Expression of Alzheimer’s Disease-associated Presenilin-1 Is Controlled by Proteolytic Degradation and Complex Formation*

Numerous mutations causing early onset Alzheimer’s disease have been identified in the presenilin (PS) genes, particularly the PS1 gene. Like the mutations identified within the β-amyloid precursor protein gene, PS mutations cause the increased generation of a highly neurotoxic variant of amyloid β-peptide. PS proteins are proteolytically processed to an N-terminal ∼30-kDa (NTF) and a C-terminal ∼20-kDa fragment (CTF20) that form a heterodimeric complex. We demonstrate that this complex is resistant to proteolytic degradation, whereas the full-length precursor is rapidly degraded. Degradation of the PS1 holoprotein is sensitive to inhibitors of the proteasome. Formation of a heterodimeric complex is required for the stability of both PS1 fragments, since fragments that do not co-immunoprecipitate with the PS complex are rapidly degraded by the proteasome. Mutant PS fragments not incorporated into the heterodimeric complex lose their pathological activity in abnormal amyloid β-peptide generation even after inhibition of their proteolytic degradation. The PS1 heterodimeric complex can be attacked by proteinases of the caspase superfamily that generate an ∼10-kDa proteolytic fragment (CTF10) from CTF20. CTF10 is rapidly degraded most likely by a calpain-like cysteine proteinase. From these data we conclude that PS1 metabolism is highly controlled by multiple proteolytic activities indicating that subtle changes in fragment generation/degradation might be important for Alzheimer’s disease-associated pathology.

In addition to processing by the presenilinase, both PS proteins are substrates for proteinases of the caspase superfamily (31)(32)(33)(34). In this pathway the 20-kDa CTF 20 serves as a substrate for caspase cleavage (33) which generates a smaller ϳ10-kDa C-terminal fragment (Fig. 1A). Caspase-mediated cleavage appears to be enhanced by a FAD mutation occurring within the PS2 gene (31).
Formation of CTF 20 is highly regulated. Overexpression of PS results in the replacement of endogenous PS fragments but not in a linear accumulation of the overexpressed fragments (23,35). Moreover, overexpression of PS1 proteins not only results in a replacement of the endogenous PS1 fragments but also blocks the synthesis of PS2 fragments and vice versa (23,35). The very tight regulation of PS processing and fragment accumulation indicates that a limited factor binds to PS proteins (35). Indeed, as described above PS proteins form a complex, which contains the PS heterodimer (28 -30, 36) and probably binding protein(s) as well. Moreover, slight changes in the tightly balanced fragment formation might be pathologically relevant (see Refs. 24,37,and 38; for review see Ref. 39). We therefore analyzed the stability of the PS precursor and its proteolytic fragments in transfected and untransfected cells and found that PS1 expression is highly controlled by multiple proteolytic degradation pathways. The formation of a heterodimeric complex protects the fragments from their rapid degradation, and complex formation appears to be required for the pathological activity of PS on A␤42 generation. Our data suggest that proteolytic degradation plays a major role in a balanced and equal expression of PS fragments. These data further indicate that slight changes in the stability of excess amounts of PS fragments might be pathologically relevant.
Antibodies-The anti-PS antibodies used are described previously (12,33,42). The epitopes of the antibodies used are shown in Fig. 1A. The anti-PARP antibody was obtained from Boehringer Mannheim. Anti-␤-catenin antibodies were obtained from Sigma. The C7 antibody to the C terminus of ␤APP and the detection of C-terminal ␤APP fragments were described previously (40). Apoptosis was induced with 1 M staurosporine (Biomol) dissolved at 1 mM in Me 2 SO. Treatment with inhibitors was performed in 4 ml of culture medium for the indicated times using cells grown to 80 -100% confluence in 10-cm dishes.

Treatment of Cells with Proteinase Inhibitors and Other Com
Analysis of PARP Cleavage as a Biological Marker for Apoptosis-Analysis of apoptosis in cell culture upon treatment with proteinase inhibitors was monitored in lysates of HeLa cells by the identification of the caspase-generated cleavage product of PARP (31).
Pulse-Chase Experiments-Stably transfected K293 cell lines were grown to confluence in 10-cm dishes. After starvation for 2 h in 4 ml of methionine-and serum-free medium (MEM, L-glutamine, 1% penicillin/ streptomycin), cells were metabolically labeled with 500 -700 Ci of [ 35 S]methionine (Amersham Pharmacia Biotech Promix) in 4 ml of methionine-free minimal essential medium for 20 -30 min as described under "Results." The cold chase was performed by replacing the labeling medium with 4 ml of chase medium (Dulbecco's minimal essential medium, 10% fetal calf serum, 1% penicillin/streptomycin, 0.8 mM Lmethionine) for the indicated periods. Cell extracts were prepared, and immunoprecipitation was performed as described (42). Proteinase inhibitors were added during starvation, labeling, and cold chase periods where indicated.
Isolation of Membrane Proteins-K293 cells were grown to confluence. Cells of five 10-cm dishes were scraped in phosphate-buffered saline and pelleted. The cell pellet was washed three times in phosphate-buffered saline. Cells were then resuspended in 5 ml of RSB buffer (10 mM Tris, pH 7.5, 20 mM KCl, 1.5 mM MgAc 2 ) containing proteinase inhibitors as described (28) and homogenized by passing the cell suspension 10 times through a 23-gauge needle. To prepare a postnuclear supernatant, the homogenate was centrifuged at 1000 ϫ g for 15 min at 4°C. Membranes from the postnuclear supernatant were then pelleted by centrifugation for 1 h at 100,000 ϫ g at 4°C. Membranes were washed in a high salt HEPES buffer (1 M KCl, 20 mM HEPES, pH 7.2, 2 mM EGTA, 2 mM EDTA, 2 mM dithiothreitol) containing proteinase inhibitors. The purified membranes were extracted with 2% CHAPS in HEPES buffer (100 mM KCl, 20 mM HEPES, pH 7.2, 2 mM EGTA, 2 mM EDTA, 2 mM dithiothreitol), containing proteinase inhibitors as described (28) for 1 h on ice. Membrane extracts were cleared by ultracentrifugation for 1 h at 100,000 ϫ g at 4°C. Protein concentrations were determined by Bio-Rad assay.
