A Novel γ-Secretase Assay Based on Detection of the Putative C-terminal Fragment-γ of Amyloid β Protein Precursor*

Alzheimer's disease is characterized by the deposits of the 4-kDa amyloid β peptide (Aβ). The Aβ protein precursor (APP) is cleaved by β-secretase to generate a C-terminal fragment, CTFβ, which in turn is cleaved by γ-secretase to generate Aβ. Alternative cleavage of the APP by α-secretase at Aβ16/17 generates the C-terminal fragment, CTFα. In addition to Aβ, endoproteolytic cleavage of CTFα and CTFβ by γ-secretase should yield a C-terminal fragment of 57–59 residues (CTFγ). However, CTFγ has not yet been reported in either brain or cell lysates, presumably due to its instability in vivo. We detected thein vitro generation of Aβ as well as an ∼6-kDa fragment from guinea pig brain membranes. We have provided biochemical and pharmacological evidence that this 6-kDa fragment is the elusive CTFγ, and we describe an in vitro assay for γ-secretase activity. The fragment migrates with a synthetic peptide corresponding to the 57-residue CTFγ fragment. Three compounds previously identified as γ-secretase inhibitors, pepstatin-A, MG132, and a substrate-based difluoroketone (t-butoxycarbonyl-Val-Ile-(S)-4-amino-3-oxo-2,2-difluoropentanoyl-Val-Ile-OMe), reduced the yield of CTFγ, providing additional evidence that the fragment arises from γ-secretase cleavage. Consistent with reports that presenilins are the elusive γ-secretases, subcellular fractionation studies showed that presenilin-1, CTFα, and CTFβ are enriched in the CTFγ-generating fractions. The in vitroγ-secretase assay described here will be useful for the detailed characterization of the enzyme and to screen for γ-secretase inhibitors.

The A␤ 1 that is invariably deposited in Alzheimer's disease (AD) is a 38 -43-residue peptide derived from a larger precursor, APP, as summarized below ( Fig. 1) (1)(2)(3)(4)(5)(6). Normal neuronal cells constitutively secrete A␤, which is detected in cerebrospinal fluid and blood (7,8). The length of most of the secreted A␤ molecules is 40 residues (A␤40) (9 -11), but a small fraction (ϳ10%) is 42 residues long (A␤42) (12). Mutations in APP (12), presenilin-1 (PS1), and presenilin-2 (PS2) (13)(14)(15) linked to familial AD (FAD) invariably increase the levels of either A␤42 alone or both A␤40 and A␤42, providing evidence that it plays an important causative role in the pathogenesis of AD (16,17). Furthermore, there is evidence that in typical late onset AD, as in FAD, there are genetic determinants that increase levels of A␤42 (18). To generate A␤, APP is sequentially cleaved by proteases referred to as ␤and ␥-secretase as described below. ␤-Secretase cleaves at the N-terminal end of the A␤ sequence, producing a secreted derivative, sAPP␤ (19), and the A␤-bearing C-terminal fragment of 99 residues, CTF␤, which is subsequently cleaved by ␥-secretase to release A␤ (2). An alternative activity, ␣-secretase, cleaves APP within A␤ (between residues 16 and 17) to the larger secreted derivative, sAPP␣ (20,21), and membrane-associated 83-residue fragment, CTF␣ (22). Cleavage of CTF␤ and CTF␣ by ␥-secretase generates A␤ and a smaller fragment of 24 -26 residues called P3, respectively (4). In addition, ␥-secretase cleavage should theoretically yield an ϳ6-kDa fragment of 57-59 residues, CTF␥ ( Fig. 1). Although the focus of the field has been on A␤, it is important to recognize that A␤42 is naturally linked to CTF␥-57. Expression of CTF␤ was shown to be toxic to neurons (23,24), and more recently, it was suggested that the 31 C-terminal residues of APP are responsible for at least part of the toxicity (25). Since CTF␥ can also serve as a precursor for this toxic 31-residue peptide, its production and turnover are important parameters that can influence the course of AD.
