γ-Secretase Associated with Lipid Rafts

Background: Intramembranous cleavages of β-carboxyl-terminal fragment (βCTF) by γ-secretase generate amyloid β-protein (Aβ). Results: Three- to six-residue peptides are released successively along with Aβ generation by lipid raft-associated γ-secretase. Conclusion: γ-Secretase cleaves βCTF through multiple interactive pathways for stepwise successive processing to generate Aβ. Significance: This cleavage model provides insights into the precise molecular mechanism of Aβ generation. γ-Secretase generates amyloid β-protein (Aβ), a pathogenic molecule in Alzheimer disease, through the intramembrane cleavage of the β-carboxyl-terminal fragment (βCTF) of β-amyloid precursor protein. We previously showed the framework of the γ-secretase cleavage, i.e. the stepwise successive processing of βCTF at every three (or four) amino acids. However, the membrane integrity of γ-secretase was not taken into consideration because of the use of the 3-[(3-cholamidopropyl)dimethylammonio]-2-hydroxy-1-propanesulfonic acid-solubilized reconstituted γ-secretase system. Here, we sought to address how the membrane-integrated γ-secretase cleaves βCTF by using γ-secretase associated with lipid rafts. Quantitative analyses using liquid chromatography-tandem mass spectrometry of the βCTF transmembrane domain-derived peptides released along with Aβ generation revealed that the raft-associated γ-secretase cleaves βCTF in a stepwise sequential manner, but novel penta- and hexapeptides as well as tri- and tetrapeptides are released. The cropping of these peptides links the two major tripeptide-cleaving pathways generating Aβ40 and Aβ42 at several points, implying that there are multiple interactive pathways for the stepwise cleavages of βCTF. It should be noted that Aβ38 and Aβ43 are generated through three routes, and γ-secretase modulator 1 enhances all the three routes generating Aβ38, which results in decreases in Aβ42 and Aβ43 and an increase in Aβ38. These observations indicate that multiple interactive pathways for stepwise successive processing by γ-secretase define the species and quantity of Aβ produced.

␥-Secretase generates amyloid ␤-protein (A␤), a pathogenic molecule in Alzheimer disease, through the intramembrane cleavage of the ␤-carboxyl-terminal fragment (␤CTF) of ␤-amyloid precursor protein. We previously showed the framework of the ␥-secretase cleavage, i.e. the stepwise successive processing of ␤CTF at every three (or four) amino acids. However, the membrane integrity of ␥-secretase was not taken into consideration because of the use of the 3-[(3-cholamidopropyl)dimethylammonio]-2-hydroxy-1-propanesulfonic acid-solubilized reconstituted ␥-secretase system. Here, we sought to address how the membrane-integrated ␥-secretase cleaves ␤CTF by using ␥-secretase associated with lipid rafts. Quantitative analyses using liquid chromatography-tandem mass spectrometry of the ␤CTF transmembrane domain-derived peptides released along with A␤ generation revealed that the raft-associated ␥-secretase cleaves ␤CTF in a stepwise sequential manner, but novel penta-and hexapeptides as well as tri-and tetrapeptides are released. The cropping of these peptides links the two major tripeptide-cleaving pathways generating A␤40 and A␤42 at several points, implying that there are multiple interactive pathways for the stepwise cleavages of ␤CTF. It should be noted that A␤38 and A␤43 are generated through three routes, and ␥-secretase modulator 1 enhances all the three routes generating A␤38, which results in decreases in A␤42 and A␤43 and an increase in A␤38. These observations indicate that multiple interactive pathways for stepwise succes-sive processing by ␥-secretase define the species and quantity of A␤ produced.
␥-Secretase is a membrane-embedded multimeric high molecular mass aspartic protease that determines the molecular species of amyloid ␤-protein (A␤), 3 a pathogenic molecule in Alzheimer disease (AD) (1). It cleaves ␤-carboxyl-terminal fragment (␤CTF) of ␤-amyloid precursor protein (APP) in the middle of the membrane and releases A␤ and APP intracellular domain (AICD). Many familial AD (FAD)-associated mutations are found in presenilin (PS) 1/2, the catalytic subunit of ␥-secretase (2). Those mutations appear to modulate the activities of ␥-secretase, leading to qualitatively and/or quantitatively altered generation of A␤ species (3). Thus, how ␥-secretase cleaves ␤CTF is a critical issue in the pathogenesis of AD. In fact, a number of clinical and preclinical AD therapeutic trials targeting A␤ and/or ␥-secretase are ongoing based on the amyloid theory (4).
␥-Secretase cleaves the substrate within its transmembrane domain (TMD), i.e. the protein hydrolysis typically occurs in the hydrophobic environment of the lipid bilayer. Because of the unavailability of water molecules within the membrane, the cleavage that occurs within the membrane has remained an enigma ever since the identification of APP (5). However, advanced structural analyses of ␥-secretase with cryo-electron microscopy and the substituted cysteine method have provided a plausible explanation; ␥-secretase itself provides the hydrophilic environment required for substrate cleavage by generating a water-accessible cavity surrounded by multiple transmembrane segments of its own components (1,6), as demonstrated in another intramembrane-cleaving protease, site-2 protease (7). This view is supported by a recent study of the crystal structure of a PS/signal peptide peptidase homologue (8).
