Polymerization of microtubule-associated protein tau under near-physiological conditions.

Neurofibrillary tangles, which form in certain degenerating neurons in the brains of patients with Alzheimer's disease, are amassed from filaments having a straight or paired helical morphology. Solubilization of these filaments reveals that they are composed of the microtubule-associated protein tau. It has not previously been shown, however, that tau will assemble to form filaments of similar morphology under conditions representative of the intracellular environment. We have succeeded in forming such filaments using tau purified from porcine or rat microtubules. The filaments are relatively straight with narrowing at irregular intervals, and are about 10 nm wide, a morphology similar to that of straight filaments seen in Alzheimer's disease neurofibrillary tangles. At tau concentrations of 1-10 μM, in vitro assembly occurs at physiological pH, ionic strength, temperature, and reducing potential, and each one of these factors modulates the reaction. Assembly is judged to be only slowly reversible by the exponential rather than normal distribution of filament lengths, and by the limited disassembly observed under conditions which inhibit polymerization. Tau purified directly from whole brain tissue rather than from microtubules does not polymerize under conditions described in this report.

The intracellular fibrillar pathology of Alzheimer's disease is characterized by the presence of filaments having a straight or paired helical morphology (Kidd, 1963;Yagishita et al., 1981). These filaments accumulate in both the somal neurofibrillary tangles (NFT) 1 and the dystrophic neuropil threads (Braak et al., 1986;Kowall and Kosik, 1987). The formation of NFT and dystrophic neurites are spatially correlated (Probst et al., 1989;Yamaguchi et al., 1990), and both lesions are highly correlated with the severity of dementia (McKee et al., 1991). Filamentous inclusions of this type are also seen in Down's syndrome (Wisniewski et al., 1985), Guamanian Parkinsonism-dementia (Hirano et al., 1968), and other disease states (Wisniewski et al., 1979). In progressive supranuclear palsy, NFT are composed primarily of filaments possessing the straight, unpaired morphology (Tellez-Nagel and Wisniewski, 1973;Bugiani et al., 1979). Although the death of polymer-laden neurons is evidenced by the presence of insoluble tangle remnants in the extracellular space, it is not known whether polymer masses disrupt neuronal function sufficiently to induce degeneration, or whether they merely form preferentially within neurons already involved in the necrotic process.
Straight filaments (SF) and paired helical filaments (PHF) form under similar conditions, as evidenced by their co-existence within individual NFT (Perry et al., 1987). SF share epitopes with PHF and copurify with PHF in protocols that exploit their resistance to SDS or protease treatments (Perry et al., 1987;Crowther, 1991). In addition, there are several reports of transitional forms of fibrils possessing stretches of straight, then paired helical morphology continuous within a single filament (Wischik et al., 1985;Perry et al., 1987;Papasozomenos, 1989;Crowther, 1991). These findings and others have led to speculation that SF and PHF are formed by similar mechanisms of assembly (Perry et al., 1987;Crowther, 1991;Wille et al., 1992).
The only known structural constituent of the PHF is the microtubule-associated protein tau (for a review of the normal biology of tau, see Lee (1990)). The presence of tau proteins has been demonstrated both by immunochemical means (Grundke- Kosik et al., 1986) and by sequencing of peptides extracted from PHF (Wischik et al., 1988;Kondo et al., 1988). Tau extracted from PHF contains more phosphorylated residues than tau isolated from normal brain (Hasegawa et al., 1992;Ksiezak-Reding et al., 1992), and these phosphorylations are frequently invoked as being involved in the polymerization process.