The secretases that ultimately cleave APP to A␤ and CTF␥ have been the subject of intense research as potential targets for the treatment of AD. A novel pepstatin A-insensitive aspartyl protease, BACE/Asp2, was recently identified as the ␤-secretase that cleaves at positions 1 and 11 of the A␤ sequence (26 -29). In addition to BACE, thimet oligopeptidase, a metalloprotease, was shown to be involved in ␤-secretase activity in COS7 cells (30). We recently showed that GPI-anchored proteins play an important role in ␤-secretase activity (31). Recently, a family of disintegrin metalloproteases, the adamalysins such as tumor necrosis factor ␣-converting enzyme and ADAM 10, has been implicated in ␣-secretase cleav-age of APP (32)(33)(34). However, tumor necrosis factor ␣-converting enzyme was shown to be specific for the phorbol esterinduced ␣-secretase activity (33). Although ADAM 10 appeared to be involved in both constitutive and inducible pathways, inhibition of this activity by a dominant-negative mutation indicated that it also plays a more important role in the inducible ␣-secretase pathway (34).
The final step in A␤ biogenesis is ␥-secretase cleavage and is the activity responsible for generating CTF␥ in vivo. This cleavage is particularly interesting as the scissile bond lies within the transmembrane domain (35,36). Since the discovery that a PS1 knockout mutant mouse was deficient in ␥-secretase activity, the possibility was raised that PS1 and PS2 are the elusive ␥-secretases (37). The finding that PS1 and PS2 show conserved and essential aspartate residues within their respective transmembrane domains led to the suggestion that they are unusual aspartyl proteases with active sites within or close to the membrane (38). Recently, an in vitro assay for ␥-secretase was described, and PS1 was found to be a part of the active protease (39). By using this assay, a biotinylated inhibitor was cross-linked to PS1 and PS2, showing that the presenilins are indeed the active subunits of ␥-secretase (40,41). However, the active enzyme was shown to be a large complex, presumably with many unidentified subunits, of which some may be also essential for ␥-secretase activity. To understand this interesting cleavage event, it is important to tease out each component of the ␥-secretase and examine its individual role in detail by a combination of biochemical and genetic methods. To address these issues, it is necessary to have a robust in vitro ␥-secretase assay in both the membrane-associated and soluble states. To develop such an assay, we incubated membranes from guinea pig and cow brains as well as from cultured cells to look for production of A␤ and CTF␥. The CTF␥ fragment was detected in all systems tested, but guinea pig brain was used because the entire APP sequence is known, and the sequence of A␤ is identical to human allowing the use of human-specific reagents for its analysis. After appropriate fractionation, we observed a consistent time-dependent generation of a putative CTF␥ fragment from these membranes. In addition to describing a useful robust cell-free assay for ␥-secretase, this study presents the initial characterization of CTF␥ from brain membranes.
Antibodies, Peptides, and Standards-Antibodies against marker proteins syntaxin-6 (43) (Transduction Laboratories) and synaptophysin (44) (Roche Molecular Biochemicals) were kind gifts from Drs. Sevlever (Mayo Clinic, Jacksonville, FL) and Lahiri (Indiana University, IN). The anti-cross-reacting determinant (anti-CRD) antibody (Ox-ford Glycosciences) reacts with a neoepitope generated after the cleavage of GPI-anchored proteins by PI-PLC. Pf998, a rabbit polyclonal antibody against the juxtamembrane cytoplasmic domain of APP (726 -744 of APP770) and the synthetic C-terminal 57 residues of APP (CTF␥-57) were kind donations from Pfizer. CTF␥-57 includes 10 residues from the transmembrane domain and the entire 47-residue cytoplasmic domain of APP and corresponds to one of the predicted CTF␥ fragments. The monoclonal antibodies BNT77, BA27, and BC05 utilized for the quantification of A␤40 and A␤42 were kind gifts of Dr. Nobu Suzuki and have been described previously (12).
The rabbit antibody O443 was raised against a maleimide-activated KLH-conjugated synthetic peptide (CKMQQNGYENPTYKFFEQMQN) prepared at the Mayo Clinic Protein Core Facility, which corresponds to the C-terminal 20 residues of APP. The animal care, injections, and bleeds were carried out by Cocalico Biologicals, Inc. (Animal approval number A3669-01), and the sera were characterized in our laboratory. The antibody detects less than 0.1 fmol (0.6 pg) of synthetic CTF␥ by Western blot analysis. The anti-PS1 antibody was a copy of that reported by Duff et al. (45) and was prepared against a KLH-conjugated synthetic peptide corresponding to residues 2-13 (CRKTELPAPLSYF).
