Three- and Four-repeat Tau Coassemble into Heterogeneous Filaments

Tau filaments are the pathological hallmark of numerous neurodegenerative diseases including Alzheimer disease, Pick disease, and progressive supranuclear palsy. In the adult human brain, six isoforms are expressed that differ by the presence or absence of the second of four semiconserved repeats. As a consequence, half of the tau isoforms have three repeats (3R tau), whereas the other half of the isoforms have four repeats (4R tau). Tauopathies can be characterized based on the isoform composition of their filaments. Alzheimer disease filamentous inclusions contain all isoforms. Pick disease filaments contain 3R tau. Progressive supranuclear palsy filaments contain 4R tau. Here, we used site-directed spin labeling of recombinant tau in conjunction with electron paramagnetic resonance spectroscopy to obtain structural insights into these filaments. We find that filaments of 4R tau and 3R tau share a highly ordered core structure in the third repeat with parallel, in-register arrangement of β-strands. This structure is conserved regardless of whether full-length isoforms (htau40 and htau23) or truncated constructs (K18 and K19) are used. When mixed, 3R tau and 4R tau coassemble into heterogeneous filaments. These filaments share the highly ordered core in the third repeat; however, they differ in their overall composition. Our findings indicate that at least three distinct types of filaments exist: homogeneous 3R tau, homogeneous 4R tau, and heterogeneous 3R/4R tau. These results suggest that individual filaments found in Alzheimer disease are structurally distinct from those in the 3R and 4R tauopathies.

Alzheimer disease (AD) 2 is the most common fatal neurodegenerative disease. It is characterized by extracellular plaques of amyloid ␤ and intracellular inclusions of tau (1). The inclusions are composed of tau filaments bundled into neurofibrillary tangles. Tau pathology in the absence of amyloid ␤ deposits is the hallmark of Ͼ20 additional neurodegenerative diseases, including progressive supranuclear palsy, corticobasal degeneration, and Pick disease (2). A clear link between tau and disease has been established through the discovery of various mutations in the tau gene (3)(4)(5) resulting in a diverse group of familial forms of frontotemporal dementia and parkinsonism. In the central nervous system of the adult human brain, six tau isoforms are expressed ranging in size from 352 to 441 amino acids (6 -8) (Fig. 1A). These isoforms are defined by the presence or absence of a 29-or 58amino acid insert in the N terminus and the inclusion or exclusion of a 31-amino acid repeat segment in the C terminus. As a consequence, three of the isoforms contain four repeats, whereas the other three isoforms contain three repeats. The repeats carry out the major known biological function of tau: the binding to and stabilization of microtubules (9,10). Upon self-aggregation into filaments, it is the repeat region that is protected from proteases (11,12), establishing the structured core. The N-and C-terminal regions by contrast remain largely disordered, forming a fuzzy coat (13). In the monomeric proteins, these regions appear to fold back onto the repeats (14), forming transient intramolecular contacts (15). These interactions, together with the overall positive charge of the protein, greatly inhibit aggregation. As a consequence, the addition of negatively charged cofactors such as heparin (16) and removal of the N and C termini, exemplified by the 4R and 3R constructs K18 and K19 (17) (Fig. 1B), accelerate aggregation. The filaments of both full-length (18) and truncated tau (19) have a cross-␤ structure in which ␤-strands run perpendicular to the long fiber axis and are spaced ϳ4.7 Å apart. This structural feature is a common denominator of fibrils belonging to the amyloid class (20). AD tau filaments are composed of all 3R and 4R isoforms (21,22). Whether these isoforms segregate into homogeneous filaments or coassemble into heterogeneous filaments is currently unknown. The role of tau filaments in disease has been controversial (23). In particular, it has been debated whether filaments might have an altogether protective function by preventing individual tau molecules from partitioning into toxic oligomers (24). Recent findings suggest that tau filaments could play an important role in the propagation of the misfolded state from one cell to the next (25). In tissue culture, filaments transferred among neighboring cells (26). In addition, injections of insoluble tau into mouse brain resulted in the spreading of misfolding through different brain regions (27). These properties are reminiscent of those observed for prion proteins (28). Although the overall evidence for prion-type spreading mechanisms in tauopathies is scarce, the cited publications have lent new urgency and relevance to better understand the structure of tau filaments. The analysis of fila-ments has been challenging due to their large size and low solubility.
