Sulindac Sulfide Is a Noncompetitive γ-Secretase Inhibitor That Preferentially Reduces Aβ42 Generation*

Nonsteroidal anti-inflammatory drugs (NSAIDs) have been known to reduce risk for Alzheimer's disease. In addition to the anti-inflammatory effects of NSAIDs to block cylooxygenase, it has been shown recently that a subset of NSAIDs selectively inhibits the secretion of highly amyloidogenic Aβ42 from cultured cells, although the molecular target(s) of NSAIDs in reducing the activity of γ-secretase for Aβ42 generation (γ42-secretase) still remain unknown. Here we show that sulindac sulfide (SSide) directly acts on γ-secretase and preferentially inhibits the γ42-secretase activity derived from the 3-[(3-cholamidopropyl)dimethylammonio]-2-hydroxy-1-propanesulfonate-solubilized membrane fractions of HeLa cells, in an in vitroγ-secretase assay using recombinant amyloid β precursor protein C100 as a substrate. SSide also inhibits activities for the generation of Aβ40 as well as for Notch intracellular domain at higher concentrations. Notably, SSide displayed linear noncompetitive inhibition profiles for γ42-secretase in vitro. Our data suggest that SSide is a direct inhibitor of γ-secretase that preferentially affects the γ42-secretase activity.

tory profile and inhibits the progression of amyloid deposition in the brains of ␤APP transgenic mice (18 -20). Contrary to these traditional views, however, it has recently been shown that a subset of NSAIDs (ibuprofen, sulindac sulfide (SSide), indomethacin, and R-flurbiprofen) selectively decrease the secretion of A␤42 from cultured cells independently of COX activity and lowers the amount of soluble A␤42 in the brains of transgenic mice (21,22). In cultured cells, the decrease in A␤42 secretion caused by SSide was accompanied by an increase in A␤38 generation, whereas the Notch site-3 cleavage activity to generate Notch intracellular domain (NICD) was not significantly affected. In contrast, other NSAIDs including sulindac sulfone (SSone), which is a metabolite of SSide and inactive for COX, had no significant effect on ␥-secretase activity. These data suggest that some of the NSAIDs may affect the pathogenetic process of AD by directly inhibiting the ␥-secretase activity, causing a shift in the cleavage site. Furthermore, in contrast to the previously developed ␥-secretase inhibitors, the treatment with NSAIDs may be free from side effects caused by inhibiting the intramembranous cleavage of other substrates. However, the molecular mechanism whereby NSAIDs inhibit A␤42 generation, as well as the target protein of NSAIDs to modulate ␥-secretase activity, still remains unclear. Here we analyzed the enzymatic property of ␥-secretase by an in vitro ␥-secretase assay using recombinant substrates and CHAPSOsolubilized membrane as an enzyme source, with special reference to the effect of NSAIDs on intramembranous cleavage of ␤APP and Notch.

MATERIALS AND METHODS
Compounds and Peptides-Synthesis of DAPT (23) was performed by a standard solution phase peptide synthesis utilizing the Cbz protecting group. The detailed synthetic procedure will be reported elsewhere.
NSAIDs used in this study (SSide, Ssone, and naproxen) were purchased from Biomol (Plymouth Meeting, PA) and were dissolved in dimethyl sulfoxide (Me 2 SO). L-685,458 and standard A␤ peptides were purchased from Bachem AG (Bubendorf, Switzerland).
Cell Culture and Treatment by NSAIDs-The expression vectors encoding ␤APP NL in pCEP4 (Invitrogen) and Notch⌬E in pCS2 were provided by Drs. K. Maruyama (Saitama Medical School) and R. Kopan (Washington University), respectively. A stable Neuro2a (N2a) cell line doubly expressing ␤APP NL and Notch⌬E (N2a NL/N) was generated as described previously (24). To analyze the effect of NSAIDs on ␥-secretase activity, N2a NL/N cells were cultured at confluency in Dulbecco's modified Eagle's medium containing 10 mM butyric acid to drive protein expression in the presence of various concentrations of NSAIDs for 48 h. Culture media were collected and subjected to BAN50/BA27 or BAN50/ BC05 ELISAs (25). Immunoblot analysis using C4 (anti-␤APP C terminus, provided from Dr. Y. Ihara (University of Tokyo)) or anti-c-Myc (Roche Applied Science) was performed as described previously (8,9).
