β-Amyloid Precursor Protein Mutants Respond to γ-Secretase Modulators*♦

Pathogenic generation of the 42-amino acid variant of the amyloid β-peptide (Aβ) by β- and γ-secretase cleavage of the β-amyloid precursor protein (APP) is believed to be causative for Alzheimer disease (AD). Lowering of Aβ42 production by γ-secretase modulators (GSMs) is a hopeful approach toward AD treatment. The mechanism of GSM action is not fully understood. Moreover, whether GSMs target the Aβ domain is controversial. To further our understanding of the mode of action of GSMs and the cleavage mechanism of γ-secretase, we analyzed mutations located at different positions of the APP transmembrane domain around or within the Aβ domain regarding their response to GSMs. We found that Aβ42-increasing familial AD mutations of the γ-secretase cleavage site domain responded robustly to Aβ42-lowering GSMs, especially to the potent compound GSM-1, irrespective of the amount of Aβ42 produced. We thus expect that familial AD patients carrying mutations at the γ-secretase cleavage sites of APP should respond to GSM-based therapeutic approaches. Systematic phenylalanine-scanning mutagenesis of this region revealed a high permissiveness to GSM-1 and demonstrated a complex mechanism of GSM action as other Aβ species (Aβ41, Aβ39) could also be lowered besides Aβ42. Moreover, certain mutations simultaneously increased Aβ42 and the shorter peptide Aβ38, arguing that the proposed precursor-product relationship of these Aβ species is not general. Finally, mutations of residues in the proposed GSM-binding site implicated in Aβ42 generation (Gly-29, Gly-33) and potentially in GSM-binding (Lys-28) were also responsive to GSMs, a finding that may question APP substrate targeting of GSMs.

Alzheimer disease (AD) 3 is the most common neurodegenerative disorder worldwide. The ␤-amyloid precursor protein (APP), a type I membrane protein, plays a central role in the pathogenesis of the disease (1). Sequential cleavage of APP by ␤and ␥-secretase generates the amyloid-␤ (A␤) peptide, which deposits as plaques in the brain of affected patients and represents one of the principal pathological hallmarks of the disease (1). ␥-Secretase is an intramembrane-cleaving protease complex, which cleaves the APP transmembrane domain (TMD) in a progressive, stepwise manner via cleavages at the ⑀-, -, and ␥-sites until it is sufficiently shortened to allow the release of A␤ from the membrane (2)(3)(4). A␤ peptides generated by ␥-secretase cleavage differ in their C termini. The major product released is A␤ 40 , whereas A␤ 38 and A␤ 42 represent minor species (1). The highly aggregation-prone, neurotoxic A␤ 42 is believed to be causative for AD by initiating a cascade of pathogenic events, which ultimately causes neurodegeneration and dementia (1). Increased production of A␤ 42 underlies the vast majority of mutations associated with familial AD (FAD), which manifests with a very early disease onset. The majority of FAD mutations have been found in PS1, the catalytic subunit of ␥-secretase (5), whereas only a few mutations were found in its homolog PS2. Few FAD-associated mutations were also found in APP, and those that affect ␥-secretase cleavage toward an increased A␤ 42 production localize to the C terminus of the APP TMD in the vicinity of the ␥-secretase cleavage sites. Fluorescence resonance energy transfer-based studies suggest that changes in the generation of A␤ 42 are due to alterations in the conformation of PS (6 -10).
Inhibition of A␤ 42 production is a major approach to therapeutically target AD. Selective A␤ 42 -lowering drugs, so-called ␥-secretase modulators (GSMs) such as certain non-steroidal anti-inflammatory drugs (NSAIDs), have been identified as promising and attractive alternatives to inhibitors of ␥-secretase, which target the active site and thus affect the processing of other physiologically important ␥-secretase substrates, such as Notch1 (11). GSMs inhibit A␤ 42 production with little effect on A␤ 40 generation and the processing of other important ␥-secretase substrates (12). Inhibition of A␤ 42 by these compounds is accompanied by an increased production of A␤ 38 (13). Because inverse modulators have also been identified (14), it was initially believed that the production of these peptides is interdependent, pointing to the possibility that A␤ 42 might represent the precursor of A␤ 38 . Evidence has been presented that A␤ 40 and A␤ 42 derive from two different product lines by stepwise cleavage roughly every three residues at positions ⑀49 -46 -␥43-␥40 and ⑀48 -45-␥42 of the A␤ domain, from which cleavages occur further downstream to generate A␤ 39 , A␤ 38 , and A␤ 37 (15,16), with A␤ 38 primarily originating from the A␤ 42 -generating product line (16). In addition, dimerization of the APP TMD mediated by its central GXXXG helixinteraction motif has been suggested to affect the formation of A␤ 42 (17). Substitution of the glycine residues to reduce APP TMD dimerization was shown to lower the production of A␤ 42 , whereas increasing that of A␤ 38 (17). Mechanistically, it was suggested that dimerization via the GXXXG motif imposes a sterical hindrance for ␥-secretase to proceed with stepwise processing such that A␤ 40 and A␤ 42 are normally released as final products. Decreasing dimerization strength would resolve this sterical constraint, now allowing ␥-secretase to efficiently proceed to more N-terminal cleavage sites, thus generating shorter A␤ species, mostly A␤ 38 (17). Interestingly, the GXXXG sequence is part of a GSM-binding site mapped to residues A␤29 -36 (GAIIGLMV) in the APP TMD (18). This region is known to be critical for A␤ aggregation (19 -21), and aggregation inhibitors interacting with this region also act as GSMs (18). However, a recent study demonstrated that dimerization per se might not be a factor that determines ␥-secretase cleavage specificity (22), and whether GXXXG mutants inhibit dimerization is controversial (23). In addition, the lack of a consistent response to GSMs regarding an inversely correlated production of A␤ 42 and A␤ 38 by PS FAD mutants observed earlier argued against a strict precursor-product relationship between A␤ 42 and A␤ 38 (24,25). Moreover, while this manuscript was in preparation, a biophysical study failed to demonstrate GSMbinding to the APP substrate, however, and suggested that the reported GSM-APP interaction (18) was unspecific (26).