Site-directed spin labeling in combination with electron paramagnetic resonance (EPR) spectroscopy, which has been systematically developed for the study of protein structure and dynamics (29 -32), has proved to be an effective tool in the investigation of tau filaments (33,34) and other amyloid fibrils (reviewed in Ref. 35). The spin label reports faithfully on the conformation of the protein backbone regardless of the size of the investigated species. Here, we have used this technique to obtain new insights into the conformations of different tau filaments. Our findings have direct implications for the structures of filaments observed in AD and other tauopathies.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-In a first step, two PCR reactions generated the DNA fragments of the large isoform of tau (htau40) and fetal tau (htau23) with NcoI and XhoI cleavage sites at their 5Ј and 3Ј ends. In these reactions, htau40 in pRK172 (provided by Dr. Goedert at the Medical Research Council Laboratory, Cambridge, UK) and htau23 in pBluescript (provided by Dr. Matt Farrer at the Mayo Clinic, Jacksonville, FL) served as templates. The fragments were subsequently cloned into the NcoI/XhoI cleavage sites of pET-28b. The two cysteines in htau40 and the single cysteine in htau23 were replaced by serines. In a second set of PCR reactions, fragments K18 and K19 (36) were generated, which then were cloned into the NcoI/XhoI cleavage sites of pET-28b. In a next step, single cysteines were introduced by site-directed mutagenesis using the QuikChange method from Stratagene/Agilent Technologies. The correctness of all sequences was verified by DNA sequencing.
Protein Expression and Purification-All proteins were expressed and purified according to previously published protocols that involved htau40 (37). In brief, proteins were expressed in the Escherichia coli strain BL21 (DE3). The bacteria were pelleted and taken up in resuspension buffer (20 mM Pipes, pH 6.5, 500 mM NaCl, 1 mM EDTA, 50 mM 2-mercaptoethanol). The cells were ruptured through consecutive steps of heating (20 min at 80°C) and sonification. Centrifugation for 20 min at 15,000 ϫ g separated soluble tau from cellular debris. The addition of ammonium sulfate to the supernatant (60% mass per volume) precipitated tau. After 1 h, the precipitate was spun down for 15 min at 15,000 ϫ g. The protein was resuspended (H 2 O, 2 mM DTT), passed through a syringe filter, and loaded onto a Mono S column (GE Healthcare). Elution occurred through the application of a NaCl gradient. After SDS-PAGE analysis, correct protein fractions were pooled and stored at Ϫ80°C. Further purification was achieved by applying the samples onto a preparative Superdex 200 column. Peak fractions were combined and precipitated overnight by an equal volume of methanol containing 5 mM DTT. In the cases of K18 and K19, a 3-fold excess of acetone (5 mM DTT) was used for precipitation instead. The samples were pelleted and redistributed into equal aliquots. After centrifugation and a final washing of pellets with methanol (2 mM DTT), proteins were stored at Ϫ80°C.
EPR Data Collection and Analysis-All protein pellets were transferred into borosilicate capillaries (0.6-mm inner diameter ϫ 0.84-mm outer diameter, VitroCom, Mountain Lakes, NJ). Measurements were carried out in a Bruker EMX spectrometer fitted with an ER 4119HS resonator. All spectra had a scan width of 150 G and an incident microwave power of 12 milliwatts. Spectra were normalized to the same number of spins using double integration. A minor background originating from Ͻ0.5% soluble tau was subtracted from all spectra.
Negative Stain Electron Microscopy-250-mesh carboncoated copper grids were placed for 1 min on 10-l drops of tau filaments (3-4 M) and for 30 s on 10-l drops of 2% uranyl acetate and subsequently air-dried on filter paper. All images were taken with a Philips/FEI Tecnai-12 electron trans-FIGURE 1. Bar diagrams of tau isoforms and constructs. A, tau isoforms are defined by the presence or absence of two inserts in the N-terminal half (striped) and the inclusion or exclusion of the second microtubule binding repeat in the C-terminal half (marked as 2). Isoform sizes range from 352 amino acids (aa) to 441 amino acids and are provided on the right. B, the truncated constructs K18 and K19 contain the repeats (four for K18 and three for K19) plus three additional residues at the C termini. NOVEMBER 26, 2010 • VOLUME 285 • NUMBER 48 mission microscope operating at 80 keV and equipped with a Gatan CCD camera.