Purification of Recombinant Substrates-cDNAs encoding the C-terminal 99 amino acids of human ␤APP or 101 amino acids of mouse Notch 1 fused to FLAG tag at the C terminus and harboring an additional Met at the N terminus was generated by PCR and subcloned into pTrcHis2A (Invitrogen) (C100-FmH and N102-FmH, respectively) (26). A cDNA encoding C100-FmH carrying I716F mutation (11,27) was generated by the long PCR protocol using a cDNA encoding C100-FLAG in pTrcHis2A vector as a template. All constructs were sequenced using Thermo Sequenase TM (Amersham Biosciences) on an automated sequencer (Li-Cor). Recombinant proteins were expressed in Escherichia coli and purified by nickel-chelating affinity chromatography as described previously (26).
Preparation of Solubilized ␥-Secretase Fractions from Cultured Cells-All cell lines were maintained as described previously (25) and grown at confluency. Membrane fractions were prepared as described previously (9,28,29). The membrane pellets were resuspended in HEPES buffer to yield a protein concentration of 5-10 mg/ml and were stored at Ϫ70°C. The membranes were solubilized by 1% CHAPSO (Wako, Osaka, Japan) combined with 1 M NDSB-256 (Calbiochem) for 60 min at 4°C and centrifuged at 100,000 ϫ g for 60 min. We defined the supernatant (ϳ5 mg of protein/ml) as the solubilized ␥-secretase fraction, which was stored at Ϫ80°C until use. All procedures were performed at 4°C.
In Vitro ␥-Secretase Assay-In vitro ␥-secretase assay was performed as described previously (26) with some modifications. Each recombinant substrate at defined concentrations was incubated together with the solubilized ␥-secretase fraction (250 g/ml) in 1ϫ ␥ buffer (HEPES buffer containing 0.25% CHAPSO, 5 mM EDTA, 5 mM 1,10-phenanthroline, 10 mg/ml phosphoramidon, Complete protease inhibitor mixture (Roche Applied Science)) with or without ␥-secretase inhibitors (including NSAIDs) at 4 or 37°C for 3 h. Control reactions were performed in the presence of 1% Me 2 SO. The reaction was stopped by boiling the reaction mixtures for 2 min. The samples were centrifuged, and the supernatants were analyzed by BAN50/BA27, BAN50/BC05, BNT77/BA27, or BNT77/BC05 ELISAs for de novo generation of A␤ (25). For the immunoblot analyses of de novo generated peptides, the total proteins in the supernatants were precipitated by trichloroacetic acid and analyzed by immunoblotting with BAN50 (anti-A␤), anti-c-Myc (Roche Applied Science), or anti-FLAG M2 (Sigma). Tris/Bicine/urea high resolution A␤ immunoblot analysis was performed as described previously (30).

