Structure of the nonhelical filament of the Alzheimer’s disease tau core

Significance

The protein tau aggregates into different fibrillar structures in different neurodegenerative diseases. The molecular basis for this structural polymorphism is unknown. Tau (297-391) was reported to form twisted filaments that resemble Alzheimer’s disease tau. We find that tau (297-391) predominantly forms non-twisting ribbons under high MgCl2 concentration and acidic pH. Solid-state NMR chemical shifts, distance restraints, and water-edited data, together with symmetry information from a recent cryoEM structure, yielded a dimeric structure in which the dimer interface is stabilized by protonated and neutral Glu residues. These results demonstrate that tau aggregation is sensitive to pH and ionic conditions, and the structure of non-twisting amyloid fibrils that are difficult to solve by cryoEM can be determined by solid-state NMR.

Abstract

The microtubule-associated protein tau aggregates into neurofibrillary tangles in Alzheimer’s disease (AD). The main type of aggregates, the paired helical filaments (PHF), incorporate about 20% of the full-length protein into the rigid core. Recently, cryo-electron microscopy data showed that a protease-resistant fragment of tau (residues 297–391) self-assembles in vitro in the presence of divalent cations to form twisted filaments whose molecular structure resembles that of AD PHF tau [S. Lövestam et al., Elife 11, e76494 (2022)]. To investigate whether this tau construct is uniquely predisposed to this morphology and structure, we fibrillized tau (297–391) under the reported conditions and determined its structure using solid-state NMR spectroscopy. Unexpectedly, the protein assembled predominantly into nontwisting ribbons whose rigid core spans residues 305–357. This rigid core forms a β-arch that turns at residues ³²²CGS³²⁴. Two protofilaments stack together via a long interface that stretches from G323 to I354. Together, these two protofilaments form a four-layered β-sheet core whose sidechains are stabilized by numerous polar and hydrophobic interactions. This structure gives insight into the fibril morphologies and molecular conformations that can be adopted by this protease-resistant core of AD tau under different pH and ionic conditions.

The microtubule-associated protein tau (1) aggregates into filamentous intracellular inclusions in many neurodegenerative diseases, chief among which is Alzheimer’s disease (AD) (2, 3). Inhibiting and disassembling these aggregates requires a molecular understanding of the pathways of tau aggregation and the structures of the aggregates. For this purpose, an experimental model that reproducibly replicates the intraneuronal tau aggregates is highly desirable. In AD, the rigid core of PHF tau adopts a C-shaped molecular structure that spans residues 306–378 (4, 5), representing 19% of the full-length 0N4R tau. Two C-shaped protofilaments stack with pseudo 2₁ screw symmetry, with the end of the microtubule-binding repeat 3 (R3) and the beginning of R4 constituting the dimer interface. In this rigid core, the amyloidogenic R3 hexapeptide, ³⁰⁶VQIVYK³¹¹, resides in the outer ring of the C-shape and stacks its sidechains with those of the first ten residues of the R’ domain in the inner ring. Outside this rigid core, the protein is dynamically disordered, and how this disordered fuzzy coat interacts with the rigid core and shapes its formation are not known. To understand the molecular mechanism of tau aggregation in AD, and to design therapeutic agents that target the PHF tau, an accurate in vitro model of AD tau would be immensely useful. Recently, a short protease-resistant fragment of AD tau, tau (297–391) (Fig. 1A) (6, 7), was reported to self-assemble into twisted filaments whose molecular structure resembles the AD PHF tau structure. This construct starts from the last nine residues of the R2 domain, which is homologous to the corresponding residues in the R1 domain. Thus, most of the amino acid sequence of tau (297–391) is shared by both 4R and 3R tau isoforms. We sought to reproduce these PHF-like fibrils using recombinant tau (297–391) and analyze them by solid-state NMR.

Fig. 1.

