Evidence for the presence of three distinct binding sites for the thioflavin T class of Alzheimer's disease PET imaging agents on beta-amyloid peptide fibrils.

Imaging the progression of Alzheimer's disease would greatly facilitate the discovery of therapeutics, and a wide range of ligands are currently under development for the detection of beta-amyloid peptide (Abeta)-containing plaques by using positron emission tomography. Here we report an in-depth characterization of the binding of seven previously described ligands to in vitro generated Abeta-(1-40) polymers. All of the compounds were derived from the benzothiazole compound thioflavin T and include 2-[4'-(methylamino)phenyl]benzothiazole and 2-(4'-dimethylamino-)phenyl-imidazo[1,2-a]-pyridine derivatives, 2-[4'-(dimethylamino)phenyl]-6-iodobenzothiazole and 2-[4'-(4''-methylpiperazin-1-yl)phenyl]-6-iodobenzothiazole, and a benzofuran compound (5-bromo-2-(4-dimethylaminophenyl)benzofuran). By using a range of fluorescent and radioligand binding assays, we find that these compounds display a more complex binding pattern than described previously and are consistent with three classes of binding sites on the Abeta fibrils. All of the compounds bound with very high affinity (low nm K(d)) to a low capacity site (BS3) (1 ligand-binding site per approximately 300 Abeta-(1-40) monomers) consistent with the previously recognized binding site for these compounds on the fibrils. However, the compounds also bound with high affinity (K(d) approximately 100 nm) to either one of two additional binding sites on the Abeta-(1-40) polymer. The properties of these sites, BS1 and BS2, suggest they are adjacent or partially overlapping and have a higher capacity than BS3, occurring every approximately 35 or every approximately 4 monomers of Abeta-(1-40)-peptide, respectively. Compounds appear to display selectivity for BS2 based on the presence of a halogen substitution (2-[4'-(dimethylamino)phenyl]-6-iodobenzothiazole, 2-[4'-(4''-methylpiperazin-1-yl)phenyl]-6-iodobenzothiazole, and 5-bromo-2-(4-dimethylaminophenyl)benzofuran) on their aromatic ring system. The presence of additional ligand-binding sites presents potential new targets for ligand development and may allow a more complete modeling of the current positron emission tomography data.

Imaging the progression of Alzheimer's disease would greatly facilitate the discovery of therapeutics, and a wide range of ligands are currently under development for the detection of ␤-amyloid peptide (A␤)-containing plaques by using positron emission tomography. Here we report an in-depth characterization of the binding of seven previously described ligands to in vitro generated

A␤-(1-40) polymers. All of the compounds were derived from the benzothiazole compound thioflavin T and include 2-[4-(methylamino)phenyl]benzothiazole and 2-(4dimethylamino-)phenyl-imidazo[1,2-a]-pyridine derivatives, 2-[4-(dimethylamino)phenyl]-6-iodobenzothiazole and 2-[4-(4؆-methylpiperazin-1-yl)phenyl]-6-iodobenzothiazole, and a benzofuran compound (5-bromo-2-(4dimethylaminophenyl)benzofuran)
. By using a range of fluorescent and radioligand binding assays, we find that these compounds display a more complex binding pattern than described previously and are consistent with three classes of binding sites on the A␤ fibrils. All of the compounds bound with very high affinity (low nM K d ) to a low capacity site (BS3) (1 ligand-binding site per ϳ300 A␤-(1-40) monomers) consistent with the previously recognized binding site for these compounds on the fibrils. However, the compounds also bound with high affinity (K d ϳ100 nM) to either one of two additional binding sites on the A␤-(1-40) polymer. The properties of these sites, BS1 and BS2, suggest they are adjacent or partially overlapping and have a higher capacity than BS3, occurring every ϳ35 or every ϳ4 monomers of A␤-(1-40)-peptide, respectively.
