Structures of AT8 and PHF1 phosphomimetic tau: Insights into the posttranslational modification code of tau aggregation

Significance

Tau aggregates in neurodegenerative brains are hyperphosphorylated, but the molecular mechanism by which phosphorylation leads to tau aggregation is unknown. We introduced glutamate mutations at two commonly phosphorylated regions of tau that are targeted by antibodies AT8 and PHF1. Full-length 0N4R tau containing these mutations formed homogeneous fibrils with a single set of chemical shifts. AT8-3E tau formed a triangular core that encapsulates the C-terminal domain, whereas PHF1-4E tau formed a triple-stranded core. Remarkably, a combined mutant adopted the same conformation as PHF1-4E tau. These results have far-reaching implications, including redundancy and dominance in the impact of phosphorylation on tau structure and the development of pathological tau aggregates from phosphorylated tau monomers.

Abstract

The microtubule-associated protein tau aggregates into amyloid fibrils in Alzheimer’s disease and other neurodegenerative diseases. In these tauopathies, tau is hyperphosphorylated, suggesting that this posttranslational modification (PTM) may induce tau aggregation. Tau is also phosphorylated in normal developing brains. To investigate how tau phosphorylation induces amyloid fibrils, here we report the atomic structures of two phosphomimetic full-length tau fibrils assembled without anionic cofactors. We mutated key Ser and Thr residues to Glu in two regions of the protein. One construct contains three Glu mutations at the epitope of the anti-phospho-tau antibody AT8 (AT8-3E tau), whereas the other construct contains four Glu mutations at the epitope of the antibody PHF1 (PHF1-4E tau). Solid-state NMR data show that both phosphomimetic tau mutants form homogeneous fibrils with a single set of chemical shifts. The AT8-3E tau rigid core extends from the R3 repeat to the C terminus, whereas the PHF1-4E tau rigid core spans R2, R3, and R4 repeats. Cryoelectron microscopy data show that AT8-3E tau forms a triangular multi-layered core, whereas PHF1-4E tau forms a triple-stranded core. Interestingly, a construct combining all seven Glu mutations exhibits the same conformation as PHF1-4E tau. Scalar-coupled NMR data additionally reveal the dynamics and shape of the fuzzy coat surrounding the rigid cores. These results demonstrate that specific PTMs induce structurally specific tau aggregates, and the phosphorylation code of tau contains redundancy.

The microtubule-associated protein tau aggregates into β-sheet-rich amyloid fibrils in many neurodegenerative diseases (1–3). In Alzheimer’s disease (AD), the tau aggregates spread from the locus coeruleus to the entorhinal cortex, continue to the hippocampus, and then to the entire neocortex (4, 5). This spatial spreading correlates with cognitive decline and is the basis for the neuropathological staging of AD. Cryoelectron microscopy (cryo-EM) structures of AD paired helical filament (PHF) tau (6–9) show the same structure among multiple individuals, with or without bound small molecules that disaggregate or image the fibrils. Tau fibrils obtained from the brains of different tauopathies (10–13) have different structures. These results suggest that specific tau conformations exist in the complex cellular milieu and are linked to specific pathology. Therefore, elucidating the chemical and environmental factors that drive the prion-like propagation and spreading of these tau conformations (14, 15) is important for understanding the fundamental process of protein aggregation and for designing disease-specific therapeutic and imaging compounds.

Tau aggregation is counterintuitive because native tau is highly charged, highly soluble, and resistant to aggregation (16). The electrostatic charges in native tau are segregated in its modular amino acid sequence: The long N-terminal domain (NT) and the short C-terminal domain (CT) are negatively charged at neutral pH, whereas the central portion of the protein, comprising a proline (Pro)-rich region (P1 and P2) and several microtubule-binding repeats (R), is positively charged (Fig. 1A). The microtubule-binding repeats are R1, R2, R3, R4, and R′ in 4R tau isoforms, whereas 3R tau isoforms lack the R2 domain due to alternative splicing. In AD PHF tau, the negatively charged terminal domains form a disordered fuzzy coat around the positively charged rigid core (17, 18). Full-length 0N4R tau, the most common isoform in adult human brains, has a net charge of +15 at neutral pH. Given this cationic nature, it is not surprising that tau interacts with many anionic cellular species, including microtubules (19), heparin (20), RNA (21), and lipid membranes (22). Similarly, charge-modifying post-translational modifications (PTMs) are common in tau. The most studied tau PTM is phosphorylation. The protein composition of AD PHF tau was identified based on its abnormal phosphorylation of S396 (23). About 45 phosphorylation sites have been found in insoluble AD PHF tau (24), most of which are Ser-Pro and Thr-Pro motifs in the regions flanking the repeats. Among these phosphorylation sites, two motifs are recognized by anti-phospho-tau antibodies: a motif involving S202, T205, and S208 in the P2 domain (25–29), recognized by the antibody AT8, and a motif involving S396, S400, T403, and S404 between R′ and CT, recognized by the antibody PHF1 (30). AT8 is the most widely used antibody for detecting phosphorylated tau in AD and other tauopathies.

Fig. 1.

