Second Generation γ-Secretase Modulators Exhibit Different Modulation of Notch β and Aβ Production*

Background: The γ-secretase complex is a drug target in the treatment of Alzheimer disease (AD). Results: Two novel second generation γ-secretase modulators (GSMs) modulate both Nβ and Aβ but not Notch intracellular domain (NICD) production. Conclusion: Second generation and NSAID-based GSMs have different modes of action regarding Notch processing. Significance: GSMs that do not affect NICD signaling are essential for the development of tolerable AD therapeutics. The γ-secretase complex is an appealing drug target when the therapeutic strategy is to alter amyloid-β peptide (Aβ) aggregation in Alzheimer disease. γ-Secretase is directly involved in Aβ formation and determines the pathogenic potential of Aβ by generating the aggregation-prone Aβ42 peptide. Because γ-secretase mediates cleavage of many substrates involved in cell signaling, such as the Notch receptor, it is crucial to sustain these pathways while altering the Aβ secretion. A way of avoiding interference with the physiological function of γ-secretase is to use γ-secretase modulators (GSMs) instead of inhibitors of the enzyme. GSMs modify the Aβ formation from producing the amyloid-prone Aβ42 variant to shorter and less amyloidogenic Aβ species. The modes of action of GSMs are not fully understood, and even though the pharmacology of GSMs has been thoroughly studied regarding Aβ generation, knowledge is lacking about their effects on other substrates, such as Notch. Here, using immunoprecipitation followed by MALDI-TOF MS analysis, we found that two novel, second generation GSMs modulate both Notch β and Aβ production. Moreover, by correlating S3-specific Val-1744 cleavage of Notch intracellular domain (Notch intracellular domain) to total Notch intracellular domain levels using immunocytochemistry, we also demonstrated that Notch intracellular domain is not modulated by the compounds. Interestingly, two well characterized, nonsteroidal anti-inflammatory drugs (nonsteroidal anti-inflammatory drug), R-flurbiprofen and sulindac sulfide, affect only Aβ and not Notch β formation, indicating that second generation GSMs and nonsteroidal anti-inflammatory drug-based GSMs have different modes of action regarding Notch processing.

directly involved in the A␤ formation, and it also determines the pathogenic potential of A␤. Because ␥-secretase mediates cleavage of many substrates involved in cell signaling, such as the Notch receptor, it is crucial to sustain these pathways while inhibiting toxic A␤-secretion. However, investigations attempting to find ␥-secretase inhibitors (GSIs) with a sufficient therapeutic window between APP and Notch processing has been extremely challenging (17). Moreover, severe side effects, probably caused by abrogated Notch signaling, were present, causing a recent large clinical phase 3 trial to be interrupted (reviewed in Ref. 18). Alternative strategies to combat A␤ production is therefore clearly needed, and ␥-secretase modulators (GSMs) represent a growing and promising class of anti-amyloidogenic drugs. Importantly, Notch inhibition is avoided using GSMs instead of GSIs. Typically, GSMs do not affect the overall rate of Notch and APP processing (⑀ and S3 cleavage). Instead, by shifting the cleavage preference of the enzyme from producing the amyloid-prone A␤42 variant to shorter and less toxic A␤ species, GSMs change the proportions of various A␤ peptides that are formed (19,20). The first GSMs, subsets of nonsteroidal anti-inflammatory drugs (NSAIDs), such as sulindac sulfide and ibuprofen, were identified in 2001 (19). Since then, many second generation GSMs have brought improvement. These compounds, which are generally structurally distinct from the NSAID family by the absence of an acidic carboxyl group, are more potent and efficient in the central nervous system than were the early GSMs (21)(22)(23)(24). Some studies reported NSAID-based GSMs interacting with APP-derived C99 peptides (25)(26)(27), but other groups have challenged this implied substrate targeting hypothesis (28 -31). During the last year, we and others identified ␥-secretase instead of APP as the principal target of second generation GSMs (21,(32)(33)(34)(35)(36). However, even though the pharmacology of A␤ generation with respect to first and second generation GSMs has been thoroughly studied, little is known about their overall effect on ␥-secretase-mediated cleavage of other substrates, such as the Notch receptor. In this study, we have developed an assay studying N␤ production and investigated, head to head, Notch and APP processing in presence of both first and second generation GSMs. These studies provide compelling evidence that it is possible to obtain selectivity between N␤ and A␤ production, a feature that may be important in development of GSMs for chronic treatment in AD.

