Impact of Lysine Acetylation Mutations on the Structure of Full-Length Tau Fibrils

Abstract

The tau protein aggregates into amyloid fibrils in Alzheimer’s disease and other neurodegenerative diseases. In these tauopathies, tau is decorated with post-translational modifications, including phosphorylation and acetylation, suggesting that these modifications may cause tau to aggregate into specific pathological structures. Here, we investigate how pseudo-acetylation of three lysine residues, K311Q, K321Q, and K369Q, affect the fibrilization and fibril structure of full-length four-repeat tau. These acetyl-mimics are in addition to four phospho-mimetic glutamate mutations at the PHF1 epitope (4E tau). The joint mutant, 4E3Q tau, formed well-ordered amyloid fibrils without anionic cofactors. The 4E3Q tau fibrils lack twists, preventing structure determination by cryoelectron microscopy and necessitating characterization by solid-state NMR. ¹³C and ¹⁵N chemical shifts indicate that pseudo-acetylation caused the protein to adopt a distinct fold from the parent 4E tau fibrils: the rigid core contains β-strands between R2 and R4 repeats and near the end of the C-terminal domain. Importantly, the C-terminal half of the R3 repeat containing the K321Q mutation is disordered, in qualitative contrast to 4E tau. Chemical shifts indicate that these structural changes likely result from the disruption of salt bridges between lysine and aspartate residues. 4E3Q tau contains an immobilized R2, which differs from AD tau. These results provide insights into the impact of acetylation on tau fibrilization and fibril structure and suggest that acetylation of these three lysine residues in AD may occur after the formation of the paired-helical filament structure.

Graphical Abstract

Introduction

Alzheimer’s disease (AD) and many other neurodegenerative diseases are characterized by pathological aggregates of the protein tau in the brain ^1–2. These tau aggregates apparently act like prions: they spread in the brain through connected pathways, recruiting healthy tau to adopt and thus amplify their pathological structure ^3–4. This spread of tau aggregates in the brain is correlated with cognitive decline and is the current basis for neuropathological staging of AD ^5–6. Recently, cryoelectron microscopy (cryoEM) studies of ex vivo tau revealed that brain tau aggregates adopt specific structures that are conserved between patients with the same disease but distinct between patients with different diseases ^7–9. This finding led to the hypothesis that the specific tau folds may define each disease at the molecular level. Understanding how and why tau forms these specific propagating structures, and how to stop their formation and spread, is therefore a central aim of the field.

Healthy tau is intrinsically disordered and binds to and stabilizes neuronal microtubules with its cationic pseudo-repeats R1, R2, R3, R4 and R’^10–11. Approximately half of tau in adult human brain lacks R2, and is known as 3R tau, while the other half contains R2 and is known as 4R tau ¹. The center of these repeats is found to comprise the rigid cores of ex vivo tau amyloid fibrils, whereas the flanking N- and C-terminal domains form a disordered “fuzzy coat” that surrounds the core ^12–13. Different disease folds are composed of distinct isoforms of tau: AD tau fibrils are composed of a mixture of 3R and 4R tau ⁷, Pick’s disease tau fibrils contain 3R tau alone ¹⁴, and progressive supranuclear palsy (PSP) ⁹ and corticobasal degeneration (CBD) ¹⁵ fibrils are formed from 4R tau alone.

Tau aggregation is electrostatically counter-intuitive, as it requires colocalization of many identical copies of the cationic repeats. In vitro, purified full-length tau is unable to aggregate into homogeneous fibrils without polyanionic cofactors added in solution or immobilized on surfaces during fibril growth ^16–18. One study induced tau aggregation by polytetrafluoroethylene beads, but the resulting fibrils were inhomogeneous, precluding high-resolution structural characterization ¹⁹. In comparison, ex vivo pathological tau is extensively post-translationally modified, reducing the cationic charge. The most common post-translational modifications (PTMs) are serine (Ser) and threonine (Thr) phosphorylation and lysine (Lys) acetylation ^20–23. Phosphorylation of Ser and Thr introduces two negative charges, where addition of an acetyl group to the NH₃ moiety of the Lys sidechain neutralizes the Lys positive charge. Both these modifications reduce the positive charge of tau and have been found to impair tau interactions with microtubules ²⁴, promote tau fibrillization in vitro, and correlate with synaptic disfunction and neurodegeneration in animal models ^25–33. Interestingly, tau phosphorylation and acetylation differ in their spatial and temporal patterns ^34–35. Phosphorylation mainly occurs in the flanking domains whereas acetylation mainly occurs in the cationic repeats. Phosphorylation chiefly affects soluble tau whereas Lys acetylation predominantly affects aggregated tau at later stages of disease ^{27, 36–37}. For example, in AD, Ac-K280 is observed primarily in late Braak stages V-VI, whereas PHF1 and AT8 phosphorylation occur in preclinical brains. While the spatial and temporal patterns of these two PTMs differ, animal and cellular models found that acetylation of certain Lys residues can perturb phosphorylation levels in local or distant regions of the protein ^38–40. This “cross-talk” complicates the biochemical dissection of the impact of individual PTMs on the physiological function and pathological aggregation of tau.