The LC99 construct in the pCEP4 vector consisting of the signal sequence of APP fused to its C-terminal 99 residues, starting at the A␤ sequence, was transfected into CHO cells (CHO C99) for use as a standard for CTF␤.
Preparation and Incubation of Active Membrane Fractions-Brains were minced and homogenized for 1 min on ice in 10 volumes of buffer A (50 mM HEPES, 150 mM NaCl, and 5 mM EDTA, pH 7.4). All subsequent steps were carried out at 4°C unless otherwise indicated. Homogenates were sequentially fractionated by centrifugation at 2500 ϫ g for 15 min to collect unbroken cells and nuclei (F1P), and the post-nuclear supernatant was spun at 10,000 ϫ g for 15 min (F2P) followed by 100,000 ϫ g for 1 h (F3P). The supernatant (F3S) was discarded, and the three pellets were washed once in buffer A and resuspended in 200 l/g initial brain weight of buffer B (50 mM HEPES buffer, 150 mM NaCl, pH 7.0). For A␤ measurements, the F2P fraction was incubated in buffer B supplemented with 5 mM EDTA, 5 mM PNT, and 1 mM thiorphan. The G9PLAP cell line was similarly prepared, except that the initial homogenate was directly centrifuged at 10,000 ϫ g for 15 min to obtain a F2P pellet that included membranes from F1P as well. For most assays, fraction F2P was rehomogenized for 10 s and incubated at 37°C for the indicated time. The reactions were stopped by chilling on ice, and the membranes were removed by centrifugation at 10,000 ϫ g for 15 min unless otherwise specified. Finally, the supernatants were analyzed for A␤ and the released C-terminal fragments of APP.
To determine the pH optimum of fragment generation, aliquots of F2P were resuspended and incubated either in 50 mM sodium citrate, 150 mM NaCl, pH 5.0 -6.5, buffer B, pH 7.0 -7.5, or 50 mM Tris, 150 mM NaCl, pH 8.0 -9.0. To make sure that the alternative buffers were not interfering with the assay, we also carried out the pH optimum analysis in buffer B adjusted to the various pH levels and essentially obtained the same results (data not shown).
Sucrose Density Gradient Fractionation of F2P-Step gradients were set up using 25, 27.5, 30, 32.5, 35, 37.5, and 40% sucrose in buffer A containing 5 mM PNT in 7 layers of 0.6 ml each. The F2P obtained from 0.75 g of guinea pig brain was resuspended in buffer A containing 20% sucrose, overlaid onto the gradient, and centrifuged for 20 h at 48,000 rpm in a Beckman SW 50.1 rotor. Eight fractions (0.6 ml) were collected from the top of the gradient and diluted 5-fold in buffer A. Membrane pellets were collected by centrifugation at 100,000 ϫ g for 1 h and then resuspended in equal volumes of buffer B. The protein concentration peaked in fractions 5 and 8. Fractions 1 and 2 containing low levels of protein were pooled for further analysis. The fractions were incubated, and the supernatants were analyzed for the release of CTF␥ as described earlier.
The membrane pellets from the fractions were analyzed for the presence of marker proteins. Markers of the Golgi, syntaxin-6 (31 kDa), synaptic vesicles and synaptophysin (38 kDa), were readily identified by their size. The anti-CRD antibody detects multiple GPI-anchored proteins as broad patches of ϳ38 and 62 kDa. The bands were considered as specific as they were only detected after PI-PLC treatment (data not shown).
Immunoassays-Western blotting and immunoprecipitation were carried out as described previously (46), except that Hammerstein grade casein (1%) in 25 mM Na 2 HPO 4 , 0.2 M NaCl, pH 7.0 was used as a blocking agent for some of the antibodies, and Bis-Tris gels with the MES running buffer system was used for protein separation. Where relevant, scanned images were quantified using the ImageQuant software (Molecular Dynamics). For comparison of treatments between experiments, treated samples were compared with controls adjusted to 100%. A␤ was measured as reported earlier by a specific and sensitive sandwich ELISA using a monoclonal antibody BNT77 against A␤11-28 for capture and horseradish peroxidase-labeled end-specific antibodies BA27 (A␤40) or BC05 (A␤42) for detection (12,31).

