γ-Secretase, Evidence for Multiple Proteolytic Activities and Influence of Membrane Positioning of Substrate on Generation of Amyloid β Peptides of Varying Length*

γ-Secretase activity is the final cleavage event that releases the amyloid β peptide (Aβ) from the β-secretase cleaved carboxyl-terminal fragment of the amyloid β protein precursor (APP). No protease responsible for this highly unusual, purportedly intramembranous, cleavage has been definitively identified. We examined the substrate specificity of γ-secretase by mutating various residues within or adjacent to the transmembrane domain of the APP and then analyzing Aβ production from cells transfected with these mutant APPs by enzyme-linked immunosorbent assay and mass spectrometry. Aβ production was also analyzed from a subset of transmembrane domain APP mutants that showed dramatic shifts in γ-secretase cleavage in the presence or absence of pepstatin, an inhibitor of γ-secretase activity. These studies demonstrate that γ-secretase’s cleavage specificity is primarily determined by location of the γ-secretase cleavage site of APP with respect to the membrane, and that γ-secretase activity is due to the action of multiple proteases exhibiting both a pepstatin- sensitive activity and a pepstatin-insensitive activity. Given that γ-secretase is a major therapeutic target in Alzheimer’s disease these studies provide important information with respect to the mechanism of Aβ production that will direct efforts to isolate the γ-secretases and potentially to develop effective therapeutic inhibitors of pathologically relevant γ-secretase activities.

The 4-kDa amyloid ␤ protein (A␤) 1 that is invariably deposited as amyloid in Alzheimer's disease (AD) is a normally secreted proteolytic product of the amyloid ␤ protein precursor (APP) (1)(2)(3). Generation of A␤ from APP requires two proteolytic events, one at the amino terminus referred to as ␤-secretase and one at the carboxyl terminus known as ␥-secretase (Fig. 1). To date, neither of the proteases responsible for these activities has been definitively identified. Comparison of the soluble A␤ secreted by cells, soluble A␤ in cerebrospinal fluid, and insoluble A␤ isolated from the AD brain has revealed that there are numerous A␤ species with extensive amino-and carboxyl-terminal heterogeneity. The major A␤ species in both conditioned cell culture media and human cerebrospinal fluid is A␤1-40 (ϳ50 -70%) although some A␤1-42 (5-20%) is also present along with minor amounts of other peptides (e.g. A␤1-28, A␤1-33, A␤1-34, A␤3-34, A␤1-37, A␤1-38, and A␤1-39) (4 -6). The importance of the longer forms of A␤, in particular A␤42, has been heightened by the fact that all of the familial Alzheimer's disease (FAD) linked mutations that have been analyzed result in an increase in the concentration of A␤42 in a wide variety of model systems (reviewed in Refs. 7 and 8).
Biophysical studies have shown that the longer forms of A␤ aggregate at a much faster rate and at lower concentrations than forms ending at A␤40 suggesting that alterations in A␤42 concentration, as occurs in FAD linked forms of the disease, may account for the observation that the longer forms of A␤ are often the initial species deposited in the parenchyma of the AD and Down's syndrome brain (9 -11).
In addition to the ␤and ␥-secretase activities that generate A␤, a third proteolytic activity referred to as ␣-secretase cleaves within the A␤ sequence at Lys 16 to release a large secreted derivative (12)(13)(14), thus precluding formation of fulllength A␤. Following cleavage of APP at the extracellular/ lumenal domain of the APP by either ␣or ␤-secretases, ␥-secretase cleavage of the resulting COOH-terminal fragment results in the release of the p3 or A␤, respectively.
The findings that all of the FAD-linked mutations in APP and the presenilin (PS) genes alter the concentration of A␤ ending at position 42 and, with the exception of the APP K670N,M671L (NL) mutation, appear to act by altering ␥-secretase activity makes understanding this activity pivotal to our knowledge of the disease process. While previous studies have shown that ␤-secretase cleavage, like most proteolytic cleavages, exhibits fairly rigid primary amino acid sequence requirements (15), similar studies of the more complex ␥-secretase activity demonstrates fairly loose specificity (16,17). These studies, however, which either examine effects on total A␤ production (16) or effects of mutations at a single residue, A␤43 (17), provide no compelling mechanisms for the observed alterations in cleavage associated with FAD-linked mutations.
One of the more unusual aspects of ␥-secretase cleavage is that based on hydropathy plots the ␥-cleavage sites lie within the putative transmembrane domain (TMD) of the APP (18). Several other proteins, SREPB (19), Notch (20), and mitochondrial inner membrane proteins (21), have been postulated to undergo such intramembranous cleavage. While recent data for SREBP site two protease cleavage (22) and Notch (20), indicate that the cleavage of these proteins occurs at hydrophobic residues near the transmembrane junction, in no instance has the site of cleavage and the location of the residues with respect to the membrane at the time of cleavage been elucidated. Thus, the concept of intramembranous proteolysis remains controversial. To date, there is no definitive evidence showing that a protease can cleave bonds buried within a membrane.