Interestingly, lowering of A␤ 42 by GSMs is not effective for the majority of the PS FAD mutants investigated so far. In particular, aggressive FAD mutants that manifest with a very early disease onset due to their strongly increased A␤ 42 production do not respond at all, whereas increased A␤ 38 production was nevertheless still observed (24,25). Whether or not APP FAD mutants respond to GSMs has not been conclusively studied yet, however (27). To address this question specifically as well as to gain a greater understanding of the mode of action of GSMs, in particular in light of the controversial data regarding the mechanism and binding site, we decided to study the action of GSMs on APP cleavage by ␥-secretase in more detail. Analysis of FAD-associated APP mutations in the ␥-secretase cleavage site region together with systematic phenylalanine-scanning mutagenesis of this region showed that all mutants respond robustly to the potent ␥-secretase modulator GSM-1 (24), which was used as the principal GSM in this study, independent of the amount of A␤ 42 produced by the mutant. Thus, unlike the majority of FAD mutants in PS1, APP TMD mutants at the ␥-secretase cleavage sites are susceptible to GSMs, suggesting that the respective APP FAD mutant carriers should positively respond to GSM-based AD therapies. However, A␤ 42 and A␤ 38 generation was not strictly coupled, further arguing against a general mechanistic relationship for the generation of these A␤ species. Furthermore, GSM treatment was also found to lower A␤ 39 and A␤ 41 generation, demonstrating considerable imprecision and flexibility in the modulation of ␥-secretase cleavage specificity in response to GSMs. Finally, we also investigated the impact of mutants close to or within the proposed GSMbinding site in the APP TMD. Because such mutants give rise to mutant A␤ species, these were also investigated in cell-free assays using purified ␥-secretase and APP substrate (28) to circumvent potentially altered cellular metabolic fates of the mutant peptides in cultured cells. In both cell-based and cellfree assays, we found that GSM treatment could still effect changes in the ratios of A␤ species generated, a finding that may question GSM-binding within the A␤ domain.
Cell Culture and cDNA Transfections-Human embryonic kidney (HEK) 293 cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum and penicillin/streptomycin on poly-L-lysine-coated plates. Cells were plated at a density of 200,000 cells/24-well plate or 1,000,000 cells/6-well plate, and the following day, cells were transiently transfected with the indicated APP cDNAs using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Suspension-adapted HEK293S cells were cultivated in instrumented 10-liter bioreactors using a Roche Applied Science proprietary serum-free hydrolysate-containing culture medium.
Analysis of Secreted A␤ from Cultured Cells-Following transfection, HEK293 cells were incubated for 24 h before media change and overnight incubation for 16 h in the presence of sulindac sulfide, flurbiprofen, or fenofibrate (all Sigma), GSM-1 (kind gift of Dr. Karlheinz Baumann, Roche), or vehicle control (dimethyl sulfoxide (DMSO)). Conditioned media were then collected and immediately analyzed by sandwich immunoassay to quantify A␤ species, or following immunoprecipitation, subjected to Tris-Bicine urea SDS-PAGE or mass spectrometry analysis. For A␤ quantitation, drug treatments were performed in triplicate, and all media samples were measured in duplicate for A␤ 38 , A␤ 40 , and A␤ 42 . For the analysis of modulation, data are always plotted as the percentage of change in the concentration of A␤ species from vehicle-treated cells, which are normalized to 100% for each cell line.
␥-Secretase in Vitro Assays-␥-Secretase was purified as described previously (28) except that HEK293S cells (kind gift of Georg Schmid and Elvira da Silva, Roche) were used as an enzyme source. In vitro ␥-secretase activity was assessed as described using purified ␥-secretase (Q-Sepharose eluate) and purified WT and mutant C100-His 6 substrates in the presence or absence of GSM-1 or fenofibrate (28).
Quantification of A␤-Secreted A␤ peptides in conditioned medium were quantified by a sandwich immunoassay using the Meso Scale Discovery SECTOR Imager 2400 as described previously (24). Quantification of total A␤ used essentially the same procedure except that for detection, 4G8 mouse monoclonal and ruthenylated anti-mouse antibodies were used in combination. For more sensitive detection of A␤ species (for the Phe mutants), the Meso Scale Discovery A␤ triplex sandwich immunoassay was used. Here, Meso Scale Discovery C-terminal specific antibodies were instead prespotted into each well, and a ruthenylated 6E10 antibody was used as the detection antibody. Meso Scale Discovery A␤ peptide standards were used for the Meso Scale Discovery triplex immunoassay. A␤ peptides generated by ␥-secretase in vitro assays were quantified using the Meso Scale Discovery sandwich immunoassay as described (28).

SDS-PAGE and Mass Spectrometry Analysis of A␤-Secreted
A␤ was analyzed from medium conditioned overnight for 16 h by combined immunoprecipitation/immunoblotting using antibodies 3552/2D8 followed by Tris-Bicine urea SDS-PAGE (33). To analyze A␤ by mass spectrometry, A␤ species were immunoprecipitated from conditioned media or from ␥-secretase in vitro assays using antibody 4G8 and subjected to matrixassisted laser desorption/ionization-time of flight mass spectrometry analysis as described previously (24,28).

FAD Mutants in the ␥-Secretase Cleavage Site Region
Respond to GSMs-To address the question whether APP FAD mutants respond to GSM treatment, we introduced the Austrian T43I (T714I), Florida I45V (I716V), and London V46I (V717I) (34 -36) mutations into APPsw-6myc, a well characterized and frequently used APP substrate (31), which was used as backbone for these and all other mutants of the APP TMD ( Fig. 1A) analyzed in this study. The cDNA constructs were transiently transfected into HEK293 cells, and levels of secreted A␤ species generated by WT APP and the APP FAD mutants were analyzed by a highly sensitive specific A␤ immunoassay, which allows the detection and quantitation of A␤ 38 , A␤ 40 , and A␤ 42 FIGURE 1. Effect of GSM-1 on A␤ species of APP FAD mutants. A, schematic of the APP amino acid sequence encompassing the A␤ peptide and the transmembrane region of APP (underlined). Amino acids are numbered from position 1 of the A␤ peptide, and arrows indicate the positions of ␥-secretase cleavage. The location of APP FAD mutations selected for analysis together with the corresponding mutant amino acids are indicated above the sequence. The amino acids subjected to phenylalanine scan are boxed in gray and contain the I45F and V46F mutants as additional FAD mutations. The reported GSM-binding site in APP is shown in italics. B, sandwich immunoassay of A␤ 38 , A␤ 40 , and A␤ 42 species from conditioned media of cells overexpressing WT or the indicated APP FAD mutants. A␤ 42 is increased for the three FAD mutants analyzed. Strikingly, in addition to a dramatic increase in A␤ 42 for T43I, A␤ 38 is also increased at the expense of A␤ 40 . Each species is plotted as a percentage of the total A␤ (A␤ 38  species (24,29). As expected, the ratios of A␤ 42 to total A␤ (i.e. the sum of A␤ 38 , A␤ 40 , and A␤ 42 ) were increased for all mutants (Fig. 1B). The strongest increase was observed for the T43I mutant. Unexpectedly, this mutant also showed a strongly increased A␤ 38 /A␤ total ratio, whereas that of A␤ 40 /A␤ total was reduced. This behavior was not observed for the I45V and V46I FAD mutants and appeared to be characteristic for the T43I mutant. We thus conclude that certain FAD mutants occurring in the APP TMD can increase A␤ 38 and A␤ 42 in parallel.