RESULTS
Defining the Stable Ordered Core in Filaments of htau40-Previous investigations on the conformation of tau filaments have revealed that residues in the third repeat of tau comprise a highly ordered core with parallel, in-register arrangement of ␤-strands (33). As these studies only included residues up to position 320, the first half of repeat 3, it remained unclear how far the crystal-like core structure would extend. In a first step toward addressing this question, we labeled the naturally occurring cysteine at position 322 with the paramagnetic nitroxide label [1-oxyl-2,2,5,5-tetramethyl-⌬3pyrroline-3-methyl]-methanethiosulfonate (38). Cross-linking resulted in the new side chain R1 ( Fig. 2A). The monomeric protein produced an EPR spectrum with three sharp lines characteristic for intrinsically disordered regions (Fig. 2B, left panel). After induction with heparin and 12 days of incubation under agitation, a single-line EPR spectrum was obtained (Fig. 2B, right panel). This spectrum is a signature feature for parallel, in-register arrangement of ␤-strands (35), indicating that the ordered core extends beyond the previously published region. Negative stain electron microscopy confirmed the presence of filaments (Fig. 2C). Next, we spin-labeled tau mutants at single cysteines ranging from position 321 to position 358. All monomeric proteins revealed EPR spectra with three sharp lines (supplemental Fig. S1), indicating a predominance of random structure. These findings are in good agreement with previous studies concluding that tau is an intrinsically disordered protein (40,41). The formation of filaments resulted in significant changes in line shape. Close inspection of all spectra (Fig. 3A) reveals two distinct regions. One region (residues 321-336), which extends to the end of repeat 3, gives rise to single-line spectra. Consequently, this region is part of the highly ordered core. The other region (residues 337-358) starting with repeat 4 produces spectra with single lines (e.g. 348), three lines (e.g. 346), or mixtures thereof (e.g. 357). Such spectral heterogeneity was previously observed for repeat 2 (34) and revealed a less stable, independent domain. Single amino acid substitutions in repeat 2 caused a conformational switch between stacked and disordered conformations (34). In line with the parallel, in-register arrangement of ␤-strands in that region, introduction of charged residues prevented folding, whereas introduction of ␤-branched residues promoted folding.
Here, we observe a similar trend for repeat 4. The substitution of Asp-348, for example, results in a single-line spectrum (representing ordered structure), whereas substitution of Val-350 produces a three-line spectrum (representing disordered structure). Importantly, single amino acid substitutions in repeat 3 (Fig. 3A) (33) produce single-line spectra, regardless of which amino acid is replaced. The signal amplitude increases linearly with increasing amounts of mobile component (34). An amplitude plot (Fig. 3B) illustrates low mobility for residues in repeat 3 and varied mobility for residues in repeat 4. Combined with our previous observations (33), these data reveal a stable, highly ordered core structure of the third repeat that is flanked by marginally stable regions that can switch between conformationally distinct states.
Preservation of Core Structure in htau23 and Smaller Tau Fragments-Our investigations thus far have focused solely on the largest isoform (htau40) containing four microtubule binding repeats. Here, we asked whether fetal tau (htau23), which lacks the second repeat, contains a similarly ordered core. To address this question, we spin-labeled six different single cysteine mutants in the third repeat (positions 309, 310, 311, 317, 322, and 326). All monomeric proteins produced EPR spectra with three sharp lines (supplemental Fig. S2). This observation is again in good agreement with the current notion that soluble tau is intrinsically disordered. Next, we induced filament formation by the addition of heparin. Similar to htau40, assembly was allowed to proceed for 12 days while agitating at room temperature. Filaments were harvested by centrifugation and placed into EPR capillaries for spectroscopic measurement. All positions investigated revealed single-line EPR spectra (Fig. 4, left column), indicating parallel, in-register arrangement of ␤-strands. Next, we asked whether smaller fragments of tau that have been used in the past as valuable models for tau aggregation (see for example Ref. 17) would reveal a similar structural arrangement. For that purpose, we produced an identical set of single cysteine mutants of the 4R and 3R constructs K18 and K19 (Fig. 1). As expected, the spin-labeled proteins in their soluble form produced spectra with three sharp lines (supplemental Fig. S2), indicating random structure. All filaments that were subsequently formed gave single-line spectra (Fig. 4, center and right columns) similar to those of their full-length counterparts. These results suggest that the parallel, in-register ar-rangement of ␤-strands is an intrinsic feature of the repeat region and may be a common theme in all filaments of isoforms and constructs containing the stable third repeat.