Fibril morphologies of tau (297–391). (A) Amino acid sequence diagrams of full-length 2N4R tau and tau (297–391), which has an additional Met residue before residue 297. (B–G) Representative negative-stain transmission electron microscopy (TEM) images of nontwisting and twisting filaments of tau (297–391). The fibrils were grown from solutions containing 100 μM protein (B) or 400 μM protein (C–G). All reaction mixtures contain 200 mM MgCl₂, and all solutions were shaken at 37˚C for 3 d to obtain fibrils. (B) Nontwisting filaments obtained at pH 5.0 under 150 rpm shaking with a 25 mm orbit. These fibrils were harvested for the ssNMR study. (C) Nontwisting filaments obtained at pH 5.0 under 420 rpm shaking with a 5 mm orbit. (D) Nontwisting filaments obtained at pH 6.0 under 180 rpm shaking with a 2-mm orbit. Images (B–D) represent different samples. (E and F) Twisted filaments obtained at pH 6.0 under 180 rpm shaking with a 2-mm orbit. These images were taken from the same grid. Only twisted filaments were observed on this grid. (G) The solution for preparing the grid for (E and F) was stored at −80 °C and thawed, then used to prepare another TEM grid. Most filaments are nontwisting. The grid of (G) is not the same as (E and F), but the same aliquot of solution was used.

Results

Several conditions have been reported for assembling tau (297–391) into PHF-like fibrils (8–10). These conditions include 1, 2, or 4 mg/mL protein monomers in 10 mM pH 7.4 phosphate buffer with 10 or 20 mM dithiothreitol (DTT). The reaction mixtures were subject to shaking speeds of 200 to 700 rpm using different shaking apparatuses, and different salts were used during fibril growth. Salt scanning to reduce lateral bundling of the filaments showed that the addition of 200 mM MgCl₂ to the buffer led to the characteristic AD PHF tau structure in 9 out of 14 constructs, including tau (297–391) (8). We thus focused on this MgCl₂ condition to produce the PHF fold.

We aggregated tau (297–391) in 10 mM phosphate buffer at pH 7.4, to which we added 200 mM MgCl₂. The solution was shaken at moderate speeds of 150–420 rpm with orbital diameters ranging from 2 to 25 mm (Fig. 1). Because the intrinsically disordered N- and C-terminal domains are absent from the construct, tau (297–391) formed fibrils readily without polyanionic cofactors, unlike full-length tau (11, 12). However, to our surprise, this procedure resulted in predominantly nontwisting ribbons rather than twisted filaments. In negative-stain transmission electron microscopy (TEM) images, these ribbons have a width of ~14 nm (Fig. 1 B and C), show no sign of width modulation, and are prone to lateral association. The addition of 200 mM MgCl₂ broke the phosphate buffer, yielding a pH of 5–6 for the reaction mixture. This acidity is consistent with the calculated concentration of soluble phosphate in the presence of 200 mM MgCl₂. The solubility products of the magnesium salts are 1.5 × 10⁻⁶ for MgHPO₄·3H₂O, 6.3 × 10⁻²⁶ for Mg₃(PO₄)₂·8H₂O, and 5.6 × 10⁻¹² for Mg(OH)₂. The acid dissociation constants (pK_a) of phosphoric acids are 2.2, 7.2, and 12.3 (13). Based on these values, 99.9% of the 10 mM phosphate in the initial pH 7.4 buffer precipitated as MgHPO₄·3H₂O (SI Appendix, SI Methods), acidifying the solution. Adjusting the MgCl₂ solution to pH 8 before adding it to the phosphate buffer yielded a pH 6 solution without immediate precipitation. However even with this modification, we still obtained predominantly nontwisting ribbons (Fig. 1D). Attempts to increase the pH of the MgCl₂ solution above pH 8 were not successful because of excessive precipitation of Mg(OH)₂. At early stages of fibril growth, we detected a small population of twisted filaments that have a crossover distance of ~77 nm and a width modulation of 8–16 nm (Fig. 1 E and F). However, when solutions containing these twisted filaments were frozen for storage at −80 °C and then thawed for additional analysis, only nontwisting ribbons were observed (Fig. 1G). Since this sample change cannot occur at low temperature, one freeze–thaw cycle appeared to be sufficient for changing the fibril morphology. Although we do not have direct evidence that individual filaments converted from twisted to nontwisting morphologies, both morphologies exist and their populations changed readily by freeze–thawing. These observations, together with the calculated pH and solubility equilibria, indicate that the MgCl₂-containing phosphate buffer is not a stable and reproducible condition for generating twisted PHF-like tau fibrils.

