Structural Determinants of Tau Aggregation Inhibitor Potency*

Background: Mechanistic insight into small-molecule Tau aggregation inhibitors is needed for their advancement as therapeutic agents. Results: Structure-activity relationship analysis identified polarizability as a common descriptor of inhibitor potency. Conclusion: Flat, highly polarizable ligands stabilize soluble oligomeric complexes at the expense of filamentous aggregates. Significance: The findings suggest a basis for rational improvement of ligand potency, whereas maintaining bioavailability. Small-molecule Tau aggregation inhibitors are under investigation as potential therapeutic agents against Alzheimer disease. Many such inhibitors have been identified in vitro, but their potency-driving features, and their molecular targets in the Tau aggregation pathway, have resisted identification. Previously we proposed ligand polarizability, a measure of electron delocalization, as a candidate descriptor of inhibitor potency. Here we tested this hypothesis by correlating the ground state polarizabilities of cyanine, phenothiazine, and arylmethine derivatives calculated using ab initio quantum methods with inhibitory potency values determined in the presence of octadecyl sulfate inducer under reducing conditions. A series of rhodanine analogs was analyzed as well using potency values disclosed in the literature. Results showed that polarizability and inhibitory potency directly correlated within all four series. To identify putative binding targets, representative members of the four chemotypes were added to aggregation reactions, where they were found to stabilize soluble, but SDS-resistant Tau species at the expense of filamentous aggregates. Using SDS resistance as a secondary assay, and a library of Tau deletion and missense mutants as targets, interaction with cyanine was localized to the microtubule binding repeat region. Moreover, the SDS-resistant phenotype was completely dependent on the presence of octadecyl sulfate inducer, but not intact PHF6/PH6* hexapeptide motifs, indicating that cyanine interacted with a species in the aggregation pathway prior to nucleus formation. Together the data suggest that flat, highly polarizable ligands inhibit Tau aggregation by interacting with folded species in the aggregation pathway and driving their assembly into soluble but highly stable Tau oligomers.

Because the appearance of Tau protein-bearing lesions in Alzheimer disease (AD) 3 correlates with neurodegeneration and cognitive decline (1,2), various approaches for inhibiting their formation are under investigation as potential therapies against disease progression. An attractive target is the Tau aggregation reaction itself, which is closely associated with lesion formation but not normal Tau function (3). Despite having the advantage of disease specificity, the approach of directly inhibiting Tau protein-protein interactions faces hurdles, including the lack of a distinguishable "binding pocket" on Tau monomer because of its natively unfolded structure, and the large surface areas that mediate Tau-Tau interactions, which could require impractically large molecules for effective antagonism (4). Under such conditions, Tau aggregation inhibitors would be expected to lack adequate binding affinity and brain bioavailability for therapeutic utility.
Nonetheless, small-molecule Tau aggregation inhibitors have been reported in the literature (5)(6)(7)(8)(9)(10). These consist of various chemotypes, including but not limited to phenothiazines (6), polyphenols (7), porphyrins (7), rhodanines (8), and cyanines (9,10), with phenothiazine derivative methylene blue showing promise for delaying progression of AD (11). Each scaffold family differs in molecular weight, hydrophobicity, and sterics, yet all inhibit Tau aggregation in vitro at micromolar or even submicromolar concentrations. These findings suggest that inhibition of Tau aggregation with small molecules is feasible, and raise the question of how inhibitory efficacy and potency is achieved across scaffold classes. The problem is significant because most inhibitors identified to date on the basis of in vitro approaches have physical-chemical characteristics inconsistent with efficient blood-brain barrier penetration, hindering their preclinical evaluation. For example, thiacarbocyanine, phenothiazine, and arylmethine inhibitors are permanent cations (6,10,12), whereas porphyrin and pthalocyanine derivatives have relatively high molecular masses (7). Unlike these molecules, aggregation antagonists with improved bioavailability could have therapeutic utility.
Descriptors of inhibitor potency traditionally have been identified through structure-activity relationship (SAR) analysis, where compound structure is systematically varied and correlated with biological activity. Using this approach, we postulated that ligand polarizability was a candidate descriptor of cyanine potency in vitro (10,13). Polarizability is an electronic property that describes how easily electron density can shift about a molecule when exposed to an external electric field, such as an adjacent dipole or ion. For planar molecules, high polarizability can support strong van der Waals interactions with flat surfaces exposed on binding partners (14). In contrast, reports of SAR analysis within rhodanine and phenothiazine families have focused on sterics rather than chemical descriptors of graded biological potency (15,16). As a result, the activity driving commonalities among diverse chemotypes have remained elusive.