High-resolution structural studies of in vitro assembled tau fibrils represent a potent tool to obtain residue-specific insights into the impact of PTMs on tau aggregation. By installing phospho-mimetic and acetyl-mimetic mutations at residues known to be frequently modified in diseased brains and determining their fibril structures, we can deduce how altered electrostatic charges and charge distributions modulate the tau fibril structure and fibrilization kinetics. Recently, we mimicked the phospho-epitopes of AT8 and PHF1 antibodies with Ser-to-glutamate (Glu) and Thr-to-Glu mutations ⁴¹. The AT8 epitope (phospho-Ser202, Thr205, Ser208) ^42–43 lies at the junction of the two proline-rich regions, P1 and P2 (Fig. 1), whereas the PHF1 epitope (phospho-Ser396, Ser404) ⁴⁴ lies at the junction of R’ and the C-terminal domain (CT). Within the PHF1 epitope, Ser400 and Thr403 phosphorylation is also enriched in AD compared to control brains ^{37, 45–46}. Thus, we installed four Glu mutations at these four residues to mimic the total charge of PHF1 phosphorylated tau. Installing these three and four phospho-mimetic mutations caused full-length 0N4R tau to form specific and distinct fibril structures ⁴¹ without anionic cofactors. AT8–3E tau formed a fibril core that ranges from R3 to the C-terminal domain (CT). This fold has not been observed in pathological brain tau. PHF1–4E tau (abbreviated as 4E tau) assembled into a triple-stranded structure, which resembles the folds of 4R tau diseases such as PSP and CBD ^{9, 15}. The absence of anionic cofactors in the fibrilization of these full-length tau proteins indicates that even a small reduction (by 3 and 4) of the net positive charges of the protein greatly increases the aggregation potential.

Figure 1.

(a) PHF1–4E 0N4R tau structure (PDB 8TTN, EMD- 41611). Many lysine sidechain densities are partially invisible (red), indicating that these sidechains have static or dynamic disorder. In contrast, K290, K311, K321 and K331 show full sidechain densities, indicating that they are immobilized. Acidic Asp or Glu residues are in close contact with some of these immobilized Lys sidechains, suggesting the formation of salt bridges. These ordered Lys residues are good targets of acetylation to disrupt the fold. (b) In vitro AD-tau fold (PDB 8Q8R; EMD-18258). K311, K321, and K369 all point towards the solvent and have disordered sidechains. These three residues are also highly acetylated based on proteomics data. Thus, they are good targets of acetyl-mimetic mutants to promote the AD fold. (c) Amino acid sequence of 4E3Q 0N4R tau, which contains four phospho-mimetic mutations at the PHF1 epitope and three acetyl-mimetic mutations at K311Q, K321Q and K369Q. (d) Sedimentation gel of cofactor-free fibrilization of 4E3Q tau over 14 days. The majority of the protein aggregated within two days (Fig S1). (e) Negative-stain TEM images of 4E3Q tau fibrils.

Compared to phosphorylation, much less is known about the effects of acetylation on tau aggregation, and no biophysical studies have reported the fibril structures of acetylated or pseudo-acetylated tau. To understand how acetylation interacts with phosphorylation to control tau fibril folds, here we investigate the fibril structure of full-length 0N4R tau in which three acetyl-mimetic Lys-to-glutamine (KQ) mutations are introduced in the cationic repeats. These mutations, K311Q, K321Q and K369Q, mimic the acetylation of these Lys residues in AD ^{35, 37}, and may disrupt key salt bridges in PHF1–4E tau and other 4R tau folds (vide infra). Adding three KQ mutations to the PHF1–4E construct thus caused a net reduction of the positive charge of wild-type full-length tau by 7. Using solid-state NMR (ssNMR), we show that this 4E3Q tau construct aggregates rapidly, forms well-ordered fibrils, and the fibril fold differs profoundly from the parent 4E tau fold, while remaining distinct from the AD tau structure.

Experimental Procedures

Cloning, expression and purification of tau

The gene for 4E3Q 0N4R tau was formed by modifying a 0N4R tau gene (GenScript) in a pET-28a vector with Gibson Assembly site-directed mutagenesis. In previous work ⁴¹, we had added S396E, S400E, T403E, S404E to 0N4R tau (Uniprot P10636–6), using amino acid numbering following the 2N4R sequence (Uniprot P10636–8). In this work we further added K311Q, K321Q, and K369Q via Gibson Assembly and confirmed mutagenesis via Sanger sequencing. The modified plasmid was transfected into E. coli BL21(DE3) competent cells. Colonies from fresh LB agar plates containing 50 μg/ml kanamycin were used to inoculate 10 ml LB starter culture containing 50 μg/ml kanamycin. After overnight growth at 37°C with 250 rpm shaking, 1 L LB with 50 μg/ml kanamycin was inoculated with the starter culture and allowed to grow at 37°C under 250 rpm shaking for 3–5 hr until reaching an OD₆₀₀ of 0.8. The cells were pelleted at 1,000 g for 10 minutes before gentle resuspension in 1 L M9 minimal media containing 2g/L ¹³C₆-D-glucose and 1 g/L ¹⁵NH₄Cl, and 50 μg/ml kanamycin. The cells were allowed to grow under similar 37°C with 250 rpm shaking conditions in M9 for 30 minutes before protein expression was induced with 1 mM IPTG and an additional 1g/L ¹³C₆-D-glucose. After 3–5 hours of expression the cells were harvested by centrifugation at 5,000 x g and pellets were frozen.