3R and 4R Tau Combine to Form Heterogeneous Filaments-Because of the structural similarities of filaments formed from 3R and 4R tau (see also Refs. 33 and 42), the question arose whether these constructs could assemble into mixed filaments. In a first step, we attached a paramagnetic label at position 310 of K18 and a nonparamagnetic label (which produces no EPR signal) at position 310 of K19. This resulted in the side chains R1 ( Fig. 2A) and R1Ј (39), respectively. Both proteins were subsequently mixed at different molar ratios. After 3 days of incubation, all filaments were sedimented at 100,000 ϫ g and transferred into EPR capillaries. At low spin concentration (10% R1), a three-line EPR spectrum was obtained (Fig. 5A, red trace) with outer peak separation of 68 G. Such a spectrum is typical for an immobilized site with little interaction between spin labels (37). At intermediate spin concentration (60% R1), the spectrum broadened and the amplitude of the central peak decreased (Fig. 5A, green trace). These changes are due to dipolar interactions between labels, which are effective at distances between 8 and 25 Å (31). At high spin concentration (100% R1), a single-line spectrum with increased amplitude was observed (Fig. 5A, black trace). This spectrum is due to spin exchange between contacting labels (35). The observation of a three-line EPR spectrum at low spin concentration (10% R1) suggests that K18 has coassembled with K19. Segregation into separate filaments can be excluded as this would have retained the contacts between individual spin labels resulting in a single-line spectrum. To obtain a more comprehensive picture of the coassembly between K18 and K19, we performed a titration series in which the concentration of K18R1 was successively increased from 5 to 100%. The spectra of the filaments were taken and amplitudes of the central lines were plotted as a function of the mole percent of K18R1 (Fig. 5B). The amplitude changes are similar to those that had been previously observed in other FIGURE 3. EPR analysis of spin-labeled htau40 filaments. A, residues 321-358 were individually replaced by the side chain R1. All spectra were taken at 150 G and normalized to the same number of spins. Note that for completeness, spectrum C322R1 is reproduced from Fig. 2. The dotted line between the left and middle columns marks the boundary between repeats 3 and 4. B, the amplitudes of all EPR spectra are plotted against respective residue numbers. All amplitudes are normalized to spectrum F346R1, which has the highest amplitude.  NOVEMBER 26, 2010 • VOLUME 285 • NUMBER 48 spin dilution experiments (33,43,44). Depending on what type of spin interaction dominates at a particular dilution, different amplitudes are obtained: 1) large amplitudes for no interaction, 2) small amplitudes for dipolar coupling, and 3) slightly elevated amplitudes for spin exchange (see above). These changes occur gradually as the concentration of paramagnetic label increases and the distances between individual labels decrease. The various spacings between labels in differently composed filaments (Fig. 5B, insets I-III) can explain these amplitude changes. A very similar amplitude plot as depicted here is obtained when instead of K18, K19 is attached to the paramagnetic label (supplemental Fig. S3). This suggests that K18 and K19 have similar probabilities of integrating into the filaments. To clearly identify the core structure of heterogeneous filaments, we next mixed equimolar ratios of paramagnetically labeled K18 and K19. The resulting filaments produced a single-line EPR spectrum (Fig. 5C), suggesting crystal-type order with parallel, in-register arrangement of ␤-strands. Electron micrographs confirmed the presence of twisted and bundled filaments (Fig. 5D). As expected for a sample with an EPR single-line signature (33), amorphous aggregates were not observed.

3R and 4R Tau Coassemble into Heterogeneous Filaments
Next, we asked whether full-length versions of 3R and 4R tau would also be able to coassemble. For this purpose, we used the same labeling strategy as for the K18/K19 constructs. Paramagnetic label was attached to htau40; nonparamagnetic label was attached to htau23. The assembled filaments produced similar spectra to those of the smaller constructs (Fig.  5E, compare with 5A). Accordingly, the amplitude plot obtained from a titration series of these isoforms revealed the same overall features (Fig. 5F) as the one obtained for K18 and K19 ( Fig. 5B and supplemental Fig. S3). Furthermore, filaments formed from an equimolar mixture of paramagnetically labeled htau40 and htau23 produced a single-line spectrum (Fig. 5G). The presence of filaments and absence of amorphous aggregates was confirmed by negative stain electron microscopy (Fig. 5H). Combined, these results demonstrate that regardless of construct size, 3R and 4R tau coassemble into heterogeneous filaments with a core of parallel, in-register arranged ␤-strands.

