enriched in the repeats but lacked β-sheet Ala signals. The SUV sample also exhibited the hallmark chemical shifts of fibrillar Y310 in the R3 hexapeptide VQIVYK. In comparison, LUV and MLV samples exhibited β-sheet signals of Ala residues in R′, indicating immobilization of R’, but lacked the fibrillar Y310 chemical shifts. Removal of cholesterol abolished the fibrillar Y310 signal, indicating that both membrane curvature and cholesterol are necessary for fibrilization. Finally, TEM images of tau incubated with cholesterol-containing SUVs showed 1 μm long filaments after weeks of incubation, which are absent from MLV and LUV samples, confirming that cholesterol-rich SUVs are the membrane conditions to induce tau fibrils.

To investigate whether tau aggregation can occur in dilute vesicle solutions and whether fibril structures differ with and without lipids, we next fibrillized 4E-tau (297–407) in the absence and presence of cholesterol-containing SUVs. 2D NMR spectra show that the protein adopted the same AD fold in the presence of lipid membranes; moreover, these C-shaped tau inserts into the membrane, as shown by protein-lipid cross peaks. Analysis of lipid- and water-edited spectra showed that the R3 hexapeptide and its steric zipper in R’ are the most dehydrated and membrane-embedded residues. Given that both filaments need to satisfy these conditions, this result suggests that paired helical filament tau might bridge multiple high-curvature vesicles (Fig. 6b). Spectral comparison with solution fibrils showed that lipids caused peak doubling to residues at the protofilament interface (³²⁶GNI³²⁸ and ³⁵⁵GS³⁵⁶), and the minor conformation increased in population with lipid concentration, indicating that the dimer interface of PHF tau is susceptible to membrane perturbation. Moreover, the lipid-perturbed residues map onto the same region that differs between AD tau and chronic traumatic encephalopathy type I tau, suggesting the intriguing possibility that lipid membranes might contribute to the structural differences between tau folds. These studies demonstrate that lipid membranes can nucleate tau assembly via curvature and cholesterol.

Figure 6.

Tau binding to lipid membranes and microtubules studied by ssNMR. (a) TEM images of tau fibrils induced by cholesterol-rich SUVs. Reproduced from⁴⁴, available under CC BY 4.0 license. (b) 2D NCA spectra of 4E tau (297–407) bound to high-curvature SUVs (magenta) versus in solution (cyan). Reproduced with permission from⁵⁴. Copyright 2026 ACS. (c) Model of the orientation and insertion of AD-fold tau in high-curvature membranes, based on the measured chemical shifts and water and lipid exposure of the protein. Reproduced with permission from⁵⁴. Copyright 2026 ACS. (d) 2D NCA spectrum of tau (198–399) bound to microtubules at 1:1 ratio. Reproduced from⁴, available under CC BY 4.0 license. The peaks were assigned to R’ and R4 residues. (e) R’ is the anchor of tau on the microtubule surface, whereas other domains become progressively more mobile toward the Pro-rich region. Zoomed-in view shows the structural model of R’(369–395) docked into the low-resolution cryoEM map of WT tau. Reproduced from⁴, available under CC BY 4.0 license.

Tau’s functional partners in the cell are microtubules. The first cryoEM maps of microtubule-bound tau¹⁸ showed continuous densities on the microtubule surface but could not definitively pinpoint which repeats in native tau occupy the densities due to low map resolution. Instead, chimeric constructs that replicate R2 four times and R1 four time were used in place of native tau to reach high-resolution maps. These data led to the conclusion that R1 and R2 are responsible for binding microtubules. However, when we bound 0N4R tau to taxol-stabilized microtubules, rigid-selective 2D NCA spectra showed a small number of signals that can be assigned to R’ and the C-terminal portion of R4 (Fig. 6d). At the same time, mobile-selective spectra detected signals assignable to P2 and R1. When tau was co-assembled with tubulin to create dynamically unstable microtubules, the R4 signals decreased while the R′ signals remained, indicating that R′ has the highest affinity for microtubules. The assigned R’ residues (K369-K395) fit into the cryoEM map of WT tau well (Fig. 6e), confirming that in native tau where all five repeats are present, R′ outcompetes the other four repeats to anchor onto the microtubule. Comparison of the amino acid sequences of these five repeats shows that R′ and R4 have higher charge density than R1–R3, which may explain their stronger electrostatic attraction with the microtubule. Interestingly, the last two residues of the microtubule-anchored R’ segment, ³⁹⁴YK³⁹⁵, ends immediately before the first residue of the PHF1 epitope, S396, strongly suggesting that PHF1 phosphorylation may weaken tau anchoring to the microtubule, initiating the cascade of self-assembly events.