EXPERIMENTAL PROCEDURES
Compounds-The GSIs dibenzazepine (DBZ) and L-685,458, as well as GSMs R-flurbiprofen and sulindac sulfide, were obtained from Sigma-Aldrich. AZ1136 was prepared as previously described (32), as was AZ4126, according to patent number WO2010132015.
Ethical Permission-All animal experiments were performed in accordance with relevant guidelines and regulations provided by the Swedish Board of Agriculture. The ethical permission was provided by the Stockholm Södra Animal Research Ethical Board.
Cell Culture-HEK293 cells stably expressing human FLAG-Notch1-⌬E (FLAG-N⌬E) (32) or APPswe were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, nonessential amino acids, 10 M Hepes, and 300 g/ml hygromycin or 100 g/ml Zeocin, respectively. For each experiment, the cells were counted and plated in T75 flasks, 6or 384-well plates (for N␤, A␤, and NICD experiments, respectively) the day before treatment. On the following day, the GSM, R-flurbiprofen (200 M), sulindac sulfide (125 M), AZ1136 (25 M), AZ4126 (400 nM), or vehicle control (Me 2 SO) was separately added to fresh cell media and incubated for 24, 16, or 5 h (for N␤, A␤, and NICD experiments, respectively) before conditioned media or cells were analyzed.
Immunoprecipitation (IP) and MS Analysis of N␤ and A␤-After a clarifying spin and addition of complete protease inhibitor (Roche Applied Science), collected media were immunoprecipitated with 40 l of ␣-FLAG M2-agarose (Sigma) or 5 g of 4G8 (Covance) that were preincubated with a mixture of protein Aand G-Sepharose beads (GE Healthcare) for FLAG-tagged N␤ (F-N␤) and A␤, respectively. The samples containing N␤ were incubated at 4°C, washed, and eluted using a FLAG-tagged protein immunoprecipitation kit (Sigma) according to the manufacturer's instructions. The A␤-containing samples were incubated at room temperature for 2 h and washed three times for 10 min at 4°C in PBS (pH 7.4). The immunoprecipitated peptides were then eluted in 100 l of 0.1% TFA in 20% acetonitril. Eluted N␤ and A␤ peptides were concentrated using C 18 ZipTips (Millipore) and re-eluted with 0.1% TFA in 50% acetonitril saturated with ␣-cyano-4-hydroxy-cinnamic acid matrix and analyzed using MALDI-TOF MS. The experiments were performed in triplicate for N␤ and duplicate for A␤ and repeated five or six times. Control experiments included IP of media from cells without FLAG-N⌬E (HEK293 cells) or beads without 4G8 antibodies attached, as well as cell media treated with two different ␥-secretase inhibitors, L-685,458, and DBZ. MALDI-TOF MS measurements were performed using a Micromass M@LDI instrument (Waters). Each spectrum represents an average of 1000 shots acquired 10 at a time. In the spectrum a total peak height of ϩ5 m/z, relative to the monoisotopic peak, was calculated for each isoform. Prior to analysis the peak heights were normalized to the sum of the peak height of all isoforms in the spectra, and an average was obtained from duplicated and triplicated samples, respectively. The data show relative changes among all the different isoforms in response to GSM treatment. It should be noted that a relative quantification cannot be interpreted as a direct reflection of an absolute or relative abundance of a species because the ionization efficiency might be different for different isoforms and because different isoforms are more hydrophobic than others.
For tandem mass spectrometry (MS/MS) analysis, F-N␤ peptides were immunoprecipitated using the ␣-FLAG M2 antibody (Sigma) coupled to magnetic beads (2). Briefly, 10 g of the antibody was added to 50 l of magnetic Dynabeads M-280 sheep anti-mouse IgG (Invitrogen). Then antibody-coated beads were added to cell media and incubated at 4°C for 2 h. After washing using the KingFisher magnetic particle processor, the F-N␤ peptides were eluted using 100 l of 0.5% formic acid. Mass spectrometry measurements were performed using a Bruker Daltonics UltraFleXtreme MALDI-TOF/TOF instrument, and all of the samples were analyzed in duplicate. By result of MS analysis, several compounds were selected for tandem MS/MS, and 2000 single shot spectra were recorded of the precursor ions and 10000 of the fragment ions.
