Main

Increasing evidence suggests that a key event in the pathogenesis of Alzheimer's disease is the altered production, aggregation and deposition of the Aβ peptide, a proteolytic fragment of 40–42 residues derived from APP. The longer isoform, Aβ42, is selectively increased in all presenilin mutations analysed and in most APP mutations that cause early-onset familial Alzheimer's disease. Aβ42 is the Aβ species initially deposited in brain, and is particularly prone to aggregation in vitro. Therefore, Aβ42 is believed by many to be the main culprit in the pathogenesis of Alzheimer's disease7.

We examined whether NSAIDs alter APP processing and generation of Aβ, particularly the Aβ42 species. We treated cells with increasing concentrations of various NSAIDs, and analysed Aβ40 and Aβ42 levels in culture medium by sensitive sandwich enzyme-link immunosorbent assay (ELISA) as described previously8. The range of NSAID concentrations was chosen on the basis of tolerated plasma concentrations achieved in humans9. Multiple NSAIDs were examined in this study, several of them sharing similar activities (see below; an overview describing the cell lines and compounds studied is provided in the Supplementary Information). For brevity, the non-selective COX-inhibitor sulindac sulphide, the active metabolite of the pro-drug sulindac with well recognized antineoplastic properties10, is presented as the prototypic NSAID in its striking preferential inhibition of Aβ42 secretion. In Chinese hamster ovary (CHO) cells stably transfected with both APP751 (wild-type APP, WT-APP) and the PS1 mutant M146L (PS1-M146L), a 50–70% reduction in the Aβ42/Aβ40 quotient was achieved at concentrations of 40–60 µM without significant reduction in total Aβ (Aβ40 + Aβ42) levels (Fig. 1a). This result was confirmed in CHO cells overexpressing WT-APP and in CHO cells transfected with the APP V717F mutation (see Supplementary Information). To exclude the possibility that this effect is specific to a particular cell type, we examined Aβ secretion in response to sulindac sulphide treatment in the human neuroglioma cell line HS683 stably transfected with APP695. A dose-dependent inhibition of Aβ42 secretion similar to that seen with CHO cells was observed (Fig. 1b). Sulindac sulphide also reduced Aβ42 secretion in kidney HEK293 cells, H4 neuroglioma cells, and primary mouse embryonic fibroblasts (Supplementary Information). Thus, this effect was observed in cell lines of rodent and human origin and is not dependent on cell type.

Figure 1: Analysis of Aβ from cultured cells treated with NSAIDs by ELISA.
figure 1

Aβ42 and total Aβ levels (Aβ40 plus Aβ42 values, Aβ40+42) were normalized to values obtained from vehicle-treated cells. Treatment with sulindac sulphide preferentially reduced Aβ42 levels in the medium of WT-APP PS1-M146L CHO cells (a) and human neuroglioma cells (b) in a dose-dependent manner. Selective inhibition of Aβ42 levels was also observed with ibuprofen (c) and indomethacin (d), but not with naproxen (e) in WT-APP CHO cells. Sulindac sulphide similarly reduced Aβ42 levels in fibroblasts deficient in COX-1 and COX-2 (f). Each panel shows the mean ± s.d. of all experiments.

Full size image

Ibuprofen reduced amyloid plaque pathology in a mouse model of Alzheimer's disease, and indomethacin seemed to slow the cognitive decline in patients with Alzheimer's disease in a placebo-controlled pilot study3,11. These two non-selective COX inhibitors also decreased the Aβ42/Aβ40 quotient by selective inhibition of Aβ42 secretion in a dose-dependent manner. In WT-APP PS1-M146L CHO cells, a 50% reduction in the Aβ42/Aβ40 quotient was reached at ibuprofen concentrations of 200–300 µM (Fig. 1c) and at indomethacin concentrations of 25–50 µM (Fig. 1d). Similarly, total Aβ levels were not significantly affected with either ibuprofen (up to 500 µM) or indomethacin (up to 150 µM) (Fig. 1c and d). No toxicity was detected by standard MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl-tetrazolium bromide) assay in CHO or HS683 cells or by [3H]-thymidine incorporation for CHO cells treated with sulindac sulphide concentrations up to 100 µM, with indomethacin up to 200 µM, or with ibuprofen up to 1 mM (data not shown). Therefore, the Aβ42 effect is not related to cytotoxicity.

