γ-Secretase Cleavage Site Specificity Differs for Intracellular and Secretory Amyloid β

The final step in Aβ generation is the cleavage of the C-terminal 99 amino acid residues of the amyloid precursor protein by γ-secretase. γ-Secretase activity is closely linked to the multi-transmembrane-spanning proteins presenilin 1 and presenilin 2. To elucidate whether the cleavage site specificities of γ-secretase leading to the formation of secreted and intracellular Aβ are identical, we made use of point mutations close to the γ-cleavage site, known to have a dramatic effect on the 42/40 ratio of secreted Aβ. We found that the selected point mutations only marginally influenced the 42/40 ratio of intracellular Aβ, suggesting differences in the γ-secretase cleavage site specificity for the generation of secreted and intracellular Aβ. The analysis of the subcellular compartments involved in the generation of intracellular Aβ revealed that Aβ is not generated in the early secretory pathway in the human SH-SY5Y neuroblastoma cell line. In this study we identified late Golgi compartments to be involved in the generation of intracellular Aβ. Moreover, we demonstrate that the presence of processed PS1 is not sufficient to obtain γ-secretase processing of the truncated amyloid precursor protein construct C99, proposing the existence of an additional factor downstream of the endoplasmic reticulum and early Golgi required for the formation of an active γ-secretase complex.

BACE1 (␤-site APP cleaving enzyme) generates the N terminus of A␤, releasing the ectodomain of APP (for review, see Ref. 4). The remaining C-terminal membrane-bound fragment C99 is further cleaved by ␥-secretase, yielding two major species of A␤ peptides, A␤40 and A␤42 (5)(6)(7). Alternatively, APP can be cleaved within its ectodomain by ␣-secretases, identified as members of the ADAM family of disintegrin metalloproteases, leading to ␣-secreted APP and a truncated non-amyloidogenic peptide (p3) (for review, see Ref. 8). The identity of ␥-secretase has been the subject of intensive investigation; however, the exact nature of ␥-secretase has not been definitively established. Recent studies implied that ␥-secretase processing requires the presence of presenilin 1 (PS1) and PS2 (for review, see Ref. 4). The PS holoproteins undergo highly regulated endoproteolytic processing to yield N-and C-terminal fragments and are thought to mediate ␥-secretase enzyme activity as part of a multimeric high molecular weight complex (9,10). It is still not proven, however, whether PS itself is ␥-secretase or an essential co-factor required for ␥-secretase enzyme activity, since other proteins of the high molecular weight complex have been shown to modulate ␥-secretase enzyme activity (11)(12)(13). Furthermore, a discrepancy exists between the predominant subcellular localization of PS1 in the endoplasmic reticulum (ER) and cis-Golgi compartment (14 -16) and PS1-dependent ␥-secretase processing at or close to the plasma membrane where little PS1 seems to be present (17). A␤ generated at or close to the plasma membrane is rapidly secreted by cultured cells as well as in biological fluids and aggregates into the characteristic extracellular protein deposits, which are thought to be the cause of AD (for review, see Ref. 18). Additionally, intracellular A␤ is generated and is discussed as important in the pathogenesis of AD. Recent studies demonstrate an intraneuronal accumulation of A␤42, the predominant species of A␤ in senile plaques (19), in AD-vulnerable brain regions (20 -22) as well as in transgenic mice expressing mutant proteins that lead to a familial form of Alzheimer's disease (23). Moreover, transgenic mice show accelerated neurodegeneration without extracellular amyloid deposition (24), indicating that intracellular amyloid-␤ peptides may play a crucial role in the development of AD. Intracellular sites with ␥-secretase activity were identified in primary neurons, neuronal cell lines, and peripheral cells (25)(26)(27)(28)(29)(30). Evidence has been obtained for the generation of A␤42 within the ER (27)(28)(29)31), where PS is predominantly localized (14 -16). However, it has also been reported that A␤42 is produced in different organelles later in the secretory pathway and that the ER is not the major intracellular site of A␤42 generation (26,30,(32)(33)(34)74). Additionally, recent work has suggested that the A␤42 generation in the ER may be independent of PS (35). Therefore, the exact site of intracellular A␤ generation and the subcellular compartments in which PS promotes A␤ generation remain so far unclear. Furthermore, little is known about the cleavage site specificity of the ␥-secretase responsible for the generation of intracellular A␤, which may play a central part in the pathogenesis of AD.