Quantification of Secreted A␤ from Cells-HEK/APPSwe cells were plated in 384-well plates and on the following day exposed to GSMs or vehicle control for 5 h, before conditioned media were analyzed as described previously (32). Briefly, A␤ levels were determined using Meso Scale Discovery (MSD) technology with C-terminally specific antibodies measuring A␤1-X (where X indicates [37][38][39][40][41][42]. For total A␤ levels, the 4G8 antibody was used. Animals and Animal Handling-Female C57BL/6 mice (Harlan Laboratories), 10 weeks old at the time of drug administration, were kept in conventional housing and fed standard rodent chew (RM1(E)SQC, SDS, Scanbur) and tap water ad libitum. The mice were randomized into different treatment groups and after acclimatization, AZ4126 (10, 25, 50, or 75 mol/kg) or vehicle (0.4 -3% dimethylacetamide in 2.7-20% HP␤CD in water) was administered once by oral gavage. The animals were anesthetized with isoflurane 2 h after treatment. Plasma was isolated from blood collected by cardiac puncture into EDTA tubes by centrifugation for 10 min at 3000 ϫ g at 4°C within 20 min from sampling. The plasma samples were frozen on dry ice and stored at Ϫ80°C until analysis. The animals were sacrificed by decapitation, and the brains were dissected into hemispheres, frozen on dry ice, and stored at Ϫ80°C until analysis of A␤40 and A␤42 and drug concentration Quantification of A␤ and Drug Concentrations in Animal Samples-Drug concentration in plasma and brain samples was determined by reversed phase liquid chromatography and electrospray MS/MS (32). Soluble A␤ was extracted from mouse brain tissue in diethylamine according to an earlier described protocol (32). A␤40 and A␤42 levels in brain were measured by a validated commercial ELISA (BioSource) according to the manufacturer's instructions. All samples were analyzed in duplicate, and data analysis was performed using GraphPad Prism 4, one-way analysis of variance followed by Dunnett's multiple comparison test. The level of significance was set at p Յ 0.05.
Autoradiography Competition Study-Autoradiography binding on 10-m frozen brain sagittal sections from rats were performed according to previously described protocols (32,37). Briefly, rat brains were sectioned, air-dried, and stored at Ϫ80°C upon analysis. Prior to the addition of 5 nM [ 3 H]AZ8349 or [ 3 H]AZ8349 together with unlabeled GSMs (500 M of R-flurbiprofen or 10 M of AZ4800 or AZ4126), prewarmed adjacent sections were incubated in 50 mM Tris-buffer (pH 7.4). The sections were incubated at room temperature for 40 min, followed by three 7-min washes in buffer at 1°C to reduce unspecific binding and then exposed to imaging plates (Fuji BAS-TR2040) for 4 days together with a plastic tritium standard (Amersham Biosciences). The plates were processed with a FLA7000 Imaging Reader (Fujifilm), and the binding was analyzed with Multigauge software V3.0 (Fujifilm).
In Vitro Cellular NICD Formation Assay-The NICD formation assay using ImageXpress scanner (Molecular Devices) was carried out as described earlier with minor modifications (32). Briefly, the cells were treated, washed, fixed with 4% paraformaldehyde, and then stained using primary ␣-NICD antibodies, C-20 (Santa Cruz Biotechnology) and Val-1744 (Cell Signaling). Thereafter, the secondary antibody, Alexa Flour 594 (Invitrogen), was added. By automatic measurements, the average fluorescence was determined and the percentage of NICD formation for Val-1744 was expressed relative to 0.5% Me 2 SO (100% control) and 500 nM L-685,458 (0% control). For NICD modulation, Val-1744 fluorescence was normalized to total C20 fluorescence and expressed relative to Me 2 SO.