Full understanding of the mechanism of Tau aggregation inhibition will require perspective from the point of view of the target (i.e. the Tau species with which the inhibitor interacts) as well as from that of compound (i.e. the descriptors of inhibitory potency). In vitro, the aggregation pathway begins with the conversion of natively unfolded Tau monomer into an aggregationprone confirmation (17). This step can be accelerated in vitro by inclusion of an "inducer" such as heparin or anionic detergent (reviewed in Ref. 18). Once aggregation competent conformations are adopted, the rate-limiting step toward fibrillization becomes nucleation. This is followed by extension, where Tau monomers add onto filament ends (17). Tau aggregation inhibitors have been reported to inhibit the forward reaction as well as drive disaggregation of mature filaments, but their molecular targets have been identified in only a few cases. For example, aldehydes can covalently modify Tau protein monomer, thereby trapping it in aggregation incompetent forms (19). Other compounds promote Cys oxidation under non-reducing conditions and extended incubation times, again yielding assembly incompetent monomer (20,21). Neither of these covalent mechanisms is predicted to have utility in vivo. However, other protein aggregation inhibitors are active in the presence of thiol reducing agents or are otherwise not associated with protein oxidation or alkylation. For example, curcumin has been reported to increase the reconfiguration rate of ␣synuclein, such that occupancy of assembly competent conformations is minimized (22,23). This mechanism implies direct but transient interaction between inhibitor and natively unfolded protein monomer. Finally, certain aromatic heterocycles have been reported to trap Tau in the form of soluble oligomeric species (7), even when tested under reducing conditions (24). Similarly, in a study of ␣-synuclein aggregation, polyphenol, phenothiazine, polyene macrolide, porphyrin, and Congo red derivatives were found to stabilize SDS-and Sarkosyl-insoluble oligomer as observed by SDS-PAGE analysis (25). However, neither the target of ligand binding nor the descriptors of inhibitor affinity were identified in these studies.
To clarify the mechanism of inhibitory action from the perspectives of both compound and protein, here we investigate the activity of four series of Tau aggregation inhibitors, composed of cyanine, phenothiazine, arylmethine, and rhodanine derivatives under reducing conditions. The results point toward ligand polarizability as a common descriptor of inhibitory potency, and at least partially folded Tau intermediates as their molecular target.
Tau Aggregation-Recombinant human Tau preparations (3 M) were incubated (37°C) without agitation in assembly buffer (10 mM HEPES, pH 7.4, 100 mM NaCl, 5 mM dithiothreitol) for up to 24 h in the presence or absence of fibrillization inducer ODS (50 M) and inhibitors (up to 10 M). Control reactions contained DMSO vehicle, which was limited to 1% (v/v) final concentration in all aggregation reactions. Following incubation, reactions were immediately assayed by either electron microscopy or SDS-PAGE, or fractionated by sedimentation as described below.
Electron Microscopy-Reaction aliquots (50 l final volume) were treated with 2% glutaraldehyde (final concentration), then mounted on grids and negatively stained with 2% uranyl acetate as described previously (32). Random fields were viewed with a Tecnai G2 Spirit BioTWIN transmission electron microscope (FEI, Hillsboro, OR) operated at 80 kV and ϫ23,000 -49,000 magnification. Total filament length is defined as the sum of the lengths of all resolved filaments per field and is reported Ϯ S.D.
SDS-PAGE-Samples were boiled for 2 min in the presence of sample buffer under reducing conditions (3.75% 2-mercaptoethanol), then applied to polyacrylamide slab gels as described previously (33).
Sedimentation-To separate soluble from insoluble Tau species, aliquots of aggregation reactions (100 l final volume) were centrifuged (100,000 ϫ g) in a Ti 42.2 rotor for 30 min at 4°C (34). Pellets (P1 fraction) were re-suspended in 100 l of assembly buffer by vigorous trituration, whereas supernatants (S1 fraction) were removed and subjected to density gradient sedimentation to further fractionate soluble Tau species. Sucrose step gradients composed of 20, 30, 40, and 50% sucrose layers (1 ml/layer) prepared in assembly buffer (35,36) were centrifuged (100,000 ϫ g) in an SW55 rotor for 2 h at 4°C. Each layer was then collected and analyzed for Tau content as described below.
Tau Protein Assay-The Tau content of fractions was determined by dot blot analysis using nitrocellulose membranes (0.2 m porosity) as described previously (37). The membranes were blocked in 5% nonfat dry milk dissolved in blocking buffer (100 mM Tris-HCl, pH 7.4, and 150 mM NaCl) for 1 h and then incubated with primary antibody Tau5 at 1:5000 dilution for 2 h. Membranes were then washed three times with blocking buffer and incubated with horseradish peroxidase-linked secondary antibody for 2 h. After washing another three times with blocking buffer, membranes were imaged using the Enhanced Chemiluminescence Western blotting Analysis System (GE Healthcare, Buckinghamshire, UK) recorded on an Omega 12iC Molecular Imaging System and quantified using Ultra-Quant software (UltraLum, Claremont, CA, USA).
Spectrophotometry-Absorbance measurements were made in neat methanol solvent as described previously (10). The maximum absorbance wavelength ( max ) was determined from fits of the data to a multi-Gaussian function (38).