Structure determination of amyloid fibrils by cryo-electron microscopy is hampered by the lack of a twist: the rotation of molecules along the hydrogen-bonded direction of a fibril vastly simplifies helical reconstruction, because a full range of molecular orientations are found within each fibril (14, 15). Therefore, most cryoEM studies select twisted filaments within their samples for analysis (16, 17), and samples containing non-twisting filaments are challenging for cryoEM structure determination (18). Since the nontwisting ribbons are the major product of these tau (297–391) fibrils obtained under these conditions, we investigated their structure using ssNMR. One-dimensional (1D) ¹³C and ¹⁵N NMR spectra and two-dimensional (2D) ¹³C–¹³C (CC) and ¹⁵N–¹³C correlation spectra (Fig. 2 and SI Appendix, Fig. S1) show narrow linewidths of 0.5 ppm for ¹³C and 1.0 ppm for ¹⁵N, indicating that the nontwisting fibril has a well-ordered molecular conformation. Three-dimensional (3D) ¹⁵N–¹³C correlation spectra (SI Appendix, Fig. S2) allowed the assignment of the chemical shifts of the rigid core (SI Appendix, Table S1). This rigid core spans residues 305–357 and consists of six β-strands in the R3 and R4 domains (Fig. 3 and SI Appendix, Table S2). The β1 and β2 strands in R3 are separated by P312, whereas the β2 and β3 strands are separated by the ³²²CGS³²⁴ triplet. The R3 VPGGG motif precedes the β4 strand, which starts at the N terminus of the R4 domain. This rigid core terminates about 22 residues before the C-terminus of the AD PHF tau rigid core. Most rigid residues in this nontwisting filament exhibit a single set of chemical shifts, except for P332, G333, and ³⁵⁵GSL³⁵⁷, which show an additional set of chemical shifts (SI Appendix, Fig. S1). Because ssNMR spectra detect all the immobilized molecules in the sample, the single set of chemical shifts for most residues indicates that the majority of the tau (297–391) fibril core adopts a single molecular conformation. This conformation is distinct from that of heparin-fibrillized full-length 0N4R tau and 0N3R tau, as manifested by the distinct chemical shifts among the three fibrils (SI Appendix, Fig. S3). For residues of tau (297–391) that are not detected in the dipolar correlation spectra, most of their signals are observed in the 2D ¹H–¹³C INEPT spectrum (Fig. 3D), indicating that the N-terminal two residues and the C-terminal 19 residues of the protein are highly mobile. The rigid core chemical shifts obtained here differ from the chemical shifts for the same construct fibrilized in the absence of salt (19), indicating that the fibril polymorph in the previous study is distinct.

Fig. 2.

Solid-state NMR spectra of nontwisting tau (297–391) fibrils. (A) 1D ¹³C and ¹⁵N CP spectra of UCN-labeled tau (297–391). (B) 2D NCACB spectra of UCN-labeled tau (297–391), showing positive intensities for the N-Cα cross peaks (orange) and negative intensities for the N-Cβ cross peaks (green). Resonance assignments obtained from 3D correlation spectra are given. (C) Aliphatic region of the 26 ms 2D CC spectrum of 50% diluted tau (297–391). This diluted sample has identical chemical shifts but narrower linewidths compared to the undiluted sample.

Fig. 3.

¹³C chemical shifts indicate that tau (297–391) nontwisting filaments contain six β-strands in the R3 and R4 domains. (A) Amino acid sequence of tau (297–391). Assigned β-sheet residues are shown in red. (B) Cβ and Cα secondary chemical shift differences, where positive difference indicates β-strand structure. Six β-strands can be identified between S305 and L357. Lines at the top indicate major long-range correlations observed from the 400 ms 2D CC spectra. The thicker the lines, the more correlations detected between the same pair of residues. (C) Chemical-shift constrained (φ, ψ) torsion angles. Error bars in the experimental torsion angles represent the uncertainty (δ) of the TALOS-N prediction. (D) 2D ¹H–¹³C INEPT spectrum, showing the presence of highly dynamic residues in tau (297–391) fibrils. The assigned residue types match the amino acid composition of the N-terminal ²⁹⁶MI²⁹⁷ and the C-terminal segment T373–E391, implying that the two termini of the protein are highly mobile.

