Evidence That Nonsteroidal Anti-inflammatory Drugs Decrease Amyloid (cid:1) 42 Production by Direct Modulation of (cid:2) -Secretase Activity*

, Chronic use of nonsteroidal anti-inflammatory drugs (NSAIDs) is associated with a lower risk of developing Alzheimer’s disease. Recent evidence indicates that some NSAIDs specifically inhibit secretion of the amyloidogenic A (cid:1) 42 peptide in cultured cells and mouse models of Alzheimer’s disease. The reduction of A (cid:1) 42 peptides is not mediated by inhibition of cyclooxygenases (COX) but the molecular mechanism underlying this novel activity of NSAIDs has not been further defined. We now demonstrate that NSAIDs efficiently reduce the intracellular pool of A (cid:1) 42 in cell-based studies and selectively decrease A (cid:1) 42 production in a cell-free assay of (cid:2) -secretase activity. Moreover, we find that pre-senilin-1 (PS1) mutations, which affect (cid:2) -secretase activity, differentially modulate the cellular A (cid:1) 42 response to NSAID treatment. Overexpression of the PS1-M146L mutation enhances the cellular drug response to A (cid:1) 42 lowering NSAIDs as compared with cells expressing wild-type PS1. In contrast, expression of the PS1- (cid:3) Exon9 mutation strongly diminishes overexpress previously

that may halt or reverse the underlying disease process are not available (1,2). Numerous epidemiological studies support the finding that chronic intake of nonsteroidal anti-inflammatory drugs (NSAIDs) can decrease the risk for AD by more than 50% (3)(4)(5)(6). This protective effect of NSAIDs has generally been ascribed to a diminution of deleterious inflammatory processes in the AD brain (3). However, recent findings suggest a direct impact of some NSAIDs on the amyloid pathology in AD. Treatment of various cultured cells with the NSAIDs sulindac sulfide, ibuprofen, indomethacin, and flurbiprofen specifically inhibited the release of the amyloidogenic A␤42 peptide (7,8). A␤42 is a proteolytic fragment derived from the ␤-amyloid precursor protein (APP) by ␤and subsequent ␥-secretase cleavage activities and is believed to play a central role in AD pathology (9,10). Short term administration of ibuprofen to APP transgenic mice lowered brain levels of A␤42 and chronic high dose ibuprofen treatment significantly reduced amyloid plaque numbers and plaque-associated pathology in aging APP transgenic mice (7,11). The reduction in A␤42 levels was achieved without affecting other APP processing pathways, like secretion of the soluble APP ectodomain (APPs), and is not the result of enhanced degradation or cell-mediated clearance of A␤42. Importantly, within the concentration range tested, A␤42lowering NSAIDs did not inhibit processing of the NOTCH receptor and presumably other substrates of ␥-secretase (7).
NSAIDs are the first small molecules reported to specifically target A␤42 without apparent overall inhibition of ␥-secretase activity, and it will be important to elucidate the molecular mechanism of this activity if more active compounds are to be developed. The NSAID effect on A␤42 levels is fully retained in cells deficient in cyclooxygenase-1 (COX1) and COX2 enzymatic activity, thereby excluding the primary pharmacological targets of NSAIDs as mediators of the A␤42 reduction (7). COX-independent mechanisms for NSAID activity are well established (12). For example, NSAIDs have emerged as potentially valuable drugs in the chemoprevention of certain cancers, and they affect several molecular pathways regulating cellular proliferation and apoptosis (13,14). Many of these mechanisms seem to be COX independent, including nuclear factor B activation (15), inhibition of lipoxygenases (16) and modulation of peroxisome proliferator-activated receptor signaling (17,18) and potentially any of these signaling pathways could contribute to the observed NSAID effect on A␤42 secretion.