As a first step toward defining the mechanism of ␥-secretase cleavage, we undertook a mutagenesis study in an attempt to define the specificity and structural requirements that produce A␤ species of varying lengths by using a large number of APP TMD mutants. The TMD mutant APPs are shown in Fig. 2. These mutations can be divided into several different categories. 1) Point mutations at positions A␤41 (I637X based on the APP695 sequence) and A␤43 (T639X) which correspond to the P1Ј positions for a ␥-secretase cleavage producing A␤1-40 and A␤1-42, respectively. 2) Deletion and insertion mutations designed to alter the localization of the ␥-secretase cleavage site within the membrane (del and ins). 3) Point mutations that alter the putative membrane stop anchor signal or increase the number of charged residues on the lumenal side of the membrane and 4) replacement of residues carboxyl to the normal ␥-cleavage sites with alanine. All mutations were made in a background of the APP695NL mutation in order to increase absolute amounts of A␤ peptides generated without affecting ␥-cleavage as this mutation increases activity at the ␤-secretase site without altering the relative amounts of either A␤40 or A␤42 (5,(23)(24)(25).

EXPERIMENTAL PROCEDURES
Generation of Mutant APPs-A two-step PCR based mutagenesis strategy was employed to generate the various TMD domain mutants. In the first step, the EcoRI/NotI fragment of pcDNA3APP695NL was replaced with a PCR product generated from wild-type APP695 using the forward primer APPϩ1803 and the reverse primers ⌬26 (5Ј-CAT-GCGGCCGCTCGTCTCTTGAACCCACATCTTCTGCA-3Ј), ⌬39 (5Ј-CA-TGCGGCCGCTCGTCTCCAACACCGCCCACCATGAGT-3Ј), and ⌬52 (5Ј-CATGCGGCCGCTCGTCTCTCAGCATCACCAAGGTGATGA-3Ј), respectively, to generate the base constructs pcDNA3APP695NL⌬26, pcDNA3APP695NL⌬39, and pcDNA3APP695NL⌬52. These mutant APPs incorporated a class IIa restriction site, BsmBI, 3Ј to A␤26, A␤39, and A␤52. When cut with BsmBI and NotI the BsmBI site is lost, leaving a 5Ј 4-bp overhang at the 3Ј end, effectively truncating the APP. To produce the various TMD mutants in this study, oligonucleotides incorporating the various mutations were generated containing a BsmBI site at their 5Ј end. PCR products were amplified using various mutant oligonucleotides and a common reverse primer using wild type APP as template. Subsequent cleavage of these products with BsmBI and NotI followed by cloning into the appropriate base vectors (pcDNA3APP695NL⌬26, pcDNA3APP695NL⌬39, and pcDNA3APP695NL⌬52) generated the desired mutants with no additional base changes. All PCR reactions were carried out using the High Fidelity PCR Kit from Boehringer Mannheim. All mutant cDNAs were sequenced to ensure that no additional mutations were incorporated. Sequences for all of the PCR primers used are available upon request.
Transient Transfection of 293 T Cells and Media Collection-Cells were plated onto 6-well culture dishes (Corning) and grown to 70 -80% confluence in Dulbecco's modified Eagle's medium (HyClone) supplemented with 10% fetal calf serum and 1% penicillin/streptomycin solution (Life Technologies, Inc.). On the day of transfection, culture media was changed to Opti-MEM (Life Technologies, Inc.). After a short period of equilibration, the media was again changed to fresh Opti-MEM (0.5 ml) and 1.5 g of DNA and 4.0 g of DOSPER liposomal transfection reagent (Boehringer Mannheim) were added to each well (premixed) in a volume of 0.5 ml of additional media. After overnight incubation, the media were discarded and replaced with Dulbecco's modified Eagle's medium, 10% fetal calf serum (1.0 ml/well). Twenty-four hours later, the media were collected and replaced with serum-free Dulbecco's modified Eagle's medium; complete protease inhibitor mixture (Boehringer Mannheim) was added to each sample. The serum-free media were collected 24 h later, and phenylmethylsulfonyl fluoride (to 1 mM) and EDTA (to 5 mM) added to each sample. In both cases, samples were immediately centrifuged at high speed to pellet cellular debris, transferred to clean Microfuge tubes, and frozen in aliquots at Ϫ80°C until analyzed.
Pepstatin treatment was conducted by adding to the media (posttransfection) pepstatin A to a concentration of 100 g/ml and Me 2 SO to 2%. Control cultures were treated with 2% Me 2 SO alone. Transfections and media collections were otherwise identical to untreated 293T cells.
