Selected Non-steroidal Anti-inflammatory Drugs and Their Derivatives Target γ-Secretase at a Novel Site

γ-Secretase is a multi-component enzyme complex that performs an intramembranous cleavage, releasing amyloid-β (Aβ) peptides from processing intermediates of the β-amyloid precursor protein. Because Aβ peptides are thought to be causative for Alzheimer's disease, inhibiting γ-secretase represents a potential treatment for this neurodegenerative condition. Whereas inhibitors directed at the active center of γ-secretase inhibit the cleavage of all its substrates, certain non-steroidal antiinflammatory drugs (NSAIDs) have been shown to selectively reduce the production of the more amyloidogenic Aβ(1–42) peptide without inhibiting alternative cleavages. In contrast to the majority of previous studies, however, we demonstrate that in cell-free systems the mode of action of selected NSAIDs and their derivatives, depending on the concentrations used, can either be classified as modulatory or inhibitory. At modulatory concentrations, a selective and, with respect to the substrate, noncompetitive inhibition of Aβ(1–42) production was observed. At inhibitory concentrations, on the other hand, biochemical readouts reminiscent of a nonselective γ-secretase inhibition were obtained. When these compounds were analyzed for their ability to displace a radiolabeled, transition-state analog inhibitor from solubilized enzyme, noncompetitive antagonism was observed. The allosteric nature of radioligand displacement suggests that NSAID-like inhibitors change the conformation of the γ-secretase enzyme complex by binding to a novel site, which is discrete from the binding site for transition-state analogs and therefore distinct from the catalytic center. Consequently, drug discovery efforts aimed at this site may identify novel allosteric inhibitors that could benefit from a wider window for inhibition of γ (42)-cleavage over alternative cleavages in the β-amyloid precursor protein and, more importantly, alternative substrates.

␥-Secretase is a multi-component enzyme complex that performs an intramembranous cleavage, releasing amyloid-␤ (A␤) peptides from processing intermediates of the ␤-amyloid precursor protein. Because A␤ peptides are thought to be causative for Alzheimer's disease, inhibiting ␥-secretase represents a potential treatment for this neurodegenerative condition. Whereas inhibitors directed at the active center of ␥-secretase inhibit the cleavage of all its substrates, certain non-steroidal antiinflammatory drugs (NSAIDs) have been shown to selectively reduce the production of the more amyloidogenic A␤(1-42) peptide without inhibiting alternative cleavages. In contrast to the majority of previous studies, however, we demonstrate that in cell-free systems the mode of action of selected NSAIDs and their derivatives, depending on the concentrations used, can either be classified as modulatory or inhibitory. At modulatory concentrations, a selective and, with respect to the substrate, noncompetitive inhibition of A␤(1-42) production was observed. At inhibitory concentrations, on the other hand, biochemical readouts reminiscent of a nonselective ␥-secretase inhibition were obtained. When these compounds were analyzed for their ability to displace a radiolabeled, transition-state analog inhibitor from solubilized enzyme, noncompetitive antagonism was observed. The allosteric nature of radioligand displacement suggests that NSAID-like inhibitors change the conformation of the ␥-secretase enzyme complex by binding to a novel site, which is discrete from the binding site for transition-state analogs and therefore distinct from the catalytic center. Consequently, drug discovery efforts aimed at this site may identify novel allosteric inhibitors that could benefit from a wider window for inhibition of ␥ (42)-cleavage over alternative cleavages in the ␤-amyloid precursor protein and, more importantly, alternative substrates.