RESULTS
␥-Secretase-mediated Notch Processing Results in Many Different Notch ␤ Peptides-To investigate how different structural classes of GSMs affect Notch processing, we used human embryonic kidney cells (HEK293) stably expressing an N-terminally FLAG-tagged human Notch1 ⌬E (FLAG-N⌬E) variant (Fig. 1A). Our first goal was to confirm that the FLAG-N⌬E variant is a ␥-secretase substrate by examining the well characterized S3 cleavage. Using ␣-FLAG M2 and neo-specific Val-1744 antibodies on immunoblotting, we obtained a robust expression of FLAG-N⌬E, as well as of NICD, which was abolished in the presence of the GSI, L-685,458 ( Fig. 1A, right panel). To study the production of secreted FLAG-tagged Notch-␤ peptides (F-N␤), conditioned media from HEK/ FLAG-N⌬E cells were immunoprecipitated using the ␣-FLAG M2 antibody and then subjected to MALDI-TOF MS analysis. The MS spectrum revealed a range of different M2-immunoprecipitated F-N␤ peptides (Fig. 1B), consistent with previous results (15). By culturing the cells in the presence of two different GSIs, L-685,458 and DBZ, we found that the majority of the F-N␤ peptides were generated in a ␥-secretase-dependent manner (Fig. 1, C and D). Interestingly, the generation of some shorter peptides, below mass to charge (m/z) 3400, were not inhibited by the GSIs and thus were probably generated through other enzymatic reactions (Fig. 1, C and D). Moreover, no peptides precipitated by M2-agarose were identified in the media of HEK293 cells that were lacking expression of FLAG-N⌬E, which displayed the specificity of the precipitation protocol (Fig. 1E).
Next, we identified which peptide sequence corresponded to which peak in the MALDI spectrum. To confirm peptide identities, the immunoprecipitates were analyzed by MALDI-TOF/ TOF. Given the natural cleavage site in the signal peptide, FLAG-N⌬E, we matched peptides, starting with RGPR before the FLAG sequence, and identified 11 different peptides corresponding to a range of species between F-N␤12 to F-N␤25 (Fig.  1F). In addition, we could not observe peptides longer than F-N␤25 (Fig. 1C), which is in line with data reported by Okochi et al. (15).
First Generation NSAID Class GSMs Affect A␤ but Not N␤ Production-To explore how the NSAID class GSMs modulate the ␥-secretase complex, we first studied their effect on APP processing by measuring the A␤ peptide generation in HEK293 cells stably expressing the APPswe mutation, HEK/APPswe. Conditioned media were subjected to IP-MALDI-TOF MS analysis, resulting in detection of A␤37, A␤38, A␤39, A␤40, and A␤42 (Fig. 1G). Like F-N␤, the generation of the A␤ peptides was abolished in the presence of either of two GSIs, L-685,458 or DBZ (Fig. 1, H and I). We treated the cells with two well characterized NSAIDs, R-flurbiprofen and sulindac sulfide or vehicle, and studied the A␤ peptide profile. Both GSMs are known to modulate ␥-secretase activity by shifting the amino acid cleavage from positions 42 to 38 (19,38), and as expected, we observed a decrease in A␤42 accompanied by a relative increase in A␤38 compared with untreated cells (Fig. 2, A-C).
In line with previous findings, treatment with sulindac sulfide resulted in a relative decrease of A␤39 (19). Interestingly, we also observed an increase in A␤37 in cells treated with R-flurbiprofen, but not with sulindac sulfide. These results were confirmed by A␤ peptide analysis using MSD technology (Fig. 2D).
To further study how NSAID class GSMs affect N␤ generation, we treated HEK/FLAG-N⌬E cells with R-flurbiprofen, sulindac sulfide, or vehicle and monitored the generation of F-N␤ peptides using IP-MALDI-TOF. Because only F-N␤16 -25 could be inhibited by GSIs and not the shorter F-N␤12-15 (Fig. 1, C and D), the latter was excluded from peak analysis. Importantly, we did not detect any differences, neither increases nor decreases, in any F-N␤ peptide level between treated and untreated cells (Fig. 2, E and F). Collectively, these results suggest that first generation NSAID class GSMs seem to affect A␤ but not N␤ production (summarized in Fig. 2G).