RESULTS
Characterization of the ␥-Secretase Activity in Vitro-We first characterized the biochemical and enzymatic properties of the ␥-secretase activity as detected by our in vitro ␥-secretase assay. For this purpose, we generated a recombinant protein substrate C100-FmH, based on the amino acid sequence of the C-terminal fragment of ␤APP fused to FLAG-myc-His 6 tag sequences at the C terminus ( Fig. 1). De novo generation of A␤ peptides from recombinant C100-FmH incubated with membranes of HeLa cells as an enzyme source required the presence of 0.25% CHAPSO, whereas Triton X-100 or SDS abolished the ␥-secretase activity (data not shown), which was consistent with the previous observations (12). Incubation of HeLa cell membranes with wild-type (wt) C100-FmH predominantly generated A␤40, in addition to A␤42 as a minor species ( Fig. 2A). However, the relatively low levels of de novo generation of A␤42 hampered detailed pharmacological and enzymatic analyses of A␤42-generating activity from wt C100-FmH. Thus, we introduced the I716F mutation into the recombinant substrate, which had been described to cause a dramatic increase in A␤42 generation in intact cell-based assays (27). De novo A␤42 generation from I716F mutant (mt) C100-FmH was significantly increased, whereas the production of A␤40 was almost totally abolished, suggesting that mt C100-FmH served as an optimal substrate for ␥ 42 -secretase ( Fig. 2A). These proteolytic activities were recovered from the solubilized membrane fraction by 1% CHAPSO containing 1 M NDSB-256. We next examined the effects of the two well characterized ␥-secretase inhibitors, L-685,458 and DAPT, on A␤ generation in our in vitro assay (12,23,31). Both compounds inhibited A␤-generating activities in a similar, concentration-dependent fashion (Fig. 2B). We then studied the A␤-generating activities in membranes of various cell lines including embryonic fibroblasts derived from PS1/2 double knockout mice (9,32). Consistent with the results of intact cell-based assays, de novo production of A␤ peptides was almost totally abolished in membrane fractions from PS1/2 double knockout cells, and these activities can be immunoprecipitated with antibodies against PS1 ( Fig. 2C and data not shown). These data suggest that recombinant wt and mt C100-FmHs were processed by the endogenous, PS-dependent ␥ 40and ␥ 42 -secretase activities that enzymatically generate A␤40 and -42 polypeptides, respectively. We further analyzed the enzymatic properties of ␥-secretase activities of intact or solubilized membrane fractions from HeLa cells in vitro. The apparent K m value for the processing of either wt or mt substrate to generate A␤40 or A␤42, respectively, by ␥-secretase activities was ϳ0.5 M (Fig. 2D). The progress curve was linear during the 6-h reaction time, and the pH dependence of ␥-secretase activity was broad, ranging from pH 5 to 9 (data not shown). Furthermore, we observed a ␥-secretase-dependent in vitro generation of the C-terminally tagged APP intracellular domain fragments (AICD-FmH) that is produced as the Cterminal counterpart of A␤, by incubation of either wt or mt C100-FmH with solubilized HeLa membranes (Fig. 2E). These data are consistent with the previous reports (12,33,34) on the in vitro ␥-secretase assays using different types of recombinant C100 with or without C-terminal tags.
Intramembranous cleavage of ␥-secretase substrates generates the intracellular domain fragments that are liberated from the membrane and mediate the signal pathways from plasma membrane to nucleus (2). Notably, a proteolytic generation of Notch intracellular domain (NICD) by PS-dependent ␥-secretase is the most well known example, and potential side effects caused by the blockade of Notch pathway by ␥-secretase inhibitors are emerging problems (35,36). To analyze the ␥-secretase activity to generate NICD in vitro, we generated a recombinant N102-FmH substrate composed of a 101-residue fragment of murine Notch1 beginning close to the S2 cleavage site and containing transmembrane domain fused to a FLAGmyc-His C-terminal tag (Fig. 1). After incubation of N102-FmH with the membrane fraction, we observed the appearance of an NICD-like polypeptide migrating slightly faster than N102-FmH, which was diminished by treatment with DAPT in a dose-dependent manner ( Fig. 2F and data not shown). Thus, the recombinant N102-FmH polypeptide is also cleaved by ␥-secretase to generate C-terminally tagged NICD (NICD-FmH) in vitro, which was consistent with the recent report by Wolfe and colleagues (37,38).
Effect of Sulindac Sulfide on ␥-Secretase Activity in Vitro-It has been reported recently (21,22) that a subset of NSAIDs lower A␤42 without affecting Notch processing in cultured cells. To gain more insights into the effect of NSAIDs on APP and Notch processing, we have chosen three NSAIDs, i.e. sulindac sulfide (SSide), sulindac sulfone (SSone), and naproxen, to treat N2a NL/N cells stably coexpressing ␤APP NL and Notch⌬E (24). We confirmed a specific decrease in A␤42 secretion by treatment with 10 -30 M SSide, whereas the secretion of A␤40 as well as Notch processing was not affected (Fig. 3). Treatment with 100 M of SSide caused cell death presumably by inducing apoptosis, resulting in marked decrease in A␤ generation as well as in total protein expression (17). The IC 50 value for A␤42 secretion of SSide was 30.6 Ϯ 2.8 M. SSone and naproxen had no effect either on A␤40 or A␤42 secretion as well as on Notch cleavage up to 100 M.