According to the "amyloid cascade hypothesis" an enhanced production or decreased clearance of toxic amyloid-␤ (A␤) 1 pep-tides is thought to be the cause of Alzheimer's disease (AD) (1). A␤ peptides are processing products (2) of the type I transmembrane protein ␤-amyloid precursor protein (␤APP) (3), which has undergone sequential cleavages by ␤and ␥-secretase enzymes. A common denominator (reviewed by Hardy (4)) for mutations causative of familial AD (FAD) has been revealed, being abnormalities in the metabolism of ␤APP that appear to lead to an elevation of the production of the A␤(1-42) peptide species. This C-terminally elongated A␤ peptide is more prone to aggregation than the shorter and more abundant A␤  species. Consequently, the prevention of A␤ production by inhibiting either of the proteases required for processing of ␤APP is currently viewed as a promising approach toward a therapy for AD. The membrane-bound aspartyl protease ␤-site ␤APPcleaving enzyme 1 (5, 6) is the major ␤-secretase required for the generation of A␤ peptides. ␤-Site ␤APP-cleaving enzyme 1 has been shown to cleave within the ␤APP ectodomain to generate membrane-bound ␤APP processing intermediates (␤APP C-terminal fragments), a prerequisite for the release of A␤ peptide by ␥-secretase as the final processing step. ␥-Secretase appears to be an unusual aspartyl protease with loose substrate specificity (7) and cleaves its substrates approximately in the middle of their transmembrane domains at a specific position relative to the membrane bilayer (8). Intramembranous substrate cleavage at multiple positions leads to the generation of A␤(1-42) and A␤  and various Cterminally truncated A␤ peptides (9,10). Presenilin 1 and 2 (11,12) are homologous polytopic membrane proteins that are the most likely candidate polypeptides to form the catalytic center of ␥-secretase. Mutations in presenilins account for the majority of cases of familial AD (13); combined with results obtained from presenilin knock-out (14), mutagenesis (15), and photoaffinity labeling studies (16 -18), compelling evidence for a catalytic function of presenilins has been generated. Apparently, presenilins alone cannot mediate ␥-secretase activity and require co-expression, together with the transmembrane proteins nicastrin (19), anterior pharynx defective 1 (20), and presenilin enhancer 2 (21), to reconstitute ␥-secretase activity (22)(23)(24).
Considerable effort has been put into the development of potent ␥-secretase inhibitors for the treatment of AD because they have the potential to block the generation of all amyloidogenic peptides from ␤APP-derived substrates (10). This class of compound suffers from the disadvantage that it does not * 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 U.S.C. Section 1734 solely to indicate this fact. discriminate between the cleavages of alternative substrates for ␥-secretase, such as the Notch receptor (25,26). Consequently, it has been demonstrated that ␥-secretase inhibitors interfere with the Notch signaling pathway, which can among other responses lead to a repression of thymocyte development (27,28). In contrast, a distinct group of non-steroidal antiinflammatory drugs (NSAIDs) has been shown to selectively reduce the production of the A␤(1-42) peptide (29). Although the production of the potentially less amyloidogenic A␤  peptide is concomitantly increased, claims have been made that neither A␤(1-40) peptide production nor the Notch receptor cleavage are adversely affected (29,30). To determine the underlying mechanism of this selective inhibition, we have profiled representative NSAID-like compounds in biochemical assays monitoring various aspects of ␥-secretase enzyme function. Taken together, our results indicate that NSAID-like compounds can act as allosteric inhibitors by directly targeting the presenilin-dependent ␥-secretase complex at a novel site discrete from the binding site for transition-state analog ␥-secretase inhibitors.