We next screened various GSMs to see whether the increased levels of A␤ 42 produced from these mutants could be lowered. As shown in Fig. 1C, all GSMs were effective. GSM-1, a previously described GSM that is effective in the low micromolar range (24,28), was overall the most potent compound with regard to modulation of the APP mutants. This GSM was capable of strongly reducing A␤ 42 production from WT APP as well as from all three FAD mutants. The reduction of A␤ 42 levels observed was ϳ90% as compared with the untreated controls. Consistent with previous results (24), GSM-1 increased the levels of A␤ 38 produced from WT APP. Likewise, a robust increase of A␤ 38 was also observed for the T43I, I45V, and V46I mutants. The NSAIDs sulindac sulfide and flurbiprofen were capable of reducing A␤ 42 production for these mutants to a similar extent to WT APP, although flurbiprofen was more potent in this regard. Sulindac sulfide was more potent than flurbiprofen with regard to the modulation of A␤ 38 levels, although all mutants responded by increasing A␤ 38 upon treatment with either compound (Fig. 1C). As GSM-1 was the most effective modulator, we focused on this compound for the analysis of A␤ species by Tris-Bicine urea SDS-PAGE, which allows an effective separation of A␤ species. The modulatory effects were confirmed, and in addition, this analysis showed that the production of A␤ 39 was reduced by GSM-1 treatment as well (Fig. 1D). Taken together, these data show that FAD mutants within the APP TMD that change the specificity of ␥-secretase cleavage and thereby increase the production of A␤ 42 are susceptible to different GSMs. GSM-1-mediated inhibition of A␤ 42 generation is accompanied by an increased production of A␤ 38 , a strong inhibition of A␤ 39 , but has little effect on A␤ 40 . Since GSM-1 elicited the most potent effects with regard to both A␤ 42 and A␤ 38 modulation, this compound was used as the principal GSM for all subsequent modulation experiments in this study.
Phenylalanine Mutants of the ␥-Secretase Cleavage Site Region Respond to GSM-1-To investigate whether mutations of the ␥-secretase cleavage site region of APP are generally susceptible to GSM-1 as shown above for a subset of FAD mutants, we next analyzed previously described phenylalanine mutants, which span the region between the ␥and ⑀-cleavage sites and thus allow a systematic analysis of the GSM-1 response to mutants within this region (37,38). These well characterized Phe mutants cover the sequence from A␤43-51 of the A␤ domain and include two other FAD mutants, the Spanish I45F (I716F) (39) and the V46F (V717F) Indiana mutation (40). Analysis of the profile of the A␤ species generated by these mutants by Tris-Bicine urea SDS-PAGE was entirely consistent with that of the previous reports (37, 38) ( Fig. 2A). Each mutant affected the profile of A␤ species produced by ␥-secretase in an individual and characteristic manner. In agreement with the previous results (37,38), the most striking changes were observed for the V44F, I45F, I47F, and V50F mutants. The I45F mutant produced the lowest amount of A␤ 40 and the highest amount of A␤ 42 , whereas the V50F mutant produced almost exclusively A␤ 40 . Similarly, the V44F and I47F mutants produced very little A␤ 42 , but interestingly, both gave rise to the production of an alternative A␤ species, which migrated somewhat slower than A␤ 42 and apparently represented A␤ 41 (37,38). Consistent with the previous reports (37,38), the T43F-V46F mutants produced higher levels of the shorter A␤ variants A␤ 38 and A␤ 39 than the T48F-V50F mutants. The highest level of A␤ 38 was observed for the V44F mutant, which did not generate detectable levels of A␤ 39 . Only small amounts of A␤ 37 were produced for WT APP and all of the Phe mutants.
Having confirmed the characteristic A␤ profile for each mutant, we next investigated whether and how the mutants would respond to GSM-1 treatment. Following drug treatment, changes in A␤ 42 and A␤ 38 were assessed by the A␤ immunoassay. As shown in Fig. 2B, as compared with the untreated controls, GSM-1 potently lowered the levels of A␤ 42 of WT APP and all the Phe mutants by ϳ70 -80%, even for the V44F mutant, which produced only extremely small amounts of A␤ 42 . Strikingly, the Phe mutants behaved differently in their response to GSM-1 with regard to its potency in increasing the levels of A␤ 38 (Fig. 2B). Although WT APP showed the expected robust increase of A␤ 38 , the T43F, V44F, and I45F mutants were less responsive to GSM-1 treatment with respect to A␤ 38 . The FAD-associated V46F as well as the M51F mutant responded to GSM-1 similarly to WT APP. In contrast, GSM-1 treatment induced the production of considerably higher levels of A␤ 38 in the I47F, T48F, L49F, and V50F mutants, the latter mutant showing the maximal increase of A␤ 38 among these mutants (ϳ6-fold increase as compared with control). Interestingly, Tris-Bicine urea SDS-PAGE analysis revealed that GSM-1 effected a strong decrease of A␤ 41 for the V44F and of A␤ 39 for the V46F mutant, which were selected to analyze the modulation of alternative A␤ species (Fig. 2C). Consistent with the results above, A␤ 38 was increased for both mutants in response to GSM-1, and A␤ 42 was decreased by GSM-1 treatment for the V46F mutant (Fig. 2C). Mass spectrometry analysis confirmed these results (Fig. 2D). Taken together, all APP Phe mutants responded robustly to the A␤ 42 -lowering capacity of GSM-1, further supporting our notion that A␤ 42 produced from APP mutant carriers harboring A␤ 42 -increasing mutations in the ␥-secretase cleavage site domain can be expected to be targetable by GSMs. Furthermore, these data show that additional ␥-secretase cleavages can be modulated, suggesting substantial flexibility in the modulation of ␥-secretase cleavage specificity.