We next examined sulindac sulphone, a second oxidative metabolite of sulindac with potent antineoplastic effects that lacks COX-inhibitory activity12. Unlike sulindac sulphide, sulindac sulphone had no effect on the Aβ42/Aβ40 quotient in WT-APP CHO cells with concentrations up to 400 µM (data not shown). This finding suggests a role for COX inhibition in the preferential reduction of Aβ42 secretion by NSAIDs. We therefore investigated whether reduction of Aβ42 levels is a characteristic common to all NSAIDs by assessing several other clinically approved NSAIDs. Naproxen is a non-selective COX inhibitor that belongs to the same structural class as ibuprofen. However, unlike ibuprofen or sulindac sulphide, treatment of WT-APP CHO cells with naproxen at concentrations up to 400 µM did not lower either the Aβ42/Aβ40 quotient or total Aβ levels (Fig. 1e). Similar negative results were obtained with aspirin, meloxicam (an NSAID with preferential COX-2 activity), SC-560 (a selective COX-1 inhibitor) and celecoxib (a selective COX-2 inhibitor) (see Supplementary Information). These data demonstrated that the capacity to lower Aβ42 secretion is not associated with all NSAIDs. Furthermore, the effective NSAID concentrations used in our experiments were clearly in excess of the levels required for complete inhibition of COX-1 and COX-2 in cell-based assays13 (see Supplementary Information), indicating that the reduction in Aβ42 levels may be independent of COX activity.

To demonstrate that the decrease in Aβ42 levels is not mediated by inhibition of COX activity and concomitant reduction in prostaglandin synthesis, fibroblasts deficient in both COX-1 and COX-2 by targeted gene deletions (COX-1-/- COX-2-/-)14 (see Supplementary Information) were examined for the effect of NSAIDs on Aβ42 levels. The basal levels of Aβ peptides were unchanged compared with littermate control mouse fibroblasts, indicating that the loss of COX-1 and COX-2 enzymes did not by itself alter Aβ42 levels (data not shown). However, treatment with sulindac sulphide reduced Aβ42 levels and altered the Aβ42/Aβ40 quotient in a similar fashion to that seen in CHO and HS683 neuroglioma cells (Fig. 1f) and identical to littermate control fibroblasts (data not shown). These results provide compelling evidence that the reduction in Aβ42 is not mediated by inhibition of COX activity, the principal mode of action of NSAIDs. Whether other NSAID-responsive pathways mediate this activity, such as by modulating lipoxygenase activity or transcription controlled by peroxisome proliferator-activated receptors (PPARs), is unknown.

In a recent study, chronic high-dose ibuprofen treatment was shown to significantly reduce amyloid pathology, neuritic dystrophy, plaque-associated gliosis and IL-1 expression in Tg2576 transgenic mice11. After 6 months of treatment, amyloid plaque numbers and Aβ levels in brain were reduced almost 50% and 40%, respectively. To determine whether the COX-independent reduction in Aβ42 seen with some NSAIDs in cultured cells could in part account for the long-term effects that have been reported, we examined whether treatment of Tg2576 mice with ibuprofen preferentially reduces Aβ42 levels. For these studies, 3-month-old female Tg2576 mice were orally dosed with 50 mg kg-1 d-1 of ibuprofen (n = 15) or 50 mg kg-1 d-1 of naproxen (n = 7), or mock treated (n = 18), and brain levels of SDS-soluble Aβ40 and Aβ42 were measured by ELISA. Administration of ibuprofen resulted in a highly significant 39% decrease in levels of Aβ42 without any changes in Aβ40 compared with mock-treated animals (Table 1 and Supplementary Information). Consistent with the tissue culture results, treatment with naproxen did not alter the levels of Aβ42 in brain at a dose that has been shown to inhibit prostaglandin production in rodent brain15. These results strongly indicate the physiological relevance of our in vitro studies. In the previous study11, a reduction in parenchymal but not vascular amyloid deposits was noted. Because vascular Aβ deposits consist primarily of Aβ40, whereas parenchymal deposits are a mixture of Aβ40 and Aβ42, the lack of effect on vascular amyloid is in agreement with our hypothesis that ibuprofen treatment may prevent amyloid pathology by decreasing the Aβ42/Aβ40 quotient in brain.

