Structures of ΔD421 Truncated Tau Fibrils

Abstract

The microtubule-associated protein tau aggregates into pathological β-sheet amyloid fibrils in Alzheimer’s disease (AD) and other neurodegenerative diseases. In these aggregates, tau is chemically modified, including abnormal hyperphosphorylation and truncation. Truncation after D421 in the C-terminal domain occurs at early stages of AD. Here we investigate the structures of ΔD421-truncated 0N4R tau fibrils assembled in vitro in the absence of anionic cofactors. Using solid-state NMR spectroscopy and cryoelectron microscopy, we show that ΔD421-truncated 0N4R tau forms homogeneous fibrils whose rigid core adopts a three-layered β-sheet structure that spans R2, R3 and R4 repeats. This structure is essentially identical to that of full-length tau containing phospho-mimetic mutations at the PHF1 epitope in the C-terminal domain. In comparison, a ΔD421-truncated tau that additionally contains three phospho-mimetic mutations at the AT8 epitope in the proline-rich region forms a fibril core that includes the first half of the C-terminal domain, which is excluded from all known pathological tau fibril cores. These results indicate that the posttranslational modification code of tau contains redundancy: both charge modification and truncation of the C-terminal domain promote a three-layered β-sheet structure, which resembles pathological four-repeat tau structures in several tauopathies. In comparison, reducing the positive charges at the AT8 epitope in ΔD421-truncated tau promotes a fibril core that includes an immobilized C-terminal domain. The absence of this structure in tauopathy brains implies that ΔD421 truncation does not occur in conjunction with AT8 phosphorylation in diseased brains.

Introduction

In healthy brain, the intrinsically disordered protein tau functions to assemble and stabilize microtubules [1, 2], which regulate the shape and polarity of cells and serve as the track for cellular transport [3]. When tau undergoes posttranslational modifications (PTMs) such as phosphorylation, acetylation and truncation, its microtubule association is impaired [4]. In Alzheimer’s disease (AD), the failure of tau to associate with microtubules disrupts axonal integrity, mis-sorts tau to the synapses [5] and correlates with tau aggregation into paired helical filaments (PHFs) and straight filaments, which make up the neurofibrillary tangles in AD [6, 7]. Understanding the molecular mechanisms by which PTMs alter the microtubule interaction of tau and induce self-assembly of the protein into pathological filaments is crucial for developing better diagnostics and therapeutics for AD.

Among tau PTMs, abnormal hyperphosphorylation is the best studied [8, 9]. About 45 phosphorylation sites have been identified in AD tau, many at proline (Pro)-preceding Ser and Thr residues. Some of the most widely used tau antibodies for postmortem diagnosis of AD target pSer-Pro and pThr-Pro motifs. For example, the antibody AT8 recognizes pS202, pT205, and pS208 in the P2 domain [10, 11], whereas the antibody PHF-1 recognizes pS396, pS400, pT403, and pS404 at the junction of R’ and the C-terminal domain [12, 13]. In addition to phosphorylation, truncation also occurs in pathological tau. Tau can be truncated by several caspases, including caspases 3, 7, and 8. Caspase-cleaved tau is known for its high neurotoxicity and is associated with AD and other tauopathies. Among the many cleaved tau species found in AD, truncation after D421 has received the most attention [14, 15]. Treatment of primary neurons by Aβ42 fibrils caused caspase cleavage of tau at D421, and the truncation product is present in AD neurofibrillary tangles, recognized by the antibody Tau-C3 [14]. Comparison of mRNA levels of multiple caspases in AD versus healthy brains found caspase 8 to be the most important for ΔD421 truncation [16]. In vitro, ΔD421 tau fibrillizes more rapidly and to a higher level compared to full-length tau [14]. Similarly, E391 truncated tau is also found in AD PHF tau and is recognized by the antibody mAb 423 [17–19]. When tau constructs that start from R3 and end at varying positions of the C-terminal domain were tested for their abilities to form fibrils, it was found that the intact C-terminal domain inhibited fibril formation, ΔD421 truncation promoted fibrilization, whereas E391 truncation allowed the formation of AD-fold fibrils under certain ionic and pH conditions [20].