sAPP Competitive ELISA-To generate and purify recombinant human sAPP, a 6-histidine tag was placed in-frame near the amino terminus of human APP695 by cleavage with KpnI and in-frame insertion of a double-stranded DNA formed by annealing two oligonucleotides. After construction in the expression vector pAG3, the APP695nterm his was stably transfected into CHO cells. For purification, after 48 h serum-free media was collected and purified using Ni 2ϩ affinity resin (Qiagen). To perform sAPP ELISAs, MaxiSorp 96-well immunoplates (Nunc) were coated overnight at 4°C with 400 ng/ml purified recombinant sAPP in 100 l of 0.1 M sodium carbonate buffer per well (pH 9.6). Plates were then blocked overnight at 4°C with 1% Block Ace, 0.05% NaN 3 in PBS (300 l/well). Blocking solution was discarded, the plates washed twice with PBS, and the wells loaded with 40 l of 1% bovine serum albumin in PBS (PBSB) and 10 l of each sample, in duplicate. Standard samples, containing from 31.5 to 8000 ng/ml sAPP in PBSB, were also loaded in duplicate. The NH 2 -terminal APP antibody 207 was added (100 l) to each well, diluted 1:32,000 in PBSB. Following overnight incubation at 4°C, wells were washed twice with PBS, 0.05% Tween 20, twice with PBS, and then loaded with 100 l of horseradish peroxidase-conjugated anti-goat IgG (1:5000 in PBSB; Rockland Immunochemicals). After a 4-h incubation at room temperature, plates were washed as above, then developed with 100 l of TMB solution (Kirkegaard & Perry). The reaction was stopped after approximately 5 min with the addition of 100 l of 6% phosphoric acid and the plates read at 450 nm. These conditions were determined empirically to optimize detection of sAPP in transfected cell lines in our laboratory, with typical expression levels placing sAPP concentrations in a linear region of the standard curve.
A␤ Sandwich ELISAs and Normalization of A␤ Measurements-For determination of A␤ concentrations we used 3 well characterized sandwich ELISA systems. Total A␤ was determined by 3160 capture and detection with either 4G8 or BNT77 (9,26), and A␤40 and 42 were determined by BAN50 capture and detection with either BA27 or BC05, respectively (5). To control for variance in transfection efficiency and the amount of APP that is appropriately inserted in the membrane and trafficked through the secretory pathway (thus APP available for ␥-secretase cleavage) A␤ levels were normalized to sAPP expression. This was accomplished by dividing the A␤ values (fmol/ml) by the sAPP values (ng/ml) resulting in a normalized A␤ value (fmol/ng). The normalized A␤ value for each mutant was then divided by the normalized A␤ value for APP695NL to give the %NL A␤.
FIG. 1. Cleavage of APP to generate A␤. The APP is cleaved at the amino terminus of the A␤ sequence by an unknown protease referred to as ␤-secretase. This cleavage releases a large secreted derivative referred to as sAPP␤. Subsequent cleavage of the APP COOH-terminal fragment (C99) by ␥-secretase generates A␤. In the ␣-secretory pathway (not shown) the APP is cleaved within the A␤ region to generate a large secreted derivative referred to as sAPP␣. Subsequent cleavage of the APP COOH-terminal fragment (C83) by ␥-secretase generates a truncated A␤ species referred to as P3.
Metabolic Labeling and Immunoprecipitation of sAPP and Fulllength APP-48 h after transfection, 293T cells were labeled for 2 h with 100 Ci/ml [ 35 S]methionine. Immunoprecipitation of full-length APP (flAPP) with Ab369W (27) from Triton X-100 cell lysates and sAPP in the 2-h conditioned media with antibody 207 were carried out as described previously (28,29). Each analysis was performed in duplicate. Immunoprecipitated proteins were separated on 10% Tris-Tricine gels. PhosphorImaging analysis of the dried gels was performed. After subtracting the number of pixels of flAPP or sAPP present in the mock transfected (vector alone) sample from each of the experimental samples, the ratio of sAPP/flAPP was calculated by dividing the pixels of sAPP by the pixels of flAPP. sAPP/flAPP ratios were then compared with the ratio of sAPP/flAPP calculated for APP695NL. TMD mutants showing sAPP/flAPP ratios greater than 50% of APP695NL ratio were assumed to have preserved processing in the secretory pathway.
Mass Spectrometric A␤ Analysis-Serum-free media (Dulbecco's modified Eagle's medium) collected for a 24-h period beginning 24 h post-transfection was used for mass spectrometric analysis. A␤ peptides were analyzed by immunoprecipitation/mass spectrometric A␤ assay as described previously (4). The A␤ peptides were immunoprecipitated from 1.0 ml of conditioned media using monoclonal anti-A␤ antibody, 4G8 (Senetek, Maryland Heights, MO), and protein G Plus/Protein A-agarose beads (Oncogene Science, Inc., Cambridge, MA) and analyzed using a matrix-assisted laser desorption/ionization time of flight mass spectrometer (Voyager-DE STR BioSpectrometry Workstation, PerSeptive Biosystem). Each mass spectrum was averaged from 500 measurements and calibrated using bovine insulin as the internal mass calibrant. For comparing the peptide levels secreted by cells expressing different mutations and between the treatments of Me 2 SO and pepstatin, synthetic A␤ (12-28) peptide (10 nM) was used as internal standard.

