The proteolytic fragments of the Alzheimer's disease-associated presenilin-1 form heterodimers and occur as a 100-150-kDa molecular mass complex.

Mutations in the presenilin (PS) genes are linked to early onset familial Alzheimer's disease (FAD). PS-1 proteins are proteolytically processed by an unknown protease to two stable fragments of approximately 30 kDa (N-terminal fragment (NTF)) and approximately 20 kDa (C-terminal fragment (CTF)) (Thinakaran, G., Borchelt, D. R., Lee, M. K., Slunt, H. H., Spitzer, L., Kim, G., Ratovitsky, T., Davenport, F., Nordstedt, C., Seeger, M., Hardy, J., Levey, A. I., Gandy, S. E., Jenkins, N. A., Copeland, N. G., Price, D. L., and Sisodia, S. S. (1996) Neuron 17, 181-190). Here we show that the CTF and NTF of PS-1 bind to each other. Fractionating proteins from 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonic acid-extracted membrane preparations by velocity sedimentation reveal a high molecular mass SDS and Triton X-100-sensitive complex of approximately 100-150 kDa. To prove if both proteolytic fragments of PS-1 are bound to the same complex, we performed co-immunoprecipitations using multiple antibodies specific to the CTF and NTF of PS-1. These experiments revealed that both fragments of PS-1 occur as a tightly bound non-covalent complex. Upon overexpression, unclipped wild type PS-1 sediments at a lower molecular weight in glycerol velocity gradients than the endogenous fragments. In contrast, the non-cleavable, FAD-associated PS-1 Deltaexon 9 sediments at a molecular weight similar to that observed for the endogenous proteolytic fragments. This result may indicate that the Deltaexon 9 mutation generates a mutant protein that exhibits biophysical properties similar to the naturally occurring PS-1 fragments. This could explain the surprising finding that the Deltaexon 9 mutation is functionally active, although it cannot be proteolytically processed (Baumeister, R., Leimer, U., Zweckbronner, I., Jakubek, C., Grünberg, J., and Haass, C. (1997) Genes & Function 1, 149-159; Levitan, D., Doyle, T., Brousseau, D., Lee, M., Thinakaran, G., Slunt, H., Sisodia, S., and Greenwald, I. (1996) Proc. Natl. Acad. Sci. U. S. A. 93, 14940-14944). Formation of a high molecular weight complex of PS-1 composed of both endogenous PS-1 fragments may also explain the recent finding that FAD-associated mutations within the N-terminal portion of PS-1 result in the hyperaccumulation not only of the NTF but also of the CTF (Lee, M. K., Borchelt, D. R., Kim, G., Thinakaran, G., Slunt, H. H., Ratovitski, T., Martin, L. J., Kittur, A., Gandy, S., Levey, A. I., Jenkins, N., Copeland, N., Price, D. L., and Sisodia, S. S. (1997) Nat. Med. 3, 756-760). Moreover, these results provide a model to understand the highly regulated expression and processing of PS proteins.

Most cases of Alzheimer's disease (AD) 1 occur sporadically, with a strong increase in risk during aging (for review see Ref. 5). However, in at least 10 -15% of cases, autosomal dominant mutations have been found to cause early onset familial AD (FAD). Mutations in three genes are known so far to cause FAD. Mutations within the gene encoding the ␤-amyloid precursor protein (␤APP) all cause the enhanced production of the 42-amino acid version of the amyloid ␤-peptide (A␤42; see Ref. 6; for review, see Ref. 5). A␤42 is a major component of amyloid plaques (7), which are the pathological hallmark of the disease (5). A␤42 exhibits enhanced neurotoxicity, which might be due to its increased ability to form insoluble fibers (8,9). Increased production of A␤42 (10 -15) was also found to result from the much more common mutations within the presenilin (PS) genes (16 -18). Since mutations within the PS genes are responsible for many FAD cases, the analysis of the cellular biology of these proteins will undoubtedly lead to a better understanding of the molecular mechanisms involved in AD (19).