Central to the neuropathology of AD is the deposition in specific brain regions of polymeric peptide/protein deposits, composed of ␤-amyloid peptide (A␤) and tau protein, termed senile plaques (SPs) and neurofibrillary tangles (NFTs), respectively (reviewed in Ref. 1).
The precise role of these deposits in the pathogenesis of AD is still unclear. However, individuals with mutations leading to increased A␤ production are at a high risk of developing AD (2), and individuals with mutations in the tau gene are linked to a class of neurodegenerative diseases including frontotemporal dementia and parkinsonism linked to chromosome 17 (3). Moreover, the accumulation of SPs and NFTs follows a predictable temporal and spatial sequence in individuals affected with AD (4,5). Detection of these deposits may therefore provide a clinically useful tool for monitoring disease progression and treatment effects. Positron emission tomography (PET) uses positron emitting radioligands to image and measure ligandbinding sites in vivo and is currently being developed as a sensitive way to track SPs and NFTs in patients (reviewed in Ref. 6).
The binding of these compounds to A␤-(1-40) fibrils generated in vitro has been investigated in a number of laboratories, and the results of these studies so far are consistent with three independent binding sites, one for each structural class of ligand, and all with high binding affinities. However, the binding stoichiometry for all these structural classes of ligand has been reported to be low, with ligand-binding sites occurring only every few hundreds (8) or even thousands (14) of A␤-(1-40) monomers, and this raises a question of whether the signal to be anticipated from further development of these radioligands will be clinically useful.
Two of the ligands 6-OH-BTA-1/PIB and FDDNP have so far * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  progressed to in vivo imaging studies (15,16), comparing small numbers of control and AD patients. The results demonstrate increased retention of labeled species in areas of the brain thought to be affected by AD. However, the precise nature of the signal in these studies remains uncertain given the low density of ligand-binding sites apparently present on A␤ fibrils. The situation is further complicated by ex vivo measurement of PIB binding to brain homogenates obtained post-mortem from patients with AD (8). A␤ peptides were not measured in this study but were interpolated from a different study by Nä slund et al. (17). The conclusion was a binding stoichiometry approaching 1:1, much higher than all reports based on the direct study of A␤ fibrils in vitro (8).
The binding properties of the ligands to A␤ fibrils have all been characterized by using essentially identical filter binding assays employing either narrow ranges of radioligand to determine K d values (10,13) or a single fixed low concentration of radioligand and varying concentrations of cold competitor (7,18). The former assay format will only generally detect very high affinity binding as lower affinity interactions dissociate during the filter washing step, whereas the latter format is a site-specific assay and will not be sensitive to novel ligandbinding sites. The results of these assays have not yet been confirmed by other ligand-binding techniques.
We have addressed these issues by developing a series of homogeneous binding assays that take advantage of the intrinsic fluorescence of both Thio T and a number of representative derivative compounds to measure their binding properties over a wide range of ligand and A␤ fibril concentrations. The results of these assays were compared with classical radioligand binding assays and are consistent with the presence of the previously described low density ligand-binding site but also demonstrate the presence of additional higher density ligandbinding sites on the fibrils. The selectivity of these additional sites and their potential significance for the innovation of selective PET radioligands for the clinical investigation of AD and its treatments are discussed.

MATERIALS AND METHODS
Compound Names and Sources-The structures of the compounds employed in this study are shown in Fig. 1  Preparation of A␤-(1-40) Fibrils-Human A␤-(1-40)-peptide (Batch number MK0611 from California Peptide Research, Napa, CA) was incubated at 0.5 mg/ml in PB buffer (10 mM sodium phosphate, 1 mM EDTA, pH 7.4) at 37°C for 48 h in an orbital shaker at 200 rpm. The formation of fibrils was confirmed by Congo Red binding (21) and Thio T binding (21) together with analysis by negative stain electron microscopy using 1% phosphotungstic acid. Fibrils were used immediately upon production or aliquoted and stored at Ϫ80°C. No difference in ligand binding behavior was observed between freshly prepared and frozen material as noted also by others (7).
