Alzheimer Presenilin-1 Mutations Dramatically Reduce Trimming of Long Amyloid β-Peptides (Aβ) by γ-Secretase to Increase 42-to-40-Residue Aβ*

Background: Mutations in presenilin-1 (PS1) cause early-onset familial Alzheimer disease (FAD). Results: The PS1·γ-secretase complex trims the C terminus of long amyloid β-peptides (Aβ), and FAD mutations significantly reduce the efficiency of trimming. Conclusion: This loss of carboxypeptidase function results in a gain of toxic Aβ42 compared to Aβ40. Significance: Understanding the effects of FAD mutations on γ-secretase function is critical for developing effective treatments. The presenilin-containing γ-secretase complex produces the amyloid β-peptide (Aβ) through intramembrane proteolysis, and >100 presenilin mutations are associated with familial early-onset Alzheimer disease (AD). The question of whether these mutations result in AD through a gain or a loss of function remains highly controversial. Mutations in presenilins increase ratios of 42- to 40-residue Aβ critical to pathogenesis, but other Aβs of 38–49 residues are also formed by γ-secretase. Evidence in cells suggests the protease first cleaves substrate within the transmembrane domain at the ϵ site to form 48- or 49-residue Aβ. Subsequent cleavage almost every three residues from the C terminus is thought to occur along two pathways toward shorter secreted forms of Aβ: Aβ49 → Aβ46 → Aβ43 → Aβ40 and Aβ48 → Aβ45 → Aβ42 → Aβ38. Here we show that the addition of synthetic long Aβ peptides (Aβ45–49) directly into purified preparations of γ-secretase leads to the formation of Aβ40 and Aβ42 whether the protease complex is detergent-solubilized or reconstituted into lipid vesicles, and the ratios of products Aβ42 to Aβ40 follow a pattern consistent with the dual-pathway hypothesis. Kinetic analysis of five different AD-causing mutations in presenilin-1 revealed that all result in drastic reduction of normal carboxypeptidase function. Altered trimming of long Aβ peptides to Aβ40 and Aβ42 by mutant proteases occurs at multiple levels, independent of the effects on initial endoproteolysis at the ϵ site, all conspiring to increase the critical Aβ42/Aβ40 ratio implicated in AD pathogenesis. Taken together, these results suggest that specific reduction of carboxypeptidase function of γ-secretase leads to the gain of toxic Aβ42/Aβ40.

teristic of Alzheimer disease (AD) (1). A large body of evidence points to A␤ as the pathogenic initiator, most notably the identification of dominant missense mutations in the amyloid ␤-protein precursor (APP) and the presenilins that cause earlyonset familial AD (FAD) (2) and the discovery of presenilin as the catalytic component of ␥-secretase responsible for generation of the A␤ C terminus (3). More than 100 FAD presenilin mutations have been identified, and these mutations can cause a decrease in ␥-secretase cleavage of substrates as well as an increase in the ratio of the more aggregation-prone 42-residue A␤ (A␤42) over the major secreted 40-residue A␤ (A␤40). These findings have led to a still unresolved controversy over whether presenilin mutations cause FAD through a loss or a gain of function (4 -8).
In deciphering the effects of presenilin FAD mutations, it is critical to consider the various proteolytic functions of ␥-secretase. First, upon assembly with the other three components of the protease complex, presenilin undergoes autoproteolysis (3,9) within a large loop located between transmembrane domains 6 and 7 to form an N-terminal fragment (NTF) and C-terminal fragment (CTF) (10), resulting in activation of the enzyme for cleavage of substrates. FAD-mutant presenilins likewise undergo NTF/CTF formation, with one notable exception (⌬exon9, which is active as a holoprotein (10)). Second, ␥-secretase has endoproteolytic activity toward membrane stubs of APP and other substrates generated upon ectodomain release by sheddases (11). This cleavage occurs within the substrate transmembrane domain to release the intracellular domain. For APP, the endoproteolytic site (called the ⑀ site) is located close to the transmembrane/ cytosolic boundary (Fig. 1A) (12), with cleavage here releasing the APP intracellular domain (AICD). Reduction of ⑀ cleavage is seen with many, although not all, FAD presenilin mutations (7,13,14).
