The Non-cyclooxygenase Targets of Non-steroidal Anti-inflammatory Drugs, Lipoxygenases, Peroxisome Proliferator-activated Receptor, Inhibitor of (cid:1) B Kinase, and NF (cid:1) B, Do Not Reduce Amyloid (cid:2) 42 Production*

Epidemiological evidence suggests that chronic use of non-steroidal anti-inflammatory drugs (NSAIDs) reduces the risk of Alzheimer’s disease. Recently, NSAIDs have been shown to decrease amyloid pathology in a transgenic mouse model of Alzheimer’s disease. This benefit may be partially attributable to the ability of NSAIDs to selectively reduce production of the amyloidogenic A (cid:2) 42 peptide in both cultured cells and transgenic mice. Although this activity does not appear to require the action of cyclooxygenases in cultured cells, it is not known whether other NSAID-sensitive targets contribute to this A (cid:2) 42 effect. In this study, we have used both pharmacological and genetic means to determine if other known cellular targets of NSAIDs could mediate the reduction in A (cid:2) 42 secretion from cultured cells. We find that altered arachidonic acid metabolism via NSAID action on cyclooxygenases and lipoxygenases does not alter A (cid:2) 42 production. Furthermore, we demonstrate that alterations in activity of peroxisome pro-liferator-activated receptors, I (cid:1) B kinase (cid:2) or nuclear factor (cid:1) B do not

Alzheimer's disease (AD) 1 is the most common form of agerelated dementia. Safe effective treatments are urgently needed. One hallmark of AD is the accumulation of amyloid ␤-protein (A␤), derived from the amyloid precursor protein (APP), in senile plaques in brain. There are two predominant isoforms of A␤ peptide, A␤40 and A␤42, differing in their C termini: A␤42, the longer isoform is more amyloidogenic and toxic to cultured cells. Virtually all mutations associated with familial AD preferentially increase amount of A␤42 produced, and hence the ratio of A␤42:A␤40 (1). Thus, decreasing production or increasing the clearance of A␤42 may be an effective way to either prevent the development of or treat AD.
Epidemiological studies demonstrate that persons with a history of non-steroidal anti-inflammatory drug (NSAID) use have a reduced risk of AD (2,3). It is generally assumed that the inflammatory responses seen in brains of AD individuals play a key role in neurodegeneration (4,5), but it has not been established that the anti-inflammatory properties of NSAIDs underlie their apparent neuroprotective effects (6,7). In line with the epidemiological findings, treatment of APP transgenic mice, which develop A␤ deposits and associated pathology, with these compounds has proven to be beneficial. Specifically, ibuprofen, curcumin, and a nitric oxide derivative of flurbiprofen were recently shown to decrease the formation of amyloid plaques and reduce inflammatory markers in a transgenic mouse model of AD (8 -10). These and other observations have led to a number of clinical trials to determine if NSAIDs or immunosuppression may be useful in the treatment of AD.
In our recent studies (11) we found that some NSAIDs, including sulindac, ibuprofen, and indomethacin, lowered the levels of the amyloidogenic A␤42 isoform, hence reducing the ratio of A␤42:A␤40, in medium from a variety of cultured cells as well as in brains of APP transgenic mice. The effective NSAIDs did not grossly affect the production or processing of the amyloid precursor protein (APP). While these NSAIDs decreased the ratio of A␤42:A␤40, they appear to increase the production of shorter A␤ peptides, such as A␤38. This suggested that NSAIDs might subtly alter the production of various A␤ species. To our knowledge, the NSAIDs are the first class of compounds that specifically reduce A␤42 production without significant alteration in A␤40 levels.
The focus of these studies was to further investigate the cellular mechanisms responsible for the A␤42:A␤40 reduction by certain NSAIDs. Our initial report established that inhibition of the cyclooxygenase enzymes COX1 and COX2, the canonical targets of NSAIDs, was not sufficient to reduce the A␤42 levels (11). We therefore proposed that NSAIDs affect the amyloid pathology by lowering the amyloidogenic A␤42 peptide through a COX-independent pathway. This hypothesis is not unparalleled, as a dual-action model has been suggested for the effectiveness of NSAIDs in colorectal cancer studies where both COX-dependent and COX-independent mechanisms have been described. To further understand the mechanism whereby NSAIDs lower A␤42 levels, we examined whether other NSAID-sensitive targets including arachidonic acid, lipoxygenases, peroxisome proliferator-activated receptors (PPAR), and nuclear factor B influence the generation of A␤42 (Fig. 1). In addition, we examined several other compounds that have been reported to lower risk of AD or to reduce amyloid pathology in transgenic mice to determine whether they also have a previously unrecognized activity against A␤42.