Although both phosphorylation and truncation occur in AD brain, the chronology of these events during disease progression is not well understood. In transgenic mice models, immunoblotting data showed that hyperphosphorylation started at 3 months, which preceded ΔD421 truncation at 6 months [21]. In Braak stage III-V AD brains, immunohistochemical and confocal microscopy data found higher levels of phosphorylated tau compared to ΔD421 tau, suggesting that phosphorylation precedes truncation. On the other hand, ΔD421 truncation induces a conformational epitope that is recognized by the early tau marker MC1 [15], which labels tau in Braak stage I and II AD brains [22]. Moreover, ΔD421 tau is phosphorylated by the glycogen synthase kinase 3β and subsequently recognized by PHF-1 [15], suggesting that truncation precedes or concurs with hyperphosphorylation. A proteomics study of tau PTMs in AD and control subjects [23] reported the frequency of tau modifications at different stages of AD, concluding that the earliest events of tau aggregation include not only phosphorylation in the proline-rich region and at the PHF-1 epitope but also C-terminal cleavage.

An important approach for investigating the impact of PTMs on tau aggregation is to determine the structure of in vitro assembled tau fibrils that contain mutations mimicking the posttranslational modifications in diseased brain. By comparing the structures of these in vitro assembled tau fibrils with the ex vivo structures, we can gain insights into the PTM code of tau. Unmodified wild-type tau is highly soluble and net positively charged at neutral pH; thus aggregation of wild-type full-length tau has traditionally required polyanionic cofactors such as heparin and RNA [24]. PTMs that reduce the positive charges of tau are expected to increase the aggregation potential of tau and may obviate the need for anionic cofactors. Indeed, we recently showed that two phospho-mimetic full-length tau constructs self-assembled into ordered amyloid fibrils without anionic cofactors [25]. We introduced three Glu mutations at the AT8 epitope (S202E, T205E, and S208E) in one construct and four Glu mutations at the PHF-1 epitope (S396E, S400E, T403E, and S404E) in the other construct. Using solid-state NMR and cryoelectron microscopy (cryoEM), we found that AT8-3E tau fibrillized into a multi-layered rigid core that extends from the R3 repeat to the C-terminus of the protein. In comparison, PHF1-4E tau formed a three-layered β-sheet core comprising R2, R3 and R4 repeats. This fold qualitatively resembles the structure of progressive supranuclear palsy (PSP) tau [26]. When both AT8 and PHF-1 epitopes were mutated to Glu, the resulting 7E mutant adopted the same structure as PHF1-4E tau.

Here we investigate the impact of D421 truncation on tau aggregation by examining two ΔD421 fibrils designed from full-length 0N4R tau. One construct, ΔD421-WT tau, has no further modifications other than truncating the last 20 residues of the protein. The second construct, ΔD421-3E tau, additionally contains the three AT8 phospho-mimetic mutations. We show that both proteins self-assemble into ordered filaments in the absence of anionic cofactors. Using solid-state NMR and cryoEM, we determined the atomic structure of the ΔD421-WT tau fibril and characterized the conformation of the ΔD421-3E fibril. Remarkably, ΔD421-WT tau adopts the same rigid core structure as PHF1-4E tau, revealing another layer of redundancy in the PTM code of tau amyloid formation.

Materials and Methods

Cloning, Expression and Purification of Phospho-mimetic Tau

The genes for ΔD421-WT 0N4R tau and ΔD421-3E 0N4R tau were produced by modifying a 0N4R tau gene (GenScript) in a pET-28a vector using Gibson Assembly site-directed mutagenesis. ΔD421-WT tau spans residues 1-421 of 0N4R tau (Uniprot P10636-6) while ΔD421-3E tau contains S202E, T205E, S208E mutations and is truncated after D421. Amino acid numbering follows the 2N4R tau sequence (Uniprot P10636-8), with K44 followed by A103. Correctness of mutagenesis was verified by Sanger sequencing before the modified plasmids were transfected into E. coli BL21(DE3) competent cells.

Colonies from freshly transformed LB Agar plates were used to inoculate a 10 mL LB starter culture. After overnight growth at 37°C with 250 rpm shaking, the starter culture was used to inoculate 1 L LB media, which was allowed to grow at 37°C with 250 rpm shaking to an OD₆₀₀ of 0.8. At this point the cells were spun down at 1,000 g for 20 minutes, and the cell pellet was resuspended in 1 L M9 minimal media containing 2 g/L ¹³C₆-D-glucose and 1 g/L ¹⁵NH₄Cl as the only carbon and nitrogen sources, respectively. After 60 minutes of shaking at 250 rpm and 37° C, the cells reached an OD₆₀₀ of 1.0, at which point protein expression was induced with 0.8 mM IPTG and an additional 1g/L ¹³C₆-D-glucose was added to the media. After 3-5 hours of protein expression at 37°C, cells were harvested by centrifugation and pellets were frozen. The Agar plates and all media contained 50 μg/mL Kanamycin.