Amino acid sequences, fibril morphologies, and ssNMR spectra of full-length 0N4R tau fibrils containing phosphomimetic mutations. (A) Amino acid sequence domains of 0N4R tau. The estimated charge at pH 7.0 for each domain is indicated. The positions of the AT8 epitope and PHF1 epitope are indicated. The full amino acid sequence shows the AT8-3E mutations (S202E, T205E, and S208E) and PHF1-4E mutations (S396E, S400E, T403E, and S404E). (B) Negative stain TEM images of full-length AT8 and PHF1 mutant tau fibrils formed in the absence of cofactors. These fibrils were used for ssNMR and cryo-EM characterization. (C) 2D NCACB spectra of AT8 and PHF1 mutant tau. Assignments are obtained from 3D correlation spectra. ¹⁵N-¹³Cα cross-peaks have positive intensities, whereas ¹⁵N-¹³Cβ cross-peaks have negative intensities due to the double-quantum Cα-Cβ magnetization transfer. A small number of negative ¹⁵N-¹³Cγ cross-peaks are observed due to direct N-Cβ polarization transfer followed by Cβ-Cγ transfer. (D) Secondary structure–dependent chemical shifts of the rigid cores of AT8 and PHF1 mutant tau fibrils. The difference between the secondary chemical shifts of Cβ and Cα (δCβ–δCα) is plotted. Positive differences indicate β-strands, while negative differences indicate coil or helical conformations.

It is well recognized that abnormal phosphorylation, by reducing the positive charges, causes tau to detach from microtubules, thus initiating its self-assembly into filamentous aggregates. Biochemical evidence has been reported that tau phosphorylation at certain sites weakens microtubule binding (31) and induces self-assembly (29, 32). However, counter-evidence that phosphorylation at other sites reduces aggregation despite weaker binding to microtubules has also been reported (33, 34). In addition, tau phosphorylation occurs in human fetal brains and hibernating animals at some of the same sites as in tauopathy brains, but these phosphorylation events are apparently reversible (35–37). Because of the overlapping patterns of tau phosphorylation under pathological and physiological conditions, high-resolution structures of phosphorylated tau assemblies would be valuable for understanding how phosphorylation impacts tau aggregation.

Controlled site-specific phosphorylation of tau is challenging because brain extracts and kinases phosphorylate many Ser and Thr residues in a statistical manner. A mixture of tau species containing different numbers and sites of phosphorylated Ser and Thr would confound the analysis of the structural consequences of specific phosphorylation events. A recent study of tau phosphorylation at the AT8 epitope while avoiding S262 phosphorylation found that the resulting protein aggregated readily (29). This study used rat brain extracts, the ERK2 kinase, and mutagenesis to obtain the desired phosphorylation pattern. An alternative strategy is to use glutamate (Glu) mutations of Ser and Thr to mimic phosphorylation (34, 38, 39). Glu mutations in the Pro-rich region and C-terminal region of tau have been shown to weaken tau binding to microtubules and abolish microtubule nucleation, similar to the behavior of phosphorylated tau. The ability of phosphomimetic tau mutants to form fibrils has been examined using heparin: Some Glu mutations promoted fibril assembly, whereas others moderately reduced fibril formation. Overall, the similar behavior of Glu mutant tau and phosphorylated tau in their interactions with microtubules suggests that Glu phosphomimetic mutation is a useful approach for understanding the effects of charge-altering PTMs on tau aggregation.

To date, no high-resolution structures of full-length tau fibrils with phosphorylated side chains or phosphomimetic Glu mutations have been reported. Fluorescence spectroscopy has been used to study the global fold of soluble tau that contains Glu mutations at AT8 and PHF1 epitopes (40). Solution NMR has been used to study the local conformational dynamics of Glu-mutated soluble tau (29, 41) or kinase-phosphorylated soluble tau (42, 43). The ex vivo tau fibril structures obtained from brain tissues stained positive for anti-phospho-tau antibodies (6, 11), but no phosphorylated side chains are observed in these structures, suggesting that the phosphorylation sites may be heterogeneous, or the purification procedure for these samples may have perturbed the phosphate groups.

Here we report the atomic structures of two phosphomimetic full-length 0N4R tau fibrils assembled without anionic cofactors. The first tau construct includes three Glu mutations, S202E, T205E, and S208E, at the AT8 epitope in the P2 domain. We refer to this construct as AT8-3E tau, similar to the nomenclature in the literature (40). The second tau construct includes four Glu mutations at S396E, S400E, T403E, and S404E, at the PHF1 epitope in the CT. We refer to this sample as PHF1-4E tau. A third sample contains all seven Glu mutations and is called AT8/PHF1-7E tau. We show that the addition of only three and four negative charges is sufficient to induce cofactor-free homogeneous fibrils that exhibit a single set of ¹³C and ¹⁵N chemical shifts, indicating a single predominant molecular conformation. Combining solid-state NMR and cryo-EM, we determined the high-resolution structures of the rigid cores of these tau fibrils. Moreover, by detecting the dynamic residues using scalar-coupled NMR experiments, we obtained information about the dynamics and packing of the fuzzy coat around the rigid core. These data indicate that the introduction of negative charges in the Pro-rich region allosterically exposes the most amyloidogenic repeat of tau while sequestering the CT domain. Introduction of negative charges at the R′-CT junction creates a three-stranded core, which shares the same topology as three-layered ex vivo 4R tau fibrils. These results provide structurally based mechanistic hypotheses for the development of hyperphosphorylated tau aggregates in human brains.

Results

AT8 and PHF1 Phosphomimetic Tau Form Ordered Aggregates in the Absence of Anionic Cofactors.