The underlying molecular mechanism of the cleavage within the membrane is another important issue in the context of developing a disease-specific therapeutic reagent. We have been concerned about the molecular mechanisms regarding how ␤CTF is cleaved by ␥-secretase within the membrane (9 -12) and proposed the stepwise successive cleavage model for A␤ generation (11). The liquid chromatography-tandem mass spectrometry (LC-MS/MS) identification of ␤CTF-derived tri-(and tetra-) peptides generated along with A␤ has lent strong support to the model (13). In this model, the initial ⑀-cleavages are followed sequentially by cleavages after every three (or four) residues, releasing A␤40 and A␤42 (A␤38) as the final products. There are two product lines as follows: A␤49 Ͼ A␤46 Ͼ A␤43 Ͼ A␤40 and A␤48 Ͼ A␤45 Ͼ A␤42 Ͼ A␤38. Similarly spaced residues with intramembrane cleavage have been identified in tumor necrosis factor-␣, a substrate of a signal peptide peptidase-like protein (14). Endoproteolysis with a spacing of three residues has also been shown to occur in PS (15). These observations indicate that the stepwise cleavage mechanism may be a characteristic of intramembrane proteolysis.
As the previous study to identify the released oligopeptides was carried out using CHAPSO-solubilized ␥-secretase (13), the mechanism underlying intramembrane cleavage by ␥-secretase remains to be clarified. Now it is known that some previous observations do not agree with the proposed model. For example, the different molecular species of A␤ and their corresponding AICDs are not produced in a one-to-one ratio (9). Cell-based expression of A␤48 generates both A␤42 and A␤40 even if there is a preference for A␤42 generation (10). These observations raise the possibility that there may be additional unidentified processing pathways for A␤ generation. Here, to address these issues, we sought to verify the stepwise cleavage model for A␤ generation using lipid raft membranes (16), where active ␥-secretase is known to reside (17).

EXPERIMENTAL PROCEDURES
Antibodies-The antibodies against A␤ used here were 82E1 (IBL), 6E10 (Covance), polyclonal antibodies specific for A␤40, A␤42, or A␤43 (IBL), and a monoclonal antibody specific for A␤38 (IBL). Antibodies against nicastrin and Pen-2 were from Sigma and Oncogene Science, respectively. The monoclonal antibodies against caveolin, flotillin, and calnexin were purchased from BD Transduction Laboratories. Anti-FLAG M2 monoclonal antibody was from Sigma.
Membrane Preparation-A microsomal fraction was prepared as described previously (11). Briefly, harvested CHO cells or cortices from 4-week-old Wistar rats were homogenized in buffer A (20 mM PIPES, pH 7.0, 140 mM KCl, 0.25 M sucrose, 5 mM EGTA) containing various protease inhibitors. Following brief centrifugation at 800 ϫ g for 10 min, the resulting postnuclear supernatants were centrifuged at 100,000 ϫ g for 1 h. The pellets containing the total membrane fraction were suspended in buffer C (50 mM PIPES, pH 7.0, 0.25 M sucrose, 1 mM EGTA).
Lipid rafts were obtained as the detergent-resistant membranes (DRMs) as described (17) with some modifications. The pellets containing the total membrane fraction were homogenized in 10% sucrose in MES-buffered saline (25 mM MES, pH 6.5, 150 mM NaCl) containing 1% CHAPSO and various protease inhibitors. The homogenate was adjusted to 40% sucrose, placed at the bottom of an ultracentrifuge tube, and overlaid with 35% and then 5% sucrose in MES-buffered saline. The discontinuous gradient was centrifuged at 39,000 rpm for 20 h at 4°C on an SW41 Ti rotor (Beckman). Lipid rafts accumulating at the 5-35% sucrose interface were carefully collected and resuspended in buffer C.
Reconstituted ␥-Secretase Assay-␤CTF (C99) tagged with FLAG at the carboxyl terminus (C99-FLAG) was prepared from Sf9 cells basically as described (18). Briefly, Sf9 cells were infected with recombinant baculovirus and cultured in the presence of 20 mM GM6001 to suppress ␣-secretase-like cleavage of C99-FLAG. C99-FLAG overexpressed in Sf9 cells was solubilized with the lysis buffer (50 mM Tris-HCl, pH 7.6, 150 mM NaCl, 1% Nonidet P-40, and protease inhibitor mixture (Roche Applied Science)) and then immunoaffinity-purified with anti-FLAG M2-agarose beads (Sigma). The quantity and purity of the obtained C99-FLAG were assessed by gel electrophoresis and Coomassie Brilliant Blue staining.
The reaction mixture contained DRMs at a protein concentration of 80 g/ml from CHO cells or 100 g/ml from rat brains, defined amounts of C99-FLAG, and 0.25% CHAPSO in buffer C supplemented with protease inhibitor mixture (0.5 mM diisopropyl fluorophosphate, 1 g/ml N ␣ -p-tosyl-L-lysyl chloromethyl ketone, 10 g/ml antipain, 10 g/ml leupeptin, 1 mM thiorphan, 100 M bestatin, 10 M amastatin, 0.1 M arphamenine, and 5 mM EDTA). This mixture was incubated at 37°C for the indicated times, and the reaction was stopped by placing the reaction mixture on ice.
To assess ␥-secretase activities in the total membrane fraction, the pellets containing the total membrane fraction were resuspended in buffer C to give a final protein concentration of 4.0 mg/ml. An equal volume of buffer C containing 2% CHAPSO and protease inhibitors (1 mM diisopropyl fluorophosphate, 2 g/ml N ␣ -p-tosyl-L-lysyl chloromethyl ketone, 20 g/ml antipain, 20 g/ml leupeptin, and 2 mM thiorphan) was added, and the fraction was kept on ice for 1 h. The fraction was diluted in 3 volumes of buffer C containing protease inhibitors (0.5 mM diisopropyl fluorophosphate, 1 g/ml N ␣ -p-tosyl-Llysyl chloromethyl ketone, 10 g/ml antipain, 10 g/ml leupeptin, 1 mM thiorphan, 133 M bestatin, 13.3 M amastatin, 0.13 M arphamenine, and 6.7 mM EDTA) and defined amounts of C99-FLAG, and then the mixture was incubated at 37°C for the indicated times. The experiments to compare ␥-secretase activities between total membranes and lipid rafts were performed within a linear range of A␤ generation according to the concen-tration of the substrate and the protein concentration in the membrane fraction.