Co-immunoprecipitations-Co-immunoprecipitations were carried out as described (28). Briefly, membranes were prepared as described above and were extracted with 2% CHAPS. To remove undissolved membrane fragments, the extracts were pelleted by ultracentrifugation for 1 h at 100,000 ϫ g at 4°C. Incubation with PS1 antibodies was performed as described (28). Immunoprecipitations were washed 4 times for 20 min in CHAPS washing buffer (0.5% CHAPS, 200 mM NaCl, 50 mM HEPES, pH 7.6) prior to SDS-polyacrylamide gel electrophoresis.
Detection of A␤40 and A␤42-Cells were plated at equal density in 6-well plates. After reaching confluence, 2 ml of conditioned media were collected for 18 h. Media were centrifuged to remove cell fragments, and aliquots were then used to determine the ratios of A␤42 to total A␤ (A␤40 and A␤42). A highly specific sandwich ELISA employing monoclonal antibodies specific for the detection of A␤40 or A␤42 was developed according to previous protocols (43).
The PS1 Precursor but Not the NTF and CTF 20 Is Rapidly Degraded by a Pathway That Is Sensitive to Inhibitors of the Proteasome-Untransfected K293 cells or K293 cells stably transfected with wt PS1 (15,26) were metabolically labeled for 30 min with [ 35 S]methionine and chased for 4 h in the presence of excess amounts of unlabeled methionine. Cell lysates were immunoprecipitated with antibody 3027 (Fig. 1A) to detect the full-length protein. Immediately after pulse labeling robust amounts of PS1 holoprotein were detected in transfected cells (Fig. 1B). Interestingly, we detected the 20-kDa CTF 20 upon a 4-h cold chase, whereas the full-length precursor almost completely disappeared (Fig. 1B). However, only a very small amount of the full-length protein was turned over into the fragment, indicating that the majority of uncleaved PS1 is rapidly degraded (Fig. 1B). In order to analyze which proteinase could be involved in the rapid degradation of the full-length protein, we treated the cells with or without inhibitors of the proteasome, calpain-like proteinases, cysteine proteinases, and inhibitors of endosomal/lysosomal proteinases during a cold chase. Immunoprecipitation with antibody 3027 revealed that degradation of the PS precursor could be partially reduced by inhibitors that specifically affect the proteasome (lactacystin) and by inhibitors that affect both the proteasome and calpains (LLnL; see Ref. 48; Fig. 1B). In contrast, degradation of PS1 was not reduced by the calpain inhibitor MDL28170 or by inhibitors of endosomal/lysosomal proteinases (leupeptin and NH 4 Cl). The involvement of the proteasome in the removal of PS1 holoprotein is supported by the reduction of PS1 degradation by 10 M lactacystin. This inhibitor is known to specifically inhibit proteasomal activity at the concentration used in this experiment (49). PS1 degradation was also reduced by MG132 (data not shown), another inhibitor known to potently block the proteasome (50). Moreover, since MDL28170 did not stabilize the PS1 precursor, whereas lactacystin did (Fig. 1B), the inhibitory effect of LLnL on PS1 degradation can most likely be attributed to proteasomal inhibition. Since only a fraction of full-length PS1 could be stabilized by proteasomal inhibitors, another proteolytic activity, which is not blocked by the inhibitors used, might also be involved in the removal of PS1 (see "Discussion"). Interestingly, we did not observe any effect of the proteinase inhibitors on the stability of the PS1 CTF 20 during the cold chase (Fig. 1B), indicating that the CTF 20 is stable (see below). To analyze the stability of steady state levels of the endogenous PS fragments, cell lysates of unlabeled cells treated with and without the inhibitors were immunoprecipitated with the polyclonal antibody 2953 to the N-terminal domain of PS1 (Fig. 1C), and the immunoprecipitated PS proteins were then identified with the monoclonal antibody PS1N (a gift from R. Nixon; see Ref. 51; Fig. 1A). As shown in Fig. 1C, inhibition of proteasomal degradation by treatment with LLnL or lactacystin did not significantly affect the amounts of NTF detected. Endosomal/lysosomal proteinases as well as other cysteine proteinases are also not involved in the degradation of the NTF, since treatment of cells with leupeptin or NH 4 Cl did not result in a stabilization of the PS1 NTF (Fig. 1C). We next immunoprecipitated the PS1 CTF 20 using the polyclonal antibody 3027. Immunoprecipitated CTF 20 was detected using the monoclonal antibody APS18. Again, inhibition of proteasomal activity with either LLnL, lactacystin, or inhibition of endosomal/lysosomal proteinases and other cysteine proteinases with FIG. 1. The PS1 precursor, but not the NTF and CTF 20 , is rapidly degraded. A, schematic representation of PS1. The epitopes of all anti-PS1 antibodies used are shown as bars. The proteases involved in proteolytic processing of PS1 are indicated in italics. Numbers in parentheses indicate the site of cleavage. aa, amino acid. The black box indicates the domain encoded by exon 9. B, the wt PS1 holo-protein, but not the proteolytic fragments, are rapidly degraded. K293 cells stably transfected with PS1 were pulse-labeled for 30 min and chased for 4 h with and without the indicated proteinase inhibitors. Cell