Alternatively, NSAIDs might directly alter ␥-secretase function. This possibility is attractive because the reduction in A␤42 level is accompanied by an increase in shorter A␤ species, particularly A␤38. This suggests a subtle alteration of ␥-secretase cleavage pattern rather than a selective inhibition of A␤42 generation (7,19). ␥-Secretase is a multiprotein complex consisting of at least four membrane-bound proteins: presenilin (PS), nicastrin, APH-1, and PEN-2. All of these proteins are required for proper maturation and activity of the complex while the enzymatic core of this activity may reside within presenilin itself (20 -22). ␥-Secretase activity displays loose sequence specificity and generates A␤ peptides of varying length, but peptides ending after 40 and 42 amino acids are the species most intensely studied. Over 150 mutations in the PS proteins, which cause the majority of familial forms of Alzheimer's disease (FAD), have been described and they all appear to increase production of A␤42 (10). It has been hypothesized that these mutations lead to subtle conformational changes of PS proteins in the ␥-secretase complex, thereby altering the stochiometry of the generated A␤ peptides (22). In this way, NSAIDs may affect ␥-secretase activity in an analogous but opposite fashion to PS mutations (7,19). In this study, we demonstrate that NSAIDs reduced intracellular A␤42 levels and selectively decreased A␤42 production in a cell-free assay of ␥-secretase activity. Furthermore, we observed that FAD mutations in the PS and APP modulate the cellular A␤42 response to NSAID treatment. These studies suggest the ␥-secretase complex, either the enzymatic machinery or its substrate, as the molecular target of A␤42-lowering NSAIDs.

EXPERIMENTAL PROCEDURES
Drugs and Antibodies-The NSAIDs sulindac sulfide, indomethacin, and ibuprofen were purchased from Biomol, naproxen was from Sigma, ␥-secretase inhibitors z-IL-CHO and Compound E were synthesized as described (23)(24)(25). All other chemicals were from Sigma except when otherwise indicated. Monoclonal antibody PSN2 generated against a synthetic peptide corresponding to residues 31-56 of PS1 and APP monoclonal antibody 26D6 recognizing amino acids 1-12 of the A␤ sequence have been described (26,27).
Generation of Retroviral Vectors-Retroviral shuttle plasmids encoding wild type human PS1, PS1-M146L and PS1-⌬Exon9 have been described previously (28). For generation of pseudo-typed retroviral vectors, GP-293 packaging cells were transiently co-transfected with retroviral shuttle plasmids and plasmid encoding VSV-G envelope glycoprotein at 1:1 ratio. Retroviral particles were collected 48 h after transfection and stored at Ϫ80°C.
Cell Lines and Cell Culture-WT-APP CHO cells, which stably overexpress wild-type human APP751, and WT-APP HS683 neuroglioma cells, which stably overexpress wild-type human APP695, have been previously described (29,30) and are the parental cell lines for all stably transfected CHO and HS683 cell lines used in this study. WT-APP CHO or HS683 cells were infected with retroviral vectors encoding wild-type PS1, PS1-M146L, or PS1-⌬Exon9 and were selected with puromycin (5 or 1 g/ml, respectively). Stable pools were analyzed without further clonal selection. WT-APP CHO cells stably overexpressing wild type human PS1 have been described (31). Comparable PS1 expression in transfected cell lines was verified by Western blotting with monoclonal antibody PSN2. CHO cell lines overexpressing APP751 harboring the V717F "Indiana" mutation or the "Swedish" mutation (K670N/M671L) have been described (29,32). All cell lines were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and 100 units ml Ϫ1 penicillin/streptomycin (Invitrogen).
ELISA-A␤ secretion was analyzed by sandwich enzyme-linked immunosorbent assay (ELISA) as described previously (24). Media were collected following conditioning for 24 h and cell debris was removed by centrifugation. Complete protease inhibitor mixture (PI, Roche Applied Science) was added, and A␤40 and A␤42 levels were quantified by BAN50/BA27 and BAN50/BC05 ELISAs or 3160/BA27 and 3160/BC05 ELISAs. All measurements were performed in duplicate.