In recent years evidence has mounted in support of a third type of proteolytic function of the ␥-secretase complex, a car-boxypeptidase activity that trims membrane-associated long A␤ peptides of 48 or 49 residues to shorter secreted forms of 38 -43 amino acids (15)(16)(17)(18)(19). Biochemical and cellular studies suggest that the endoproteolytic ⑀ cleavage of APP substrate occurs first and that the resultant long A␤ peptides are primarily trimmed in intervals of three amino acids along two product lines (Fig. 1A), with A␤49 leading to A␤46, A␤43 and A␤40 and A␤48 leading to A␤45, A␤42, and A␤38 (this last event removing a tetrapeptide). The expected tri-and tetrapeptide products have been detected by mass spectrometry (18). Despite the evidence, however, other studies suggest that A␤ and AICD production can be dissociated (20,21) or that there is no precursorproduct relationship between A␤42 and A␤38 (22,23). The former contention has been disproven by the determination of equimolar A␤ and AICD production (24), and the latter has been unambiguously settled recently with the demonstration that synthetic A␤42 is converted to A␤38 by isolated ␥-secretase (25).
In this study we developed a method for examining the carboxypeptidase cleavage of synthetic A␤48 and A␤49 by isolated or purified enzyme complexes, independent of their initial formation through ⑀ proteolysis of APP substrate. In so doing, we confirm that these long A␤ peptides are indeed intermediates toward A␤40 and A␤42, provide support for the dual-pathway model, and determine the degree to which long A␤ peptides contribute to the critical A␤42-to-A␤40 ratio. This biochemical system also allowed evaluation of the role of the membrane in the carboxypeptidase activity. Most notably, kinetic analysis of A␤40 and A␤42 production from long A␤ peptides by FAD mutant presenilin-1⅐␥-secretase complexes revealed striking reductions in carboxypeptidase activity that have important implications for resolving the loss-of-function versus gain-offunction controversy and providing a unifying model for the pathogenic mechanism of presenilin mutations.

EXPERIMENTAL PROCEDURES
␥-Secretase Preparations-CHAPSO-solubilized membranes from S20 cells, which are CHO cells overexpressing all four ␥-secretase components, were prepared as previously described (26). Briefly, 20 confluent 15-cm plates were scraped and lysed using a French pressure cell at 1000 p.s.i. The lysate was spun at low speed to remove nuclei and unbroken cells and then at 100,000 ϫ g. The resulting membrane pellet was washed in sodium bicarbonate buffer and solubilized in 2 ml of 1% CHAPSO. For purified ␥-secretase preparations, membranes from 160 confluent 15-cm plates of S20 cells were isolated, washed, and solubilized as described above, and ␥-secretase complexes were purified by sequential affinity purifications via nickel-nitrilotriacetic acid-agarose beads (Sigma) and M2 immobilized anti-FLAG-agarose beads (Sigma). For the inactive mutant enzyme controls, both solubilized membranes and purified complexes were prepared from CHO D1 cells (overexpressing all four ␥-secretase components but with D257A mutant PS1 containing an N-terminal Myc tag) in the same way as above, except the first affinity purification step involved anti-Myc antibody-agarose beads (Sigma). For the analysis of FAD PS1 mutant complexes, membranes were prepared from CHO cells overexpressing Pen2, Aph1, nicastrin, and either WT or mutant PS1 as previously reported (14). For these preparations, membranes were isolated from 20 confluent 15-cm plates and resuspended in 160 l of 1% CHAPSO.
␥-Secretase Assays-For assays performed in CHAPSO (27), solubilized membranes were incubated with substrate in HEPES buffer at pH 7.0 with 0.1% phosphatidylcholine, 0.025% phosphatidylethanolamine, 0.00625% cholesterol, and a final CHAPSO concentration of 0.25%. For assays performed in vesicles (28), purified ␥-secretase, 0.1% phosphatidylcholine, 0.025% phosphatidylethanolamine, and 0.00625% cholesterol were solubilized in 0.25% CHAPSO and incubated with styrene-based Biobeads (Bio-Rad) for 2 h at 4°C to remove detergent. Substrate was added to the resulting detergent-solubilized preparations or proteoliposomes followed by incubation at 37°C. To validate trimming of long A␤ substrates, reactions were carried out for 4 h, which was within the time frame needed for maximal product formation; to analyze the rate of trimming by WT or FAD mutant PS1⅐␥-secretase and to examine the effects of modulators (GSM-1 (22) and compound 2 (29), obtained courtesy of T. Golde (University of Florida)) on the trimming process, reactions were carried out for 1.5 h, which was within the linear range of product formation. Synthetic, purified long A␤ substrates were purchased from Anaspec, and APP C100-FLAG substrate was prepared as previously described (27,30). A␤40 and A␤42 generated from A␤49-45 were detected using A␤40-and A␤42-specific ELISAs (Invitrogen).