lysates were immunoprecipitated with antibody 3027 to detect the full-length PS1 protein as well as CTF 20 . Degradation of the PS1 holoprotein is reduced (but not blocked) by LLnL and lactacystin, whereas the same inhibitors do not affect the levels of CTF 20 . C, steady state levels of NTF and the CTF 20 (1st and 2nd panels) are resistant to proteolytic degradation. K293 cells were treated with and without the indicated proteinase inhibitors for 18 h (leupeptin and NH 4 Cl) or 24 h (LLnL and lactacystin). Both fragments were detected by a combined immunoprecipitation/immunoblotting protocol (26,33,35). Minor differences in the levels of the corresponding fragments are due to experimental variations. The activity of proteasome inhibitors was confirmed by the detection of stabilized ␤-catenin (52). Arrowhead indicates the high molecular weight variants of ␤-catenin observed after inhibition of the proteasome (3rd panel). Inhibition of lysosomal proteinases (4th panel) was proved by the detection of stabilized C-terminal fragments of ␤APP (40,53). Aliquots of cell lysates from the experiment described in C were immunoblotted with the anti-␤-catenin antibody and with anti-C7 to the C terminus of ␤APP (40). D, rapid degradation of de novo synthesized PS1⌬exon9. K293 cells stably transfected with PS1⌬exon9 were pulse-labeled for 30 min and chased for 4 h with and without the indicated proteinase inhibitors. Cell lysates were immunoprecipitated with antibody 3027 to detect the PS1⌬exon9 protein. Degradation of the PS1⌬exon9 protein is reduced but not blocked by LLnL and lactacystin. leupeptin or NH 4 Cl did not result in significant changes in the amounts of CTF 20 detected (Fig. 1C). To prove if the proteasome and endosomal/lysosomal inhibitors were indeed active, we analyzed the turnover of ␤-catenin and C-terminal fragments of ␤APP within the same samples. As reported previously (52) we observed the high molecular weight ubiquitinated variants of ␤-catenin in the samples that were treated with proteasome inhibitors but not in samples treated with inhibitors of the endosomal/lysosomal pathway (Fig. 1C). In contrast, treatment with endosomal/lysosomal inhibitors stabilized the C-terminal fragments of ␤APP as described previously (40,53). In addition LLnL but not lactacystin lead to an augmentation of the ␤APP fragments, which is again in agreement with previous results (54). We therefore conclude from these data that the PS1 NTF and CTF 20 are not degraded by the proteasome or endosomal/lysosomal proteinases as well as other cysteine proteinases but are rather stable over long periods.
By having established the degradation of the wt PS1 holoprotein, we next wanted to investigate the proteolytic degradation pathway of a mutant uncleaved PS1 derivative. In order to study this process, we used K293 cells stably expressing PS1⌬exon9 (15), which is known to accumulate as an uncleaved holoprotein due to the lack of the cleavage site within exon 9 (23). The cell line was pulse-labeled with [ 35 S]methionine for 30 min and chased for 4 h in the presence of excess amounts of unlabeled methionine. Cell lysates were immunoprecipitated with antibody 3027 to the C terminus of PS1. Robust amounts of a PS1 holoprotein were detected in the cells stably transfected with the PS1⌬exon9 cDNA (Fig. 1D). During the chase period the PS1⌬exon9 protein was rapidly degraded (Fig. 1D). As observed for the wt holoprotein, only proteasome inhibitors reduced the degradation of PS1⌬exon9. Again, only a fraction of PS1⌬exon9 could be stabilized by proteasome inhibitors indicating the potential participation of an additional proteolytic activity, which is insensitive to the inhibitors used (see "Discussion").
NTFs Not Incorporated into the Heterodimeric Complex Are Degraded by the Proteasome and Do Not Affect the Pathological Generation of A␤42 upon Proteasome Inhibition-Previously it has been observed that a recombinant NTF (NTF 298 ) of PS1 artificially produced due to a stop codon inserted at amino acid 298 (the predominant cleavage site of PS1; see Ref. 26) can only be detected upon a short pulse labeling but not by steady state metabolic labeling (44). To test if that phenomenon can be explained by proteolytic degradation, we performed a pulsechase experiment. K293 cells stably expressing NTF 298 (44) were metabolically labeled with [ 35 S]methionine for 20 min and chased in the presence of excess amounts of unlabeled methionine for 2 h. Cell lysates were immunoprecipitated with antibody 2953. This revealed very high levels of the recombinant NTF 298 during the short pulse ( Fig. 2A). In agreement with the previous results, endogenous NTF could not be labeled during the pulse period (Ref. 44 and data not shown). Upon a 2-h cold chase, NTF 298 was quantitatively removed, indicating very efficient proteolytic degradation of this fragment ( Fig. 2A). To identify the proteinase involved, we treated cells during the labeling and cold chase period with proteinase inhibitors. Whereas proteasome inhibitors efficiently blocked the degradation of NTF 298 , all other inhibitors did not affect the rapid removal of the recombinant fragment ( Fig. 2A). Thus, the proteasome is involved in the degradation of the recombinant NTF 298 .