Tau was purified using a similar protocol as described before for full-length tau constructs ^{41, 47–48}. Cells were thawed and homogenized by vortexing in ice-cold lysis buffer containing 20 mM sodium phosphate at pH 7.4, 50 mM NaCl, 5 mM DTT, 0.1 mg/ml lysozyme and 1x cOmplete^™ protease inhibitor cocktail tablet (Roche) per 50 ml lysis buffer. Lysis was accomplished using a probe sonicator at 5s on / 5s off for 5 minutes of process time on ice, followed by 20 minutes in a boiling water, which caused most of non-tau proteins to aggregate. Boiled lysate was sedimented at 15,000 x g for 40 minutes, and the supernatant was applied to a cation exchange column (self-packed SP Sepharose Fast Flow resin, GE healthcare) equilibrated with 20 mM sodium phosphate buffer at pH 7.4, 50 mM NaCl, 2 mM DTT. After washing with equilibration buffer, tau was eluted with a gradient of the same buffer containing 1 M NaCl. Fractions containing tau were further purified by reverse-phase HPLC on a Zorbax 300SB-C3 column, 21.2 × 250 mm, 7 μm particle size, with a gradient of 5–50% acetonitrile over 45 minutes. Pure tau eluted at 33% acetonitrile and was lyophilized. The final yield was ~35 mg purified 4E3Q tau from 1 L labelled M9 media.

Cofactor-free fibrillization

Lyophilized 4E3Q tau was dissolved at 1.6 mg/mL in 50 mM K₂HPO₄ : KH₂PO₄ buffer, pH 6.8, containing 300 mM NaCl, 5 mM DTT, and 1x cOmplete^™ protease inhibitor cocktail tablet (Roche) per 40 ml fibrillization buffer. To ensure complete dissolution, the 17 ml monomer solution was bath sonicated for 5 minutes, before transferring into a 500 mL glass bottle and shaking at 180 rpm at 37°C with a 25 mm orbital diameter. Aliquots for sedimentation assays and TEM were taken at day 0 and every 2–3 days afterwards and frozen. To maintain reducing conditions, the bottle was flushed with N₂ gas after adding the monomer solution; each time the bottle was opened to take aliquots, 2 mM fresh DTT was added followed by another N₂ flush. To limit proteolysis, 1x fresh protease inhibitor cocktail was added at day 7. After 14 days, final aliquots for TEM and cryoEM were taken and frozen, and the remaining fibrillization reaction was ultracentrifuged at 100,000 g for 1 hour in a TLA-55 rotor. The pellet was partially dehydrated in a desiccation chamber before being manually transferred into a 3.2 mm pencil-style magic-angle-spinning (MAS) rotor using a needle. Additional water was added to the fibrils within the rotor as needed and incorporated by centrifugation in an Eppendorf microcentrifuge. About 25 mg of ¹³C, ¹⁵N-labeled fibrils were packed into the 3.2 mm MAS rotor.

Sedimentation assay

Sedimentation assays were conducted by ultracentrifuging a 20 μL aliquot of fibrillization reaction mixture for 60 min at 100,000 x g. The 20 μL supernatant was carefully removed, and the (invisible) pellet was resuspended with 20 μL of fibrillization buffer. 20 μL of 2x SDS sample buffer was added to both pellet and supernatant samples, then the solution was boiled at 95°C for 20 minutes. 2 μl of the solution was loaded onto the polyacrylamide gel and separated by SDS-PAGE. Protein was stained with Coomassie blue and imaged with an iPhone camera. Gel band densitometry was performed in ImageJ by integrating the color above the background for monomer and putative dimer peaks.

Transmission electron microscopy

Aliquots of fibril solutions were adsorbed onto freshly glow-discharged 200-mesh formvar/carbon-coated copper grids (Ted Pella), washed twice with 100 mM sodium acetate, then stained with 0.7% w/v uranyl formate for 15–30 s. TEM images were acquired on an FEI Tecnai T12 electron microscope. Three clear images were used to measure the apparent width of the negatively stained fibrils. The widths were manually measured from 10–20 fibrils in ImageJ.

Cryoelectron microscopy

2.5 μl aliquots of the day 14 fibrillization reaction were thawed and applied to glow-discharged R1.2/1.3 400 mesh carbon gold grids (SPI Supplies). The grids were plunged into liquid ethane after a 5 second wait and 5 second blot in a 4°C 95% relative humidity Thermo Fisher Scientific Vitrobot Mark IV system. The grids were transferred without warming to the Krios G3i with Bioquantum K3 camera with (Gatan) at the MIT.nano facility. Images were recorded at 300 kV with 20 eV energy filter, at a magnetization of 81,000 x for a nominal pixel size of 1.06 Å. Total electron exposure was 31 el/Ų and defocus was varied between −0.2 and −2.0 μm. Data was analyzed with the RELION-4.0 software ⁴⁹. Frames of raw EM movies were gain corrected, aligned, and dose weighted using RELION’s motion correction program ⁵⁰. Contrast transfer function parameters were estimated using CTFFIND-4.1⁵¹. Filament particle picking was attempted both manually and using a modified version of Topaz ^52–53. Particles were extracted in boxes of 640 pixels downscaled to 256 pixels for initial reference-free 2D classification. Multiple rounds of 2D classification were attempted, discarding only clearly non-fibrillar classes, which led to a series of high-resolution but non-twisting 2D classes (Fig. 2). Additional attempts to observe a twist by using larger boxes and hence longer views did not lead to the identification of twisting fibrils (Fig. S2). As no apparent helical twist was present in these samples, 3D helical reconstruction was not possible. Twenty high-resolution classes were used for fibril width measurements, counting pixels in RELION and multiplying them by the nominal pixel size in Å.