Kinetic Analysis-Samples of WT and FAD mutant membrane preparations were run on 4 -12% Bis-Tris gels followed by Western blotting with mAb1563 (Millipore), which detects PS1 NTF. The signal was captured using ECL (GE Healthcare), densitometry was performed on the films using ImageQuant (GE Healthcare), and the amount of enzyme added to each reaction was normalized based on the measured level of PS1 NTF (or holoprotein in the case of the uncleavable ⌬exon9). Time course experiments were first performed with the FAD mutants to establish the linear range of product formation. Reactions were then carried out with varying concentrations of substrate within this linear time frame. A␤40 and A␤42 products were quantified by specific ELISAs (Invitrogen), and nonlinear regression analysis with K m and V max determination was performed using GraphPad Prism 4.
Cloning, Cell Culture, Transfections, and Analysis-All CHO cell lines were grown in DMEM with 10% fetal bovine serum. For the analysis of long A␤ trimming in cells, APP truncated at residues 49 and 48 (A␤ numbering) and harboring the Swedish double mutation were cloned and expressed using the Tet-On system (Clontech). 100,000 CHO cells/well were seeded into 24-well plates. Upon reaching ϳ50% confluence, cells were transiently co-transfected with a 1:5 ratio of the pTet-On Advanced plasmid (Clontech) and a pTRE-Tight plasmid (Clontech) carrying the truncated APP sequence. Six hours post-transfection, the medium was changed to DMEM complete containing 1 g/ml doxycycline to induce the expression of the truncated APP. Inhibitor control transfections were also treated with 10 M DAPT at this time. After 24 h cells and conditioned media were harvested. A␤40 and A␤42 in the media were measured by specific ELISAs (Meso Scale Discov-ery). Cells from each well were lysed in radioimmune precipitation assay buffer, and the intracellular protein concentration was measured by a BCA assay (Pierce). Western blot analysis for APP⑀49 and APP⑀48 in cell lysates was carried out using anti-APP antibody 22C11 (Millipore).
Statistical Analysis-Comparisons of more than two groups were carried out using one-way analysis of variance (ANOVA) and Dunnett's post hoc test using WT values as the control group. Statistical significance between two groups was determined by unpaired two-tailed Student's t test.

RESULTS
␥-Secretase Trims ⑀-Site A␤ Peptides to A␤40 and A␤42-We first examined the production of A␤40 and A␤42 from synthetic A␤48 and A␤49 in vitro, which allowed us to demonstrate the C-terminal trimming of ⑀-site A␤ peptides by ␥-secretase using isolated enzyme and substrate. We could also directly examine the model in which A␤48 processing leads primarily to A␤42 and A␤49 primarily to A␤40 (18). We performed these assays in CHAPSO detergent (27,30) as well as with purified enzyme complexes reconstituted into lipid vesicles (28) to ensure that the results are not artifacts arising from detergent solubilization and to determine if the membrane is critical to the trimming process. We and others have previously described the use of urea polyacrylamide gel electrophoresis followed by Western blotting with an N-terminally directed anti-A␤ antibody to detect the range of C-terminal variants of A␤ produced in in vitro ␥-secretase reactions (7,14,17,31). However, trimmed A␤ generated from the synthetic long A␤ substrates could not be detected by this method because the signal from the excess substrate streaked throughout the lane, completely obscuring any signals from the A␤ products (data not shown). The levels of A␤40 and A␤42 produced from each synthetic long A␤ were instead quantified using specific ELISAs. The data reveal that A␤48 and A␤49 are indeed trimmed by ␥-secretase to A␤40 and A␤42 in both the CHAPSO-solubilized ( Fig. 1B) and proteoliposome systems (Fig. 1C). Control reactions with A␤ substrate alone (in which no enzyme was added to the reaction mixture) give the base-line levels of crossreactivity of the large amounts of long A␤ substrates in the assays with each ELISA detection antibody. Although there is some cross-reactivity, the results indicate that the signal obtained in the reactions with enzyme present is not simply background. Furthermore, the addition of the ␥-secretase transition-state analog inhibitor L-685,458 to the assays results in inhibition of A␤40 and A␤42 production. Heat-inactivated or inactive PS1 D257A (D1) ␥-secretase showed no A␤ production above background. In addition, no A␤40 or A␤42 was detected from solubilized membranes or proteoliposomes incubated without substrate, ensuring that the A␤40 and A␤42 produced is not due to cleavage of endogenous substrate present in the FIGURE 1. ␥-Secretase trims synthetic A␤49 and A␤48 to A␤40 and A␤42 in vitro. A, APP substrate is thought to be cleaved sequentially by ␥-secretase at the ⑀, , and ␥ sites, indicated by arrowheads. These cleavage events result in A␤ peptides with the indicated C termini. Ihara and co-workers (16,18) have proposed A␤40-and A␤42-generating pathways (top and bottom, respectively), in which ⑀ cleavage that produces AICD50 -99 primarily leads to A␤40, whereas ⑀ cleavage that produces AICD49 -99 mainly produces A␤42. B and C, A␤40 and A␤42 production from synthetic A␤49 and A␤48 and CHAPSOsolubilized membranes from CHO cells overexpressing the four ␥-secretase components (B) or purified ␥-secretase complexes reconstituted into lipid vesicles (C). A␤40 and A␤42 were detected using specific ELISAs. Assays were performed in triplicate, with control reactions as follows: ϩ I reactions contain 1.5 M ␥-secretase inhibitor L-685,458; D1 reactions contain ␥-secretase with an inactive D257A mutant PS1; boiled reactions contain heat-inactivated ␥-secretase; ␥ reactions have no substrate added to the assay mixtures; A␤ reactions have no enzyme added to the reaction mixtures. For all of the non-control reactions, the level of A␤40 produced was significantly different from that of A␤42 (p Ͻ 0.01, Student's t test). n ϭ 3; error bars, S.D. enzyme preparations. ␥-Secretase modulators (GSMs) also had the effect on trimming that they typically have on A␤ production (32); levels of A␤42 generated from A␤48 and A␤49 were decreased by two different GSMs (22,29), whereas A␤40 levels were not altered, providing further validation of C-terminal trimming by ␥-secretase in vitro (Fig. 2, A and B). However, only the more potent GSM-1 compound showed a robust dose-re-sponse effect. Kinetic analysis of the levels of A␤42 generated from APP C100FLAG and A␤48 in the presence and absence of GSM-1 indicates that this modulator reduces the V max in both cases and does not alter the K m of these conversions (Fig. 2C). This result is consistent with a recent report demonstrating that modulators increase the k cat of A␤42 conversion to A␤38 (25).  29) have previously been shown to selectively lower A␤42 levels and concomitantly increase A␤38 levels (22,29). A, as expected, both compounds lower A␤42 and increase A␤38 in a concentration-dependent manner using APP C100-FLAG as substrate in CHAPSO-solubilized ␥-secretase assays, as detected using a Bicine/urea-polyacrylamide gel electrophoresis system. B, both compounds decreased the amount of A␤42 generated from A␤48 and A␤49 without altering A␤40 levels. n ϭ 3; error bars, S.D. * indicate values that were significantly different from the A␤42 value with no compound added (p Ͻ 0.05). A␤40 values were not significantly different at any concentration tested. C, cleavage reactions were performed in CHAPSO with the indicated concentrations of substrate. The levels of A␤42 generated from C100-FLAG (left) or A␤48 (right) were measured by ELISA. ϩGSM, reactions contained 25 M GSM-1; ϪGSM, reactions with vehicle alone. V max values for C100 and A␤48 substrates were significantly reduced in the presence of GSM-1 (p Ͻ 0.05, Student's t test). n ϭ 2; error: S.D.
The ratios of A␤42 to A␤40 produced from each A␤ substrate in each system (taking into account the background signal that each substrate alone has in each ELISA) are shown in Table 1 and indicate that ␥-secretase cleavage of A␤49 primarily leads to A␤40, with an A␤42/40 ratio of ϳ1:7, and cleavage of A␤48 primarily leads A␤42, with an A␤42/40 ratio of ϳ9:1 in detergent and ϳ6:1 in vesicles (with no statistically significant differences in trimming between the CHAPSO and vesicle assays). These results are consistent with the model proposed by Ihara and co-workers (18) of two pathways in which A␤49 is primarily converted to A␤40 and A␤48 primarily to A␤42. We show that C-terminal trimming along these pathways is an inherent property of ␥-secretase whether it is solubilized in detergent or incorporated in a membrane and also demonstrate that A␤40 and A␤42 can be produced by a small degree of crossover between the two pathways.