EXPERIMENTAL PROCEDURES
Materials-Sulindac sulfide, sulindac sulfone (Calbiochem), and Rflurbiprofen (Sigma Aldrich) were obtained from commercial sources. The synthesis of Merck A (31, 32) has been described, and the tritiated version [ 3 H]-Merck A was prepared as follows. A solution of FMOC-4,5dehydro-Leu-OH (0.50 g), L-phenylalaninamide, 1-(3-dimethylaminopropyl)-3-ethylcarbodiimide hydrochloride (0.32 g), and 1-hydroxybenzotriazole hydrate (0.23 g) in N,N-dimethylformamide (5 ml) was stirred at room temperature for 2 h. The reaction mixture was diluted with ethyl acetate, washed with aqueous citric acid, aqueous sodium bicarbonate and brine, dried (MgSO 4 ), filtered, and evaporated in vacuo. The resulting solid was treated with 50% piperidine-N,N-dimethylformamide solution, stirred for 10 min, and then evaporated in vacuo. A portion of the resulting amine (253 mg) was treated with 2R-benzyl-5S-tert-butoxycarbonylamino-4R-(tert-butyldimethylsilanyloxy)-6-phenyl-hexanoic acid (33) (485 mg), 1-(3-dimethylaminopropyl)-3-ethylcarbodiimide hydrochloride (191 mg), 1-hydroxybenzotriazole hydrate (136 mg), and N,N-dimethylformamide (5 ml) and stirred for 2 h. The reaction mixture was diluted with ethyl acetate, washed with aqueous citric acid, aqueous sodium bicarbonate and brine, dried (MgSO 4 ), filtered, and evaporated in vacuo. Purification by column chromatography yielded 550 mg (76%) of the amide as a white solid. A portion of the foregoing amide (250 mg) was treated with tetrabutylammonium fluoride (1.0 M in tetrahydrofuran, 4 ml) and stirred overnight. The reaction mixture was diluted with ether, water, and citric acid solution. The resulting precipitate was collected by filtration, washed with water, ether, and dried to give the alkene (200 mg, 90%). A portion of this compound (3 mg, 4.48 ϫ 10 Ϫ3 mmol) was dissolved in N,Ndimethylformamide (1 ml) and mixed with 10% palladium on carbon (5 mg). The reaction mixture was degassed thoroughly, cooled to Ϫ 78°C, and then stirred with carrier-free tritium gas at room temperature for 2 h. Upon the completion of the reaction, the unreacted tritium gas was removed, and the catalyst was filtered off using a Whatman auto-vial syringeless filter. Any labile tritium in the filtrate was removed by the means of exchanging with absolute ethanol (10 ml), and the solvents were evaporated in vacuo. The residue was subjected to high-performance liquid chromatography purification on Zorbax RX-C8 at 220 nm, eluted with 48% aqueous (0.1% trifluoroacetic acid) and 52% acetonitrile isocratically in 30 min at 5 ml/min. The combined fractions of high-performance liquid chromatography collection was concentrated by passing through a Sep-Pak C 18 cartridge to afford 79.7 mCi of [ 3 H]-Merck A in 10 ml of ethanol (specific activity, 97.2 mCi/mg).
Solubilized ␥-Secretase Enzyme Assay-The generation of A␤(1-40) and A␤  peptides in CHAPSO-solubilized ␥-secretase enzyme preparations using an exogenous recombinant C100Flag substrate has been described previously (34). Essentially, ␥-secretase enzyme from SH-SY5Y neuroblastoma P2 cell membranes was CHAPSO-solubilized according to Beher et al. (32) in 1% (w/v) CHAPSO, 50 mM 2-[Nmorpholino]ethanesulfonic acid-NaOH, pH 6.0, 0.15 M NaCl, 5 mM MgCl 2 , and 1ϫ EDTA-free protease inhibitor mixture (Roche Molecular Biochemicals) and adjusted with the same buffer without CHAPSO to give a final detergent concentration of 0.5% CHAPSO (w/v). For the in vitro ␥-secretase reaction, 40 l of solubilized enzyme (0.6 -0.7 mg/ml protein) were incubated for 90 min at 37°C with recombinant C100Flag substrate at concentrations indicated in the figure legends and in the presence of, additionally, 20 mM HEPES, pH 7.3, 2 mM EDTA, and 0.1% bovine serum albumin in 100 l of final volume. A␤ peptides were quantified by an electrochemiluminescence assay in a 96-well plate format (Origen M-Series analyzer, Igen) as described (25) using 25 l of the reaction for A␤  and 50 l for A␤(1-42) detection. Nonspecific background was defined by the signal obtained when the assay was performed in the presence of 10 M Merck A, a specific ␥-secretase inhibitor (31), and compared with Me 2 SO vehicle controls (1% v/v). Dose-response experiments for various inhibitors were performed as indicated in the figure legends in the presence of 1% final Me 2 SO concentrations. For enzyme kinetic studies, peptide standards containing synthetic human A␤  or A␤(1-42) peptides (California Peptide Research Inc., Nappa Valley, CA) were diluted under the same buffer conditions as the enzyme reaction.