Computational Chemistry-Semi-empirical descriptors were calculated for each analyzed compound using E-DRAGON 1.0, an online implementation of the DRAGON 5.4 molecular descriptor generator (39) that computes Ͼ1600 descriptors categorized into 20 logical blocks (40). The starting E-DRAGON descriptor set was then pruned on the basis of reported dye SAR studies (41,42) to yield a focused molecular property set comprising 278 descriptors representing five logical blocks: 48 constitutional descriptors, 33 connectivity indices, 154 functional group counts, 14 charge descriptors, and 29 molecular properties. The pruned descriptor set was then augmented with clogP and topological polar surface area values were calculated using the Molinspiration Property Calculation Service, and with polarizability (␣) values calculated at the quantum level using density functional theory methods implemented in Gaussian 09 (G09) (43). We described these methods in detail previously (14,37). Quantum calculations were performed in implicit solvent using the density field theory functional B3LYP and the 6 -311ϩϩG(d,p) basis set to more accurately model the photophysical properties of dye molecules (44).
To correlate descriptor and IC 50 data, Genetic algorithm-PLR analysis was performed using the Virtual Computational Chemistry Laboratory, an online portal for computational chemistry tools (last accessed 1 June 2013). PLR models were optimized on the basis of leave-one-out cross-validation (Q 2 loo ). External validation of PLR models was performed in R version 2.13.0. The test set was chosen using the Kennard-Stone algorithm (as implemented in the ken.sto function of the soil.spec package). Statistical analysis was performed as described previously (45), where R 0 2 , RЈ 0 2 , k, and kЈ correspond to the correlation coefficients and slopes of linear regressions constrained through the origin (46). Data Analysis-Concentration-effect data from either filter or electron microscopy assays were normalized to DMSO vehicle control reactions and fit to the function, where y and y max represent the minimum and maximum aggregation measured in the presence and absence of inhibitor (at concentration x), respectively, n is the Hill coefficient, and IC 50 is the concentration of inhibitor that results in 50% of maximal inhibition. IC 50 values are reported Ϯ S.E. of the estimate.
All measured parameters are reported as mean Ϯ S.D. of biological replicates. Differences between groups were analyzed by one-way analysis of variance and Tukey's post hoc multiple comparison test.

Structure Activity Relationship Analysis of Cationic Tau
Aggregation Inhibitors-To test whether polarizability was a descriptor of Tau aggregation inhibitory potency, experimentation focused initially on cyanine, phenothiazine, and arylmethine inhibitor families. These series were chosen because representative members were commercially available that varied narrowly in size and sterics, but more widely in potential descriptors of inhibitor activity such as polarizability. The cyanine series included eight molecules (1-8) with varying heterocycle and substituent composition, and one containing two thiacarbocyanine moieties in macrocyclic configuration (9) ( Table 1). The inhibitory potencies of cyanines 1-8 against ODS-induced aggregation of full-length recombinant human 2N4R Tau under near-physiological conditions of Tau concentration, pH, ionic strength, and reducing conditions were reported previously (10), whereas the concentration-effect relationship for 9 assayed under identical conditions is shown in Fig. 1a. ODS detergent was used as aggregation inducer in these experiments because of its efficacy with full-length Tau isoforms (47) and because of the reported ability of micellar detergents to depress small-molecule aggregation associated with promiscuous activity (48). Overall, the nine cyanines were found to vary in IC 50 value over nearly 2 orders of magnitude Յ10 M (Table 1). When these values were compared against polarizability (␣) calculated for the ground state using ab initio quantum methods (Table 1), a strong log-linear correlation was observed (R 2 ϭ 0.72; Fig. 2a). In contrast, only weak correlations were observed between IC 50 and clogP, topological polar surface area, or M r (shown for clogP only in Fig. 2b). These data indicate that increasing electron delocalization within the ground-state cyanine ring system correlated positively with aggregation inhibitory potency within this narrow series.
The approach was then extended to phenothiazine and arylmethine derivatives. The phenothiazines included the established aggregation inhibitor 10 (methylene blue) and four ana-logs (11)(12)(13)(14) that differed solely in the strengths of the electron donor/acceptor groups flanking the phenothiazine nucleus ( Table 2). As a result, this series varied over relatively narrow ranges of molecular weight and hydrophobicity (Table 2). When assayed under identical conditions as the cyanine inhibitors described above, only the two phenothiazines with greatest ␣ value inhibited Tau aggregation with IC 50 Յ 10 M (Fig.  1b). Similarly, the arylmethine series comprised four triarylmethines that varied solely in the composition of electron donor/ acceptor groups (15)(16)(17)(18) and Bindschedler's green (19) (Table  2). When assayed under identical conditions as the cyanine inhibitors described above, only the three arylmethines with greatest ␣ inhibited Tau aggregation with IC 50 Յ 10 M (Fig.  1c). Within the phenothiazine and triarylmethine series, the rank order of potency paralleled the strengths of constituent electron donor moieties (-N(CH 3 Table 1.  (Table 1) (10) were plotted as a function of a, polarizability and b, clogP values calculated by ab initio and semi-empirical methods, respectively. Solid lines represent best fit of the data to linear regression, with the resultant correlation coefficient reported as R 2 . Cyanine inhibitory potency correlated directly with polarizability but not clogP. -OH) established on the basis of their Hammett substituent constants (49). Overall, these data implicate polarizability as one descriptor of inhibitory potency in three distinct scaffolds, and identify electron donating substituents as being effective drivers of ligand polarizability.