Mutants of the GXXXG Motif in the APP TMD Respond to GSMs-We next investigated whether mutation of the glycine residues of the GXXXG motif, which had been implicated in the production of A␤ 38 and A␤ 42 (17) and which lie within the proposed GSM-binding site of APP (18), would affect GSMinduced changes of ␥-secretase cleavage specificity. APPsw-6myc constructs containing the G29A, the G33A, or the stronger G33I substitution described previously (17) were transiently transfected into HEK293 cells, and the ratios of the secreted A␤ 38 , A␤ 40 , and A␤ 42 species to total A␤ were examined by the A␤ immunoassay to assess changes in ␥-secretase cleavage specificity. As shown in Fig. 3A, the G29A mutant showed a normal A␤ 40 /A␤ total ratio but an increased A␤ 38 / A␤ total ratio and a decreased A␤ 42 /A␤ total ratio. The G33A mutant displayed an even higher A␤ 38 /A␤ total ratio, but unlike the G29A mutant, it showed a normal A␤ 42 /A␤ total ratio similar to that of the WT control. In further contrast to this mutant, a decreased A␤ 40 /A␤ total ratio was immediately apparent for the G33A mutant, and this effect was observed even more strongly for the G33I mutant. For this mutant, A␤ 38 became the principal species representing ϳ90% of the total A␤ measurable as compared with A␤ 40 , which represented only ϳ10% of total A␤, whereas A␤ 42 was almost undetectable. Thus, the G33I mutant showed a strong change in ␥-secretase cleavage specificity,  38 (lower panel) in cells expressing WT or the indicated APP Phe mutants. The more sensitive Meso Scale Discovery triplex assay was used here to detect A␤ peptides with low abundance such as A␤ 42 for the V44F mutant. All mutants respond to treatment by decreasing A␤ 42 to a similarly dramatic extent. However, the increase in A␤ 38 depends upon the amino acid position within the TMD, with a clear pattern of increasing response the further away the Phe mutation is from position 44. Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, **, p Ͻ 0.01, *, p Ͻ 0.05. C, Tris-Bicine urea SDS-PAGE immunoblot of WT and V44F and V46F mutant A␤ species in the presence of GSM-1 or vehicle. For the V44F mutant, the species migrating as A␤ 41 could be lowered upon GSM-1 treatment, as could the A␤ 39 species produced by the V46F mutant. D, mass spectrometry of V44F and V46F mutant A␤ species in the presence of GSM-1 or vehicle. For the V44F mutant, the increased generation of A␤ 41 in place of A␤ 42 was confirmed, and this species could be lowered by GSM-1 treatment. The spectra for the V46F mutant also confirm that A␤ 38 was indeed robustly increased upon GSM-1 treatment and that both A␤ 39 and A␤ 42 are lowered. The intensities of the A␤ 40 peaks were set at 100% in the spectra. DMSO, dimethyl sulfoxide.
apparently at the expense of the normally major A␤ 40 species. These results were confirmed using the C99-6myc substrate (31) to preclude any effects of dimerization of full-length APP via its ectodomain that may have affected its downstream processing and to rule out that the Swedish mutation at the ␤-secretase cleavage site of APPsw-6myc may have influenced the results (Fig. 3A). We next investigated whether the Gly mutants might also affect other A␤ species, which were not measured by our A␤ immunoassay. Tris-Bicine urea SDS-PAGE analysis was, however, hampered by an aberrant electrophoretic migration behavior of the mutant peptides, which precluded a clear assessment of the identity of the bands observed and revealed altered biochemical properties of the mutant peptides as compared with WT A␤ (data not shown). We therefore analyzed the profile of A␤ species by mass spectrometry, which confirmed the A␤ profiles of WT and the Gly mutants and additionally revealed substantial levels of A␤ 37 and even shorter species for the G33I mutant (Fig. 3B) in full agreement with previous results (17). Thus, although overall the Gly mutants favor the production of shorter species, mostly A␤ 38 , they do not consistently change the cleavage specificity of ␥-secretase toward a reduced production of A␤ 42 but rather affect that of  40 , and A␤ 42 species that were isolated from conditioned media of cells overexpressing WT and the indicated glycine mutants of APPsw-6myc (upper panel) and C99-6myc (lower panel). For both substrates, the increase in A␤ 38 correlates with the reported decrease in dimerization strength. This correlates with a decrease in A␤ 40 , but there is no correlation to A␤ 42 . Each species is plotted as a percentage of the total A␤ (A␤ 38 ϩ A␤ 40 ϩ A␤ 42 ) measured for each cell line. Bars represent the mean of 3 experiments with error bars indicating the S.E. B, mass spectrometry of A␤ species from WT and the indicated Gly mutants of APP. For A␤ 38 , A␤ 40 , and A␤ 42 , the spectra closely reflect the immunoassay measurements. However, Gly-33 substitutions exhibit increased A␤ 39 production, and G29A and G33I both exhibit increased A␤ 37 production. Indeed, for G33I, there is a general shift toward shorter A␤ peptides, with A␤ 37 the predominant species. The intensities of the highest A␤ peaks were set at 100% in the spectra. Magnified parts of the spectra are shown on the right-hand side to better visualize the A␤ 42 peaks. C, sandwich immunoassay showing the effect of 1 M GSM-1 treatment on A␤ 42 and A␤ 38 in cells expressing WT or the indicated APP Gly mutants. Both mutants respond to treatment by decreasing A␤ 42 by ϳ70% and increasing A␤ 38 , although the magnitude of the increase depends upon the starting ratio of A␤ 38 . Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, **, p Ͻ 0.01, *, p Ͻ 0.05. D, sandwich immunoassay showing the effect of 50 M fenofibrate treatment on A␤ 42 , A␤ 38 , and A␤ 40 in cells expressing WT or the G33I mutant of APP. Both WT and the G33I mutant respond to treatment by increasing A␤ 42 , and this increase is more pronounced for the mutant. A␤ 38 could also be decreased for both WT and G33I, although the decrease was less pronounced for the mutant. Strikingly, A␤ 40 could also be increased for G33I, in contrast to the decrease that is observed for WT. Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, **, p Ͻ 0.01, *, p Ͻ 0.05. A␤ 40 . This was also reflected by the average amounts of the individual A␤ species measured in additional experiments (Table 1).