Radioligand Binding Assays-All radioligand binding experiments were performed using CHAPSO-solubilized SH-SY5Y neuroblastoma P2 cell membranes prepared as above. For radioligand binding, 10 g of solubilized enzyme was incubated in the presence of In Vitro Generation of AICD and A␤ Peptides in Membranes-P2 membranes from SH-SY5Y neuroblastoma cells overexpressing the ␥-secretase substrate precursor SPA4CT (37) (SP-LEC99) were prepared according to Beher et al. (32). For in vitro generation of ␤APP intracellular domain (AICD) and A␤ peptides, 240 g of membrane protein was incubated for 2 h at 37°C with increasing amounts of NSAID-like compounds as indicated in the figure legends, in 120 l of 0.5% CHAPSO, 20 mM HEPES, pH 7.3, 2 mM EDTA, and 1ϫ EDTA-free protease inhibitor mixture (Roche Molecular Biochemicals). Me 2 SO vehicle (0.5% v/v) or 10 M Merck A was used as a control, and for quantitation of de novo production of A␤(1-40) and A␤(1-42) peptide, 0.5 l and 5 l aliquots of the corresponding reactions, respectively, were analyzed by an electrochemiluminescence assay as described for the solubilized ␥-secretase enzyme assay above. AICD generation in 15 l of reaction was quantified by Western blot analysis using the antibody R7334 (25), which is directed at the C terminus of ␤APP. Primary antibodies were detected with Alexa Fluor 680-conjugated goat antirabbit F(abЈ) 2 fragments (Molecular Probes, Eugene, OR) and quantified on a LI-COR Odyssey infrared imager (LI-COR Biosciences Inc., Lincoln, NE).
Surface-enhanced Laser Desorption/Ionization Time-of-Flight Mass Spectrometry-Immunocapture of A␤ peptides generated in the solubilized enzyme assay from recombinant C100Flag substrate was performed analogous to methods described previously using monoclonal antibodies 6E10 or W0-2 (10). Briefly, after incubation for 90 min at 37°C, 2.5 ml of the in vitro ␥-secretase reaction was adjusted to 0.5% (v/v) Triton X-100, 25 mM HEPES, pH 7.3 (10 ml final volume). After centrifugation for 10 min at 4,000 ϫ g, the supernatant was incubated overnight at 4°C with the antibody-coupled surface-enhanced laser desorption ionization protein chip (Ciphergen Biosystems, Fremont, CA), which was processed for surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (10).

RESULTS
Selected NSAID-like Compounds Act as Noncompetitive Inhibitors of A␤ Peptide Formation-To profile the NSAID sulindac sulfide and the NSAID-like derivative R-flurbiprofen for their inhibitory potential toward ␥-secretase enzyme activity, we evaluated these compounds in an exogenous substrate enzyme assay (16). Because the assay uses CHAPSO-solubilized membranes from human neuroblastoma cells as a source of enzyme and a recombinant C100Flag substrate, it allows a direct examination of the effects of compounds on enzyme catalysis. Furthermore, the cell-free nature of the assay allowed the use of these compounds at higher concentrations (Ͼ100 to 300 M), where normally decreased cell viability would be ob-served. 2 Sulindac analogs were chosen because sulindac sulfide, but not its close analog sulindac sulfone (serving as a less potent control; data not shown), has been described to selectively lower A␤(1-42) production in cells (29). R-Flurbiprofen, the cyclooxygenase-inactive enantiomer of the NSAID S-flurbiprofen, was chosen as a structurally distinct agent described to selectively lower A␤(1-42) production (38,39).