Structure-Activity Relationship Analysis of Rhodanine Aggregation Inhibitors-Cyanines, phenothiazines, and arylmethines all share permanent cationic character, which may limit their ability to cross the blood-brain barrier. In contrast, ideal Tau aggregation inhibitors would likely have neutral charge to enhance brain penetrability. To test whether polarizability could modulate inhibitory activity in a neutral chemotype, a structure-activity relationship analysis of rhodanine derivatives was performed. IC 50 values for these compounds, which spanned nearly 4 orders of magnitude, were taken from the literature (8). However, because of the computational challenges raised by adamantyl, boronyl, and ferrocenyl moieties, only 45 of the reported 52 molecules were analyzed herein (Table 3). Many members of this library were predicted to be charged at physiological pH, but this resulted from ionizable pendent R groups rather than from the core scaffold itself (which was neutral).
Because of the moderate size of this dataset, and its substantial structural diversity, a cheminformatics approach was taken to identify the best combination of descriptors for predicting inhibitory activity. On the basis of a training set of 39 molecules, the best PLR model rationalized pIC 50 (i.e. -log IC 50 ) in terms of 35 molecular descriptors (x variables) collapsed into six linear combinations (latent t variables) ( Table 4). On the basis of internal leave-one-out cross-validation (Q 2 loo (50)) the correlation was adequately strong (Table 4). Moreover, when the model was applied to an external test set (i.e. six compounds not used in the calibration), the resulting correlation between predicted and observed IC 50 values met target criteria of slope and goodness of fit (46) necessary for predictive utility (Table 5, Fig. 3). Together, the internal and external validation experiments indicated that an acceptable rationalization of rhodanine activity over a broad concentration range was achieved.
The five highest-weighted and therefore top-ranked descriptors identified by the model as contributing to rhodanine potency were AlogS, ␣, BLTF96, AlogP, and topological polar surface area ( Table 4). The second highest weighted of these was ␣, and its positive coefficient indicated that increasing polarizability correlated directly with rhodanine potency. This result was consistent with the cyanine, phenothiazine, and arylmethine series described above. The next three descriptors   NOVEMBER 8, 2013 • VOLUME 288 • NUMBER 45 related to compound hydrophobicity. That inhibitory potency correlated directly with AlogP (51) and inversely with topological polar surface area (52), suggested that compound hydrophobicity also contributed to rhodanine activity. However, the positive correlation with BLTF96, another hydrophobicity index that varied inversely with octanol-water partition coefficient (53), indicated that the contribution of hydrophobicity may be complex and/or have an optimum. In fact, the positive correlation between the highest weighted descriptor, AlogS (54), and potency revealed that for this series, which contained a number of large hydrophobic analogs, maintenance of aqueous solubility was of paramount importance. Taken together, this analysis confirmed polarizability as an important molecular descriptor for Tau aggregation inhibitor potency that was shared among multiple chemotypes.

Tau Aggregation Inhibitor Structure and Mechanism
Aggregation Inhibitors Stabilize Soluble Tau Oligomers-The foregoing identified flat, highly polarizable compounds as mediators of Tau aggregation antagonist activity. To gain insight into their mechanism of action, the products of recombinant human 2N4R Tau aggregation reactions performed in the presence and absence of the most potent commercially available inhibitor from each compound family (cyanine 1, phenothiazine 10, triarylmethine 15, and rhodanine 28A) were separated by sedimentation and quantified by immunoblot analysis. In the absence of ODS inducer, Tau protein did not aggregate, and so nearly all Tau remained in the supernatant fraction (i.e. the S1 fraction) after centrifugation (Fig. 4a). In the presence of ODS, however, Tau aggregated to form filaments as reported previously (47), and the majority of protein product migrated with the insoluble fraction (Fig. 4a). In contrast, when ODS and 1, 10, or 15 were present, the distribution of reaction products shifted toward the soluble fraction (p Ͻ 0.001; Fig. 4a). Rhodanine 28A was the least efficacious inhibitor under these conditions, but it too shifted reaction products to the soluble fraction (p Ͻ 0.05; Fig. 4a). These data show that all four inhibitors acted to stabilize soluble forms of Tau protein when present during the aggregation reaction.