Figure 2.

(a) Representative cryoelectron micrograph of 4E3Q tau fibrils with zoomed inset. (b) High-resolution reference-free 2D class averages, showing no apparent twist. The lack of helical symmetry (i.e. twist) precludes helical reconstruction. White text indicates the estimated resolution of the classes in Å. (c-d) Additional class averaging attempts with 80 nm boxes to detect signs of twisting. To accommodate the larger views, the raw images were down-sampled to give a Nyquist frequency of 5.088 Å. (c) Single filaments with ~5 nm widths that slowly bend but do not twist. (d) Putative dimeric filaments that do not bend nor twist. These filaments do not show any apparent twists and thus lack helical symmetry.

Solid-state NMR

Solid-state NMR experiments were conducted on a Bruker AVANCE NEO 800 MHz (18.8 T) NMR spectrometer in the Francis Bitter Magnet Lab, using a BlackFox 3.2 mm HCN MAS probe. ¹³C chemical shifts were referenced externally to the adamantane CH₂ chemical shift at 38.48 ppm on the tetramethylsilane (TMS) scale, and ¹⁵N chemical shifts were referenced externally to the ¹⁵N peak of ¹⁵N-acetylvaline at 122.0 ppm on the liquid ammonia scale. Typical radiofrequency (rf) field strengths were 50–83 kHz for ¹H, 50–62.5 kHz for ¹³C, and 25–36 kHz for ¹⁵N. ¹H chemical shifts were calibrated externally using a hydrated DSS-containing POPC membrane sample. Reported temperatures are estimates of sample temperatures based on the measured water ¹H chemical shift. Experiments were conducted under 14 kHz MAS (Table S2).

Dipolar-coupling based polarization transfer experiments were used to selectively detect and characterize immobilized residues of the protein, while refocused-INEPT based experiments were used to selectively detect mobile residues. Resonance assignment was conducted using three-dimensional (3D) NCACX, NCOCX, CONCA, and CONCACB correlation experiments, in which polarization transfer was conducted using ¹H-¹⁵N or ¹H-¹³C cross polarization (CP), ¹⁵N-¹³C specific-CP ⁵⁴, ¹³C-¹³C CORD spin diffusion ⁵⁵, and Cα-Cβ DREAM ⁵⁶. These 3D experiments were conducted in blocks of 1–3 days each, during which rf carrier frequencies were adjusted for field drift between blocks. This enabled the multiple blocks of 3D data to be added in the time domain before Fourier transformation. Spectra were processed in Topspin 4.3 and were apodized either using the Gaussian window function with LB and GB values of −20 and 0.05 or using the QSINE window function with an SSB value of 3.5. Two-fold linear prediction was applied to truncated indirect dimensions in the 3D datasets. All spectra are plotted with the lowest contours chosen to show only a small amount of noise and with each successive contour at 1.2 x the intensity of the previous contour. Chemical shift assignment was conducted in NMRFAM-SPARKY⁵⁷. Secondary chemical shifts were calculated using TALOS-N ⁵⁸ and reported as ${\mathrm{\Delta }\delta }_{C\beta }-{\mathrm{\Delta }\delta }_{C\alpha }\equiv \left({\delta }_{C\beta ,expt}-{\delta }_{C\beta ,coil}\right)-({\delta }_{C\alpha ,expt}-{\delta }_{C\alpha ,coil})$. Positive ${\mathrm{\Delta }\delta }_{C\beta }-{\mathrm{\Delta }\delta }_{C\alpha }$ values indicate β-sheet whereas negative values indicate α-helical conformations.

Results

Our previous study showed that phospho-mimicry of the PHF1 epitope near residue 400 caused full-length 0N4R tau to form cofactor-free fibrils whose three-stranded fold resembles the topology of several ex vivo 4R tau such as PSP and globular glial tauopathy ⁴¹. Key to this in vitro 4E tau structure as well as the ex vivo 4R tau structures are several Lys-aspartate (Asp) salt bridges within the fibril cores. In the 4E tau structure, K311 and K321 show well-ordered sidechain densities that point to D295 and D283, respectively (Fig. 1a). In comparison, K311, K321 and K369 sidechains are disordered in the AD tau fold and point to the solvent ⁷ (Fig. 1b), consistent with mass spectrometry proteomics data that these three residues are highly acetylated in AD brain ³⁷. We hypothesized that acetylation of Lys residues that are disordered in AD tau but ordered in 4E tau should disrupt the triple-stranded structure of 4E tau that is characteristic of 4R tauopathies and may shift the protein structure toward the AD fold. To test this hypothesis, we mutated K311, K321, and K369 to glutamine (Q) in 4E tau to mimic lysine acetylation (Fig. 1c) and conducted cofactor-free fibrillization of this mutant full-length tau. Although several other lysine residues such as K280 and K353 are also commonly acetylated in AD tau, we did not mutate these residues because they do not show different sidechain disorder between 4E tau and AD tau, and because introducing a large number of mutations may complicate the interpretation of the effects of site-specific acetylation on the rigid core structure.