Both sulindac sulfide and R-flurbiprofen inhibit ␥(42)-secretase activity (Fig. 1, A and B tivity was also observed, however, which suggests that these compounds produce a full ␥-secretase inhibition. That appears to be distinct mechanistically from the selective modulation of A␤(1-42) production observed at lower concentrations.
To investigate the mode of inhibitor action, enzyme kinetic analyses were performed with R-flurbiprofen (Fig. 1, C-E). The data from these studies revealed that the inhibition of both ␥(42) and ␥(40) activities cannot be overcome by increasing the substrate concentration in the assay. The V max for both ␥(42)and ␥(40)-secretase activities were considerably decreased (the latter one at the highest concentration used; Fig. 1C) in the presence of increasing concentrations of R-flurbiprofen (Fig. 1,  C and D). This is confirmed by Lineweaver-Burk transformation of the R-flurbiprofen A␤(1-42) inhibition curve (Fig. 1E), which reveals a change in V max without affecting the K m . In accordance with Takahashi and colleagues (40), similar results were obtained for sulindac sulfide (supplementary information). Such a mode of noncompetitive inhibition with respect to substrate has been observed previously for structurally diverse ␥-secretase inhibitors resembling aspartyl protease transition state analogs or other small molecules (41). Combined, this suggests that both NSAID-like inhibitors and prototypical ␥-secretase inhibitors, such as the transition state analog inhibitor L-685,458, target a site independent of the substrate binding site of the enzyme.
Inhibitor Profile of NSAID-like Compounds Mimics That of ␥-Secretase Inhibition-␥-Secretase cleavage of ␤APP-derived substrates results in the production of various C-terminally truncated A␤ peptides (9,10). To investigate whether the inhibition of A␤(1-40) production observed in the cell-free enzyme assay (Fig. 1, A and B) is accompanied by an inhibition of the production of all A␤ peptide species, we performed mass spectrometric analyses of the recombinant substrate enzyme reactions (Fig. 2). R-Flurbiprofen was chosen because this compound is less selective compared with sulindac sulfide (Fig. 1, A  and B). The spectra obtained after treatment with 1 mM Rflurbiprofen were similar to the samples incubated in the presence of 10 M of the potent ␥-secretase inhibitor Merck A. For both compounds, a decrease of the production of all C-terminally truncated A␤ peptides was observed, with Merck A being more potent as expected. This included the A␤(1-38) peptide that was shown previously to be up-regulated when NSAIDlike compounds are used at concentrations where selective inhibition of A␤(1-42) peptide production occurs (29). In good accordance with this, a reduction of A␤  in the presence of 0.3 mM R-flurbiprofen was accompanied by an enhanced production of C-terminally truncated A␤ peptides such as A␤(1-38) and A␤(1-37) ( Fig. 2; average peak intensities relative to the Me 2 SO control spectra obtained from W0-2 and 6E10 captures were: A␤(1-42), Ϫ15.7%; A␤(1-40), ϩ0.5%; A␤(1-38), ϩ25.5%; and A␤(1-37), ϩ108.5%).