To obtain long-range correlations that constrain the three-dimensional structure of the nontwisting tau (297–391), we measured 400 ms 2D CC spectra of a 1,3-¹³C labeled sample, a uniformly ¹³C, ¹⁵N-labeled protein diluted by an equal amount of unlabeled protein (denoted as a 50% diluted sample), and a uniformly ¹³C, ¹⁵N-labeled protein without dilution (Fig. 4 and SI Appendix, Figs. S4–S6). These spectra together yielded 21 unambiguous long-range contacts, with unique chemical shift assignments for both dimensions of the 2D spectra (SI Appendix, Table S3). In addition, we found 16 long-range contacts with ambiguous chemical shift assignment, but they can be disambiguated either based on unambiguous contacts involving the same residue pair or during structural calculation. Together with additional ambiguous contacts (SI Appendix, Table S4), these long-range contacts indicate close proximity between β1 and β4 strands, between β2 and β3 strands, and between β3 and β5 strands. To measure intermolecular contacts, we compared the 400 ms 2D CC spectra of the diluted sample and the undiluted ¹³C, ¹⁵N-labeled sample (SI Appendix, Fig. S6). Intensities of intermolecular correlation peaks in the diluted sample should decay to about half (S/S₀ ~0.5) of the signal intensities of the fully labeled sample. We found multiple intermolecular long-range contacts between β3 and β5 strands, such as S324–I354 and N327–D348 (SI Appendix, Table S5). In addition, some of the medium-range correlations between β4 residues such as V339 and Q336 are attenuated by dilution (SI Appendix, Fig. S6), implying that they are involved at intermolecular interfaces.

Fig. 4.

Representative interresidue cross peaks in the 400 ms 2D CC spectrum of 1,3-¹³C-labeled tau (297–391). Long-range contacts (i − j ≥ 5) are assigned in magenta, medium-range contacts are assigned in blue and sequential contacts are assigned in gray. (A) S341-I308 cross peaks indicate close contact of the β1 and β4 strands, while the P322–L344 cross peak indicates close contact of the R3 PGGG motif and the β4 strand. (B) V318–I328 and V318/S320–G326 cross peaks indicate a β-arch at ³²¹KCG³²³. (C) Cross peaks involving Y310. Some Val sidechains show two resolved methyl peaks, suggesting the formation of steric zippers. Although the indirect dimension’s chemical shift assignments are ambiguous, V313, T319, and V306 can be reasonably ruled out based on structural arguments, leaving V339 and V337 as the only two Val residues that can interact with Y310.

Combining these interresidue distance restraints, intermolecular distance restraints, and dihedral angle restraints (20), we calculated the structure of the nontwisting tau (297–391) ribbons using XPLOR-NIH (Table 1 and SI Appendix, Table S6). Inspired by a recent dimeric cryoEM structural model of a spindle-like twisted filament of tau (297–391) formed under similar conditions (9), we used a dimer model to accommodate the intermolecular contacts observed in the 2D spectra. A monomer model can be independently ruled out by the lack of intermolecular contacts as well as by poor agreement with the (φ, ψ) torsion angles for the R4 hexapeptide ³³⁷VEVKSE³⁴² (SI Appendix, Fig. S7). In the lowest-energy ensemble of tau (297–391) (Fig. 5B), the dimer interface spans residues G323 to I354. Each protofilament adopts a β-arch architecture with the turning point at residues ³²²CGS³²⁴. The β1 and β2 strands, separated by P312, form the exterior of the four-layered fibril core. The internal two layers are composed of β3, β4, and β5 strands. The main disordered region among these three strands is a prominent ³³¹KPGGGQ³³⁶ bulge. The side of the β5 strand that does not stack with the β3 strand of the other monomer is rich in Arg and Lys residues, suggesting that this region is unlikely to be shielded by the C-terminal domain of the protein. Between the two layers in each monomer, the β1 and β4 strands stack in an antiparallel fashion, with the R3 hexapeptide residue Y310 packing against V337 in R4, potentially mediated by CH–π interaction (Fig. 5C and SI Appendix, Fig. S8). This Y310–V337 interaction is consistent with observed cross peaks between Tyr Cβ, Cδ, and Cζ and Val Cγ in the 2D CC spectra (Fig. 4C). The V339 sidechain in β4 is in close contact with the I308 sidechain in β1. The β2 and β3 strands also stack in an antiparallel fashion, stabilized by polar interactions between S320 and S324 and by van der Waals interactions between V318 and I328.