Analysis of Intracellular A␤ Production-Intracellular A␤ levels were analyzed by bicine/urea A␤ Western blot analysis as described (7,33). One 15-cm dish of WT-APP PS1-M146L CHO cells was lysed in 1 ml 2% CHAPS, 50 mM Tris, 150 mM NaCl containing 1ϫ PI and A␤ was immunoprecipitated with monoclonal antibody 26D6. Samples were separated on bicine/urea gels, transferred to nitrocellulose membranes, and immunoblotted with 26D6 antibody. Standard A␤1-40 and A␤1-42 peptides (Sigma) were separated on the same gel for identification of the corresponding A␤ species. Representative radiograms are shown.
␥-Secretase in Vitro Assay-␥-Secretase in vitro assay was performed using purified detergent solubilized membranes from CHO-overex-pressing APP695NLϩI as described (23,34). Samples were incubated for 2h at 37°C in the presence of selected compounds or Me 2 SO vehicle to allow in vitro production of A␤ and the APP intracellular cytoplasmic domain (AICD, CTF␥) and subsequently stored at Ϫ80°C. A␤40 and A␤42 levels were then quantified by ELISA. Net activity during the 2-h incubation period was defined by subtracting background values obtained from samples immediately frozen at time 0. Corrected values were then normalized to the Me 2 SO control condition and expressed as % control.
Dose Response Experiments and Statistical Analysis-The A␤42 response of individual cell lines to NSAID treatment was compared in dose response experiments. All cell lines intended for comparison were cultured and treated in parallel at similar cell densities. Cells were cultured in serum-containing medium and pretreated overnight with increasing concentrations of sulindac sulfide, ibuprofen, indomethacin, or Me 2 SO vehicle. Medium was changed and treatment was continued for another 24 h. A␤40 and A␤42 levels in conditioned medium were then analyzed by ELISA. Duplicate A␤42 measurements from each drug concentration were averaged and normalized to the Me 2 SO control condition. These experiments were repeated 6 -10 times, and results were analyzed by two-way ANOVA with Bonferroni post-tests using cell line and drug concentration as categorical variables. Calculations were performed with GraphPad Prism software (GraphPad Software).

Matrix-assisted Laser Desorption/Ionization Time-Of-Flight
Mass Spectrometry-MALDI-TOF was performed on A␤ peptides immunoprecipitated from conditioned medium of CHO cells as described (35) with the following modifications. P i , phosphoramidon, and a synthetic A␤1-22 peptide that served as an internal control were added, and all of the A␤1-x were immunoprecipitated from conditioned medium by overnight incubation with anti-mouse IgG-agarose beads and 26D6. Extraction from the beads was with formic acid/water/isopropyl alcohol 1:4:4 (v/v/v). Eluted material was mixed 1:1 with ␣-cyano-4-hydroxycinnamic acid solution prior to spotting for spectrometry. Treatment-induced changes in A␤ species distribution were determined by normalization of peak heights to A␤40.

Sulindac Sulfide Reduces Intracellular A␤42 Production-
A␤42-lowering NSAIDs have been shown to reduce secretion of A␤42 peptides from a variety of cell lines, and this effect is not caused by enhanced degradation or clearance of A␤42 from tissue culture media (7,8). These changes in the secreted pool of A␤42 should be reflected by similar changes in the intracellular pool of A␤42 if NSAIDs affect A␤42 production by direct modulation of ␥-secretase activity. We therefore analyzed intracellular A␤42 levels after treatment of CHO cells stably transfected with both APP751 (wild-type APP, WT-APP) and the PS1 mutant M146L (PS1-M146L) with the A␤42-lowering NSAID sulindac sulfide. A␤ was immunoprecipitated from cell lysates and separated on a gel system that permits resolution of individual A␤ species. Treatment of cells with 60 M sulindac sulfide consistently reduced intracellular A␤42 levels and strongly decreased the A␤42/A␤40 ratio with minimal effects on A␤40 levels ( Fig. 1).