Although these results demonstrate the trimming of ⑀-site A␤ peptides in vitro, we nonetheless sought to validate our findings in a cellular assay. Previous studies have attempted to examine the trimming of A␤49 and A␤48 by ␥-secretase in cells by direct expression of these long A␤s with a signal sequence (15). Although this system was used to demonstrate that A␤49 expression primarily leads to A␤40 secretion, the 1:1 ratio for A␤42/40 observed from A␤48 was inconclusive, as the levels of secreted A␤ were highly variable and apparently below the detection limits, likely due to the observed poor expression of A␤48. Alternatively, APP truncated at position 49 (APP⑀49) can be expressed at high levels, inserted into the membrane, and sorted to the cell surface (33). Moreover, APP⑀49 was processed by ␤or ␣-secretase and then ␥-secretase to generate quantifiable amounts of secreted A␤ or p3, an N-terminally truncated A␤ generated through ␣/␥-secretase cleavage. Therefore, we generated constructs for the expression of APP⑀49 and the previously unreported APP⑀48 and introduced them into CHO cells overexpressing all four ␥-secretase components (14). Expression of these truncated APPs was confirmed by Western blot of the cell lysates (Fig. 3A), and the levels of p3/A␤40 and p3/A␤42 secreted into the media were measured by specific ELISAs (Fig. 3B). We used 4G8, targeting the middle region of A␤, as the detection antibody in the ELISA because it detects both p3 and A␤, as secretion of A␤ alone from these cells was too low for quantification. The results confirm our in vitro findings, as APP49 expression primarily led to p3/A␤40 secretion, and APP48 expression led to mainly p3/A␤42 secretion. p3/A␤ production from each construct could be inhibited by the ␥-secretase inhibitor DAPT. In principle, we could obtain A␤42/40 ratios by subtracting the background observed from endogenous APP in untransfected cells; however, this may not be appropriate, as the overexpressed truncated APPs may compete with endogenous substrate for binding to secretases. The inability to accurately quantify ⑀-independent A␤40 and A␤42 production in this cellular system emphasizes the value of direct biochemical analysis with isolated or purified enzyme. In any event, whether production of p3/A␤ from endogenous substrate is completely inhibited or uninhibited by the overexpression, the conclusions about product preference (A␤40 from APP⑀49 and A␤42 from APP⑀48) still stand.
␥-Secretase Trims -Cleaved A␤s to A␤40 and A␤42-We next explored the cleavage of synthetic -site A␤s 45 and 46 as well as A␤47, a long A␤ that is not naturally observed (Fig. 4). A␤45 and A␤46 were also trimmed to A␤40 and A␤42 in a ␥-secretase-dependent manner. In addition, the proportions of A␤40 and A␤42 generated from each long A␤ were again consistent with a dual pathway model; using A␤46 as a substrate primarily resulted in A␤40 production with an A␤42/40 ratio of   NOVEMBER 7, 2014 • VOLUME 289 • NUMBER 45 JOURNAL OF BIOLOGICAL CHEMISTRY 31047 ϳ1:7, whereas A␤45 led to A␤42 almost exclusively, with a large A␤42/40 ratio of ϳ40:1. A higher ratio of A␤42/40 generated from A␤45 than from A␤48 is also consistent with the Ihara model, which posits only one cleavage event between A␤45 and A␤42. Importantly, we again found that the A␤42/40 ratios from each substrate were the same in both the solubilized and proteoliposome systems (Table 1). These results consistently indicate that dual-pathway C-terminal trimming is an intrinsic property of ␥-secretase and that the membrane is not essential for the enzyme to trim with apparent precision along these two pathways. We also attempted to confirm these results in cells by expressing APP⑀46 and APP⑀45 in the same CHO cell line that was used for APP⑀49 and APP⑀48 expression; however, we could not detect an increase in the level of secreted p3/A␤ above endogenous background when these constructs were expressed, likely due to the inability of these truncated APPs to remain inserted in the membrane. With synthetic A␤47 as a substrate for isolated ␥-secretase, we found that the levels of A␤40 and A␤42 produced were substantially reduced compared with that seen with the natural long A␤ substrates. This is consistent with the fact that the expected major cleavage products of A␤47 based on the Ihara model are A␤44 and A␤41, which, like A␤47, are not