Tau purified directly from brain or from brain microtubules (MT) has been reported to form a variety of polymers resembling SF or PHF. Dialysis of porcine MT tau for several days against 6 -8 M urea produced polymers ranging in width from 5 to 35 nm, which included a subset resembling PHF (Montejo de Garcini et al., 1986;Montejo de Garcini and Avila, 1987). The effects of urea were attributed to deamination of glutamine residues or carbamylation of lysine residues, although producing these modifications by enzymatic or chemical means did not fully reproduce the effects of urea treatment alone. Ureatreated tau was reported to assemble independent of NaCl concentration in the range of 0.1-1 M. Using tau purified directly from bovine whole brain, 10-nm filaments were formed in the presence of the cross-linking enzyme, transglutaminase, under conditions optimized for enzymatic activity (Dudek and Johnson, 1993). It is unlikely that this enzyme is required for tau polymerization in vivo, however, since monomeric tau can be readily solubilized from isolated PHF (Greenberg and Davies, 1990;Lee et al., 1991). Polymer formation has also been demonstrated using bacterially expressed human recombinant tau. Two groups using deletion constructs roughly equivalent to the MT binding domain of tau and similar acidic conditions produced several polymer species, which included a subset possessing the twisted morphology of PHF (Wille et al., 1992;Crowther et al., 1992). Full-length tau constructs did not assemble under these conditions. More recently, however, using conditions of neutral pH and high ionic strength (1.25 M CH 3 CO 2 -K ϩ ), full-length tau constructs were observed to form filaments, some of which resembled PHF (Crowther et al., 1994).
The establishment of causal relationships between the assembly of tau into SF and PHF, and potential modulating factors such as phosphorylation or other enzymatic or chemical treatments, would benefit from an in vitro assembly system in which these polymers can be demonstrated to form under physiologically relevant conditions. We have defined the conditions in which tau purified from rat or porcine MT will assemble into a homogeneous population of filaments resembling SF. Although kinetically a relatively slow process, in vitro filament formation is observed under essentially physiological conditions.

MATERIALS AND METHODS
Fresh bovine brains were obtained from John Morrell Meat Packing, Montgomery, AL. Fresh porcine brains were obtained from Bryan Meat Packing, Westpoint, MS. Human brain was provided by Dr. Richard Powers of the Brain Resource Center, University of Alabama at Birmingham. Sprague-Dawley rats were obtained from Charles River Laboratories, Wilmington, MA, and killed by decapitation. Purified NFT  were provided by Dr. Khalid Iqbal, Institute for Basic Research in Developmental Disabilities, Staten Island, NY. Neurofilament antibodies (Amersham Corp.) and protein were provided by Dr. Robert Goldman, Northwestern University Medical School, Chicago, IL.
Protein Purification-Microtubules were purified from brain by two cycles of temperature-dependent assembly and disassembly essentially as described previously (Shelanski et al., 1973), with glycerol added to 25% during the first warm incubation only. Tubulin was further purified by phosphocellulose chromatography (Weingarten et al., 1975), using phosphocellulose precycled as described (Sloboda et al., 1976). Taxol-stabilized microtubules (Vallee, 1982) were made by incubating purified porcine tubulin at 5 mg/ml with 10 M taxol for 30 min at 37°C, then diluting 1/20 for electron microscopy.
Tau was isolated using protocols similar to those published by others (Sandoval and Weber, 1980;Johnson et al., 1989), exploiting the protein's stability to heat treatment (Cleveland et al., 1977a) and solubility in perchloric acid (Lindwall and Cole, 1984). For isolation from microtubules, pellets were resuspended in cycling buffer (100 mM PIPES, 1 mM EGTA, 1 mM MgCl 2 , pH 6.9) supplemented with 0.8 M NaCl and 2 mM DTT, stirred on ice for 30 min, boiled for 10 min, stirred on ice for 30 min, and centrifuged at 100,000 ϫ g for 45 min. Supernatants were concentrated over an Amicon YM10 ultrafiltration membrane, and loaded on a Bio-Gel A-1.5 sieve column (32 ϫ 430 mm, run at 15 ml/h), equilibrated with buffer A (20 mM MES, 80 mM NaCl, 2 mM EGTA, 1 mM MgCl 2 , 0.1 mM EDTA, pH 6.8) supplemented with 0.8 M NaCl and 2 mM DTT (buffer Aϩ). Fractions containing tau were brought to 2.5% perchloric acid, stirred on ice for 30 min, and centrifuged at 100,000 ϫ g for 30 min. Supernatants were dialyzed against buffer A and concentrated by ultrafiltration. Residual DTT was estimated to be Յ0.2 M. All procedures except boiling were carried out at 4°C.