Following previous protocols for fibrillizing full-length 4E tau ⁴¹, we dissolved 1.6 mg/ml (40 μM) of lyophilized 4E3Q tau in 50 mM potassium phosphate buffer at pH 6.8 with 300 mM NaCl, and shook the solution for two weeks at 37°C. Reducing conditions were maintained throughout the reaction by adding 5 mM DTT every two days with nitrogen gas flushing to remove excess oxygen. Proteolysis was limited by adding protease inhibitor cocktails on day 0 and day 7. Sedimentation gels showed that the majority of soluble monomers on day 0 were recruited into sedimentable aggregates within the first two days of shaking incubation, with no further obvious changes over the remainder of the two-week fibrillization (Fig. 1d, Fig. S1). No proteolysis was observed during the reaction. Negative-stain transmission electron microscopy (TEM) data showed abundant straight fibrils that are 100–400 nm long and 9–15 nm wide (Fig. 1e, Fig. S2). Fibrils of 20–30 nm widths are also observed, which may be a wider view of the same fibrils or multiple filaments that are laterally associated. These widths may be over-estimates, as the disordered fuzzy coat that wrap the fibril core may also interact with the uranyl formate stain and thus be visible.

CryoEM micrographs of 4E3Q tau fibrils showed long filaments of either 5–8 nm in width or apparently paired into 10–12 nm width (Fig. 2a). Automated picking followed by reference-free 2D classification (Fig. 2b, Fig. S2) revealed high-resolution (~2.8 Å) views of 4.5–5.5 nm wide fibrils, with staggered 4.8 Å repeats along the fibril axis (Fig S3), indicating a cross-β amyloid fibril. Again, these views showed no observable twist, as the intensity pattern across the width of the fibril did not change along the fibril axis. Instead, the fibrils appear to bend before any twist could be observed. Reclassification with larger 80-nm diameter circular masks showed either single fibrils with ~5 nm widths that slowly bend but do not twist over the 80 nm views (Fig. 2c), or a variety of putative dimers that do not bend but also do not twist or vary over 80 nm along the fibril axis (Fig. 2d, Fig. S2). Thus, both negative-stain TEM and cryoEM data indicate that 4E3Q tau fibrils do not twist. Amyloid fibrils with very small or no twist have been reported for various proteins, including tau ^59–60 and the β-amyloid peptide ⁶¹. This lack of twist precludes helical reconstruction by cryoEM ⁶². Thus, we turned to MAS solid-state NMR to characterize the structure of these 4E3Q fibrils.

Two-dimensional ¹³C-¹³C (CC) and ¹⁵N-¹³C correlation spectra that selectively detect rigid residues provided the NMR fingerprint of the fibril core. The 2D NCACB and CC spectra show well-resolved signals with narrow linewidths of 1.0–1.2 ppm for ¹⁵N and 0.5–0.7 ppm for ¹³C (Fig. 3), indicating that 4E3Q tau fibrils adopt a homogeneous molecular conformation. Compared to 4E and 3E tau (Fig. 4c–f), 4E3Q tau exhibits distinct chemical shift patterns, indicating distinct molecular conformations. 4E3Q has a large number of serine peaks, similar to 3E tau but different from 4E tau, which has a much smaller number of serine peaks. 4E3Q tau has two well-resolved and assigned Thr peaks, compared to one in 4E tau and seven in 3E tau (Fig. 4b, d, f). Finally, 4E3Q tau has two resolved and assigned Ala peaks, compared to none in 4E tau and seven in 3E tau (Fig. 4b). In addition, the 4E3Q tau spectra also exhibit weak and broad Ala and Thr peaks at random coil chemical shifts as well as a strong Ala peak at α-helical chemical shifts. None of these peaks are observed in the rigid-selective 2D and 3D ¹⁵N-¹³C correlation spectra, suggesting that some of the disordered semi-mobile residues are partially rigidified, as previously found in other tau fibrils ⁴⁷.

Figure 3.

2D fingerprint solid-state NMR spectra of 4E3Q tau fibrils. (a) 2D NCACB spectrum. Positive peaks (blue) represent one-bond N-Cα correlations while negative peaks (red) mostly result from two-bond N-Cβ correlations, which are established by N-Cα polarization transfer followed by direct Cα-Cβ polarization transfer under the signal-inverting double-quantum DREAM sequence. (b) Aliphatic region of the 2D CC spectrum acquired with a short, 23 ms CORD, ¹³C-¹³C mixing period. Assignment for well-resolved resonances from 3D spectra are indicated. For clarity, not all assignments are indicated.

Figure 4.

Comparison of the 2D fingerprint NMR spectra of (a, b) 4E3Q tau, (c, d) 4E tau ⁴¹ and (e, f) 3E tau ⁴¹ fibrils. All three proteins were modified from full-length 0N4R tau sequences, and all were fibrillized without anionic cofactors under similar shaking conditions. (a, c, e) 2D NCA spectra. Note the different numbers and chemical shifts of the Gly region. (b, d, f) Ser/Thr region and Ala region of the 2D CC spectra. All assigned resonances for these fibrils are indicated in the 2D NCA spectra. In the 4E3Q tau spectra, random coil Thr and Ala resonances are indicated, capturing some partially mobile Thr and Ala without secondary structure that are not observed in the more rigid-selective 2D NCA and 3D correlation spectra. Similarly, α-helical Ala signals are observed in the 2D CC spectra of 4E3Q and 4E tau fibrils, representing intermediate mobile Ala residues, which are not observed in the 3D spectra.