Transmembrane Domain Mutations: Effects on Total A␤ Lev-
els-APP appears to be a substrate for ␥-secretase only after it has been cleaved by ␣or ␤-secretase (30). Because cleavage of APP by these activities requires normal membrane insertion and trafficking through the secretory pathway, we have only performed in depth analyses of APP TMD mutants that showed preserved processing by these activities and normalized A␤ production to sAPP levels. Processing in the secretory pathway was assessed by metabolic labeling experiments and determination of the sAPP/flAPP ratio in comparison to APP695NL as described under "Experimental Procedures" (data not shown). To examine A␤ production in the mutants that showed preserved processing in the secretory pathway, sAPP, total A␤, A␤1-40, and A␤1-42 were measured in the conditioned media by ELISA. Comparison of normalized total A␤ shows that most mutants resulted in modest to moderate decrements in A␤ production (Table I). Even strikingly non-conservative mutations at residues 637 (A␤41) or 639 (A␤43) had relatively modest effects on total A␤ production. Basic (Lys) or acidic (Glu) amino acid substitutions at both sites, and substitutions of amino acids with a large hydrophobic side chain (Phe and Trp) or even a proline at position 637, did not prevent ␥-secretase cleavage. In addition, production of A␤ was only modestly decreased by most mutations at or near the lumenal transmembrane domain junction or by deletion or insertion mutants. The notable exceptions being the large decrements in A␤ production by del640 -43, 624 -626E, and ins625-631 mutants.
Six mutants (I637R, T639P, 649 -651E, 649 -651D, del625-631, and del640 -647) showed dramatic decreases in A␤ levels (Ͻ5% of APP695NL A␤, data not shown). Pulse-chase analyses showed that these mutations produced equivalent levels of full-length APP, but that the sAPP/flAPP ratio was markedly diminished (Ͻ1% compared with APP695NL, data not shown). Although the precise mechanism for the impaired processing of these mutations is not definitively established by these studies, based on previous studies (16) and our metabolic labeling data it is likely that these mutants either alter membrane insertion or prevent proper trafficking through the secretory pathway. In agreement with this, one of these mutations (649 -651E) has previously been shown to result in a non-transmembrane, membrane-associated cell surface, full-length APP (31).
Effects on Cleavage Site-Comparison of the relative amounts of A␤1-40 to A␤1-42, measured, respectively, by BAN50/BA27 or BAN50/BC05 ELISA, shows that many of these mutants dramatically alter the relative amounts of the major A␤ species produced. These data are expressed as the percent of total A␤ that A␤1-40 and A␤1-42 species represent (Table I). For the I637X mutants, detection of A␤1-42 (43) species could be impaired as the 42(43) end specific BC05 detection antibody recognizes an epitope partially determined by the residue at A␤41; thus, altered A␤42 peptides could be produced but not detected (5). To illustrate how these data indicate shifts in cleavage it is useful to look at the effects on %A␤1-40 and %A␤1-42 in the ins625-628 mutant. Comparison of the %A␤1-40 and %A␤1-42 illustrates that the major shift in cleavage is a reduction in processing at the A␤40 cleavage site (2% of total versus 51% of total in the APP695NL construct) while A␤42 cleavage is preserved (4.6% of total versus 3.3% of total in the APP695NL construct). These analyses also show that only 7% of total A␤ is accounted for by A␤1-40 and 1-42 versus 54% in the APP695NL construct, indicating that the vast majority of A␤ peptides generated from this mutant are cleaved at a different site or sites than the ones normally utilized. By these criteria, almost all of the mutants show shifts in ␥-secretase cleavage. In the last column of Table  I, the statistically significant shifts in cleavage for each of these mutations are summarized.
Mass Spectral Analysis-In order to determine exactly how cleavage is shifted by the TMD mutations, A␤ produced from 293T cells expressing APP695NL or select TMD mutant APPs was analyzed by immunoprecipitation/mass spectrometric analysis (4). The monoclonal antibody, 4G8, was used to immunoprecipitate A␤ from serum-free conditioned media. This antibody recognizes A␤17-24 (32), a region of the A␤ not altered by the mutants used in this study, making it highly unlikely that this analysis would be biased by selective immunoprecipitation of different A␤ species. The molecular masses of various A␤ peptides were measured in these analyses by using internal mass calibrants, bovine insulin and A␤12-28 peptide. These masses were then used to identify A␤ peptides produced by the TMD mutant APPs and infer the ␥-secretase cleavage sites as illustrated in Fig. 3. The relative peak intensity was used to determine the relative abundance of A␤ peptides within each spectra resulting from each TMD mutant APP. For clarity, all A␤ peptides are numbered according to the cleavage sites in wild type A␤ species; thus A␤1-43 in del625-628 is a 39-amino acid A␤ peptide derivative. Representative mass spectra for APP695NL, I637F, I637P, and ins644 -647 are shown in Fig. 3A.
The mass spectrum of A␤ produced from APP695NL shows that A␤1-40 was the major A␤ species and that the minor A␤ species were A␤1-42, A␤1-39, A␤1-38, and A␤1-37. Based on comparison to APP695NL, several of the different TMD mutations analyzed dramatically shift the ␥-secretase cleavage site. Mass spectrometric A␤ analyses of APP695NL and TMD mutants are schematically summarized in Fig. 3B. The largest shifts in cleavage site utilization are seen in the ins625-628 and del625-628 mutants, while less dramatic effects are seen with the del644 -647 and ins644 -647 mutations. G625K also increases long A␤ production, while G625P increases production of shorter A␤ peptides. Introduction of Lys at position 637 or 639 has only minor effects on cleavage, while substitution of Ala at 639 increases both long A␤ (A␤1-42) and short A␤ (A␤1-38) production. Pro or Phe substitutions at Ile 637 dramatically shift cleavage away from A␤40. Finally, A␤ production is only minimally affected by the 640 -648A mutant. Notably, the mass spectral data and the ELISA data are remarkably consistent with the exception being that small amounts of A␤1-42 detected by ELISA are not always detected by mass spectrometric analysis (for example, in 640 -648A) due to the slightly lower detection efficiency of A␤1-42 (4).