To characterize the Tau species stabilized by inhibitors, S1 supernatants from each experiment were subfractionated on discontinuous gradients containing 20 -50% sucrose. These conditions were chosen because they had been reported to resolve Tau monomer, which remains mostly in the lowest density sucrose layer, from soluble oligomeric Tau complexes, which appear mostly in denser sucrose layers (36). Indeed, when the supernatant from the control Tau reaction lacking inducer was fractionated on the gradient, the overwhelming majority of immunoreactivity migrated within the least dense layer (20% sucrose; Fig. 4b). In contrast, soluble products resulting from the aggregation control prepared in the presence of ODS migrated predominantly in the denser layers, consistent with the formation of small soluble aggregates (Fig. 4b). The presence of aggregation inhibitors increased the levels of Tau migrating in the denser fractions still further. In particular, Tau levels in the 40% sucrose layer were significantly elevated for 1, 10, and 15 (Fig. 4b). In contrast, the differences in Tau levels produced by 28A reached statistical significance only in the 20% sucrose layer (Fig. 4b). Together, the sedimentation data were consistent with Tau aggregation inhibitors acting to stabilize soluble Tau species of varying density.   To further characterize soluble Tau species, aliquots of each S1 fraction were analyzed by SDS-PAGE after boiling under reducing conditions. In the absence of inhibitor, filamentous Tau aggregates were efficiently solubilized in SDS/2-mercaptoethanol-containing sample buffer to yield monomer migrating at ϳ67 kDa (Fig. 5). In contrast, the presence of inhibitors depressed the amount of Tau migrating in the monomer position, with 1, 10, and 28A reaching statistical significance under these conditions (Fig. 5). These data suggested that the soluble Tau species stabilized by aggregation inhibitors resisted SDS solubilization. To further characterize this behavior, the time course of cyanine 1-mediated effects of Tau migration in SDS-PAGE was investigated. Compared with aggregation reactions containing DMSO vehicle alone, incubation in the presence of 1 led to decreasing levels of solubilized monomer and increasing levels of slowly migrating species within 1 h (Fig. 6). After 2 h incubation with 1, SDS-soluble monomer decreased to Ͻ25% of control levels, whereas slowly migrating species appeared that did not enter the gel (Fig. 6). After a 4-h incubation, the majority of Tau was rendered SDS-insoluble (Fig. 6). Together with sedimentation data, these findings indicate that cyanine 1 and Tau interact rapidly to form highly stable oligomeric complexes that resist solubilization in SDS under reducing conditions.

Cyanine-mediated Oligomer Formation Requires Inducer and the Tau MTBR Region-Because
Tau oligomers were stabilized most strongly by cyanine 1, this compound was selected as a probe for Tau-ligand interactions. Moreover, because of its speed and simplicity, loss of SDS solubility was used as a secondary assay for quantifying interactions between 1 and a range of Tau constructs that would be difficult to capture using aggregation assays alone. To identify the regions of Tau protein that mediate inhibitor activity, the ability of 1 to depress SDS solubility of Tau truncation mutants was investigated. Because all Tau constructs used to characterize inhibitor activity up to this point were tagged with an N-terminal His tag, we first compared the performance of recombinant wild-type 2N4R and 2N4R ⌬6His proteins in the presence or absence of compound 1 (see Fig. 7a for maps of 2N4R Tau and the truncation mutants used herein). As before, incubation of 2N4R Tau protein in the presence of 1 under aggregation conditions lowered its SDS solubility, and the absence of the His tag did not change this   NOVEMBER 8, 2013 • VOLUME 288 • NUMBER 45 JOURNAL OF BIOLOGICAL CHEMISTRY 32605 pattern (Fig. 7b). When quantified from replicate analysis (n ϭ 3 biological replicates), the null hypothesis was accepted (Fig.  7c). These data indicate that the His tag did not mediate interaction between Tau protein and 1. Therefore, all subsequent experiments continued to leverage His-tagged Tau constructs to expedite their purification and analysis. These included a series of truncations that systematically deleted Tau sequences from the C terminus through the MTBR region (Fig. 7a). Results showed that deletion of all residues C-terminal to the third (i.e. 2N4R ⌬344 ) or fourth (i.e. 2N4R ⌬376 ) MTBR did not affect the ability of 1 to lower SDS solubility of Tau (Fig. 7, b and  c). However, deletion of sequences C-terminal to the second MTBR (i.e. 2N4R ⌬314 ) yielded a significant reduction in efficacy, whereas deletion before (i.e. 2N4R ⌬252 ) or after (i.e. 2N4R ⌬283 ) the first MTBR completely destroyed it (Fig. 7, b and  c). Conversely, the MTBR region alone (i.e. 2N4R 252-376 ) supported full efficacy (Fig. 7, b and c). These data revealed that sequences in the MTBR region, and especially within or adjacent to the first two imperfect repeats, mediated the loss of SDS solubility driven by cyanine 1.