The expressed tau proteins were purified using similar procedures as described before [27]. Briefly, cells were thawed and homogenized by vortexing in ice-cold lysis buffer, which contained 20 mM Na₂HPO₄ (pH 6.8), 50 mM NaCl, 10 mM DTT, and one cOmplete^™ protease inhibitor cocktail tablet (Roche) per 40 mL of lysis buffer. The lysate was sonicated on ice using a probe sonicator (5 seconds on, 5 seconds off for a total of 10 minutes). Following sonication, the lysate was incubated in a boiling water bath for 20 minutes, then centrifuged at 15,000 g for 60 minutes to remove cell debris and aggregated proteins. The supernatant was purified by using a cation-exchange column (self-packed with SP Sepharose Fast Flow resin from GE Healthcare) with a gradient elution between buffer A and buffer B: A = 20 mM Na₂HPO₄ (pH 6.8), 2 mM DTT; B = 20 mM Na₂HPO₄ (pH 6.8), 2 mM DTT, 1 M NaCl. Eluted fractions containing the tau protein were further purified by reverse-phase HPLC using an Agilent Zorbax 300SB-C3 column (21.2 x 250 mm, with 7 μm particle size) using an acetonitrile gradient of 5-50% over 50 minutes. Collected HPLC fractions were pooled and lyophilized. This process typically yielded around 30-40 mg of purified tau protein per liter of M9 medium for each construct.

Cofactor-Free Fibrillization

Lyophilized tau was dissolved at 1.2 mg/mL in 50 mM K₂HPO₄ : KH₂PO₄ buffer, pH 6.8, containing 300 mM NaCl, 10 mM DTT, and 1x cOmplete^™ protease inhibitor cocktail tablet (Roche) per 40 ml fibrillization buffer. Freshly dissolved tau protein was vortexed for 45 seconds and sonicated for 10 minutes to fully dissolve the protein. The resulting solution (15-30 mL) was transferred into a 250 mL glass bottle that had been purged with nitrogen gas to eliminate any oxygen. The solution was shaken at 200 rpm with a 25 mm orbital diameter at 37°C for 10 days. To maintain a reducing environment for the two cysteine residues (C291 and C322), 5 mM DTT was added every two days, with nitrogen gas purging each time the bottle was opened. After the first 5 days, a 1x cOmplete^™ protease inhibitor cocktail was added to prevent proteolysis. After 10 days, aliquots of the fibrillization reaction were frozen for TEM and cryoEM experiments. The remaining solution was centrifuged at 100,000 g for 2 hours at 4°C using a TLA-55 rotor. The pellet was packed into 3.2 mm MAS NMR rotors by centrifugation, supplemented with manual packing using a thin needle to transfer the remaining sticky material in the pipette tip into the rotor. A small amount of water was added to the rotor as needed to ensure homogeneous hydration of the fibrils.

Negative-Stain Transmission Electron Microscopy

Unconcentrated tau fibril solutions were diluted 2-fold and adsorbed for 1 min onto freshly glow-discharged 200-mesh formvar/carbon-coated copper grids (Ted Pella). The grids were washed twice with 100 mM sodium acetate to remove trace phosphate buffer and then stained with 0.7% (w/vol) uranyl formate for 15 s. TEM images were taken on an FEI Tecnai T12 electron microscope.

Solid-State NMR Spectroscopy

Solid-state NMR experiments were conducted on an 800 MHz (18.8 T) spectrometer with a Bruker NEO and AVANCE III console and a BlackFox 3.2 mm ¹H/¹³C/¹⁵N magic-angle-spinning (MAS) probe. ¹H chemical shift was internally calibrated using DSS. ¹³C chemical shifts were first calibrated externally to the adamantane CH₂ chemical shift at 38.48 ppm on the tetramethylsilane (TMS) scale, then subsequently calibrated via ¹H by setting the ¹³C sr value in the TopSpin software to sr(¹H)/4-153 to give the ¹³C chemical shift on the TMS scale. ¹⁵N chemical shifts were first calibrated externally to the ¹⁵N signal of ¹⁵N-acetylvaline at 122.0 ppm on the liquid ammonia scale, then subsequently calibrated via ¹H by setting the ¹⁵N sr to be equal to sr(¹H)/10 – 40. Reported sample temperatures were estimated from the measured water ¹H chemical shift on the DSS scale using the equation T_sample (K) = 96.9 x (7.83 – δ_H2O) [28]. Samples were spun at 14 kHz MAS.

