The “Jaws” of the Tau-Microtubule Interaction*

Tau is the major microtubule-associated protein in neuronal axons. It aggregates into “neurofibrillary tangles” during the course of Alzheimer disease. Binding to microtubules and microtubule assembly requires the “repeat domain” in the C-terminal half of Tau, as well as the two regions flanking the repeats. Here we report the NMR characterization of a 198-residue Tau fragment composed of the four tandem repeats and the flanking domains and containing the full microtubule binding and assembly activity of Tau. NMR secondary chemical shifts and dipolar couplings detect the highest propensity for β-structure within the four-repeat region, whereas the flanking domains are largely random coil, with an increased rigidity in the proline-rich region. Chemical shift perturbation experiments identify two motifs in the upstream flanking domain, 225KVAVVRT231 and 243LQTA246, and one downstream of the repeats, 370KIETHKTFREN380, which strongly contribute to the binding to the acidic outside of microtubules, as well as to the binding of other polyanions such as heparin. This is consistent with the “jaws” model of Tau-microtubule interactions and highlights the importance of the regions flanking the repeats for both microtubule binding and pathological Tau aggregation.

Tau, the major MAP 2 of neurons, is localized predominantly in axons. Tau stabilizes MTs and is necessary for neurite outgrowth (1,2). Tau occurs in six main isoforms in the human brain (ranging between 352 and 441 amino acid residues), obtained by alternative splicing, and differs by having 3 or 4 semi-conserved repeats of ϳ31 residues in the MT assembly domain and 0 -2 insertions in the N-terminal projection domain ( Fig. 1) (3,4). The repeat domain is important for pro-moting microtubule assembly, but it binds to microtubules only with low affinity. However, strong MT binding and efficient assembly is achieved when the repeat domain is combined with the flanking domains, whereas the flanking domains alone bind to microtubules but do not promote MT assembly. This leads to the proposition of the "jaws" model of Tau whereby the flanking regions are considered as targeting domains, responsible for positioning Tau on the MT surface, and the repeats are considered as catalytic domains for MT assembly (5,6).
In Alzheimer disease, Tau aggregates into neurofibrillary tangles, which consist of "paired helical filaments" (PHFs) (7,8). The repeat domain coincides with the core of PHFs (9 -11) and promotes PHF assembly in vitro (12). For PHF aggregation, two hexapeptides at the beginning of the second and third repeats ( 275 VQIINK 280 and 306 VQIVYK 311 ) are crucial (13,14). The role of the repeat domain both in MT assembly and PHF aggregation illustrates a remarkable overlap between physiological and pathological functions of Tau. Both functions are regulated by phosphorylation, whereby the phosphorylation of the repeat domain (KXGS motifs) efficiently detaches Tau from microtubules and prevents PHF aggregation, whereas phosphorylation in the flanking domains has a modulating effect (15).
To gain a structural understanding of the MT binding and PHF aggregation activities, we performed a detailed NMR characterization of a 198-residue fragment (construct K32), containing the repeat domain and the two flanking domains. Tau is a natively unfolded protein so that x-ray crystallography cannot be applied, leaving NMR spectroscopy as almost the only method that allows detailed structure information. The results presented here provide a view of the jaws in Tau that are essential for efficient binding of Tau to microtubules.

Expression and Isotopic Labeling of Recombinant Tau
Constructs-Human Tau constructs were expressed in the vector pNG2 (a derivative of pET-3a; Merck) in Escherichia coli strain BL21(DE3) as described (5) (see Fig. 1). K32 comprises residues (Met)Ser 198 -Tyr 394 , equivalent to the four repeats and the two flanking regions (198 residues). The expressed proteins were purified from bacterial extracts by making use of the heat stability of the protein and by FPLC SP-Sepharose chromatography (Amersham Biosciences). To label the Tau proteins with 15 N-and 13 C-stable isotopes, the E. coli culture expressing the K32 protein was grown on rich growth medium based on chemolithoautotrophic bacteria labeled with 13 C and 15 N isotopes (Silantes, Munich, Germany). Protein samples uniformly enriched in 15 N were prepared by growing E. coli bacteria in minimal medium containing 1g liter Ϫ1 of 15 NH 4 Cl.
The cell pellets were resuspended in boiling extraction buffer (50 mM MES, 500 mM NaCl, 1 mM MgCl 2 , 1 mM EGTA, 5 mM DTT, pH 6.8) complemented with a protease inhibitor mixture. The cells were disrupted with a French pressure cell and subsequently boiled for 20 min. The soluble extract was isolated by centrifugation, and the supernatant was dialyzed against two changes of cation exchange chromatography buffer A (20 mM MES, 50 mM NaCl, 1 mM EGTA, 1 mM MgCl 2 , 2 mM DTT, 0.1 mM phenylmethylsulfonyl fluoride, pH 6.8) and loaded on an FPLC SP-Sepharose column. The proteins were eluted by a linear gradient of cation exchange chromatography buffer B (20 mM MES, 1 M NaCl, 1 mM EGTA, 1 mM MgCl 2 , 2 mM DTT, 0.1 mM phenylmethylsulfonyl fluoride, pH 6.8). In the second purification step, the K32 breakdown products were removed by applying the gel filtration column Superdex G-75 with PBS buffer (137 mM NaCl, 3 mM KCl, 10 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , pH 7.4, with 1 mM DTT). NMR samples contained 0.9 to 1.5 mM 15 N-or 15 N/ 13 C-labeled protein in 95% H 2 O, 5% D 2 O, 50 mM phosphate buffer, pH 6.8, with 1 mM DTT.