Tau Aggregation Inhibitor Structure and Mechanism
The MTBR region implicated in inhibitor-Tau interactions also mediates Tau aggregation propensity (55), suggesting that aggregation intermediates could be direct targets of inhibitor action. To test this hypothesis, cyanine 1 interactions with 2N4R Tau and aggregation-modulating mutants (Fig. 8a) were studied. First we investigated ODS dependence, because entry into the aggregation pathway involves conversion of natively disordered Tau into aggregation competent conformations, and anionic inducers such as ODS promote this step (reviewed in Ref. 18). Indeed, incubation of 2N4R Tau in the absence of ODS yielded no detectable filaments (data not shown), whereas the presence of ODS yielded abundant filaments biased toward shorter lengths (Fig. 8b). These data confirmed that detectable 2N4R fibrillization was inducer dependent under these conditions. In contrast, the presence or absence of ODS and accompanying fibrillization had no effect on Tau solubility in SDS when analyzed by SDS-PAGE (Fig. 8, c and d). However, the presence of ODS was necessary for Tau SDS solubility to be lowered by cyanine 1 (Fig. 8, c and d). These data suggest that cyanine does not interact efficiently with natively unfolded Tau, but with conformers populated as a result of interaction with ODS micelles.   (76). The construct corresponding to the MTBR region alone (2N4R 252-376 ) is referred to as K18a. b, representative gel migration pattern visualized by Coomassie Blue stain. c, quantification of SDS-soluble Tau monomer remaining after 4 h incubation (n ϭ 3; reported Ϯ S.D.), normalized to each DMSO vehicle control of the construct. The black bar corresponds to the 2N4R Tau DMSO vehicle control normalized to itself, which was taken to represent 100% SDS solubility, and therefore the negative control for inhibitor activity on 2N4R Tau. The blue bar correspond to cyanine-1-mediated depression of SDS-soluble 2N4R Tau normalized to its DMSO vehicle control, and therefore represents the positive control for inhibitor activity on 2N4R tau. ***, p Ͻ 0.001, for comparison of each normalized construct versus 2N4R negative control. #, p Ͻ 0.05; ###, p Ͻ 0.001, for comparison of each normalized construct against the 2N4R positive control. Loss of SDS solubility by cyanine required sequences within the MTBR region.
The second step in the aggregation pathway is reportedly nucleation, which is mediated in part by the 275 VQIINK 280 (i.e. PHF6*) and 306 VQIVYK 311 (i.e. PHF6) hexapeptide motifs located in the MTBR region (56, 57) (Fig. 8a). Missense mutations that introduce Pro residues into these motifs (e.g. 2N4R I277P,I308P ) have been reported to depress ODS-mediated filament formation almost completely (30). Indeed, incubation of 2N4R I277P,I308P with ODS inducer yielded only small aggregates devoid of fibrillar morphology (Fig. 8b). Nonetheless, 2N4R I277P,I308P supported loss of SDS solubility in the presence of cyanine 1 (Fig. 8, c and d). These results imply that Tau species residing at or beyond the nucleation step are not required for 1-mediated effects on Tau solubility.
The MTBR region also contains two Cys residues that can modulate 2N4R Tau aggregation propensity when oxidized (58). Some compounds foster this reaction when incubated with Tau under non-reducing conditions (21). To assess the contribution of Cys residues to cyanine activity, the interaction of 2N4R C291A,C322A , a double mutant that could not be oxidized or modified at these positions under any circumstances, was investigated. As reported previously (28), fibrillization of 2N4R C291A,C322A resembled wild-type 2N4R Tau, except that a smaller number of longer filaments were produced (Fig. 8b), suggesting a lower nucleation rate. Nonetheless, 2N4R C291A,C322A interacted with cyanine 1 to yield depressed levels of SDS-soluble Tau (Fig. 8, c and d). This result indicates that cyanine activity is not mediated solely through Cys oxidation or covalent binding to these nucleophiles. Overall, these experiments support cyanine 1 interacting with intermediates stabilized by ODS inducer, but not necessarily with species at or beyond the nucleation stage of the aggregation pathway.

Cyanine Activity Extends to Familial Tauopathy Mutants-
In addition to 2N4R Tau, humans express alternatively spliced isoforms that lack products of exons 2, 3, and 10 ( Fig. 9a). Moreover, certain missense mutations cause Tau lesion formation in familial forms of frontotemporal lobar degeneration (59). A successful aggregation inhibitory strategy should extend to these forms of Tau as well. Therefore, the ability of cyanine 1 to depress SDS solubility of purified Tau isoform 0N3R and tauopathy missense mutants 2N4R R5L , 2N4R G272V , 2N4R P301L , 2N4R V337M , and 2N4R R406W was investigated (Fig. 9a). No statistically significant differences were observed between these proteins and 2N4R Tau with respect to compound-mediated depression of SDS solubility (Fig. 9, b and c).
To test whether depression of SDS solubility correlated with aggregation inhibitory efficacy, 2N4R, 0N3R, and the tauopathy mutants were subjected to aggregation conditions in the presence and absence of 1. Levels of insoluble Tau were then estimated after centrifugation. As before, the majority of 2N4R Tau was rendered insoluble when incubated with DMSO vehicle, whereas the presence of 1 significantly depressed recovery of insoluble Tau (p Ͻ 0.001, Fig. 9d). In contrast, only a minority of fibrillization-incompetent mutant 2N4R I277P,I308P was recovered in the pellet fraction in the presence of DMSO vehicle (p Ͻ 0.001 relative to 2N4R), and the presence of 1 did not alter this distribution (Fig. 9d). Compared with these boundary examples, 0N3R and the tauopathy missense mutations most closely resembled 2N4R Tau with respect to recovery of insoluble Tau (Fig. 9d). Although 1 significantly depressed recovery of all investigated mutants in insoluble form, the probability of rejecting the null hypothesis was lowest for 2N4R G272V (p Ͻ 0.01) and 2N4R P301L (p Ͻ 0.05), consistent with the especially The black bar corresponds to the 2N4R Tau DMSO vehicle control normalized to itself, which was taken to represent 100% SDS solubility, and therefore the negative control for inhibitor activity on 2N4R Tau. The blue bar corresponds to cyanine-1-mediated depression of SDS-soluble 2N4R Tau normalized to the DMSO vehicle control, and therefore represents the positive control for inhibitor activity on 2N4R Tau. ***, p Ͻ 0.001, for comparison of each condition (gray bars) versus 2N4R negative control. ###, p Ͻ 0.001, for comparison of each condition against the 2N4R positive control. Depletion of SDS-soluble Tau species by 1 required the presence of ODS inducer, but not Cys residues or intact PHF6 and PHF6* sequence motifs.