Aliquots from the reaction mixture were transferred into the SDS sample buffer and subjected to quantitative Western blotting. The remaining part of the reaction mixture was mixed with trichloroacetic acid and kept on ice for 15 min. The suspension was centrifuged at 100,000 ϫ g, and the supernatant was filtered through a PVDF or nitrocellulose membrane (0.22-m pore size, Merck). The filtrate was injected into an HP1100 (Hewlett-Packard) system equipped with an STR ODS-II column (4.6 ϫ 150 mm, Shinwa Chemical Industries) to be concentrated. The column was washed with 0% B for 7 min, and the peptides were eluted as a mixture by using a steep 0 -90% B gradient for 0.1 min (A, 0.05% formic acid; B, 0.05% formic acid in 100% acetonitrile). The eluate was subjected to LC-MS/MS analyses.
Western Blotting and Quantification-The proteins were separated on either a 16.5% conventional Tris/Tricine gel or an 11% Tris/Tricine long gel (20 cm length) containing 8 M urea, followed by Western blotting with the antibodies to A␤ as described previously (11). The blots were developed by an ECL system (GE Healthcare). The signals were detected using a LAS 4000 mini-luminescent image analyzer (Fuji Film) and quantified using MultiGauge software (Fuji Film) with defined amounts of each synthetic A␤ species as standards. For the detection of other proteins, the proteins separated on an SDSpolyacrylamide gel were transferred onto a PVDF membrane.

LC-MS/MS Quantification of Tripeptides and Other
Oligopeptides-The quantification of the expected peptides by LC-MS/MS was performed as described previously (13). A Quattro Premier TM XE tandem quadrupole mass spectrometer accompanied by ultra-performance liquid chromatography (Waters) equipped with a column (ACQUITY UPLC HSS T3, 1.8 m, 2.0 ϫ 150 mm) was used to identify and quantify the tripeptides and other oligopeptides. To quantify each analyte, the precursor ion-product ion pairs were monitored using the device's multiple-reaction monitoring mode as follows: m/z ϭ 304. Mass Spectrometric Analyses of Shorter A␤ Species and AICDs-After 3 h of incubation of the reconstituted A␤ generation system, the produced A␤ was immunoprecipitated with 6E10 and protein G-coupled to Sepharose beads. The produced AICD was immunoprecipitated subsequently with anti-FLAG M2-agarose beads. A␤ and AICD bound to the recovered beads were eluted with 30% acetonitrile in 1% trifluoroacetate. Molecular masses of the peptides were determined with a matrixassisted laser desorption ionization-TOF-mass spectrometer, 4800 Plus MALDI TOF/TOF TM analyzer (AB SCIEX) using ␣-cyano-4-hydroxycinnamic acid as a matrix (18).

Generation of A␤ and AICD by Lipid Rafts-Our
previous study showed that lipid rafts contain all four components (PS1/2, nicastrin, Aph-1, and Pen-2) required for the active ␥-secretase complex and exhibit higher ␥-secretase activity in the membrane-based cell-free A␤ generation system using an endogenous substrate (17). This indicates that lipid rafts are one of the sites involved in A␤ generation within the cell. The following studies provided several lines of evidence for this assumption (22,23). In addition, the lipid composition of the rafts is favorable for ␥-secretase activities (24). Thus, we decided to use lipid rafts as a model system for the intramembrane cleavage of ␤CTF by ␥-secretase.
Lipid rafts were prepared from CHO cells by sucrose density gradient centrifugation in the presence of 1% CHAPSO. The raft fraction was determined by its flotation, the presence of flotillin (a lipid raft marker), and the absence of calnexin (a non-DRM marker). The majority of mature nicastrin, Pen-2, and the carboxyl-terminal fragment of PS were fractionated into the lipid raft fraction (Fig. 1A) (17).
An in vitro reconstituted ␥-secretase assay was performed using lipid rafts together with 0.25 M C99-FLAG that was purified from Sf9 cells. ␥-Secretase activity was assessed by the amounts of A␤ generated. Although only small amounts of proteins were recovered, the raft fraction yielded specific ␥-secretase activity 10-fold higher than that of the total membrane-associated ␥-secretase (Fig. 1B). A␤ was produced in a time-dependent manner by incubation at 37°C (Fig. 2, A and B). The production of A␤40 and A␤42 proceeded in a similar profile (Fig. 2, A and C). L-685,458 completely suppressed the A␤ production, indicating that it was mediated by ␥-secretase. The molecular species of A␤ produced were examined by a long SDS-urea gel (Fig. 2D). The most robustly produced was A␤40, followed by A␤42. In addition to A␤40 and A␤42, significant amounts of A␤43 and small amounts of A␤45 were detected. A weak signal migrating at the A␤46 position Multiple Pathways of Stepwise ␤CTF Processing by ␥-Secretase FEBRUARY 21, 2014 • VOLUME 289 • NUMBER 8 represented the combined signals of A␤46 plus larger species. Moreover, significant amounts of A␤38 were produced. The A␤38 band was just beneath the bands of C99-FLAG on a long SDS-urea gel, and its identity was confirmed with an A␤38specific antibody (data not shown, see Fig. 6A). A␤41 and A␤44 were undetectable.
The analysis of the produced AICD, a counterpart of A␤, by immunoprecipitation (IP)/TOF-mass spectrometry clearly showed that raft-associated ␥-secretase generated the two counterparts, AICD(50 -99) and AICD(49 -99) (Fig. 2F), as expected. Other longer or shorter AICDs were undetectable or below the detection limit. Thus, the molecular species of A␤ and AICD produced by ␥-secretase associated with lipid rafts in vitro were overall indistinguishable from those produced by CHAPSO-solubilized ␥-secretase (13,18).