All solid-state NMR experiments were conducted on a single sample for each protein. Immobilized residues in the fibrils were measured using dipolar-coupling based polarization transfer experiments. Two-dimensional (2D) ¹³C-¹³C (CC) correlation spectra were measured using ¹H-¹³C cross polarization (CP) followed by CORD mixing [29]. 2D ¹⁵N-¹³Cα (NCA) correlation spectra and three-dimensional (3D) NCACX, NCOCX and CONCA correlation spectra were measured to assign the resonances of rigid residues (Table S4). The NCACX and NCOCX spectra used 4-5 ms ¹⁵N-¹³C SPECIFIC CP [30] and 82 ms ¹³C-¹³C CORD mixing. The CONCA experiment contained CO-N and N-CA SPECIFIC-CP transfer steps. For many dipolar 2D and 3D experiments, a short ¹H-¹³C CP contact time of 150 μs was used to selectively detect the most rigid β-sheet signals while suppressing the signals of mobile residues. Experiments that require extensive signal averaging to obtain sufficient signal-to-noise ratios were run in blocks of 1-3 days with field drift correction between blocks. The multiple 3D spectra were added in the time domain before Fourier transformation.

2D and 3D correlation spectra were processed in TopSpin 4.3 using either QSINE apodization with SSB = 3 or GM apodization with LB = −20 Hz and GB = 0.05. Spectra were plotted with 1.2 x multiplication factor between successive contour levels. Chemical shift assignment was conducted in CCPNMR [31].

CryoEM Data Acquisition

3 μl aliquots of fibrillization reactions were thawed and applied to glow-discharged R1.2/1.3 300-mesh carbon gold grids (SPI Supplies). The grids were plunge-frozen in liquid ethane using a Thermo Fisher Scientific Vitrobot Mark IV system. CryoEM data were acquired at the MIT.nano facility. All images were recorded at a dose of 30 electrons per Ų using the EPU software (Thermo Fisher Scientific) and were converted to tiff format using relion_convert_to_tiff prior to processing. Images were recorded on a Krios G3i with Bioquantum K3 camera (Gatan) using an energy slit of 20 eV on a Gatan energy filter.

Helical Reconstruction of ΔD421 tau fibrils

Helical reconstructions of the two ΔD421 tau fibrils were performed using the RELION-4.0 software [32, 33]. The frames of raw EM movies were adjusted for gain, aligned, weighted, and combined using RELION’s motion correction program [34]. CTFFIND-4.1 [35] was used to estimate the parameters of the Contrast Transfer Function (CTF). Filaments were manually selected in RELION 4.0. The picked particles were initially extracted in boxes of 768 pixels downscaled to 512 pixels. For ΔD421-WT tau, reference-free 2D classification was conducted on these particles to assess distinct views, estimate crossover lengths, and select segments for further processing. The chosen 2D classes were re-extracted in full-resolution boxes of 512 pixels. Other rounds of reference-free 2D classification were performed using T values of 20-50 to pick the twisting fibrils. About ten classes that show clear 4.75 Å cross-β monomer separation were chosen. A featureless cylinder was used for the first round of 3D classification using K = 1 and T = 20. Then several rounds of 3D classification were used to select particles that gave the best reconstruction, with parameters K = 1-3 and T = 20-30. 3D auto-refinement was applied to enhance image resolution and optimize the helical twist and rise. Bayesian polishing [36] and CTF refinement [37] were applied to all rendered maps to further improve resolution. The final maps were sharpened using standard post-processing techniques in RELION. For ΔD421-3E tau, we conducted the same analysis but were unable to obtain a high-resolution map due to the limited twist and width modulation of the fibril.

Structure Calculation and Validation

Atomic structural models were constructed in COOT [38] with three rungs for each structure. ISOLDE was used for coordinate refinement [39]. Model refinement was conducted by raising the temperature to 300 K for 1 minute in the ISOLDE menu inside Chimera-X [20]. Additional information about data processing, model refinement, and validation is listed in Table S3. The ΔD421-WT tau structure has been deposited into the PDB with the accession code 2XYZ.