Table 1.

NMR restraints and structural calculation statistics for nontwisting tau (297–391) filaments

	Protein
NMR distance and dihedral restraints
Distance restraints
Total NOE	373
Intermolecular correlations	44
Intramolecular	5
Dimeric interface	21
Along fibril axis	12
Intramolecular and along fibril axis	6
Interresidue correlations	329
Medium range, unambiguous (2 ≤ |i – j| ≤ 4)	200
Medium range, ambiguous	18
Long range, chemical shift unambiguous (|i – j| ≥ 5)	21
Long range, disambiguated through unambiguous peaks from the same residues	12
Long range, disambiguated through intermediate structural calculation	4
Long range, ambiguous	73
Total dihedral-angle restraints
φ	45
ψ	45
Structure statistics
Violations (mean ± SD)
Distance restraints (Å)	0.26 ± 0.06
Dihedral-angle restraints (°)	0.78 ± 0.20
Max. dihedral-angle violation (°)	8.00
Max. distance-restraint violation (Å)	1.32
Deviations from idealized geometry
Bond lengths (Å)	0.005 ± 0.000
Bond angles (°)	0.53 ± 0.01
Impropers (°)	0.32 ± 0.04
Average pairwise rms deviation (Å)*
Backbone (Å)	1.72 ± 0.22
Heavy atom (Å)	2.06 ± 0.24

^*Pairwise all-against-all rmsd was calculated among the 10 lowest-energy refined structures, excluding ³³²PGGG³³⁵ and ³⁵⁵GSL³⁵⁷, which exhibit polymorphism.

Fig. 5.

Structure of the nontwisting tau (297–391) fibril core. (A) β-strand locations in the nontwisting tau (297–391) ribbons. (B) Ensemble of ten lowest-energy structures of nontwisting tau (297–391) fibrils. (C) Lowest-energy structure model of nontwisting tau (297–391) fibrils. All residues of the top protofilament are annotated, and the dimer interface residues (G323–I354) of the bottom protofilament are labeled in blue. ³³²PGGG³³⁵ and ³⁵⁵GSL³⁵⁷ (gray) exhibit chemical shift doubling, thus their conformations are more disordered in the structural model. Asp and Glu residues that show large sidechain carbonyl chemical shifts indicative of deprotonated carboxyl groups are indicated by a negative charge symbol.

Between the two monomers, the dimer interface is stabilized by numerous hydrophobic interactions, for example, between L325 and V350/I354 (SI Appendix, Fig. S8) and by polar interactions among N327, H329, and D348. The dimer interface is centered at a pair of interdigitating E338 residues, suggesting that this residue may be protonated and neutral in the non-twisting fibril. This is supported by the absence of a downfield carboxyl ¹³C chemical shift for E338 (SI Appendix, Fig. S1), which is expected for deprotonated carboxyl groups. In comparison, large carboxyl ¹³C chemical shifts are observed for D314, E342, and D345. In the structural model, D314 and E342 sidechains reside in the spacious regions that flank the disordered ³³¹KPGGG³³⁵ loop, and D345 is exposed to the fibril surface. Therefore, acidic residues that are located in spacious regions of the fibril have deprotonated carboxyl chemical shifts, whereas the tightly constrained E338 is protonated. H329 Cγ and Cδ2 chemical shifts indicate a protonated and cationic imidazole (SI Appendix, Table S1). These protonated E338 and H329 sidechains are consistent with the acidic pH of the fibril growth solution induced by MgCl₂ (Fig. 1).

Discussion

The above data indicate that it is difficult to reproduce twisted tau fibrils under the high-Mg²⁺ phosphate buffer condition. The pH equilibria for phosphate buffer and the solubility equilibria for magnesium phosphate dictate that the combination of phosphate and Mg²⁺ will precipitate most of the phosphate ions and cause acidification of the resulting solution. In the recent in vitro screening of conditions to generate PHF tau (8), 16 out of 47 conditions used 200 mM MgCl₂ in a 10 mM pH 7.4 phosphate buffer. Our data suggest that many of these conditions will be difficult to reproduce.