The amino acid sequence of 0N4R tau with AT8 and PHF1 mutations is shown in Fig. 1A. We over-expressed the proteins in Escherichia coli and purified them by heat denaturation, cation-exchange column chromatography, and HPLC. Fibrils were assembled from monomer solutions at 0.4 mg/mL for AT8-3E tau and 1.6 mg/mL for PHF1-4E tau and AT8/PHF1-7E tau. The phosphate buffer (pH 6.8, 50 mM) contained 300 mM NaCl, 5 mM DTT, and protease inhibitors. The solutions were incubated at 37 °C with shaking at 250 rpm for 14 d, during which DTT was added every 2 d to prevent cysteine oxidation. No anionic cofactors were used. Fibrils began forming after 1 wk, and the reactions were stopped after 2 wk. Transmission electron micrographs revealed 200-nm long and 10- to 15-nm wide fibrils for the AT8 and PHF1 mutants (Fig. 1B). The cofactor-free fibrilization of these phosphomimetic tau samples differs from wild-type (WT) full-length tau, which does not form fibrils without polyanionic cofactors (44). Thus, the introduction of three or four negative charges is sufficient to induce self-assembly of full-length tau.

AT8 and PHF1 Phosphomimetic Tau Fibrils Have Different β-Sheet Conformations.

To determine the position of the rigid core in the amino acid sequences of AT8-3E tau and PHF1-4E tau fibrils and characterize their secondary structures, we measured 2D and 3D correlation magic-angle-spinning (MAS) NMR spectra. two-dimensional (2D) ¹³C-¹³C (CC) and NCACB spectra yielded fingerprints of the rigid cores. Both AT8-3E and PHF1-4E tau fibrils show well-resolved spectra (Fig. 1C), indicating that the rigid cores are well ordered. The 2D CC spectra exhibit fewer than 20 Ser and Thr peaks for AT8-3E tau and fewer than 10 peaks for PHF1-4E tau (SI Appendix, Figs. S1 and S2), suggesting that AT8-3E tau has a larger fibril core. To assign the ¹³C and ¹⁵N chemical shifts, we measured 3D NCACX, NCOCX, and CONCA correlation spectra (SI Appendix, Fig. S3). For AT8-3E tau, we assigned 106 residues from the beginning of R3 (V306) to the C terminus (G440), except for residues H330 to I354, which do not exhibit signals in the dipolar spectra, indicating that this segment is dynamically disordered (Fig. 1D and SI Appendix, Table S1). The assigned chemical shifts indicate the existence of ten β-strands, separated by Pro and Gly-rich segments such as ³²²CGS³²⁴, ³⁶⁴PGGG³⁶⁶, and ³⁸⁹GAE³⁹¹. The caspase cleavage site D421 shows non-β-strand chemical shifts and acts as the break between the eighth and ninth β-strands. For PHF1-4E tau, we assigned 68 residues from the beginning of R2 (V275) to the middle of R4 (L357) (SI Appendix, Table S2). This rigid core contains seven β-strands. For both fibrils, most of the high-intensity peaks are assigned, and a single set of chemical shifts are observed, indicating that both fibrils adopt a single predominant molecular conformation. Strikingly, when Glu mutations are introduced to both AT8 and PHF1 sites, the protein shows an almost identical 2D NCA spectrum as PHF1-4E tau (SI Appendix, Fig. S4), indicating that the 7E mutant has a similar structure as the 4E mutant.

Fibril Core Structures of AT8-3E Tau and PHF1-4E Tau.

With the rigid cores identified by NMR chemical shifts, we next turned to cryo-EM to obtain the three-dimensional folds of these phosphomimetic tau fibrils. Because AT8/PHF1-7E tau has the same spectra as PHF1-4E tau, we focused on the two primary mutants for structure determination. Fibrils were manually picked from ~11,000 micrographs for each sample (SI Appendix, Fig. S5 and Fig. 2). Manual alignment of the 2D classes led to an initial estimate of a cross-over length of 150 nm for AT8-3E tau and 80 nm for PHF1-4E tau. This cross-over length estimate was crucial for successful 3D classification and refinement, which resulted in a 2.6 Å resolution map for AT8-3E tau and a 2.4 Å map for PHF1-4E tau (SI Appendix, Table S3). The AT8-3E tau fibril core has a complex multilayered triangular shape, whereas the PHF1-4E tau fibril core exhibits a simple three-layered β-strand fold (Fig. 3). A single structural model was found for each fibril, consistent with the presence of a single set of chemical shifts. For both samples, the structures comprised one protofilament. The refined helical cross-over distance is 123 nm for AT8-3E tau and 79 nm for PHF1-4E tau (Fig. 2).

Fig. 2.

Cryo-EM data of two phosphomimetic full-length tau fibrils. (A–C) AT8-3E tau fibril data. (A) Representative micrograph. (B) Helical reconstruction of AT8-3E tau fibrils with a cross-over length of 123 nm. (C) Projected slice of the central 4.8 Å of the final 3D reconstructed map of AT8-3E tau. (D–F) PHF1-4E tau fibril data. (D) Representative micrograph. (E) Helical reconstruction of PHF1-4E tau fibrils with a cross-over length of 79 nm. (F) Projected slice of the central 4.8 Å of the final 3D reconstructed map.

Fig. 3.