For isolation of tau directly from whole brain, frozen tissue was thawed and homogenized (1:1, w/v) in a Waring blender in buffer Aϩ supplemented with an additional dry weight of NaCl sufficient to bring the homogenate to 0.8 M NaCl. Following centrifugation at 150,000 ϫ g for 45 min, the supernatant was boiled 10 min, iced 30 min, and centrifuged as above. The supernatants were then brought to 60% saturation with (NH 4 ) 2 SO 4 , stirred on ice 45 min, and centrifuged at 100,000 ϫ g for 30 min. Pellets were resuspended in a total of 50 ml of buffer Aϩ, dialyzed against the same solution, concentrated by ultrafiltration, and loaded on the sieve column. Procedures described above for isolation of tau from microtubules were then followed.
Conditions for Tau Polymerization-In general, tau stored at Ϫ80°C in buffer A was thawed at 4°C, diluted into 100 mM Tris pH 7.2 containing DTT or ␤-mercaptoethanol (␤ME), and incubated at 37°C. 100 mM MES was used for experiments requiring buffering below pH 7. Final tau concentrations were in the range of 1-10 M. Significant variations in time, temperature, pH, ionic strength, and concentration of reducing agent are described in the text.
Electron Microscopy-Samples were deposited in 10-l aliquots onto 400-mesh nickel grids coated with 0.4% Formvar, rinsed with 5 drops of H 2 O, and stained with 5 drops of 2% uranyl acetate, the last drop sitting 1 min prior to blotting. Grids were examined using a JEOL JEM-100CX transmission electron microscope operated at 80 kV. For filament length measurements, micrographs obtained at a nominal magnification of 15,000 ( Fig. 6) or 10,000 (Figs. 7 and 9) were digitized using a SIT68 camera (MTI), and lengths were determined using software from Universal Imaging Corp. Only filaments contained completely within a field and measuring at least 50 nm were included in data sets. Fields selected at random were chosen at low illumination and without the aid of the 10X binoculars, so that Formvar integrity could be assessed without viewing the filaments present. Filament width measurements were made manually using micrographs similar to those shown in Fig.  1 (F-K). Negatively stained microtubules (24-nm assumed diameter, with 6 -7 discernible protofilaments) were used for size reference.
For colloidal gold labeling, following deposition on grids and an H 2 O rinse, grids were inverted for 1 h on a 10-l drop of the primary antibody (Tau-2; Papasozomenos and Binder, 1987) diluted in borate saline (0.1 M H 3 BO 3 , 25 mM Na 2 B 4 O 7 , 75 mM NaCl). Grids were then rinsed with borate saline, inverted on a drop of secondary antibody (Ted Pella, 10 nm colloidal gold conjugated goat anti-mouse, diluted 1/125), rinsed with borate saline supplemented with 1.5 M NaCl, and stained with 2% uranyl acetate.
Electrophoresis-Proteins were separated by SDS-PAGE (Laemmli, 1970) and stained with Coomassie Blue or transferred to nitrocellulose (Towbin et al., 1979). For Western blotting, antibody and blocking (2% nonfat dry milk) incubations were performed in borate saline. Protein concentrations were determined using the method of Lowry (Lowry et al., 1951) after samples in Laemmli sample buffer were precipitated with 10 volumes of 10% perchloric acid, 1% phosphotungstic acid.

RESULTS
Polymerization of MT tau was accomplished by diluting the purified protein into neutral Tris or MES buffers in which sulfhydryl reactivity was limited by the presence of reducing agents. Polymers formed under a variety of conditions are shown in Fig. 1. Polymerization was demonstrated using tau from juvenile rat (indicated by postnatal age), adult rat, and adult pig, at protein concentrations of 1.6 -6.5 M. Incubations were done at 37°C for 5-26 h, using 5-25 mM ␤ME or 2-20 mM DTT. The filaments formed resemble the SF known to reside within the NFT, in the sense that they are non-helical and unpaired. The obvious flexibility of some filaments (Fig. 1D), however, makes them straight in a strictly non-Euclidean sense, and distinguishes them from the rather rigid appearing PHF  and SF (Crowther, 1991). The tau polymers at their widest extent have an average width of 10.5 nm, with a measured range of 6.5-13 nm. Filaments narrow for variable distances at irregular intervals ( Fig. 1, H and J) to about 50% of their widest extent. These narrowings likely represent the crossover points of a slightly twisted filament. SF isolated from NFT also show a modulation of width interpreted as resulting from a twisting of the long axis (Crowther, 1991). The relatively large range of filament diameters observed is believed to result from this variable twisting combined with the final disposition of filaments on the support film, and not from multiple filament morphologies. Differences in staining intensity, which may create the appearance of multiple filament populations within a single field (Fig. 1A), are commonly observed. These are due to the unequal contributions of positive and negative staining across the grid surface, and again, do not indicate the presence of multiple filament morphologies.