high aggregation propensities reported for these species (27). These data indicate that the inhibitory mechanism identified herein will likely extend to the multiple Tau species implicated in neurofibrillary lesion formation.

DISCUSSION
Aggregation Antagonism from the Perspective of Ligand-Many seemingly unrelated Tau aggregation antagonists have been disclosed in the academic (7) and patent (5) literatures, some of which show efficacy at high nanomolar/low micromolar concentrations. Despite structural diversity, many of these compounds share the absorbance characteristics of dyes (i.e. they absorb electromagnetic radiation in the visible spectrum), a property that stems from delocalized -electron distribution (60). Here we found that ligand polarizability, a metric of electron delocalization in the ground state, correlated with inhibitory potency within a series of cyanine, phenothiazine, arylmethine, and rhodanine inhibitors. Although polarizability was not the only descriptor of affinity, and by itself was not a predictor of affinity among scaffolds, its modulation may offer a route for maximizing potency within individual chemotypes. The series investigated here suggest two strategies for doing so. The first is to increase the size of the conjugated -electron network. This approach, which was leveraged in the rhodanine series, also may contribute to the reported activities of polyene, porphyrin, pthalocyanine, and other large inhibitors (7,24).
However, polarizability also can be increased through appropriate positioning of electron donating and withdrawing groups, as illustrated by the phenothiazine and arylmethine series. The rank order of potency paralleled the strengths of constituent electron donor moieties in both of these series. This strategy has the potential to maximize polarizability whereas minimizing ligand size, and therefore facilitate blood-brain barrier penetration (61). For in vivo applications, it will be important to do so within neutral chemotypes that, unlike the permanent cations used herein, support passive diffusion into brain.
Aggregation Antagonism from the Perspective of the Tau Target-Previously, we provided evidence that the ODS-mediated Tau aggregation paradigm used to characterize inhibitors resembles a heterogeneous nucleation pathway (reviewed in Ref. 18). In the absence of inhibitors, natively unfolded Tau enters this pathway by interacting with ODS micelles to yield assembly competent conformations (Fig. 10). Once populated, the rate-limiting step becomes formation of a thermodynamic nucleus on the micelle surface, which then extends by endwise addition of free monomers (Fig. 10). Here we found that inhibitor-mediated disappearance of insoluble Tau filaments was accompanied by the appearance of soluble oligomeric complexes that resisted denaturation in SDS. Similar Tau oligomers were previously reported to form in the presence of pthalocyanine (24) and also methylene blue (7). More gener- FIGURE 9. Oligomer-forming efficacy of cyanine inhibitor extends to tauopathy mutants and 3R Tau. Recombinant 2N4R tauopathy mutants 2N4R R5L , 2N4R G272V , 2N4R P301L , 2N4R V337M , and 2N4R R406W and Tau isoform 0N3R (3 M each) were incubated (16 h, 37°C) with ODS inducer in the presence and absence of cyanine 1, then subjected to SDS-PAGE. a, map of missense mutants used in the analysis (shown in "orange") relative to segments derived from alternatively spliced exons E2, E3, and E10. b, representative SDS-PAGE migration pattern visualized by Coomassie Blue stain. c, quantification of SDS-soluble Tau remaining after 16 h incubation normalized to DMSO vehicle control (n ϭ 3; reported Ϯ S.D.). The black bar corresponds to the 2N4R Tau DMSO vehicle control normalized to itself, which was taken to represent 100% SDS solubility, and therefore the negative control for inhibitor activity on 2N4R Tau. The blue bar corresponds to cyanine-1-mediated depression of SDS-soluble 2N4R Tau normalized to the DMSO vehicle control, and therefore represents the positive control for inhibitor activity on 2N4R Tau. ***, p Ͻ 0.001, for comparison of each construct (gray bars) versus the 2N4R negative control. In contrast, no difference (p Ͻ 0.05) was detectable between each construct and the 2N4R positive control, consistent with 1 being capable of depleting SDS-soluble Tau species from all six Tau species examined. d, the products of Tau aggregation reactions (16 h, 37°C) conducted in the presence of cyanine 1 or DMSO vehicle alone were centrifuged (100,000 ϫ g for 30 min), and the percentage of total Tau protein migrating in the pellet (P) fractions quantified by dot blot analysis using monoclonal antibody Tau5 (n ϭ 3; reported Ϯ S.D.). *, p Ͻ 0.1; **, p Ͻ 0.01; and ***, p Ͻ 0.001, for comparison of insoluble Tau in the presence versus absence of cyanine 1; ###, p Ͻ 0.001, for comparison of insoluble Tau in the absence of 1 among Tau constructs. The presence of 1 shifted reaction products out of the pellet fraction for all shown Tau constructs except 2N4R I277P,I308P .