To examine whether SSide modulates ␥-secretase activity by direct or indirect mechanisms (e.g. altering the trafficking of substrates or enzymes, affecting the secretion or degradation of A␤42 peptides, or modifying the transcription of ␥-secretaserelated genes), we analyzed the in vitro ␥-secretase activity in solubilized membrane fraction in the presence of NSAIDs. We observed an inhibition of ␥ 42 -secretase activity by SSide in a dose-dependent manner. The IC 50 value of SSide for inhibiting ␥ 42 -secretase activity in vitro was 20.2 Ϯ 2.6 M (Fig. 4A). We found a decrease in slope by the increase of the concentration of SSide in the plot of rate against the enzyme concentration, suggesting that SSide is not an irreversible or pseudo-irreversible inhibitor (Fig. 4B). Moreover, when we dialyzed the solubilized ␥-secretase fraction pretreated with SSide against CHAPSO buffer without SSide, ␥-secretase activity was almost totally recovered (Fig. 4C). From these data, it was strongly suggested that the genuine molecular target of SSide is the ␥-secretase complex and that SSide works as a reversible ␥-secretase inhibitor.
In contrast to the results in cultured cells, the application of SSide at low concentrations (1-25 M) caused a transient, but significant, increase in A␤40 generation in vitro (Fig. 4A). Moreover, SSide diminished the de novo generation of A␤40, in addition to that of A␤42, at high concentrations (50 -100 M). Thus, SSide has an inhibition potency against ␥ 40 -secretase activity at high concentrations, whereas it elevates the ␥ 40secretase activity at sub-inhibitory doses. It was reported that the decrease in A␤42 secretion by SSide was accompanied by a dose-dependent increase in A␤38 secretion (21). To determine whether the decrease in A␤42 (plus A␤40) production caused by high concentrations of SSide affects that of A␤38 in vitro, we analyzed the de novo generated A␤ species by high resolution immunoblotting (30). We observed a dose-dependent increase in A␤38 generation in the low concentration ranges (1-25 M) accompanied by a decrease in A␤42 production, although the in vitro generation of A␤ peptides including A␤38 was entirely inhibited by high concentrations of SSide (data not shown). These results suggest that SSide is a bona fide ␥-secretase inhibitor directly affecting the membrane-embedded protease complex, exhibiting distinct inhibitory potencies against A␤38-, A␤40-, and A␤42-generating activities of ␥-secretase.