EXPERIMENTAL PROCEDURES
Cell Lines and Culture Conditions-Chinese hamster ovary (CHO) cells stably expressing APP751 with or without human presenilin 1 M146L (designated APP-WT and PS1ML, respectively) were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, penicillin (100 units/ml), and streptomycin (100 g/ml) in a 37°C, humidified incubator with 5% CO 2 . One day prior to drug treatment cells were plated at ϳ1 ϫ 10 5 cells/ml. Embryonic fibroblasts from wild-type, IKK2 knockout (kindly provided by I. Verma) or p65/RelA knockout (a gift from E. Shaulian) mice were maintained as above. Primary cultures of ALOX5 and ALOX15 knockout fibroblasts were generated from neonatal mice resulting from homozygous ALOX5 or ALOX15 knockout (mice purchased from The Jackson Laboratory) breeding. Fibroblasts were infected with 60 pfu per cell of adenovirus encoding wild-type APP751 for 2 h before treatment with NSAIDs (11).
Enzyme-linked Immunosorbent Assays-For most experiments human A␤ ELISA were performed as described (12,13). For each experiment duplicate or triplicate samples were analyzed. Each drug was tested at several doses in a minimum of two independent experiments with sulindac sulfide as a positive control. Selective reduction in A␤42 was determined by calculating the A␤42:A␤40 ratio for each sample. These ratios were normalized to vehicle control in each experiment. These normalized ratios were pooled and analysis of variance used to determine if the A␤42:A␤40 ratio was significantly different from vehicle with p Ͻ 0.05. Error bars in figures represent standard errors.
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 (14) with the following modifications. Complete protease inhibitor, phosphoramidon and a synthetic A␤1-22 peptide, that served as an internal control, were added, and all A␤1-x were immunoprecipitated from conditioned medium by overnight incubation with anti-mouse IgG agarose beads and 26D6, which recognizes the N terminus of A␤. 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. Spectra shown are representative of at least two experiments performed with duplicate samples. Treatment-induced changes in A␤ species distribution were determined by normalization of peak heights to A␤40.
Bicine/Urea SDS-PAGE and Western Blots-Conditioned medium was immunoprecipitated as for MALDI-TOF analysis, except without the addition of synthetic A␤22. Bicine/urea gels were performed as described in Wiltfang et al. (15) with the following modifications. The final acrylamide concentration in the separating gel was 10% T, 5% C. The comb gel was eliminated from our protocol. In addition, the 30% sucrose and 5% 2-mercaptoethanol were replaced with glycerol and dithiothreitol. Proteins were transferred to nitrocellulose in 10 mM CAPS, pH 11, with 10% methanol and boiled for 5 min. Immunoblotting was with 26D6 APP antibody and the signal detected by horseradish peroxidase-conjugated goat anti-mouse IgG followed by enhanced chemiluminescence. Gels shown are representative of a minimum of three experiments with 2-3 replicates per experiment. Bands intensities were quantified using a CCD camera and software (Syngene).

Reduction in A␤42 Is Not Generalized Anti-inflammatory
Actions-To confirm that the A␤42-reducing activity was specific to NSAIDs and not mimicked by other anti-inflammatory compounds, we investigated the effects of glucocorticoids on A␤42 levels in CHO cells stably transfected with wild-type human APP (APP-WT). In clinical trials, these compounds did not slow the rate of cognitive decline in AD individuals (16,17). At the concentrations tested, neither prednisone nor dexamethasone reduced the amount of A␤42 or the A␤42:A␤40 ratio (Table I).