naturally observed ␥-secretase products. C-terminal Trimming by FAD-mutant ␥-Secretase-Having established a system to examine ␥-secretase trimming of synthetic long A␤ from the ⑀ to ␥ sites, we next examined the effects of FAD mutations in PS1 on this trimming process. To accomplish this, we determined the levels of A␤40 and A␤42 that five FAD-mutant PS1⅐␥-secretase complexes produced from A␤49 and A␤48 and their rates of formation compared with wild-type (WT) complexes. In this system, the levels of A␤40 and A␤42 generated are solely based on the efficiency of trimming, independent of effects on cleavage at the ⑀ site (which could either alter the rate of ⑀ site cleavage or the specificity of cleavage at positions 48 and 49). The PS1 mutations are G384A, ⌬exon9, L166P, A246E, and L286V, and all except L286V have been previously shown to reduce the endoproteolytic cleavage at the ⑀ site that leads to AICD production (14). However, all five mutations have been shown to generate an increased proportion of long A␤ peptides (A␤42ϩ) compared with WT enzyme (14). Membranes from CHO cells stably overexpressing the four ␥-secretase components with either WT or mutant PS1 were solubilized in CHAPSO, and equal amounts of each enzyme were used in the reactions based on PS1 NTF levels (or PS1 holoprotein in the case of the uncleavable but proteolytically active ⌬exon9) in each enzyme preparation. We first performed time course experiments to determine the linear range for A␤40 and A␤42 production from A␤48 and A␤49 followed by kinetic analysis of C-terminal trimming by each mutant. The rates of A␤40 and A␤42 production for each mutant were measured across different concentrations of substrate, and curves were generated for each possible conversion (A␤49 3 A␤40, A␤49 3 A␤42, A␤48 3 A␤40, and A␤48 3 A␤42) using non-linear regression analysis (Fig. 5). The K m and V max values were then calculated from these curves ( Table 2). Because each reaction contained equal amounts of enzyme, k cat is proportional to V max , and V max /K m is, therefore, a measure of FIGURE 4. ␥-Secretase trims synthetic A␤45, A␤ 46, and A␤47 to A␤40 and A␤42 in vitro. A, enzyme assays using CHAPSO-solubilized membranes from CHO cells overexpressing the four ␥-secretase components. B, enzyme assays using purified ␥-secretase complexes reconstituted into lipid vesicles. Control reactions are the same as in Fig. 1. The level of A␤40 generated in all non-control reactions was significantly different from that of A␤42 (p Ͻ 0.01 for A␤45 and A␤46 substrates, p Ͻ 0.05 for A␤47 substrate, Student's t test). n ϭ 3; error bars, S.D.

Presenilin Mutations Reduce Amyloid-␤ Trimming
catalytic efficiency (Table 2). When the efficiencies of these conversions by WT PS1 are set to 100% and compared with the efficiencies of the mutants ( Table 3), all of the FAD mutant complexes displayed clear and substantial reduction of the normal carboxypeptidase trimming function of ␥-secretase, with efficiencies ranging from 2 to 9% of WT for A␤49 trimming and from 17 to 40% of WT for A␤48 trimming. The K m and V max values in Table 2 indicate that these reductions in trimming efficiencies are primarily the result of dramatic decreases in V max , suggesting that the mutations affect the turnover of the long A␤s more than their affinity for the enzyme. Moreover, the trimming of these synthetic ⑀-cleaved A␤s was altered in ways that contribute to an increase in the A␤42/A␤40 ratio. First, as shown in Table 3, the trimming of A␤49 to A␤40 was significantly more reduced than the trimming of A␤48 to A␤42 by all of the mutant complexes when compared with WT. The cata-lytic efficiency values in Table 2 show that although the WT enzyme trimmed A␤49 to A␤40 ϳ2.5 times more efficiently than A␤48 to A␤42, the ⌬exon9 and L286V complexes trimmed A␤48 to A␤42 more efficiently than A␤49 to A␤40, and the L166P, G384A, and A246E complexes trimmed A␤48 to A␤42 with nearly equal efficiency as A␤49 to A␤40.
In addition to these differences seen between the processing of A␤49 and A␤48, the L166P and ⌬exon9 mutations resulted in a significantly greater reduction of the major A␤49 to A␤40 conversion compared with the crossover A␤49 to A␤42 conversion, and the ⌬exon9 and L286V mutations led to a greater reduction of the crossover A␤48 to A␤40 conversion compared with the major A␤48 to A␤42 conversion (Table 3).