Compound Preparation-All compounds were prepared as 1-10 mM Me 2 SO stocks before dilution into assay buffer. The maximum final concentration of Me 2 SO in the assays was 0.5%. Because of the relatively low solubility of the ligands (with the exception of Thio T) in aqueous solution, all assays were performed in PB supplemented with 10% ethanol (PBE) and at a maximum concentration of 5 M compound.
Intrinsic Fluorescence Binding Assays-Intrinsic fluorescence intensity (FLINT) changes associated with ligand binding to A␤-(1-40) were recorded in a Ultra Evolution 384 plate reader (Tecan, Reading, UK) using a 340 -440-nm filter pair. For assays measuring the binding of Thio T, a 450 -505-nm filter pair was used. Two formats were used for the FLINT assays. FLINT1 was performed using a fixed concentration of ligand (50 or 100 nM) and varying concentrations of A␤-(1-40) polymer. FLINT2 was performed using a fixed concentration of A␤-(1-40) (500 nM) and varying concentrations of ligand (0 to 2 M) as detailed under "Results." Fluorescence polarization measurements were recorded in a Ultra Evolution 384 plate reader using a 360-nm polarizing excitation filter and 440-nm emission filters. The assays employed a fixed concentration of BTA-1 (50 nM) and varying concentrations of A␤-(1-40) (0 -50 M).
All fluorescent assays were performed in PBE in a final volume of 80 l and were incubated for 2 h at 20°C before reading. All data points were performed at least in quadruplicate and were analyzed using Grafit (Erithacus Software Limited, Horley, UK) to obtain K d values using the single site ligand-binding module.
Reactions were performed in PBE in a final volume of 80 l and were incubated for 2 h at 20°C before measurement in a Ultra Evolution 384 plate reader using a 450 -505-nm filter pair. All data points were performed in quadruplicate, and the fluorescent signal in the absence of competitor defined a fractional binding of 1. Data were analyzed using Grafit to obtain IC 50 values using full 4-parameter curve fits. K i values were derived from the Cheng-Prusoff equation (22), was the concentration (1000 nM) and K d (750 nM) of Thio T used in the assay.
Radioligand Binding Assays-A fixed concentration of A␤-(1-40) (40 nM) was titrated against a range of [ 3 H]Me-BTA-1 concentrations (0.05-40 nM) in PB buffer supplemented with 10% ethanol (PBE) for 3 h at 20°C. The final reaction volume was 500 l. Nonspecific binding was determined in the presence of 50 M Thio T. The bound and free fractions were separated by vacuum filtration through GF/B glass filters (Whatman, Maidstone, UK) using an M-48T harvester (Brandel, Gaithersburg, MD) followed by four 1-ml washes with PBE. Filters containing the bound ligand were mixed with 3 ml of Ultima Gold MV scintillation liquid (PerkinElmer Life Sciences) and incubated for 2 h before counting in a TriCarb 2100TR liquid scintillation counter (PerkinElmer Life Sciences). All data points were performed in triplicate. The specific binding signal under these assay conditions was ϳ70%. Data were analyzed using Grafit to obtain the apparent dissociation constant (K d ) and the maximal number of binding sites (B max ) using the single site ligand-binding module.
Radioligand Competition Assays-Competition assays employed a fixed concentration of A␤-(1-40) (40 nM) and [ 3 H]Me-BTA-1 (4 nM) and used varying concentration ranges of cold competitor dependent on the ligand. Reactions were performed in PBE and were incubated for 2 h at 20°C before harvesting and counting as described above. Nonspecific binding was determined in the presence of 50 M Thio T. BF1 (200 nM) was used to determine nonspecific binding for assays involving Thio T. All data points were performed in triplicate, and the specific binding signal in the absence of competitor defined a fractional binding of 1. Data were analyzed using Grafit to obtain IC 50 values using full 4-parameter curve fits. K i values were derived as stated above using the Cheng-Prusoff equation (22), where [L] was the concentration (4 nM) and K d (4.2 nM) of the radioligand used in the assays.