We next asked the question, why endogenous NTFs derived from proteolytic processing of the holoprotein are resistant to the proteasome, whereas the recombinant NTF terminating after amino acid 298 and the full-length protein are highly sensitive to proteasome-mediated degradation. Previously, we and others (28 -30) have shown that the CTF 20 and NTF bind to each other and form a heterodimeric complex. To investigate whether the endogenous NTF can be replaced by the recombinant NTF 298 within the heterodimeric complex, we performed co-immunoprecipitation experiments. In order to avoid artifacts due to the addition of an epitope tag, which might interfere with protein-protein interactions, we first investigated if the endogenous NTF can be discriminated from the recombi-FIG. 2. NTF 298 is rapidly degraded by the proteasome and is not incorporated into the PS1 heterodimeric complex. A, the NTF 298 is degraded by the proteasome. K293 cells stably transfected with the NTF 298 cDNA were pulse-labeled for 20 min and chased for 2 h with and without the indicated proteinase inhibitors. Cell lysates were immunoprecipitated with antibody 2953 to detect the NTF 298 . Degradation of NTF 298 is efficiently blocked by LLnL and lactacystin. B, the endogenous NTF and NTF 298 can be discriminated based on their differential migration behavior on 10% SDS gels. Cell lysates from K293 cells stably expressing NTF 298 treated with and without LLnL and untransfected K293 cells as a control (Ctr.) were subjected to immunoprecipitation with antibody 2953, and NTFs were detected with the monoclonal antibody PS1N. The recombinant NTF 298 (arrowhead) migrates just below the endogenous NTF (arrow) and is strongly augmented upon treatment with LLnL. LLnL treatment does not affect the stability of the endogenous NTF (upper band; arrow) which is consistent with the data presented in Fig. 1C 28 -30). In cells transfected with the NTF 298 cDNA an additional band is detected that migrates below the prominent endogenous PS1 NTF (see B). Cell lysates were depleted from the PS1 heterodimeric complex by two additional rounds of co-immunoprecipitation using antibody 3027 (IP-Depl.). After depletion, the cell lysate was again immunoprecipitated (Re-IP) but now with an antibody to the N-terminal domain of PS1 (antibody 2953). In cells stably expressing NTF 298 but not in the untransfected control cells, this reveals robust amounts of NTF 298 . Since NTF 298 is rapidly degraded (A), cells were treated with LLnL to allow its efficient detection. nant NTF 298 . Cell lysates from either untransfected cells or NTF 298 -transfected cells treated with and without LLnL were immunoprecipitated with antibody 2953 and detected with antibody PS1N. Inhibition of proteasomal activity was used to allow the efficient detection of the recombinant NTF 298 (see above). Under these conditions the steady state levels of endogenous PS1 fragments are detected (44), in contrast to the pulsechase experiment shown in Fig. 2A. Immunoblotting with antibody PS1N revealed in untransfected cells a predominant endogenous NTF and a much less intense band migrating just below it (Fig. 2B). In cells stably transfected with NTF 298 , the lowest band was slightly elevated under control conditions, indicating expression of a transgene-derived gene product. Moreover, upon treatment with LLnL this band was strongly augmented, whereas the higher molecular weight band was not affected (Fig. 2B). This is consistent with our finding that NTF 298 but not the endogenous NTF is stabilized by inhibition of the proteasome. Therefore, the lower band corresponds predominantly to the recombinant NTF 298 , and the upper band represents the endogenous NTF. The differential migration might be explained by post-translational modifications of the long lived endogenous NTF leading to an increased molecular weight. After demonstrating that both fragments can be efficiently and reproducibly discriminated based on their differential migration behavior on SDS gels, membrane fractions of LLnL-treated cells stably expressing NTF 298 were subjected to immunoprecipitation (Fig. 2C, upper panel). In parallel, endogenous PS fragments were immunoprecipitated from membrane fractions prepared from untransfected cells that were also treated with LLnL (Fig. 2C, lower panel). Membrane fractions were immunoprecipitated with antibody 3027 (to the large loop) or 2953 (to the N terminus) and immunoblotted with antibody PS1N to the N terminus of PS1. Antibody 3027 coimmunoprecipitated the endogenous NTF ( Fig. 2C; upper and lower panel, 2nd lane) which is consistent with data observed previously (28 -30). Again, we detect two predominant polypeptides in the cell line expressing NTF 298 (Fig. 2C, upper panel). The lower band corresponds to NTF 298 , whereas the higher band is equivalent to the endogenous NTF. Note that the lower band is strongly augmented upon immunoprecipitation with antibody 2953 but not with antibody 3027, suggesting that the recombinant NTF 298 might not be co-immunoprecipitating with the heterodimeric complex. We then wanted to deplete the cell lysate from the heterodimeric complex and to analyze if in such a depleted extract the recombinant NTF 298 is retained. We therefore re-immunoprecipitated the cell lysates with antibody 3027 (Fig. 2C). After two rounds of consecutive re-immunoprecipitations we observed very low amounts of the coimmunoprecipitating NTF indicating at least a partial depletion of the lysate from the heterodimeric complex (Fig.  2C). The depleted lysate was then immunoprecipitated with an antibody to the N-terminal domain of PS1 (antibody 2953). This revealed a very prominent NTF in the transfected cell line but not in untransfected control cells (Fig. 2C, compare last  lane in the upper and lower panel). In contrast to the immunoprecipitation with the antibody 2953 before depletion, we now observe predominantly the lower molecular weight band (Fig.  2C, upper panel; compare last lane with 1st lane), which corresponds to the recombinant NTF 298 (Fig. 2B). When the same blot was decorated with antibodies to the large loop of PS1 (antibody 3027), we detected almost no CTF 20 in the final immunoprecipitation (data not shown), again demonstrating that the majority of NTF 298 is not bound to endogenous PS fragments. Therefore, the overexpressed NTF 298 does not replace the endogenous NTF within the heterodimeric complex but can be immunoprecipitated as a free fragment after the PS complex has been largely removed. These results indicate that heterodimer formation is required for the observed stability of endogenous PS1 fragments. Fragments lacking their binding partners are rapidly degraded by the proteasome.