Two types of aggregation often seen in preparations of polymerized tau are also shown in Fig. 1. The first is most common in the juvenile tau samples and appears to involve the precipitation of nonfilamentous tau onto tau filaments (Fig. 1B). Because of the electron opaqueness of these aggregates, we cannot rule out the possibility that they are composed of highly compacted short filaments. They are, however, usually associated with a single discernible tau filament. The appearance of such aggregates is usually associated with an obvious decrease in the density of filaments seen by electron microscopy, and confounds attempts to quantify polymerization by centrifugal separation of soluble and filamentous tau. Bundling of tau filaments is also occasionally observed (Fig. 1C). Both forms of aggregation appeared to be reduced when the Tris assembly buffer was replaced with borate saline (data not shown), although this appears to reduce polymer yields.
To date, we have tested 15 tau isolates for their ability to form filaments. Although all of the tau preparations purified from twice-cycled MT (11) exhibit the ability to assemble, none of the preparations purified directly from whole brain extracts (4) are assembly competent (see "Materials and Methods" for differences in purification protocols). The assembly incompetent tau included isolates from bovine, porcine, and human brain. Tau purified by both methods is presumed to have the same primary structure, but is likely to exist in different states of phosphorylation owing to the activity of phospatases and kinases present during MT cycling (Tsuyama et al., 1986;Burns, 1991). Differences in levels of specific post-translational modifications may be involved in stimulating or suppressing polymerization.
The morphology of in vitro assembled tau polymers was compared to that of Alzheimer's SF and PHF (Fig. 2). NFT were isolated using an SDS extraction protocol , and processed for electron microscopy in the same manner as the tau filaments. Most of the filaments found in these preparations exhibited the typical PHF morphology ( Fig. 2A). Occasional SF were observed, and, rarely, a NFT fragment composed almost exclusively of SF was observed (Fig. 2B). A comparison of SF and tau filaments (Fig. 2C) revealed two filament populations, which were similar in width and which presented a smooth surface lacking detectable substructure by this staining method. The PHF at their widest extent were significantly wider than the SF or the in vitro assembled tau filaments. The SF found in purified NFT preparations were apparently more rigid than tau filaments (Fig. 1D) and often appeared to be broken when forced by inter-filament contacts to attempt significant bending. This may result from SF containing a greater number of phosphorylated residues, analogous to the increased rigidity of paracrystals formed from phosphorylated tau (Hagestedt et al., 1989).
Owing to the fact that our polymers are also similar in size to neurofilaments, and that neurofilaments are a known contaminant of the cycled MT from which we purify our assembly competent tau (Berkowitz et al., 1977), we tested our tau iso-  H-K) was polymerized as described under "Materials and Methods" and stained with uranyl acetate. In addition to dispersed preparations, typical patterns of non-filamentous (B) and inter-filamentous (C) aggregation are shown. Taxol-stabilized MT (arrowheads) were co-precipitated on some grids for size reference. Bar equals 500 nm (A-E) or 100 nm (F-K). lates for possible contamination by neurofilament proteins. A Coomassie stain of two typical tau preparations separated by SDS-PAGE revealed dye binding only to the tau bands (Fig. 3,  A and B). A Western blot of an identical gel using a monoclonal antibody directed against neurofilament light chain showed binding to 40 ng of a partially purified neurofilament preparation ( Fig. 3E; binding was also detected at 10 ng), but no binding to the tau proteins loaded at 5-10 g/lane (Fig. 3, C  and D). The maximum potential neurofilament contamination was calculated to be Ͻ0.1%.
To further confirm that in vitro assembled filaments were composed of tau, EM localization was performed using a tau monoclonal antibody in conjunction with gold-conjugated secondary antibodies. Gold particles were specifically associated with filaments when filaments were preincubated with the Tau-2 monoclonal antibody (Fig. 4, A and B), but not when the primary antibody was omitted (Fig. 4, C and D). Treatment of filaments with 0.5 M KCl to remove proteins bound nonspecifically to the surface of filaments did not effect labeling (Fig. 4, E and F).