Pepstatin Treatment of Selected Mutants-Given the loose sequence specificity exhibited by ␥-secretase, we postulated that the ␥-secretase activity responsible for generating the various A␤ species was unlikely to be due to a single protease. Therefore, we treated cells transfected with several of the TMD mutants that shift cleavage with pepstatin, an inhibitor of normal ␥-secretase activity. Although it is not known whether pepstatin directly inhibits ␥-secretase, pepstatin treatment of cells transfected with APP COOH-terminal fragments LC99, APP695wt, or APP695NL results both in accumulation of COOH-terminal fragments and a decrease in A␤ production 2 without significantly altering sAPP secretion in either APP695wt or APP695NL (Table II), consistent with it being an inhibitor of ␥-secretase activity. Because cells are rather impermeable to pepstatin, it is necessary to treat them with high 2 C. Eckman, unpublished results. shown. Spectra are normalized to the most abundant A␤ peptide species and peaks in the spectra are labeled with A␤ peptide sequence numbers based on the wild type A␤ sequence. Background peaks are labeled with an asterisk (*). B, mass spectrometric analysis of A␤ peptide secreted by the TMD mutants are schematically depicted. A␤ peptides detected in each mutant are indicated by arrows above the cleavage sites, while the height of the arrows indicates relative peak intensity. The large arrows indicate peaks with relative intensity greater than 60%, the medium arrows represent peaks with relative intensity between 20 and 60%, and the small arrows stand for peaks with relative intensity less than 20%. These results are averaged from multiple measurements.
concentrations (100 g/ml) in the presence of 2% Me 2 SO in order to obtain effective intracellular concentrations (33). Again sAPP, total A␤, and A␤1-40 and 1-42 were measured by ELISA from conditioned media. Treatment of cells with vehicle alone and vehicle with pepstatin had no overt toxic effect in the time course studied, and although changes in sAPP levels were observed in individual experiments in some mutants, none of these changes reached statistical significance (Table II). Pepstatin treatment of cells transfected with the APP695NL mutant decreased total A␤ and A␤1-40 equally while A␤1-42 was less affected. The effects on three of the TMD domain mutants: del625-628, G625P, and I637F were similar to the effect on APP695NL. In contrast, pepstatin treatment did not alter A␤ production from the I637P mutant to any significant extent. The effects of pepstatin on T639K, a mutant that had only subtle effects on cleavage site utilization, were also distinct from APP695NL. A␤1-40 was only modestly inhibited and A␤1-42 was not inhibited. Although A␤1-40 produced from ins625-628 represents only a minor fraction of the total A␤ produced, pepstatin actually increased the amount of A␤1-40 produced from 15 to 50 fmol/ml.
To definitively identify how pepstatin altered cleavage of these TMD domain mutants, A␤ secreted from treated and untreated cells was analyzed by mass spectrometry (Fig. 4). For APP695NL, G625P, I637F, and T639K, the relative peak heights of each of the major A␤ species was similar before and after pepstatin treatment, indicating that cleavage was inhibited equally at each site (Fig. 4, data shown only for APP695NL). Consistent with the ELISA analysis, the relative peak height of A␤1-42 was not reduced as much by pepstatin treatment in APP695NL and G625P. The effect of pepstatin treatment on del625-628 also revealed little difference in relative peak heights of the major A␤ species except that peaks corresponding to A␤1-45 and A␤1-46 were increased suggesting enhanced utilization of minor cleavage sites (Fig. 4, G and  H). In the ins625-628 and I637P TMD mutants, there were some marked shifts in the relative peak intensities of some A␤ species after pepstatin treatment. In ins625-628 the A␤1-33 peak was decreased more than the A␤1-37 peak (Fig. 4, E and F), while in I637P the relative peak intensity of A␤1-37 decreased and the relative peak intensity of A␤1-43 increased (Fig. 4, C and D). Thus, there appears to be differential sensitivity of certain ␥-secretase cleavages to pepstatin and increased utilization of alternative sites by pepstatin-insensitive proteases in certain mutants when pepstatin-sensitive cleavage is inhibited. DISCUSSION

APP Transmembrane Domain Mutants Shift ␥-Secretase
Cleavage-An unanticipated result of this analysis is that many TMD mutations alter the major sites of ␥-secretase cleavage. The most striking finding resulting from the analyses of the mutations, ins625-628 and del625-628, is that the major ␥-secretase cleavage site appears to be determined by the length of the TMD lumenal to the normal ␥-secretase sites. In wild type APP, the major ␥-secretase site carboxyl to A␤40 lies 12 amino acids from the lysine at A␤28 (KA␤28) that is predicted to delineate the lumenal TMD boundary. In these mutations, the normal ␥-site is shifted either 4 amino acids (del625-628) closer or farther away (ins625-628) relative to KA␤28. Although the primary sequence surrounding the normal ␥-site is unaltered, the major cleavage shifts to the 13 th amino acid (A␤37) from KA␤28 in ins625-628 or the 11 th amino acid (A␤43) from KA␤28 in del625-628. Only a minor portion of A␤ produced from either mutant is cleaved at the V-I bond at A␤40 (14% in del625-628 and 2% in ins625-628). Mutations altering charge at the presumptive lumenal border of the transmembrane domain (625-626K, 624 -626E, and 624 -626D) had a similar effect as the del625-628 mutant increasing production of longer A␤ peptides. These mutants might be predicted to decrease the length of the TMD proximal to the ␥-secretase site by several amino acids resulting in increased cleavage at sites distal to A␤40.