Previously it was demonstrated that NTF 298 containing a FAD-associated point mutation does not cause an increased generation of A␤42 (44). 2 However, here we demonstrated that this fragment is highly unstable, and most of it is rapidly removed by the proteasome. Therefore, an accurate determination of its pathological effects on A␤42 formation might not be possible without inhibition of its proteasomal degradation. In order to investigate if stabilized mutant PS1 NTF 298 could affect the rate of A␤42 generation, cells were treated with and without LLnL (which efficiently stabilizes NTF 298 as shown in Fig. 2A), and the conditioned media were analyzed for the ratio of A␤42:total A␤ (A␤40 and A␤42) using a sandwich ELISA which allows the quantitative and specific detection of both peptides. Four cell lines expressing either the Swedish mutant ␤APP alone (45) or together with the wt NTF 298 , the mutant NTF 298 Y115H (44), or PS1⌬exon9 (as a positive control for enhanced production of A␤42 caused by PS mutations) were used (15). Swedish mutant ␤APP was co-expressed to facilitate A␤ detection as described previously (15). Fig. 3 demonstrates that as expected the PS1⌬exon9 mutation results in an increased ratio of A␤42:total A␤ (A␤40 and A␤42) (see Refs. [13][14][15], whereas the expression of NTF 298 Y115H had no effect on A␤42 generation as demonstrated before (44). As expected from previous results (55,56), treatment with LLnL results in an increase of the ratio of A␤42:total A␤ (A␤40 and A␤42), which is due to the inhibition of calpains (56). This is not only observed in cells expressing endogenous PS1 but also in cell lines expressing mutant and wt NTF 298 as well as PS1⌬exon9 (Fig. 3). However, stabilization of mutant NTF 298 by proteasome inhibition did not result in any increase of the ratio of A␤42:total A␤ (A␤40 and A␤42) as compared with wt NTF 298 (Fig. 3). From these data we conclude that a mutant recombinant NTF does not contribute to A␤42 production even when its degradation is blocked by proteasomal inhibitors. Moreover, these results might indicate that complex formation is required for the pathological activity of PS1 (see "Discussion").
Endogenous CTF 20 Is Proteolytically Degraded upon Previous Caspase Cleavage-As shown above, the endogenous PS1 CTFs are very stable (Fig. 1C). However, we and others (31)(32)(33)(34) have shown previously that the endogenous PS1 CTF 20 can be an in vivo substrate for an additional cleavage mediated by a proteinase of the caspase superfamily even without the induction of apoptosis (Fig. 1A). Therefore, we next wanted to analyze if the cleavage of the PS1 CTF 20 by caspases makes the resulting CTF 10 accessible for further proteolytic degradation. We investigated the levels of CTF 10 within the samples shown in Fig. 1C (where K293 cells were treated with LLnL, lactacystin, NH 4 Cl, or leupeptin) by immunoblotting with antibody 3027, which efficiently detects CTF 10 (33). As shown in Fig. 4A, treatment of cells with LLnL caused a marked accumulation of CTF 10 that was barely detectable in the untreated cells suggesting that CTF 10 is rapidly removed by proteolysis. The specific proteasome inhibitor lactacystin also enhanced the levels of CTF 10 but to a significantly lesser extent than LLnL (Fig.  4A), suggesting that the proteasome might only play a minor role in the degradation of CTF 10 . Interestingly, treatment of cells with leupeptin led to a pronounced increase of CTF 10 , whereas NH 4 Cl had almost no effect (Fig. 4A). From these results it appeared that a cysteine proteinase activity might contribute to the final degradation of the PS1 CTF. Since high concentrations of leupeptin (100 M) might inhibit the proteasome as well (57), we titrated the concentration of leupeptin. As shown in Fig. 4B, leupeptin inhibited degradation of CTF 10 at concentrations as low as 10 M, making it unlikely that leupeptin was inhibiting the proteasome. In addition, immunoblotting with anti-␤-catenin antibodies failed to detect high molecular weight variants of this protein upon treatment with leupeptin (see Fig. 1C) further demonstrating that leupeptin does not inhibit the proteasome under the conditions used in this experiment. Moreover, E64, a specific cysteine proteinase inhibitor that does not inhibit the proteasome (58) also blocked degradation of CTF 10 (Fig. 4C). Taken together, these data therefore demonstrate that CTF 10 is predominantly degraded by a cysteine proteinase. Since calpains are cysteine proteinases that can also be inhibited by LLnL (see above and see Ref. 48), we also treated cells with the calpain inhibitor MDL28170. As observed for leupeptin or LLnL treatment, MDL28170 significantly stabilized the CTF 10 (Fig. 4C), therefore indicating that calpain-like proteinases are most likely involved in its degradation. Previously, we and others (32)(33)(34) have noticed that the amounts of CTF 10 are highly variable depending on the cell line or tissue analyzed. We therefore analyzed CTF 10 degradation in another cell type. HeLa cells were treated with the cysteine proteinase inhibitor leupeptin and the proteasome inhibitor lactacystin. Although HeLa cells expressed higher levels of CTF 10 under control conditions (Fig. 4D), treatment with leupeptin still allowed its augmentation (Fig. 4D). Again, as observed for K293 cells, the effect of lactacystin on CTF 10 stabilization was only modest if detectable at all, although the drug was active in HeLa cells as judged from the accumulation of high molecular weight ␤-catenin variants (see Fig. 4E). The same result was obtained when the cysteine proteinase inhibitor E64 was used in these experiments (not shown). Based on the results in Fig. 4, A-C, we could not exclude that treatment with the variety of inhibitors used might have induced apoptosis resulting in an enhanced production of CTF 10 rather than increased stabilization. We therefore investigated the possible induction of apoptosis by monitoring cleavage of poly(ADPribose) polymerase (PARP) (31) within the samples analyzed above. Whereas staurosporine as positive control for the induction of apoptosis caused a quantitative cleavage of PARP and an enhanced generation of CTF 10 , no PARP cleavage was observed in the cells treated with the proteinase inhibitors (Fig.  4D), demonstrating that apoptosis was not the cause of the observed CTF 10 accumulation. As a further control we treated HeLa cells with Z-VAD, an inhibitor of caspase-1 like proteinases, to inhibit de novo production of CTF 10 by blocking the