Results

ΔD421-WT and ΔD421-3E tau assemble into conformationally distinct amyloid fibrils

This study focuses on two 0N4R tau constructs that are truncated after D421 (Fig. 1a). ΔD421-WT tau has no other modifications whereas ΔD421-3E tau contains Glu mutations at S202E, T205E and S208E. We assembled the fibrils from 1.2 mg/mL monomer solutions in phosphate buffer (50 mM, pH 6.8) containing 300 mM NaCl, 10 mM DTT, and protease inhibitors. The reaction mixtures were shaken at 200 rpm at 37°C for ten days. This fibrilization condition is the same as that used before to fibrillize full-length 0N4R tau containing phospho-mimetic AT8-3E and PHF1-4E mutations [25]. Fibril formation began after 3-4 days, visible by eye, which is moderately faster than AT8-3E and PHF1-4E tau, which had a lag phase of 5-6 days. Transmission electron micrographs of the fibrils after ten days show relatively isolated and homogeneous filaments with lengths of 0.5-1.0 μm and widths of 9-12 nm (Fig. 1b, c).

Figure 1. Amino acid sequences, fibril morphologies, and 2D solid-state NMR spectra of ΔD421 tau fibrils.

. (a) Domains and amino acid sequence of 0N4R tau. Residues from S422 to L441 are truncated in the present study. Three AT8 epitope residues are mutated to Glu (S202E, T205E, S208E) to create the ΔD421-3E sample. The unmodified protein is termed ΔD421-WT tau. (b, c) Negative stain TEM images of ΔD421-WT tau (b) and ΔD421-3E tau (c) fibrils. Both fibrils were formed without anionic cofactors. (d) 2D NCA spectrum of ΔD421-WT tau fibrils. (e) 2D NCA spectrum of ΔD421-3E tau fibrils. Resonance assignments are obtained from 3D correlation spectra. The ambiguous assignment SKCGS in (e) can be either ²⁸⁹SKCGS²⁹³ in R2 or ³²⁰SKCGS³²⁴ in R3. Capital letter denotes the tentatively assigned residue within the segment.

To characterize the structure of the rigid core of these ΔD421 tau fibrils, we measured 2D NCA and CC spectra. ΔD421-WT tau fibrils show well-resolved 2D fingerprint spectra (Fig. 1d, Fig. 2a), indicating a well-ordered rigid core. In comparison, the ΔD421-3E spectra have lower sensitivity (Fig. 1e, Fig. 2b). The 2D NCA spectra resolve about ten Gly peaks for ΔD421-WT and six for ΔD421-3E, while the 2D CC spectra resolve about twenty Ser and Thr signals for each protein (Fig. 2). To assign the ¹³C and ¹⁵N chemical shifts, we measured 3D NCACX, NCOCX, and CONCA correlation spectra (Fig. S1). For ΔD421-WT tau, we were able to assign 52 residues from the beginning of R2 to S356 in the middle of R4 (Table S1). Most of the high-intensity peaks in the spectra were assigned, and a single set of chemical shifts were found, indicating that the protein adopts a single predominant conformation. Interestingly, the Cα and Cβ secondary chemical shifts of ΔD421-WT are similar to those of PHF1-4E tau (Fig. 3a, b), with β-strand breakers occurring at similar positions of the amino acid sequence. The ¹³C and ¹⁵N chemical shift differences are small except for the two CGS motifs in R2 and R3 (Fig. 3c), whose chemical shifts are interchanged between the two proteins (Table S1). Otherwise, many peaks in the 2D NCA spectra of the two proteins superimpose well (Fig. 4). This chemical shift similarity suggests that ΔD421-WT tau has a similar rigid-core structure as PHF1-4E tau.

Figure 2. Aliphatic region of the 2D ¹³C-¹³C correlation spectra of ΔD421 tau fibrils.

(a) ΔD421-WT tau. (b) ΔD421-3E tau. Resonance assignment is obtained from 3D correlation spectra.

Figure 3. Cα and Cβ secondary chemical shifts of ΔD421 tau fibrils.

(a) ΔD421-WT tau chemical shifts. (b) Previously published PHF1-4E tau chemical shifts [25]. (c) ¹³C (top) and ¹⁵N (bottom) chemical shift differences between ΔD421-WT tau and PHF1-4E tau. Residues that do not have assignment in both proteins are shaded in grey. The two CGS motifs, ²⁹¹CGS²⁹³ and ³²²CGS³²⁴, have the only significant chemical shift differences. (d) ΔD421-3E tau chemical shifts. β-sheet conformation is detected for a small number of R’ and C-terminal residues. No R2 and R3 residues can be unambiguously assigned due to low spectral sensitivity.

Figure 4. ΔD421-WT and PHF1-4E tau fibrils have similar NMR fingerprints.