AZ4126 Is a Novel Second Generation GSM That Modulates A␤ Formation Both in Vitro and in Vivo-We next characterized the pharmacology of a novel and chemically distinct second generation GSM, AZ4126, in both in vitro and in vivo assays. In a first series of experiments, HEK/APPswe cells were exposed to AZ4126, as well as the previously described GSM AZ1136 (32), and their effect on secreted A␤ production was investigated. Both AZ1136 and AZ4126 caused a reduction in A␤42 and A␤40, although the latter more potently than the former (AZ1136: IC 50A␤42 ϭ 990 Ϯ 150 nM, IC 50A␤40 ϭ 1400 Ϯ 50 nM; AZ4126: IC 50A␤42 ϭ 6.0 Ϯ 3.1 nM, IC 50A␤40 ϭ 22.5 Ϯ 8.7 nM; Fig. 3, A and B, and Ref. 32). This pattern of A␤42 and A␤40 inhibition is consistent with the behavior of second generation GSMs (21,32). Importantly, total A␤ was not affected at the concentration range at which modulation occurred, suggesting that these molecules are bona fide GSMs (Fig. 3A and Ref. 32). Further, the effect of AZ4126 was also investigated in vivo by administration of a single dose at different concentrations by oral gavage to wild type (C57BL/6) mice. Two hours after drug administration, both A␤40 and A␤42 levels were measured in diethylamine-extracted brain homogenates. AZ4126 reduced both A␤40 and A␤42 in a dose-dependent manner, with up to 36 and 40% for A␤40 and A␤42, respectively (Fig. 3, C and D). The concentration of AZ4126 in both plasma and brain followed a dose-dependent pattern (Fig. 3E). Thus, AZ4126 readily reaches the brain and exhibits the expected CNS A␤ lowering activity. Moreover, we have previously shown that several identified compounds, including AZ1136, display a competitive interaction to established second generation GSMs but not to first generation GSMs. To confirm that AZ4126 belongs to the second generation GSMs, we performed radioligand displacement binding experiments with an established tritiated second generation GSM, [ 3 H]AZ8349, as tracer (32). Although AZ4800, a structural derivative to AZ4126, and AZ4126 caused a displacement of [ 3 H]AZ8349, R-Flurbiprofen did not (Fig. 3F). Combined, the A␤ profile of displacement results both in vitro and in vivo together clearly manifests AZ4126 as a second generation GSM.

Second Generation GSMs Affect Both N␤ and A␤ Production-
Because we had found that the first generation GSM displays a clear differentiation with regard to A␤ and N␤ generation, we wanted to compare the effect of second generation GSM on the generation of these peptides. First, we analyzed the effect of AZ GSMs on A␤ peptide formation in HEK/APPswe cells using IP-MALDI-TOF MS. In line with MSD data (Fig. 3B), both AZ1136 and AZ4126 caused a differential reduction in A␤40 and A␤42 (2.3-and 6.4-fold decreases for A␤40 and 8.3-and 16.5-fold decreases for A␤42, respectively). The MALDI-TOF MS also revealed a unique A␤ pattern in response to the two AZ GSMs (Fig. 4, A-C): AZ1136 relatively increased both A␤37 and A␤39 (3,5 and 2.1 times, respectively), whereas A␤38 was unaffected, consistent with previously published results (32). AZ4126 also increased A␤37 but with a higher magnitude (4.8-fold), decreased A␤39 (1.8-fold), and marginally induced the levels of A␤38 (1.6fold). Importantly, MSD analysis resulted in the same A␤ pattern as the MS results (Fig. 3B).
The effect of AZ GSMs on N␤ formation was studied using HEK/FLAG-N⌬E cells and IP-MALDI-TOF. Interestingly, both AZ1136 and AZ4126 caused a relative decrease in F-N␤24 (2.0 and 2.3 times, respectively) and F-N␤25 (2.6-fold for both) levels, in comparison with untreated cells. However, we did not observe as distinct differences in the production of shorter F-N␤, as observed for shorter A␤ peptides in response to the same compounds. AZ1136 increased F-N␤18 1.7-fold, whereas AZ4126 increased F-N␤21 only 1.3-fold. Examined together, these data suggest that second generation AZ GSMs modulate N␤ production, although to a lower extent compared with their effect on A␤ production (summarized in Fig. 4G).