We assigned the majority of strong resonances of 4E3Q tau sequence-specifically using rigid-selective 3D NCACX, NCOCX, CONCA, and CONCACB correlation experiments (Fig. S4). These 3D experiments allowed the unambiguous assignment of 84 residues from V275 to A437, encompassing R2, R3, R4, R’, and the CT (Table S1). No peak doubling was observed, thus there is no evidence of multiple folds for the rigid core. In addition, 21 weaker spin systems were assigned but could not be attributed to specific residues in the protein sequence. The 84 unambiguously assigned residues comprise the majority of R2 and the first part of R3, the majority of R4 and the first part of R’, and two small segments in the CT (Fig. 5a). A large portion of R3, from residues K317 to Q336, is missing from the rigid-selective spectra. This finding is surprising, as R3 is usually the most fibrillization-prone repeat and is an integral part of nearly all tau fibril structures to date. Because the vast majority of reasonably intense peaks in the rigid-selective spectra were assigned, we conclude that this second half of R3 is excluded from the rigid core and thus is either statically or dynamically disordered in the 4E3Q tau fibrils. This result is reminiscent of the 3E tau fibril ⁴¹ (Fig. 5c), which excludes H329-K353 at the end of R3 and the first half of R4 from the rigid core. These residues neither show signals in the rigid-selective ssNMR spectra nor display densities in the cryoEM maps ⁴¹. It is unclear whether the R3 disorder in the 4E3Q tau fibrils is due to the K321Q mutation in the center of this region, or due to the specific fold adopted by this acetyl-and phospho-mimetic construct. This question will require future studies.

Figure 5.

Cα and Cβ secondary chemical shifts of 4E3Q tau and several other 0N4R tau fibrils indicate the β-strand locations in the amino acid sequence. (a) 4E3Q tau. The three KQ mutation sites are indicated in red in the amino acid sequence. (b) PHF1–4E tau ⁴¹. Three Lys-Asp/Glu salt bridges are indicated. (c) AT8–3E tau ⁴¹. (d) 4E-tau (297–407) ⁶³, which adopts the AD fold. Red bars for V313, K321, K343 and H299 indicate large qualitative changes in the secondary chemical shifts of these residues among the four tau constructs. Compared to PHF1–4E tau, the three KQ mutations in 4E3Q tau mobilized the C-terminal half of R3, immobilized the N-terminal region of R’, and immobilized the C-terminal half of CT. Color coding of β-strands match that of the models in Figure 6.

Deviations of the backbone Cα and Cβ chemical shifts from random coil values provide information about the protein secondary structure. For tau fibrils, Cβ chemical shifts are often larger (downfield) from the random coil values whereas Cα chemical shifts are smaller. Both are indicative of β-sheet (ϕ, ψ) torsion angles of these rigid residues. Using the measured secondary chemical shifts, we assigned ten β-strands in 4E3Q tau fibrils. The last β-strand lies in the CT, well separated from the nineth strand at the beginning of the R’ repeat (Fig. 5a). These β-strands are separated by proline, glycine, and unassigned residues, which are putatively disordered. These secondary chemical shifts allowed comparison of the 4E3Q secondary structure with those of 4E and 3E tau fibrils ⁴¹ (Fig. 5b, c). The 4E tau β-strands start at the beginning of R2 and end in the middle of R4, whereas 3E tau β-strands are located in the R3 and from the C-terminal half of R4 to the C-terminus of the protein. The 4E tau β-strand locations are similar to those of ex vivo 4R tau fibrils whereas the 3E tau β-strand positions represent a distinct fold unseen in brain tau. Finally, the chemical shifts of AD-fold tau, recently obtained from a short construct (residues 294–407) (Fig. 5d), show mostly contiguous β-strands in R3, R4 and the N-terminal half of R’ ^{41, 63}.

Comparing 4E3Q to 4E tau, we find broad agreement of the β-strand locations, but key differences exist at V313, K343 and the C-terminal half of R3 (Fig. 5a, b). The 4E tau fold contains two sharp turns at the ³¹¹KPV³¹³ segment (Fig. 6a), which are absent in the PSP and CBD tau folds (Fig. 6b, c). The secondary chemical shifts indicate that these sharp turns are relaxed in the 4E3Q fibrils, most likely due to the K311Q mutation, which removes the potential for forming the K311-D295 salt bridge that is present in 4E tau. K343 exhibits helical chemical shifts in 4E3Q tau, indicating a break in the β-strand, whereas in 4E tau this residue is part of a regular β-strand. Interestingly, K343 conformation also differs between the two classes of ex vivo 4R tau folds (Fig. 6b, c): in PSP tau, K343 is in the middle of a regular β-strand whereas in CBD tau K343 occurs immediately after a sharp turn, which exposes the charged sidechains of E342 and K343 to solvent. Thus, the non-β-strand chemical shifts of K343 in 4E3Q tau fibrils suggest that the local conformation of K343 in 4E3Q tau may be similar to that of the CBD fold.

Figure 6.

Schematics of the three-dimensional folds of in vitro PHF1–4E tau fibrils (a), ex vivo PSP tau (b), CBD tau (c), and AD tau (d). Dashed ovals indicate the positions of K311, K321 and K369 and their salt bridges with acidic residues. In PHF1–4E tau, K311 and K321 form salt bridges (pink ovals) that should be removed by their acetylation. In PSP and CBD tau, K311 forms salt bridges that should be removed by acetylation. KQ mutations at these residues are thus expected to change the fibril core structure. In contrast, in AD tau (d), all three Lys residues point to the solvent (blue ovals) and are known to be acetylated in AD brain.