In the present study we examined the cleavage site specificity of intracellular ␥-secretase as well as the subcellular compartments in which the direct A␤ precursor C99 (36) is processed by ␥-secretase to intracellular A␤. To address the cleavage site specificity of ␥-secretase we made use of point mutations close to the ␥-cleavage site known to have a dramatic effect on the 42/40 ratio of secreted A␤ (37,38) and determined their effect on the 42/40 ratio of intracellular A␤. We found that these mutations only marginally influenced the 42/40 ratio of intracellular A␤, indicating that the cleavage site specificity of intracellular ␥-secretase is less affected by point mutations close to the ␥-cleavage site. To investigate the compartments involved in ␥-secretase processing, we used different experimental strategies including treatments with organelle-specific toxins and expression of C99 proteins directed to different cellular compartments as well as subcellular fractionation. Our results show that generation of intracellular A␤ was eliminated when C99 was retained in the ER and early Golgi despite the presence of processed endogenous PS1, indicating that PS1 is not sufficient to mediate ␥-secretase processing in the early secretory pathway. Instead, we obtained several lines of evidence that an active ␥-secretase complex is located in late Golgi compartments.
Cell Culture and Transfections-Human SH-SY5Y neuroblastoma cell line was maintained in Dulbecco's modified Eagle's medium (high glucose) (Sigma) containing 10% fetal calf serum (PAA Laboratories) and 1% nonessential amino acid solution (Sigma). 80% confluent cells were transfected with the expression vector pCEP4 (Invitrogen) alone or the pCEP4 vector carrying the SPDAC99 inserts using Lipofectin (Invitrogen) as described by the producer. Stable transfectants were selected using 300 g/ml hygromycin (PAA Laboratories). For each construct, at least two independent cell lines were established.
Preparation of Cell Lysates and Collection of Conditioned Media-Fresh culture medium (5 ml) was added to a confluent monolayer of cells in a 10-cm culture dish. Conditioned media were collected after 14 -16 h. The conditioned media were centrifuged at 4°C for 1 min at 13,000 rpm, and the supernatants were used for immunoprecipitation of soluble secreted A␤. In parallel to the conditioned media, cell lysates were prepared. Cells were harvested and lysed in lysis buffer (50 mM Tris/HCl, pH 7.5, 150 mM NaCl, 1% Nonidet P-40, 1% Triton X-100, and 2 mM EDTA) supplemented with protease inhibitor mixture (Roche Molecular Biochemicals). The 13,000-rpm supernatants were used for immunoprecipitation of detergent-soluble intracellular A␤.
Treatments with Organelle-specific Toxins-Stably transfected SH-SY5Y cells were preincubated with brefeldin A (BFA) (10 g/ml, Sigma) or monensin (5 M, Sigma) for 1.5 h followed by a 10-h incubation period in the presence of the drugs in fresh medium.
Metabolic Labeling and Preparation of Cellular Membranes-After 1.5 h of preincubation in methionine-free minimum essential Eagle's medium (Sigma) in the presence or absence of BFA (10 g/ml), C99 WT-transfected SH-SY5Y cells were incubated for 6 h in methioninefree nonessential amino acid solution containing 5% fetal calf serum and 133 Ci/ml [ 35 S]methionine (Amersham Biosciences) in the presence or absence of BFA (10 g/ml). Cellular membranes were prepared as described by Mercken et al. (46) and subjected to immunoprecipitation as described above.