To characterize further the inhibitory mechanism of ␥ 42secretase activity by SSide, we performed the double-reciprocal plot analysis (Fig. 5A). We found that the K m value remained at a constant level, but the V max value was decreased under increasing concentrations of SSide, suggesting that SSide displayed a noncompetitive inhibition for ␥ 42 -secretase activity. Because transition state analogue ␥-secretase inhibitors (i.e. pepstatin or L-685,458) displayed linear noncompetitive inhibition profiles, a two-binding site model for intramembranous cleavage by ␥-secretase has been proposed (34). In this model, ␥-secretase complex is predicted to harbor a docking/anchoring site of substrates as well as a separate catalytic site. To determine whether SSide affects ␥ 42 -secretase activity by interacting with the catalytic site of ␥ 42 -secretase, we analyzed the inhibition profile of SSide by coincubation with L-685,458, a transition state analogue inhibitor that is expected to occupy the active site of ␥ 42 -secretase. If SSide binds to the docking/ anchoring site, coincubation with SSide and L-685,458 would result in a synergistic inhibition of ␥ 42 -secretase and cause an increase in the inhibition slope. Unexpectedly, however, an addition of L-685,458 had no effect on the slope, raising the possibility that SSide may inhibit the ␥ 42 -secretase activity in a similar mechanism to that of a transition state analogue inhibitor, L-685,458 (Fig. 5B). Thus, SSide would compete for a We next analyzed the effect of SSide on intracellular domain generation from recombinant substrates of ␤APP and Notch in vitro (Fig. 6). We observed that the inhibition kinetics of AICD-FmH generation from C100-FmH by SSide was approximately similar to that of the A␤ generation; the proteolytic activity to release AICD-FmH from wt C100-FmH was increased by treatment with 10 -25 M SSide, whereas it was completely inhib-ited at 100 M. In contrast, AICD-FmH production from mt C100-FmH was inhibited by SSide in a dose-dependent fashion at 10 -100 M. We then analyzed the effect of SSide on NICD-FmH generation from recombinant substrate in vitro. The endoproteolytic cleavage of N102-FmH was inhibited by high concentrations of SSide (250 -500 M), whereas it was not affected by SSide at concentrations up to 100 M, which was consistent with the results obtained in cultured cells (Fig. 3B). These results demonstrate that SSide can inhibit the ␥-secretase activity for Notch as well as for ␤APP, although its inhi- bition potency for Notch is much weaker than that for ␤APP, especially for ␥ 42 -cleavage. DISCUSSION It has been shown that a subset of NSAIDs selectively lowers the secretion of A␤42, although the molecular mechanism whereby NSAIDs affect the ␥-secretase activity remained unclear (21,22). In this study, we established an in vitro ␥-secretase assay using recombinant wild type as well as mutant C100 as substrates and analyzed the effect of NSAIDs. We found that SSide, but not its metabolite SSone nor naproxen, directly inhibits the ␥-secretase activity derived from membrane fractions of HeLa cells in a dose-dependent manner. Moreover, we showed that SSide is a bona fide ␥-secretase inhibitor that has the highest inhibition potency against A␤42-cleaving activity compared with those for A␤38, A␤40, or Notch, with noncompetitive inhibition kinetics.
It has been extensively documented that NSAIDs exhibit various molecular targets, of which the primary target is COX, that converts arachidonic acid to prostaglandins (17). In addition, SSide has been shown to inhibit IB kinase ␤ activity, activate PPAR␥, inactivate PPAR␦, inhibit Ras signaling, and reduce the proliferation and induce apoptosis of cancer cells (17). It has been shown that the A␤42-lowering effect of NSAIDs is independent of COX-inhibiting activity in cultured cells (21). We confirmed the A␤42-specific inhibition of A␤ secretion in culture cells by SSide, although the analysis of the inhibition profile at high concentrations was difficult because of the cell toxicity. To examine whether SSide directly inhibits ␥-secretase, we employed an in vitro assay system using solubilized membranes as enzyme sources and recombinant C100 polypeptides as substrates, and we demonstrated that SSide has the capacity to inhibit the total ␥-secretase activity, although cleavage at A␤42 site was most effectively inhibited. In addition, in vitro A␤ generation took place in the absence of any NTPs, suggesting that kinase activities (e.g. IB kinase ␤) are not involved in the regulation of ␥-secretase by SSide; we also observed that addition of ATP does not change the level of de novo A␤ generation, suggesting that ␥-secretase activity does not require energy. 2 Finally, we showed that SSide displayed noncompetitive inhibition kinetics in vitro, which is a common characteristic of a number of ␥-secretase inhibitors (34). From these data, we postulate that SSide works as a ␥-secretase inhibitor that directly affects its activity by binding to the membrane-embedded protease complex.