Release of the "Predicted" Five Tripeptides and One Tetrapeptide by Raft-associated ␥-Secretase-To learn how ␤CTF is cleaved by ␥-secretase within membranes, we first examined using LC-MS/MS whether the same five tripeptides and one tetrapeptide released in the CHAPSO-solubilized system in a stepwise manner (13) were produced by raft-associated ␥-secretase. As shown in Fig. 3 B, five tripeptides, IAT, VIV, ITL, TVI, and VIT, were released concomitantly with A␤ generation. Their levels increased linearly in a time-dependent manner. The levels of VVIA, a tetrapeptide, also increased in a time-dependent manner, although its amount was much smaller than those of the five tripeptides (ϳ1/5 that of TVI and ϳ1/9 that of ITL). The generation of all six peptides was completely suppressed by the addition of L-685,458 at 1 M (data not shown). Consistent with our model, the quantitative relationships IAT Ͻ VIV Ͻ ITL and VVIA Ͻ Ͻ TVI Ͻ VIT, as seen previously in the CHAPSO-solubilized system (13), were invariably maintained. These results indicate that raft-associated ␥-secretase cleaves the intramembrane region of ␤CTF in a stepwise successive manner at every three or four residue to generate A␤40 and A␤42 (and then A␤38).
According to the stepwise cleavage model, the differences between the amounts of the successively released peptides determine the amounts of the A␤ species produced (13); it is assumed that A␤40 ϭ IAT (when further cleavage does not occur) and A␤42 ϭ TVI Ϫ VVIA (Fig. 3A). Thus, the amounts of A␤40 and A␤42 calculated from the released peptide quantified by LC-MS/MS were compared with those A␤ levels quantified by Western blotting using end-specific A␤ antibodies to further test the cleavage model. These two A␤ measures assessed by two different methodologies were roughly consistent (Fig. 3C).
Generation of Other Tripeptides, Tetrapeptides, and Pentapeptides Indicates Multiple Pathways for Stepwise Processing-As lipid rafts exhibited extremely high ␥-secretase activity for A␤ generation, we searched, using LC-MS/MS, more extensively and systematically for ␤CTF TMD-derived oligopeptides in addition to the six peptides already found.
Sixteen oligopeptides out of 29 peptides examined were released in a time-dependent manner besides the six peptides (Fig. 4A). Surprisingly, the levels of VIVI, VVIAT, and VIVIT increased to almost the same extent as VVIA. Of the peptides FIGURE 1. Preparation of lipid rafts and their A␤ generation activity. A, distribution of marker proteins and the components of ␥-secretase following sucrose density gradient centrifugation. Lipid rafts were prepared from CHO cells by sucrose density gradient centrifugation in the presence of CHAPSO. After centrifugation, the 5-35% sucrose interface (fraction 2), the layers containing 5, 35, and 40% sucrose (fractions 1, 3, and 4, respectively), and the pellet (ppt) were carefully collected. An aliquot from each fraction was subjected to Western blotting. Fraction 2 contained lipid rafts. The components of ␥-secretase were enriched in this fraction. B, comparison of in vitro A␤ generation activity between total membranes (TM) and lipid rafts (DRM). Total membranes and lipid rafts prepared from CHO cells were subjected to the A␤ generation assay by the addition of exogenous C99-FLAG. The generated A␤ was quantified by Western blotting with 82E1. It should be noted that the raft-associated ␥-secretase exhibited extremely high and specific A␤ generation activity compared with that associated with total membranes. Data are represented as means Ϯ S.E. from three independent experiments. FIGURE 2. A␤ and AICD species generated by the reconstituted lipid raft A␤ generation system. Lipid rafts prepared from CHO cells were subjected to the reconstituted A␤ generation assay. Following incubation for the indicated times, the reaction mixtures were subjected to either Western blotting or IP/TOF mass spectrometry. A, Western blotting with 82E1 (total A␤) and A␤40-and A␤42-specific antibodies. A␤ peptide standards were applied onto each rightmost lane. B and C, quantification of the signal intensities for total A␤ (closed circles in B), A␤40 (closed squares in C), and A␤42 (open triangles in C) on the blots, which were expressed as picomoles of A␤ generated per ml of the reaction mixture. Data are represented as means Ϯ S.E. from four independent experiments. D, Western blotting with 82E1 using a long SDS-urea gel. Authentic A␤ species from A␤37 to A␤49 were loaded onto the rightmost three lanes (M1: A␤37, A␤38, A␤40, A␤42, A␤45, and A␤46 to A␤49; M2: A␤39, A␤41, and A␤44). In this gel system, A␤46 and longer A␤ species (A␤47 to A␤49) were not separated. A␤40 was robustly produced by lipid rafts. A␤38, A␤42, A␤43, A␤45, and A␤46 and longer species were also produced in a time-dependent manner. Asterisks in D indicate C99-FLAG or its carboxyl-terminally truncated fragments. E, IP/TOF-mass spectrometry for shorter A␤ species. An aliquot of the reaction mixture was subjected to Western blotting with 82E1 (upper panel). A␤ in the reaction mixture was immunoprecipitated with 6E10 and subjected to TOF-mass spectrometry (lower panel). Among the candidate A␤s, only A␤(1-16) and A␤(1-15) were identified after incubation. However, L-685,458 at 2 M suppressed A␤ generation almost completely (upper panel) but not generation of the shorter A␤s, indicating that the generation of A␤(1-16) and A␤(1-15) was not mediated by ␥-secretase (26). A␤ (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17) and longer species were undetectable in this system. F, IP/TOF-mass spectrometry for AICDs. An aliquot of the reaction mixture was subjected to Western blotting with FLAG antibody (upper panel). AICD produced was subjected to IP/TOF mass spectrometry (lower panel). The signals corresponding to AICD(50 -99) and AICD(49 -99) were identified after incubation. They were suppressed by L-685,458 at 2 M, indicating ␥-secretase-mediated generation.