Having determined the changes in ␥-secretase cleavage specificity for the Gly mutants, we next asked whether they would show a GSM response. A␤ 42 levels could be robustly lowered for the G29A and G33A mutants by GSM-1 treatment (Fig. 3C). Concomitantly, GSM-1 treatment caused an increase of A␤ 38 levels for the G29A and G33A mutants. Probably due to the higher A␤ 38 starting levels of the mutants, the increase was attenuated as compared with WT APP. Given its favored production of shorter A␤ species such as A␤ 38 , the G33I mutant was tested for its ability to respond to the inverse GSM fenofibrate (Fig. 3D). This showed that fenofibrate treatment caused a dramatic increase of A␤ 42 (Fig. 3D), which had almost been undetectable for this mutant at baseline (see above). Furthermore, A␤ 38 could still be lowered, despite its high baseline level for this mutant. Surprisingly, A␤ 40 was increased in response to fenofibrate treatment as well (Fig. 3D), a modulation that, to our knowledge has so far not been observed for inverse GSMs. Thus, these data show that substitution of the glycine residues of the GXXXG motif does not interfere with the modulation of A␤ levels by GSMs.
APP TMD Mutants Respond to GSMs in a Cell-free ␥-Secretase Assay Using Purified Components-A potential complication in the analysis of A␤ produced from Gly mutant APP is that these particular A␤ peptides are mutated within the A␤ domain itself and that their altered biochemical properties might potentially have an impact on the levels of Gly mutant A␤ peptides detectable in cultured cells downstream of production (e.g. altered secretion, enhanced degradation, or aggregation). In particular, the G33I mutant A␤ 42 has been described as a highly aggregation-prone peptide (21). We therefore also carried out cell-free assays using purified recombinant APP C-terminal fragment-based C100-His 6 substrates containing selected Phe mutants, the G29A, G33A, or G33I mutants, and purified lipid-reconstituted ␥-secretase (28). We first monitored the production of A␤ peptides from C100-His 6 substrates containing the I45F or V50F mutants, i.e. substrates that should generate high amounts of both A␤ 38 and A␤ 42 and low amounts of A␤ 40 (I45F) or almost exclusively A␤ 40 (V50F). Importantly, in contrast to the Gly mutants within the A␤ domain, A␤ species generated from the Phe mutants are WT in sequence. Therefore, altered effects with respect to the cellular metabolism of A␤ do not apply to these mutations. As shown in Fig. 4A, the production of A␤ 38 , A␤ 40 , and A␤ 42 peptides as assessed by A␤ immunoassay from WT, I45F, and V50F C100-His 6 substrates was consistent with the results obtained for these mutants in cultured cells and the previous reports by others (37,38). We next monitored whether the I45F and V50F mutants would respond to GSM-1 treatment, which indeed proved to be the case (Fig.  4B). The responses of the two mutants to GSM-1 with respect to A␤ 42 and A␤ 38 generation were very similar to those observed in cultured cells, thus confirming the results described above and validating the cell-free assay for the analysis of APP TMD mutant substrates.
Analysis of the A␤ species generated in the in vitro assay revealed that the Gly mutants caused an increased relative production not only of A␤ 38 , but surprisingly also of A␤ 42 , whereas concomitantly reducing that of A␤ 40 (Fig. 5A), although we noticed a greater variability of G33I mutant A␤ 42 in this assay (Table 1). Mass spectrometry analysis of mutant A␤ peptides confirmed this unexpected result and revealed that G29A, and in particular the G33I mutant, addi-

Levels of the individual A␤ peptides (A␤ 38 , A␤ 40 , A␤ 42 ) measured for mutants of the proposed GSM-binding site
Levels of individual A␤ species were measured by A␤ sandwich immunoassay using 2D8 as capture antibody and C-terminal-specific antibodies to A␤ 38 , A␤ 40 , and A␤ 42 for detection (n ϭ number of independent experiments). Total A␤ levels measured by using 4G8 as a capture antibody (which detects all species including e.g. A␤ 37 ) were similar to those of the WT control except for the G33I mutant, which showed an ϳ30 -40% reduction of total A␤ (data not shown). Surprisingly, A␤ could not be detected for the K28E mutant in cultured cells and the cell-free assay system by SDS-PAGE (data not shown). This may reflect a potential alteration of the folding of Lys-28 mutant A␤ peptides (45), which could possibly have affected their detectability in some of the previous reports (22,46,47).  tionally generated A␤ 37 . Although A␤ 42 was robustly produced by ␥-secretase for the G29A, G33A, and G33I mutants, it was less well detectable by mass spectrometry for the latter mutant. This is most likely because of very low ionization efficiency due to the increased hydrophobicity of this mutant peptide (Fig. 5B). Thus, although increased generation of A␤ 38 from Gly mutants was a consistent observation in both cultured cells and the cell-free in vitro assay using purified components, the A␤ 42 levels detected differed between these two systems. Assessment of A␤ production in the latter assay system shows that ␥-secretase can in fact generate substantial amounts of A␤ 42 from the Gly mutant substrates as solely A␤ production is assessed in this system. We next assessed the response of the Gly mutants to GSM-1. Consistent with previous results (28), WT APP showed a clear response with lowered A␤ 42 and increased amounts of A␤ 38 . Again, GSM-1 effected a change in the A␤ species produced by the Gly mutants by lowering the production of A␤ 42 and increasing that of A␤ 38 (Fig. 5C). This was clearly observed for the G29A mutant and in attenuated form for the G33A mutant ( Fig. 5C) but unexpectedly, not observed for the G33I mutant, which instead showed an ϳ20% reduction of A␤ 40 in response to GSM-1 (data not shown). Although A␤ 42 production could not be increased further by fenofibrate treatment for the G33I mutant, it responded effectively to this inverse GSM with respect to both A␤ 38 and A␤ 40 generation (Fig. 5D), similar to the results obtained for this mutant in cultured cells. Taken together, these data show that the Gly mutants are susceptible to pharmacological modulation of ␥-secretase cleavage specificity both in cultured cells as well as in the cell-free system, FIGURE 5. Profile of A␤ peptides generated in vitro from purified C100-His 6 GXXXG mutants and their response to GSMs. A, sandwich immunoassay of A␤ 38 , A␤ 40 , and A␤ 42 species that were generated in cell-free assays (28) from WT and the indicated Gly mutant substrates. Comparable with results from cultured cells (Fig. 3A), the increase in A␤ 38 correlates with a decrease in A␤ 40 for these mutants. However, unexpectedly and in contrast to the cell culture data, robust amounts of A␤ 42 could be detected for all Gly mutants. Each species is plotted as a percentage of the total A␤ (A␤ 38 ϩ A␤ 40 ϩ A␤ 42 ) measured for each substrate. Bars represent the mean of 13-18 experiments with error bars indicating the S.E. B, mass spectrometry of A␤ species that were generated in cell-free assays (28) from WT and the indicated Gly mutant substrates. The data reflect the immunoassay measurements for A␤ 38 , A␤ 40 , and A␤ 42 , with the exception of low detectable amounts of A␤ 42 for the G33I mutant, possibly due to reasons described under "Results." In parallel with a decrease in A␤ 40 , a mild-robust increase in A␤ 37 is observed for the Gly mutants in vitro. The intensities of the highest A␤ peaks were set at 100% in the spectra. A magnified part of the spectrum of the G33I mutant is shown on the right-hand side to better visualize the A␤ 42 peak. C, sandwich immunoassay showing the effect of 2.5 M GSM-1 treatment on A␤ 42 and A␤ 38 generated from WT and Gly mutant substrates in cell-free assays. In parallel with the observed increase in A␤ 38 for the G29A and G33A mutants, the response of A␤ 38 to GSM-1 is significant but attenuated. Likewise, the decrease of A␤ 42 is diminished but still significant for these mutants. Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, **, p Ͻ 0.01, *, p Ͻ 0.05. D, sandwich immunoassay showing the effect of 150 M fenofibrate treatment on A␤ 42 , A␤ 38 , and A␤ 40 generated from WT and G33I mutant substrates in cell-free assays. Here, A␤ 42 does not respond to modulation, whereas both A␤ 38 and A␤ 40 do. Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, **, p Ͻ 0.01, *, p Ͻ 0.05. n.s. denotes not significant.