Subcellular Fractionation and Sucrose Density Centrifugation-Subcellular fractionation and sucrose density equilibrium centrifugation of SH-SY5Y cells stably expressing SPC99-WT were performed essentially as described (43) with minor modifications. Postnuclear membranes were applied onto a continuous sucrose gradient (0.2-2 M) and 17 1-ml fractions were collected from the bottom of the gradient after centrifugation overnight at 100,000 ϫ g av (27,000 rpm, Beckman SW 28.1 rotor). Individual fractions were diluted into 5 mM HEPES, pH 7.3, 0,15 M NaCl, and membranes were sedimented by centrifugation for 1 h at 80,000 rpm in a MLA-80 rotor (Beckman). The final membrane pellets were re-suspended in 200 l of phosphate-buffered saline, 5% glycerol and stored at Ϫ80°C until further use.
Electrochemiluminescence Assay-Equal aliquots of each fraction were adjusted 2% CHAPSO and 0.5% CHAPSO for detection of A␤40 and A␤42 peptides, respectively. Membrane-associated A␤ peptides were released by incubation for 20 min at 37°C and measured by an electrochemiluminescence assay as described (10,47) in a 96-well plate format. A␤ peptides were captured with biotinylated monoclonal antibody 4G8 (Senetek) followed by detection of A␤40 with ruthenylated G2-10, whereas ruthenylated G2-11 was used for detection of A␤42. Nonspecific background signal was determined by using non-biotinylated capture antibody 4G8 in the presence of ruthenylated detection antibodies G2-10/G2-11 and subtracted.

Intracellular and Secreted A␤42/40 Ratios Are Differently Influenced by Point Mutations Close to the ␥-Cleavage
Site-Point mutations close to the ␥-cleavage site of APP as well as pathogenic mutations in the genes encoding PS1 and PS2 have been shown to alter the product ratio of A␤42 and A␤40 (A␤42/ 40) (37-39, 48 -50). However, in these studies, the authors analyzed the 42/40 ratios of secreted A␤, which is generated via a pH-sensitive and endocytosis-dependent pathway (17,(51)(52)(53). Far less is known about the cleavage site specificity of the ␥-secretase responsible for the generation of intracellular A␤ (A␤i). To investigate the cleavage site specificity of intracellular ␥-secretase, we selected two point mutations close to the ␥-cleavage site (Fig. 1), known to have a strong effect on the 42/40 ratio of secreted A␤ (A␤sec42/40). These are the point mutations I45F and V50F, which have been identified by phenylalanine-scanning mutagenesis of the transmembrane domain of APP (38).
Consistent with previously published data for non-neuronal cells (38), both mutants (C99 I45F and C99 V50F) had an opposite effect on the generation of the A␤ species secreted into the media of stably transfected human SH-SY5Y neuroblastoma cells. C99 I45F is mainly processed to secreted A␤ ending residue 42 (A␤sec42) ( Fig. 2A), resulting in a dramatic increase of the A␤sec42/40 ratio compared with C99 WT (relative ratio, 20.4 Ϯ 3.6 (I45F), p Ͻ 0.001, n ϭ 6) ( Fig. 2C) (Table I). In contrast, the V50F mutation is mainly cleaved after amino acid residue 40 ( Fig. 2A), causing a decrease of the A␤sec42/40 ratio compared with C99 WT (relative ratio, 0.3 Ϯ 0.1 (V50F), p Ͻ 0.001, n ϭ 6) ( Fig. 2C) (Table I). As a negative control we used SH-SY5Y cells stably transfected with the expression vector alone measuring the amount of endogenous secreted and intracellular A␤ (data not shown).