Microtubule Preparation-Porcine brain tubulin was purified as described (5,16) and incubated at concentrations higher than 200 M in microtubule assembly buffer (100 mM Pipes, pH 6.9, 1 mM EDTA, 1 mM MgSO 4 , 1 mM DTT) in the presence of 1 mM GTP at 37°C for 5 min. After addition of 100 M paclitaxel (Sigma), the polymerization was performed for 20 min at 37°C. The integrity of microtubules was checked by standard negative stain electron microscopy.
NMR Spectroscopy-NMR spectra were acquired at 5°C on a Bruker Avance 900 spectrometer equipped with a cryo-probe. Aggregation did not occur under these low temperature conditions. Three-dimensional (HA)CANNH and HNN experiments (4 scans, 1.2-s recovery delay) were collected to obtain sequence-specific assignments for the backbone of K32. NMR data were processed and analyzed using NMRPipe (17) and Sparky 3 (University of California, San Francisco). Secondary shift values were calculated as the differences between measured C ␣ or CЈ (as obtained from a three-dimensional HNCO experiment) chemical shifts and the empirical random coil value for the appropriate amino acid type (18,19). Using the program NMRView version 5 (20), random coil values for the primary sequence of K32 were generated. Random coil values for histidines, glutamates, and aspartates were taken from Wishart and Sykes (21), as the chemical shifts of these residues are particularly sensitive to pH and the pH in the studies by Wishart and Sykes (22) (pH 5) is more similar to the one used here (pH 6.9). Furthermore, correction factors for residues occurring in the sequence before a proline residue were also taken from Wishart and Sykes (21). To estimate the ␤-sheet propensity in contiguous segments of Tau, the observed C ␣ and CЈ chemical shifts were normalized by the empirically determined secondary shift expected for that residue type in a fully ␤-sheet conformation (22), summed and normalized by the number of residues in the segment (23). To probe the structural consequences of binding of heparin to K32, C ␣ chemical shifts were measured using a three-dimensional (HA)CANNH with a K32:heparin ratio of 4:1.
Titration of K32 with heparin was carried out with uniformly 15 N-labeled protein containing 0.15 mM K32, pH 6.9 (100 mM Pipes, 1 mM EDTA, 1 mM MgSO 4 , 1 mM dithiothreitol). Heparin (average molecular weight 3350, 5.8 dissaccharide subunits, charge/subunit Ϫ2.5) was from Sigma. Complex formation was monitored by recording two-dimensional 1 H- 15  One-bond N-H RDCs (D NH ) were determined by using the two-dimensional inphase-antiphase (IPAP)-HSQC sequence (24) in 300 mM sodium chloride. D NH values were calculated as the difference between splittings measured in the isotropic phase and in a sample, in which K32 had been aligned in 8 mg/ml Pf1 bacteriophage (Asla, Riga, Latvia) (deuterium splitting 4.4 Hz). RDCs were not corrected for the negative gyromagnetic ratio of 15 N. 15 N R 2 and R 1 relaxation rates were measured with pulse sequences based on Farrow et al. (25) at 5°C on a Bruker Avance 700-MHz spectrometer. For T 1 , where the expected range of values is small, only two relaxation periods with 20 and 320 ms were used (26). The spin lock frequency was 1.5 kHz. For R 2 measurements, the relaxation delays were set to 20, 100, 200, and 260 ms. Relaxation times were calculated from the ratio of the signal integrals (R 1 ) or by fitting a exponential function to the decaying signal integrals (R 2 ). Heteronuclear NOE values are reported as the ratio of peak heights in paired spectra collected with and without an initial period (4 s) of proton saturation during the 5-s recycle delay. Errors were calculated on the basis of the S/N ratio for each residue.

RESULTS
Backbone Assignment of K32-NMR resonances in 1 H, 15 N HSQC spectra of the construct K32, consisting of all four repeat motifs (R1-R4) of Tau and the two flanking regions P2 and RЈ ( Fig. 1), were recorded at 5°C and pH 6.9. The resonances were sharp and showed only a limited dispersion of chemical shifts (supplemental Fig. S1), reflecting a high degree of backbone mobility and the lack of well defined secondary and tertiary structure, in agreement with circular dichroism and Fourier transform-infrared spectroscopy measurements.
