Generation of Aβ38 and Aβ42 Is Independently and Differentially Affected by Familial Alzheimer Disease-associated Presenilin Mutations and γ-Secretase Modulation*

Alzheimer disease amyloid β-peptide (Aβ) is generated via proteolytic processing of the β-amyloid precursor protein by β- and γ-secretase. γ-Secretase can be blocked by selective inhibitors but can also be modulated by a subset of non-steroidal anti-inflammatory drugs, including sulindac sulfide. These drugs selectively reduce the generation of the aggregation-prone 42-amino acid Aβ42 and concomitantly increase the levels of the rather benign Aβ38. Here we show that Aβ42 and Aβ38 generation occur independently from each other. The amount of Aβ42 produced by cells expressing 10 different familial Alzheimer disease (FAD)-associated mutations in presenilin (PS) 1, the catalytic subunit of γ-secretase, appeared to correlate with the respective age of onset in patients. However, Aβ38 levels did not show a negative correlation with the age of onset. Modulation of γ-secretase activity by sulindac sulfide reduced Aβ42 in the case of wild type PS1 and two FAD-associated PS1 mutations (M146L and A285V). The remaining eight PS1 FAD mutants showed either no reduction of Aβ42 or only rather subtle effects. Strikingly, even the mutations that showed no effect on Aβ42 levels allowed a robust increase of Aβ38 upon treatment with sulindac sulfide. Similar observations were made for fenofibrate, a compound known to increase Aβ42 and to decrease Aβ38. For mutants that predominantly produce Aβ42, the ability of fenofibrate to further increase Aβ42 levels became diminished, whereas Aβ38 levels were altered to varying extents for all mutants analyzed. Thus, we conclude that Aβ38 and Aβ42 production do not depend on each other. Using an independent non-steroidal anti-inflammatory drug derivative, we obtained similar results for PS1 as well as for PS2. These in vitro results were confirmed by in vivo experiments in transgenic mice expressing the PS2 N141I FAD mutant. Our findings therefore have strong implications on the selection of transgenic mouse models used for screening of the Aβ42-lowering capacity of γ-secretase modulators. Furthermore, human patients with certain PS mutations may not respond to γ-secretase modulators.

duction do not depend on each other. Using an independent non-steroidal anti-inflammatory drug derivative, we obtained similar results for PS1 as well as for PS2. These in vitro results were confirmed by in vivo experiments in transgenic mice expressing the PS2 N141I FAD mutant. Our findings therefore have strong implications on the selection of transgenic mouse models used for screening of the A␤ 42 -lowering capacity of ␥-secretase modulators. Furthermore, human patients with certain PS mutations may not respond to ␥-secretase modulators.
Alzheimer disease is the most abundant form of dementia, and increasing numbers of patients are to be expected in the near future. Amyloid ␤-peptide (A␤) 5 is a central player in the disease pathology. Originally it was purified as the building block of the disease-defining amyloid plaques. Now it is becoming clear that amyloid plaques are probably not the major neurotoxic entity in the disease rather this is an assembly of soluble oligomeric A␤ species (1). These assemblies initiate the so-called amyloid cascade and finally induce abnormal phosphorylation of tau and subsequent formation of paired helical filaments (2). A␤ is generated by proteolytic processing of the ␤-amyloid precursor protein (APP). Two proteases, ␤-secretase and ␥-secretase, perform the cleavages on the N and C termini of the A␤ domain, respectively (3). ␤-Secretase is a conventional aspartyl protease, whereas ␥-secretase is a rather unusual aspartyl protease capable of intramembraneous cleavage by utilization of a novel active site (4) that is signified by a highly conserved GXGD motif that includes one of the two active site aspartyl residues (5). ␥-Secretase is a complex composed of four subunits, presenilin 1 (PS1) or PS2, APH-1a or APH-1b, PEN-2, and nicastrin. PS harbors the catalytically active center of the protease. Numerous familial Alzheimer disease (FAD)-associated mutations occur within the PSs. They all increase the A␤ 42 /A␤ 40 ratio. Since A␤ 42 aggregates much faster than A␤ 40 , these mutations affect the kinetics of oligomer formation and aggregation (1).