Second Generation GSMs Do Not Modulate NICD Formation-We finally explored whether the Notch S3 cleavage, liberating the more biologically relevant NICD peptide, also was affected by second generation GSMs, because we observed that they modulate N␤ production. To address this question, we used a previously developed immunocytochemistry assay based on HEK/FLAG-N⌬E cells and the C20 antibody that recognizes total NICD (32). Here, we included an anti-NICD antibody that specifically recognizes NICD beginning at position Val-1744 (V1744-NICD). We performed dose-response experiments and analyzed the amount of V1744-NICD in the nucleus relative to untreated cells and confirmed that AZ GSMs do not inhibit NICD formation (Fig. 5A). L-685,458 inhibited V1744-NICD formation with similar potency (IC 50 Val-1744 ϭ 10.0 nM) as previously determined for its effect on total NICD formation (IC 50 C20 ϭ 6.0 nM (32)). By calculating the ratio of the signal from both NICD antibodies, we could determine whether the relative amount of V1744-NICD to total NICD was affected. If GSMs modulate NICD formation, the amount of V1744-NICD and hence the ratio would change. None of the GSMs showed any change of V1744-NICD to total NICD ratio (Fig. 5B), suggesting that AZ GSMs do not modulate NICD formation. These results were also confirmed by Western blot analysis (Fig. 5C).

DISCUSSION
The insights in the mode of action of first and second generation GSMs are not fully understood, even though they represent a prioritized therapeutic approach in AD drug discovery. However, a growing body of evidence suggests that many second generation GSMs have a mode of action that is different from first generation NSAID class GSMs, because the former appear to target ␥-secretase instead of APP (21,(32)(33)(34)(35)(36). Even though the pharmacology of A␤ generation associated with GSMs has been thoroughly studied, less is known about their effect on other substrates, such as the Notch receptor. Here, we describe a head-to-head comparison of first and second generation GSMs affecting ␥-secretase processing of APP and Notch. Whereas both the first generation GSMs R-flurbiprofen and sulindac sulfide and the second generation GSMs AZ4126 and AZ1136 display A␤ modulation, only the second generations GSMs modify N␤ production. The modulatory effect of AZ4126 and AZ1136 on A␤ and N␤ differs both with regard to efficacy and profile, and the overall impact is more pronounced on A␤ generation compared with N␤ production. These data suggest that ␥-secretase-targeting GSMs are not only modulat-

. Characterization of novel AZ second generation GSMs in vitro and in vivo.
A, dose-response curve of AZ4126 displaying A␤ modulation, where A␤ formation is set relative to 0.5% Me 2 SO (100%) and 0.5 M L-685,458 (0%) controls. The curve represents the means of two experiments with error bars indicating S.D. B, detection of secreted A␤ peptides in conditioned medium from HEK/APPswe cells treated with AZ1136, AZ4126, or vehicle using MSD technology. A␤ peptide formation is determined as a percentage of total A␤ (i.e., the sum of A␤37-42). The bars represent the means of three experiments with error bars indicating S.D. C and D, AZ4126 reduces brain A␤40 (C) and A␤42 (D) levels in female C57BL/6 mice in a dose-dependent manner. Levels of A␤ in the brain were measured in mice 2 h after peroral administration of AZ4126 (10, 25, 50, or 75 mol/kg) or vehicle (n ϭ 7/group). Statistical analysis was one-way analysis of variance followed by Dunnett's multiple comparison test. *, p Յ 0.05; **, p Յ 0.01. E, plasma and brain exposure levels of AZ4126 in C57BL/6 mice 2 h after peroral dosing. Cp, plasma concentration; Cu,p, unbound plasma concentration; Cbr,corr, brain concentration corrected for plasma; Cu,br,corr, unbound brain concentration corrected for plasma. F, displacement of [ 3 H] AZ8349 GSM by different GSMs on sagittal rat brain sections. 5 nM of [ 3 H] AZ8349 was incubated in the absence or presence of 500 M R-flurbiprofen, 10 M of the reference compound AZ4800 (32), or 10 M AZ4126, and the binding in the autoradiograms was quantified as optical density (photo-stimulated luminescence/mm 2 ). AZ4126 as well as the reference compound AZ4800 displace [ 3 H] AZ8349 but not R-flurbiprofen, indicating distinct interaction sites. The bars represent the means of two experiments with error bars indicating S.D.
ing A␤ but also modify N␤ production and possibly also other ␥-secretase-dependent A␤-like peptides.