To enable the study of the structure and dynamics of Tau, the assignment of the NMR resonances was required. K32 includes 198 residues with a calculated mass of 21,029 Da. Assignment of an intrinsically disordered protein of this size is not a trivial task because of the very limited chemical shift dispersion. Because of the strongly repetitive primary sequence (five types of amino acids make up more than 50% of the protein), resonances from residues located in the repeat domain were particularly affected by overlap. Moreover, the presence of 11 prolines, out of a total of 46 residues (almost 25%) in the basic region P2, further complicated the assignment. These complications could be overcome by the use of high resolution three-dimensional (HA)CANNH and three-dimensional HNN experiments (supplemental Fig. S2), in combination with the previously determined assignment of the isolated four repeats (on referred to below as construct K18) (27). No resonances for Ser 198 and Ser 199 were detectable. This could be due to enzymatic cleavage of these N-terminal residues or line broadening caused by the conformational flexibility of the N terminus. In addition, residues Gly 272 , Gly 303 , Gly 334 , and Gly 366 (surrounded by two glycines in the sequence motif PGGG, which occurs at the C-terminal end of each repeat region) and Gly 304 /Gly 335 could be identified in the spectra, but their unambiguous assignment was not possible because of high resonance overlap. In total, about 95% of K32 residues were assigned unambiguously (prolines excluded). The achieved assignment constitutes ϳ45% of the sequence of hTau40, the longest Tau isoform in the human brain (441 residues). Comparison with the assignment of K18 showed that the positions of peaks belonging to the four repeats remained largely unaffected by the presence of additional flanking regions (data not shown). Moreover, with the exception of the 10 N-terminal residues of K18, signal intensities within the repeat domain were not significantly altered. This underlines the character of Tau as an intrinsically disordered protein.
Secondary Structure Propensity in K32-NMR chemical shifts, in particular of C ␣ atoms, are very sensitive probes of secondary structure both in globular and intrinsically disordered proteins (22,28). These shifts show small but distinct deviations from random coil values for K32 (Fig. 2). Several continuous stretches (containing 7-11 residues) with negative secondary chemical shifts were observed, in particular Lys 274 -Leu 284 (R1/R2), Ser 305 -Asp 315 (R2/R3), and Gln 336 -Asp 345 (R3/R4). Continuous stretches of negative secondary chemical shifts indicate a propensity for ␤-structure. According to a quantitative analysis of the C ␣ secondary chemical shifts, the ␤-structure-like conformations are populated 22, 25, and 19% of the time for residues Gly 273 -Leu 284 , Gly 304 -Leu 315 , and Gly 333 -Asp 345 , respectively. These elements of ␤-structure are located at the beginning of R2, R3, and R4. The corresponding region in repeat 1 does not show a strong propensity for ␤-structure, most likely because of the presence of proline residues Pro 247 and Pro 251 . Residual ␤-structure was also observed for residues Val 256 -Ser 262 and Gln 351 -Leu 357 , which are located roughly in the center of repeats R1 and R3. In the RЈ, the region downstream of the repeat motifs, which shows only a weak sequence homology to the repeats, no strong tendency to populate either helical or ␤-structure was detected. On the other hand, quite large negative secondary chemical shifts were present in the proline-rich region P2, with Pro 223 -Arg 230 forming a continuous stretch. The pattern of CЈ secondary chemical shifts was very similar to that observed for C ␣ (supplemental Fig. S3) indicating that either C ␣ or CЈ chemical shifts can be used for detection of transient structure in unfolded proteins. Comparison of K32 with K18, i.e. the construct in which the two flanking domains are missing, shows an almost identical pattern of secondary chemical shifts for residues Ala 246 -Ile 371 demonstrating that the flanking regions do not significantly FIGURE 1. Bar diagrams of full-length Tau and the fragment K32. Full-length adult Tau (hTau40) contains 441 residues and can be subdivided into several domains (5). The N-terminal "projection domain" comprises residues 1-197 and does not contribute to MT binding. The C-terminal "assembly domain" contains the residues important for microtubule nucleation and stabilization. The four repeats (31 or 32 residues) are denoted as R1-R4. R2 (encoded by exon 10) is absent in three-repeat isoforms because of alternative splicing. The basic and proline-rich region preceding the repeats is subdivided into P1 and P2, separated by the chymotryptic cleavage site behind Tyr 197 , which divides the projection domain and assembly domain. Fragment K32 comprises residues Ser 198 -Tyr 394 with a preceding methionine (198 residues), i.e. domains P2, R1-R4, and RЈ. P2 contributes a major part of the microtubule-binding affinity and contains many of the proline-directed phosphorylation sites (e.g. AT8 site ϭ S202 p ϩ T205 p , AT100 site ϭ T210 p ϩ S212 p , which represents the binding site of the Pin1 proline isomerase (59), and the AT180 site ϭ S231 p ϩT235 p , which contributes to a conformational change upon proline isomerization). The C-terminal flanking region is denoted RЈ because it shows weak sequence homology to the other repeats ("5th repeat"). The two hexapeptide motifs 275 VQIINK 280 and 306 VQIVYK 311 in R2 and R3, which are necessary for PHF formation (13), are indicated by black rectangles. affect the secondary structure propensity in the repeat domain of Tau.