Pharmacological inhibition of the secretases is a major therapeutic task. Unfortunately, the development of ␤-secretase inhibitors seems to be rather complicated (6), and ␥-secretase inhibitors caused major side effects in animal models and during clinical trials, due to the reduction of the biological function of ␥-secretase in Notch signaling (3). However, modulators of ␥-secretase activity, namely a subset of the nonsteroidal anti-inflammatory drugs (NSAIDs) such as ibuprofen or sulindac sulfide, selectively reduce the production of the aggregationprone A␤ 42 while leaving Notch signaling intact (7)(8)(9)(10). Concomitantly, they increase A␤ 38 production (10). Therefore, there seems to be an equilibrium of A␤ 42 and A␤ 38 generation. Moreover, compounds such as fenofibrate, which are known to exhibit an opposite modulating activity by increasing A␤ 42 production, reduce A␤ 38 production (11), again suggesting an interdependence of A␤ 42 and A␤ 38 production. NSAIDs most likely modulate ␥-secretase activity directly, because they are active in cell-free assays (7,8,(12)(13)(14) and they are known to affect the conformation of PS probably by an allosteric mechanism (15). Moreover, their modulating activity in terms of A␤ 42 reduction is affected by some FAD-associated PS1 mutations. The PS1 ⌬exon9 mutation decreases the sensitivity of the ␥secretase to sulindac sulfide, whereas the PS1 M146L mutation enhances its sensitivity (13).
We now investigated the equilibrium of A␤ 38 and A␤ 42 generation in cells expressing wild type (WT) PS or FAD-associated PS mutations as well as in transgenic mice expressing WT PS2 or a FAD-associated PS2 mutation. Surprisingly, we found that A␤ 38 and A␤ 42 generation do not depend on each other. Moreover, A␤ 38 and A␤ 42 respond differentially to ␥-secretase modulation depending on the PS mutation expressed.

EXPERIMENTAL PROCEDURES
Antibodies-Monoclonal antibodies against the PS1 N terminus (PS1N) and against the PS2 large loop (BI.HF5c), as well as the poly-and monoclonal antibodies to A␤ (3552, 2D8), were described previously (16). The C-terminal-specific anti-A␤ 38 antibody was obtained from Meso Scale Discovery, and C-terminal-specific anti-A␤ 40 (BAP24) and anti-A␤ 42 (BAP15) antibodies were a kind gift from Dr. Manfred Brockhaus (Roche Applied Science).

Independent Generation of A␤ 38 and A␤ 42 by ␥-Secretase
cDNA Constructs-cDNA constructs encoding PS1 and PS2 FAD mutants were generated by PCR-mediated mutagenesis using oligonucleotide primers encoding the respective mutations and cloned into pcDNA3.1/zeo(ϩ) (Invitrogen).
Cell Culture-Human embryonic kidney (HEK) 293 cells stably co-expressing Swedish mutant APP (HEK293/sw) together with the indicated PS variant were cultured as described before (17). HEK293/sw cells were stably transfected with the indicated PS cDNAs using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Pools of stably transfected PS cells were investigated to avoid clonal variations. Cells were plated at a density of 200,000 cells/well in poly-Llysine-coated 24-well plates and incubated for 24 h. Thereafter, cells were incubated in 500 l of fresh medium containing either 50 M sulindac sulfide (Sigma), 100 M fenofibrate (Sigma), 5 M GSM-1 (Roche Applied Science), or vehicle (Me 2 SO) for 16 h before analysis of conditioned medium by sandwich immunoassay. Treatments were performed in triplicate, and all media samples were measured in duplicate for A␤ 38 , A␤ 40 , and A␤ 42 .
Mass Spectrometry Analysis-Immunoprecipitation-MS analysis of A␤ species was carried out as described previously (18). Briefly, A␤ species were immunoprecipitated from conditioned medium from each cell line with antibody 82E1 for 4 h at 4°C. Immunoprecipitates were washed four times with immunoprecipitation-MS buffer (0.1% N-octylglucoside, 140 mM NaCl, 10 mM Tris, pH 8.0) and two times with distilled water. Immunoprecipitated peptides were eluted with 0.3% trifluoroacetic acid in 40% acetonitrile saturated with a-cyano-4-hydroxy cinnamic acid. The dissolved samples were dried on a stainless plate and subjected to MALDI-TOF MS analysis using Voyager DE STR (Applied Biosystems).