(a) 2D NCA spectrum of ΔD421-WT tau. (b) 2D NCA spectrum of PHF1-4E tau. (c) Overlay of the two spectra, showing a high degree of chemical shift similarity.

Compared to ΔD421-WT tau, the ΔD421-3E spectra have lower sensitivity, making it difficult to conduct resonance assignment. We unambiguously assigned 23 residues from N359 in R4 to I417 in the C-terminal domain (Table S2, Fig. 3d). In addition, we identified the chemical shifts of an SKCGS motif, which may correspond to either ²⁸⁹SKCGS²⁹³ in R2 or ³²⁰SKCGS³²⁴ in R3. No other residues could be unambiguously assigned. Both ΔD421 tau fibrils show distinct chemical shifts from the AD-fold tau chemical shifts as obtained from the phospho-mimetic 4E-tau (297-407) construct [40]. Thus, truncation at D421, with or without the AT8 mutations, does not induce the C-shaped AD fold.

Fibril core structures of wild-type ΔD421 tau and water accessibility

To determine the three-dimensional fold of the rigid core of the two ΔD421 tau fibrils, we measured cryoEM data and conducted helical reconstruction. We manually picked fibrils in 5,000 micrographs for each sample and used a featureless cylinder as the initial reference (Fig. S2, Fig. S3). For ΔD421-WT tau, 3D classification and refinement resulted in a 2.9 Å map (Fig. 5), showing a triple-stranded fold that is nearly superimposable with the PHF1-4E tau fibril structure (Fig. 5e). The heavy-atom RMSD between the two structural models is 1.4 Å while the backbone RMSD is 1.35 Å for residues G273 to H362. For ΔD421-3E tau, helical reconstruction failed because the fibril lacks twist, which made it impossible to generate an initial 3D model (Fig. S3).

Figure 5. CryoEM structure of ΔD421-WT tau fibrils.

(a) Representative micrograph. (b) Helical reconstruction, giving a crossover length of 92.2 nm. (c) Projection of the central ten slices of the 3D reconstructed map of ΔD421-WT tau. (d) Atomic model of ΔD421-WT tau fibril core overlaid with the electron density map. (e) Comparison of the ΔD421-WT tau and PHF1-4E 0N4R tau structural models. While both show a triple-stranded structure, ΔD421-WT tau is slightly expanded in the lateral direction compared to PHF1-4E tau.

To obtain information about the water accessibilities of the three-layered core of ΔD421-WT tau, we measured water-edited NMR spectra (Fig. S4). The lowest water accessibility is found for the R3 hexapeptide ³⁰⁶VQIVY³¹⁰ in the central strand, whereas the ²⁹³SKDN²⁹⁶ moiety on the exterior R2 strand and the ³⁵⁴IGSL³⁵⁷ segment on the exterior R4 strand are the most accessible to water (Fig. 5a). These trends are similar the water accessibilities of PHF1-4E tau (Fig. 5b). The inverted Ω motif ³²⁰SKCGSLG³²⁶ is less water accessible in ΔD421-WT tau than in PHF1-4E tau. On the exterior R2 surface of the ΔD421-WT rigid core, the ³⁰¹PGGG³⁰⁴ motif and the ²⁷⁵VQII²⁷⁹ segment are poorly accessible to water, indicating that these residues are shielded by the fuzzy coat. Our data do not indicate which segments of ΔD421-WT tau shield the rigid core; but the fact that truncation at ΔD421 does not cause large changes to the water accessibility compared to PHF1-4E tau imply that segments other than the last twenty residues of the full-length protein shield these rigid cores. In both proteins, R1 and R’ are disordered and are immediately N- and C-terminal to the triple-stranded core, thus it is likely that these two repeats, common to both proteins, are chiefly responsible for shielding the rigid core of both proteins from water. To obtain information about the fuzzy coat, we measured a 2D ¹H-¹⁵N INEPT spectrum of ΔD421-WT tau (Fig. 7). The spectral pattern is broadly consistent with that of PHF1-4E tau, but detailed chemical shift and intensity differences are observed. For example, several N-terminal residues show higher intensities in ΔD421-WT tau than PHF1-4E tau, indicating that the fuzzy coat dynamics and residual conformation of ΔD421-WT tau is not identical to that of PHF1-4E tau. Future studies will be necessary to obtain more detailed information about the fuzzy coat of these differently modified tau proteins.

Figure 7. 2D ¹H-¹⁵N INEPT spectra of ΔD421-WT tau and PHF1-4E tau.