In contrast to the dramatic effect of both point mutations on the 42/40 ratio of secreted A␤, both mutations only marginally influenced the 42/40 ratio of intracellular A␤ (A␤i42/40) (Fig.  2B). The C99 WT protein produced roughly 1.5-fold more A␤i40 than A␤i42 (A␤i42/40-WT, 0.65 Ϯ 0.27, n ϭ 9) (Table I), indicating an elevated A␤i42/40 ratio relative to the ratio of A␤sec42/40 (Table I), consistent with previously published data (27,52,54,55). However, the point mutation at position 45 of A␤ that dramatically influenced the 42/40 ratio of secreted A␤ by a factor of 20.4 compared with the WT-protein showed only a minor effect on the 42/40 ratio of intracellular A␤. The ratio of A␤i42/40 was increased by a factor of 1.9 compared with C99 WT (relative ratio, 1.9 Ϯ 0.5 (I45F), p Ͻ 0.01, n ϭ 9) (Table I). A similar phenomenon was obtained for C99 V50F. Compared with C99 WT, the A␤sec42/40 ratio was reduced to 30%, whereas the A␤i42/40 ratio of C99 V50F was not significantly decreased compared with the A␤i42/40 ratio of the WT construct (relative ratio, 0.8 Ϯ 0.5 (V50F), not significant, n ϭ 9) (Table I), indicating that both point mutations have a minor effect on the cleavage site specificity of the ␥-secretase responsible for intracellular A␤ generation. To verify our results that intracellular ␥-secretase is less affected by point mutations close to the ␥-cleavage site, we used C99 V46F, a familial Alzheimer's disease-linked mutation at position 717 of APP770 ( Fig. 1) (56). The A␤sec42/40 ratio of C99 V46F (A␤sec42/40-V46F: 0.67 Ϯ 0.17, p Ͻ 0.01, n ϭ 7) was increased by a factor of 3.1 relative to the WT protein (Table I), consistent with previous findings (37,38), whereas the A␤i42/40 ratio was not affected compared with the A␤i42/40 ratio obtained for the WT  (Table I), further confirming our results obtained for C99 I45F and C99 V50F.
Generation of A␤ Is Inhibited by C99 Proteins Bearing an ER/Intermediate Compartment (IC) Retrieval Signal-To analyze the subcellular compartments involved in A␤ generation, we used different experimental strategies. To determine whether ␥-secretase processing can occur within the ER, we introduced a dilysine motif at the C terminus of C99 (C99 KK) that leads to the retention of proteins in the ER and IC (30,57,58). As shown in Fig. 3A the expression levels of C99 KK proteins (C99 WT KK, C99 I45F KK, C99 V50F KK) were comparable with the expression levels of C99 proteins without the ER/IC retrieval signal (C99 WT, C99 I45F, C99 V50F) and did not affect the levels of endogenous APP (Fig. 3A). However, significant amounts of A␤i could only be recovered from cell lysates of cells expressing C99 constructs without an ER/IC retrieval signal (C99 WT, C99 I45F, C99 V50F) (Fig. 3A). Cell lines expressing C99 proteins bearing the dilysine motif did not produce A␤i levels above the background level of A␤i produced from the endogenous APP, as verified by the comparison with cells stably transfected with the expression vector alone (Fig.  3A). These results clearly show that C99 has to be transported out of the ER to get processed by ␥-secretase, indicating that the ER is not a major intracellular site of ␥-secretase activity. This finding is supported by the observation that A␤ secretion was blocked to endogenous background levels when C99 was retained in the ER and IC (Fig. 3B), providing evidence that A␤ was not generated in the ER and rapidly secreted into the conditioned media.
Brefeldin A Treatment Abolishes Intracellular A␤ Generation-To verify our data that A␤ cannot be generated early within the secretory pathway, we blocked protein transport using BFA. BFA blocks anterograde protein transport out of the ER, resulting in a fusion of the proximal Golgi (cis-and medial-Golgi) with the ER (59,60). The efficacy of BFA treatment was verified by different observations as follows (i) BFA treatment affected the maturation of endogenous APP as seen by a complete change of the observed band pattern of APP immunoreactivity (Fig. 4A). (ii) The generation of A␤sec and ␣-secreted endogenous APP (sAPP␣) was blocked in the presence of BFA, whereas A␤ secretion could be detected in the conditioned media of untreated cells (Fig. 4B). (iii) An increase of endogenous APP was observed in cell lysates of treated cells (Fig. 4C), indicating that the secretory pathway was efficiently blocked by BFA treatment.