Multiple Pathways of Stepwise ␤CTF Processing by ␥-Secretase
FEBRUARY 21, 2014 • VOLUME 289 • NUMBER 8 examined, only those three increased prominently. This indicates that ␥-secretase releases significant amounts of pentapeptides in addition to tri-and tetrapeptides. The levels of GVV, VVI, VIA, IVI, TLV, ATVI, and IVIT increased significantly in a time-dependent manner (Fig. 4A). ATV, LVM, TVIV, VITL, GVVIA, and VIVITL increased slightly but significantly. Their release was almost completely suppressed by 1 M L-685,458 (data not shown), indicating that their generation is mediated by ␥-secretase. Thus, A␤ generation by raft-associated ␥-secretase accompanied the release of many three-to six-residue peptides in addition to the above six peptides (Fig. 4B). There should be novel routes by which A␤48 and A␤47 are processed to A␤43, releasing a pentapeptide VIVIT and a tetrapeptide VIVI, respectively, and by which A␤43 is processed to A␤38, releasing a pentapeptide VVIAT (Fig. 4B). The presence of small amounts of the hexapeptide VIVITL suggests that the cleavage at every three residues is rarely skipped.
Overall, there are three routes to generate A␤43, i.e. those via A␤46, A␤47, and A␤48 (see Figs. 4B and 9). The comparison of the amounts of the released peptides (VIV, VIVI, and VIVIT) indicates that the routes via A␤46, A␤47, and A␤48 contribute to A␤43 generation in the ratio 50:6:ϳ5. There are also three routes by which to generate A␤38. Most A␤38 is generated either from A␤42 (ϳ50%) or A␤43 (ϳ36%) by the release of a tetrapeptide (VVIA) or a pentapeptide (VVIAT), respectively. A minor fraction of A␤38 could be generated from A␤41 (ϳ14%) by the release of a tripeptide (VVI).
To validate the above assumption for multiple cleavage pathways by ␥-secretase, the levels of A␤38 and A␤43 quantified by Western blotting were compared with those calculated from peptide amounts quantified by LC-MS/MS. There are two routes processing A␤43 (see Figs. 4B and 9). Thus, it is assumed that A␤43 ϭ (VIV ϩ VIVI ϩ VIVIT) Ϫ (IAT ϩ VVIAT) and A␤38 ϭ VVIAT ϩ VVIA ϩ VVI. The results show that the measures obtained from the two methodologies are roughly consistent (Fig. 3C).
Notably, the release of VIVIT cross-links the A␤40 and A␤42 product lines at the A␤43 level, making the two lines interactive (see Fig. 4B). The release of VVIAT also links the A␤40 product line to A␤38 generation. The minor release of a tetrapeptide makes it possible to link the A␤ product lines. Ultimately, the clipping of these oligopeptides permits production of multiple alternative pathways for stepwise cleavages, linking the A␤ product lines in an interactive manner.
Uniform Suppression of Peptide Release by ␥-Secretase Inhibitors-L-685,458, a transition state analog inhibitor, uniformly suppressed all the A␤ species and peptides generated by the raft-based ␥-secretase in a dose-dependent manner, as expected (data not shown). DAPT, a nontransition state analog inhibitor, similarly suppressed A␤ generation (Fig. 5). Unexpectedly, there were no accumulations of A␤43 and A␤46 as observed with the cell system (Fig. 5A) (11). The release of the major and minor peptides was also uniformly suppressed, but FIGURE 3. Release of the predicted five tripeptides and one tetrapeptide according to the stepwise cleavage model by raft-associated ␥-secretase. A, schematic illustration of the ␥-secretase-mediated stepwise cleavages of ␤CTF, for which the results from the CHAPSO-solubilized system (13) are shown. Following ⑀-cleavages that generate A␤49 and A␤48 and their counterparts AICD(50 -99) and AICD(49 -99), respectively, ␥-secretase successively cleaves A␤49 and A␤48 in the direction from the ⑀to the ␥-cleavage sites by releasing tripeptides and finally produces A␤40 and A␤42, respectively. A small fraction of A␤42 is further converted to A␤38 by the release of a tetrapeptide. According to the model, the quantitative relationships among the peptides, IAT Ͻ VIV Ͻ ITL and VVIA Ͻ TVI Ͻ VIT, should be maintained. The differences between the successively generated tripeptides determine the amounts of the different A␤ species produced. B, quantification by LC-MS/MS of the predicted five tripeptides and one tetrapeptide released in the reaction mixture of the lipid raft A␤ generation system. The release of three tripeptides (IAT, VIV, and ITL) from the A␤40 product line and two tripeptides (TVI and VIT) and a tetrapeptide (VVIA) from the A␤42 product line proceeded in a time-dependent manner (p Ͻ 0.01, Spearman's rank correlation). Data are represented as means Ϯ S.E. from four independent experiments. Note that the quantitative relationships ITL Ͼ VIV Ͼ IAT and VIT Ͼ TVI Ͼ Ͼ VVIA were maintained at every time point. C, quantitative comparison of A␤ levels between Western blotting and LC-MS/MS. The reaction mixture was divided into two halves. One-half was not incubated; the other was incubated for 60 min. One-half of the reaction mixture with or without incubation was subjected to Western blot analyses using A␤ species-specific antibodies VIA and VIVITL did not exhibit a significant decrease (Fig. 5, B  and C). Sulfonamide, which induced a build-up of A␤43 and A␤46 within the cells (28), also failed to duplicate in this system (data not shown). It is likely that an intact cell system is required for the accumulation of longer A␤ species by DAPT, because significant DAPT-induced accumulations of longer A␤ species were not observed even by use of isolated membranes.