A␤
despite differences in the relative amounts of each A␤ species generated.
The K28E APP Mutant Responds to GSMs-It has been speculated that GSMs may shift the position of the substrate in the membrane plane relative to the ␥-secretase active site. A␤ 42raising GSMs and A␤ 42 -lowering GSMs would either pull the substrate out from the membrane or, respectively, sink it into the membrane (18). Interestingly, all so far described A␤ 42lowering GSMs have a carboxyl group, whereas A␤ 42 -raising GSMs lack this group. The positively charged lysine 28 residue of the A␤ domain located directly at the extracellular border of the membrane could potentially form a salt bridge with the carboxyl group of A␤ 42 -lowering GSMs, which as a consequence might position the ␥42 site away from the ␥-secretase active site. To address such a potential mechanistic contribution of Lys-28 in GSM binding, we substituted this residue by the negatively charged glutamate. As shown in Fig. 6A, assessment of the A␤ ratios revealed that A␤ 38 was the major species detected for the K28E mutant (ϳ60% of the total A␤). The A␤ 40 /A␤ total ratio was strongly reduced, as was the A␤ 42 /A␤ total ratio, indicating a strong preference for ␥-secretase cleavage at the ␥38 site (see Table 1 for average A␤ levels measured in additional experiments). Mass spectrometry confirmed this result and additionally revealed a substantial increase of the shorter A␤ species A␤ 37 and A␤ 33 (Fig. 6B). Despite the already high A␤ 38 levels, they could still be significantly increased by GSM-1 treatment (Fig. 6C). The levels of A␤ 42 were too low such that changes were at the limit of detection, and no significant changes were observed for the levels of A␤ 40 (data not shown). These data suggest that the proposed electrostatic interaction does not play an essential role for the mechanism of ␥-secretase modulation. Finally, treatment of the K28E mutant with fenofibrate showed that A␤ 42 could be tremendously increased, whereas A␤ 38 could still be substantially lowered. Interestingly, fenofibrate caused an increase of A␤ 40 as well (Fig. 6D). This result shows that the K28E mutant responds to GSMs of two different classes.
As above, to investigate modulation of ␥-secretase cleavage by an independent assay, which assesses production in the absence of cellular metabolism, we also investigated the K28E mutant in the in vitro system using a C100-His 6 K28E mutant substrate. In the cell-free assay, a slightly increased A␤ 42 / A␤ total ratio and a decreased A␤ 40 /A␤ total ratio were observed for this mutant, whereas the A␤ 38 /A␤ total ratio was similar to that of the WT substrate ( Fig. 7A; see Table 1 for average A␤ levels of all experiments performed). Mass spectrometry confirmed this A␤ profile (Fig. 7B). Thus, unlike in cultured cells, the K28E mutant behaved as a comparably normal ␥-secretase substrate in the cell-free assay. GSM-1 behaved as a modulator on the K28E mutant, lowering A␤ 42 and increasing A␤ 38 production. As compared with WT, the response of the K28E mutant to GSM-1 was attenuated, however (Fig. 7C). As shown in Fig. 7D, we also observed inverse modulation for the K28E mutant with fenofibrate. Taken together, despite the observed differences in the amounts of A␤ species detected in cultured cells versus the cell-free system, the K28E mutant was clearly responsive to GSMs both in cultured cells and in vitro using purified components, suggesting that Lys-28 of the A␤ domain is unlikely to play a major role for the action of GSMs.

FIGURE 6. Profile of A␤ peptides generated in cells from the APP K28E mutant and its response to GSMs.
A, sandwich immunoassay of A␤ 38 , A␤ 40 , and A␤ 42 species that were isolated from conditioned media of cells overexpressing WT and the K28E mutant of APP. This substitution results in a dramatic increase in A␤ 38 and a decrease in A␤ 40 and A␤ 42 , the latter being close to the detection limit of the immunoassay. Each species is plotted as a percentage of the total A␤ (A␤ 38 ϩ A␤ 40 ϩ A␤ 42 ) measured for each cell line. Bars represent the mean of 3 experiments with error bars indicating the S.E. B, mass spectrometry of A␤ species from WT and the K28E mutant of APP. The spectra closely reflect the immunoassay measurements for A␤ 38 , A␤ 40 , and A␤ 42 , but similar to the G33I mutation, there is again a shift toward shorter A␤ peptides, with A␤ 37 the predominant species and A␤ 33 also becoming a major species. The intensities of the highest A␤ peaks were set at 100% in the spectra. Magnified parts of the spectra are shown on the right-hand side to better visualize the presence or absence of A␤ 42 40 , and A␤ 38 in cells expressing WT or the K28E mutant of APP. Both WT and the K28E mutant respond to treatment by increasing A␤ 42 , and this increase is more pronounced for the mutant. Strikingly, A␤ 40 could be increased for K28E, similar to G33I, and A␤ 38 could also be effectively decreased for both WT and K28E. Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, **, p Ͻ 0.01, *, p Ͻ 0.05.