caspase activation cascade (Refs. 31-34; for review see Ref. 59). Treatment of HeLa cells with Z-VAD resulted in a significant reduction of CTF 10 (Fig. 4E). However, when cells were treated with Z-VAD and leupeptin simultaneously, CTF 10 could still be detected in amounts that were increased compared with untreated cells (Fig. 4E), despite the inhibition of de novo production of CTF 10 . Simultaneous treatment of Z-VAD and lactacystin did not lead to a stabilization of CTF 10 further supporting the observation that a cysteine proteinase rather than the proteasome plays the major role in CTF 10 degradation. The activity of lactacystin was monitored by the detection of high molecular weight variants of ␤-catenin, stabilized upon proteasome inhibition (Fig. 4E). Taken together, these data rule out that treatment with the various proteinase inhibitors induced apoptosis resulting in an accumulation of CTF 10 due to enhanced caspase cleavage. Moreover, these data demonstrate that the CTF 10 is predominantly degraded by a cysteine proteinase and, depending on the cell line used, possibly also to a lesser extent by the proteasome. The much higher levels of CTF 10 detected in HeLa cells as compared with K293 cells FIG. 3. Mutant NTF 298 does not increase the production of A␤42 even upon its stabilization. K293 cells stably expressing Swedish ␤APP (Ctr.) or Swedish ␤APP with the PS1⌬exon9 mutation (PS1⌬exon9) (15) were incubated for 18 h (upper panel) with and without LLnL. The ratios of A␤42 to total A␤ (A␤42 and A␤40) were determined by a sandwich ELISA according to Suzuki et al. (43). Bars represent means Ϯ S.E. of three independent experiments. As described previously (13)(14)(15)(16)(17), the PS1⌬exon9 mutation causes a significant increase of A␤42 production demonstrating the specificity of the ELISA. Lower panel, K293 cells stably expressing Swedish ␤APP and NTF 298 with (NTF 298 Y115H) and without the Y115H mutation were treated with LLnL or the vehicle alone. Note, that the mutant NTF 298 (NTF 298 Y115H) does not induce A␤42 generation even upon its stabilization with LLnL. Treatment with LLnL increases the ratio of A␤42 to total A␤ as observed previously (56). might suggest that the cysteine proteinase is less abundant or active in HeLa cells. This difference might also account for the different levels of CTF 10 detectable in various cell lines or tissues. In addition, these data also indicate that CTF 20 must be first cleaved by caspases to make the resulting CTF 10 accessible for its final degradation.
C-terminal Fragments Not Incorporated into the PS1⌬exon9 Complex Are Rapidly Degraded by the Proteasome-Upon treatment of K293 cells expressing PS1⌬exon9 with LLnL and lactacystin, surprisingly we detected not only the CTF 10 fragment (which is known to be generated from the PS1⌬exon9 holoprotein; Ref. 33) but also a higher molecular weight Cterminal fragment (CTF ⌬9 ) that was identified by antibody 3027 (Fig. 5A, upper panel). This fragment strongly accumulated upon inhibition of proteasome activity by LLnL and lactacystin but not of other inhibitors such as leupeptin and NH 4 Cl (Fig. 5A). This further supports the differential degradation of distinct CTFs by the proteasome and cysteine proteinases (see above). To prove further the origin of CTF ⌬9 , the blot shown in Fig. 5A was reprobed with APS18 (amino acids 314 -334; see Ref. 28). Antibody APS18 is known not to recognize CTF 10 since its epitope is N-terminal to the caspase cleavage site after amino acid 345 (28,32,33). As shown in Fig. 5A this antibody did identify CTF ⌬9 but not CTF 10 . Co-migration of the precipitated CTF ⌬9 with CTF 20 revealed that CTF ⌬9 has a lower molecular weight as the corresponding endogenous CTF 20 (Fig. 5B). CTF ⌬9 was also not observed upon staurosporine treatment (data not shown). This strongly suggests that CTF ⌬9 is generated by an alternative cleavage of PS1⌬exon9 N-terminal to Asp-345 but not by caspase-mediated processing, thus indicating that this fragment corresponds to the endogenous CTF 20 . It is therefore likely that CTF ⌬9 might be generated by an alternative presenilinase cleavage. However, this fragment was overlooked so far since it is completely removed by proteolysis. CTF ⌬9 can probably not be incorporated into a FIG. 4. The caspase-generated CTF 10 is predominantly degraded by a cysteine proteinase. A, untransfected K293 cells were treated with and without the indicated inhibitors as described in Fig.  1C. CTF 10 was detected by reprobing the blot shown in Fig. 1C with antibody 3027. LLnL and leupeptin stabilize large amounts of CTF 10 , whereas the proteasome-specific inhibitor lactacystin has less effect. The activity of proteinase inhibitors within these samples was confirmed (see Fig. 1C). B, titration of the leupeptin concentration required for the inhibition of CTF 10 degradation. C, the cysteine proteinase inhibitor E64, which does not inhibit proteasomal degradation, efficiently blocks degradation of CTF 10 . MDL28170 inhibits degradation of CTF 10 as well. CTF 10 was isolated and detected as described in A. D, HeLa cells were treated with and without the indicated inhibitors for 18 h. CTF 10 was isolated and detected as described (A). Under control conditions HeLa cells contain high levels of CTF 10 . However, the levels of CTF 10 can be further increased upon treatment with leupeptin. Staurosporine was used to induce apoptosis and therefore the production of CTF 10 (31)(32)(33). Lower panel, treatment of HeLa cells with the inhibitors used in the upper panel does not induce apoptosis as monitored by the lack of PARP cleavage. Staurosporine was used as a positive control to induce apoptosis and cleavage of PARP (31)(32)(33). E, inhibition of de novo synthesis of CTF 10 still allows its stabilization upon parallel treatment with leupeptin. HeLa cells were treated for 24 h with Z-VAD alone to inhibit caspase-mediated generation of CTF 10 . Simultaneous incubations of Z-VAD with leupeptin, but not with lactacystin, inhibited degradation of CTF 10 . * indicates a C-terminal fragment, which is, like CTF 10 , specifically detected after stabilization by inhibiting its degradation or its enhanced production upon induction of apoptosis. This band most likely represents a CTF that is generated by additional caspase cleavage at an aspartic acid residue N-terminal to the major caspase cleavage site at amino acid 345. The activity of lactacystin was monitored by the detection of high molecular weight derivatives of ␤-catenin (arrowhead; lower panel).