Density Gradient Centrifugation of A␤ Fibrils-A␤ fibril preparations were fractionated on gradients consisting of 700 l each of 60, 50, 40, 30, 20, 10, and 280 l of 5% Optiprep (Sigma) prepared in 13 ϫ 51-mm ultraclear centrifuge tubes (Beckman Instruments, High Wycombe, UK). A 300-l sample of A␤ fibrils (0.5 mg/ml) in PB was carefully layered on top of the Optiprep gradient and then fractionated at 50,000 rpm for 2 h and 30 min at 22°C in an Optima Max-E ultracentrifuge using an MLS-50 swinging bucket rotor (Beckman Instruments). Following centrifugation, the bottoms of the tubes were pierced with a 21-gauge syringe needle, and 12 ϫ 400-l fractions were collected for analysis. Fraction 1 represents the most dense part of the gradient and fraction 12 the least dense part of the gradient.
The fractions were analyzed for thioflavin T binding by mixing 80 l of each fraction with 2 l of 100 M thioflavin and reading the fluorescence after 2 h of incubation in an Ultra Evolution 384 plate reader using a 450 -505-nm filter pair. In addition, 5 l of each fraction was separated by SDS-PAGE followed by blotting onto nitrocellulose as described previously (23). Blots were probed with the mouse antihuman ␤-amyloid antibody 6E10 (Merck) at a 1:1000 dilution followed by an anti-mouse horseradish peroxidase-conjugated antibody (Sigma) at a 1:10,000 and were then developed with SuperSignal West Dura extended duration substrate (Perbio, Cheshire, UK). Images of the blot were captured using a Chemigenius2 imaging system (Syngene, Cambridge, UK) and quantified using Genetools (Syngene). Radioligand binding was evaluated by mixing 50 l of each fraction with 450 l of 4 nM [ 3 H]Me-BTA-1 prepared in PBE. The reactions were incubated for 2 h at 20°C before harvesting and counting as described above by vacuum filtration.

Determination of Binding Constants Using Fluorescence
Assays-The compounds used in this study (Fig. 1) all share similar fluorescent properties, with excitation and emission wavelengths of around 340 and 440 nm, respectively, but that are distinct from those of Thio T (450:480 nm) (data not shown). In addition, the compounds display increases in their intrinsic FLINT when bound to A␤-(1-40) fibrils along with slight shifts in their excitation and emission wavelengths consistent with ligand binding into a hydrophobic pocket on the fibrils. The exception to this was BTA-1, which displayed a slight decrease (ϳ5-10%) in FLINT when bound to the A␤-(1-40) fibrils.
The increases in FLINT upon ligand binding to A␤-(1-40) fibrils were exploited in designing simple, homogeneous assays to directly measure compound binding. Two assay formats were employed, FLINT1 and FLINT2. In FLINT1 assays, compound binding was measured by using a fixed concentration of ligand (50 or 100 nM) and varying concentrations of A␤-(1-40) polymer (Fig. 2, A and C). A fixed concentration of A␤-(1-40) polymer (500 nM) and varying concentrations of ligand were used in the FLINT2 assays (Fig. 2, B and D). The binding constants K d1 and K d2 obtained from the FLINT1 and FLINT2 assays are summarized in Table I.
The binding of BTA-1 to A␤-(1-40) fibrils was determined by fluorescent anisotropy measurements using a fixed concentration of ligand (50 nM) and varying concentrations of polymer (Fig. 2E). The apparent K d value from this assay was 5.23 M. The anisotropy value of the compound at saturating concentrations of polymer was 0.4, which is toward the maximum limit for a fluorophore (24) and probably reflects the large Stokes radii of the A␤-(1-40) polymers.