Generation of a Novel C-terminal APP Fragment from
Guinea Pig Brain Membranes-To assay for ␥-secretase activity, guinea pig brains were homogenized and fractionated as described under "Experimental Procedures" into F1P, F2P, and F3P and resuspended in equal volumes of buffer B. Since the yield of membranes from CHO cells was low, these membranes were only fractionated into 10,000 and 100,000 ϫ g pellets (F2P and F3P). Equal volumes of each fraction were incubated for 0 and 2 h at 37°C, chilled on ice, and centrifuged at 100,000 ϫ g to remove the membranes. Supernatants were examined by Western blotting with the O443 antibody raised against the C-terminal 20 residues of APP (Fig. 1). Western blots detected a soluble C-terminal fragment of ϳ6 kDa from the F1P and F2P of guinea pig brain and the F2P fraction of CHO cells, which increased considerably after 2 h at 37°C (Fig. 2A). The fragment was not detected upon incubation of F3P ( Fig. 2A). For further characterization, we immunoprecipitated the supernatant obtained from the F2P fraction of guinea pig brain with Pf988, an antibody against the juxtamembrane domain of APP (residues 726 -744; Fig. 1), and we analyzed the recovered APP C-terminal fragments by Western blotting with O443 as above. The capacity to immunoprecipitate the putative CTF␥ with Pf998 shows that it includes epitopes from the juxtamembrane region on the cytoplasmic domain of APP (Fig. 2B).
We also measured the levels of A␤ in the supernatant using a well characterized sandwich ELISA (12). The data obtained from 10 independently processed brain samples in three experiments detected moderate levels of A␤40 (62.3 Ϯ 3.3 pM; Fig. 3). A␤42 was also detected at low levels of 5.4 Ϯ 1 pM consistent with the in vivo findings that A␤42 constitutes 5-10% of the secreted A␤ (12). The increase in A␤40 was highly significant (p ϭ 4.7 ϫ 10 Ϫ12 ) but A␤42 was not (p ϭ 0.18), presumably due to the large variance in background levels (two-tailed t test).
To determine the stability of CTF␥, the supernatant obtained after removal of membranes from the incubation mix was incubated further for 2 and 4 h in the absence of membranes. The CTF␥ fragment was degraded by over 50% after 2 h and by over 90% after 4 h at 37°C. Addition of PNT, a metalloprotease inhibitor, partially protected the fragment from degradation after 4 h (Fig. 4). In addition, PNT increased the yield of the C-terminal fragment by over 2-fold, suggesting that one or more metalloproteases in the preparation degrade the fragment and/or the enzyme responsible for its generation (Fig. 4). Although several protease inhibitors including phenylmethylsulfonyl fluoride, E64, leupeptin, and ALLN were tested, protection was only seen by EDTA and PNT, making metalloproteases the only identified class important for degrading the released CTF␥ in vitro (data not shown). The pH optimum of the observed ␥-secretase activity was between pH 7 and pH 7.5, but activity was detected in a broad range from pH 5 to pH 9 (Fig. 5). This pH optimum is similar to that recently reported for ␥-secretase activity using membranes purified from HeLa cells (39).
We attempted to determine the cleavage site(s) that generate CTF␥ in collaboration with Dr. Rong Wang (Rockefeller University, New York). Presumably due to its low levels and un-FIG. 1. Major APP processing pathways. The APP holoprotein is a type I integral membrane protein with a large N-terminal extracellular domain, a single transmembrane domain, and a short cytoplasmic tail. A␤ depicted as a violet box is partly extracellular (28 residues) and partly embedded within the membrane (12-14 residues). Antibodies O443 and Pf998 are against the final 20 residues and 19 juxtamembrane residues, respectively. APP is cleaved to sAPP␣ and CTF␣ by ␣-secretase or sAPP␤ and CTF␤ by ␤-secretase. The membrane-bound CTF␣ and CTF␤ are further cleaved within the transmembrane domain by ␥-secretase (green arrow) to generate P3 and A␤, respectively. Most (90%) A␤ and P3 end at residue 40 and a small fraction (5-10%) after residue 42 of the A␤ sequence. In addition, ␥-secretase cleavage should yield a C-terminal fragment, CTF␥, of 57-59 residues.