To obtain further insight into the secondary structure propensity, we measured one-bond N-H residual dipolar couplings (RDCs) in K32, which had been partially oriented in a Pf1 bacteriophage alignment medium. In an anisotropic environment, like magnetically oriented Pf1 phage, the large one-bond internuclear dipolar interactions no longer average to zero. In contrast to the bell-like smooth distribution of dipolar couplings, which is expected for random flight chains (29), a very specific distribution of positive and negative couplings was observed for K32 (Fig. 2B). In the region R1-R4 two stretches of high positive RDC values were evident, 274 KVQIINKKLDL 284 and 306 VQIVYKPVD 314 . These regions correlate well with the elements of high ␤-structure propensity at the beginning of R2 and R3, as identified by NMR secondary chemical shifts (Fig. 2). Note that the ␤-regions at the beginning of R2 and R3 encompass the two PHF-promoting hexapeptide motifs (13), whereas the ␤-region in R4 overlaps with the "module B" motif that determines the packing of Tau into twisted PHFs (30). The regions with large RDCs end with residues displaying negative RDCs: Lys 254 , Ser 285 , Asn 286 , Ser 316 , Lys 317 , Gly 326 , Ile 328 , Lys 347 , Gly 355 , Arg 379 , Glu 380 and Gly 389 . The same residues show positive secondary chemical shifts, and we previously proposed that they are involved in turn structures (27). In the proline-rich region P2, particularly large and positive RDC values were found. The RDC magnitude for residues Ser 214 -Ser 235 , a region in which eight prolines are located, is about twice the size than in the rest of the protein.
In contrast to RDCs, which are influenced by both structure and dynamics, 15 N spin relaxation times can directly probe the motion of the protein backbone in a site-specific manner (31). 15 N R 1 and R 2 spin relaxation rates are sensitive to motions of the backbone occurring on the pico-to nanosecond as well as micro-to millisecond time scale. The 15 N on-resonance R 1 rates observed in K32 at 5°C follow a smooth bell-like distribution with an average value of 3.7 Hz, underlining the character of Tau as a natively unfolded protein (Fig. 6). At the termini the relaxation rates are reduced, in agreement with an increased flexibility. Some residues such as Ile 308 -Lys 317 show slightly higher R 1 values, reporting on motional restrictions in these regions. On the other hand, heteronuclear steady-state 15 N NOE values were all close to zero (supplemental Fig. S4), indicating that abundant pico-to nanosecond dynamics occur along the polypeptide chain of K32, and fully formed regular secondary structure elements are not present. In contrast to R 1 rates, R 2 spin relaxation times are also affected by chemical exchange. Thus, a difference plot between R 2 and R 1 rates identifies residues that exchange between multiple conformations in solution. In K32, above average chemical exchange contributions are most evident for residues Gly 201 -Ser 210 , Ser 237 -Ser 241 , Gly 271 , Lys 290 -Gly 292 , Lys 298 , Gly 302 , Val 309 -Lys 321 , Gly 323 , Gly 333 , Asn 368 , and Phe 378 .
Interaction of Tau-K32 with the PHF Inducer Heparin-NMR signals of backbone amides constitute excellent probes of complex formation (32), providing maps of interaction interfaces. We were particularly interested in the interaction of Tau with its physiological target microtubules, and with the poly-anion heparin that induces the formation of Alzheimer-like PHFs. The binding of heparin to K32 was monitored by twodimensional 1 H-15 N HSQC spectra. This allows one to identify the residues that are important for the binding to polyanions. At high heparin concentrations chemical shift changes of some residues were accompanied by a decrease of the intensity of their signals, indicating chemical shift exchange intermediate on the NMR time scale. The chemical shift perturbation experiments showed the strongest shifts in the following regions (Fig.  3A): Thr 217 -Leu 253 (most pronounced for 243 LQTA 246 ), 275 VQIINKKLDLSN 286 , 298 KHV 300 , 372 ETHKLTFRENAK 383 . The chemical shift differences correlate very well with the signal intensity ratios (residue intensity of spectrum with ligand: residue intensity of the unperturbed spectrum of unbound protein) of these spots (Fig. 3B). For strongly shifting residues the intensity ratios were generally below 70% of their original values. In the regions around Lys 225 , Thr 244 , Asn 279 , and Thr 373 , the signal intensity was reduced even down to about 40%, indicating that these residues are strongly involved in exchange processes intermediate on the NMR time scale. Besides the binding sites deduced from the chemical shift changes, 212 TPSLP 216 , 306 VQIVYKPV 313 , and 325 LGNIHHK 331 were identified as being important for the heparin binding by exchange broadening (Fig. 3B). In most regions of K32 strongly affected by heparin, positively charged residues are present as follows: Arg 211 , Arg 242 , Lys 280 , Lys 281 , Lys 298 , His 299 , Lys 311 , His 329 , His 330 , Lys 331 , His 374 , Lys 375 , and Arg 379 . This indicates that the binding to heparin is of electrostatic nature. Because of the presence of two lysine residues and one arginine, chargecharge interactions might also occur with the stretch 224 KKVAVVR 230 , but in this case a hydrophobic contribution is possible as well.
To probe the structural impact of binding of heparin to Tau, we determined the C ␣ secondary chemical shifts of Tau in presence of heparin (Tau:heparin 4:1). NMR chemical shifts can be measured with high accuracy such that even minute changes in secondary structure can be detected. However, the C ␣ chemical shifts of K32 with and without heparin were virtually identical (supplemental Fig. S5), indicating that heparin binding does not induce a significant increase in ␤-structure under these conditions.