The binding isotherms for all of the ligands in all assay formats were consistent with a single population of binding sites for each of the ligands. The apparent K d values from the assays were found to be strongly dependent on the assay format, with the FLINT2 assays consistently reporting lower values. In addition, the rank order of binding constants for the individual compounds was conserved between the two assay formats (Table I).
Comparison of the K d2 values with the literature values obtained previously indicates that although fluorescence assays are reporting significantly higher values, the rank order of binding affinities is generally well conserved. The exception to this was Thio T whose K d2 value was essentially identical to that reported using a similar assay format by Naiki et al. (25) and LeVine (26), indicating that the present set of assays was accurately reporting binding constants.
Comparison of Binding Constants Is Consistent with Two Classes of Binding Sites on the A␤ Fibrils-The observation that the binding constants were dependent on assay format was unexpected as both assay formats should report approximately the same K d values. However, an explanation for this difference became apparent when the number of potential binding sites on the A␤ polymer was considered. The binding data previously reported with these ligands demonstrated that the binding site density on the A␤ fibrils was Ͻ Ͻ1 site per A␤ peptide monomer (8,14). This meant that the effective concentration of binding sites on the fibrils was potentially significantly less than the concentration of A␤ peptide monomers added to the FLINT1 assay. Because calculation of the K d1 values are based on the A␤ peptide monomer concentration, this leads to an overestimation of the binding constant by a factor dependent on the binding site density. In contrast, the calculation of K d2 values is directly dependent on the ligand concentration, and as a consequence the FLINT2 assay does not overestimate the binding constants. Furthermore, the ratio K d1 /K d2 allows an estimation of the ligand-binding site density on the A␤-(1-40) fibrils (Table I). This sum clearly splits the ligands into two distinct groups. Thio T, IMPY-H, and IMPY-Me have a ratio of ϳ30 -40 (i.e. 1 ligand-binding site per ϳ35 A␤-(1-40) monomers), whereas TZDM, TZPI, and BF1 have a ratio of ϳ3-5 (i.e. 1 ligandbinding site per ϳ4 A␤-(1-40) monomers). These data suggested that there were at least two distinguishable classes of binding sites on the A␤ fibrils and are supported by the observation that the latter group of compounds all contain a halogen substitute on their aromatic ring.
In summary, the data from the intrinsic fluorescence assays are as follows: (a) consistent with a single affinity binding mode for each ligand and accurately reporting intrinsic binding constants; (b) determining K d values significantly higher than that previously described; and (c) when combined with the evidence of the binding site densities and selectivity the data, suggested that there was more than one class of binding site on the A␤ fibrils: a low density site (every ϳ35 A␤ monomers) and a higher density site (every ϳ4 A␤ monomers) that are termed BS1 and BS2, respectively.
Fluorescent Competition Assays Support Presence of BS1 and BS2-Ligand binding to A␤-(1-40) fibrils was further characterized by using a fluorescent competition assay (FLINT3) employing Thio T as a reporter for BS1. The K i values obtained from these assays were again significantly higher than those reported in the literature (Table I). However, they were similar to the K d2 values providing additional evidence that the fluorescent assays were accurately reporting binding data and consistent with our thesis that there were additional ligandbinding sites on the A␤ fibrils.
The competitive binding curves were consistent with displacement from a single type of site for each ligand on the  (Fig. 3) and also indicated that the compounds could again be separated into the same two groups identified from the FLINT1/2 assays such that IMPY-H and IMPY-Me (and BTA-1) were able to displace Ͼ80% of the bound Thio T (Fig.  3A), whereas TZDM, TZPI, and BF1 were able to displace Ͻ50% of the bound Thio T (Fig. 3B). Full displacement curves for the IMPY ligands could not be accurately obtained due to compound insolubility at higher concentrations, but it is apparent that they are displacing Ͼ50% of the Thio T and that their K i values are consistent with their K d2 values. The ability of the ligands TZDM, TZPI, and BF1 to partially displace Thio T may be a consequence of the spatial arrangement of higher density BS2 relative to BS1 on the A␤ fibrils such that they periodically either partially or completely overlap with BS1.