High-resolution structures of the rigid cores of phosphomimetic 0N4R tau fibrils. (A) The AT8-3E tau rigid core adopts a complex triangular structure that spans the R3 to the CT, except for the last 7 residues of R3 and the first half of the R4, which are disordered. (B) The PHF1-4E tau rigid core spans R2, R3, and R4, which form antiparallel stacked β-strands that are separated by PGGG motifs.

Based on the chemical shift–derived β-strand positions and the cryo-EM densities, we generated atomic models of the two phosphomimetic tau fibrils. Out of 383 residues of 0N4R tau, AT8-3E tau comprises 115 residues in its rigid core, spanning residues K305-H330 and Q351-G440 (Fig. 3A). The discontinuity between H330 and Q351 is in excellent agreement with the lack of signals for these residues in the dipolar NMR spectra (Fig. 1D). Thus, the 3% of the particles used for helical reconstruction are consistent with dipolar NMR spectra that detect all rigid proteins. The triangular shape of AT8-3E tau rigid core is bounded by the first 25 residues of R3 (V306-H330) at the top, the second half of R4 (Q351-G367) on the left, and the first half of CT (R406-S416) on the right. This triangle is filled by four layers of β-strands in the interior. After the R4 ³⁶⁴PGGG³⁶⁷ segment on the left side of the triangle, the protein makes a sharp U-turn, packing the first half of R′ against the second half of R4. This R′ β-strand continues until it reaches the ³²²CGS³²⁴ motif in R3 and then makes a sharp right turn at H388 to stack the second half of R′ (H388-S400) against the R3 strand. This antiparallel R′-R3 stacking is stabilized by hydrophobic interactions between L315 and V398, a salt bridge between K317 and E391, and another salt bridge between K311 and D402 (SI Appendix, Fig. S6B). Notably, K311 is part of the amyloidogenic R3 hexapeptide, whereas D402 is part of the PHF1 epitope. C-terminal to D402, the protein spirals inward, turning sharply at H407, G415, D421, L428, E431, and S435. The chemical shifts of these residues are consistent with the non-β-strand conformations in the cryo-EM densities.

PHF1-4E tau adopts a simpler β-sheet structure than AT8-3E tau: The triple-stranded core is bookended by the PGGG motifs at the end of the R2 and R3 repeats (Fig. 3B). The outer layers are composed of R2 residues K274-P301 and R4 residues V337-H362. R3 forms the central strand and establishes numerous side-chain contacts with R2 and R4 strands. The antiparallel R3-R2 interface is stabilized by hydrophobic interactions such as I308-I297, polar interactions such as S316-S289, and salt-bridge interactions at K311-D295, D314-K290, and K321-D283. The R3-R4 interface is more polar, stabilized by side-chain pairs such as K317-D348, Q307-N359, and S305-T361. The three β-strands are nearly ideally matched with the three repeats, with the terminal PGGG in each repeat forming U-turns between the layers.

Although the three-dimensional folds of AT8 and PHF1 tau mutants are qualitatively different, the R3 domain adopts an uncannily similar conformation (SI Appendix, Fig. S6A). In both fibrils, the ³²⁰SKCGSLG³²⁶ segment forms an inverted-Ω fold, with the L325 side chain filling the space within the loop. One notable difference between the two fibrils is K311. In PHF1-4E tau, the K311 side-chain points to the same side of the β-strand as Y310, leading to a bent backbone that is consistent with the non-β-strand chemical shifts of ³¹¹KPV³¹³. This bent backbone allows K311 to form a salt bridge with D295 in R2. In AT8-3E tau, the K311 side chain lies on the opposite side of Y310 and P312, consistent with a canonical β-strand structure and the chemical shifts of these residues. In this orientation, K311 forms a putative salt bridge with D402.

Water Pockets Exist Inside the Fibril Cores.

To investigate whether water pockets exist inside the rigid cores, we measured water-edited two-dimensional (2D) CC and NCACB correlation spectra (SI Appendix, Fig. S7). By transferring water ¹H magnetization to the protein and detecting it through protein ¹³C signals, we probe the water accessibilities of individual residues. For AT8-3E tau, G440, V399, P397, and T403 are better hydrated than their neighboring residues, indicating the presence of a small water pocket that hydrates the C terminus (SI Appendix, Fig. S7D). L441 is not observed in either dipolar NMR spectra or cryo-EM densities, indicating that this C-terminal residue is disordered. On the left side of the triangular core, R′ residues K369-H388 are well hydrated, suggesting the presence of a water channel bounded by the disordered R4. At the top of the triangle, residues I308 and Y310 are water-inaccessible, suggesting that the R3 hexapeptide is shielded by the unidentified densities (Fig. 3A). In comparison, the H407-S416 stretch on the right side of the triangle is relatively water-accessible. This suggests that the parallel ridge of unidentified densities outside this segment is only present in some of the fibrils. Model building using Model Angelo (45) suggests that these unidentified densities may correspond to R2 residues K281-Q288, with a correlation coefficient of 0.3 between the map and the model (SI Appendix, Fig. S8). Since solid-state NMR spectra report the water accessibility of all proteins, whereas the cryo-EM structural model comes from a small percentage of all fibrils, these extra densities may reflect a subset of samples in which R2 is ordered and packed against the H407-S416 segment. Consistent with this interpretation, the N410-S416 segment has an average water-edited intensity ratio of ~0.25, whereas the R3 hexapeptide has an average water-edited intensity ratio of less than 0.2. These indicate that the R3 hexapeptide is more shielded from water by the extra densities compared to the CT segment H407-S416.