FIG. 5. Characterization of a CTF generated by an alternative cleavage of PS1⌬exon9. A, kidney 293 cells stably transfected with
PS1⌬exon9 were treated with and without the indicated proteinase inhibitors as described in Fig. 1C. Cell lysates were immunoprecipitated with antibody 3027. CTFs were detected by immunoblotting using the same antibody (upper panel). In addition to CTF 10 , a CTF of higher molecular weight (CTF ⌬9 ) is detected. This fragment is exclusively stabilized by proteasome inhibitors but not by leupeptin. * indicates a C-terminal fragment, which is, like CTF 10 , specifically detected after stabilization by inhibiting its degradation or its enhanced production upon induction of apoptosis. This band most likely represents a CTF that is generated by additional caspase cleavage at an aspartic acid residue N-terminal to the major caspase cleavage site at amino acid 345. Upon reprobing of the same blot with antibody APS18 only CTF ⌬9 is detected (lower panel). ** indicates the position of CTF 10 not detectable with APS18. B, comparison of the molecular mass of CTF ⌬9 and CTF 20 . CTF 20 was isolated from untransfected K293 cells treated with the indicated inhibitors for 24 h, and CTF ⌬9 was isolated from K293 cells stably expressing PS1⌬exon9 as described in A. The fragments were detected using the monoclonal antibody APS18. As expected from the results shown in Fig. 1C, LLnL and lactacystin do not affect levels of CTF 20 . Note that expression of PS1⌬exon9 inhibits the formation of the endogenous CTF 20 (23,42,60). PS1 complex, since our previous work demonstrated that the PS1⌬exon9 holoprotein substitutes the heterodimeric complex and forms a complex by itself (28). CTF ⌬9 might thus behave like a "free" fragment (similar to NTF 298 ) that is subjected to rapid proteolysis by the proteasome. DISCUSSION The observation that many PS1 mutations cluster close to the cleavage sites of the presenilinase (3,4) indicates that proteolytic processing of PS1 might be very important for its physiological and pathological function. Moreover, fragment formation appears to be highly regulated. Overexpression of PS1 inhibits synthesis of endogenous PS1 and the homologous PS2 fragments (23,35). Furthermore, overexpression of PS proteins does not result in a linear increase of fragment formation. Instead, endogenous fragments are replaced by the transgene-derived fragments at approximately the same levels (23,35). Although the available data regarding the effect of mutations on PS processing are quite controversial (see Refs. 24,37,and 38; for review see Ref. 39), it appears that proteolytic processing of PS proteins might be disturbed by FAD-associated mutations. Cell type specificity may be responsible for these controversial observations; however, in the majority of cell lines including primary neurons and in all human tissues analyzed, PS proteins are proteolytically processed in the pres-ence and absence of FAD mutations (4). Interestingly, FADassociated mutations also affect proteolytic processing via the caspase-mediated pathway. In that case the Volga-German mutation causes an increased alternative cleavage of PS2 by a member of the caspase superfamily (31). Together with the highly regulated fragment formation, these data indicate that subtle changes in the balanced fragment formation may result in early onset AD. One would therefore expect physiological control mechanisms that prevent aberrant accumulation of PS fragments and/or its precursor. We therefore searched for proteolytic enzymes that are involved in the degradation of the PS holoprotein as well as its proteolytically processed fragments. PS fragments were found to be very stable when they are generated by proteolytic processing from the holoprotein (Fig.  6). This might be explained by the recent finding that PS fragments bind to each other to form a heterodimeric complex ( Fig. 6; see Refs. 28 -30 and 36). Upon complex formation, PS fragments are then either transported to a compartment that protects them from proteolytic degradation or undergo structural changes that make the dimer resistant to proteolysis. In that regard it is interesting to note that NTF 298 is highly unstable (data shown in this work; see also Refs. 44 and 60). This is due to the rapid removal by the proteasomal degradation pathway (Fig. 6). The sensitivity of NTF 298 to proteolysis FIG. 6. Schematic summary of the proteolytic processing and degradation pathways. PS proteins that form a complex (shaded box) are stable, whereas all PS molecules not involved in complex formation are rapidly removed. For details see "Discussion." can be explained by its inability to be incorporated into the PS1 heterodimeric complex (Fig. 6) as demonstrated by the lack of co-immunoprecipitation of NTF 298 with the PS1 heterodimeric complex. Rapid proteolysis of such free fragments is further supported by the proteasomal degradation of CTF ⌬9 which is most likely derived from an alternative presenilinase cleavage of the PS1⌬exon9 holoprotein (which forms a complex by itself; see Ref. 28) and thus represents a free CTF (Fig. 6). In the presenilinase pathway PS proteins are cleaved within the domain encoded by exon 9 of PS1 (23). The cleavage is heterogeneous and occurs after amino acids 291, 292, and 298 (26). The heterogeneous cleavage of PS1 might suggest