Tau-Microtubule Interactions and the Jaws Model-The binding of K32 to MTs was characterized using NMR chemical shift perturbation as described (27), where two-dimensional 1 H-15 N HSQC spectra of K32 were recorded in the presence of increasing amounts of taxol-stabilized MTs. The influence of temperature was probed by performing NMR titrations at 5 and 20°C. As shown previously, taxol-stabilized microtubules are stable at 5°C for several hours, sufficient for the time course of the NMR experiment, and Tau binds to them (27). The overall signal intensity in two-dimensional 1 H-15 N HSQC experiments decreases rapidly at high MT concentrations, indicative of an increasing amount of K32 strongly bound to MTs. Because of very fast relaxation, the NMR signals of MT-bound K32 are no longer observable. Nevertheless, continuous chemical shift changes and additional exchange-induced broadening of specific regions of K32 were observed with increasing MT concentrations. It is immediately apparent from Fig. 4 that both the P2-and the RЈ-flanking regions, as well as the repeat domain itself, strongly contribute to the interaction with microtubules. The strongest chemical shift changes, pointing to a strong involvement into the MT-binding process, were observed for residues 225 KVAVVRT 231 in domain P2, 240 KSRLQTAPV 248 in domain R1, 275 VQIINKKLDLS 285 and 297 IKHV 300 in R2, and 370 KIETHK 375 in RЈ. The strong contribution of the flanking domains is consistent with their role as microtubule-targeting domains (jaws). In addition, however, other isolated residues were strongly perturbed in the presence of MTs. The location of the jaws of MT binding became clearer, when the reduction in 1 H-15 N HSQC signal intensities caused by addition of MTs was taken into account (Fig. 4B). For most of the residues from Leu 215 to Lys 331 , the intensity was reduced to below 40% at a K32:MT ratio of ϳ2:1. Particularly low signal intensities were observed for residues 222 EPKKVAVVRT 231 , 275 VQIINKKLDL-SNVQSK 290 , 306 VQIVYKPVDLSKVTSK 321 , and 377 TFREN 381 . Both signal intensity ratios and chemical shift changes indicate that repeat R4 and the outermost residues in K32, Gly 200 -Ala 209 and Thr 386 -Tyr 394 , do not significantly bind to MTs.
At 20°C, the overall magnitude of chemical shift changes induced by the presence of MTs was decreased (Fig. 4A). In particular, the chemical shift changes of residues 240 KSR-LQTAPV 248 , 275 VQIINKKLDLS 285 , and 370 KIETHK 375 were less pronounced. In comparison to the results at 5°C, large chemical shift changes remained for 225 KVAVVRT 231 and 297 IKHV 300 . Signal intensity ratios between the free and the MT-bound state of K32 were less informative than at 5°C. Overall, the reduction in signal intensity in the MT-bound state was lower than at 5°C, and signal intensities were more uniform along the whole K32 sequence (data not shown).
The binding of Tau to MTs occurs in two structurally and kinetically distinct steps, comprising a first binding phase with low stoichiometry but tight binding (K d ϳ0.1 M) and a second phase of weaker affinity but higher and nonsaturable stoichiometry (33). Solution NMR spectroscopy cannot detect Tau molecules that are fully bound to MTs because of the large molecular weight of the complex and the accompanying fast signal decay. To test if the binding sites identified by the NMR titrations of K32 with MT are characteristic also for the strong primary binding of Tau to MTs, we repeated the K32-MT titration at increased ionic strengths. Previous studies have shown that in the presence of sodium chloride, the MT-binding strength of Tau is reduced and the weaker, nonsaturable binding is strongly attenuated (34). In agreement with these results, the addition of 100 mM NaCl to the Tau-MT solution (K32: MT ϭ 2:1) increased the average ratio of signal intensities to about 80% (data not shown). A further increase to 300 and 600 mM NaCl raised the average signal intensity ratio to a maximum value of 85%. This indicates that at 300 mM NaCl, the off-rate is sufficiently increased, such that most K32 molecules now contribute to the NMR signal. Fig. 5A shows the chemical shift changes induced in K32 upon addition of MTs at 5°C in the presence of 300 mM NaCl. The same residues that show large chemical shift changes without NaCl were also strongly affected at 300 mM NaCl. Because of the higher ionic strength, the interactions with the positively charged binding hot spots of K32 were weakened. The most strongly shifting resonances were now located in the stretch 225 KVAVVRT 231 in the upstream flanking domain, suggesting that there is a hydropho-bic MT-binding contribution originating from this region. In addition, signals from residues 337 VEVK 340 and 350 VQSKI 354 were more strongly affected than in the absence of sodium chloride. Taken together, the high salt titration of K32 with MTs demonstrates that the binding profile identified by our NMR studies reports on the primary, strong binding of Tau to microtubules. Note that at 20°C and in the presence of 300 mM NaCl no MT binding was detected anymore, as judged by NMR chemical shifts and signal intensity ratios (data not shown).