The PHF1-4E tau fibril exhibits low water accessibility for internal hydrophobic steric zippers such as ³⁰⁶VQIVY³¹⁰ and high water accessibility for surface residues such as ²⁹³SKDN²⁹⁶ (SI Appendix, Fig. S7E). Interestingly, the inverted Ω motif ³²⁰SKCGSLG³²⁶ is well hydrated even though they are packed on both sides by R2 and R4 residues. This suggests the presence of two water pockets, one between R2 and R3 and bounded by K281, D283, K321, and L325 and the other between R3 and R4 and bounded by S324, H329, E338, K340, and E342. These water pockets solvate the hydrophilic side chains in the interior of this fibril. In contrast, many surface-facing hydrophobic residues such as ³³²PGGGQV³³⁷ and the R2 hexapeptide ²⁷⁵VQIINK²⁸⁰ are water-inaccessible, suggesting that disordered segments are present to shield these hydrophobic residues from water exposure (vide infra).

The Three Phosphomimetic Tau Fibrils Have Different Fuzzy Coat Dynamics.

The dynamically disordered fuzzy coat of tau fibrils is invisible in dipolar NMR spectra and cryo-EM densities but is important for mediating the interactions of the β-sheet core with the environment (46). The tau fuzzy coat also contains most of the known phosphorylation sites, including AT8 and PHF1. Thus, it is important to investigate the dynamics and shape of the fuzzy coat and their perturbation by phosphorylation. We measured scalar-coupled 2D ¹³C-¹³C TOCSY and ¹H-¹⁵N INEPT spectra under MAS to selectively detect highly mobile residues in the three fibrils (Fig. 4). All three fibrils exhibit many sharp signals, indicating the presence of highly dynamic residues. The 2D ¹³C-¹³C TOCSY spectra resolved the signals of Glu residues that precede proline, Glu(Pro), from those that do not (Glu). Interestingly, the relative intensities of these Glu(Pro) and Glu peaks differ among the three samples (SI Appendix, Table S4A). AT8-3E tau has a higher Glu(Pro) intensity than Glu, whereas the other two fibrils have weaker Glu(Pro) intensities than Glu. Other pairs of residues, including Ser(Pro) versus Ser, Thr(Pro) versus Thr, and Ala(Pro) versus Ala, also show different intensity ratios among the three fibrils. The chemical shifts of most of the dynamic residues (SI Appendix, Table S4B) are near the random coil values.

Fig. 4.

The fuzzy coat of the three phosphomimetic tau fibrils has different dynamics. (A) Glu region of the 2D ¹³C-¹³C TOCSY spectra of AT8, PHF1, and AT8/PHF1 mutant tau fibrils. For AT8-3E tau, the Pro-preceding Glu (E(P)) peak is more intense than the Glu (E) peak. But for the other two fibrils, the E(P) peak is weaker than the E peak. (B) Diagnostic regions of the 2D ¹⁵N-¹H INEPT spectra of the three tau fibrils, showing intensity variations of the N-terminal residues E3, T17, and G43 among the three samples. These intensity variations indicate that AT8/PHF1-7E tau has a highly dynamic NT1, whereas AT8 and PHF1 mutant tau fibrils have partially immobilized NT1. (C) Distribution of Glu (blue lines) and Glu(Pro) (pink lines) residues in the amino acid sequence and summary of the mobilities of the three phosphomimetic tau fibrils based on the NMR spectra. (D) Models of fuzzy coat arrangement around the rigid core of the three tau mutants, based on the TOCSY and INEPT spectral intensities, and accounting for the NT/CT interactions found from previous FRET and solution NMR data (40, 41). Light blue: the most mobile segments; light green: the semimobile segments; light orange: the semirigid domains.

Similarly, 2D ¹⁵N-¹H INEPT spectra resolved many N-terminal residues, whose intensities show structurally informative variations among the three samples (SI Appendix, Fig. S10). AT8/PHF1-7E tau has much higher intensities than the other two mutants for N-terminal residues, particularly those in the NT1 domain, indicating that the combined mutant has a highly dynamic N terminus. In comparison, AT8-3E tau has the weakest intensities for NT1 residues among the three proteins, indicating that NT1 is partly immobilized in this mutant. These TOCSY and INEPT intensity patterns together indicate that NT1 and R1 are partially immobilized in AT8-3E tau whereas NT1, R′, and CT are partially immobilized in PHF1-4E tau (Fig. 4C). AT8/PHF1-7E tau has the most dynamic N-terminal region among the three samples, whereas its R′ and CT are partially immobilized. These results are schematically depicted in Fig. 4D. We hypothesize that the partially immobilized NT1 in AT8-3E tau lies close to R3, whereas the dynamic NT domain in AT8/PHF1-7E tau is well separated from the β-sheet core, which sequesters R3 in the interior.

Discussion

Distinct Tau Phosphorylation Patterns Have Distinct Amyloid Structures, and the Phosphorylation Code Contains Redundancy and Dominance.