The presence of BFA also dramatically affected the genera-tion of A␤i. A␤i could be detected in untreated cells expressing C99 WT, I45F, and V50F (Fig. 4C). However, A␤i formation was strongly impaired in BFA-treated cells (Fig. 4C), indicating that protein trafficking beyond the early secretory pathway is required for A␤i generation and, thus, confirming our results obtained with C99 KK mutant proteins. To verify that the lack of ␥-secretase processing in BFA-treated cells is not simply caused by diminished endoproteolysis of endogenous PS1, which seems to be important for PS function (61), we analyzed the amount of N-and C-terminal PS1 fragments (PS1-NTF and PS1-CTF, respectively) in cells treated or not treated with BFA. Therefore, C99 WT-expressing cells were metabolically labeled in the presence and absence of BFA, and cellular membranes were immunoprecipitated with antibodies directed to the N or C terminus of PS1. Processing of PS1 to its N-and C-terminal  fragments was not inhibited in the presence of BFA as shown in Fig. 4D, indicating that the presence of processed PS1 is not sufficient to obtain ␥-secretase activity.
Monensin Increases Intracellular A␤ Levels-To investigate further the intracellular compartments that are involved in A␤i generation, we incubated SH-SY5Y cells expressing the C99 WT protein or the mutant proteins (C99 I45F, C99 V50F) in the presence or absence of monensin. Monensin inhibits the maturation of newly synthesized proteins in the trans-Golgi and blocks their transport out of the trans-Golgi network (TGN) (62,63). The analysis of the conditioned media of treated and untreated cells showed that secretion of A␤ as well as secretion of sAPP␣ (␣-secreted endogenous APP) was completely blocked in the presence of monensin (data not shown), similar to treatment of cells with BFA. In contrast to BFA, monensin treatment strongly increased A␤i levels (Fig. 5A), indicating that late Golgi compartments are involved in ␥-secretase processing of C99. A␤ levels were increased by a factor of 3.0 in monensin-treated cells stably expressing the C99 WT protein compared with untreated cells (A␤i ϩ monensin/A␤i Ϫ monensin, 3.0 Ϯ 0.9 (WT), p Ͻ 0.05; n ϭ 4). Similar results were obtained for C99 I45F and C99 V50F (A␤i ϩ monensin/A␤i Ϫ monensin, 2.5 Ϯ 0.4 (I45F), p Ͻ 0.05, n ϭ 3; 2.9 Ϯ 0.3 (V50F), p Ͻ 0.01; n ϭ 3) (Fig. 5B).

Generation of A␤ by C99 Proteins Bearing a TGN-sorting
Signal-The results obtained so far provide evidence that protein transport to late Golgi compartments (trans-Golgi, TGN) is essential for ␥-secretase processing of C99. To confirm these results, we established SH-SY5Y cells, stably expressing a C99 protein, bearing the sorting signal of TGN38 for recycling between the cell surface and the TGN (C99 SDYQRL) (64,65). It was previously shown that the addition of the amino acid motif SDYQRL to the C terminus of APP leads to an accumulation of APP in the TGN (34,40). We found that A␤i levels were increased by a factor of 1.5 in cells stably expressing the C99 SDYQRL construct compared with C99 WT (relative ratio A␤i/ C99, 1.5 Ϯ 0.3 (SDYQRL), p Ͻ 0.05, n ϭ 7) (Fig. 6A), confirming the role of the TGN in ␥-secretase processing of C99.