Instead, our ssNMR data indicate that in this high-Mg²⁺ phosphate solution, tau (297-391) forms nontwisting filaments with a dimeric molecular structure. The nontwisting ribbon morphology developed over the course of several days under moderate shaking and became the predominant species after an initial period in which nontwisting ribbons coexisted with a small amount of twisted filaments. In good agreement with our structural model, a recent cryo-EM study of the same tau (297–391) construct found spindle-like twisted filaments (9) whose molecular structure is very similar to the current structure of the nontwisting fibrils (Fig. 6 A and B and SI Appendix, Fig. S8). Both fibril cores span residues from the R3 to the C-terminal half of R4, and both exhibit a U-turn at C322 that stacks the N-terminal segment of R3 with the N-terminal segment of R4. Both fibrils have a dimer interface lined by residues G323–I354. The main distinction between the two models is the center of the dimer interface, which is E338 in the nontwisting filament and Q336 in the spindle-like filament. This difference suggests that a small shift in the sidechain packing registry may be sufficient to give rise to different fibril morphologies. The sidechain chemical shifts of E338 suggest that this residue is most likely protonated and neutral under our sample conditions, supporting the sequestration of two E338 residues in close proximity at the dimer interface.

Fig. 6.

Schematic models of several in vitro and ex vivo tau fibril cores. (A) Tau (297–391) nontwisting fibril. (B) Tau (297–391) spindle-like fibril reported recently by cryoEM (9). (C) Corticobasal degeneration tau (21). (D) Pick’s disease (PiD) 3R tau protofilament (22).

How well can we differentiate the nontwisting fibril structure from the spindle structure, given the similarity between the two models? The nontwisting filaments have extremely high conformational homogeneity, as manifested by the ¹³C linewidths of about 0.25 ppm in the 1,3-¹³C labeled sample. Nevertheless, the size of this protein (95 amino acid residues) is sufficiently large that resolving all the chemical shifts in the dense 2D CC spectra to assign long-range correlations and intermolecular correlations (SI Appendix, Figs. S4 and S5) de novo is challenging. Three lines of evidence suggest that the difference between the two structural models is likely real. First, we measured water-edited 2D ¹³C–¹³C correlation spectra, which show that ³²²CGS³²⁴ and ³⁵⁴IGSL³⁵⁷ at the two long ends of the dimer are highly water accessible, whereas the R4 hexapeptide ³³⁷VEVKSE³⁴² at the center of the dimer is dry (SI Appendix, Fig. S9 and Table S9). Thus, there is a gradient of decreasing water accessibility from the two ends of the dimer to the center. The high water accessibility of C322, I354, and L357 implies that these residues are exposed to the fibril surface. This finding is inconsistent with the spindle model, which shields ³²²CGS³²⁴ by a long overhang formed by residues ³⁵⁶SLDNIT³⁶¹. Second, in the nontwisting fibrils, residues after L357 are not observed in the dipolar NMR spectra, indicating that they are dynamically disordered. This contradicts the spindle model, which shows the ³⁵⁶SLDNIT³⁶¹ segment to be a well-ordered β-strand. Third, the nontwisting fibril model is partly constrained by multiple ambiguous intermolecular correlations in the center of the dimer, including four correlations between K340 and V337 and four correlations between K340 and Q336 (SI Appendix, Table S5 and Fig. S8). These contacts are inconsistent with the spindle model, which show distances of more than 10.5 Å for the corresponding contacts. Although these cross peaks are ambiguous in terms of chemical shift assignment, many alternative assignment can be reasonably ruled out during iterative structure calculation (23, 24). Together, these lines of evidence suggest that the two structural models are slightly different in the sidechain registry. Further verification of these differences may be achieved using site-specifically labeled samples that probe the center of the dimer interface.

The four-layered dimer structure of the nontwisting tau (297–391) filament has several similarities with the structures of 4R and 3R tau aggregates extracted from CBD, argyrophilic grain disease (21) and PiD (22). The ³²²CGS³²⁴ turn also manifests as a turning point in the four-layered CBD tau structure and the two-layered PiD tau structure (Fig. 6 C and D). This R3–R4 β-arch is stabilized by steric zippers between the R3 hexapeptide motif and the N-terminal region of R4 (25–27) in both the nontwisting tau (297–391) and CBD tau. Similar sidechain stacking was also found in heparin-induced 3R tau fibrils (12). The recurring presence of these R3–R4 hexapeptide interactions in vitro and ex vivo suggests that this structural motif is energetically stable. In comparison, the R3–R4 β-arch is absent from the AD PHF tau structure (SI Appendix, Fig. S7C) (8), which instead stacks the R3 hexapeptide against the ³⁷⁴HKLTF³⁷⁸ segment of the R’ domain.