DISCUSSION
Tau is an interesting protein in several respects. As an intrinsically disordered protein, it challenges the paradigm that a folded structure is necessary for function. Tau is a major microtubule-associated protein involved in the formation of axons and stabilization of microtubules as tracks for axonal transport. In the form of paired helical filaments, Tau constitutes one of the hallmarks of Alzheimer disease. Therefore, key questions concerning the structure and function of Tau protein include the following. What are the binding sites of Tau to microtubules? How is the aggregation of Tau into PHFs facilitated? To address some of these questions we investigated a 198-residue fragment (K32), comprising the four repeat motifs and the two flanking basic domains P2 and RЈ (Fig. 1). The repeat domain forms the core of the microtubule-binding domain and also the core of PHFs (3,9). However, maximal binding of Tau to microtubules requires the two flanking regions P2 and RЈ. This was interpreted in terms of jaws of the Tau-microtubule interaction (6).
Residual ␤-Structure in Repeat Domain and Random Coil Structure in Flanking Domains-Backbone resonance assignment of K32 is challenging, because of the small chemical shift dispersion, the repetitive nature of the repeat domain, and the large number of prolines in domain P2. We were able to assign 95% of non-proline residues of K32, equivalent to 40% of all residues of full-length Tau. This extends and is complementary to recent work of Lippens and co-workers (36) who reported a partial assignment of Tau. In this study, the unambiguously assigned residues (ϳ30 -40%) were mainly located in the acidic N-terminal header domain (residues 1-120) and in the C-terminal tail domain (residues 390 -441), whereas only a few residues in the K32 region could be assigned unambiguously. Thus, our assignment of K32 was essential for the investigations reported here and will be useful for future studies of Tau, such as the structural consequences of phosphorylation or the influence of FTDP-17 mutations.  Assignment of backbone chemical shifts in proteins gives access to detailed information on secondary structure. The difference between experimental chemical shifts and random coil values, which are specific for the sequence of the protein under investigation, allows site-specific detection of even small propensities for helical or ␤-structure. In K32, the initial stretches of R2, R3, and R4 have the most pronounced ␤-sheet propensity, with average values of 22, 24, and 19% for residues Gly 273 -Leu 284 , Gly 304 -Leu 315 , and Gly 333 -Asp 345 , respectively. This agrees well with other biochemical and structural studies showing that these regions are involved in the assembly of PHFs and their molecular packing (13,30). In addition, residues Lys 224 -Arg 230 , Val 256 -Ser 262 , Gln 351 -Leu 357 , and Val 363 -Glu 372 preferentially populate extended structures ( Fig. 2 and Fig. 7). No ␣-helical structure was detected in the repeat domain in this study. This is contrary to other NMR studies of Tau-derived fragments that were performed, however, in the presence of helix-inducing agents such as trifluoroethanol or SDS (37)(38)(39). In addition, no strong tendency to populate ␣-helical or ␤-structure was detected in the regions flanking the repeat domain. In the N-terminal P2 domain (Ser 198 -Leu 243 ), this is most likely because of the large number of proline residues. This feature, combined with the presence of many residues with comparably large negative secondary chemical shifts ( Fig.  2A), hints to a tendency for formation of a polyproline helix in this region. Indeed, phosphorylation at residues Thr 231 , Ser 235 , and Ser 237 seems to induce polyproline helix conformations for residues Val 229 -Ser 238 (40).
Residual dipolar couplings can be measured in a weakly aligned macromolecule, for which the large one-bond internuclear dipolar interactions no longer average to zero (41,42). RDCs are exquisitely sensitive to bond vector orientation and offer many new opportunities in NMR studies of proteins in solution. In highly dynamic systems, such as intrinsically disordered proteins, RDCs define local conformational behavior as well as long range order (43). In the repeat domain of Tau, large RDCs correlate with regions that preferentially populate ␤-structure as indicated by NMR chemical shifts. In particular, in the two hexapeptide motifs at the beginning of repeats R2 and R3, which are crucial for aggregation of Tau into PHFs, RDCs are about twice the size than in the rest of the repeat domain (Fig. 2B). The C-terminal ends of these regions are characterized by positive secondary chemical shifts and negative RDCs. This suggests the formation of turns at these sites. The largest positive RDCs were observed, however, for residues Ser 214 -Ser 235 in the center of the proline-rich domain P2. For the analogous case of ␣-synuclein (the protein that irreversibly aggregates in the course of Parkinson disease), we have shown previously that large RDCs in the C-terminal domain were because of an increased rigidity of the protein backbone, which was induced by long range interactions (44). Thus, long range interactions within K32 might be responsible for the increased magnitude of RDCs in the P2 domain. Alternatively, the increased RDC magnitude of Ser 214 -Ser 235 might be due to the large number of prolines in this region, which make the backbone more rigid. Comparison of chemical shifts and signal intensity ratios between K32 and K18 (the construct lacking the flanking P2 and RЈ domains), steady-state NOE values, and 15 N transverse relaxation rates do not provide any evidence for a folding back of P2 or RЈ on the repeat region. This, however, does not exclude very transient interactions. To probe for transient long range interactions, several single cysteine mutants of K32 would be required, to which a paramagnetic nitroxide radical can be specifically attached. This work is in progress. Noteworthy, the strongly increased RDC magnitude of Ser 214 -Ser 235 is not accompanied by above average 15 N R 1 rates or steady-state NOE values, indicating that motions in the nanoto microsecond time scale (toward which R 1 rates are not sensitive) are reduced in this region.