To compare the effect of GSMs on N␤ and A␤ formation, we established an antibody pulldown approach that we followed with MALDI-TOF MS analysis. This strategy enabled detection of the expected A␤ peptides, as well as 11 different F-N␤ spe-cies, ranging from F-N␤12 to F-N␤25, of which F-N␤16 -25 were generated in a ␥-secretase-dependent manner. These observations are well in line with previously reported results (15). MALDI-TOF analysis of A␤ peptides revealed some clear commonalities but also differences in the pharmacological profile among the compounds examined. Both AZ1136 and SEPTEMBER 21, 2012 • VOLUME 287 • NUMBER 39 AZ4126 cause a very potent reduction in A␤42 and A␤40 and a parallel increase in A␤37. In contrast, R-flurbiprofen and sulindac sulfide are associated with a clear relative reduction in A␤42 but not A␤40 and a concomitant increase of A␤38, suggesting that these different classes of GSM compounds modulate A␤ production through different mechanisms. In line with our previous study, we also found that AZ1136 treatment increases the relative levels of A␤39, whereas AZ4126 causes an increase in A␤38 (32). The pharmacological profiles of R-flurbiprofen and sulindac sulfide display differently; R-flurbiprofen primarily increases A␤37, whereas sulindac sulfide reduces A␤39. Thus, besides the differences in A␤ modulation between the first and second generation GSMs, clear differences are found within each class of GSMs with regard to A␤ modulation. Further studies are needed to explain the mechanism that gives rise to these differences and thus shed more light on the pharmacology of these different GSMs with respect to A␤ modulation.

Second Generation GSMs Modulate Both N␤ and A␤ Formation
The fact that both AZ1136 and AZ4126, but neither R-flurbiprofen nor sulindac sulfide, affect N␤ production is interesting. Similar to their effect on A␤40 and A␤42, AZ1136 and AZ4126 reduce both F-N␤24 and F-N␤25, but the relative effects are not as discriminating. Strikingly, even though AZ4126 decreases A␤40 and A␤42 approximately twice as much as AZ1136 at their respective tested concentration, both compounds appear to have the same efficacy in reducing F-N␤24 and F-N␤25. Regarding the shorter F-N␤ peptides, we observed that most of these peptides were similarly unaffected by the AZ GSMs. In contrast to their very clear relative elevation of A␤37, AZ41126 and AZ1136 do not cause a clear increase in the same F-N␤ peptide. However, AZ1136 increases F-N␤18 slightly, whereas F-N␤21 is elevated by AZ4126. Thus, it seems like both compounds share a general pharmacological profile on APP and Notch processing, specifically, a decrease in the longer A␤ and F-N␤ peptides (A␤40/A␤42 and F-N␤24/F-N␤25) and an increase in some of the shorter peptides, but that disparities do exist between the compounds, pertaining to the efficacy of the process and the specificity of which peptides are to be modulated. Our data therefore suggest that it is possible to generate ␥-secretase targeting GSMs that are preselective for A␤ over N␤ production. ␥-Secretase targeting is not restricted to these compounds, because there are several well characterized GSIs of different chemical classes that have been shown to bind to the ␥-secretase complex and exhibit different potencies in APP and Notch processing. Such features are assigned the Notch-sparing GSIs, such as BMS-708163, PF-3084014, Begacestat, ELND-006, and Semagacestat, which differ in their magnitude of selectivity of APP toward Notch with 3-1473 times (see review in Ref. 18).