Analysis of filament assembly revealed that the extent of polymerization was dependent on incubation times and temperatures (Fig. 5). Increasing incubation times resulted in the presence of longer filaments on the grid surface. When tau was incubated at 37°C (Fig. 5, F-J), the filaments formed appeared to be greater in length and number than those formed at 22°C (Fig. 5, A-E). In this preparation of P14 rat tau, no polymerization was evident after a 4-h incubation at 4°C. In subsequent experiments with adult rat tau, however, some polymerization was observed at 4°C (data not shown).
Electron micrographs similar to those seen in Fig. 5 were digitized so that filament lengths could be measured (Fig. 6). The histograms generated confirm that the average filament length increases with time, and, at any given time, filaments in the 37°C sample are longer than those assembled at 22°C. The data also show that filaments display an exponential length distribution, rather than the Gaussian distribution that is usually seen in microtubule populations (Symmons and Burns, 1991). This type of distribution would be consistent with a filament population that was adding nucleation sites at a constant rate and exhibiting limited subunit dissociation.
The dependence of filament assembly on ionic strength was also examined. As neutral solutions (25 mM Tris, 40 mM NaCl) were supplemented with 0 -75 mM KCl, the filament mass subsequently found on grid surfaces was markedly decreased (Fig. 7). All samples were supplemented with 500 mM KCl prior to their deposition on grids, so this decrease is not due to differential adherence to the grid surface, which could have otherwise resulted from the variable salt content of the samples. Filaments were observed at low density when KCl was added at 100 mM (Tris ϩ Na ϩ K ϭ 165 mM), but complete inhibition was routinely observed at higher salt concentrations (data not shown).
With respect to filament depolymerization, increasing KCl concentrations well above assembly-inhibiting concentrations caused only limited filament disassembly. Filaments were still observed when samples preassembled for 24 h were raised to over 500 mM KCl and incubated another 24 h (Fig. 7). This suggests that assembled filaments display only limited subunit dissociation, consistent with the data obtained on filament length distributions (Fig. 6). FIG. 3. Purification of tau to apparent homogeneity. A-D, microtubule tau from P14 rat (5 g, A and C) or adult rat (10 g, B and D) was separated on 4 -20% gels and stained with Coomassie Blue (A and B) or blotted with a monoclonal antibody that binds the 68-kDa neurofilament protein (C and D). Neurofilament protein was also blotted (40 ng, E) or stained with Coomassie Blue (15 g, F). Molecular size markers (in descending order: 208, 101, 71, 44, 29, and 18 kDa) are indicated to the left, and the neurofilament triplet proteins on the right. Lanes C-E were excised from the same blot.
FIG. 4. Antibody labeling of tau polymers. Porcine tau was assembled overnight, and filaments were either deposited directly on grids (A-D) or incubated in 500 mM KCl for 30 min at 37°C prior to being deposited on grids (E and F). Grids were then incubated with the tau monoclonal antibody, Tau-2 (A, B, E, and F), or control buffer (C and D), followed by incubation with a secondary antibody conjugated to 10 nm colloidal gold. Grids were then stained with uranyl acetate. Two examples of gold labeling for each condition are shown. Bar equals 0.2 m.
The effect of pH on tau polymerization is shown in Table I. Tau from P14 and adult rat was incubated overnight in 20 mM ␤ME buffered at variable pH, then assayed for filament formation by electron microscopy. Assembly occurred over the broad pH range between 6 and 11, but was largely inhibited at pH 5.6 and completely inhibited at lower pH. Samples that were incubated overnight at inhibitory pH and then raised to neutral pH exhibited normal assembly (data not shown).