GSM-1 Affects Multiple A␤38-generating Pathways-GSM, a compound that suppresses A␤42 generation without affecting the total A␤ amount generated, would be beneficial for therapeutic development, because it could avoid the adverse effects elicited through other ␥-secretase substrates, especially Notch (1). According to the stepwise processing model, decreased A␤42 and increased A␤38 levels caused by nonsteroidal antiinflammatory drugs could be explained by an enhancement of the final (fourth) processing step of the A␤42 product line, i.e. the increased release of VVIA (13). However, the effect may not be so straightforward, because three routes generating A␤38 were found here. Thus, we sought to clarify which route among the three was affected with GSM-1.
GSM-1 caused a dose-dependent decrease in the A␤42 levels and a reciprocal increase in the A␤38 levels without affecting the total A␤ levels (Fig. 6A). The A␤40 levels were almost constant up to 1 M of GSM-1 and then declined a little. Peptide analyses by LC-MS/MS showed that although the levels of all five major tripeptides were not altered, the levels of VVIA increased remarkably (ϳ2.4-fold increase at 2.5 M of GSM-1) in a dose-dependent manner (Fig. 6B). Additionally, the pentapeptide VVIAT and the tripeptide VVI increased in a dosedependent manner remarkably (ϳ2.4-fold) and moderately (ϳ1.8-fold), respectively (Fig. 6C). The increases in the amounts of the released peptides VVIA, VVIAT, and VVI are in the ratio 7:4:1. These results indicate that GSM-1 enhances the cleavage step not only from A␤42 to A␤38 but also from A␤43 to A␤38 and to a lesser extent from A␤41 to A␤38. The former two contributed to a large increase in A␤38 levels upon treat- ment with GSM-1. Consistent with these observations, the Western blotting analysis showed a dose-dependent decrease in the A␤43 levels upon treatment with GSM-1 (Fig. 6A). The A␤41 levels were not evaluated because they were under the detection limits on the SDS-urea gel, and a specific antibody was not available.
Regarding the putative A␤40 product line, the levels of IAT did not alter (13), but the levels of GVV tended to increase upon treatment with GSM-1, which may reflect a slight decrease in A␤40 levels at high concentrations of GSM-1 (Fig. 6). The levels of all other oligopeptides were not changed by GSM-1. These results indicate that GSM-1 acts as an enhancer to the final step of multiple processing pathways, including all three routes converging to A␤38, rather than affecting only the A␤42-generating pathway.
Brain Raft-associated ␥-Secretase-Because active ␥-secretase is known to be localized to lipid rafts in human brains as well (29), we next examined whether the cleavage pathways found in the CHO raft are also at work in the brain raft. As it is possible that the post-mortem time affects the activity of ␥-secretase (29), we used rat brains for this purpose. The raft fractions prepared from rat cortices were subjected to an in vitro reconstituted ␥-secretase assay. As shown in Fig. 7A, A␤ was produced in a time-dependent manner. Both A␤40 and A␤42 increased during incubation, and the generation of A␤ was suppressed almost completely by L-685,458. The LC-MS/MS analyses of oligopeptides released during incubation showed that 15 oligopeptides out of 22 peptides generated by the CHO raft were released by the brain raft in a time-dependent manner (Fig. 7, B and C). These included two pentapeptides and one hexapeptide observed in the CHO raft. The relative quantitative relationships among the peptides were very similar in the CHO and brain rafts, except for IVI and VVIAT (Figs. 4A and 7C), the amounts of which were smaller in the brain raft. The most abundant were the original five tripeptides, followed by VVIA (Fig. 7B). Quantification showed that the amounts increased mostly in the order IAT ϭ VIV Ͻ ITL and VVIA Ͻ Ͻ TVI Ͻ VIT. These results indicate that the two major stepwise cleavage pathways, which generate A␤40 and A␤42 (A␤38), respectively, are conserved in brain rafts as well. The peptides that were generated to lesser extents by the CHO raft were not found in the brain raft, probably because of the detection limit. This is because the CHO raft exhibited much higher ␥-secretase activity than the brain raft (Figs. 3, 4, and 7). Thus, the cleavage mechanisms appear to be largely shared by the CHO and brain rafts, and the unusual oligopeptides such as penta-and hexapeptides are detectable in both systems.
Treatment with L-685,458 suppressed the release of the five major tripeptides and of other minor oligopeptides (Fig. 8A). The suppression was uniform and dose-dependent. DAPT also FIGURE 5. Uniform suppression by DAPT of oligopeptide release by raft-associated ␥-secretase. Lipid rafts from CHO cells were subjected to the A␤ generation assay for 60 min in the presence of DAPT at the indicated concentrations. A, effect of DAPT on the generation of various A␤ species was examined by Western blotting with 82E1 after electrophoresis on a long SDS-urea gel. DAPT suppressed generation of all the A␤ species uniformly in a dose-dependent manner. Authentic A␤ species from A␤37 to A␤49 were loaded onto the rightmost three lanes. Asterisks indicate C99-FLAG and/or its carboxyl-terminally truncated fragments. B, effects of DAPT on the release of the predicted five tripeptides and one tetrapeptide by raft-associated ␥-secretase were examined with LC-MS/MS. The release of the peptides was suppressed uniformly by DAPT in a dose-dependent manner (p Ͻ 0.01, Spearman's rank correlation). The suppression profiles for oligopeptides were very similar to those for the A␤ species. C, LC-MS/MS quantification of other oligopeptides released by ␥-secretase associated with the lipid raft in the presence of DAPT. The release of these peptides was also uniformly suppressed by DAPT. Statistically significant dose-dependent suppression was observed for all these peptides (p Ͻ 0.05 for GVV and p Ͻ 0.01 for others, Spearman's rank correlation) except for VIA (p ϭ 0.0788) and VIVITL (p ϭ 0.6815). Data are represented as means Ϯ S.E. from four independent experiments.
suppressed the generation of all the peptides uniformly and in a dose-dependent manner (Fig. 8 B). Again, we were unable to observe transient accumulations of longer A␤ species in brain rafts.