Conflicting reports have been published on sulindac sulfide regarding the inhibition of ␤APP intracellular domain (AICD) generation, which is a direct measure of ⑀-cleavage of ␤APPderived processing products in proximity to the cytosolic surface of the membrane. Whereas Takahashi and colleagues (40) observed an inhibition of AICD generation after an initial elevation at low concentrations, Weggen and colleagues (30) reported that sulindac sulfide was not able to inhibit AICD generation. To address this issue, we modified our cell-free ␥-secretase assay (31) to establish an AICD assay that allows the parallel measurement of AICD generation with A␤(1-40) and A␤(1-42) peptide production from the same sample. The results indicate that both R-flurbiprofen (Fig. 3, A and B) and sulindac sulfide (Fig. 3,  C and D) can inhibit the ⑀-cleavage of ␤APP, causing a dose-dependent reduction in the production of AICD at concentrations where selectivity is lost and a concomitant inhibition of A␤(1-40) generation is observed. Thus, at modulatory concentrations these compounds only affect A␤(1-42) production and not the genera- mM R-flurbiprofen, respectively, using monoclonal antibodies 6E10, W0-2, or purified nonimmune mouse IgG. Captured peptides were directly analyzed by surface-enhanced laser desorption/ionization timeof-flight mass spectrometry. All spectra were normalized to the average intensity height of the peak for the single charged bovine insulin peptide species (m/z, 5733.6) and calibrated internally using the single and double positively charged species of bovine insulin. A higher magnification of the 4200 -4300 mass range is shown for the Me 2 SO control and 0.3 mM R-flurbiprofen reaction. Note that all peptides derived from recombinant C100Flag substrate contain an additional methionine residue at their N-termini, and the last C-terminal amino acid indicates the last amino acid of the native A␤ sequence. Asterisks indicate A␤ peptides oxidized at methionine Ϫ1 or 35. Masses of these A␤ peptides are increased by 16 mass units, as expected for methionine sulfoxide derivatives. The spectra shown are representative of at least two independent experiments using monoclonal 6E10 antibody for capture. Identical spectra were obtained using monoclonal W0-2 antibody for capture, and none of the peaks labeled was captured by nonimmune mouse IgG (data not shown). tion of A␤  or AICD, but at higher inhibitory concentrations the production of all these species is reduced, in accordance with classical ␥-secretase inhibition. The ⑀-cleavage of ␤APP is equivalent to the ␥-secretase-mediated S3 cleavage of the Notch receptor (42), which is a critical event in the Notch signaling cascade (43). Therefore, it will be of future interest to see whether these compounds can produce a similar inhibition of the cleavage of a Notch substrate under comparable conditions. NSAID-like Inhibitors and Noncompetitive Antagonism of Inhibitor Binding-Aspartyl protease transition-state analog inhibitors have been shown to label presenilins (16,17) and directly interact with the critical transmembrane aspartates, forming the presumed active site (44). A tritiated version of the transition-state analog inhibitor Merck A (31) has been successfully exploited to study the relationship of various inhibitor binding sites on solubilized enzyme (35). To gain information on the inhibition mechanism of certain NSAID-like compounds, radioligand binding competition studies were performed using tritiated Merck A (Fig. 4). Primary evaluation of the assay demonstrated that the binding of [ 3 H]-Merck A to solubilized enzyme displays saturable kinetics and was specific because it could be blocked by an excess of unlabeled parent compound Merck A (Fig. 4A). Scatchard transformation of the same specific binding data revealed a straight downward line suggestive of single-site binding of the radioligand at the concentration range used in the assay (Fig. 4B). Sulindac sulfide (Fig. 4C) appeared to be the most potent compound at displacing [ 3 H]-Merck A (K i ϭ 86.6 M). Potencies of R-flurbiprofen and the control sulindac sulfone were lower (K i Ͼ1 mM), with the latter, as expected, being the least potent compound. To exclude that an interaction of these acidic compounds with the charged filters is the cause for competition, both benzoic acid and the NSAID N-acetylsalicylic acid (aspirin) were analyzed in the same assay. These compounds have pK a values that are comparable with the previously characterized NSAID derivatives, but in contrast, neither displaced the radioligand [ 3 H]-Merck A from the enzyme (data not shown).