DISCUSSION
In this study, we have investigated the impact of GSMs on a variety of APP mutants to better understand their mode of action as well as the cleavage mechanism of ␥-secretase. A summary of our principal findings is shown in Table 2. We establish that naturally occurring pathogenic mutations of the ␥-secretase cleavage site region in the APP TMD, which affect the precision of ␥-secretase cleavage toward an increased production of A␤ 42 , respond to GSM treatment. These mutants even respond as strongly as WT APP following treatment with GSM-1, a well characterized potent GSM (24), which was selected as the principal GSM in this study. FAD patients with mutations in APP that affect ␥-secretase cleavage should therefore be susceptible to GSM treatment. In addition, phenylalanine-scanning mutagenesis analysis of the ␥-secretase cleavage site region revealed responsiveness to GSM-1 for all mutants. We found that even very strong mutants among those, such as the I45F mutant that produces high amounts of A␤ 42 , responded to GSM-1. The I45F mutant represents the most aggressive APP FAD mutation identified so far with an extremely early disease onset of 31 years (39). Based on our data, we therefore conclude that treatment with GSMs might provide a successful therapeutic option also for these mutant carriers and others of the ␥-secretase cleavage site domain with pathogenic A␤ 42 production that may be identified in the future. Interestingly, our results contrast with those obtained recently for PS mutations (24,25). Most of the PS FAD mutations investigated were not responsive to NSAIDs that act as A␤ 42 -lowering GSMs, such as sulindac sulfide. In particular, very strong and aggressive mutations producing high levels of A␤ 42 , such as PS1 L166P, and others, were not susceptible to the A␤ 42 -lowering capacity of sulindac sulfide. This mutant was also not responsive to the more potent GSM-1 (24). Thus, our data indicate that the mode of action of GSMs is different for FAD mutants of APP and PS, i.e. different for substrate and protease. We also note that AD mouse models expressing human APP FAD mutants as transgene, such as some of the ones investigated here (e.g. V46I or V46F), should be more suitable for in vivo validation of GSMs than models that additionally express strong PS FAD mutations.
Recent data suggested that GSMs target the substrate rather than the protease by binding to the A␤ domain at residues 29 -36 (18). Thus, it appeared possible that the modulatory capacity of GSMs could be different from that observed for the protease if they were targeting the substrate. Moreover, the GXXXG motif lying in this region was shown to determine A␤ 42 and A␤ 38 production in an inverse and interdependent manner via dimerization of the TMD. Our data do not support a mechanistic coupling of A␤ 38 and A␤ 42 production, however. The T43I and I45F (FAD) mutants both simultaneously increased A␤ 38 and A␤ 42 production and lowered that of A␤ 40 , suggesting that the production of A␤ 38 and A␤ 42 is not necessarily coupled in an inverse manner. Furthermore, the analysis of Phe mutants showed that although A␤ 42 production was lowered for all mutants upon GSM-1 treatment, the concomitant increase in A␤ 38 production was attenuated for T43F, V44F, and I45F mutants, indicating uncoupling effects. Interestingly, for the T43I, I45V, V46I, and V46F FAD mutants, the increased production of A␤ 38 induced by GSM-1 was accompanied by a reduced production of A␤ 39 in addition to A␤ 42 , which might indicate a (precursor-product) relationship between these two peptides. Moreover, as evident for the V44F mutant, which produced a high amount of A␤ 41 , we FIGURE 7. Profile of A␤ peptides generated in vitro from the C100-His 6 K28E mutant and its response to GSMs. A, sandwich immunoassay of A␤ 38 , A␤ 40 , and A␤ 42 species that were generated in cell-free assays (28) from C100-His 6 WT and K28E mutant substrates. Each species is plotted as a percentage of the total A␤ (A␤ 38 ϩ A␤ 40 ϩ A␤ 42 ) measured for each substrate. The K28E mutant shows a mildly increased A␤ 42 ratio as compared with the WT. Bars represent the mean of 5 experiments with error bars indicating the S.E. B, mass spectrometry of A␤ species that were generated in cell-free assays (28) from WT and K28E mutant substrates. The spectra closely reflect the immunoassay measurements for A␤ 38 , A␤ 40 , and A␤ 42 , showing a slightly higher relative peak for the A␤ 42 species in the K28E mutant. Note also the higher relative peak of 〈␤ 37 in this mutant in comparison with the WT spectrum. The intensities of the A␤ 40 peaks were set at 100% in the spectra. C, sandwich immunoassay showing the effect of 2.5 M GSM-1 treatment on A␤ 42 and A␤ 38 generated from WT and K28E mutant substrates in cell-free assays. Both A␤ 38 and A␤ 42 species could be significantly modulated for this mutant in vitro, but to a lower extent as compared with the WT. Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, *, p Ͻ 0.05. D, sandwich immunoassay showing the effect of 150 M fenofibrate on A␤ 42 and A␤ 38 generated from WT and K28E mutant substrates in cell-free assays. The K28E mutant is, like the WT, also susceptible to inverse modulation. Bars represent the mean of 3 experiments with error bars indicating the S.E. Statistical significance is calculated by paired Student's t test (two-tailed distribution); ***, p Ͻ 0.001, **, p Ͻ 0.01.
found that also, this A␤ species could be lowered by GSM-1 treatment. Thus, not only A␤ 42 production, but also A␤ 39 and A␤ 41 production, can be lowered in response to a GSM. Taken together, these data are difficult to reconcile with the model that production of A␤ 42 and A␤ 38 is interdependent. The relationship between A␤ 42 and A␤ 38 is apparently more complex, and production of A␤ 38 from A␤ 42 (16) may in fact occur only in the WT situation.
Although our results obtained for mutants of the GXXXG dimerization motif are to a large extent consistent with previously reported findings (17), we noted some differences. Although we observed an increase in A␤ 38 for all mutants, we did not detect a consistent concomitant change in A␤ 42 with respect to both A␤ 42 /A␤ total ratios and absolute levels of this species. As particularly evident for the G33A and the G33I mutants, the increase in A␤ 38 rather correlated with reduced A␤ 40 levels. The decrease of A␤ 40 may be explained for the G33I mutant by the concomitant increase of A␤ 37 , which could be derived from the A␤ 40 product line but not for the G33A mutant, where the A␤ 37 increase was not observed. The Gly mutants were responsive to GSM-1, which effectively lowered A␤ 42 for G29A and G33A, while increasing A␤ 38 levels. Likewise, the G33I mutant, which showed the highest A␤ 38 levels among the Gly mutants investigated, responded effectively to the inverse GSM fenofibrate, which lowered A␤ 38 and increased A␤ 42 to detectable levels for this A␤ 38 -biased mutant. Interestingly, fenofibrate also effected a strong increase of A␤ 40 for this mutant. Such a property has to our knowledge not yet been observed for an inverse GSM. These observations suggest that the glycine residues in the proposed GSM-binding site of APP do not play an essential role for potential GSM binding at this site. In addition, these data suggest that pharmacological modulation of ␥-secretase cleavage specificity appears not to be linked in an essential manner with GXXXG-dependent APP TMD interactions, whether involving dimerization or not (17,23).