A␤42-lowering NSAIDs Inhibit A␤42 Production in a Cellfree Assay of ␥-Secretase Activity-We next examined the effect of several NSAIDs on A␤ production in a cell-free assay of ␥-secretase activity. If A␤42-lowering NSAIDs act through di-FIG. 1. Analysis of intracellular A␤ levels after treatment with the A␤42-lowering NSAID sulindac sulfide. CHO cells overexpressing WT-APP and PS1-M146L were treated with 60 M sulindac sulfide or Me 2 SO vehicle. A␤ was immunoprecipitated from cell lysates and analyzed by bicine/urea SDS-PAGE. Sulindac sulfide treatment decreased intracellular A␤42 to undetectable level but had essentially no effect on A␤40 from a representative experiment. A␤-(1-40) and A␤-(1-42) peptide standards, used to identify corresponding intracellular A␤ species, are shown on the right. ␥-Secretase Is the Molecular Target of A␤42-lowering NSAIDs rect modulation of ␥-secretase activity, they would be expected to inhibit A␤42 production in cell-free assays, as demonstrated previously for well characterized ␥-secretase inhibitors (23,36). As predicted, the ␥-secretase inhibitors z-IL-CHO and Compound E both inhibited ␥-secretase activity, with significant reductions in both A␤40 and A␤42 production. NSAIDs, such as indomethacin and sulindac sulfide, that have been shown to affect A␤42 production in cell-based assays (7) displayed very similar activities in the cell-free ␥-secretase assay (Fig. 2). At these doses, even when A␤42 production is virtually abolished as seen with sulindac sulfide and indomethacin, no measurable effect was observed on AICD/CTF␥ production (data not shown). Naproxen, an NSAID without A␤42-lowering properties (7), had no effect on A␤42 production (Fig. 2).
The PS1-M146L Mutation Enhances the Cellular A␤42 Response to NSAID Treatment-In our previous study, the NSAID effect on A␤42 was observed across multiple cell lines and cell types, including those expressing various PS mutations. In view of the in vitro ␥-secretase assay results, we now asked whether different PS1 mutants might subtly influence the cellular A␤42 response to NSAID treatment. We choose to first analyze the PS1-M146L mutation as preliminary observations suggested that the level of A␤42 reduction was different from wild-type PS1 cells. For these experiments, CHO cell lines were treated with three increasing concentrations of sulindac sulfide (20 -60 M) and A␤40 and A␤42 levels in culture media were measured by ELISA. The A␤42 response of individual cell lines was then compared by two-way ANOVA (see "Experimental Procedures" for details). Total A␤ levels (A␤40ϩA␤42) were not significantly affected at these concentrations of sulindac sulfide and treatment did not cause toxicity as reported previously (7). ANOVA demonstrated that overexpression of wild type, human PS1 resulted only in minimal, non-significant changes in the A␤42 response as compared with parental cells expressing endogenous, hamster PS1 ( Fig. 3 and Table I). However, overexpression of the PS1-M146L mutation strongly enhanced the cellular A␤42 response to sulindac sulfide treatment. ANOVA showed a highly significant statistical difference in the A␤42 reduction as compared with either parental WT-APP CHO cells expressing endogenous PS1 or WT-APP WT-PS1 CHO cells overexpressing wild-type, human PS1 (p Ͻ 0.001, Fig. 3 and Table I). At all three concentrations tested, the reduction in A␤42 levels was almost 20% greater in PS1-M146L cells than that seen in the other two cell lines. Similar trends toward enhanced A␤42 reduction with the PS1-M146L mutant as compared with WT-PS1 were also observed with indomethacin and ibuprofen and, importantly, in HS683 neuroglioma cells expressing the same PS constructs (data not shown). These results indicated that the enhanced A␤42 reduction is distinctively associated with the PS1-M146L mutant, and is not limited to the NSAID sulindac sulfide or to a specific cell type.