(a) Spectrum of ΔD421-WT tau fibrils (red contours). (b) Spectrum of PHF1-4E tau fibrils (blue contours). Each spectrum is overlaid with the chemical shifts (black circles) of the fuzzy coat of full-length 2N4R tau filaments measured by HR-MAS solution NMR [61]. These solution NMR chemical shifts are assigned in magenta for C-terminal residues in (b). In (a), residues from 422 to 441, which are truncated in ΔD421-WT tau, are annotated in orange. Three N-terminal residues (E3, T17 and G42, assigned in blue) show higher intensities in ΔD421-WT tau than PHF1-4E tau, indicating that the N-terminal domain is more mobile in ΔD421 tau. (c) Overlay of the ΔD421-WT tau and PHF1-4E tau spectra. Although the two spectra are broadly similar, many specific chemical shift and intensity changes are observed.

Discussion

Tau cleavage occurs in many domains of the protein and is observed in both healthy and diseased brains [41–43]. In healthy brains, the majority of detergent-soluble tau is full-length, while a small fraction is truncated in the C-terminal domain [44]. Truncated tau is mainly found inside cells but can also be released into the interstitial fluid, cerebrospinal fluid, and even the bloodstream [45–47]. Many proteolytic enzymes, including caspases, calpains, thrombin, cathepsins, and asparagine endopeptidase, are involved in generating pathological tau fragments [41]. Among these, caspases, whose levels are elevated in AD brain [16, 48], are unique in their ability to cleave tau at the conserved D421 [14, 15, 49]. This cleavage alters the tau conformation and accelerates filament assembly. The presence of D421-truncated tau strongly correlates with the clinical dementia index and Braak staging of AD [14, 50].

Here we sought to understand the impact of D421 truncation on the structure of tau aggregates. Our main result is that in the absence of other modifications, D421 truncation leads to the same structure as PHF1-4E tau and PHF1/AT8-7E tau (Fig. 8). All three proteins form a three-layered β-sheet core spanning R2, R3 and R4 repeats, suggesting that this is an energetically stable structure. Because D421 truncation decreases the number of negative charges in the C-terminal domain whereas PHF1-4E mutations increase it, this structural identity cannot be simply attributed to charge modification of the protein. Instead, we suggest a conformational origin to this recurring triple-stranded fold. Our recent analysis of PHF1-4E tau dynamics by NMR suggested that the charge-modified C-terminal domain stabilizes the triple-stranded core by increasing the electrostatic attraction with the cationic repeats [25]. In comparison, truncation of the last twenty residues of the C-terminal domain may decrease the conformational disorder of the remaining portion of the C-terminal domain, thus similarly stabilizing the cationic repeats and promoting the same triple-stranded structure.

Figure 8.

Amino acid sequences and three-dimensional folds of the rigid cores of modified 0N4R tau fibrils. (a) Amino acid sequence of unmodified wild-type 0N4R tau. The net charge at pH 7.0 for each domain is indicated. (b) ΔD421-WT tau adopts a three-layered β-sheet structure spanning R2, R3 and R4 repeats. This fold resembles the PSP tau structure. (c) PHF1-4E tau adopts the same three-layered β-sheet structure [25]. (d) AT8-3E and PHF1-4E phospho-mimetic tau adopts the same three-layered β-sheet structure. (e) AT8-3E tau incorporates the C-terminal domain, R’, R4 and R3 into the rigid core. This structure is distinct from any tauopathy tau structures known to date.

Although the ΔD421-WT tau structure is essentially identical to the PHF1-4E structure (Fig. 5e), 2D NCA fingerprint spectra of the two proteins show local chemical shift differences (Fig. 4). For example, the PHF1-4E tau spectrum has additional Gly peaks that are absent in the ΔD421-WT spectrum, and the two CGS motifs have interchanged ¹⁵N chemical shifts between the two proteins. These chemical shift differences may reflect the slight lateral expansion of the ΔD421-WT tau fibril core compared to PHF1-4E tau seen in the cryoEM map (Fig. 5e). The two structures align well at the center of the three-layered β-sheet, but the ΔD421-WT tau structure is wider, separating the R2 and R3 PGGG motifs further away from each other. The end-to-end distance between G303 Cα and P332 Cα is 86 Å in PHF1-4E tau and increases to 90 Å in ΔD421-WT tau.