The insertion and deletion mutants designed to alter the length of the TMD distal to the normal ␥-cleavage sites had much subtler effects. The major effect of the ins644 -647 mutation was to decrease cleavage at A␤40 and increase cleavage at A␤38. The major effect of the del644 -647 mutant was to modestly decrease A␤1-42 production. This is similar to the effect seen in the I637K and T639K mutants, where A␤1-42 production is decreased. In all three cases, increased positive charge has been placed closer to the normal A␤42 cleavage site. Replacement of residues 640 -648 with alanine (640 -648A) had only a minor effect on A␤ production shifting cleavage away from the A␤40 site (28% A␤1-40) indicating that this region is not as crucial in determining the membrane positioning of the ␥-secretase sites in comparison to the region amino to the ␥-secretase cleavage site. This finding is similar to a previous report showing that substitution of APP residues 635-642 or 636 -653 with the corresponding residues of the epidermal growth factor receptor TMD did not significantly impair A␤ production (16). Nevertheless, alterations in this region do result in subtle, but important, shifts of ␥-cleavage. All FADlinked mutations in this region, V641I, V641F, V641G, and I640V (based on the APP695 sequence), replace hydrophobic residues in the transmembrane domain with other hydrophobic residues and each results in increases in A␤42 (5,17,34;reviewed in Refs. 7 and 8). In this study, a similar substitution, T639A, increased A␤42 cleavage. This result is similar to a recent report demonstrating that hydrophobic substitutions at A␤43 increased A␤42 production relative to A␤40 (17). However, that study only looked at ratios of A␤42:A␤40; thus, alternative cleavages would have been missed, and it is unclear from the data presented whether the increase in the ratio is due to a decrease in A␤40 production or an increase in A␤42 production. Of the mutations on the carboxyl side of the ␥-site, only del640 -643 had dramatic effects on A␤, significantly decreasing total A␤ production and markedly increasing %A␤1-42 production. This mutant's more dramatic effect compared with deletions or insertions downstream could be explained by the fact that this region may represent the end of the APP TMD (16). Taken together, these data indicate that mutations in the lumenal portion of the TMD have effects on cleavage that are much more profound than mutations which alter charge or TMD length on the cytoplasmic side of the normal ␥-secretase site.
␥-Secretase Is Not a Single Proteolytic Activity-Our data shed some light on the proteolytic mechanism responsible for generation of A␤ peptides of various lengths. The finding that the del625-628 preferentially secretes an A␤ species of 39 amino acids ending at A␤43 is evidence against a mechanism in which a single endoprotease cleavage occurs carboxyl to the major ␥-sites followed by trimming by carboxypeptidase. However, for any given mutation, we cannot rule out the possibility that shorter A␤ species are generated through carboxypeptidase degradation of longer A␤ peptides. Furthermore, the differential effect of pepstatin on select TMD mutants is most consistent with the action of at least two and possibly more proteolytic activities. This differential inhibition is most clearly demonstrated in the I637P mutant. This mutant is readily processed into A␤ but is almost completely pepstatin insensitive. Mass spectral analysis indicates that the ␥-secretase that generates A␤1-43 in the I637P mutant is pepstatin-resistant, while a different ␥-secretase that cleaves this mutant at A␤1-37 is pepstatin-sensitive. When cleavage is inhibited at A␤37 by pepstatin a corresponding increase is seen in the cleavage at A␤43, resulting in almost identical amounts of total A␤ secreted. Similarly, other TMD domain mutants as well as APP695NL exhibit both a pepstatin-sensitive and a pepstatinresistant ␥-secretase activity.
Previous reports based on differential inhibition of A␤40 and A␤42 cleavage have indicated that these A␤ species may be generated by distinct proteases. Unfortunately, interpretation of these previous studies has been difficult. The compound, MDL 21760, used in several studies could have altered trafficking of APP as it markedly increased sAPP (35,36). In other studies, reporting an A␤1-40 inhibitory effect of calpain 1 inhibitor, no controls for either APP synthesis or sAPP production were reported (37,38). In fact, only one potential ␥-secretase inhibitor, a substrate-based difluoroketone compound, altered ␥-secretase activity without apparent alteration of other secretase activity (39). In this study, pepstatin had no toxic effects and no significant effect on sAPP production from APP695NL; yet, it was a less effective inhibitor of A␤1-42 (ϳ35% of control) production than A␤1-40 (ϳ20% control).