The PS1-⌬Exon9 Mutation Diminishes the Cellular A␤42 Response to NSAID Treatment-We next examined CHO cells overexpressing the PS1-⌬Exon9 mutation to assess whether another PS1 mutation would similarly alter the cellular drug response to NSAID treatment. We chose this mutation because it results in an exon deletion rather than a missense substitution as seen in virtually all other FAD PS1 mutations and because the clinical manifestation of this mutation is varied (37). Surprisingly, we found that the PS1-⌬Exon9 mutation strongly diminished the A␤42 response to sulindac sulfide treatment. Two-way ANOVA analysis of 10 independent doseresponse experiments showed a highly significant attenuation of the A␤42 response at all three concentrations tested as compared with parental WT-APP CHO cells expressing endogenous PS1 or WT-APP WT-PS1 CHO cells overexpressing wildtype, human PS1 (p Ͻ 0.001, Fig. 4, panel A, and Table II). This result was confirmed with HS683 neuroglioma cells overexpressing the PS1-⌬Exon9 mutation (Fig. 4, panel B, and Table  II). In sum, these findings strongly indicated that PS1 mutations that alter A␤42 generation are able to modulate the cellular A␤42 response to NSAIDs treatment.
It has recently been demonstrated that inhibition of A␤ production by the transition state analogue ␥-secretase inhibitor L-685,458 was also attenuated in cells expressing the PS1-⌬Exon9 mutant (38). Furthermore, in contrast to cells expressing wild type PS1, cells expressing PS1-⌬Exon9 produced significant amounts of A␤1-43 peptides and, paradoxically, secretion of A␤1-43 was increased in a dose-dependent manner by inhibitor treatment. Our ELISA for detection of A␤42 does not efficiently discriminate between A␤42 and A␤43 peptides (39). Hence, our results could be confounded by the A␤43 effects of the ⌬Exon9 mutation. We therefore investigated the profile of A␤ peptides secreted by WT-APP CHO cells overexpressing PS1-⌬Exon9 treated with 60 M sulindac sulfide or Me 2 SO vehicle by MALDI-TOF mass spectrometry (Fig. 5). We confirmed the prominent release of A␤1-43 peptides in cells overexpressing PS1-⌬Exon9. However, when compared with vehicle control, treatment with sulindac sulfide did not increase FIG. 2. A␤42-lowering NSAIDs inhibit A␤42 production in a ␥-secretase in vitro assay. CHAPSO-solubilized membrane preparations of APP-transfected CHO cells were incubated for 2 h with 100 M of indomethacin, sulindac sulfide, naproxen, ␥-secretase inhibitors (z-IL-CHO and Compound E), or Me 2 SO vehicle and A␤ production was quantified by ELISA. Indomethacin and sulindac sulfide displayed selective inhibition of A␤42 production up to 95%. In contrast to ␥-secretase inhibitors, A␤42-lowering NSAIDs did not significantly affect A␤40 production. Consistent with previous results from cell-based assays, naproxen did not impair A␤40 or A␤42 production. *, p Ͻ 0.05; ANOVA with Dunnett's post-hoc test. n ϭ 3.  (Table I). ***, p Ͻ 0.001 Bonferroni post-tests.

␥-Secretase Is the Molecular Target of A␤42-lowering NSAIDs
A␤1-43 release indicating that the impairment to inhibit A␤42 secretion in cells expressing PS1-⌬Exon9 as detected by ELISA was accurate (Fig. 5).