FIG. 2. A 6-kDa C-terminal fragment of APP is released on incubation of membrane fractions. Guinea pig brain was fractionated into 2,500 ϫ g (F1P), 10,000 ϫ g (F2P), and 100,000 ϫ g (F3P) fractions, as described under "Experimental Procedures." CHO cells were processed into two fractions of 10,000 ϫ g (F2P) and 100,000 ϫ g (F3P). Fractions were incubated for 0 or 2 h in the presence of 5 mM PNT, and the membranes from the incubation mixture were removed by centrifugation. Supernatants were directly analyzed on Western blots using O443 antibody (A). B, the incubated guinea pig brain F2P fraction was immunoprecipitated using Pf998 and then analyzed by Western blotting with O443. Lysates from CHO cells expressing LC99 (C99) mark the positions of the CTF␤ and the 83-residue CTF␣ fragments. favorable flying characteristics, the exact size of CTF␥ could not be detected by matrix-assisted laser desorption/ionization time of flight (MALDI-TOF) mass spectroscopy. However, the fragment comigrates with a 57-residue synthetic peptide corresponding to one of the predicted CTF␥ fragments on polyacrylamide gels (Fig. 6). Judging from the separation of CTF␣ from CTF␤ and the molecular weight markers, small changes of over five residues (0.5 kDa) should be readily detected in this size range. Thus, the cleavage size is likely to be close to the C-terminal end of the A␤ sequence and occur within the transmembrane domain of APP as predicted for ␥-secretase activity (Fig. 6).
Caspase Inhibitor Does Not Reduce CTF␥ Production-Previous studies have shown that caspase can cleave within the cytoplasmic domain of full-length APP in cells undergoing apoptosis (25,(47)(48)(49). This cleavage resulted in an ϳ3-kDa fragment and was inhibited by a broad spectrum caspase inhibitor, Z-VAD-FMK (49). It is unlikely that the observed 6-kDa fragments are generated by caspase, as the fragment is much larger than reported, and this cleavage occurs in the membrane pellet fraction after washing away the cytoplasm. However, we cannot rule out that some caspase is bound to the membrane pellet. To rule out the possibility that the CTF␥-like fragment is a product of caspase cleavage, we included Z-VAD-FMK, a broad spectrum caspase inhibitor, in our incubations. Z-VAD-FMK did not inhibit the generation of the putative CTF␥ (Fig.  6). We also incubated the 57-residue synthetic CTF␥ as well as the brain membrane fraction with caspase Ϯ Z-VAD-FMK.
Caspase-3 cleaved the synthetic CTF␥ as shown by the reduction in band intensity, and Z-VAD-FMK blocked this reduction, indicating that the inhibitor was active under the conditions used (Fig. 6). Similarly, reduction in the yield of the 6-kDa CTF␥ was also observed when brain membranes were spiked with purified caspase, which was restored in the presence of Z-VAD-FMK (Fig. 6).
␥-Secretase Inhibitors Reduce CTF␥-A number of protease inhibitors were tested to determine the class of proteases involved in the generation of the putative CTF␥ and to understand the role of ␥-secretase in generating this fragment. Previous reports have already identified several inhibitors that lower ␥-secretase activity as defined by a reduction in the production of A␤ and by the increase in CTF␣ and CTF␤ levels in cell lysates (2). Three of these inhibitors are pepstatin-A, MG132, and Wolfe-1 (a ␥-secretase substrate-based difluoroketone inhibitor). F2P membranes treated with the inhibitors pepstatin-A (Fig. 7A, lanes 5-8),   Fig. 7A, lanes 9 -12), and MG132 (Fig. 7B, lanes 7-12) showed a dose-dependent reduction in the yield of the 6-kDa fragment generated from brain membranes in vitro. These data provide additional evidence, indicating that the putative CTF␥ fragment is a product of ␥-secretase cleavage.
CTF␥ Is Generated in a Fraction Enriched in CTF␣, CTF␤, and Presenilin-CTF␣ and CTF␤ are the postulated substrates for ␥-secretase activity, and PS1/PS2 are reported to be an integral part of purified ␥-secretase (39). Furthermore, reports show that PS1 binds APP and its C-terminal fragments, consistent with its role in ␥-secretase activity (50 -52). To examine the role of PS1 in the generation of CTF␥, we analyzed brain membrane fractions. As described earlier, we initially obtained three membrane pellet fractions F1P, F2P, and F3P generated after sequential centrifugation. The fractions were probed with antibodies against APP and PS1 and marker proteins as described under "Experimental Procedures". The cell-surface GPI-anchored proteins (53) and the Golgi-marker syntaxin-6 (43) were preferentially found in F3P, which also had the highest concentration of protein (Fig. 8). In contrast, most of the ␥-secretase activity together with CTF␣ and CTF␤ was in F2P, suggesting that ␥-secretase activity was not enriched in the plasma membrane or in Golgi vesicles. Although the Nterminal fragment of PS1 was observed in all three fractions, its mobility was somewhat retarded in F3P. The reason for this shift in mobility is not known but may be due to post-translational modification such as phosphorylation of PS1 (54).