This protease-resistant domain of tau has now been found to adopt three fibril morphologies with associated molecular structures: paired helical filaments with an AD-mimetic C-shaped structure (8); spindle-like twisted filaments with a Q336-centered dimer structure; and nontwisting filaments with the E338-centered dimer structure. We attribute the presence of these multiple fibril morphologies and molecular structures to variations in pH and ionic conditions of the fibril growth solution. The E338-centered dimer structure for the nontwisting ribbon is stabilized by the acidic pH of the fibril growth solution, which protonates E338 and removes electrostatic repulsion at the center of the dimer. This acidic pH results from the precipitation of the 10 mM phosphate buffer ions as MgHPO₄ upon the addition of 200 mM MgCl₂ (SI Appendix, SI Methods), leaving Cl^- as the main anion in the solution. The fibril growth reaction is thus unbuffered other than by the protein itself (28). In addition to pH-induced changes in the charge state of Glu and Asp, depletion of the hydrogen phosphates and dihydrogen phosphates from solution may affect charge–charge interactions of Arg and Lys residues, which have a strong propensity to form salt bridges with phosphates (29, 30). The involvement of phosphate ions in tau (297–391) self-assembly is hinted by the fact that the protein does not form fibrils in Tris buffer. Finally, we speculate that the precipitated MgHPO₄ salts might act as heterogeneous nucleation centers for fibril formation, not unlike the acceleration of fibril growth by glass beads and Teflon beads for tau and α-synuclein (31, 32). Overall, depending on the precise pH, the amount of the MgHPO₄ precipitate, and the ionic conditions of the solution, different nuclei and hence different fibril structures may become dominant in a particular reaction.

The dependence of protein aggregation on pH and salt has been reported for other amyloid proteins. Aβ(11–25) changed the registry of its antiparallel fibrils between pH 7.4 and 2.4 (33), and Aβ(1–40) exhibited pH-dependent rates of primary nucleation (34). α-synuclein fibrilization is slow at neutral pH but fast at acidic pH (31, 35). This pH-dependent rate was attributed to the presence of many Asp and Glu residues in the C-terminal region of the protein. Unprotonated Glu and Asp residues at neutral pH cause electrostatic repulsion, thus slowing down fibril growth (28). The pK_a values of these Asp and Glu residues were deduced from the measured pH of the fibril growth solution and found to be elevated compared to the solution pK_a values, favoring aggregation. The morphology of the α-synuclein fibrils obtained from seeded growth also depends on the pH: the fibril structure can be primarily dictated by the solution pH rather than by the structure of the seed (36). These results are in good agreement with our interpretation that the tau (297–391) fibril structures are sensitive to the solution pH. In addition to pH, ionic conditions also affect the morphology and structure of amyloid proteins. High salt concentrations (35) and multivalent cations (37) accelerated α-synuclein aggregation. Therefore, pH, ionic strength, and ionic types, through charge–charge interactions between the ions and protein sidechains, can all impact the kinetics and structure of amyloid fibrils. Tau (297–391) contains 6 His, 5 Asp, 5 Glu, 14 Lys, and 2 Arg residues, which are expected to be sensitive to pH changes and salt conditions of the fibril growth solution.

The stabilization of the E338-centered dimer structure under our experimental conditions suggests that the spindle fibril with a Q336-centered dimer structure (9) may be a metastable state on pathway to the nontwisting ribbon. This hypothesis results from three observations: 1) the conversion of a small population of twisted filaments into nontwisting ribbons after freeze–thawing; 2) the previously reported coexistence of the spindle-like filaments and straight filaments (9); and 3) a Q336-centered dimer structure may contain protonated E338 and H329 to achieve charge balance. Both the spindle fibrils and nontwisting ribbons may be specific to the acidic condition of the fibril growth condition.