In our earliest attempts at in vitro polymerization of P14 rat tau, it was readily apparent that reducing agents must be included in the incubation buffer if filaments of sufficient length to be recognized as such were to be formed. The relation between ␤ME concentration and the length of filaments assembled from P14 tau is shown qualitatively in Fig. 8. As ␤ME concentration was lowered from 0.1 M (Fig. 8B) to 0.1 mM (Fig.  8E), a decrease in maximum filament length was observed. At FIG. 6. Filament lengths display an exponential distribution. P14 rat tau was incubated in 10 mM DTT at 22°C (A-E) or 37°C (F-J) for the indicated number of hours (t). Histograms were generated using filament measurements obtained as described under "Materials and Methods." Average filament length (x) and sample size (n) are also given for each data set. Bin width is 200 nm.
low ␤ME concentrations, a concomitant increase in particles smaller than those recognized as filaments was often observed (Fig. 8, D and E). These may represent minimal length polymers. Similar data were produced using porcine tau and DTT as the reducing agent (Fig. 9). Although in this experiment the average filament length clearly decreased as the concentration of DTT was lowered (Fig. 9F), the length of filaments observed at the lowest level of reducing agent was more variable than that seen in other experiments (Fig. 8), 2 with filaments as long as 2.4 m recorded. This variation reflected the variable extent to which different preparations of MT tau were observed to polymerize, with those preparations that produced the highest level of polymer typically exhibiting a more uniformly short population of filaments at low concentrations of reducing agents. Although high concentrations of reducing agents were used in many experiments to maximize the assembly of long filaments, it should be emphasized that the range of reducing potentials represented in Fig. 9 would clearly encompass values expected to occur in the cytoplasm. Physiological concentrations of glutathione (1-10 mM, Hwang et al., 1992), the tripeptide that constitutes most of the redox buffering capacity of the cytoplasm, also promoted the assembly of MT tau purified from P11 rat, adult rat, and adult porcine brain (data not shown). DISCUSSION We have succeeded in assembling a homogeneous population of tau filaments that appear to be morphologically related to the SF seen in Alzheimer's disease. It has not previously been demonstrated that tau will form polymers of this nature under conditions typical of the intracellular environment. Previous studies describing the assembly of tau filaments possessing straight or paired helical morphologies (widths ranging from 10 to 25 nm) have depended on the use of non-physiological conditions. Assembly of bovine or porcine tau was previously achieved only after the chemical (Montejo de Garcini et al., 1986) or enzymatic (Dudek and Johnson, 1993) modification of the protein. Assembly of recombinant tau constructs containing either the complete or partial tau sequence relied on either high salt concentrations (1.25 M CH 3 CO 2 -K ϩ ; Crowther et al., 1994) or an acidic pH (Wille et al., 1992, Crowther et al., 1992. None of these previous studies utilized reducing agents in their assembly reactions. In this paper we demonstrate that in a reducing environment, the assembly of 10-nm tau filaments occurs when the variables of temperature, pH, and ionic strength are adjusted to physiological values. Of these variables, only ionic strength is limiting at physiological values, with optimal assembly occurring at lower salt concentrations. Although filament densities observed at physiological ionic strength were low, significant polymer mass would likely accumulate in situ on a longer time scale associated with the pathogenesis of the disease state. We believe that tau polymers formed in vitro, as well as SF formed in vivo, are ribbon-like in the sense that in cross-section their width is about twice their thickness, and that their measured profile depends on the extent to which they are lying either flat or on edge. Consistent with this interpretation is the two-fold difference in the range of widths measured for tau filaments. An example of a filament probably laying partially on edge is seen in Fig. 1J, where the central segment is very narrow but the ends are seen to flare to more typical dimensions. This interpretation of the filament morphology implies that the actual maximun width of these tau polymers is closer to the range maximum (13 nm) than the average measured width (10.5 nm). Although values of 15 nm reported for the width of SF in thin sectioned tissue are somewhat greater (Metuzals et al., 1981;Yagishita et al., 1981), overestimates because of the deposition of positive stain may result from this technique (Ruben et al., 1993).
Because of their similar dimensions, the possibility that the polymers assembled from our tau preparations were actually neurofilaments was addressed. Coomassie-stained gels of representative tau samples revealed no neurofilament or other protein contaminants, and Western blots of the same samples revealed no proteins cross-reactive with an anti-neurofilament protein antibody. The polymers formed were heavily labeled by an anti-tau antibody, even after incubating filaments in high salt to remove peripherally bound proteins. There were also differences between assembly conditions described in this paper and those previously reported for the assembly of neurofilaments. The assembly of tau filaments was favored at low ionic strength and inhibited at high ionic strength, while neurofilaments are assembled at high ionic strength (Geisler and Weber, 1981) and disassembled at low ionic strength (Hisanaga and Hirokawa, 1988). In addition, lowering the temperature or raising the pH of buffers used for filament assembly had no effect on the morphology of tau filaments (data not shown), but resulted in the formation of neurofilaments exhibiting abnormal morphologies (Aebi et al., 1988).