The active F2P was further fractionated by centrifugation to equilibrium in a sucrose step gradient ranging from 20 to 40% sucrose as described above. The eight fractions were examined for ␥-secretase activity by incubating membrane pellets as described earlier. In addition, the membranes were analyzed on Western blots for APP, CTF␣, CTF␤, and PS1 (Fig. 9). The peak  1) and 2 h (lanes 2-5). Z-VAD-FMK was included in lanes 3 and 5, and caspase-3 was included in lanes 4 and 5. Note that caspase-3 reduces the yield of CTF␥ in lane 4, but the yield is recovered when Z-VAD-FMK is added in lane 5. The synthetic CTF␥-57 (lane 6) migrates with the fragment generated from brain membranes (lanes 2-5). The synthetic CTF␥ is degraded by caspase-3 (lane 7), which is protected by Z-VAD-FMK (lane 8).
of ␥-secretase activity as determined by the relative intensity of the CTF␥ fragment generated was in fraction 5. Interestingly, PS1, CTF␣, and CTF␤ are all enriched in this fraction. How-ever, it is important to note that PS1 was distributed in several fractions with low ␥-secretase activity (Figs. 8 and 9). Recent findings show that most of the PS1 is in the endoplasmic reticulum and intermediate compartments with small amounts in the Golgi and cell surface (55). PS1 was also detected in detergent-resistant glycosphingolipid-enriched membranes (DIGs (56)), which was also enriched in cell-associated A␤ (57). Our preliminary observations 2 show that PS1, BACE, and the ADAM 10 protease (Kuzbanian) are all enriched in detergentinsoluble glycosphingolipid-enriched membranes isolated from guinea pig brain. 3 Since we did not detect ␥-secretase activity in detergent-insoluble glycosphingolipid-enriched membranes prepared by flotation of Triton X-100-extracted membranes, we homogenized guinea pig brain in carbonate buffer to strip peripheral proteins, and generated a membrane fraction enriched in caveolin as described by Lisanti and coworkers (58). The ␥-secretase activity was enriched in the caveolin-rich fractions suggesting that it may be present in DIG-related membranes. 3 Solubilization of the Active ␥-Secretase-The in vitro ␥-secretase activity is lost when either F2P or membrane fraction 5 purified by sucrose density gradients is extracted with several detergents such as Triton X-100, methyl ␤-cyclodextrin, digitonin, Nonidet P-40, CHAPS, and octyl ␤-glucoside (data not shown). The data presented compare ␥-secretase activity in fraction 5 in the presence of Brij 35, Big CHAP, and CHAPSO (Fig. 10A). ␥-Secretase activity, determined as an increase in CTF␥ production, appeared equivalent to the detergent-free control in CHAPSO, reduced in Brij 35, and almost absent in Big CHAP. Thus, the detergent extraction profile of this in vitro activity is similar to the in vitro assay recently reported by Li et al. (39). Since dissolving the activity is essential for its further purification and characterization, we removed the CHAPSO-insoluble membranes by centrifugation at 100,000 ϫ g for 60 min and compared ␥-secretase activity in the supernatant with activity in the mixture (Fig. 10B). Our data indicate that most of the activity was extracted from the sucrose gradient purified fraction 5 in this detergent, as judged by the approximately similar intensities of the CTF␥ band obtained by incubation of equivalent amounts of the CHAPSO-solubilized supernatant alone and the CHAPSO-treated membrane mix. Solubilization of the activity is necessary for further biochemical analysis of ␥-secretase in vitro.  8. CTF␥ is generated in a fraction distinct from the cell surface and the Golgi apparatus. Guinea pig brain fractions F1P, F2P, and F3P were analyzed on Western blots stained using reagents shown on the right side of each panel. Note that full-length APP is enriched in F3P, whereas CTF␣, CTF␥, and some of the CTF␤ are enriched in F2P. The N-terminal fragment of PS1 at ϳ30 kDa was seen in all fractions. However, the fragment was slightly shifted up in F3P. The anti-CRD antibody against PI-PLC-cleaved GPI-anchored proteins detected two major groups of bands in the 38-and 62-kDa range primarily in F3P. Each of these groups presumably represents several GPI-anchored proteins in these size ranges. The Golgi marker, syntaxin-6 at 31 kDa, was also primarily in F3P. The synaptic vesicle marker, synaptophysin, was primarily detected in F1P and F2P indicating that most of it is pelleted at 10,000 ϫ g, although a sizable quantity is also seen in F3P. A total protein stain (Ponceau S) detected the strongest signal in F3P, although Fig. 2 shows that most of the CTF␥ is generated in F2P.