Interconversion of twisted and nontwisting filaments has been reported for other amyloid proteins. Atomic force microscopy data of preformed β-lactoglobulin fibrils in solutions of varying ionic strengths (38) showed that initially twisted filaments with short crossover lengths unwound to adopt longer pitches as the ionic strength increased. This change was attributed to the changing balance between elastic torsional energy, which favors nontwisting morphologies, and electrostatic repulsion, which favors fibril twist around the long axis. Thus, changing ionic strength can lead to substantial changes in fibril ultrastructural morphology by modulating the electrostatic shielding of charged residues on the fibril surface. The change of twist for preformed β-lactoglobulin fibrils is slow, requiring several weeks to occur. For tau (297–391), a conversion from the spindle fibril to the nontwisting fibril after freeze–thawing may be facilitated by the rectangular shape of the fibril core and the absence of the fuzzy coat, both of which favor side-on nucleation (39).

In AD, ex vivo tau fibrils appear to be predominantly twisted, while nontwisting fibrils have not been documented. However, ribbon-like tau fibrils with no twist or only occasional twists have been reported in multiple system tauopathy with presenile dementia (40), in an unclassified 4R tauopathy associated with familial parkinsonism and progressive respiratory failure (41), and in an AD mouse model (42). Although the modern cryoEM technique is highly efficient for determining amyloid fibril structures, the technique requires fibril twists, thus nontwisting fibrils have largely escaped attention. Indeed, in electron micrographs that show PHF-like twisted fibrils formed by tau (297–391) under high MgCl₂ concentration, nontwisting ribbons were also present but were not analyzed further (8). Solid-state NMR can play a key role in characterizing these nontwisting filaments.

CryoEM data of tau aggregates so far show that tau misfolds into different structures in different diseases but adopts the same structure across patients afflicted by the same disease (43). This implies that the ex vivo tau fibril structures represent distinct prion-like folds that propagate in the brain. The specific tau aggregate structure observed in each disease may be the best able to propagate or nucleate in the cellular environments that are characteristic of a particular disease state or be the best able to evade cellular defense mechanisms (44). Understanding how the environment modulates tau fibril structures is therefore essential to the elucidation of the tau aggregation mechanism. The current data indicate that differences in pH and ionic conditions can change the molecular structures of the fibril cores dramatically or subtly. The full pH range of 5 to 7 is biologically relevant, with endosomes and lysosomes presenting acidic environments of pH 4.5–5.5, whereas the cytosol is approximately neutral. The subtle difference in the molecular structures of the nontwisting tau (297–391) and spindle-like tau (297–391) suggests that the conformational polymorphism of tau aggregates can result from finely tuned energetic and kinetic balances at the tertiary and quaternary structure levels. It should be possible to reveal the chemical code of tau aggregation by applying precise experimental control. Correlating the structures of tau with environmental conditions as well as posttranslational modifications should provide detailed insights into the chemical basis of tau aggregation.

Materials and Methods

Tau (297–391) was expressed in Escherichia coli and purified using cation exchange chromatography followed by HPLC. Fibrils were formed in 10 mM phosphate buffer containing 200 mM MgCl₂ under 150 rpm shaking at 37 °C. Three isotopically labeled fibril samples were produced: a uniformly ¹³C, ¹⁵N-labeled (UCN) fibril, a 50% diluted UCN fibril consisting of a 1:1 mixture of UCN-labeled protein and unlabeled protein, and a 1,3-¹³C-labeled fibril. Negative-stain TEM images were taken to assess fibril morphology. Solid-state NMR spectra were measured on 800 and 900 MHz spectrometers. ¹³C and ¹⁵N chemical shifts were assigned using 2D CC and NCACB correlation spectra and 3D NCACX, NCOCX, and CONCACB spectra. Long-range interresidue contacts were identified using 400 ms 2D CC spectra. Water-edited spectra were measured using the 2D DREAM CC experiment with a 4 ms ¹H mixing time. Tau (297–391) fibril structure was calculated using XPLOR-NIH.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

The NMR chemical shifts and 2D CC, NCACB, and 400 ms CC spectra shown in this work have been deposited to the Biological Magnetic Resonance Data Bank under accession code 31059. The atomic structural model of the nontwisting tau (297–391) has been deposited in the Protein Data Bank under accession number 8G58.