Following the time course of polymerization at different temperatures reveals three points of interest. First is the observation that all filament populations exhibited an exponential distribution of filament lengths, a fact confirmed by the length distributions generated over a range of DTT concentrations. This type of distribution is in contrast to the Gaussian distributions observed in equilibrated microtubule populations (Symmons and Burns, 1991) and is consistent with a filament population that exhibits limited subunit dissociation and adds nucleation sites at a constant rate. This interpretation is supported by the observation that tau filaments incubated at salt concentrations sufficient to inhibit assembly displayed only limited disassembly, again suggesting a relatively slow rate of subunit dissociation. A second point of interest is that the polymer mass formed increases as the temperature increases. This parallels the results of circular dichroism studies, which 2 D. M. Wilson and L. I. Binder, unpublished data.

FIG. 7. Polymerization of tau filaments is dependent on ionic strength.
Porcine tau in buffer A was diluted 1:1 into neutral 50 mM Tris, 20 mM DTT supplemented with a variable amount of KCl. Samples were incubated for 24 h at 37°C and then supplemented with another 500 mM KCl. One aliquot was taken immediately for EM processing (open circles), and another after an additional 24-h incubation at 37°C (filled circles). Random fields were digitized, after which filaments were measured and their lengths were summed. The average total polymer length/field Ϯ S.E. is plotted as a function of the initial cation concentration, n Ն 10. demonstrated that the secondary structure content of tau also increased as a function of temperature (Ruben et al., 1991). It is likely, therefore, that prior to polymerization the tau molecule assumes a more highly ordered conformation. The third point is that tau filaments appear to lengthen with time while maintaining constant radial dimensions. This suggests that filaments are elongating by the endwise addition of subunits, as opposed to the lateral condensation of preformed protofilaments. This was not evident in previous studies, which have all examined polymers from single time points.
A role for tau dimers in the assembly of PHF-like polymers has previously been suggested (Wille et al., 1992). Treatment of tau constructs with phenylenedimaleimide to induce non-reducible cross-linking of cysteine residues resulted in the formation of tau dimers as demonstrated by SDS-PAGE, and these dimers assembled to form filaments identical to those formed by untreated tau (Wille et al., 1992). Our observation that filament lengths decrease as ␤ME or DTT concentration decreases suggests that tau dimers forming in a less reducing environment may inhibit the polymerization process. It has previously been shown in MT populations, however, that a factor that stimulates assembly (microtubule-associated proteins) reduces the average MT length (Sloboda et al., 1976). This illustrates the inverse relationship that can exist between assembly rate and filament length and points out that our data are open to more than one interpretation, since we measured filament lengths as opposed to total polymer mass. Although disulfide bonds resistant to cytoplasmic reduction are found in globular proteins, they are not expected in a protein like tau, which is believed to exist in an extended conformation with limited secondary structure (Cleveland et al., 1977b. Since the region of tau that contains the cysteine residues is known to be a relatively hydrophobic part of the molecule (Ruben et al., 1991), however, it is possible that conformational changes associated with dimerization and/or polymerization trap the relevant cysteines in a central hydrophobic domain not easily accessed by reducing agents.
The involvement of disulfide bonds may also be indicated by the pH dependence of assembly (Table I). The inhibition of a Tau purified from microtubules was diluted 1:8 into 100 mM MES or Tris buffered at the indicated pH. Samples were incubated 16 h at 37°C in 20 mM ␤ME, then viewed by electron microscopy. b Filaments were observed, but at very low density.
FIG. 8. Tau polymerization is dependent on reducing potential. P14 rat tau was incubated overnight at pH 7.2 with ␤ME added to 1.0 M (A), 0.1 M (B), 10 mM (C), 1.0 mM (D), 0.1 mM (E), or 0.0 mM (F). Samples were processed for EM, and representative fields are shown. Bar equals 1 m.