To date, several ␥-secretase inhibitors have been documented to inhibit secretion of A␤40 and A␤42 in two different patterns. The peptide aldehydes and peptidomimetic inhibitors containing a difluoroketone or alcohol group increase A␤42 secretion at sub-inhibitory doses and diminish it at a high concentration, whereas they inhibit A␤40 generation in an absolutely dose-dependent manner (39 -41). However, the rank order of inhibitory potencies of several peptide aldehydes against A␤40 and A␤42 are at similar levels, suggesting that a single ␥-secretase complex would generate A␤40 and A␤42 (40). In addition, the transition state analogue inhibitors of aspartyl proteases containing a hydroxyethylene isostere also inhibited the secretion of A␤40 and A␤42 to similar extents (12,41). The compounds display similar inhibition kinetics between a cellfree system (i.e. incubation of the membrane fraction that harbors both enzyme and substrate) and an in vitro assay using recombinant substrates, suggesting that these compounds act directly on ␥-secretase and that the differences in the inhibition kinetics might depend on their binding sites or target molecule(s) within the ␥-secretase complex.
The molecular mechanism underlying the reciprocal regulation in A␤40 and A␤42 generation at low concentrations of peptide aldehyde inhibitors and SSide still remains unknown. One possible explanation is that a partial loss of ␥ 40 -secretase function by the low concentrations of peptide aldehydes or difluoroketone peptide mimetics would shift the substrate supply to ␥ 42secretase that is still active, thereby leading to overproduction of A␤42. In sharp contrast, SSide exhibited entirely novel inhibition profiles to preferentially inhibit A␤42 generation, which was accompanied by an increase in the production of A␤38 as well as A␤40 at sub-inhibitory concentration ranges in vitro. We speculate that SSide may act on a component that is distinct from those affected by peptide aldehydes or difluoroketone protease inhibitors. Alternatively, SSide and other inhibitors may exert opposite effects on a component that is involved in the determination of the position of a scissile bond to be cleaved in ␥-secretase complex. Unexpectedly, a coincubation study with the transition state analogue inhibitor, L-685,458, showed a direct competition with SSide, raising the possibility that SSide might directly act on the catalytic site, although the structure of SSide is not similar to any known substrates. An alternative possibility would be that SSide binds to the noncatalytic site of ␥-secretase complex and allosterically regulates the catalytic site in a way to dissociate substrates and active site-specific inhibitors, showing an apparent direct competition. Such reciprocal regulation of different proteolytic activities by protease inhibitors or substrates has been observed in proteasome that harbors three distinct proteolytic activities (i.e. chymotrypsin-like, trypsin-like, and peptidylglutamyl peptide-hydrolyzing activities) (42)(43)(44). Ritonavir, an inhibitor of human immunodeficiency virus-1 protease, competitively inhibits the chymotrypsin-like activity, whereas trypsin-like activity is enhanced (43). Extensive studies using active site-specific inhibitors suggested that proteasome effectors/substrates (e.g. ritonavir) that cause reciprocal regulation might act on noncatalytic sites, rather than through binding to an active site. Further studies using derivatives of SSide that contain affinity moiety (e.g. photoreactive groups) are needed to obtain definitive proof that SSide acts directly on ␥-secretase.
It has been documented that almost all ␥-secretase inhibitors abolish the site-3 cleavage of Notch in cultured cells, with the exception of SSide and a nonpeptidic isocoumarin derivative, JLK6 (21,45). However, it has been shown that JLK6 fails to inhibit ␥-secretase activity in vitro, suggesting that JLK6 is not a direct inhibitor of ␥-secretase (46). We found that SSide has the capacity to inhibit the endoproteolysis of Notch at the S3 site at much higher concentrations compared with those for A␤ inhibition in vitro, whereas other peptidic inhibitors (e.g. DAPT) abolished Notch cleavage with similar potencies to those for A␤ generation. These results suggest that ␥and Notch secretases are pharmacologically distinct but related. Thus, it may be possible to avoid the envisaged side effects of ␥-secretase inhibitors caused by inhibition of Notch signaling by developing derivatives of SSide. Further attempts to define the molecular mechanisms of inhibition on ␥-secretase activity by SSide and to screen its derivatives specifically relevant to ␤APP cleavage will facilitate not only the development of a novel therapeutic drug for AD but also our understanding of the unusual intramembranous proteolytic activity of ␥-secretase that cleaves membrane-spanning proteins at multiple positions.