FRET Measurements Indicate That BS1 and BS2 Are Independent Sites on the A␤ Fibril-Further supporting evidence for the spatial separation but close proximity of BS1 and BS2 was obtained from fluorescent resonance energy transfer (FRET) measurements. The differing fluorescent properties of BF1 (BS2 probe, excitation/emission maxima 340:440 nm) and Thio T (BS1 probe, excitation/emission maxima 450:480 nm) were again exploited, but this time as a potential donor-acceptor pair.
By using an excitation wavelength of 340 nm, which specifically excites BF1, the emission spectra of A␤-(1-40) fibrils incubated with either Thio T or Thio T/BF1 were recorded. Calculation of the difference spectrum between the samples demonstrated the presence of an emission maximum at ϳ480 nm, corresponding to the expected wavelength for the Thio T emission (Fig. 4). This experiment was repeated by replacing BF1 with BTA-1, which has similar excitation/emission maxima but should directly compete with Thio T for BS1. No fluorescent signal was observed at 480 nm with the BTA-1. These data clearly demonstrate that only BF1 is able to efficiently transfer energy to Thio T consistent with BS1 and BS2 representing spatially independent ligand-binding sites on the A␤ fibrils.
Radioligand Assays Indicate a Third Class of Binding Site on the A␤ Fibrils-In order to ascertain whether the observed differences in binding affinities between the literature and fluorescent assays reported here were because of the different assay formats, the binding of [ 3 H]Me-BTA-1 to A␤-(1-40) fibrils was determined by using a filter binding assay. The resulting binding isotherm was consistent with the presence of a single high affinity binding site on the polymer with an apparent K d of 4.2 nM and a maximal binding (B max ) of 0.13 when measured at 40 nM A␤ monomer concentration (Fig. 5A). The K d value is slightly higher than the previously reported IC 50 value for this compound of 20 nM (7); and the low B max value, equating to ϳ1 ligand-binding site per 300 A␤-(1-40) monomers, is consistent with that found for other similar compounds such as 6-OH-BTA-1/PIB and BTA-1 (8).
The data from this set of experiments were consistent with the identification of the previously described very low density site on A␤-(1-40) fibrils. Given that the ligand-binding characteristics of this site were clearly differentiated from BS1 and BS2, it is suggested that this represents a third class of binding site, termed BS3.
In competitive binding assays, all of the ligands tested were able to displace [ 3 H]Me-BTA-1 in a manner consistent with a single class of binding sites on the A␤ fibril (Fig. 5B). The K i values derived from these assays were significantly lower than the values derived from both the FLINT2 and Thio T fluorescent competition assays. However, again the relative order of binding affinities was conserved between the radioligand and fluorescent assays (Table I).  In addition, the radioligand competition assays have allowed a direct comparison, under a standard set of conditions, of ligand binding to this low density site on the A␤-(1-40) fibrils. The K i values from the current set of experiments are broadly in agreement with the previously published values for these compounds both in terms of potency and rank order of binding affinity.
The inability to detect ligand binding at BS3 by using fluorescent techniques may be due to a number of related factors. The relative contribution of this site to a binding signal (whether FLINT or anisotropy) would be small relative to the more densely packed BS1 and BS2. By assuming that a ligand produced an identical FLINT increase when bound to any of the sites, the observed signal from BS3 would be a tenth or a hundredth smaller than that observed at BS1 and BS2, respectively. In addition, it is possible that ligand binding to BS3 may significantly quench or produce no change in ligand fluorescence making it difficult to detect by either FLINT or anisotropy measurements.