The striking finding that three different modifications of tau – ΔD421 truncation, PHF1-4E mutation, and PHF1/AT8-7E mutation – gives rise to the same fibril core structure indicates that the three-layered β-sheet is an energetically stable fold of tau and the PTM code of tau contains significant redundancy (Fig. 8). The triple-stranded β-sheet fold is adopted when the C-terminal domain is modified by charge mutations or is partially truncated. When the C-terminal domain is not modified and when charge mutations are introduced in the Pro-rich region, then a qualitatively different structure results (Fig. 8e) [25, 51]. The AT8-3E fold includes an immobilized C-terminal domain in a multi-layered structure that is distinct from all ex vivo pathological tau structures known to date. In comparison, the three-layered β-sheet fold is similar to the PSP tau fold [26]. These results suggest that the intact and unmodified C-terminal domain is protective against pathological tau aggregation. This conclusion is supported by fluorescence resonance energy transfer data of soluble tau, which found that the negatively charged N- and C-terminal domains fold over the positively charged microtubule-binding repeats, forming a paperclip-like conformation that inhibits aggregation [52]. Phospho-mimetic mutations and C-terminal truncations remove this protective effect, increasing the aggregation potential of the protein [53].

When the ΔD421 truncation is combined with AT8 mutations in the Pro-rich region, the protein no longer adopts the triple-stranded fold and instead shifts its rigid core to include the first half of the C-terminal domain, as evidenced by the measured chemical shifts (Fig. 3d). Since no tau fibrils from diseased brains include the C-terminal domain in the rigid core [26], the conformation of ΔD421-3E tau implies that the ΔD421 truncation does not occur in conjunction with AT8 phosphorylation in diseased brains. Indeed, biochemical studies have shown that phosphorylation has modulatory effects on the D421 truncation level in cells [54]: production of truncated tau was reduced in cell lines expressing AT8 phospho-mimetic tau but the effect was reversed by the introduction of PHF1 phospho-mimetic mutations. Similarly, experiments that treated crude tau extracts with recombinant caspase-3 found that AT8 mutations suppressed the formation of truncated tau, while the combination of AT8 and PHF1 mutations enhanced it. Therefore, these biochemical data indicate that AT8 phosphorylation interferes with enzymatic ΔD421 truncation, consistent with the implication of our structural result. At the same time, AT8 phosphorylation does not occur in healthy adult brain [11], thus the C-terminal immobilized conformation of ΔD421-3E tau is also unlikely to exist in healthy adult brain. However, AT8 phosphorylation is known to occur in fetal tau in a developmentally regulated manner [10, 55], thus it is possible that the C-terminal immobilized conformation observed here might be transiently populated in fetal brain. Our data indicate that PTMs play an important role in shaping the tau aggregates, and AT8 phosphorylation appears to be more involved in indirect regulation of tau aggregation rather than direct participation in the nucleation process.

Tau monomers are highly heterogeneous in human brain due to the presence of six isoforms, each of which can be modified. The coexistence of many chemically different tau species implies the coexistence of many different aggregation pathways with presumably different intermediates. Thus, why only one mature fibril fold is found in each disease [26, 56–59], despite the heterogeneity of the protein sequence and the heterogeneity of the cellular environment, remains an unanswered question. Our finding that differently modified tau can adopt the same rigid-core structures in vitro indicate redundancy in tau’s PTM code, which may partially explain the limited number of ex vivo tau fibril structures. Future studies, potentially combining experiments with computational simulations of the energy landscape, charge distribution and folding pathways of tau [60], will be important for understanding the molecular mechanism of this redundancy in the tau PTM code.

Supplementary Material

Figures of additional NMR spectra, representative cryoEM micrographs and 2D and 3D classes of the two fibrils, and tables of assigned NMR chemical shifts, cryoEM refinement statistics, and NMR experimental parameters, are provided.

Figure 6. Water accessibility of ΔD421-WT tau fibrils obtained from water-edited NMR spectra.

(a) Residue-specific water accessibilities of ΔD421-WT tau fibrils. (b) Residue-specific water accessibilities of PHF1-4E 0N4R tau fibrils. The S/S₀ scale bar is chosen to reflect the full range of measured S/S₀ values within each sample while overlapping between the two samples, to facilitate visual comparison of the water accessibilities of the two fibrils.

Data Availability:

The NMR chemical shifts of ΔD421-WT tau and ΔD421-3E tau have been deposited into the Biological Magnetic Resonance Bank (BMRB) with the accession codes 52914 and 52915, respectively. The ΔD421-WT tau structure has been deposited into the Protein Databank with the accession code 9MR8. It cryoEM map has been deposited into the Electron Microscopy Databank with the accession code EMD-48555.