Quantification of Secreted A␤-Secreted A␤ peptides in conditioned medium were quantified by a sandwich immunoassay using the Meso Scale Discovery Sector Imager 2400. Streptavidin-coated 96-well Multi-Array plates were blocked in blocking buffer (0.5% bovine serum albumin, 0.05% Tween in phosphate-buffered saline, pH 7.4) and then incubated with biotinylated 2D8 capture antibody diluted 1:1000 in blocking buffer.
Plates were washed twice with phosphate-buffered saline-Tween before the addition of media samples and A␤ peptide standards (Bachem). Ruthenylated C-terminal-specific anti-A␤ 38 (Meso Scale Discovery), anti-A␤ 40 (BAP24), or anti-A␤ 42 (BAP24) antibodies were diluted 1:1000 in blocking buffer and used as detection antibodies. Plates were incubated at room temperature for 2 h before washing twice with phosphate-buffered saline-Tween and twice with phosphate-buffered saline. For detection, Meso Scale Discovery Read buffer was added, and the light emission at 620 nm after electrochemical stimulation was measured using the Meso Scale Discovery Sector Imager 2400 reader. The corresponding concentrations of A␤ peptides were calculated using the Meso Scale Discovery Discovery Workbench software. Ratios of each A␤ species as a percentage of total A␤ (A␤ 38 , A␤ 40 , and A␤ 42 ) were then calculated, and graphs were plotted using the GraphPad Prism software.
Compound Synthesis and Administration-The ␥-secretase inhibitor MRK-560 (19 -21) was synthesized starting from commercially available materials following the procedures depicted by Castro Pineiro et al. (22) and Churcher et al. (23). The ␥-secretase modulator GSM-1 was synthesized starting from commercially available materials following the procedures depicted by Hannam et al. (24). Both compounds were dissolved in 5% ethanol, 10% solutol and administered by oral gavage at a dosing volume of 10 ml/kg.
In Vivo Experiments-Mice homozygous for huAPPSw (line 147.72H) (25) and mice homozygous for both huAPPSw and huPS2 mutant N141I (line B6.152H) (26) were used in these experiments. In these lines, the expression of human APPSw and human mutant PS2 is driven by the Thy-1 and prion promoters, respectively (25). Studies were conducted with mice aged 2-3 months, because at this age cortical A␤ is primarily in a soluble form (27). Mice were sacrificed 4 h after a single oral administration of the drugs or vehicle. Brains were collected and frozen on dry ice until analysis of soluble cerebral A␤. All in vivo experiments were conducted in strict adherence to the Swiss federal regulations on animal protection and to the rules of the Association for Assessment and Accreditation of Laboratory Animal Care.
Extraction and Quantification of Cerebral A␤-For determination of A␤ 38 , A␤ 40 , A␤ 42 , and total A␤, brain hemispheres were homogenized in 9 volumes of 1.0% diethylamine/50 mM sodium chloride, incubated for 3 h on ice, and then centrifuged at 100,000 ϫ g (1 h, 4°C). Supernatants were aliquoted FIGURE 1. PS1 FAD mutants differentially affect A␤ 38 and A␤ 42 generation. A, cell lysates of HEK293/sw cells expressing endogenous PS (Endo) and HEK293/sw cells stably overexpressing PS1 WT or the indicated PS1 FAD mutants were analyzed for PS1 derivatives (holoprotein and N-terminal fragment (NTF)) and PS2 C-terminal fragment (CTF) by immunoblotting. PS1 holoprotein is visible in all cells overexpressing PS1 mutants in addition to PS1 NTF (top panel). Note that PS1 ⌬exon9 does not undergo endoproteolysis. Stable expression of PS1 results in the displacement of PS2 CTFs (lower panel). B, sandwich immunoassay of A␤ 38 , A␤ 40 , and A␤ 42 species that were isolated from conditioned medium of cells overexpressing PS1 WT and the indicated PS1 FAD mutations. Each species is plotted as a percentage of the total A␤ measured for each cell line. Data are arranged from left to right in order of A␤ 42 levels, and this correlates well with the age of onset of the different mutations. A␤ 42 levels are inversely correlated with A␤ 40 levels, in contrast to A␤ 38 , which shows no correlation. For clarity, A␤ 38 quantitation is enlarged in the lower panel. Data are plotted as the mean of three experiments with error bars indicating the S.E. C, qualitative MALDI-TOF MS of A␤ species produced by cells expressing PS1 variants. MALDI-TOF MS was performed on A␤ species immunoprecipitated from conditioned medium. A␤ 38 , A␤ 40 , and A␤ 42 could be detected as distinct and most abundant A␤ species secreted by cells expressing all PS1 variants, and the spectra are consistent with the immunoassay data. The A␤ 42 signal was low for PS1 WT but slightly increased for the mild mutations such as M146L, ⌬exon9, A285V, and L424R. For the severe mutations such as L166P, P117L, M233V, and Y256S, a peak corresponding to A␤ 42 was more readily observed. Interestingly, for the mutations ⌬exon9, ⌬IM, L166P, and Y256S, a peak corresponding to the predicted mass of A␤ 43 was also detected. Furthermore, the mutation M233V produced A␤ 39 as a distinct species, although its amount compared with other species is less as investigated by gel electrophoresis (data not shown). However, A␤ 38 , A␤ 40 , and A␤ 42 were the most abundant species secreted by the majority of PS1 variants. and stored at Ϫ80°C until assayed. Brain A␤ levels were determined using the Liquid Phase Electrochemiluminescent method as described in Narlawar et al. (28).