To probe the intrinsic dynamics of Tau, we measured 15 N transverse spin relaxation rates. In particular, the glycine residues Gly 201 , Gly 271 , Gly 292 , Gly 302 , Gly 323 , and Gly 333 were strongly affected by chemical exchange (Fig. 6), suggesting that they function as flexible linkers between more rigid parts of the molecule. Gly 271 , Gly 302 , and Gly 333 are part of the PGGG motifs just upstream of the hexapeptides in the repeats. On the basis of secondary chemical shifts, we previously suggested that these residues form turns. In addition, pronounced chemical exchange was observed for the stretch Val 309 -Lys 321 , which encompasses the hexapeptide motif of R3. This region has the strongest propensity to form ␤-structure according to secondary chemical shifts, suggesting that the hexapeptide motif of R3 exchanges between a random coil conformation and ␤-structure in solution.
In the fibrillar state, the repeat domain of Tau assumes the characteristic cross-␤-structure with a high content of ␤-strands (45)(46)(47). By NMR secondary chemical shifts, we could clearly identify several regions in K32, primarily located in the repeat domain, with a pronounced tendency to populate ␤-structure. Therefore, it is tempting to speculate that these regions serve as seeds of aggregation. Dimer formation as the initial step of aggregation might be initiated by intermolecular contacts between the regions of high ␤-structure propensity. The intermolecular interaction will stabilize the ␤-structure and ultimately convert these parts into ␤-sheets in PHFs. This idea is supported by the prediction of aggregation hot spots using the program Zyggregator (48). It predicts three hot spots of aggregation for the Tau peptide Asn 265 -Glu 338 as follows: residues Lys 280 -Val 287 , Val 305 -Lys 311 , and Gly 335 -Lys 340 . In addition, based on experimental aggregation propensities derived from Tau-related peptides, the two hexapeptides in the beginning of repeats R2 and R3 ( 275 VQI-INK 280 and 306 VQIVYK 311 ), 226 VAVVR 230 , and 392 IVYK 395 are predicted to have high aggregation propensities (49). These stretches correlate well with the regions of experimentally determined ␤-structure propensity.
Regions Flanking the Repeat Domain Are Involved in Binding to PHF-inducing Polyanions-Because Tau is a highly soluble protein, it normally resists aggregation; however, aggregation can be enhanced by different polyanions, presumably because they overcome the charge repulsion of the cationic Tau molecules. Examples are sulfated glycosaminoglycans, heparin, RNA, and acidic peptides resembling the C-terminal domain of tubulin (50 -52). It was therefore of interest to determine which residues in Tau are involved in the interaction with microtubules and polyanions such as heparin.
Previously, we have shown that heparin binds to the hexapeptide motifs in the repeat domain, as well as to positively charged patches such as the KK motif at the beginning of R2 ( 280 KK 281 ) and the KH ( 267 KH 268 and 298 KH 299 ) or HHK ( 329 HHK 331 ) motifs near the end of the repeats R1, R2, and R3. These studies were performed on K18, a 130-residue fragment of Tau that just comprised the repeat domain (27). These studies show that heparin binds to the same positively charged patches in the repeat domain, even when the two flanking regions P2 and RЈ are present. In particular, the fact that the hexapeptide VQIINK in the beginning of repeat 2, which is important for aggregation of Tau into PHFs, preferentially populates ␤-structure and binds to the polyanion heparin also in K32 underlines the hypothesis that polyanions enhance aggregation of Tau by reducing electrostatic repulsion between different Tau molecules. In contrast to a recent study (53), we could not detect any increase in ␤-structure in Tau upon addition of heparin. Our measurements were performed with a 4:1 ratio of Tau:heparin, which was described as optimum for induction of Tau aggregation by heparin (54). Besides a different Tau:heparin ratio, Sibille et al. (53) used full-length Tau and a specific low molecular weight version of heparin.
Here we show that both flanking domains, P2 and RЈ, are also involved in the interaction with heparin (Fig. 3). Most notably, Lys 224 -Arg 230 , the only region outside of the repeat domain for which NMR secondary chemical shifts indicated a propensity to form ␤-structure, binds to heparin. Similar to the hot spot of aggregation in the beginning of repeat 2 ( 274 KVQINNK 280 ), the 224 KKVAVVR 230 sequence contains two adjacent lysine residues separated from a third positively charged residue (Arg 230 ) by several uncharged amino acids. Therefore, the mechanism of interaction with heparin might be similar, and regions outside of the repeat domain may play a role in aggregation of Tau into PHFs.
The Jaws of the Tau-Microtubule Interaction-The microtubule-associated protein Tau is a critical regulator of diverse microtubule functions (5,55,56). The repeat domain with its four repeats is essential for microtubule assembly; however, in the absence of the two flanking regions P2 and RЈ, the repeat domain binds only weakly to microtubules. The flanking domains, on the other hand, bind to microtubules even in the absence of the repeats. This has led to the proposition of the jaws model of Tau whereby the regions flanking the repeats are considered as targeting domains, responsible for positioning Tau on the microtubule surface, and the repeats that act as catalytic domains for microtubule assembly (5,6). The aim of this study was to provide a structural view of the residues located in the flanking regions, which are important for binding of Tau to microtubules. The 198-residue fragment K32, which consists of the repeats and the flanking domains P2 and RЈ, is well suited for this aim, as it captures the full microtubule binding and assembly activity of the longest Tau isoform, hTau40, in the human brain.