Because the GXXXG mutant substrates give rise to mutated A␤ peptides, whose altered biochemical properties may affect the levels detectable in cultured cells downstream of production, we also analyzed these using our recently described validated cell-free in vitro system consisting of purified ␥-secretase and purified APP substrate (28). In this assay system, solely the production of A␤ is analyzed independent of e.g. altered secretion or degradation. As compared with cultured cells, similar results were obtained regarding the production of A␤ 38 and A␤ 40 in this system. In contrast, however, the rather substantial A␤ 42 production observed for all mutants in the cell-free system using purified components shows that ␥-secretase can per se generate A␤ 42 from Gly mutant substrates. It is important to note that the cell-free assay system was fully validated by the I45F and V50F mutant substrates. These mutants represent two extremes of APP mutants regarding the production of A␤ 38 and A␤ 42 and, in contrast to the Gly mutants, generate A␤ 38 , A␤ 40 , and A␤ 42 peptides without internal mutations. Both mutants behaved exactly as in cultured cells, proving that the ␥-secretase enzyme itself is in the correct conformation in the in vitro assay and thus further validating the cell-free assay used. It remains possible that the altered biochemical properties of Gly mutant A␤ 42 as compared with WT A␤ peptides may differentially affect the fate of this peptide in the two systems and thus account for the observed differences. Alternatively, it is also possible that slight conformational alterations may occur for

Summary of mutations, their effects on A␤ generation, and their responsiveness to modulation
The relative change in the A␤ 42 /A␤ total and A␤ 38 /A␤ total ratio as determined by A␤ immunoassay for each mutant as compared with WT is indicated by arrows (for the phenylalanine mutants, also refer to Fig. 2A for a representative Tris-Bicine urea gel analysis). Additionally, the magnitude of response to GSMs as compared with WT is indicated (WT response is equivalent to ϩϩϩ). All responses summarized in this table refer to GSM-1, except for some mutants in the proposed GSM-binding site. For mutants therein that have a strong A␤ 38  substrates carrying mutations within the A␤ domain regarding the ␥42 site cleavage, such as that of certain Gly mutant substrates, in the in vitro system. In agreement with the modulation results from cultured cells, responsiveness to GSM-1 was also observed for the Gly mutants in the cell-free system, although differences regarding the respective GSM response were noticed for the G33I mutant. An interesting residue that might contribute to GSM binding in the APP TMD is lysine 28. This residue, which lies directly adjacent to the GSM-binding site in the APP TMD, might form a salt bridge between the positively charged ⑀-amino group of the lysine side chain and the negatively charged carboxyl group of GSMs, which is essential for their A␤ 42 -lowering activity (41). This ionic interaction might change the position of the APP TMD relative to the active site of ␥-secretase, thus mediating a change in its cleavage specificity. However, our data show that the K28E mutant was susceptible to GSM-1 treatment, suggesting that an ionic interaction mediated by Lys-28 does not contribute to a potential GSM-APP interaction. The K28E mutant was also responsive to the inverse GSM fenofibrate, which lacks the carboxyl group and thus is apparently effective in the absence of an ionic interaction. Interestingly, mass spectrometric analysis revealed A␤ 33 and A␤ 37 as major A␤ species (species that are not detected by our A␤ immunoassay), possibly indicating that the K28E mutant A␤ might be turned over to shorter A␤ peptides in cultured cells. Analysis of the K28E mutant substrate in the cell-free system, however, i.e. in the absence of cellular metabolism, revealed that this mutant is normally processed by purified ␥-secretase with only minor changes in the profile of A␤ species as compared with the WT APP control. Importantly, GSM-1 was also effective on the K28E mutant in this system, further suggesting that the membrane-flanking lysine residue does not play a major role for the mode of action of carboxyl group-containing GSMs.
With respect to the mechanism of ␥-secretase cleavage, our data show an uncoupling of A␤ 38 and A␤ 42 generation for APP mutations located at different sites in the APP TMD, including mutations at or within the proposed GSM-binding site. All mutants allow a change of ␥-secretase cleavage specificity with respect to the generation of A␤ 38 and A␤ 42 in response to GSMs. GSM-mediated modulation of ␥-secretase cleavage specificity was shown to occur largely independent of the glycine residues of the GXXXG motif within the proposed GSMbinding site of APP, which were implicated in the generation of A␤ 42 (17). Although these data do not entirely exclude GSM binding to this site, they suggest that the glycine residues are unlikely to play an essential mechanistic role for the mode of action of GSMs, irrespective of the current controversy regarding the GSM-APP interaction (18,26). The GSM response of APP TMD mutants shown here, irrespective of their site and the amounts of A␤ 42 generated, may favor a substrate-independent targeting mechanism of GSMs. Binding studies with more potent high affinity GSMs rather than the currently existing low affinity compounds (18) will provide important answers by clarifying whether GSMs target the enzyme, which was initially suggested by several previous studies (6,27,(42)(43)(44). Mechanistically, NSAIDs that lower A␤ 42 were suggested to allosterically alter the conformation of PS (6,10,44), and conformational changes of PS opposite to that induced by such NSAIDs were also observed for PS FAD mutants (6,7,10). It is thus conceivable that many aggressive PS FAD mutations are locked in a conformation that makes the PS-substrate interaction refractory to the A␤ 42 -lowering capacity of GSMs (24,25). Clearly, as shown in this study, mutations in the APP substrate are permissive to GSMs, suggesting that the substrate is conformationally more flexible than the ␥-secretase enzyme, allowing APP substrate positioning such that the ␥42 site is less exposed to the active site of ␥-secretase. Thus, unlike the situation for PS FAD mutants, AD mouse models carrying APP FAD mutant transgenes should be useful for the in vivo evaluation of GSMs, and APP FAD mutant carriers are expected to be susceptible to GSM-based therapeutic strategies for AD treatment.