RESULTS
To study whether A␤ 42 and A␤ 38 production is coupled, we first investigated 10 selected PS1 FAD-linked mutations. Mutations were selected that cover a wide range of disease onsets (from ϳ50 years in the case of the PS1 M146L and A285V mutations to less than 30 years in the case of the PS1 P117L and Y256S mutations). cDNA constructs encoding PS1 mutations as well as WT PS1 were stably transfected into HEK293 cells stably expressing Swedish mutant APP to facilitate analysis of secreted A␤ species. Cell lysates were then investigated for the replacement of endogenous PS2, which served as an indicator for successful ectopic expression of PS1 (29). As shown in Fig.  1A, endogenous PS2 was fully replaced by PS1, demonstrating ectopic expression of all PS1 mutations. Due to overexpression, PS1 holoprotein accumulated for all mutants. In each case PS1 was endoproteolysed, with the exception of PS1 ⌬exon9, which is known not to undergo proteolytic processing (29) (Fig. 1A).
We then determined the total levels of secreted A␤ 38 , A␤ 40 , and A␤ 42 . As expected, the PS1 mutations cover a wide spectrum of effects on A␤ 42 production. This can range from a rather mild increase, as in the case of the PS1 M146L mutation, to a very severe increase as in the case of the PS1 L166P, G384A, M233V, P117L, and Y256S mutations (Fig. 1B). In line with our previous findings (17), the increase of A␤ 42 production seems to occur at the expense of A␤ 40 , because these levels gradually decrease with increased A␤ 42 levels. These findings are also largely consistent with an inverse correlation of A␤ 42 levels with age of onset (30). In contrast, A␤ 38 levels failed to show any negative correlation to A␤ 42 levels (Fig. 1B, enlarged in lower  panel), suggesting that A␤ 42 is generated independently of A␤ 38 . Because alternative A␤ species with C termini other than amino acids 38, 40, or 42 could not be detected by the sandwich immunoassay, we also investigated A␤ species by qualitative mass spectrometry analysis. Although some additional species such as A␤ 39 and A␤ 43 were occasionally observed upon expression of certain mutations, A␤ 38 , A␤ 40 , and A␤ 42 were the most abundant species for the majority of mutants (Fig. 1C).