The atomic-resolution structures obtained here provide detailed insights into the impact of electrostatic interactions on tau aggregation, the relationship between different phosphorylation sites, the propagation of post-translationally modified tau, and the dynamics of the fuzzy coat (Fig. 5). Our data show that the addition of a small number of negative charges to the amino acid sequence is sufficient to cause soluble tau to aggregate into specific structures without the need for anionic cofactors. The introduction of three negative charges in the P1-P2 junction resulted in a large β-sheet core that spans R3 to CT, except for ~20 residues between R3 and R4. Thus, reduction of positive charges in the Pro-rich region allosterically sequestered R′, the high-affinity MT-binding anchor of tau (47). It also exposed the most amyloidogenic segment of tau, ³⁰⁶VQIVYK³¹¹. The introduction of four negative charges at the R′-CT junction caused a triple-stranded β-sheet core that is evenly divided among R2, R3, and R4 repeats. Neither structures include the mutated residues, indicating that charge alterations impact the fibril core structure allosterically. The formation of these cofactor-free full-length tau fibrils underscores the notion that the native tau sequence is finely balanced between intrinsic disorder and order; even small electrostatic changes can tip the balance toward aggregation.

Fig. 5.

Proposed aggregation pathways of AT8 and PHF1 phosphorylated tau. (A) Schematic of the residual conformation of soluble tau bearing Glu mutations, obtained from previous FRET data (40). (B) Schematic of AT8-3E tau and PHF1-4E tau structures and their fuzzy coat arrangement determined in this work. The color scheme of the dynamic domains of the protein is the same as in Fig. 4D: light blue: the most mobile segments; light green: semi-mobile segments; light orange: semi-rigid domains. The + and − signs approximately indicate the number of charges of each domain at pH 7. (C) Three-dimensional folds of three ex vivo brain tau fibrils (13). The pink shade indicates the ³²²CGS³²⁴ motif, the green shade indicates the R4-R′ U-turn, the gray shade indicates the R2-R3 U-turn, and the blue circle indicates the unassigned cofactor density.

The qualitatively different structures of AT8-3E tau and PHF1-4E tau indicate that distinct phosphorylation patterns in the monomers lead to distinct amyloid structures. At the same time, the near identity of the NMR spectra of PHF1-4E tau and AT8/PHF1-7E tau indicates that the phosphorylation code of tau has redundancy: Different PTM patterns can lead to the same structure. When both AT8 and PHF1 sites are mutated, the resulting fibril adopts the same conformation as when PHF1 sites alone are mutated. Thus, the structural effects of AT8 mutation are abolished by PHF1 mutation, and the three-layered PHF1 tau structure dominates the triangular AT8-3E tau structure. This dominance could occur because specific interactions involving the AT8 sites are overridden by those of PHF1 or because PHF1-driven nucleation and elongation are faster than those of AT8. We can rationalize the former by the molecular interactions of S396, S400, T403, and S404 in the AT8 fold (SI Appendix, Fig. S6). S396 is embedded in a hydrophobic core surrounded by V393 and V398; thus, its replacement would likely destabilize the hydrophobic core of the CT. S400 interacts with I313 in R3, while T403 is adjacent to the D402 that forms a salt bridge with K311 in the R3 hexapeptide. Thus, mutations of these two residues would weaken CT-R3 interactions. Finally, phosphorylation or Glu mutation of S404 may cause the formation of a salt bridge with R406. Therefore, mutations of these four residues would destabilize the AT8 fold, yielding another polymorph. If the CT is dissociated from the R3, then the latter could form alternative steric zippers with R2 and R4, giving rise to the PHF1 fold (Fig. 5B). We expect specific molecular interactions such as those observed here to underlie the redundancy and dominance in tau PTM. This hypothesis can be tested using molecular dynamics simulations (48, 49). We postulate that this redundancy and dominance of PTMs of tau may be important in the brain by allowing specific pathological folds to emerge from the heterogeneous environment of the cell.

Pathological and Physiological Phosphorylation of Tau.

The high-resolution structures of these pseudo-phosphorylated tau fibrils provide a structural framework for understanding how tau phosphorylation might lead to pathological tau aggregates and how non-pathological aggregates could exist transiently. Tau is highly phosphorylated not only in tauopathy brains (23, 25, 30, 50) but also in normal brains in a developmentally controlled manner (35, 51, 52). Differentiating pathological from physiological phosphorylation requires detailed knowledge of the distribution of phosphorylated residues, extent of phosphorylation at each residue, the time course of phosphorylation events, and the spatial distribution of phosphorylated tau in diseased and healthy brains. Human fetal tau is highly phosphorylated but with different site-specific frequencies compared to AD tau (35). For example, antibody staining data show that S202, S396, and S404 are phosphorylated in both fetal and AD brains, whereas pT205 and pS208 are more prevalent in AD tau (36, 53). Mass spectrometry data show that pT205 is enriched in seeding-competent fractions of AD tau while the other AT8 residues are not (54). In hibernating animals, tau is phosphorylated at some of the AT8 and PHF1 residues such as S199, S202, and S404, but this phosphorylation is reversible after animal arousal (37, 55). The phosphorylated sites in hibernating animals are enriched in sarkosyl-soluble and seeding-incompetent fractions of AD tau (54), consistent with the fact that these phosphorylation events have little deleterious consequences. These data on physiological phosphorylated tau are consistent with emerging mass spectrometry data of the order of phosphorylation during the progression of AD (54). These data suggest that phosphorylation of N-terminal residues such as S198, S199, and S202 occurs early and exists in sarkosyl-soluble fractions of AD, whereas phosphorylation of S396, S400, T403, and S404 in the C-terminal region occurs later in AD. AT8-positive tissues are preferentially enriched in pretangles, whereas PHF1-positive inclusions are preferentially enriched in intracellular and extracellular neurofibrillary tangles (56). Finally, AT8 immunoreactive neurons have been observed from 1 to 100 y, increasing with age (4). Taken together, this biochemical evidence indicates that phosphorylation at AT8 and PHF1 epitopes is involved in both pathological and physiological tau, but to different degrees. This duality should be considered in understanding phosphomimetic tau structures.