Comparing 4E3Q tau to 3E tau (Fig. 5a, c), we find that both proteins contain long disordered regions within their rigid cores, and both contain β-strands within the C-terminal domain. However, major differences exist in their secondary structures. 4E3Q tau contains an immobilized R2 repeat, contains the majority of its rigid residues in the R2-R4 repeats, and has a disordered segment in the C-terminal half of R3. In comparison, 3E tau excludes R2 from the rigid core, contains a disordered segment in the N-terminal half of R4, and has an immobilized R’-CT domain. Thus, the 4E3Q tau fold is qualitatively different from the 3E tau fold. The 4E3Q tau conformation may involve stacking of the R2-R3 segments with the R4-R’ segments, while the small stretch of the immobilized CT may be wrapped around the rest of the rigid core. In comparison, the 3E tau structure mainly consists of an R’-CT fold that is packed against part of the R3 repeat. The phospho-mimetic 4E mutations near residue 400 appear to create a zone of disorder in both 4E3Q and 4E tau fibrils, which is absent in 3E tau. This is consistent with the fact that in the 3E tau fold, residues near 400 maintain the interactions between the R3 β-strands and the CT. The negative charges introduced by the Glu residues thus seem to destabilize these interactions and prevent the R’-CT juncture from joining the core.

The presence of the R2 repeat in the rigid core of 4E3Q tau makes the protein structurally more similar to the PSP and CBD fold than to the AD fold (Fig. 5d, Fig. 6d). However, local conformational similarities are observed between 4E3Q tau and AD tau. K343 in the AD fold immediately follows a sharp turn to expose E342, K343, and D345 sidechains to the solvent. The non-β-strand chemical shifts of K343 in 4E3Q tau are consistent with such a turn. Likewise, V313 shows characteristic β-strand chemical shifts in 4E3Q tau, which are consistent with the β-strand chemical shifts of V313 in the AD fold. Therefore, these chemical shifts indicate that while 4E3Q tau did not adopt the AD fold due to the retention of R2 in the rigid core, the pseudo-acetylation perturbed the local conformations of some residues to be more similar to the AD fold, by disrupting Lys-Asp salt-bridges at K311, K321 and K369.

The presence of immobilized CT residues in 4E3Q tau fibrils that may stack against the R2-R’ rigid core raised the question of the mobility of the disordered linker from residues 375 to 418. To answer this question, we measured a J-based 2D ¹H-¹⁵N INEPT spectrum, which detects the most mobile residues in the fibrils (Fig. 7). The spectrum exhibits a small ¹H chemical shift dispersion, with all backbone amide ¹H peaks between 8.0 and 8.6 ppm, which is characteristic of intrinsically disordered regions in proteins. Thus, the most mobile residues of 4E3Q tau fibrils lack discernible structure. Compared to the chemical shifts of the mobile residues of heparin-fibrillized 2N4R tau protofilaments ⁶⁴, we find general agreement, with the exception of the C-terminal residues, which are largely absent from the 4E3Q tau spectrum. We thus conclude that the N-terminal residues 1–200 in 4E3Q tau are disordered and mobile, whereas the C-terminal domain has intermediate mobility or is rigid, consistent with the rigid-selective spectra. The number of immobilized CT residues in 4E3Q tau is less than in 3E tau but more than in 4E tau, whose CT residues were not observed in mobile-selective spectra nor in rigid-selective spectra ⁴¹. To our knowledge, no data is available so far on the mobile domains of ex vivo 4R tau fibrils. Future studies will be required to understand the dynamic structure of these mobile domains in tau fibrils.

Figure 7.

2D ¹H-¹⁵N J-INEPT MAS spectrum of 4E3Q tau fibrils, overlaid with the solution NMR chemical shifts of 2N4R tau protofibrils ⁶⁴ (red circles). Many C-terminal residues (assigned in blue), including the C-terminus L441, do not show signals in the spectrum, indicating that the C-terminal domain is immobilized.

Discussion

The first studies of tau acetylation focused on the use of the CREBS-binding protein (CBP) or its homolog, p300, to acetylate tau ^{23, 25}. But it was soon discovered that tau has intrinsic acetyltransferase activity that involved the use of its own cysteine residues, C291 and C322 ²⁶. Solution NMR has been used to quantify the in vitro tau acetylation levels by CBP, p300, or without added acetyltransferases ^65–66. These experiments found 10–40% acetylation at practically all of the 44 Lys residues of full-length tau. Higher levels of acetylation occurred in the center of the repeat domains where the two cysteines reside, including K280, K281, K290, K294, K298, K311, K317 and K321. Interestingly, the acetylation patterns were broadly similar with or without additional acetyltransferases, although the acetylation levels are lower without the enzymes. The lack of specificity of enzymatic acetylation is surprising given that studies of human tissue found that different diseases are characterized by different acetylation epitopes. For example, Ac-K311 is found in AD and 3R-only tauopathies³⁴ but not 4R tauopathies.