Conclusions and Outlook

These studies of tau assemblies have yielded numerous insights into the conformational propensity, structural plasticity and dynamic properties of this central protein in neurodegeneration. Driven by electrostatic attraction, WT 0N4R and 0N3R tau self-assembles into homogeneous and monomorphic fibrils in the presence of heparin. 0N4R tau adopts an R2–R3 β-arch that is seen in many 4R tauopathy folds, in which R2 is conformationally more plastic than R3. Charge-modifying phospho-mimetic mutations and truncations in the C-terminal region led to three-layered tripartite folds that closely resemble PSP tau. Heparin-fibrillized 0N3R tau and AT8–3E 0N4R tau fibril cores both include CT, which shields the aggregation-prone repeats, suggesting that these folds, if found in vivo, may be protective or less prone to spreading. More extensively truncated tau can form the AD fold under certain pH and ionic conditions. But replicating the C-shaped AD fold does not automatically lead to serial amplification of the aggregates. Thus, specific fuzzy coat dynamics, cofactors, or environmental conditions must be present to reproduce the prion-like propagation that defines pathological tau. Electrostatic interactions also drive tau interaction with lipid membranes and microtubules. Cholesterol-containing high-curvature membranes induced tau filaments adopting the AD fold, suggesting that tau mis-sorting to the presynaptic junction that is rich in small lipid vesicles¹⁹ might trigger tau aggregation in neurons. Finally, NMR data of native tau show that the R’ segment N-terminal to the PHF1 epitope is the highest-affinity microtubule-anchoring domain of the protein, suggesting that stabilizing this segment may prevent the cascade of events leading to loss-of-function and self-assembly of tau.

Tau oligomers have also been important objects of research because they arise early during aggregation, can be seeding-competent, and are frequently associated with synaptic dysfunction, cellular stress, and pathological spread^45–46. However, these oligomers are transient, heterogeneous, and difficult to define structurally. In comparison, targeting tau fibrils or seed-competent aggregates is often a stronger therapeutic strategy⁴⁷ because fibrils are more stable, better characterized, and more closely linked to neuropathological burden in disease⁷. Tau fibrils are also more amenable to conformation-selective antibodies that preferentially recognize and deplete seeding-competent tau from diseased brain while sparing control brain tau⁴⁸. In addition, tau-aggregate burden measured by positron emission tomography ligands is strongly associated with cognitive decline and has shown stronger prognostic performance than phospho-tau fluid measurement in symptomatic AD, supporting the clinical relevance of targeting aggregated tau^49–50. Thus, although oligomers may mediate early toxicity, fibrillar and other seeding-competent tau aggregates currently offer a more reproducible, structurally grounded, and clinically actionable target for intervention.

These studies demonstrate the central role of ssNMR spectroscopy for extracting a wide range of information about the structures and dynamics of tau assemblies. Rigid-selective NMR spectra reveal the number of polymorphs whereas mobile-selective spectra characterize the heterogeneous dynamics of the fuzzy coat. Structures of non-twisting filaments that are inaccessible to cryoEM reconstruction can be determined using ssNMR, and tau binding to lipid membranes and microtubules can be decisively investigated. Integrating ssNMR with cryoEM is a fruitful approach for observing the entire population of fibrils while efficiently determining the 3D folds of twisted filaments. 2D fingerprint solid-state NMR spectra can now be used as a structural screening tool³⁴. Complementarily, solution NMR experiments facilitate the study of tau dynamics by providing important reference information about soluble tau.

Biographies

N.E.M.

Nadia El Mammeri is a CNRS researcher at CBMN (CNRS-University of Bordeaux). She earned her Ph.D. in Biochemistry and Biophysics from the University of Bordeaux in 2020. From 2021 to 2025, she conducted postdoctoral research at the Massachusetts Institute of Technology in Mei Hong’s laboratory, developing expertise in integrative structural biology, especially combining cryoEM and solid-state NMR. She started her independent research program at CNRS in 2025, focusing on neuronal aging-related amyloid proteins. Her group combines cryoEM with solid-state NMR spectroscopy to characterize the structures and dynamics of these assemblies and to capture their conformational diversity in function and pathology.

M.H.

Mei Hong is the David A. Leighty (1970) Professor of Chemistry at the Massachusetts Institute of Technology (MIT). After receiving her Ph.D. at the University of California Berkeley and conducting postdoctoral research at MIT, she joined the faculty at Iowa State University as an Assistant (1999), Associate (2003) and Full (2005) Professor. She returned to MIT in 2014 and is currently the Director of the Francis Bitter Magnet Laboratory and Associate Editor of the Journal of the American Chemical Society. Mei Hong develops and applies advanced solid-state NMR spectroscopic techniques to elucidate the structure and dynamics of biological macromolecules, including membrane proteins, amyloid proteins, and plant cell walls.