Table 1 Acute dosing with ibuprofen lowers brain Aβ42
Full size table

NSAIDs are unique in their ability to change the Aβ42/Aβ40 quotient by selectively decreasing Aβ42 levels. Recently, γ-secretase inhibitors have been developed that either inhibit the generation of all Aβ species or show preferential activity against Aβ40. Several of these inhibitors have been demonstrated to bind presenilins, leading to the hypothesis that presenilins are actual γ-secretases16. Because presenilins are essential not only for γ-secretase cleavage of APP but also for proteolytic processing of the Notch receptor17, γ-secretase inhibitors may concomitantly inhibit Notch cleavage, resulting in adverse side effects18. Furthermore, treatment with γ-secretase inhibitors leads to the accumulation of high amounts of carboxy-terminal fragments (CTFs) of APP that may be neurotoxic19,20. We therefore examined multiple parameters in APP processing as well as Notch intramembrane cleavage to determine whether any of these cellular pathways are altered in response to NSAID treatment. WT-APP CHO cells were treated with increasing concentrations of sulindac sulphide as described above. Western blot analysis with an APP C-terminal polyclonal antibody did not show any changes in the levels of full-length APP or APP CTF species cleaved by α- or β-secretase (see Supplementary Information). Similarly, secretion of the APP ectodomain (APPs) was not altered, as determined by western blotting analysis of conditioned medium using two different APP monoclonal antibodies (see Supplementary Information). Furthermore, by pulse-chase metabolic labelling, APP turnover was unchanged after treatment with sulindac sulphide (see Supplementary Information).

We next examined APP internalization because processing of APP in the endocytic pathway is hypothesized to be a major route for the generation and secretion of both Aβ40 and Aβ42 (ref. 21). Using a previously described APP internalization assay21, we found the ratio of cell surface APP to internalized APP to be virtually identical in cells treated with sulindac sulphide compared with vehicle-treated cells (see Supplementary Information). Finally, we analysed Notch intramembrane cleavage and formation of the Notch intracellular cytoplasmic domain (NICD) in HEK293 cells after transfection with the Myc-tagged NΔEMV complementary DNA, an artificial Notch receptor construct that undergoes constitutive cleavage22. Consistent with the results for APP CTFs shown above, treatment with sulindac sulphide did not impair Notch cleavage or NICD formation (Fig. 2). Similarly, treatment with ibuprofen (500 µM) or indomethacin (150 µM) did not result in either accumulation of APP CTFs or inhibition of Notch cleavage (data not shown). These results demonstrate that NSAID treatment did not significantly perturb APP processing or γ-secretase activity. However, the preferential decrease in Aβ42 could be explained by minor changes in γ-secretase activity that were not detectable in the preceding assays.

Figure 2: Analysis of Notch processing after treatment with sulindac sulphide in HEK 293 cells.
figure 2

NICD formation from a constitutively cleaved Notch construct (NΔEMV) was not impaired in cells treated with sulindac sulphide (125 µM SS). The left lane shows control transfection with an NICD construct to identify the cleavage fragment.

Full size image

To determine how NSAIDs may alter Aβ42 levels, two additional sets of experiments were performed. First, we analysed the full spectrum of Aβ species secreted by cells treated with sulindac sulphide by immunoprecipitation with mass spectrometry (IPMS). In contrast to ELISA measurements, IPMS provides definitive information about the length and identity of the individual Aβ peptides, for example Aβ(1–38), Aβ(1–40), and so on. The IPMS results extended our findings in significant ways. As expected, treatment with 75–100 µM sulindac sulphide reduced the level of Aβ(1–42) whereas Aβ(1–40) was essentially unaffected. Remarkably, the decrease in Aβ(1–42) secretion was accompanied by a dose-dependent increase in Aβ(1–38) species, resulting in a two-fold increase in the Aβ(1–38)/Aβ(1–40) quotient (Fig. 3a and b). Other Aβ peptide species (Aβ(1–37) and Aβ(1–39)) did not show consistent changes after treatment. To confirm the IPMS results, Aβ was immunoprecipitated from conditioned media of WT-APP PS1-M146L CHO cells and fractionated on a gel system that allows the resolution of individual Aβ species23. As anticipated, a large reduction in an immunoreactive band corresponding to Aβ(1–42) was accompanied by a dose-dependent increase in an immunoreactive species corresponding to Aβ(1–38) (Fig. 3c).

Figure 3: Aβ species in medium after NSAID treatment.
figure 3

a, Representative mass spectra of Aβ peptides from WT-APP CHO cells treated with 100 µM sulindac sulphide (SS) or DMSO vehicle showing a decrease in Aβ(1–42) species but an increase in Aβ(1–38) species. b, Quantitative mass spectrometry analysis of Aβ peptide levels calculated as a quotient, Aβ(1–x)/Aβ(1–40). c, The changes in Aβ(1–42) and Aβ(1–38) levels in medium were confirmed by bicine/urea SDS–PAGE. Standard Aβ(1–38), Aβ(1–40) and Aβ(1–42) peptides were used to identify Aβ species (right lane). The lane with control peptides was intentionally offset because the exposure time was longer to provide comparable band intensities. d, Reduction in Aβ(1–42)/Aβ(1–40) was accompanied by an increase in Aβ(1–38)/Aβ(1–40). Duplicate measurements are shown.