Comparison of Phosphomimetic Tau Structures with Brain Tau Structures.

Comparison of the AT8 and PHF1 mutant tau structures with pathological brain tau structures is consistent with the current knowledge about the time course of phosphorylation at the two epitopes. The three-dimensional fold of AT8-3E tau does not have any analog among brain tau structures known to date. This uniqueness is consistent with biochemical evidence that the CT domain is cleaved early on during the development of AD, which would prevent the formation of the AT8-3E tau structure. It is also consistent with the observation that phosphorylation of the AT8 epitope occurs in physiological functions of tau and in the early stages of disease progression, neither of which have been studied structurally. Despite the unique three-dimensional fold, the R4-R′ U-turn at residues ³⁶⁰Ile-His³⁷⁴ resembles the structure of the same domain in Pick’s disease, globular glial tauopathy (GGT), and CBD-seeded tau in SH-SY5Y cells (57) (SI Appendix, Fig. S6C and Fig. 5C).

In contrast to AT8-3E tau, the three-stranded structure of PHF1-4E tau has the same topology as three-layered 4R brain tau (SI Appendix, Fig. S11), consistent with biochemical evidence that phosphorylation of the PHF1 epitope is a late-stage event during disease progression. The three-layered 4R brain tau fibrils use R2, R3, and R4 as the constituents of the three β-strands, similar to PHF1-4E tau. However, the precise β-strand stacking of brain tau is shifted relative to PHF1-4E tau. Using PSP tau and CBD tau to represent three-layered and four-layered brain tau proteins, it can be seen that both proteins have the same ³⁰²GGG³⁰² turn as in PHF1-4E tau, which induces the R2-R3 steric zipper. But both brain tau fibrils contain an unassigned cofactor near one of the two CGS motifs, which disorders the β-strands. In PSP, GGT, and GPT tau fibrils, the unassigned density lies in a spacious region formed by the ³²¹KCGS³²⁴ segment (Fig. 5C and SI Appendix, Fig. S11B), which forms a 90° turn. This turn differs from the AT8 and PHF1 tau ³²²CGS³²⁴ motif, which forms part of an inverted Ω that leaves R3 straight as a whole. The unassigned density in the three-layered brain tau structures may shift the R3 and R4 forward relative to cofactor-free PHF1 tau to wrap around the density. This perturbation may in turn lead to a backward shift of the R2 to stack against the R3 ³²²CGS³²² motif. In the four-layered CBD tau, the unassigned cofactor is found near ²⁹¹CGS²⁹³ (SI Appendix, Fig. S11C) and may cause a forward shift of R4 to stack against R2 to wrap around the density. We hypothesize that the absence of cofactors in the PHF1-4E tau is one of the reasons for the structural differences between brain tau and phosphomimetic tau. These differences are reflected by heavy-atom rmsd of 8.5 to 10 Å between PHF1-4E tau and the three-layered 4R brain tau structures for residues 306 to 362.

The CT domain in the AT8-3E tau structure is associated with R3 via residues at the PHF1 epitope. Thus, addition of PHF1-phosphorylated tau monomers or CT-truncated monomers to an AT8-3E tau fibril should disrupt the propagation of the AT8-3E tau structure (Fig. 5B). Future experiments will be required to understand how tau sequences bearing phosphorylation at the AT8 and PHF1 motifs develop into pathological brain tau fibrils. One possible fate of the AT8-3E tau fibril structure is that it might exist transiently and be cleared by the cell after dephosphorylation by phosphatases, which are active in the developing brain and in hibernating animals after arousal.

Shape and Dynamics of the Fuzzy Coat and Model of How Phosphomimetic Mutations Give Rise to These Fibril Structures.

In addition to the rigid core structures, we obtained information about the dynamics and shape of the fuzzy coat by selectively detecting the NMR signals of highly mobile residues. Previous fluorescence resonance energy transfer (FRET) and EPR data of unmodified monomeric tau found that the protein has a residual conformation that is shaped like a paperclip. The CT lies in the center, flanked by the NT on one side and the MT-binding repeats on the other (58). When S199, S202, and T205 are mutated to Glu, FRET data showed that the NT swung outward, loosening the paperclip (Fig. 5A) (40, 41). Mutating S396 and S404 to Glu caused the CT to approach the NT. Mutating all five residues to Glu caused the NT to approach the microtubule-binding repeats as well as the CT, tightening the paperclip. Because the residual conformation of soluble tau contributes to the nucleation of aggregates, and the chemical shifts of dynamic residues in soluble tau are in good agreement with those in fibrillar tau (59–62), it is reasonable to approximate the dynamic conformations of the fuzzy coat in fibrils by those in soluble tau (Fig. 4D). For AT8-3E tau fibrils, we hypothesize that the partially immobilized NT stack against R3 to account for the unresolved electron density on the exterior of R3. The R2 domain, which is also partially immobilized, likely shields the CT, accounting for the unresolved densities that line the H407-G415 segment. The P1 and P2 domains are highly mobile (Fig. 4C) and may constitute the most exposed surface of the fibril. In PHF1-4E tau, the NT is semi-mobile, suggesting that this domain might stack against R′, which could in turn lie outside R4. In the combined mutant, the negative charges at the AT8 epitope may repel the NT from the fibril core, thus explaining the increased dynamics of the NT observed in the INEPT spectra.