High-resolution structural studies can shed light on the acetylation specificity of tau. In AD, biochemical studies indicate that K311 is acetylated; consistently, the high-resolution structure shows the K311 sidechain to extend into the solvent ⁷. In comparison, in CBD tau, the K311 sidechain points towards the acidic sidechain of D295 to form a putative salt bridge ¹⁵. Consistent with this, mass spectrometry data found no K311 acetylation in CBD ^34–35. Thus, ex vivo tau fibril structures are in partial agreement with biochemical data of acetylation sites in brain tau. However, ex vivo structural data lack information about the time course of acetylation: it is not known whether K311 or other Lys residues that show disordered sidechains in the cryoEM maps is acetylated before or after the formation of the fibril core. Mass spectrometry data of multiple patient brains at different stages of disease have been used to suggest the time course of Lys acetylation during disease progression ³⁷. For example, Ac-K280 is found primarily in mature intracellular tau aggregates in Braak stages V-VI ³⁶, whereas PHF1 and AT8 phospho-epitopes stain tau at early stages of AD. But again, these PTM data do not indicate whether acetylation occurred before or after maturation of the fibrils.

While many studies have found that acetylation and acetyl-mimics promote tau aggregation and pathology, data that are inconsistent with this trend have also been reported. Transgenic mice expressing full-length human tau with K274Q and K281Q mutations led to tau mis-localization, impaired synaptic plasticity and memory deficits through pathways correlated with those seen in human AD ^30–31. In a Drosophila model expressing human tau, full-length K280Q tau exacerbated tau pathology, as seen by locomotion tests, and increased tau expression and phosphorylation at S262 ²⁸. Full-length K163Q, K280Q, K281Q, K369Q tau reduced phosphorylation at the AT8 epitope, increased phosphorylation at S262, and decreased microtubule binding ³⁸. Similarly, AT8 phosphorylation of full-length tau in QBI-293 cells was reduced when the protein was co-expressed with CBP acetyltransferase and when K280 and K281 were deleted or mutated to Gln ⁴⁰. K280Q and K281Q mutations caused 50% and 100% loss of microtubule polymerization in vitro, promoted heparin-based aggregation in vitro and facilitated seed-dependent aggregation in QBI-293 cells ⁴⁰. Thus, acetylation generally perturbs the tau phosphorylation landscape, inhibits microtubule binding, and promotes tau pathology and aggregation. On the other hand, some studies have found certain acetylation or acetyl-mimics to inhibit tau fibrillization. For example, K259Q and K290Q mutations to the aggregation-prone P301L tau increased seed-dependent fibrillization, whereas K321Q and K353Q mutations inhibited seed-dependent fibrillization ⁶⁷. In a similar contradiction, enzymatic acetylation accelerated polytetrafluoroethylene bead-induced fibrillization of 3R tau but inhibited 4R tau fibrillization ⁶⁶.

Two reasonable mechanisms could explain the inconsistent effects of acetylation on tau aggregation and pathology. First, in cell and animal models, distinct acetylation patterns of tau may modify the phosphorylation patterns of the protein ^{38–40, 68}, which can affect downstream aggregation and pathological pathways. Second, specific Lys acetylation may either stabilize or destabilize specific tau fibril folds. Without high-resolution structural characterization, this would appear only as acceleration or inhibition of the aggregation propensity. In vivo, the relative importance of these two mechanisms is unclear. In contrast, by removing phosphatases and kinases, in vitro studies prevent the indirect phosphorylation-mediated mechanism from playing a role. Thus, high-resolution structural studies of in vitro assembled acetyl-mimetic tau can allow the disentangling of these two mechanisms and may resolve the seemingly contradictory patterns. In 4E3Q tau, the entire region around K321Q, which is rigidified by the K321-D283 salt bridge in the parent 4E tau (Fig. 1a), is disordered. Based on this observation, we hypothesize that K321Q should disrupt an important salt bridge in the unknown P301L-seeded fold, thereby slowing fibrillization ⁶⁷.

In conclusion, we have determined the backbone conformation of cofactor-free 4E3Q tau fibrils containing three acetyl-mimetic mutations at K311Q, K321Q, and K369Q. The observed β-strand positions in the amino acid sequence indicate that these acetyl-mimetic mutations disrupted internal salt bridges in the parent 4E tau fold, suggesting that they may do so as well in ex vivo 4R tau folds. The three KQ mutations accelerated fibril formation, dramatically altered the protein fold compared to the parent 4E tau and changed the local conformation of specific residues to be more similar to AD tau. However, the retention of the R2 repeat in the rigid core makes the 4E3Q fold distinct from the AD fold. It remains to be seen whether the acetylation observed in ex vivo AD tau occurs before or after formation of the C-shaped structure. However, our results better support the model that in AD, these three Lys residues, K311, K321 and K369, are acetylated after rather than before fibril formation. If acetylation had occurred before the formation of the AD fold, then we would expect it to better promote the eventual AD fold. Instead, our data show that these acetyl-mimetics do not fully drive tau towards the AD fold, thus it is more likely that acetylation of these residues in AD occurred after fibrillization. This conclusion is also consistent with the fact that these Lys residues occur at solvent-exposed sites in the AD tau structure. Future studies will be required to elucidate the impact of acetylation on the structures and assembly pathways of pathological tau fibrils.

Supplementary Material

Additional sedimentation assay and electron microscopy data, as well as 3D correlation spectra and tables of NMR chemical shifts and experimental conditions. This material is available free of charge via the Internet at http://pubs.acs.org.