Epidemiological studies (18) have also suggested that antihistamine usage reduces the incidence of AD. Since histamine responses trigger inflammatory changes, we wanted to determine if blocking histamine action altered A␤42 levels. Therefore, we treated A␤-producing PS1ML cells with 0.5-20 M of the H-2 histamine receptor antagonist cimetidine. This treatment did not alter the levels of A␤ detectable by ELISA (Table  I) or bicine/urea gel analysis of conditioned medium.
Recent studies indicated that the NSAID-like compound curcumin might be beneficial in a mouse model of AD (9). Curcumin has both anti-inflammatory and anti-oxidant properties. APP-WT CHO cells were treated with curcumin to determine if it affected the production of A␤ species. Between 1 and 100 M, there was no reduction in A␤ levels or A␤42:A␤40 (Table I).
Counter to the effect of NSAIDs, there was a slight increase (20%) in A␤42:A␤40 at doses between 20 and 100 M. This increase in A␤42:A␤40 could be attributed to a reduction in A␤40 levels. Higher doses were toxic to the cells so A␤ could not be evaluated.
COX inhibition Is Not Required for NSAID Action on A␤42-We previously reported that the reduction in A␤42 levels secondary to NSAID treatment was accompanied by an increase in the amount of A␤38 species (11). However, several  NSAIDs, such as naproxen and aspirin, did not reduce A␤42 levels. Whether these NSAIDs affect shorter A␤ species was not investigated as the inactivity of naproxen in our previous study had only been determined by ELISA. Here we confirmed the lack of effect on A␤ peptides by both MALDI-TOF spectrometry and bicine/urea SDS-PAGE methods. In these assays, naproxen did not reduce A␤1-42 or increase A␤1-38, whereas sulindac sulfate, ibuprofen, flurbiprofen, and indomethacin were all effective ( Fig. 2 and data not shown). In our earlier work, we used COX-deficient cells to demonstrate that COX enzymes are not required for sulindac sulfide to reduce A␤42 levels. However, both of these approaches only concluded that COX inactivation is insufficient to reduce A␤42 levels. To address whether COX inhibition is required to alter A␤ production we used R-flurbiprofen, which does not inhibit COX and is not efficiently isomerized to S-flurbiprofen. R-flurbiprofen potently reduced A␤42 levels and increased production of shorter A␤s (Fig. 2C), in agreement with a recent report (19).
Arachidonic Acid and Lipoxygenases Do Not Affect A␤42 Levels-NSAIDs inhibit not only COX but can alter the activities of several lipoxygenases, which could increase arachidonic acid levels and effect the processing of arachidonic acid into HETEs and leukotrienes (Fig. 1). To mimic the effect of NSAIDs on increasing arachidonic acid levels, we treated the PS1ML cells with exogenous arachidonic acid and assayed the levels of the various A␤ species by ELISA and bicine/urea gels. We also treated cells with the phospholipase A2 activator, melittin (20). This treatment mobilizes endogenous arachidonic acid from cellular membranes. Arachidonic acid, either added to the extracellular medium or induced by melittin, increased the total amount of A␤ generated by ϳ20% as compared with control cells. In the case of arachidonic acid there was a slight reduction in the A␤42:A␤40 ratio, because of an increase in A␤40 and not a decrease in A␤42 (Table II). Melittin treatment had no effect on A␤42:A␤40 ratio.
Several NSAIDs, including ibuprofen and indomethacin, can inhibit the activity of 5-lipoxygenase (21). 5-lipoxygenase expression may be elevated in aging human brains (22), so we wanted to determine if the NSAID effect on A␤42 was mediated by inhibition of this enzyme. We examined whether other 5-lipoxygenase inhibitors could reduce the A␤42 levels detected by our sensitive ELISA assay (Table II). The doses chosen for these compounds were above those published to inhibit 5-lipoxygenase, but below toxic concentrations. The inhibitors nordihydroguiaretic acid, MK-886, and caffeic acid, which preferentially inhibit 5-LOX, were without effect on total A␤ or A␤42 production. Since at elevated doses these pharmacological inhibitors have multiple activities, which could mask the desired effect, we also used a genetic approach to eliminate 5-lipoxygenase activity. Neonatal fibroblasts from 5-lipoxygenase-deficient mice were infected with an adenovirus encoding APP695 and then treated with sulindac sulfide. As shown in Fig. 3A, sulindac retained the ability to inhibit A␤42 production in these cells, verifying that 5-LOX is not necessary for the NSAID effect on A␤42.