To investigate the basis for the competitiveness of sulindac sulfide and R-flurbiprofen, dose-responses for [ 3 H]-Merck A binding were analyzed in the presence of various fixed concentrations of these compounds and the less active control sulindac sulfone (Fig. 4, D-F). A decrease in the maximum binding (B max ) of the transition state analog inhibitor to solubilized enzyme was observed that was insurmountable and thus characteristic of noncompetitive antagonism. As expected, the rank of order reflected the K i values of the NSAID derivatives and was observed with as little as 50 M sulindac sulfide (Fig. 4D). DISCUSSION ␥-Secretase is a prime target for a pharmacological intervention in the disease progression of AD. Whereas classical ␥-secretase inhibitors, which neither discriminate between the cleavages of alternative substrates for ␥-secretase nor the actual cleavage position in ␤APP-derived substrates, have been shown conclusively to target presenilins (16 -18), substantial evidence for such an interaction of NSAID-like inhibitors with the presenilin-dependent ␥-secretase complex has just begun to evolve (40,45). The noncompetitive mode of inhibition observed in solubilized membrane preparations with either R-flurbiprofen or sulindac sulfide demonstrates that recent findings that were restricted to the latter (40) can be extended to structurally diverse NSAID-like inhibitors. In this respect, it appears that NSAID-like inhibitors share similarities with classical nonselective inhibitors (41) by interacting with the enzyme at a site discrete from the initial substrate binding site. In contrast with nonselective inhibitors, NSAID derivatives are characterized by a distinct window of modulation, where a selective inhibition of A␤(1-42) generation is observed. When these compounds are used at higher concentrations, however, a classical ␥-secretase inhibition occurred, as seen by an inhibition of A␤(1-40) generation. This finding is in good accordance with a previous study (40) but does not confirm claims that these compounds are truly selective inhibitors of ␥(42) activity (29,39). In particular, R-flurbiprofen has been a useful compound in our studies because it has a relatively modest window for inhibition of A␤(1-42) over A␤ . Accordingly, the mass spectrometric characterization of the recombinant substrate cleavage unequivocally demonstrated that when used at 1 mM, treatment with R-flurbiprofen resulted in a full inhibition of the production of all A␤ peptides, including various C-terminally truncated species, again similar to the profile obtained with a prototypical ␥-secretase inhibitor. Consistent with these observations, ⑀-cleavage of ␤APP, which leads to the generation of AICD, was also inhibited by sulindac sulfide and R-flurbiprofen. This inhibition occurred at concentrations where the selectivity for inhibition of A␤(1-42) over A␤  is lost, which indicates that an inhibition of A␤  production in this assay reflects a complete ␥-secretase inhibition. It is noteworthy that the cell-free AICD/␥-secretase assay simultaneously detects various ␥-secretase cleavage products generated in the same sample under identical conditions. Thereby interassay variations in inhibitor potencies that are commonly observed, even for potent ␥-secretase inhibitors, can be avoided. It is likely that these variations, in addition to the different compound concentration ranges chosen for the actual assays, might explain to some extent the conflicting evidence described in the literature (30,40).