the involvement of an endoproteinase with relaxed sequence specificity. This is further supported by our surprising finding that PS1⌬exon9 can undergo a presenilinase-like cleavage, although the original cleavage site within the domain encoded by exon 9 is absent. Relaxed sequence specificity is also observed for other proteinases, such as the ␣-secretase (61). The cleavage site of ␣-secretase appears to be determined by its distance to the cell membrane (61), a characteristic phenomenon that might be very similar to the PS-cleaving enzyme. On the other hand, the observed cleavage of PS1⌬exon9 might also represent a less frequently used cleavage site. CTF ⌬9 , in contrast to the uncleaved PS1⌬exon9 protein (28), is not incorporated into the complex and therefore is sensitive to proteolytic degradation like NTF 298 . If we assume that CTF ⌬9 represents a free CTF 20 , it appears that both PS fragments are rapidly removed by the proteasome as soon as they are not incorporated into the PS1 complex. Interestingly, both fragments are not attacked by the leupeptin-sensitive cysteine proteinase that degrades CTF 10 (see below), demonstrating differential degradation of PS-derived peptides.
The stable CTF 20 that is incorporated into the heterodimeric PS complex can be proteolytically attacked by caspases (data presented in this work and see Ref. 33). Caspases cleave CTF 20 and generate the alternative CTF 10 (Fig. 6). CTF 10 is then predominantly removed by a cysteine proteinase activity. This proteinase can be inhibited by LLnL, MDL28170, E64, and leupeptin but not by NH 4 Cl. Furthermore, treatment with the proteasome-specific inhibitor, lactacystin, only stabilizes the caspase-generated fragment to a minor extent. Therefore, it is likely that CTF 10 is degraded by a calpain-like cysteine proteinase.
PS1 holoproteins (wt as well as PS1⌬exon9) are rapidly degraded by a pathway that is sensitive to inhibitors of the proteasome similar to the PS2 holoprotein (31). Moreover, inhibition of the proteasome results in the accumulation of polyubiquitinated PS1 holoproteins (data not shown; see Ref. 55). As previously observed for the cystic fibrosis transmembrane conductance regulator protein (62), inhibition of proteasomal degradation did only partially prevent the degradation of the PS1 holoproteins. Thus, it is possible that an additional proteinase activity that is not inhibited by any of the compounds used in this study is involved in the degradation of PS1 holoproteins. This proteinase activity might be related to the recently identified proteolytic system that can compensate for the loss of proteasome function (63).
In addition to Ratovitski et al. (60), who demonstrated a long-lived PS1⌬exon9 pool, we show that most of the de novo synthesized PS1⌬exon9 is rapidly removed. This might again be due to the fact that excess amounts of PS1⌬exon9 do not assemble into the PS1⌬exon9 complex characterized previously (28). Thus, uncomplexed PS1⌬exon9 holoprotein is probably unstable like the PS1 holoprotein that is not incorporated into the PS complex (28).
Interestingly, a mutant NTF 298 does not cause an increased production of A␤42 even after its stabilization by inhibition of the proteasome. Furthermore, we have previously shown that a highly similar recombinant NTF is not sufficient to rescue the phenotype of a mutant PS-homologue (sel-12) in Caenorhabditis elegans (64). This suggests that either the PS heterodimeric complex or the full-length protein is required for the pathological and biological activity. However, in respect to the extremely low amounts of the full-length protein and its instability, we find it more likely that the heterodimeric complex (28 -30, 36) is biologically and pathologically active. The fact that PS1⌬exon9 is pathologically and biologically active as an uncleaved protein is explained by the recent finding that this protein but not the wt PS1 holoprotein forms a PS complex by itself (28). The recently described PS1 ⌬exon 4 mutation (65), which results in the production of an N-terminally truncated PS1 molecule, also does not cause the enhanced production of A␤42. 3 In addition, we have recently demonstrated that Nterminally truncated PS2 species derived from alternative splicing do not induce A␤42 generation as well (66). These data again support our hypothesis that the heterodimeric complex or the full-length protein is required for the pathological activity. Taken together, our data demonstrate that in addition to the highly balanced fragment formation, several proteolytic systems including the proteasome, caspases, and a leupeptin-sensitive cysteine proteinase degrade PS to avoid aberrant accumulation of excess amounts of PS fragments and holoprotein (Fig. 6). Efficient removal of the PS holoprotein and its proteolytic fragments might be very important to avoid aberrant A␤ generation specifically in the light of the recent findings that the lack of PS1 expression leads to significant reduction of A␤40 and A␤42 generation (21). The pathological mechanism of FAD-associated mutations might therefore be closely related to the half-life of the PS precursor or its proteolytic fragments. Future work on PS processing should therefore carefully determine whether FAD-associated PS mutations exert their pathological activity due to subtle changes in PS stability.