Recent reports indicate that some NSAIDs activate the murine leukocyte-type 12-lipoxygenase and its human homologue 15-lipoxygenase (23,24), so we examined the effect of modulating 15-lipoxygenase on A␤42:A␤40. The 15-lipoxygenase inhibitor baicalein had no effect on A␤42 levels (Table II). As above, we cultured fibroblasts from L-12-lipoxygenase-deficient mice and assayed A␤ levels after adenoviral infection. Once again we saw a marked reduction in A␤42 with sulindac treatment (Fig.  3B), indicating that this enzyme is also not required for the amyloid-altering effects of NSAIDs.
Peroxisome Proliferator-activated Receptors Do Not Alter A␤42 levels-The peroxisome proliferator-activated receptors (PPARs) are sensitive to NSAIDs and the eicosanoids produced by arachidonic acid metabolism. Recent studies have also indicated that PPAR activation inhibits microglial activation and the production of proinflammatory molecules and may attenuate A␤ toxicity (25,26). NSAIDs activate PPAR␣ and PPAR␥ (27), and inhibit PPAR␦ (28). Each or these PPARs must heterodimerize with the RXR nuclear transcription factor to bind DNA and alter transcription. If NSAIDS reduce A␤42 by activating PPAR␣ then the PPAR␣ agonist 8(S)-HETE should also reduce A␤42 levels from APP-expressing cells. However, 8(S)-HETE did not reduce A␤42 levels secreted from APP-WT cells. Likewise the PPAR␥ agonists ciglitazone, 15-deoxy-⌬-prostaglandin J 2 or BRL49653 (29) failed to reduce A␤42 levels (Table  III). In addition, we examined the ability of a PPAR␥ antagonist, GW9662 (30), to block the response to sulindac sulfide. This antagonist had no effect on the ability of sulindac to reduce A␤42 levels (Fig. 4, A and B). We also tested the A␤42reducing potential of the RXR agonist LG10305, which was ineffective in reducing A␤42 levels.
To confirm the pharmacological approach, APP-WT CHO cells were transfected with either PPAR␥ or PPAR␦ cDNAs and then stimulated with 15-deoxy-⌬-prostaglandin J 2 or cyclic-prostaglandin I2, respectively. Overexpression of these receptors, with or without activator, did not alter the ratio of FIG. 2. A␤42 lowering potential of NSAIDs does not correlate with the ability to inhibit COX. A, MALDI-TOF analysis of A␤ reveals that not all NSAIDs alter A␤42 and A␤38 levels. PS1ML cells were treated with 300 M naproxen, 350 M ibuprofen, 250 M flurbiprofen, or Me 2 SO control. Ibuprofen and flurbiprofen reduced A␤42 and increased A␤38 species while naproxen had no effect on any of the A␤ species. B, bicine-urea SDS gel analysis of A␤ from NSAID-treated cells. Results from naproxen, ibuprofen, and flurbiprofen treatment were similar to that obtained by MALDI-TOF. Migration of A␤ standard peptides is indicated on the right. C, bicine-urea SDS gel analysis of A␤ with flurbiprofen enantiomers. Cells were treated with 250 M racemic, purified S-or R-flurbiprofen. Both enantiomers reduced A␤42 as effectively as racemic compound. A␤42:A␤40 (Fig. 4C). Because pharmacologic and genetic manipulations of the various PPARs had no affect on A␤42: A␤40, these receptors are unlikely to mediate the NSAID A␤42 response.