The differential inhibition observed at A␤40 and A␤42 sites could be due to a number of mechanisms. Since A␤1-42 has previously been shown to be specifically generated in the endoplasmic reticulum (although endoplasmic reticulum-derived A␤1-42 is not secreted), it is possible that this difference simply reflects a varying degree of organelle penetrance (a similar argument could be made for any of the other published inhibitors) (40 -42). Alternatively, this differential inhibition could be viewed as additional evidence that multiple proteases generate both A␤40 and A␤42, but that pepstatin-sensitive cleavage is responsible for a higher percentage of activity at the A␤40 site. The finding that differential inhibition of A␤1-40 and A␤1-42 was constant among several of the TMD mutants which alter cleavage site preference (del625-628, G625P, I637F), but not the T639K mutant supports this notion of multiple proteases as it is most consistent with enhancing the A␤40 and A␤42 cleavages by a pepstatin-insensitive protease. Based on our results, we would predict that for wild type For each pair of spectra, Me 2 SO versus pepstatin, the peak heights of A␤ peptides are first normalized to the internal standard, A␤(12-28) (labeled as 12-28 (Std)). The spectra are then plotted using the relative peak intensities; e.g. the highest peak within the pair of spectra is plotted as 100%. In this way, the relative intensity of peaks between the two spectra can be compared. Peaks in the spectra are labeled with A␤ peptide sequence numbers based on the wild type A␤ sequence. Thus, A␤  in ins625-628 is a 41-amino acid peptide whereas A␤  in del625-628 is a 39-amino acid peptide, respectively. Background peaks and unidentified peaks are labeled with asterisk (*) and question marks (?), respectively. substrate a pepstatin-sensitive protease cleaves both A␤40 and A␤42, with preferential specificity for cleavage at A␤40 and that additional pepstatin-insensitive proteases account for ϳ20% of the cleavage events at A␤40 and ϳ35% of the cleavage events at A␤42. Thus, when the major A␤40 cleavage site is "protected" (see below) from cleavage in the del625-628 mutant (14% A␤1-40 versus 51% in APP695NL, Table II) and shifted to A␤1-42(43) (48% versus 3.3% in APP695NL, Table  II), a remarkably similar degree of inhibition is seen on both species following pepstatin treatment. A final mechanism in which a single pepstatin-sensitive protease cleaves both sites but is differentially inhibited could also be considered. This mechanism, however, is not consistent with known models of proteolytic inhibition by pepstatin and would be an unprecedented mechanism of action for a protease inhibitor that acts in a competitive fashion.
Implications for Models of ␥-Secretase Cleavage: the Length of the Lumenal Portion of the TMD Is the Prime Determinant of Cleavage-Until the ␥-secretases are definitively identified and the location of the residues at the time of cleavage defined, the issue of intramembranous proteolysis is likely to remain controversial. Unless ␥-secretases are completely novel proteases that can cleave proteins in the hydrophobic environment of the membrane, mechanisms must exist that permit membraneembedded regions of polypeptides to transfer from the lipid bilayer into a hydrophilic proteinaceous environment that supports proteolysis. This could occur in one of two fashions. First, as has been generally proposed, ␥-secretase activity may be due to the action of proteases or proteolytic complexes that are capable of creating a pore or hydrophilic pocket within the membrane. Based on our data, in this model the interaction of the protease with the TMD domain of the APP lumenal to the ␥-secretase site would appear to be a critical factor in determining cleavage (see Fig. 5A). A second possibility is that these cleavages require translocation of the cleavage sites out of the membrane (Fig. 5B). One of the similarities between the ␥-secretase-like cleavages in APP, Notch and SREBP, is that the potentially intramembranous cleavage occurs only after an initial cleavage or cleavages that remove significant portions of the protein amino-terminal to the intramembranous cleavage site. Thus, it is possible that the sites are normally protected by the membrane prior to the primary lumenal cleavage of the holoprotein. Once the lumenal domain is cleaved, the protein either assumes a different conformation or is actively translocated resulting in exposure of the intramembranous site to the cytoplasm where it can be cleaved. Such a cut-expose-cut model for the APP is illustrated in Fig. 5B. In this model one could envision slippage or translocation of the intramembranous site in either direction; for APP we propose that the ␥-site is exposed to proteases that are associated directly or indirectly with the membrane and have active sites facing the cytoplasm. This model, while subject to some potential thermodynamic penalties associated with translocation of the APP TMD, is consistent with our data showing that length of the TMD lumenal to the ␥-cleavage site has more profound effects on ␥-secretase cleavage than mutations carboxyl to the normal ␥-sites, offers a very simple explanation as to why these sites are resistant to proteolysis in the intact holoprotein, is consistent with another model of intramembranous cleavage (21), and does not implicate an unprecedented type of proteases in this cleavage.
In either model proposed above, it is apparent that the prime determinant of ␥-secretase cleavage is the length of the transmembrane domain proximal to the ␥-secretase cleavage site. In an intramembranous cleavage model, we would predict that the position of the active site of the protease is relatively fixed within the membrane and that the protease recognizes determinants in the APP TMD lumenal to the normal cleavage site. Thus, varying the length of the lumenal TMD alters the residues that contact the active site. In contrast, the membrane serves as the delineating factor in the cut-expose-cut model, residues buried within the membrane are protected from cleavage while exposed residues are cleaved.