Based on the results from genetic rescue experiments of the mutant PS homologue (the sel-12 gene (23)) in Caenorhabditis elegans and gene deletions in mice, PS-1 is most likely involved in cell fate decisions via the Notch signaling pathway (2,3,24,25). In C. elegans, wild type human PS-1 and PS-2 were found to rescue all aspects of the sel-12 mutant phenotype (2,3). The sel-12 gene is known to facilitate Notch signaling (23); there-fore, the results from the rescue experiments strongly suggest that human PS proteins play an important role in the Notch signaling cascade of vertebrates as well. This conclusion was further supported when knock-outs of the PS-1 gene were generated in mice, since the loss of PS-1 expression resulted in a phenotype reminiscent of Notch knock-outs (24,25). Surprisingly, all FAD-associated point mutations of PS-1 tested so far exhibited a strongly reduced ability to rescue the sel-12 mutant phenotype in C. elegans (2,3), indicating that theses mutations might change functionally important amino acids (23). This notion is supported by the finding that FAD-associated mutations occur at positions that are highly conserved during evolution in all PS genes analyzed so far (for review, see Ref. 21). In contrast, expression of PS-1 lacking exon 9 due to a naturally occurring FAD causing splicing mutation (26) rescued the sel-12 mutant phenotype surprisingly well (Refs. 2 and 3; also see below).
Interestingly, PS proteins have been found to occur predominantly as stable C-terminal and N-terminal fragments (CTF and NTF; see Fig. 1), whereas only low levels of unclipped PS holoprotein can be detected within all cell lines and tissues analyzed to date (1,27,28). Surprisingly, mutations within the TM domains of the PS-1 NTF appear to result not only in the hyperaccumulation of the NTF by itself but also in the accumulation of the complementary CTF (4). Because it is highly unlikely that mutations that occur far away from the cleavage site (28) of PS-1 directly influence the rate of cleavage, other mechanisms allowing the accumulation of both fragments in a stoichiometrically regulated manner must be considered. In this regard, it is interesting to note that overexpression of PS proteins does not result in a linear increase of fragment formation (1,28). Moreover, expression of the ⌬exon 9 mutation (26), which is known to inhibit conventional proteolytic processing of PS-1, also markedly decreases the formation of the endogenous PS fragments (1,4,29). These results indicate a highly regulated mechanism that allows the accumulation of only certain levels of both fragments. Any disturbance in this regulation, such as hyperaccumulation of the PS fragments or of the unclipped PS-1 ⌬exon 9 protein (4), appears to be associated with early onset FAD, probably due to the enhanced production of A␤42. However, nothing is known about the nature of the regulation mechanism. We have therefore analyzed whether PS-1 fragments interact with each other. We find that the NTF of PS-1 co-immunoprecipitates with the CTF. Moreover, both fragments form a 100 -150-kDa complex in untransfected cells. Binding of PS fragments might therefore explain their concomitant accumulation in transgenic animals expressing mutations within the N-terminal portion of the PS protein. These results might also suggest that the highly regulated fragment formation could be due to the formation of a stoichiometric high molecular weight complex.
Isolation of Membrane Proteins-Kidney 293 cells were grown to confluence. Cells of 10 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 protease inhibitors as described (30) and homogenized with 30 strokes in a glass Dounce homogenizer. 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 DTT) containing protease inhibitors (30). 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 DTT, containing protease inhibitors as described (30)) for 1 h on ice. Alternatively, proteins were also extracted with 1% Triton X-100 or 0.5% SDS in HEPES buffer. For SDS extraction, K ϩ was substituted by Na ϩ in the HEPES buffer. 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.