Evidence That BS1, BS2, and BS3 Are Uniformly Distributed on the A␤ Fibrils-Although the FRET data were able to demonstrate that BS1 and BS2 were in close proximity on the same A␤ fibril, the radioligand data do not directly imply that BS3 is also distributed on the same population of fibrils. To address this issue, we separated the A␤ fibrils by density gradient ultracentrifugation using a method adapted from Ward et al. (27), and we assayed the resulting fractions for A␤ immunoreactivity and thioflavin T and [ 3 H]Me-BTA-1 binding (Fig. 6).
Based on the immunoreactivity of the anti-A␤ antibody 6E10, the centrifugation step split the fibrillar preparation into two peaks centered on fractions 3 and 5. The vast majority (ϳ80%) of the signal from the blot is present in fraction 3, and this is consistent with the expected high molecular mass of the A␤ fibrils and similar to that found previously using this method (27). Fraction 5 is a small shoulder on the main peak of A␤ immunoreactivity indicative of a second population of high molecular mass species in the fibrillar preparation, although of lesser mass than fraction 3.
The ligand binding properties of both Thio T (BS1 probe) and [ 3 H]Me-BTA-1 (BS3 probe) to these fractions display an identical pattern both to each other and also to the signal obtained from the A␤ immunoblot. These findings support the idea that the ligand-binding sites are uniformly distributed across the A␤ fibril preparation and are not restricted to morphologically distinct subpopulations of polymers. DISCUSSION The development of potentially useful PET ligands for the detection of SPs is being widely pursued because of the clinical importance such imaging would have (6). However, until now, the binding properties of these prospective PET ligands to A␤ fibrils have not been rigorously examined, raising doubts over the true significance of the clinical investigations performed to date. We have developed and performed a range of assays that, for the first time, have allowed a full analysis of the binding of a series of Thio T derivatives to A␤-(1-40) fibrils in vitro. Our data are consistent with the presence of a previously described binding site (BS3) to which all of the compounds bind with high affinity, but it also demonstrates the presence of two additional high affinity and much more abundant binding sites (BS1 and BS2). The finding of these additional binding sites on the A␤-(1-40) fibrils may help resolve issues surrounding the nature of the signal detected in vivo by PET with these ligands.
A number of lines of evidence support the presence of multiple ligand-binding sites on the A␤-(1-40) polymers. First, both the K d and K i values from two independent fluorescencebased assay formats are significantly higher than the values reported from previous and the currently described radioligand binding assays. This provides direct evidence of additional li- gand-binding sites on the A␤-(1-40) fibrils. In addition, the FLINT assays also demonstrated that the binding site densities of BS1 and BS2 are significantly higher than that of BS3.
Finally, the selectivity of compounds containing a halogen substitution on their aromatic ring systems (TZDM/TZPI/BF1) provides a structural discriminator for BS1 and BS2. The ability of BS3 to accommodate both halogen-and nonhalogen-containing structures suggests a hybrid-binding site that occurs periodically only when BS1 and BS2 are in complete register on the fibril. However, the significantly different binding constants for ligands between BS1/2 and BS3 would argue that the latter site is a spatially distinct periodic feature of the polymer. A schematic demonstrating the potential arrangement of binding sites on the A␤-(1-40) fibril is shown in Fig. 7. This model proposes that all three types of binding sites are present together on the majority of A␤ fibrils in the polymerized mixture and takes into account the FRET data demonstrating the close spatial proximity of BS1 and BS2 and the congruent distribution of fluorescence and radioligand-binding sites following fractionation of polymers by density gradient centrifugation.
A high resolution structure of amyloid fibrils formed from A␤ peptides is not available, making it difficult to associate the ligand-binding sites with structural features on the polymer. The increase in the intrinsic fluorescence of the ligands is indicative of compound binding into a hydrophobic pocket on the polymer and suggests that BS1 and BS2 are associated with the two hydrophobic clusters (residues 17-21 and 30 -40) on the A␤ peptide (reviewed in Ref. 28).