We next examined to what extent GSM-1 affects intramembrane cleavages of ␤CTF by brain raft-associated ␥-secretase (Fig. 8C). The LC-MS/MS analyses of the released peptides showed a dose-dependent increase in VVIA and VVIAT, as observed in the CHO raft, whereas VVI did not increase. The levels of all other peptides stayed the same. The increase in VVIAT was remarkable, being almost equivalent to that in VVIA. These results indicate that GSM-1 causes an increase in A␤38 levels largely through an enhancement of both the A␤43to-A␤38 and A␤42-to-A␤38 cleavage steps in brain rafts.

DISCUSSION
In this study, we have shown using a lipid raft system that the intramembranous cleavage of ␤CTF by membrane-associated ␥-secretase proceeds in a stepwise successive manner similar to that observed with ␥-secretase in a CHAPSO-solubilized form FIGURE 6. GSM-1 enhanced the release of VVIA as well as VVIAT and VVI by raft-associated ␥-secretase. A, effect of GSM-1 on A␤ generation by raft-associated ␥-secretase. Lipid rafts from CHO cells were subjected to the A␤ generation assay in the presence of GSM-1 at the indicated concentrations. Following incubation for 60 min, the reaction mixtures were subjected to Western blotting with 82E1 for total A␤ and species-specific antibodies to A␤40, A␤42, A␤38, and A␤43. Relative A␤ generation activities compared with no treatment with GSM-1 are shown. Whereas the levels of total A␤ did not alter, the levels of A␤42 and A␤43 decreased, and those of A␤38 increased significantly with GSM-1 in a dose-dependent manner (p ϭ 0.0032, 0.0179, and 0.0017, respectively, Spearman's rank correlation). The A␤40 levels were relatively stable under treatment with GSM-1 at most concentrations, but very high concentrations of GSM-1 caused a decrease in A␤40 (p ϭ 0.0006, Student's t test compared with no treatment with GSM-1). B, effects of GSM-1 on the release of the predicted five tripeptides and one tetrapeptide by the raft-associated ␥-secretase were examined with LC-MS/MS. Only the release of VVIA, which reflects the conversion of A␤42 to A␤38, increased significantly with GSM-1 in a dose-dependent manner (p ϭ 0.0149, Spearman's rank correlation). C, effects of GSM-1 on the release of other oligopeptides were examined with LC-MS/MS. Most of the peptides were released in unchanged amounts after treatment with GSM-1, but the release of VVIAT, which reflects the conversion of A␤43 to A␤38, and of VVI, reflecting the conversion of A␤41 to A␤38, increased remarkably and significantly in a dose-dependent manner (p ϭ 0.0051 and 0.0039, respectively, Spearman's rank correlation). Data are represented as means Ϯ S.E. from four independent experiments. **, p Ͻ 0.01 and *, p Ͻ 0.05. (13). However, the stepwise cleavage occurs not only at every three residues but also at four or five residues, and even cleavage at a six residues was observed (Fig. 9). Thus, the cropping of these oligopeptides links the two major A␤ product lines identified previously (13) at several points, leading to multiple interactive pathways for the stepwise cleavages.
It is possible that the integration of ␥-secretase in the raft membrane allows it to maintain its conformation adequately for exhibiting higher activity. If so, this could cause qualitative differences in the cleavages between the raft system and the CHAPSO-solubilized system. It is also possible that the extremely high activities of raft-associated ␥-secretase (see Fig.  1B; the activities amounted to 78% of the total cell membrane activities 4 ) reveal minor cleavage pathways that were below the detection limit in the CHAPSO-solubilized system. For example, although A␤37 was not detected in the CHAPSO-solubilized system (13), the detection of GVV and its response to ␥-secretase inhibitor treatment indicate that a very small fraction of A␤40 is processed to A␤37 in the A␤40 product line. This agrees with the previous observation that A␤37 was produced by the SH-SY5Y in vitro-reconstituted A␤ generation system (30). Thus, when all the available data are taken into account, it is reasonable to consider the progressive stepwise cleavage to be a basic mechanism of the intramembranous cleavage of ␤CTF by ␥-secretase, whereas the quantitative variability derived from the employed system may result in some differences of the molecular species produced.
Quantitative analyses of the released oligopeptides indicate that the major processing pathways are the two successive tripeptide-releasing pathways starting at ⑀-cleaved A␤49 and A␤48 and ending at A␤40 and A␤42, respectively (Fig. 9). A␤ generated by these pathways amounts to ϳ75% of total A␤ generation. The released amounts of four tetra-and pentapeptides (VVIA, VIVI, VVIAT, and VIVIT) are less than those of the above tripeptides. The amounts of VVIA correspond to 19% of A␤42 (TVI ϩ TVIV) assumed to be generated, indicating the conversion of ϳ20% of A␤42 generated to A␤38. The amounts of VVIAT and VIVIT released are ϳ10% those of the tripeptides IAT and VIT released from the same A␤ through the major pathway. VIVI is released prominently, but we do not know its origin, because the corresponding AICDs are barely detectable (9,18).