Because at least two, and possibly more, proteolytic activities appear to contribute to ␥-secretase cleavage, defining the substrate specificity is problematic until the secretases are definitively identified. Nevertheless, it is almost certain that any individual ␥-secretase would exhibit at least some sequence specificity as even nonspecific proteases such as proteinase K preferentially cleave certain substrates (43,44). Consistent with this, altering sequence at the normal ␥-cleavage site has dramatic effects on recognition of the normal cleavage site by the major pepstatin-sensitive ␥-secretase activity which results in either increased cleavage by a pepstatin-sensitive protease at sites other than A␤40 and A␤42, increased cleavage by a pepstatin-insensitive protease at normally utilized sites, increased cleavage by a pepstatin insensitive protease at A␤40 and A␤42, or a combination of these alterations.
Multiple Cellular Factors Could Influence ␥-Secretase Cleavage-If ␥-secretase cleavage is primarily dependent upon the location of the ␥-secretase cleavage site with respect to the membrane or active site of a protease within the membrane, then it is likely that a number of cellular factors could influence this cleavage including membrane thickness or composition and interaction with other proteins. PSs, which are important regulators of ␥-secretase activity (reviewed in Refs. 7 and 8), could influence ␥-secretase cleavage of APP by altering trafficking of APP to different cellular microdomains where cleavage would be influenced by membrane thickness, by directly interacting with APP carboxyl-terminal fragments and positioning them within the membrane, by translocating the ␥-site out of the membrane, or by altering the position of ␥-secretase with respect to the APP transmembrane domain. Alternatively, the possibility that PSs, which do not resemble any known protease, could in fact be ␥-secretase has not been excluded, in which case altered interaction between PS and APP could provide a simple explanation for shifts in cleavage induced by FAD-linked PS mutants and would be entirely consistent with decreased production of A␤ from PS 1 knockout mice (45).
Conclusions-␥-Secretase appears to represent a membrane protein secretase defined as a proteolytic activity that cleaves a proteolytically sensitive region of a transmembrane protein resulting in secretion of the proximal portion of the cleaved protein (46). Compared with other secretase activities, ␥-secretase activity appears to be distinct in that it does not cleave its substrate within an extracellular/lumenal "stalk" like region proximal to the membrane, but rather within a hydrophobic region purportedly within the TMD of the APP COOH-terminal fragment. The demonstration in this study that the ␥-secretase is not a single proteolytic activity and that the membrane plays a critical role in determining ␥-secretase cleavage of APP will be important in developing strategies for isolation of the various ␥-secretase activities, which remain a major therapeutic target in AD. Additional studies will be needed to determine whether the pathological shifts in A␤ cleavage are caused by the alterations in the major pepstatin-sensitive ␥-secretase activities or by additional proteases, which might play a role in pathogenic processing. Given that the common effect of all early onset FAD-linked mutations is to modestly increase long A␤ production, this study offers important insight into how various A␤ peptides are generated and suggests possible mechanisms whereby various FAD-linked mutations might shift ␥-secretase cleavage. Because other biologically important cleavage events in SREBP and Notch may be intramembranous, it will be important to perform similar studies on such proteins to determine if similar, atypical, proteolytic mechanisms are responsible for such cleavages. Together, these studies should also offer additional insights into the important question of how transmembrane domains, in general, are degraded. In this model, the APP COOH-terminal fragment generated after ␤-secretase cleavage associates with a protease whose active site lies within the membrane. Based on our data, the interaction of the intramembranous protease with the TMD domain of the APP lumenal to the ␥-secretase site would appear to be a critical factor in determining cleavage. Thus, del625-628 and ins625-628 mutants would alter the residues in contact with the active site resulting in different cleavages. B, a cut-expose-cut model. In this model, prior to cleavage by ␣or ␤-secretase, the ␥-site is buried within the membrane, inaccessible to proteolysis. After cleavage to release sAPP, the APP COOH-terminal fragment location within the membrane is altered to expose the ␥-site to the cytoplasm where it could be cleaved by either cytoplasmic proteases or membrane-associated proteases with active sites facing the cytoplasm. This translocation could occur because this is the preferential conformation of the APP COOH-terminal fragment within the membrane or it could be facilitated by interaction with PS, ␥-secretase itself, or other transmembrane proteins. Such a model provides a simple explanation for the cleavages observed in the del625-628 and ins625-628 mutant APPs. Based on this model, thickness of the membrane could influence cleavage site or alternatively interaction with other proteins in the membrane, such as PS, could alter the exposure of the ␥-site. Because the transmembrane region of a protein is not static within the membrane, it is likely that in either model there is some resonance of the COOH-terminal fragment with respect to the membrane. Such resonance could result in either different residues presented to the active site of an intramembranous protease or exposure of different residues to the cytoplasm, which could account for the invariable production of "ragged" ends. Scissors indicate proteolytic events. ␤ and ␥ indicate ␤and ␥-secretase cleavage, respectively.