NFB Is Not Required to Mediate the Sulindac Sulfide-induced Reduction in A␤42-Several NSAIDs can inhibit the activity of the NFB transcription complex at one or more levels: the inhibition of IKK and the prevention of NFB DNA binding (31,32). Inhibition of IKK would prevent IB phosphorylation and degradation and prevent translocation of NFB to the nucleus. The NFB complex is a dimer that can be composed of several proteins. The most common NFB complex contains p65/RelA and p50. Tomita et al. (33) reported that overexpression of p65/RelA increased production of A␤42 but not A␤40 indicating a potential A␤42-specific signaling pathway that can be modulated by NSAIDs. Because elimination of p65/RelA by genetic knockout prevents induction of most NFBresponsive genes (34), we used mouse embryonic fibroblasts from IKK␤ or p65-knockout mouse embryos to test whether the NFB pathway was required for the NSAID-mediated reduction in A␤42. As before, the MEFs were infected with APP695 adenovirus and then treated with sulindac sulfide. In both cell lines, sulindac inhibited the production of A␤42 similar to control fibroblasts (Fig. 3), indicating that these proteins are not required to mediate the NSAID response. DISCUSSION The unexpected finding that NSAIDs reduce the production of A␤42 in vitro in cell culture medium and in vivo in brains of transgenic mice opened two key questions. The first question was whether the effect was mediated by the anti-inflammatory action of these compounds, even if not due to COX inhibition. The second is the mechanism by which NSAIDs lower A␤42 levels. The goal of this study is to examine a number of candidate cellular pathways that may be responsible for this NSAID effect on A␤42 generation.
NSAIDs are potent anti-inflammatories that are primarily known to inhibit COX enzymes. However, our initial report (11) demonstrated that A␤42 was reduced by sulindac in cells lacking COX enzymes. Furthermore, several NSAIDs that inhib-   ited COX could not reduce A␤42:A␤40. These findings indicate that inhibition of COX was not sufficient to reduce A␤42 levels. Further confirmation of this hypothesis was obtained from a study treating cultured cells with the R-enantiomers of flurbiprofen and ibuprofen (19) and from our results with R-flurbiprofen using both MALDI-TOF and Western blotting analyses. The fact that R-enantiomers of these NSAIDs, which are inactive against COX, were able to reduce A␤42 production argues against COX inhibition as the basis for A␤42 reduction.
To examine whether NSAIDs may act through a general anti-inflammatory mechanism, we screened several compounds unrelated to conventional NSAIDs with known antiinflammatory activity. One of the proposed bases for the neuroprotective benefits of NSAIDs is to inhibit the inflammatory responses seen in brains of AD individuals, thus we thought it is pertinent to analyze this potential mechanism. This notion led to a treatment trial in AD individuals with the corticosteroid, prednisone, which was found to be without benefit (17). In our cultured cell system, we found that glucocorticoids, including prednisone and dexamethasone, did not reduce A␤42 levels. On the other hand, our finding that curcumin does not alter A␤42 levels is notable in light of the recent report showing a significant reduction in amyloid pathology in transgenic mice treated with curcumin chronically (9). As curcumin has multiple properties, these findings taken together suggest that curcumin may reduce A␤ load in brain by a mechanism distinct from the A␤42 property seen in NSAIDs. Although there was initial suggestion that H2 antagonists, such as cimetidine, lowered the risk of AD, a recent analysis of the Cache county epidemiological data and a placebo-controlled study concluded that use of these drugs did not affect AD risk or progression (35,36). These more recent analyses are more aligned with our finding that cimetidine had no effect on A␤42. However, as our cell culture experiments are performed under basal conditions, we cannot draw any conclusion as to the benefit of the aforementioned anti-inflammatory compounds in cells already challenged by inflammatory stimuli. It is possible that mediation of inflammatory responses may be beneficial in treating AD, and this benefit could be dependent or independent of A␤42 production. These findings therefore highlight the fact that there likely are multiple activities from these anti-inflammatory compounds and that multiple mechanisms can account for the reduction in A␤ levels, amyloid pathology, and/or ameliorate AD risk.