FIG. 9. Filament lengths are a function of the reducing potential. A-E, porcine MT tau was incubated 24 h with the indicated concentration of DTT. Samples were deposited on grids, and filaments from 10 randomly chosen fields were digitized and measured. Bin width is 300 nm. F, the average length of each filament population is plotted as a function of DTT concentration. assembly seen below pH 6.0 could be due to decreased sulfhydryl reactivity, although in this pH range effects due to histidine protonation cannot be ruled out. It is interesting to note that the assembly of tau deletion mutants was also observed to be pH-dependent. Under non-reducing conditions, constructs encompassing most of the MT binding domain were not observed to assemble at neutral pH, but did form filaments at pH 5.0 -5.5 (Wille et al., 1992) or pH 4.5-5.0 (Crowther et al., 1992). We have also observed that under non-reducing conditions, tau preparations that do not form filaments of significant length at neutral pH (Fig. 8F) will form long filaments at pH 5.5 (data not shown). Taken together these data suggest that, although a sufficiently low pH will completely block polymerization, under conditions more conducive to assembly the effects of pH and reducing agent on disulfide stability are additive, and increasing either [␤ME] or [H ϩ ] will increase the average filament length.
Although tau purified from MT was observed to form filaments using our assembly conditions, tau purified from whole brain appeared assembly-incompetent. Another study, however, was able to demonstrate the presence of 10-nm filaments in samples of whole brain tau incubated with transglutaminase, an enzyme that catalyzes intermolecular cross-links between glutamic acid and lysine residues (Dudek and Johnson, 1993). The formation of filaments morphologically similar to those shown in our study suggests that whole brain tau is capable of associating in a manner similar to that which precedes the polymerization of MT tau. However, in the absence of the cross-linking enzyme, it appears likely that in the case of whole brain tau, this association is reversible. This implies that differences in the post-translational modification of tau purified by these two methods can directly affect the stability of cohesive tau interactions, and that transglutaminase can induce assembly independent of these modifications.
Phosphorylation affects many of the structural and functional properties of tau. Given that we are unable to isolate normal tau in a phosphorylation state identical to that of tau that has been incorporated into PHF and SF, it is not surprising that we are unable to exactly duplicate these morphologies. Tau filaments assembled in vitro did not resemble PHF and exhibited greater flexibility than SF. Specific phospates, which might allow for the lateral association of subunits during PHF assembly, could be absent or occluded by other phosphates present in the tau purified from cycled MT. Likewise, the increased phosphorylation of tau incorporated into SF and PHF could contribute to increased structural rigidity, analogous to the increase in the rigidity of tau paracrystals induced by phosphorylation (Hagestedt et al., 1989). We believe it is likely that if the phosphorylation state of SF and PHF could be duplicated, then identical structures would form under our assembly conditions. Using the conditions for tau polymerization defined in our study has several potential advantages over previously reported procedures. First, reasonable filament yields were obtained at relatively low tau concentrations (1-10 M). Previous studies have used concentrations of 50 M (Ͻ10 filaments/field; Montejo de Garcini and Avila, 1987) to 250 M (Crowther et al., 1994). One study did report assembly at a tau concentration of 2.5 M, but only in the presence of transglutaminase (Dudek and Johnson, 1993). Since this enzyme could potentially induce polymer accumulation in a manner independent of physiologically relevant post-translational modifications, this protocol might not be suitable for defining those protein modifications which result in changes in filament morphology or rates of assembly. Second, unlike conditions reported for the assembly of tau fragments (Wille et al., 1992;Crowther et al., 1992), the conditions defined in this study are conducive to assembly of the full-length tau protein. This is important if the contribution of all regions of the tau sequence to the assembly process are to be assessed. Third, use of our assembly conditions results in the polymerization of a morphologically homogeneous population of filaments. Therefore, analyzing changes in morphology resulting from protein modification may be more straightforward than when using conditions which are reported to result in heterogeneous filament populations (Montejo de Garcini et al., 1986;Wille et al., 1992;Crowther et al., 1992Crowther et al., , 1994. Finally, if chemical or enzymatic treatments of tau are to be screened for their potential ability to modulate tau polymerization in vivo, they should preferably be examined under conditions as close to physiological as possible.