Interestingly, three pathways, each releasing a tri-, tetra-, or pentapeptide, converge at the levels of A␤38 and A␤43 (Fig. 9). This indicates that the peptide bonds between A␤38 and A␤39 (Gly-38 -Val-39) and between A␤43 and A␤44 (Thr-43-Val-44) are favorably attacked by ␥-secretase. Because Gly is a small amino acid with a strong helix-destabilizing effect, the double-Gly configuration of Gly-37-Gly-38 may induce helix instability and facilitate cleavage at this position to release A␤38. In the case of A␤43, Thr-43 is a relatively hydrophilic residue embedded in a stretch of hydrophobic residues, and it may promote cleavage at Thr-43-Val-44.
The presence of multiple interacting pathways shown here could provide a better explanation for several previous observations apparently inconsistent with the stepwise processing model. The observation that the produced amounts of A␤ species and their corresponding AICDs (AICD(50 -99) and A␤40; AICD(49 -99) and A␤42) are correlated but not in a one-to-one ratio (9) could be explained by the presence of multiple minor cleavage pathways linking the major pathways in an interactive manner. The observation that the exogenously expressed A␤48 produces both A␤40 and A␤42, whereas A␤49 produces predominantly A␤40 (10), fits the present cleavage model well. The FIGURE 7. Time-dependent A␤ generation and the release of various oligopeptides by brain raft-associated ␥-secretase. Lipid rafts prepared from the brain of rats were subjected to an in vitro A␤ generation assay by the addition of C99-FLAG. A, Western blotting with 82E1 (total A␤) and A␤40-and A␤42-specific antibodies. Authentic A␤ species were applied onto the rightmost lane. The increases in A␤ were suppressed by L-685,458 at 1 M (an asterisk in A), indicating ␥-secretase-mediated generation of A␤. B, quantification by LC-MS/MS of the predicted five tripeptides and one tetrapeptide released by brain raft-associated ␥-secretase. All the peptides increased in a time-dependent manner, as expected. C, LC-MS/MS quantification of other oligopeptides released by brain raft-associated ␥-secretase. Oligopeptides similar to those released by the CHO raft-associated ␥-secretase were released following incubation, although GVV and VIA were below the detection limit. significant release of the pentapeptide from A␤43 found here may explain why a decrease in A␤42 and an increase in A␤38 are uncoupled in unusual cases (21,31).
The release of the pentapeptide VVIAT by ␥-secretase was reported quite recently by Okochi et al. (32) with a cell-based system and an in vitro reconstituted system using synthetic A␤ as a substrate. This study further identified two pentapeptides and estimated the relative contribution of those pentapeptides in A␤-producing pathways, using ␤CTF and analyzing the released oligopeptides systematically by LC-MS/MS. This indicates that cropping of a pentapeptide is not an unusual pathway specific for A␤38 generation. Furthermore, the release of VIVIT links the A␤40 and A␤42 product lines and may have an important role to define the A␤ species generated (see below).
One may think that FAD-associated mutations simply enhance the A␤42 product line, resulting in increased A␤42; however, that may be unlikely, because multiple pathways can be involved in the production and metabolism of A␤42. There are many potential sites where mutations may produce effects. Those effects could modify the cleavage pattern in various ways that would lead to the common phenotype. Only the mutations associated with increased A␤42 (in most cases) may have been selected as FAD mutations. In this context, it would be interesting to determine whether the release of VIVIT, which brings FIGURE 8. Uniform inhibition by L-685,458 and DAPT of oligopeptide production and enhancement by GSM-1 of VVIA and VVIAT release in the brain raft-associated ␥-secretase assay. The reconstituted A␤ generation assay with brain lipid raft-associated ␥-secretase was performed in the presence of the indicated concentrations of L-685,458 (A), DAPT (B), or GSM-1 (C). The reaction mixture was subjected to LC-MS/MS analyses to quantify the amounts of oligopeptides released. Left panel shows the predicted five tripeptides and one tetrapeptide released, and right panel shows other oligopeptides released. The release of all peptides was suppressed uniformly by L-685,458. DAPT also suppressed uniformly the release of all peptides. Only the peptides quantified consistently in significant amounts are shown here. GSM-1 enhanced remarkably the release of VVIA and VVIAT in a dose-dependent manner (p ϭ 0.0016 and 0.0033, respectively, Spearman's rank correlation). The release of VVI increased slightly by GSM-1 but not significantly. Data are represented as means Ϯ S.E. from four independent experiments. **, p Ͻ 0.01. about the conversion of A␤48 to A␤43, is influenced by FAD mutations, because FAD-associated mutations of PS cause an increase in A␤43 (33) in addition to increases in A␤42 and AICD(49 -99), a counterpart of A␤48 (9). This remains to be investigated in the future and is beyond the scope of this study.
GSM-1 enhances all three routes generating A␤38, resulting in a large increase in A␤38 along with a significant decrease in both A␤42 and A␤43 (Fig. 6). As A␤43 is also pathogenic and neurotoxic (33), GSM-1 could be a better candidate reagent for AD therapeutic drugs. It has been reported that GSM-1 binds to the transmembrane domain 1 of PS1 (34) and alters the kinetics of ␥-secretase activity for cleaving A␤42 (32). However, our results indicate that the effects of GSM-1 are not necessarily limited to the cleavage of A␤42. GSMs may rather target the mid-portion of the transmembrane domain of ␤CTF and induce its cleavage to proceed one step further, resulting in the generation of shorter A␤ species.
In conclusion, this study has revealed the overall profile for A␤ generation pathways: the successive tripeptide-cropping pathway is the framework by which membrane-integrated ␥-secretase cleaves ␤CTF. The recent NMR investigations of the TMD of ␤CTF lend substantial support to the model (35,36). However, the pathways are not as simple as we previously thought. In particular, the concomitant release of tetra-and pentapeptides leads to cross-talk between the stepwise processing pathways, giving rise to diverse A␤ species generated through variable pathways. These observations extend our understanding of the intramembranous cleavage of ␤CTF by ␥-secretase and may contribute toward the eventual development of an efficient therapeutic strategy against AD.