A major focus of this study is to test candidate cellular pathways that may be responsible for the observed reduction in A␤42 by examining the non-COX pathways known to be affected by NSAIDs. Although the primary target of NSAIDs is COX inhibition, NSAIDs are pleiotropic compounds with other known effects. These pathways are summarized in the schematic diagram in Fig. 1. For example, one such pathway is through altered AA metabolism. Inhibition of COX blocks the conversion of AA to prostaglandin H2 and results in a build up of AA that has been mobilized from the plasma membrane by the action of phospholipase A2 (37). Phospholipase A2 has been reported to increase APPs secretion, which presumably would preclude A␤ release (38). In addition, NSAIDs regulate arachidonic acid not only by inhibiting COX enzymes but also via interaction with lipoxygenases. 5-Lipoxygenase activity increases in the brain with age and is inhibited by indomethacin, ibuprofen, or sulindac (21,22). On the other hand, some NSAIDs increase activity and expression of 15-lipoxygenase. Consequently, we investigated the role of lipoxygenases and AA metabolism in A␤42 generation via pharmacologic and genetic means. Mobilization of AA by supplementation of culture medium with exogenous AA or by activation of PLA2 by melittin did not reduce A␤42 levels but showed a slight increase in total A␤ levels. In our studies, neither 5-nor 15-LOX appeared to influence A␤42:A␤40 in conditioned medium. We cannot exclude the possibility that inflammatory leukotrienes generated by lipoxygenases are an exacerbating factor in Alzheimer's disease, but this does not appear to be manifested in A␤42 levels.
PPARs are another important class of NSAID targets that have been the focus of some attention, especially in the cancerrelated literature. The PPAR family of nuclear receptors consists of PPAR␣, PPAR␥, and PPAR␦ (also denoted as PPAR␤). The NSAIDs indomethacin, flufenamic acid, fenoprofen, and ibuprofen appear to directly activate the transcriptional activity of PPAR␣ and PPAR␥ (27). PPARs also have important roles in modulating lipid and glucose metabolism. In this regard, ␥-secretase activity can be strongly influenced by the cholesterol content of the plasma membrane (39), and insulin affects not only glucose metabolism but also APP processing and A␤ levels (40). In addition, PPAR-␥ agonists have antiinflammatory properties, and in particular, these agonists inhibited A␤ stimulated release of proinflammatory products from microglia (25). Despite interesting potential links to amyloid or NSAID activity, our studies described here were not able to observe any selective effects of PPAR overexpression, activation, or inhibition on lowering A␤42 levels.
Lastly, PPARs also regulate transcriptional activation by STAT, AP-1, and NFB transcription factors. In this context, overexpression of NFB/p65 has been shown to increase the proportion of A␤1-42 (33). Moreover, there is evidence that NSAIDs can antagonize NFB more directly. NFB may have profound effects in Alzheimer's disease progression by elevating levels of inducible nitric-oxide synthase, tumor necrosis factors, interleukins, and complement factors, all of which are known to be elevated in AD brains (3). Our results showed that NSAID-induced A␤42 effects were intact in IKK␤ or p65-deficient fibroblasts, much as has been seen in COX-deficient MEFs, thereby arguing against this pathway as a major contributor to the A␤42 reduction.
In summary, our studies failed to identify a potential candidate cellular pathway, known to be affected by NSAIDs, that mediates the A␤42 effect. Our studies therefore reinforce the concept that the ability to reduce A␤42 production is not shared by all NSAIDs, but rather it is due to a novel secondary activity. Our studies did not address whether NSAIDs can directly target either APP or the ␥-secretase complex itself. In view of our results, it is tempting to suggest that NSAIDs may conformationally alter ␥-secretase activity, much as presenilin mutations preferentially increase A␤42 levels. This notion would be consistent with the subtle switch in A␤ peptides following treatment of cultured cells with certain NSAIDs to favor the shorter species. In related studies we have observed that the A␤42 lowering activity of NSAIDs can be modified by presenilin mutations and that A␤42 generation from an in vitro ␥-secretase assay can be modified by NSAIDs (42). A recent report by Takahashi et al. (41) using in vitro assay also suggested that NSAIDs alter the activity of ␥-secretase. Taken together, these observations lead us to favor the notion that NSAIDs alter A␤42 production by subtly modifying APP cleavage either at the level of the substrate or a component of the ␥-secretase complex, rather than indirectly, such as by modulating cellular signaling pathways. Identification of the molecular target, however, will be difficult for two reasons. First, the precise nature of ␥-secretase remains to be elucidated and second, NSAIDs reduce A␤42 production with low affinity. Nevertheless, the findings presented here are an important first step in the long term goal of identifying compounds that selectively lower A␤42 in vivo but have negligible COX inhibition to limit the well known gastrointestinal side effects as potential candidates for AD treatment.