We also attempted to examine the effects of FAD mutations on the trimming of -site A␤s (A␤46 and A␤45) in an effort to determine if a specific step in the trimming process is reduced FIGURE 5. A␤49 and A␤48 conversions to A␤40 and A␤42 by ␥-secretase are dramatically reduced by PS1 FAD mutations. A, equal amounts of each enzyme were used in the reactions, normalizing for PS1 NTF levels in each enzyme preparation based on densitometry of PS1 NTF Western blots (or of full-length PS1 for the uncleaved but proteolytically active ⌬exon9 mutation). B, cleavage reactions with WT (top panels) or the indicated FAD-mutant PS1 (bottom panels) were carried out in CHAPSO with the indicated concentrations of substrate. Left panels, A␤49 substrate; right panels, A␤48 substrate. Levels of A␤40 (solid lines) and A␤42 (dashed lines) were determined by ELISA, and data were fit using GraphPad prism 4 non-linear regression analysis. n ϭ 3; error bars, S.E. Note the difference in scale for A␤ levels from WT (top panels) and FAD-mutant PS1 (bottom panels). more than others. However, due to the high background that the A␤45 and A␤46 substrates have in the A␤42 ELISA (Fig. 4), we were not able to detect any A␤42 signal above background when we attempted to monitor the rates of trimming of these substrates by the FAD mutants (data not shown). In addition, due to the low efficiency of the conversion of A␤45 to A␤40 (Fig. 4), we were unable to detect any A␤40 production from A␤45 by the mutants (data not shown). We were, however, able to quantitate the trimming of A␤46 to A␤40. When compared with wild type, each mutant showed a substantial reduction of the rate of A␤46 conversion to A␤40 (Table 4). Moreover, these low rates of A␤46 trimming by the mutants were comparable with those observed for A␤49 ( Table 4), suggesting that the individual cleavage step from A␤49 to A␤46 is not one that is significantly impaired by the FAD mutations.
FAD Mutant Complexes Increase A␤42/40 from a Fixed A␤49/48 Mixture-Some FAD mutations in PS1 have been shown to lead to an increased proportion of ⑀ site cleavage at A␤48 compared with cleavage at A␤49 than is seen for WT PS1 (34). Because, as we have demonstrated above, A␤48 is the ⑀ product that primarily leads to A␤42, this alteration in the ⑀ cleavage site is one mechanism by which these mutations may increase A␤42/40. However, not all FAD PS1 mutations have this effect (35), and our data suggest that the more substantial reduction in the efficiency of trimming of A␤49 to A␤40 compared with that of A␤48 to A␤42 by the FAD mutant complexes should be sufficient to increase the A␤42/40 ratio regardless of the effects on the specificity of ⑀ site cleavage. We measured the levels of A␤42 and A␤40 generated from a fixed mixture of A␤49 and A␤48, with 70% A␤49 and 30% A␤48, as this has been shown to be the normal proportions generated from WT PS1 ␥-secretase complexes (34). The A␤42/40 ratios generated by isolated WT and mutant ␥-secretase complexes are shown in Fig. 6. The ratios for the mutant complexes are 2-4-fold higher than the ratio generated by the WT complex, suggesting that the differential reduction in the rates of trimming of each sub-   5. Equal amounts of enzyme were used in each reaction; therefore, K cat is proportional to V max , and V max /K m is a measure of catalytic efficiency. For all of the mutants, the V max and catalytic efficiency values for each conversion were significantly lower than the values obtained with WT enzyme (p Ͻ 0.01, one-way ANOVA and Dunnett's post test). No significant changes in K m values were measured. n ϭ 3; error, S.D.

DISCUSSION
Our findings have important implications for the normal biochemical function of the ␥-secretase complex as well as for the mechanism of pathogenesis of FAD presenilin mutations. First, we demonstrate that the carboxypeptidase activity is an intrinsic function of the enzyme independent of the membrane. Synthetic A␤ peptides of 45-49 amino acids in length were converted to A␤40 and A␤42 in a ␥-secretase-dependent manner whether the enzyme was isolated from membranes and detergent-solubilized or purified and reconstituted into lipid vesicles. Little or no difference in the ratio of A␤42/A␤40 was seen between detergent-solubilized and membrane-incorporated protease complexes. Our results are consistent with the model that A␤48 and A␤49, formed upon initial ⑀ endoproteolysis by ␥-secretase, are intermediates toward A␤40 and A␤42.