ally, the phenomenon has been observed with ␣-synuclein in the presence of certain polyalcohol, flavonoid, polyene, diazo, porphyrin, and phenoxazine derivatives (25,(62)(63)(64). The reciprocal relationship between Tau protein oligomers and filaments suggests they lie off the aggregation pathway (Fig. 10). Sequestration of Tau would be expected to raise the apparent filament critical concentration and to foster endwise disaggregation of mature filaments, both of which have been observed in the presence of cyanine inhibitor (38,65). Flat, highly polarizable molecules appear to share this soluble-oligomer stabilizing activity.
The molecular characteristics outlined above are especially appropriate for interacting with protein surfaces through dispersion effects (reviewed in Ref. 66). The ODS-mediated Tau aggregation paradigm tested herein yields multiple targets that could interact with ligands in this way (Fig. 10). Traditionally, secondary assays for Tau-inhibitor interaction have been used to identify candidate interacting species residing along this pathway. For example, NMR spectroscopy can detect direct interactions between pthalocyanine and Tau protein (24). More generally, direct interactions between diazo dyes, phenothiazines, polyalcohols, and pthalocyanine with monomeric ␤-amyloid (67) and ␣-synuclein (25,68,69) have been reported as well. However, this approach requires high-micromolar to low-millimolar concentrations of ligand, and it is not clear that natively unfolded protein structures could support such interactions at the low protein concentrations used to demonstrate aggregation inhibition. Using thioflavin dye displacement as another secondary assay, direct interaction between phenothiazines, cyanines, and arylmethines with mature Tau fibrils has been reported as well (14). Although such binding can be high affinity, it is difficult to rationalize aggregation inhibitory activity in terms of this target alone.
In contrast, here we used loss of SDS solubility as the secondary assay for Tau-ligand interactions, allowing us to investigate cyanine interactions at near-physiological bulk Tau concentrations and reducing conditions. The results showed that natively unfolded monomer alone was incapable of supporting oligomer formation. Rather, the absolute requirement for ODS inducer suggested that folded conformations residing along or off the aggregation pathway were the true substrates for inhibitor binding. Moreover, that intact PHF6 and PHF6* hexapeptide motifs were not required for interaction with cyanine further narrowed candidate binding partners for cyanine inhibitor to those existing prior to nucleation (Fig. 10). Because we previously reported that polarizability is a descriptor of high affinity binding to a cross-␤-sheet structure (14), it is tempting to propose that the target has ␤-sheet character. However, ␣-helices also can present appropriate surfaces for binding -delocalized ligands (70), and these too have been proposed to form in conjunction with anionic surfaces (71).
Therapeutic Implications-The mechanism of Tau aggregation inhibitors proposed here has favorable implications for therapy of tauopathies. First, it can act at physiologically relevant bulk Tau concentrations. Indeed, cyanine inhibitors have been reported to depress Tau aggregation in ex vivo mouse models of tauopathy (10,72). Although oligomer formation has been linked to toxicity in some biological models (73), the physical characteristics of inhibitor-stabilized oligomers differs (24). Consistent with this hypothesis, treatment of brain slices with cyanine inhibitor did not induce apoptotic responses at the low concentrations needed to clear aggregates (10,72). Second, it supports interaction with both 3R and 4R isoforms as well as missense tauopathy mutants, suggesting it could be broadly applicable to both AD and frontotemporal lobar degeneration diseases. Third, it does not rely on interaction with natively unfolded Tau monomer, which is an important binding partner of microtubules. In fact, ϳ50% of proteins contain long stretches of unfolded structure (reviewed in Ref. 74), which could potentially cross-react with inhibitors that target such regions. Finally, the mechanism predicts that inhibition could be useful at early stages of disease before filaments are rendered irreversibly insoluble through cross-linking (75).
In summary, we have identified polarizability as a potential link among structurally diverse Tau aggregation inhibitors. The compounds act to rapidly stabilize soluble oligomeric species at the expense of filamentous aggregates. The proposed mechanism suggests design considerations for optimizing inhibitors with potential therapeutic utility. The Tau aggregation pathway begins with conversion of natively unfolded monomer (U X ) to an assembly competent conformation (U C ). In vitro, step 1 requires the presence of an inducer. Without inhibitor, the rate-limiting step to aggregation is the formation of a dimer, which represents the thermodynamic nucleus (N). The dimer becomes a fibril (F) through extension via endwise polymerization. The presence of cyanine dye inhibits fibril formation by shifting the equilibrium to an off-pathway oligomer (O s ) that is soluble on the basis of centrifugation but SDS-insoluble as determined by SDS-PAGE.