Full size image

Two potential mechanisms may explain this unprecedented change in Aβ production after NSAID treatment. Sulindac sulphide and other NSAIDs could reduce Aβ42 secretion by shifting γ-secretase activity towards production of Aβ38. Alternatively, NSAIDs may stimulate the activity of an exopeptidase that converts Aβ42 into shorter Aβ species such as Aβ38. In the second experiment, we therefore asked whether the turnover of Aβ42 in culture medium could be enhanced after NSAID treatment. Up to this point, only the levels of Aβ peptides had been assessed and it is possible that NSAIDs may facilitate the degradation or cell-mediated clearance of Aβ42 (refs 24,25,26). To examine this possibility, untransfected CHO cells were cultured in conditioned medium obtained from WT-APP CHO cells supplemented with increasing concentrations of sulindac sulphide. Because untransfected CHO cells do not produce Aβ to any significant level, all the measurable Aβ peptides from this culture paradigm were derived exogenously. Under this condition, sulindac sulphide did not reduce the levels of Aβ42 compared with controls (Fig. 4), indicating that NSAIDs did not selectively target Aβ42 for catabolism, either by activating exopeptidase activity or by selective clearance. Future studies will be required to address this complex mechanism as there are no precedents for this pharmacological alteration in Aβ C-terminal cleavages, putatively attributable to altered γ-secretase activity.

Figure 4: Turnover of Aβ42 after NSAID treatment.
figure 4

Aβ40 and Aβ42 were assayed from untransfected CHO cells incubated with conditioned medium obtained from CHO cells transfected with WT-APP and treated with increasing concentrations of sulindac sulphide (SS) (see Methods). The levels of Aβ40 and Aβ42 from the conditioned medium were virtually identical to vehicle-treated controls analysed by western blotting (top) or ELISA (bottom), indicating that Aβ turnover (clearance or degradation) was not affected by NSAID treatment. The ELISA results represent mean ± s.d. of all experiments.

Full size image

Our observations suggest a mechanism through which NSAIDs might confer protection from Alzheimer's disease that is independent of direct anti-inflammatory properties. Because not all NSAIDs lower Aβ42 levels, this Aβ42 activity may be an important criterion to consider in clinical trials of NSAIDs for the treatment of Alzheimer's disease. Our results suggest an explanation for the variable data obtained from epidemiological studies and recent negative clinical results with COX-2-selective NSAIDs27, as we find that many NSAIDs (including these selective COX-2 inhibitors) lack the ability to lower Aβ42. However, our results do not exclude the potential benefit of NSAIDs in reducing the inflammatory response in the Alzheimer's disease brain. Finally, in contrast to the current generation of γ-secretase inhibitors, NSAIDs do not perturb APP or Notch processing but rather seem to induce a subtle shift in γ-secretase activity. However, significant gastrointestinal and renal toxicity associated with long-term COX-1 inhibition limit the clinical utility of current NSAIDS as Aβ42-lowering agents. Because the Aβ42 effect we described is independent of COX activity, compounds with optimized Aβ42 reduction and little to no inhibition of COX-1 activity are likely to be identified. Such agents would represent a new generation of ‘anti-amyloid’ drugs that selectively target production of the highly amyloidogenic Aβ42 species without inhibiting either COX activity or the vital physiological functions of γ-secretase.