Why do Glu mutations at these two regions of the amino acid sequence produce these specific polymorphs? Both sets of Glu mutations occur at the boundary of two domains where the net electrostatic charge changes substantially: The AT8 mutations occur at the P1-P2 junction, whereas the PHF1 mutations occur at the R′-CT junction. It is known that tau phosphorylation reduces microtubule binding and increases tau accumulation in the cytoplasm (26, 31). Separated from its cellular partner, the R′ domain, which has a net positive charge of +3.4 at neutral pH, could associate with CT, which has a similar number of negative charges (−2.8). This self-association would be electrostatically balanced and would be stabilized by salt bridges such as between D430-E431 in CT and Y394-K395 in R′ and between D421 in CT and H374 in R′ (Fig. 3A). The R′-CT association could recruit R4 via the conformationally stable R4-R′ U-turn, while the negatively charged CT could recruit the next possible cationic repeat, R3. The rest of the protein is more intrinsically disordered. Thus, it would coat the R3-CT core, culminating with the negatively charged NT stacking against the positively charged R3.

When negative charges are added to the R′-CT junction, the modified CT acquires a larger number of negative charges than can be balanced by the positive charges in R′. We hypothesize that this may be the reason for the anionic CT to be distributed over the exterior of three cationic repeats (R2-R4). Although the structure of cofactor-free WT tau fibrils has not been determined due to the high solubility of unmodified full-length tau, the residual conformation of Glu-mutated soluble tau deduced from the FRET data is broadly consistent with the fibril structures and dynamics determined here (Figs. 4D and 5 A and B). This suggests that the AT8 and PHF1 tau fibril structures reflect the intrinsic propensity of the Glu-mutated monomers, representing phosphorylation mimicry, instead of being the result of polymorphism. Finally, both AT8 and PHF1 epitopes occur in the disordered regions of these fibrils, outside the rigid cores. Thus, the different sizes and shapes of the Glu side chain from the phosphorylated Ser and Thr side chains are not expected to impact the rigid core structures. The larger negative charge of phosphorylated Ser compared to Glu at neutral pH could affect the fibril structure. However, since phosphorylation of Ser and Thr residues in brain tau is often incomplete and low-frequency, 100% Glu mutation at each residue may acquire a similar number of negative charges as partially phosphorylated Ser and Thr residues. Future studies will be required to determine whether phosphorylated tau fibrils have the same structures as the phosphomimetic tau fibrils.

Conclusions

We have determined the structures of two full-length tau fibrils that contain phosphomimetic mutations at two of the most common phosphorylated segments in tauopathies. We hypothesize that these structures form as a result of charge distribution in the amino acid sequence and the relative disorder of the tau domains. Future experiments will determine whether these two phosphomimetic tau structures will evolve into disease-relevant structures under specific cellular conditions (63). The AT8-3E tau aggregate may also exist as a transient species that is reversible in healthy brains. Our results suggest that the distinct tau conformations in tauopathies may be caused by distinct patterns of PTMs (64). In the brain, tau aggregates have heterogeneous PTMs and variable isoforms. It is not yet understood which PTMs occur prior to, during, or after fibril assembly. Therefore, the structures of homogeneous phosphomimetic tau fibrils provide important insights and testable hypotheses about the molecular species during the development of tau aggregates in the human brain.

The integration of solid-state NMR and cryo-EM in this study is synergistic for understanding both the rigid core structure and the fuzzy coat dynamics. The remarkable agreement between the chemical-shift derived β-strand locations in the amino acid sequence and the cryo-EM densities indicate that all the proteins detectable by dipolar NMR experiments have a predominant structure that is the same as the less than 10% of the particles used for helical reconstruction. The three-dimensional folds of these structures are readily determined by helical reconstruction, whereas the dynamics of the protein and the water accessibility of the rigid core are readily obtained from solid-state NMR.

Materials and Methods

Three full-length 0N4R tau constructs were cloned, expressed, and purified in this work. These tau constructs contain three Glu mutations for AT8-3E tau (S202E, T205E, and S208E), four Glu mutations for PHF1-4E tau (S396E, S400E, T403E, and S404E), and all seven Glu mutations for AT8/PHF1-7E tau. The proteins were expressed in E. Coli and purified using cation exchange chromatography and HPLC. The fibril growth solutions were pH 6.8 phosphate buffer with 0.4-1.6 mg/mL protein, without any anionic cofactor. The solutions were shaken at 250 rpm and 37 °C for 14 d to obtain fibrils, during which DTT was added every 2 d to maintain a reducing environment. Solid-state NMR 2D and 3D correlation spectra were measured on an 800 MHz Bruker AVANCE spectrometer using a 3.2-mm BlackFox ¹H/¹³C/¹⁵N MAS probe. Negative-stain TEM images of unconcentrated fibril solutions were measured on an FEI Tecnai T12 electron microscope, whereas cryo-EM images were measured on a Krios G3i microscope. The spectra of dynamic residues were measured under MAS using scalar coupling–based experiments.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

CryoEM structures data have been deposited in PDB and EMDB (8TTL and 8TTN in PDB, and 41610 and 41611 in EMDB) (65, 66). All other data are included in the manuscript and/or SI Appendix.