An APP Mutation That Affects ␥-Secretase Cleavage (APPV717F) Enhances the Cellular A␤42 Response to NSAID Treatment-Finally, in view of the results from cells expressing PS1 mutations, we investigated whether FAD APP mutations would likewise alter the cellular A␤42 response to NSAID treatment. WT-APP CHO cells, CHO cells overexpressing the APPV717I mutant, which affects ␥-secretase cleavage (40), and CHO cells overexpressing Swedish mutant APP, which affects ␤-secretase cleavage (41), were treated with sulindac sulfide and analyzed as above. The A␤42 response of CHO cells overexpressing Swedish mutant APP was not significantly different from WT-APP CHO cells. However, similar to the PS1-M146L mutation, CHO cells overexpressing the APPV717F mutant displayed a significant enhancement in the A␤42 reduction at 40 -60 M (p Ͻ 0.01) and a trend toward greater reduction at 20 M as compared with WT-APP CHO cells. (Fig. 6 and Table III). DISCUSSION Accumulation and aggregation of A␤ peptides in the cerebral cortex is believed to be an early and crucial event in the pathogenesis of AD. A␤ peptides are generated by sequential proteolytic cleavage of APP by ␤and ␥-secretase activities, which are therefore considered prime targets for therapeutic intervention (1,2). A number of small molecule ␥-secretase inhibitors have been identified, and they are able to block A␤ production with high potency in cultured cells and in APP transgenic mouse models of AD and have entered early stages of clinical testing (1,42). However, ␥-secretase activity also cleaves a number of additional substrates besides APP such as Notch and ErbB-4, and ␥-secretase inhibitors suppress Notch processing in vivo (43)(44)(45). Consequently, it is unclear whether inhibition of ␥-secretase activity can be used clinically without causing serious side effects (46).
NSAIDs that specifically inhibit A␤42 production do not appear to affect Notch processing (7). Elucidation of their mechanism of action therefore seems to be highly desirable not only for pharmacological intervention in Alzheimer's disease but also for further understanding ␥-secretase function. Since cyclooxygenases are the main pharmacological target of NSAIDs, inhibition of COX and suppression of prostaglandin synthesis seemed to be an obvious mechanism by which NSAIDs could reduce A␤42 secretion. However, such a mechanism appeared less likely when it became apparent that only a subset of NSAIDs lower A␤42 levels, whereas all NSAIDs by definition inhibit COX. Consistent with this interpretation, we subsequently showed that the NSAID effect on A␤42 was preserved in cells lacking cyclooxygenase activity (7). Given these results and, in particular, our recent negative correlation of the NSAID A␤42 effect with other known non-COX targets of NSAIDs (57), we therefore considered a more immediate effect of A␤42-lowering NSAIDs on ␥-secretase activity.
If NSAIDs were to inhibit ␥-secretase activity, then perhaps precedents can be found in the transition state analogs and other small molecule inhibitors that modulate or directly interact with the ␥-secretase complex or its substrates. Three characteristics appear to define these ␥-secretase inhibitors. First, these compounds have been shown to reduce A␤ production in both cell-based and cell-free in vitro ␥-secretase assays. The latter use membrane preparations or solubilized microsomes for A␤ production and many ␥-secretase inhibitors suppress A␤ formation in these in vitro assays with equal or higher potency as compared with cell-based assays (47). However, it is noteworthy that a group of A␤-lowering isocoumarins reduced A␤ secretion only in cell-based assays but not in in vitro assays (48,49), suggesting that these compounds do not directly affect ␥-secretase activity. Second, FAD mutations in the PS and APP proteins that alter ␥-secretase cleavage by selectively increasing A␤42 generation have been shown to decrease the efficiency of ␥-secretase inhibitors to block A␤ secretion from transfected cells (38,50). The interpretation of these findings is that the mutations have induced subtle conformational changes in the PS-containing ␥-secretase complex or its substrate, APP, which alter the inhibitor interaction with the enzymatic machinery or its substrate (50). And third, several transition state and nontransition state inhibitors have been demonstrated by affinity labeling to directly bind to PS (25,36,51,52).