Methods

Cell culture and drug treatment

CHO cells stably transfected with human APP751 or transfected with both human APP751 and human mutant PS1 (M146L), CHO cells transfected with human mutant APP751 (V717F), human neuroglioma cells HS683 transfected with APP695, HEK 293 cells transfected with APP695, and spontaneously immortalized embryonic fibroblasts from COX1- and COX-2-deficient animals were maintained in DMEM medium supplemented with 10% fetal bovine serum and 100 U ml-1 penicillin/streptomycin (Life Technologies). The NSAIDs sulindac sulphide (50 mM, Biomol), sulindac sulphone (50 mM, Biomol), naproxen (100 mM, Cayman Chemical), aspirin (2.5 M, ICN Biomedicals), meloxicam (50 mM, Calbiochem) and SC-560 (50 mM, Calbiochem) were dissolved in dimethyl sulphoxide (DMSO). Indomethacin (50 mM, Biomol) and (S)-ibuprofen (250 mM, Biomol) were dissolved in ethanol. Celecoxib was prepared from commercial Celebrex capsules (Searle) by solvent extraction followed by recrystallization. Nuclear magnetic resonance (NMR) and mass spectrometry verified the identity and purity of celecoxib. Purified celecoxib was dissolved in ethyl acetate (10 mM). For analysis of Aβ secretion, APP processing and Notch cleavage, cells were cultured in serum-containing medium and pretreated overnight with the specific NSAIDs. Medium was changed and treatment was continued for another 24 h. All experiments were repeated 2–4 times in duplicate or triplicate and results either from a representative experiment or from all experiments (mean ± s.d.) are shown.

ELISA

Aβ was analysed by sandwich ELISA as described previously8. Media were collected after conditioning for 24 h, and cell debris was removed by centrifugation. Complete protease inhibitor cocktail (Roche) was added and Aβ40 and Aβ42 were quantified by BAN50/BA27 and BAN50/BC05 ELISAs or 3160/BA27 and 3160/BC05 ELISAs. All measurements were performed in duplicate.

Adenoviral infection of COX1-/- COX-2-/- cells

The adenoviral vector expressing APP695 has been described previously28. Primary embryonic fibroblasts derived from mice deficient in COX-1 and COX-2 were infected with 100 plaque-forming units per cell in serum-free medium for 2 h. Medium was changed and cells were then treated with drugs as described above.

APP processing, Aβ clearance and Notch cleavage

Methods and antibodies for analysis of steady-state APP expression, APPs secretion, APP turnover and APP internalization are described in the Supplementary Information. Aβ turnover after NSAID treatment was examined by culturing untransfected CHO cells with conditioned medium obtained from WT-APP-transfected CHO cells. CHO cells were pretreated with sulindac sulphide overnight. Medium was then replaced with the conditioned medium supplemented with increasing concentrations of sulindac sulphide and incubated for another 24 h. Aβ40 and Aβ42 levels in the medium were then analysed by bicine/urea Aβ western blot analysis and ELISA as documented below.

The Myc-tagged, amino-terminal-truncated Notch-1 construct (NΔEMV, in which methionine 1,726 has been mutated to valine to eliminate translation initiation at that site) and the construct containing only the NICD domain have been described22. To monitor formation of NICD, the NΔEMV construct was transiently transfected into HEK293 cells. Cells were then treated with drugs for 36 h and subsequently pulse labelled with [35S]-methionine for 30 min and chased for 2 h. Cell lysates were immunoprecipitated with monoclonal antibody 9E10 (Calbiochem) against the Myc-epitope sequence, fractionated by SDS–PAGE and analysed by phosphor imaging.

Mass spectrometry

Secreted Aβ peptides were analysed by IPMS as described29. One millilitre of conditioned medium was immunoprecipitated with monoclonal antibody 4G8 (Senetek) and molecular masses and concentrations of Aβ peptides were measured with an ultraviolet-laser desorption/ionization time-of-flight mass spectrometer. To compare the concentrations of individual Aβ species in conditioned medium, synthetic Aβ(12–28) peptide (Sigma) was added to the samples as an internal standard, and relative peak heights were calculated.

Western blotting of Aβ peptides

Bicine/urea Aβ western blot analysis was performed as described23. One millilitre of conditioned medium was immunoprecipitated with APP monoclonal antibody 26D6 recognizing amino acids 1–12 of the Aβ sequence26. Samples were separated on bicine/urea gels, transferred to nitrocellulose membranes and immunoblotted with 26D6 antibody. Standard Aβ40, Aβ42 and Aβ38 peptides (Sigma) were separated on the same gel for identification of the corresponding Aβ species. Representative radiograms are shown.

Ibuprofen treatment of Tg2576 transgenic mice

Female Tg2576 mice overexpressing APP695 containing the ‘Swedish’ mutation were treated at 3 months old, when they show high levels of soluble Aβ in brain but no signs of Aβ deposition30. NSAIDs were dissolved in DMSO, mixed with a sucrose drink (Kool-Aid) and fed orally to the animals. Controls were administered Kool-Aid with DMSO. The total amount of ibuprofen or naproxen delivered was 50 mg kg-1 d-1. This daily dose was divided equally and administered every 4 h for 3 d. Two hours after the final dose, animals were killed and SDS-soluble Aβ40 and Aβ42 were analysed by ELISA as described previously30.