To further investigate whether A␤ 38 and A␤ 42 production are directly linked to each other, we treated cells with the NSAID sulindac sulfide. Sulindac sulfide is known to modulate ␥-secretase activity by shifting the cleavage from amino acid 42 to amino acid 38 (10). In line with previous findings, the PS1 M146L mutation showed slightly greater sensitivity to treat-  ment with sulindac sulfide, whereas the PS1 ⌬exon9 and PS1 L166P mutations failed to respond to the A␤ 42 -lowering activity (13, 31) (Fig. 2). Furthermore, almost all other mutations with the exception of PS1 A285V showed only very little lowering of A␤ 42 levels (Fig. 2). This suggests that these mutations render ␥-secretase resistant to the modulating activity of sulindac sulfide. However, when we investigated the levels of secreted A␤ 38 , we found that all PS1 mutations, including those that showed no reduction of A␤ 42 upon sulindac sulfide treatment such as PS1 ⌬exon9, L424R, and L166P, increased A␤ 38 generation. Thus, some mutant ␥-secretase complexes exhibit a selective resistance to the A␤ 42 -lowering activity of sulindac sulfide but remain sensitive to the A␤ 38 -modulating activity. This effect remained robust up to 100 M sulindac sulfide, the maximum concentration tolerated by the cells (data not shown). Independent A␤ 38 and A␤ 42 production was further supported by treatment of cells with fenofibrate. Fenofibrate is a lipid-regulating drug that has been shown to exhibit opposite effects on ␥-secretase modulation by increasing A␤ 42 and decreasing A␤ 38 as compared with NSAIDs like sulindac sulfide and indomethacin (11). Interestingly, all PS1 mutations were less sensitive to the A␤ 42 -increasing activity of fenofibrate compared with WT PS1 (Fig. 3). Moreover, the very aggressive mutations such as PS1 L166P, G384A, M233V, P117L, and Y256S showed only a minor increase of A␤ 42 (Fig. 3). However, although some of the more aggressive mutations responded to a lesser degree with regard to the A␤ 42 -increasing activity, they showed similarly robust decreases of A␤ 38 (Fig. 3). Other mutations, like PS1 ⌬exon9, ⌬IM, L424R, and L166P, showed a somewhat diminished but still significant reduction of A␤ 38 (Fig. 3). Thus, some PS1 mutations render the ␥-secretase complex in a way that it becomes more or less resistant to A␤ 42 modulation, but, strikingly, this occurs independently of the A␤ 38modulating activity.
The observed effects on ␥-secretase modulation are not restricted to PS1 mutations and the use of sulindac sulfide. We tested a novel NSAID derivative with a lower IC 50 for ␥-secretase modulation (Fig. 4A, GSM-1). Treatment of cells expressing WT PS1 with as little as 5 M of this compound caused a significant increase in A␤ 38 and a significant decrease in A␤ 42 (Fig.  4B). Consistent with sulindac sulfide treatment, cells expressing L166P showed no change in A␤ 42 levels despite a highly robust and significant increase in A␤ 38 (Fig. 4B).
To determine whether this effect was specific for PS1, we investigated cells expressing WT PS2 and PS2 harboring the "Volga German" mutation (PS2 N141I). Treatment of cells expressing WT PS2 with GSM-1 caused a significant increase in A␤ 38 and a significant decrease in A␤ 42 (Fig. 4C); however, not much change in A␤ 42 was observed for PS2 N141I despite a significant increase in A␤ 38 (Fig. 4C). Interestingly, WT PS2-expressing cells showed a much stronger increase of A␤ 38 compared with cells expressing WT PS1 (Fig. 4, B and C).
To prove that ␥-secretase modulators differentially affect A␤ 42 and A␤ 38 production in vivo, we compared the effect of the in vivo active modulator GSM-1 in transgenic mice expressing APP-Swe or APP-Swe/PS2 N141I (Fig. 5). The modulator GSM-1 significantly reduced brain A␤ 42 levels (by 69% at 30 mg/kg) in the APP-Swe mice together with a corresponding increase of brain A␤ 38 levels (Ͼ4-fold at 30 mg/kg) in a dosedependent manner. No significant effect on brain A␤ 40 could be observed (Fig. 5A). When tested in the double transgenic APP-Swe/PS2 N141I line, GSM-1 did not induce any significant reduction of brain A␤ 42 . However, it led to a significant dose-dependent increase of brain A␤ 38 . At the highest dose of 30 mg/kg, a significant reduction of brain A␤ 40 of 27% could also be detected in the double transgenic line (Fig. 5B). Additionally, the effect of the ␥-secretase inhibitor MRK-560 in these two transgenic lines is different for the mutant PS2; although brain A␤ 40 and A␤ 42 is reduced to the same extent (86 and 87%, respectively) in the single APP-Swe transgenic line, there is a pronounced difference in the APP-Swe/PS2 N141I double transgenic line (95 and 44% for A␤ 40 and A␤ 42, respectively). This result is consistent with a previous finding in mice transgenic for the PS1 L166P mutation (31), suggesting that reduced sensitivity to ␥-secretase inhibitors in respect to A␤ 42 production may be common among strong FAD-associated PS mutants. Finally, the modulator GSM-1 does not alter the total amount of brain A␤, although total A␤ is reduced by the inhibitor MRK-560 (by 71% in the single transgenic line and by 64% in the double transgenic line).