Although the cell-free enzyme assay data suggested a direct interaction of NSAID-like inhibitors with the ␥-secretase complex, the radioligand competition data provide, for the first time, direct evidence for such a mechanism. In this paradigm, the transition state analog inhibitor served as a probe for the integrity of the active center because it binds to the catalytic residues in ␥-secretase (44). It is noteworthy that this type of competition can only be detected in the radioligand displacement assay because NSAID derivatives do not appear to compete for photolabeling of presenilin 1 and rather increase the capture efficiency in affinity precipitation studies. 3 Nevertheless, the noncompetitive antagonist behavior of sulindac sulfide and R-flurbiprofen in the radioligand binding paradigm suggests that these compounds introduce conformational changes into the complex that affect the binding of the radioligand itself or less likely the retention of the enzyme complex on the charged filters. This allosteric modulation of ␥-secretase by NSAID derivatives could result either in a direct change of enzyme conformation or, for example, the oligomerization state of the complex. The latter is conceivable because ␥-secretase has been proposed recently to exist as a dimer (46). How this leads to a change of enzyme cleavage specificity from position 42 to 38 before full inhibition of the enzyme is not yet known. However, the degree of fractional enzyme occupancy of a given dimeric or multimeric assembly of the active enzyme complex can potentially explain this unconventional pharmacological profile. Conditions where only a certain fraction of allosteric sites is occupied in a single enzyme dimer/multimer could lead to a modulation of cleavage specificity. Further saturation of remaining allosteric sites at higher compound concentrations would subsequently lead to a full enzyme inhibition. This would even be applicable to in vivo occupancy because these NSAID-like compounds appear to selectively reduce A␤  production in vivo after repeated and subchronic dosing in the Tg2576 transgenic mouse model for amyloidosis (29,39). At a dose of 50 mg/kg/day, R-flurbiprofen reduced formic acid-extractable brain A␤(1-42) by 60%, with drug concentrations reaching 2.5 M in the brain 2 h after the last dose. Considering the poor potencies observed consistently in cellular and cellfree assays, this presents a large discrepancy between in vivo efficacy and in vitro potency. A variety of arguments, including the selective accumulation of these compounds in ␥-secretaseenriched compartments, have been discussed (39), and because the degree of fractional enzyme occupancy in brain is unknown, further studies are needed to address this question.
Comparing cell-free enzyme inhibition and radioligand displacement data, the IC 50 for inhibition of A␤(1-42) generation and the K i for displacement of the transition state analog inhibitor by sulindac sulfide are within a similar range (IC 50 for ␥(42) ϭ 34 M; K i ϭ 86.6 M). Because the assays use comparable conditions, we speculate that the inhibition of transition state analog inhibitor binding may occur at concentrations where a selective inhibition of A␤(1-42) production occurs. This is conceivable because the same pharmacological entity generates A␤(1-42) and A␤ , and therefore one would predict that at half-maximal modulation of A␤  production, half of the total number of enzyme molecules is occupied. The NSAID-like compounds act as noncompetitive antagonists for the active site-directed inhibitor, and therefore this should yield a half-maximal inhibition of radioligand binding.
Further studies will be required to elaborate the exact mechanism of ␥-secretase inhibition by NSAID-like compounds, but because the transition state analog radioligand binds to the active site, it is reasonable to conclude that NSAID-like inhibitors target an independent allosteric site on the ␥-secretase enzyme complex. The active site is formed by presenilins (15,16), and the ␥-secretase enzyme complex consists of four components, presenilins, nicastrin, presenilin enhancer 2, and anterior pharynx defective 1 (22)(23)(24). Therefore, it still remains to be elucidated which of those polypeptides is the molecular entity targeted by NSAID derivatives. It can be predicted that the development of more potent NSAID-like inhibitors will facilitate the synthesis of novel molecular probes that are needed to address this question. This will be a challenge because the information currently available on the structureactivity relationships for selective inhibition of the ␥(42)-enzyme activity is limited. Only certain classes of NSAIDs and derivatives thereof appear to be associated with this phenomenon (29,39), and their potencies are modest. Furthermore, these compounds do not share any of the structural features that are found in potent ␥-secretase inhibitors (reviewed by Harrison and Beher (47)), and not all NSAIDs modulate the enzyme. Thus far, it is only known that the activity for selective inhibition of A␤  can be separated from cyclooxygenase inhibition, because inactive derivatives of NSAIDs such as R-flurbiprofen can still act as ␥(42)-secretase inhibitors.
In conclusion, we interpret the data obtained in this study as meaning that NSAID-like inhibitors can directly introduce allosteric changes into the presenilin-dependent ␥-secretase complex by binding to a novel site discrete from the active center, thereby preferentially lowering the generation of A␤ . This mode of inhibition is distinct from any effects on cyclooxygenases, or as recently suggested, Rho kinases (48). These findings highlight the possibility that novel NSAID-like inhibitors targeted at the presenilin-dependent ␥-secretase complex can be developed that discriminate between both different ␥-secretase cleavages and other substrates, and thus offer an advantage.