We used solution-state NMR to identify which residues contribute to the jaws of the Tau-microtubule interaction. Solution-state NMR can only observe molecules that are rapidly tumbling in solution, excluding the direct observation of Tau molecules completely bound to MTs. Nevertheless, information about the residues important for binding to microtubules can be obtained when the exchange between the fully bound form and the free state is sufficiently fast, such that conformational characteristics of the bound state are transferred to the soluble, unbound protein. Here, we investigated changes in NMR chemical shifts and signal intensity ratios for increasing amounts of MTs. The observed signal is an average of the resonances originating from the unbound and bound forms of Tau. The magnitude of chemical shift changes will depend on the chemical environment in the bound state and the concentration of the bound species. The latter depends on the rates with which Tau binds and detaches from MTs. In addition, residues that are involved in the binding process will be affected by conformational exchange, potentially affecting the NMR signal intensity of these residues. Therefore, we have two tools available, NMR chemical shift and signal intensity changes that allow identification of the binding sites of Tau to MTs. These two tools show that the binding behavior of Tau to heparin and taxol-stabilized MTs is similar (Fig. 4 and Fig. 7). Within the repeats the same residues that were identified as binding sites on the basis of an NMR titration of K18 (the construct comprising only the repeats) with MTs (27) are important for binding of K32 to MTs. In particular, residues 275 VQIINKKLDLS 285 (comprising the hexapeptide in the beginning of repeat 2) show large chemical shift changes, in agreement with biochemical analyses that reported a strong reduction of MT binding affinity upon mutation of Lys 274 or Lys 281 to alanine (57). In addition, clusters of positive charges upstream of the PGGG motifs at the end of repeats 1, 2, and 3 contribute to the MT binding, thereby providing attachment points for these repeats to tubulin. Repeat 4, on the other hand, does not significantly contribute to MT binding (Fig. 4).
For high affinity binding of Tau to MTs, the regions flanking the repeats are required (5). Our NMR studies show that 225 KVAVVRT 231 and 240 KSRLQTAPV 248 in the proline-rich domain and 370 KIETHKLTFREN 381 in the region downstream of the repeats are binding sites for both heparin and microtubules (Fig. 7). In the proline-rich domain the most strongly affected residues are 226 VAV 228 and 243 LQT 245 . In agreement with the NMR data, deletion analysis mapped the microtubule binding activity of the proline-rich region to residues Lys 224 -Asn 255 and in particular to the stretch 225 KVAVVRT 231 . Moreover, site-directed mutagenesis indicated that Lys 224 , Lys 225 , and Arg 230 are important for MT binding and assembly (58). It is noteworthy that the region 225 KVAVVRT 231 is conserved between Tau and two other microtubule-associated proteins MAP-2 and MAP-4 (58). Interestingly, the two motifs 225-231 and 240 -248 flank the motif 231 TPPKSP 236 that contains two major phosphorylation sites (by GSK3␤ and/or cdk5) and the residue Pro 232 , whose isomerization by Pin-1 is important for the accessibility to phosphatases (59).
Over 30 phosphorylation sites have been identified in Tau, for many of which the level of modification is elevated in Alzheimer disease (60 -62). Prominent sites are located in the flanking domains, e.g. Ser 199 , Ser 202 , Thr 205 , Thr 212 , Ser 214 , Thr 231 , and Ser 235 before the repeats, Ser 396 , Ser 404 and others after the repeats (not contained in K32). The major sites within the repeats are located in the KXGS motifs, i.e. Ser 262 , Ser 293 , Ser 324 , and Ser 356 . These sites are phosphorylated by the kinase MARK, which results in the detachment of Tau from microtubules (35). Ser 262 , Ser 293 , and Ser 324 are separated by only four residues from the charged patches at the C-terminal end of repeat 1, 2, and 3, and Thr 231 is right in the middle of the MT binding region in the proline-rich domain. Thus, phosphorylation might lead to intramolecular charge neutralization of the positive spots required for MT binding. The required bending in the backbone would easily be possible because of the high conformational flexibility of Tau. Alternatively or in combination, phosphorylation can induce other conformational changes, such as the stabilization of a polyproline II helix. Indeed, CD and NMR data on peptides derived from the proline-rich region of Tau suggest that phosphorylation induces a polyproline II helix for residues Val 229 -Ser 238 (40). Residues Val 229 -Ser 238 are located between the two MT-binding elements of the proline-rich domain ( 225 KVAVVRT 231 and 240 KSR-LQTAPV 248 ) and include the Pin1 target site of proline isomerization that is a major regulator of the conformation of Tau. Thus, stabilization of a polyproline II helix in this region would increase the distance between the two MT-binding sites, such that they no longer fit to their interaction sites on the MT surface.
In conclusion, we have provided a detailed view of the structural properties and of the sites of interaction with microtubules and polyanions of a fragment of Tau, which contains the full microtubule binding and assembly activity of the longest Tau isoform, hTau40, in the human brain. It will be interesting to see how genetic mutations and phosphorylation affect the hot spots of aggregation and the jaws of the Tau-microtubule interaction. These studies are currently in progress.