Subcellular Distribution of A␤i40, A␤i42, and Endogenous PS1-To analyze the subcellular distribution of A␤i40 and A␤i42 independent of sorting signals or drug treatments, membranous organelles of SH-SY5Y cells transfected with C99 WT were separated by sucrose density centrifugation. A␤i40 and A␤i42 were measured by a highly sensitive electrochemiluminescence (ECL) assay (10,47), whereas the corresponding fractions were probed by Western blotting for endogenous PS1, C99, C83 (␣-cleaved C-terminal fragment), endogenous APP, and several well characterized organelle marker proteins. ERrich fractions were detected as expected at the bottom of the gradient using an antibody against the ER marker protein calnexin (66). Calnexin-reactive ER vesicles were most en-

FIG. 4. Effects of brefeldin A treatment on A␤ generation, maturation of endogenous APP, and endoproteolysis of PS1. SH-SY5Y
cells stably transfected with the indicated SPC99 constructs or the expression vector alone (control) were preincubated with 10 g/ml BFA for 1.5 h. After preincubation the cells were incubated for additional 10 h in fresh medium containing 10 g/ml BFA. Control cells were incubated in the absence of BFA. A, cell lysates of treated and untreated cells, expressing C99 WT, were prepared and directly loaded on a 8% Tris-Tricine gel. The immature (APP immat) and mature (APP mat) isoforms of endogenous APP were visualized by immunoblotting with antibody W02. B, endogenous ␣-secreted APP (sAPP␣) and A␤ secreted into the conditioned media (A␤sec) were immunoprecipitated and detected by Western blot using antibody W02. C, cell lysates were prepared after BFA treatment and immunoprecipitated with the antibody W02. A␤i, C99, and endogenous APP were detected by immunoblotting with antibody W02. B and C, respectively, represent different exposure times of the same Western blot analysis. D, SH-SY5Y cells stably transfected with C99 WT were metabolically labeled in the presence or absence of BFA (10 g/ml). Cellular membranes were prepared and immunoprecipitated with the indicated antibodies. Precipitated proteins were separated on a 12.5% SDS page. PS1-NTF and PS1-CTF were visualized by autoradiography. riched in fractions 4 -6 ( Fig. 7A). Golgi-containing fractions were identified by blotting for ␤-COP, a Golgi marker protein (67,68) most abundant in the early Golgi compartments, found in fractions 12 and 13 (Fig. 7B). TGN-rich fractions were defined by the TGN marker protein syntaxin 6 (69), enriched in fractions 7-10 ( Fig. 7C).
Using these organelle marker proteins we found that the ␥-secretase substrates C99 and C83 accumulated in TGN-fractions, as defined by the enrichment of the TGN marker protein syntaxin 6 (fractions 7-10) (Fig. 7, A and C). Exogenous expressed C99 was detected to a smaller extent in ER fractions (fractions 4 -6), defined by the marker protein calnexin and the presence of immature endogenous APP (APP immat) (Fig. 7, A and C). As expected C83, the ␣-cleaved C-terminal fragment could not be detected in the ER-rich vesicles (Fig. 7A), supporting previously published data that ␣-secretase cleaves later within the secretory pathway (70 -72). PS1-NTF was detected across the whole sucrose gradient; however, an accumulation of PS1-NTF as well as PS-FL was observed in fractions 7-12 (Fig.  7B), overlapping with the TGN marker protein syntaxin 6 (fractions 7-10) (Fig. 7C). To detect all A␤ peptides ending at position 40 and 42 including N-terminal-truncated A␤ species, for example Glu11-A␤ (73), we used the antibody combinations G2-10/4G8 and G2-11/4G8, respectively. A␤i40 and A␤i42 as measured by ECL are almost identically distributed across the gradient (Fig. 7D). A peak for A␤i x-40 and A␤i x-42 (intracellular A␤ peptides with variable N terminus ending at residue 40 and 42, respectively) was observed in fractions 7-9 (Fig. 7D), suggesting a TGN localization as defined by the accumulation of the TGN marker protein syntaxin 6 in fractions 7-10 ( Fig.  7C). Neither A␤i40 nor A␤i42 could be detected in fraction 5, which showed the strongest staining for the ER marker protein calnexin (Fig. 7A). This excludes the presence of significant amounts of A␤i40 and A␤i42 in the ER, consistent with our findings that A␤ cannot be generated in the ER by analyzing C99 KK mutant proteins. Additionally, A␤i40 and A␤i42 could not be detected in ␤-COP-rich fractions (fractions 12 and 13), FIG. 5. Effect of monensin treatment on A␤i generation. SH-SY5Y cells stably expressing the indicated SPC99 constructs were preincubated for 1.5 h with 5 M monensin, and monensin treatment was performed for an additional 10 h in fresh medium. A, cell lysates of treated (ϩ) and untreated (Ϫ) cells were immunoprecipitated with the antibody W02 followed by Western blot analysis using antibody W02. The figure represents different exposure times of the same Western blot analysis as indicated by the black line within the picture. B, quantification of at least three independent experiments. A␤i detected in the Western blot of treated and untreated cells was quantified densitometrically. The amount of A␤i in the presence of monensin was divided by the amount of A␤i in the absence of monensin for each construct and each experiment (A␤i ϩ monensin/A␤i Ϫ monensin). Thus, the value for untreated cells was 1.0 and was set to 100%. Gray columns represent the mean values; black vertical bars give the S.D. The asterisks indicate the statistical significance (two-sided Student's t test) relative to untreated cells (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). C, SH-SY5Y cells stably expressing C99 WT were incubated with monensin as described. Equal volumes of cell lysates were immunoprecipitated (IP) with the antibody G2-10, specific for A␤40, and the antibody G2-11, specific for A␤42. The immunoprecipitated A␤i40 and A␤i42 peptides were detected with antibody W02 in the Western blot. D, ratios of A␤i42/40 after monensin treatment relative to untreated cells of six independent experiments. For this, the same graphical presentation was used as described under B. supporting our data that A␤ cannot be generated in early Golgi compartments.