DISCUSSION
Since the first description of the ␥-secretase-modulating activity of NSAIDs, it was widely assumed that A␤ 42 and A␤ 38 generation are mutually dependent as decreased/increased A␤ 42 production correlated with increased/decreased A␤ 38 generation (10,11). Our findings now demonstrate that A␤ 38 generation is not coupled to A␤ 42 production. This is shown by the lack of a negative correlation of A␤ 38 levels with A␤ 42 levels in the 10 different PS1 mutations investigated. Moreover, some FAD-associated PS mutations make the ␥-secretase complex at least partially resistant to the A␤ 42 -lowering activity of NSAIDs or the A␤ 42 -increasing activity of fenofibrate, while they are still sensitive to the corresponding A␤ 38 -modulating activity. There appears to be no correlation between the severity of the mutation and the responsiveness to A␤ 38 modulation.
Currently, the cellular mechanisms that allow a selective modulation of these cleavages are not fully understood. However, studies using fluorescence resonance energy transfer methods suggest that NSAIDs allosterically affect the conformation of PS and, probably as a consequence, the PS1/APP interaction (15). Moreover, FAD-associated PS mutations also affect the structure of the active site by changing the spatial relation of the PS1 N-and C-terminal fragments, in a manner opposing the effect of A␤42-lowering NSAIDs (15,32). Fianlly, NSAIDs modulate ␥-secretase activity in cell-free assays (7,8,(12)(13)(14). All this suggests that NSAIDs directly or indirectly affect PS conformation and ␥-secretase activity. However, other data imply that NSAIDs may affect intramembrane dimerization of APP and thereby influence ␥-secretase processing at its variable sites (33). Our data support the idea that NSAIDs affect PS activity directly or indirectly. A conformational change induced by NSAIDs may affect substrate cleavage at selective sites, which is well known for FAD-associated PS mutations. If mutant PS adopts a certain pathological conformation, which in the case of the aggressive mutations allows production of very high levels of A␤ 42 , this structure may not be changed any more by NSAIDs or fenofibrate. Under these conditions, further and efficient trimming of the substrate at position 38 of the A␤ domain may still be possible, thus still permitting the A␤ 38 -modulating activity of NSAIDs. In addition, NSAIDs may affect substrate structure/position within the active site by changing its conformation.
Our data have strong implications for animal models selected for in vivo analysis of NSAIDs. So far, only transgenic mice with a FAD-associated APP mutation have been investigated, and in these cases NSAIDs show their expected activity (8,11,34). However, many laboratories use transgenic mice overexpressing APP in combination with more or less aggressive PS mutations to enhance and accelerate amyloid plaque formation. Our data from the PS2 N141I mouse are in agreement with recent data from the PS1 L166P mouse (31). Such Total A␤ levels were unchanged, and A␤ 40 levels were unchanged at 3 and 10 mg/kg GSM-1, although they were significantly reduced at 30 mg/kg GSM-1. Interestingly, the inhibitor MRK-560 is less efficacious in reducing A␤ 42 levels in APP-Swe x PS2 N141I mice. Vehicle, MRK-560 (3 mg/kg), or GSM-1 (3-30 mg/kg) were dosed orally, and concentrations of brain A␤ 38 , A␤ 40 , A␤ 42 , and total A␤ were determined 4 h later. All values were expressed as a percentage of vehicle-treated controls (n ϭ 6 mice/group Ϯ S.E., except for MRK-560, n ϭ 3 Ϯ S.E.). Statistical significance is calculated by one-way analysis of variance followed by DunnettЈs post-test. *, p Ͻ 0.05; **, p Ͻ 0.01. Note that brain A␤ 38 levels were below limit of detection (ϽLOD) after treatment with MRK-560. models will lead to a false interpretation of the in vivo activity of NSAIDs and potentially other drugs, because the A␤ 42 -modulating activity of such drugs may be strongly compromised. Moreover, our findings also indicate that patients carrying aggressive PS1 or PS2 FAD mutations will not respond to NSAID treatment.