Second, our results are completely consistent with the dualpathway model originally hypothesized by Ihara and co-workers (16,18) in which A␤49 3 A␤46 3 A␤43 3 A␤40 and A␤48 3 A␤45 3 A␤42 3 A␤38. A␤49 substrate resulted in A␤40 along with a small level of A␤42, and A␤48 gave A␤42 as well as some A␤40. Nevertheless, the production of A␤42 from A␤49 and A␤40 from A␤48 reveals that crossover between the two pathways does occur to some degree, and therefore these crossover pathways contribute to the overall A␤42-to-A␤40 ratio. In addition, A␤46 trimming results in A␤40 in high preference to A␤42, and A␤45 is cleaved to A␤42 to the virtual exclusion of A␤40, again consistent with the dual-pathway model. Interestingly, use of A␤47 as substrate resulted in only small degrees of conversion to either A␤40 or A␤42. This peptide, along with its expected trimmed products A␤44 and A␤41, is virtually absent in analyses of A␤ from in vitro ␥-secretase assays, cell culture, or brain tissue. The corresponding tripeptide intermediates have also not been detected (18). Further study of A␤47 is needed to determine if A␤44 and A␤41 are indeed produced as expected.
Most importantly, we uncovered a surprising and striking effect of PS1 FAD mutations on the carboxypeptidase function of the ␥-secretase complex. Our biochemical system provided a means to study trimming by these mutant complexes independently of ⑀ proteolysis. All mutant complexes that we examined, located in different regions of PS1 and associated with average ages of onset from 24 to 53 years displayed dramatic reductions in rates of conversion of A␤49 and A␤48 to A␤40 and A␤42. Most unexpectedly, the rates of A␤49 conversion to A␤40 with the PS1 mutants were extremely low, with catalytic efficiencies of 2-9% that of wild-type enzyme. These relative conversion rates of A␤49 to A␤40 by the PS1 mutants were much lower than those seen from A␤48 to A␤42, which gave catalytic efficiencies ranging from 17 to 40% that of wild-type enzyme. As A␤49 is the major ⑀ cleavage product leading to A␤40 (34), this difference in FADmutant PS1⅐␥-secretase in handling A␤49 and A␤48 leads to increased A␤42/A␤40, primarily through reduction in A␤40.
Differences were also revealed in the rates of A␤49 conversion to A␤40 and A␤42, as the crossover conversion of A␤49 to A␤42 was not decreased as much as its primary conversion to A␤40 for two of the PS1 mutations (L166P and ⌬exon9). Moreover, the crossover conversion of A␤48 to A␤40 was decreased more than the major conversion to A␤42 for two PS1 mutations (L286V and ⌬exon9). These effects, although minor in comparison to the overall difference in handling between A␤49 and A␤48, likewise contribute to net increases in A␤42/A␤40. Interestingly, some (34) but not all (35) PS1 mutations have been reported to shift ⑀ cleavage to increase AICD51-99 versus AICD50 -99, the other products generated along with A␤48 and A␤49, respectively (Fig. 1). Shifting ⑀ cleavage toward A␤48 in this way would also increase A␤42/A␤40. Finally, a new report showed that PS1 FAD mutations can also slow the conversion of A␤42 to A␤38 (25). Thus, multiple effects of PS1 FAD mutations all conspire to raise A␤42/A␤40.
Despite these multiple effects, the most striking and consistent change is the decreased carboxypeptidase trimming of ⑀ cleavage products A␤48 and A␤49 to A␤40 and A␤42, most particularly the dramatic reduction in the A␤49 (or A␤46) to A␤40 conversion. This loss, although not complete, is severe and clearly leads to increases in A␤42/A␤40 by virtue of reducing A␤40 formation, providing a simple reconciliation of the loss-of-function versus gain-of-function controversy. These mutations do cause a loss of function: a specific loss of carboxypeptidase function, particularly the ability to trim A␤49 or A␤46 to A␤40. Our findings are consistent with recent reports suggesting that FAD-mutant presenilins can cause a reduction in the conversion of A␤43 and A␤42 to A␤40 and A␤38, respectively (7,25). This loss of carboxypeptidase function results in a gain of function; that is, the elevation of the critical A␤42/A␤40, thereby increasing the propensity of A␤ to aggregation and neurotoxicity (36). It should be noted that this specific loss of function also elevates longer A␤ peptides (7,14) and that the gain of neurotoxic function may be through these forms of A␤. Investigation of this possibility is, therefore, warranted and currently under way.