The presence of multiple classes of binding sites on the A␤ fibrils is also supported by two studies using the ligand FDDNP. The first study using a radioligand filter binding assay indicated that [18F]FDDNP was not displaceable by Thio T from A␤-(1-40) fibrils (29). However, a recent study (30) demonstrated that, in a fluorescent competition assay using Thio T and A␤-(1-40) fibrils (in an essentially identical format to that described here), FDDNP was able to displace Thio T. These findings are consistent with the presence of both independent and common/overlapping binding sites on the fibrils.
The increase in the intrinsic fluorescence of Thio T has been exploited through its use as a histological dye in the postmortem diagnosis of AD (31,32). Our findings would suggest that the signal obtained in the post-mortem tissue is primarily associated with its interaction with BS1 and would also suggest that the fluorescent signal associated with the use of these Thio T derivatives in histological studies, when the compounds are used at relatively high concentrations (Ͼ100 nM), is not from BS3 but primarily from the large increase in FLINT associated with the higher density BS1 and BS2.
An approximate rule for the success of a radioligand in in vivo imaging experiments makes use of the term binding potential (BP) which is the sum of B max /K d ; values Ͼ10 generally produce a sufficient signal for PET studies (8). The in vivo value of B max necessary for the calculation of the BP has not been determined, however, Nä slund et al. (17) have calculated the regional distribution of total A␤-(1-40) and A␤-(1-42) levels (soluble and plaque) in formic acid extracts of post-mortem brains from both control and AD cohorts. Their findings suggest that levels of total A␤ load approached 1-3 M in the cortical regions of AD brains. By using the mid-point of the data set together with the K d of 4.2 nM for [ 3 H]Me-BTA-1, a BP of ϳ500 is obtained. However, when the in vitro binding stoichiometry of the radioligand binding to BS3 is taken into account, the BP is reduced to ϳ2, which is unlikely to generate a significant PET signal in vivo. Although higher B max values have been reported for BTA-1 and PIB binding to AD brain homogenates, the levels of A␤ peptides in this study were not explicitly measured (8).
One key question that arises from the present set of studies is as follows: what might be relative potential contributions of BS1 and BS2 compared with BS3 in the signal obtained from a PET study? Using BTA-1 (K i 200 nM), BF1 (K d 110 nM), and Me-BTA-1 (K d 4.2 nM) as model compounds for BS1, BS2, and BS3 and taking into account the stoichiometry of these sites (1:35, 1:4, and 1:300), the levels of A␤ peptides required to produce a BP of 10 are 70, 4, and 12 M respectively. The values for all of the compounds are clearly at the top end of the reported concentrations for A␤ peptides but suggest that BS2 may make a more significant contribution than BS1 and potentially BS3 to the observed PET signals.
These data also suggest that the design of higher affinity compounds targeting BS2 may result in improved PET ligands for imaging plaque deposits. In addition, the radioligand assays currently widely in use for compound screening and selection may be targeting the "wrong" site on the fibrils, the low density BS3, and that fluorescence-based screens, such as those using Thio T, may identify compounds with more desirable properties for in vivo imaging studies.
In conclusion, our data support a complex model for the binding of PET radioligands being designed for tracking SPs in AD, with the presence of multiple ligand-binding sites on A␤ fibrils. The development of therapy for AD is presently hampered by the fact that there is no sensitive test of the progression and regression of the underlying pathology before cognition is affected (6). Understanding the properties of A␤ fibrils will assist in the more rational design of more effective specific PET radioligands to track the progression of this disease and accelerate the development of treatments.  7. A model of binding sites on A␤ fibrils for the Thio T class of ligands. BS3 is the previously described independent low density site. BS1 and BS2 are the newly identified higher density sites that are independent from BS3. The arrangement of BS1 and BS2 schematically reflects how differences in their relative polymer density can lead to the sites periodically occurring in close proximity and/or partially overlapping. The short and long arrows indicate how both displacement and FRET can occur between BS1 and BS2, respectively.