Although some conventional ␥-secretase inhibitors preferentially inhibit A␤40, and can increase A␤42 production under conditions where A␤40 is partially inhibited (24,(53)(54)(55), the subtle switch in A␤ C-terminal cleavage induced by A␤42lowering NSAIDs to favor shorter A␤ peptides is not seen with any of the published ␥-secretase inhibitors. Nevertheless, we found in this study that these NSAIDs exhibit features that are in common with previously described ␥-secretase inhibitors in  4. The PS1-⌬Exon9 mutation diminishes the A␤42 response to sulindac sulfide treatment. CHO or HS683 glioma cells overexpressing wild-type APP alone (WT-APP), wild-type APP, and wild-type PS1 (WT-APP WT-PS1), or wild-type APP and the FAD mutant PS1-⌬Exon9 (WT-APP WT-PS1-⌬Exon9) were treated with increasing concentrations of sulindac sulfide and A␤42 secretion was quantified by ELISA. Dose response experiments were analyzed by two-way ANOVA with WT-APP cells as control. A, CHO cells overexpressing the PS1-⌬Exon9 mutation demonstrated significantly diminished A␤42 reduction as compared with parental WT-APP or WT-APP WT-PS1 cells (n ϭ 10, Table II). B, HS683 cells overexpressing the PS1-⌬Exon9 mutation also exhibited significantly diminished A␤42 reduction to sulindac sulfide as compared with parental WT-APP or WT-APP WT-PS1 cells (n ϭ 6, Table II). ***, p Ͻ 0.001; **, p Ͻ 0.01 Bonferroni post-tests.
␥-Secretase Is the Molecular Target of A␤42-lowering NSAIDs two notable ways. First NSAIDs lower A␤42 both in cell-based assays and in vitro. Second, the ability of NSAIDs to reduce A␤42 is altered by PS and APP mutations. Specifically, in regards to the first characteristic, we observed that A␤42lowering NSAIDs reduced the intracellular pool of A␤42 and also inhibited A␤42 but not A␤40 production in an in vitro ␥-secretase assay. These results establish that the A␤42 reduction is a rapid effect and does not require secondary transcription, and that NSAIDs affect A␤42 production rather than the release of A␤42 into the medium. Similar to ␥-secretase inhibitors and also in contrast to A␤-lowering isocoumarins, NSAIDs lower A␤42 both in cell-based as well as in cell-free assays. The latter confirms recent results by Takahashi et al. (56) demonstrating that sulindac sulfide behaves as a reversible, noncompetitive inhibitor of A␤42 production in a modified ␥-secretase in vitro assay system. However, our studies differ from those of Takahashi and colleagues in two ways. First, we did not observe a biphasic behavior in A␤40 production with an increase at low concentrations of sulindac sulfide (1-25 M, data not shown) and a decrease at higher concentrations (100 M) as reported (56). Although certain peptide aldehyde or peptidomimetic ␥-secretase inhibitors have been shown to increase A␤42 but not A␤40 production at sub-inhibitory doses, these effects are present in both ␥-secretase in vitro and cellbased assays (24,(53)(54)(55). In contrast, neither of us detected any increase in A␤40 levels from cell-based assays with either sulindac sulfide or other A␤42-lowering NSAIDs, suggesting that these findings may be confined to the in vitro assay employed in that study (7,56). Second, Takahashi et al. (56)  Dose-response experiments were performed as described in the text and analyzed by two-way ANOVA with WT-APP CHO or WT-APP HS683 cells as control group. n ϭ 10 (CHO); n ϭ 6 (HS683). **, p Ͻ 0.01 Bonferroni post-tests. ***, p Ͻ 0.001.  6. The APPV717F mutation, which affects ␥-secretase cleavage, enhances the A␤42 response to sulindac sulfide treatment. CHO cells overexpressing wild-type APP (WT-APP), cells overexpressing the APPV717F mutant (APPV717F) or cells overexpressing Swedish mutant APP (APPswedish) were treated with increasing concentrations of sulindac sulfide and A␤42 secretion was quantified by ELISA. Dose response experiments were analyzed by two-way ANOVA with WT-APP cells as control. APPV717F cells but not APPswedish cells exhibited significantly greater A␤42 reduction at 40 -60 M (p Ͻ 0.01) and a trend toward greater reduction at 20 M as compared with WT-APP CHO cells. (n ϭ 8 -10, Table III); **, p Ͻ 0.01 Bonferroni post-tests. reported that production of the APP intracellular cytoplasmic domain (AICD) was similarly perturbed by sulindac sulfide in their in vitro assay system, again with an unusual biphasic response: a severalfold increase at low concentrations and almost complete inhibition at 100 M. As before, we did not find any effects of sulindac sulfide on AICD/CTF␥ production in cell-based assays (60 M) or in our in vitro assay (58) (up to 400 M). The apparent discrepancies between our results with sulindac sulfide and the results of Takahashi et al. may be related to the use of different ␥-secretase in vitro assays. The experiments by Takahashi et al. (56) employed solubilized membranes as a source of ␥-secretase activity and a recombinant peptide consisting of the 100 C-terminal amino acids of APP as substrate. In contrast, our assay system utilized solubilized membrane preparations from transfected cells that contain both ␥-secretase activity and full-length APP as substrate.