DISCUSSION
The ␥-secretase is defined as proteolytic activity resulting in cleavage of APP CTFs releasing the C terminus of A␤ and A␤-like peptides (5)(6)(7)(8). PS1 and PS2 have been found to be essential components of the high molecular weight complex mediating ␥-secretase activity (9,10). The cleavage site specificity of the ␥-secretase enzyme determines the ratio of A␤40 and A␤42. Pathogenic point mutations in APP and presenilins increase the A␤42/40 ratio of secreted A␤ (37, 48 -50).
In the present study we show that point mutations close to the ␥-cleavage site of APP have a minor effect on the cleavage site specificity of intracellular ␥-secretase compared with their dramatic effect on the cleavage site specificity of the ␥-secretase responsible for the generation of A␤sec. The point mutations I45F and V50F that revealed a dramatic effect on the A␤sec42/40 ratio only marginally affected the A␤i42/40 ratio. The relevance of this observation was confirmed by the use of the familial Alzheimer's disease mutant V717F of APP770 (56). Although secreted A␤42 was increased in the presence of the mutant, we observed no increase in intracellular A␤42 production, consistent with recent data (35). Altered ␥-secretase cleavage site specificity may be caused by different proteolytic activities for intracellular and secreted A␤. Wilson et al. (35) report that intracellular A␤ generation in the ER is independent of PS. However, PS-independent A␤i42 production in the early secretory pathway is unlikely to be involved in intracellular A␤ generation in the human SH-SY5Y neuroblastoma cells, because we found late Golgi compartments responsible for intracellular A␤ generation. Thus, it is not unlikely that presenilins are involved in intracellular A␤ generation, and cleavage site specificity may be affected by different co-factors of the high molecular weight PS-containing complexes (9,10). Alternatively, minor changes present in subcellular compartments like lipid composition (75,76), protein glycosylation, or other modifications of ␥-secretase complex proteins (11,13,77,78) may be sufficient to modify cleavage-site specificity.
Our finding that point mutations increasing ␥-cleavage at position 42 have a more pronounced effect on secretory A␤ compared with intracellular A␤ may indicate that familial Alzheimer's disease-linked mutant proteins initiate accumulation of A␤42 in the extracellular A␤ pool and that intracellular A␤ production has a minor role in the pathogenesis of AD. However, we cannot exclude that slightly elevated intracellular A␤42 levels or additional mechanisms that we did not observe in our in vitro cell culture system may be sufficient to start A␤ aggregation. Indeed, intracellular A␤42 induces neurotoxicity in primary rat and human neurons (79,80). Moreover, studies on AD-vulnerable brain regions as well as animal models expressing familial Alzheimer's disease-related mutant proteins implicated intraneuronal accumulation of A␤42 (20,22,23).