DISCUSSION
Tau filaments are the pathological hallmark of numerous tauopathies. Although in most diseases, filaments are composed of either 3R or 4R tau, in AD, filaments contain all isoforms. Although the isoform composition of tau filaments had been determined nearly 20 years ago (21,22), until now, it was not clear whether the 3R and 4R isoforms in AD might form homogeneous or heterogeneous filaments (45). Such information, however, is important as either aggregate would imply fundamentally different mechanisms of fibrillization. In this study, we set out to obtain molecular insights into the structure of different tau filaments and to address the question of single-filament composition. First, we demonstrated that homogeneous filaments of both 3R and 4R tau share a highly ordered core structure that coincides with the third repeat and is characterized by parallel in-register arrangement of ␤-strands. This conformation is preserved regardless of whether full-length isoforms (htau40 and htau23) or smaller constructs (K18 and K19) are used. Second, we unequivocally showed that when present in the same sample, 3R and 4R tau coassemble into heterogeneous filaments. This coassembly occurs for both truncated constructs and full-length isoforms. Because heterogeneous filaments contain the same ordered third repeat core structure as their homogeneous counterparts, the third repeats of 3R and 4R tau must be able to replace each other. Together, these results suggest that at least three fundamentally different types of tau filaments exist: 1) filaments of 3R tau, 2) filaments of 4R tau, and 3) filaments of 3R/4R tau (Fig. 6). This classification is based on composition and places emphasis on the core-forming repeat regions.
Each 3R and 4R tau can represent three different isoforms that are defined by the inclusion or exclusion of the two inserts in the N-terminal region (Fig. 1A). This region is in the nonstructured periphery of the filament and may not contribute to the stable core (13). However, it is evident that each filament type can be further classified into subtypes composed of different isoforms at various ratios. Our dilution experiments have shown that 3R and 4R tau coassembled at all ratios tested. As a consequence, a staggering number of filaments exist based on composition alone. Despite a shared parallel, in-register arrangement of ␤-strands, all three types of filaments must vary in conformation. This is because the second repeat of 4R tau contributes to the filament core (34,46). Its presence or absence will result in different packing interactions for each filament type. In heterogeneous 3R/4R filaments, for example, the second repeat cannot solely pack against itself. Hence, 4R tau will assume different conformations depending on whether it is incorporated into homogeneous or heterogeneous filaments. Conformational plasticity of 4R tau was previously demonstrated as htau40 assumed different structures when grown onto wild type filaments as opposed to mutant filaments (47). Conformational plasticity is a common property of amyloid proteins resulting in structurally distinct fibrils of homogeneously composed proteins (see for example Refs. 48 -52). It is likely that similar structural polymorphisms exist for each type of tau filament. Compositional diversity and conformational plasticity allow for a plethora of tau filament structures that may exceed those of other amyloid fibrils.
Ample experimental evidence suggests that phenotypic diversity in prion diseases is encoded in different misfolded conformations of the same protein (48,(53)(54)(55)(56). The particular fragmentation and growth properties of different fibrils have provided a physical basis of how protein conformation could determine phenotype (51,57). As pointed out above, the filaments in different tauopathies have characteristic compositions of 3R and 4R tau. Filaments in Pick disease contain 3R tau. Filaments in progressive supranuclear palsy and corticobasal degeneration contain 4R tau. Filaments in AD incorporate both 3R and 4R tau. Because we have demonstrated that 3R and 4R tau coassemble into heterogeneous filaments, filaments in AD could be distinct from those in the 3R and 4R tauopathies. It was previously shown that filaments in AD contain roughly equimolar ratios of 3R and 4R tau (21). This ratio reflects the expression pattern of tau in the adult human brain (7). Hence, the majority of individual filaments in AD brain may be composed of evenly mixed 3R and 4R isoforms. However, because the expression patterns of tau isoforms can vary between different neuronal subpopulations (6), the specific composition of individual tau filaments could vary in a cell-specific manner. The free interchangeability of 3R and 4R tau would allow for a broad spectrum of heterogeneous filaments with diverse physical and biological properties.
Although the repeat region in tau may determine the stability of filaments (58), the N-terminal inserts could influence the interactions with cellular components. Here, we demonstrate that 3R and 4R tau can coassemble into heterogeneous filaments, expanding the number of possible filament structures. If future experiments corroborate interneuronal spreading of tau misfolding in a prion-like fashion (59), the large compositional and conformational diversity of tau filaments could be an important contributor to the large phenotypic diversity of tauopathies.