Given that NSAIDs appear to induce a very subtle shift in ␥-secretase activity, it is highly likely that differences in the in vitro assays, such as the cell type used as a source of ␥-secretase activity, solublization conditions, and source of substrate, could result in slightly different effects of NSAIDs on ␥-secretase activity. Nevertheless, despite these differences, both studies strongly support the hypothesis that some NSAIDs directly modulate ␥-secretase activity.
A second unexpected characteristic shared by NSAIDs and ␥-secretase inhibitors was seen when cells expressing FAD mutations in the PS and APP proteins were treated with A␤42-lowering NSAIDs. The PS1-M146L and the APPV717F "Indiana" mutations that alter ␥-secretase cleavage to increase A␤42 production, but not the "Swedish" APP mutation that affects ␤-secretase cleavage, significantly enhanced the A␤42 reduction after NSAID treatment as compared with wild type PS1 or APP. In striking contrast, the PS1-⌬Exon9 mutation significantly attenuated the A␤42 response to sulindac sulfide indicating that PS1 mutations modulate the drug response both positively and negatively. The greatly reduced A␤42 response in cells expressing PS1-⌬Exon9 in this report is especially informative in light of a recent study with the ␥-secretase inhibitor L-685,458 (38). It was shown that the ability of L-685,458 to reduce A␤ secretion, especially the A␤42 species, from cells expressing PS1-⌬Exon9 was substantially attenuated as compared with cells expressing wild-type PS1, a situation highly reminiscent of what we observed with the A␤42lowering NSAID sulindac sulfide.
Finally, photoaffinity derivatives of the transition-state analog L-685,458 label the N-and C-terminal fragments of PS1, which form mature non-covalent heterodimers. L-685,458 does not bind to immature full-length protein but does bind to the PS1-⌬Exon9 mutant, which lacks the endoproteolytic cleavage site and accumulates as full-length protein (36). Therefore, the prominent deficiency of the ␥-secretase inhibitor L-685,458 and the NSAID sulindac sulfide to reduce A␤42 production from cells expressing the PS1-⌬Exon9 mutant indicates that these compounds are both conformationally affected by the same PS mutation. Importantly, this observation leads us to speculate that these compounds may share the same molecular target and even potentially an overlapping binding region. This notion is further supported by recent findings that sulindac sulfide and L685,458 act in competition and not synergistically on ␥-secretase activity (56). However, unambiguous identification of the molecular target for A␤42-lowering NSAIDs will require the development of an NSAID-like photoaffinity reagent. Until such labeling studies are completed, we would propose that NSAIDs represent a class of compounds that lower A␤42 production by direct modulation of ␥-secretase activity or the enzyme substrate, i.e. APP. As such, complete elucidation of the exact molecular mechanism by which NSAIDs lower A␤42 generation holds considerable promise for the development of novel A␤42-specific ␥-secretase inhibitors.