FIG. 9. The peak of PS1, CTF␣, and CTF␤ coincide with the peak of CTF␥ generation. F2P from guinea pig brain was further fractionated on sucrose density gradients, and eight fractions were collected from the top of the tube as described under "Experimental Procedures," and equal volumes of each fraction were analyzed by Western blotting with O443 (top three panels) and anti-PS1 N terminus (bottom panel). Fractions 1 and 2 were mixed and loaded in the 1st lane, and the remaining fractions were loaded in individual lanes. The top panel shows full-length APP, the 2nd panel CTF␣, and the 3rd panel CTF␥ generated after incubation for 2 h. CTF␤ can be detected as a faint band above CTF␣ in the 2nd panel. The bottom panel shows the ϳ30-kDa PS1-NTF. Note that PS1-NTF, APP, CTF␣, CTF␤, and generation of CTF␥ peak in fraction 5.

DISCUSSION
The current study reports a novel ␥-secretase assay from brain membranes and is the initial description of CTF␥, a previously undescribed fragment predicted as a product of ␥-secretase cleavage of APP. The CTF␥ fragment is normally not detected in cell lysates and brain homogenates, probably due to its instability in vivo. A possible exception is a report of a faint 5.8-kDa CTF in human brain homogenates (22). However, CTF␥ is readily detected in an in vitro ␥-secretase assay based on fractionated brain membranes or cell homogenates. The in vitro detection of CTF␥ strongly suggests that ␥-secretase cleavage is due to a specific endoproteolytic cleavage and not random degradation of the cytoplasmic tail domain of APP as suggested previously (59).
An in vitro assay for ␥-secretase has been recently reported by looking for A␤ as a product using a purified substrate consisting of the C-terminal 99 residues of APP (39). In the presence of thiorphan, an inhibitor of A␤ degradation in the brain (60), we were also able to detect the generation of A␤40 (Fig. 3) as well as A␤42 in our in vitro assay. Since a pool of CTF␥ is not already present in brain or cell lysates at detectable levels, the low background allows the characterization of ␥-secretase activity from fractionated membranes more easily than the measurement of A␤. Based on the reports that ␤-secretase accounts for Ͻ10% of the secretory processing of APP (22), using CTF␥ as an end point for ␥-secretase activity may be 10 times more sensitive than the measurement of A␤. In addition, the measurement of CTF␥ is not complicated by aggregation like A␤ or by the presence of alternative fragments such as P3 (Fig. 1).
Subcellular fractionation studies based on the in vitro generation of CTF␥ suggest that ␥-secretase activity is localized in fractions enriched in CTF␣ and CTF␤. PS1 is apparently enriched in these fractions, consistent with the currently favored hypothesis that presenilins include the active site of ␥ secretase. However, high levels of PS1 are still seen in fractions showing little in vitro ␥-secretase activity. Thus, if PS1 is indeed the active subunit of ␥-secretase as suggested, only a small subset of it, presumably in a special compartment, is involved in ␥-secretase activity. The restriction of the activity may be due to either the observed enrichment of the substrate CTF␣ and CTF␤ in this compartment, limitations in essential components of the ␥-secretase complex other than presenilins, or unfavorable conditions in the isolated organelle.