The analysis of the intracellular sites of ␥-cleavage shows that intracellular A␤ generation was inhibited when cells were treated with BFA or were stably transfected with constructs bearing an ER/IC retrieval signal despite the presence of processed PS1. These results are consistent with previously published data that A␤i generation was inhibited after BFA treatment in the mouse neuroblastoma cell line N2a and COS1 cells as well as in different cell lines transiently transfected with truncated APP constructs bearing an ER/IC retrieval signal, including kidney 293 cells and N2a cells (26,30,32,34). However, A␤42 generation was also found within the ER using very similar approaches (27,29,31). A␤42 generation within the ER seems to be more abundant in differentiated neurons as shown for rat primary neurons (28) and differentiated NT2N cells (27,31,81) derived from the human embryonal carcinoma cell line NT2 (82). Interestingly, the less differentiated NT2 cells fail to produce intracellular A␤ (54). However, a neuronal phenotype cannot explain the different results obtained for NT2N cells infected with SFV/APP695 KK (27)   The vertical bar indicates that the samples were loaded onto two separate 10 -20% Tris-Tricine gels, blotted onto the same nitrocellulose membrane. As a standard (std) 1.0 ng of recombinant C100FLAG (10) was loaded on each gel. Aliquots of individual fractions were directly loaded on a 12% SDS-gel and immunoblotted for ␤-COP and PS1 (PS1-FL and PS1-NTF) (B) and on a 10% SDS-gel for the detection of endogenous immature and mature APP (APP immat, APP mat) and syntaxin 6 (C). A, B, and C, the data shown are representative of at least three independent experiments. D, the individual fractions were analyzed by an ECL assay to determine the subcellular distribution of A␤i40 and A␤i42. A␤i x-40 and A␤i x-42 were detected by the use of the antibody combination 4G8/G2-10 and 4G8/G2-11, respectively. The graph shows the average of duplicate ECL measurements and is representative of at least three independent experiments. When we take into account that even very closely related cell culture lines or primary cells show significant differences in ␥-secretase 42 activity in different subcellular compartments, it appears that intracellular A␤42 generation can be variable between different cell types. However, our data clearly show that the ER/early Golgi is not the major intracellular site for ␥-secretase processing in the human SH-SY5Y neuroblastoma cells. Our findings are further confirmed by our subcellular fractionation studies. Neither A␤i40 nor A␤i42 accumulation was detected in the ER-and early Golgi-containing fractions. Moreover, we obtained several lines of evidence that late Golgi compartments are involved in ␥-secretase processing. First, A␤i levels were dramatically increased when we blocked protein transport late in the secretory pathway using monensin, consistent with studies proposing the involvement of late Golgi compartments in ␥-secretase processing (26,32,34,84). Second, A␤i generation was increased in cells expressing a C99 mutant directed to the TGN. Third, subcellular fractionation of organelles from C99 WT-transfected SH-SY5Y cells revealed that A␤i40 and A␤i42 accumulated in TGN-rich fractions. Interestingly, we found that PS1 and the direct ␥-secretase substrate accumulated in TGN fractions as well, indicating the possibility of an interaction between PS1 and C99, resulting in ␥-secretase processing in these compartments. These results are supported by findings that PS1 forms complexes with APP C-terminal fragments in Golgi-and TGN-rich fractions and that de novo A␤i generation was found in the same Golgi-/TGNrich vesicles (85). In accordance to our ER-related data we conclude that the presence of processed PS1 is not sufficient to obtain ␥-secretase processing and that at least one additional factor is required for the formation of an active ␥-secretase complex in late Golgi compartments. Such additional factors could be the transmembrane proteins aph-1, pen-2, and nicastrin that have been recently shown to be essential ␥-secretase complex components (11,13,77,78). Interestingly, the fully mature glycosylated form of nicastrin preferentially interacts with PS1 (86). In contrast, only a small proportion of PS1 is bound to the immature species of nicastrin, indicating that mature nicastrin may be essential for the formation of a functional ␥-secretase complex (86) and, thus, confirming our results that an active ␥-secretase complex is formed in late Golgi compartments where fully glycosylated mature nicastrin is present.
The identification of differences in ␥-secretase cleavage site specificity for intracellular and secretory A␤ generation shows that the ␥-secretase system is even more complex than previously assumed. This may have important implications for AD. It shows that at least partly different ␥-secretases exist. Because ␥-secretase cleaves a number of different substrates, like Notch (for review, see Ref. 8) and low density lipoprotein receptor-related protein (LRP) (83), this finding may help to target A␤ processing in a manner avoiding cross-inhibition with other ␥-secretase substrates.