␥-Secretase-mediated NICD formation is a key event in Notch receptor activation. Generating GSMs that do not affect NICD signaling is crucial toward developing tolerable AD therapeutics. The observation that many GSMs are Notch-sparing, that is, they do not affect the total amount of NICD generated, is applicable to the GSMs used in this study (Fig. 4A) (32). However, it may not be only the total amount of NICD generated that ought to be considered but also the specificity of the ␥-secretase cleavage event at the S3 site. Indeed, Tagami et al. (39) recently reported that NICD exists in two distinct forms at the N terminus, which results in NICDs with quite different stabilities and thus different signaling properties. In this study we could not find evidence for modulatory effect by AZ1136 and AZ4126 on Notch S3 cleavage, suggesting that the AZ GSMs are selective for S4 cleavage modulation. . The effect of second generation GSMs on A␤ and F-N␤ generation. A, MALDI-TOF MS spectrum of A␤ using 4G8 immunoprecipitated conditioned medium from HEK/APPswe cells treated with AZ1136, AZ4126, or vehicle. The intensities of the highest peak were set to 100% in the spectrum. B, A␤ peak distribution under the influence of second generation GSMs. Each A␤ peak is plotted as a percentage of total A␤ (i.e., the sum of A␤37-42). The bars represent the means of five or six experiments with error bars indicating S.D. C, scatter plots of the A␤ peptide distribution under the influence of second generation GSMs. The data are from five or six experiments and plotted as percentages of total A␤ (i.e., the sum of A␤37-42). D, MALDI-TOF MS spectrum of F-N␤ using ␣-FLAG immunoprecipitated conditioned medium from HEK/FLAG-N⌬E cells treated with AZ1136, AZ4126, or vehicle. The intensities of the highest peak were set to 100% in the spectrum. E, F-N␤ peak distribution under the influence of second generation GSMs. We excluded F-N␤12-15 peaks from peak analysis because they were not ␥-secretase-dependent. Each F-N␤ peak is plotted as a percentage of total F-N␤ (i.e., the sum of F-N␤16 -25). The bars represent the means of five or six experiments with error bars indicating S.D. F, scatter plots of the F-N␤18, 21, 24, and 25 peptide distributions under the influence of second generation GSMs. The data are from five or six experiments and plotted as percentages of total F-N␤ (i.e., the sum of F-N␤16 -25). G, a summary of the effect of second generation GSMs on A␤ and F-N␤ peptide formation. Our finding that second generation GSMs modulate both A␤ and N␤ supports the theory that the ␥-secretase complex is the molecular target of second generation GSMs (21,(32)(33)(34)(35)(36). In contrast, there has been no consensus as to what is the binding site of NSAID-based GSMs. Some studies report that these compounds interact with APP-derived C99 fragment (25)(26)(27), whereas others have challenged this hypothesis (28 -31). Our data, showing that R-flurbiprofen and sulindac sulfide affect A␤ but not N␤ production, indicate that APP is the target of NSAID-based GSMs. These data contrast with a previous report showing that sulindac sulfide decreases the ratio of N␤25/(N␤21 ϩ N␤25) (40). These discrepancies could be explained by differences in technical strategy. However, our results fit very well with a previous finding that NSAIDs such as sulindac sulfide interact and interfere with dimerization of the APP transmembrane domain, thereby affecting A␤ generation (27). APP forms dimers with the help of three different dimerization sites: two in the ectodomain and a third formed by three GXXXG motifs in the APP transmembrane domain. Recently, it was reported that a slight decrease in the dimerization strength of the GXXXG motifs gave rise to an enormous decline of A␤42 formation (41). Although it has been claimed that Notch can dimerize through its epidermal growth factor repeats in the ectodomain, Vooijs et al. (42) also report that most surface Notch molecules are monomeric. Thus, it is plausible that NSAIDs interact with mechanisms affecting dimerization of APP, resulting in no modulation in N␤. Further studies with competitive experiments and photo-probe labeling on NSAID-based GSMs are needed to fully understand the binding target of the NSAIDs.
Modulation by second generation GSMs may also affect other substrate-releasing A␤-like peptides, such as APP-like proteins 1 and 2, CD44, and interleukin-1 receptor II (43)(44)(45). Therefore, it will be of great interest to monitor these peptides during the research development of this class of drugs. However, the biological relevance of the A␤-like peptides is currently unclear. However, the APP-like protein 1 A␤-like peptide, which is less amyloidogenic than A␤42, is present in human cerebrospinal fluid and is purported to function as a surrogate marker for A␤42 in response to ␥-secretase-targeting drugs (46). Thus, the selectivity pattern of a given GSM should be a major consideration in biomarker development.
In summary, we report that second generation but not first generation GSMs affect N␤ production, although to a much lower level than they affect A␤ formation. However, second generation GSMs do not affect the Notch S3 cleavage site or the NICD formation, which is a crucial factor toward developing safe and tolerable AD therapeutics.