Although high levels of A␤ are secreted by several cell lines in culture, CTF␥ is not readily detected in cell lysates, indicating that it is either rapidly degraded in the cytoplasm or sequestered in a manner that prevents its detection. Similar fragments from other membrane proteins such as Notch and SREBP are transported into the nucleus where they act as transcription factors, are rapidly degraded, and are difficult to detect (61,62). The C-terminal tail of APP includes a sequence (KFFEQ) that resembles a motif present in soluble proteins that are rapidly degraded in lysosomes (63). It will be useful to determine whether this sequence is responsible for rapid CTF␥ turnover in cells and to identify alternative cellular pathways for its degradation, if any. Recent reports show that PS1 specifically binds a ␥-secretase inhibitor indicating that the active site of ␥-secretase lies within the transmembrane domains of PS1 and PS2 (40,41). It was proposed that presenilins are unusual aspartyl proteases, but this is not definitive as the aspartyl protease inhibitor, pepstatin A, can inhibit other classes of proteases at high concentrations (64). In addition, other ␥-secretase inhibitors are known to inhibit several classes of proteases including serine proteases, cysteine proteases, and the proteasome (2). The data showing ␥-secretase activity in fractions enriched in PS1 agree with studies using knockout mice deficient in PS1/ PS2 that fail to generate A␤ and accumulate CTF␣ and CTF␤ (37,65,66). The detection of CTF␥ provides another powerful tool for the analysis of ␥-secretase activity in vitro and for identifying mechanisms involved in its regulation and for developing drugs that block this activity.
Although presenilins have been identified as being essential for ␥-secretase activity and are shown to contain the active sites, it is probably not sufficient, as it has not been possible to reconstitute the activity with pure PS1/PS2 and purified ␥-secretase is a multisubunit complex (39). The mechanism of the endoproteolytic cleavage of the membrane-spanning domains of proteins may identify the role of each component. For example, it has been suggested that the cleavage is initiated by the sliding of the peptide out of the membrane into the cytoplasm (67). However, the energy required for this sliding reaction is likely to be very high. It is also likely that the reaction is mediated by a water molecule within the active site of the enzyme for the hydrolytic reaction. Thus, a water molecule may be held between the two transmembrane aspartate residues on PS1/PS2 that serve to hydrolyze within the transmembrane domain. It is possible that some of the unidentified subunits are involved in either transport of water or the APP for hydrolysis. The mechanisms for maintaining a supply of the water molecules in the membrane for the hydrolysis may provide useful insights into the mechanisms of ␥-secretase cleavage and identify additional therapeutic targets for inhibiting this activity. A similar intramembrane proteolytic cleavage has been described for the sterol regulatory element binding protein (SREBP). The protease that cleaves SREBP has been identified as S2P by genetic complementation of a mutant cell line. The predicted active site of this enzyme resembles that of a metalloprotease with the exception that hydrophobic residues flank the domain, suggesting its intramembranous location (62). The development of an in vitro assay for S2P may be useful for comparison with ␥-secretase and also provide useful insights into intramembrane proteolysis.
Since the generation of CTF␥ is closely tied to the generation of A␤, the mutations that increase the production of A␤42 should also increase the 57-residue CTF␥ starting at residue 43 of the A␤ sequence. Since FAD mutations that increase A␤ by preventing its degradation have not been reported, the correlation between CTF␥-57 and AD is as strong as A␤42. In addition, CTF␥ includes a peptide sequence that is reported to FIG. 10. The in vitro ␥-secretase activity can be solubilized in select detergents. Membranes from fraction 5 shown in Fig. 9 were diluted and collected by centrifugation as described under "Experimental Procedures." A, the fraction was incubated in pairs in the absence (lanes 1 and 2) and presence (lanes 3-14) of the indicated detergents for 0 and 2 h. Activity was recognized as an increase in the CTF␥ level at 2 h (even lanes) over 0 h (odd lanes) for each condition. Note that CTF␥ levels did not increase in the presence of Big CHAP (lanes 7-10). B shows that CTF␥ is generated in the CHAPSO-soluble fraction. CHAPSO-resistant membranes were removed by spinning at 100,000 ϫ g for 1 h, and the supernatants were incubated for 0 (lanes 1 and 4) and 2 h (lanes 2 and 5). The CHAPSO-treated membranes were incubated for 2 h (lanes 3 and 6) as in A. Note that removal of CHAPSO-resistant membranes did not reduce the yield of CTF␥. The CHAPSO-resistant membrane pellet did not generate any CTF␥ (data not shown).
induce apoptosis in neurons as discussed in the Introduction. The localization of the enzyme, the biological activity of the fragment, the cleavage site involved in its production, the mechanism of its turnover, and its role in AD are important unanswered questions that will be facilitated by this initial description of CTF␥.