Glycerol Velocity Gradients-Glycerol velocity gradient centrifugation was performed as described by Hay et al. (31). Briefly, 1-2 mg of membrane proteins were loaded on a linear 5-25% (v/v) glycerol velocity gradient (31) in gradient buffer (100 mM KCl, 20 mM HEPES, pH 7.2, 2 mM EGTA, 2 mM EDTA, 2 mM DTT, 0.2% CHAPS). Gradients were centrifuged at 40,000 rpm for 16 h at 4°C in a SW40 rotor (Beckman L-70 ultracentrifuge). After centrifugation, 13 fractions of 1 ml were collected from bottom to top. Proteins were precipitated with an equal volume of 20% trichloroacetic acid, and the precipitated proteins were washed with 90% acetone. Protein pellets were solubilized in sample buffer containing 4 M urea and incubated for 10 min at 65°C (32). Proteins were separated on SDS-urea gels (32) and transferred to polyvinylidene difluoride membranes.
Antibodies-The polyclonal antibodies, 2953 and 3027, used in this study were described previously (29,32). The monoclonal antibody PS1N to the N terminus of PS-1 is described by Capell et al. (33). The monoclonal antibody APS 18 to the large loop of PS-1 was raised to a peptide corresponding to amino acids 314 -334 of PS-1. Epitopes of all antibodies used are indicated in Fig. 1.
Immunoblotting-Immunoblotting was carried out as described (32). Bound antibodies were detected by enhanced chemiluminescence (Amersham Corp.) or the ECL-PLUS system (Amersham Corp.).
Co-immunoprecipitations-For co-immunoprecipitation, membranes were prepared as described above and extracted either with 2% CHAPS, 1% Triton X-100, or 0.5% SDS. For SDS extraction, K ϩ was substituted by Na ϩ in the HEPES buffer. To remove undissolved membrane fragments, the extracts were pelleted by ultracentrifugation for 1 h at 100,000 ϫ g at 4°C. Incubation with PS-1 antibodies was performed as described (29,32). SDS extracts were diluted 10 ϫ prior to antibody addition; CHAPS and Triton X-100 extracts were immunoprecipitated without further dilution. Immunoprecipitations of CHAPS-extracted proteins were washed 4 ϫ for 20 min in CHAPS washing buffer (0.5% CHAPS, 200 mM NaCl, 50 mM HEPES, pH 7.6). Immunoprecipitations of SDS-extracted proteins were washed as described (30). Immunoprecipitations of Triton X-100-extracted proteins were washed 4 ϫ for 20 min in STEN buffer only (30).

RESULTS
To determine whether PS-1 fragments occur as a complex, we analyzed membrane protein fractions from human K293 cells. In most experiments, we specifically used untransfected K293 cells to allow the analysis of endogenous PS proteins under in vivo conditions. Moreover, this cell line is highly appropriate for the analysis of the biochemistry of the FADassociated proteins (␤APP and PS-1/PS-2), since ␤APP metabolism and the effects of ␤APP and PS mutations originally sorted out in K293 cells (13,30,34,35) and other peripheral cell lines such as COS (15) and CHO (14) were completely confirmed in neuronal cells, primary cell cultures, human and mouse brain tissue, cerebrospinal fluid, and plasma (10 -13, 15, 35).
Identification of a Presenilin Complex-Membrane preparations from K293 cells were extracted with CHAPS, and the proteins were separated on a continuous 5-25% glycerol velocity gradient as described previously (31). To determine the apparent molecular weight of isolated proteins, molecular mass markers of 29 -205 kDa were separated on parallel gradients ( Fig. 2A). Gradients were fractionated and proteins precipitated with trichloroacetic acid. An aliquot of each fraction was then analyzed by immunoblotting using antibodies to the C and N termini of PS-1 (epitopes of all antibodies used are indicated in Fig. 1). Antibody 3027 to the large loop of PS-1 (29) detected a prominent, approximately 20-kDa CTF (Fig. 2C). In addition to the 20-kDa CTF, we also consistently found smaller amounts of a CTF of 23 kDa (labeled with an asterisk in Fig. 2C), which corresponds to the previously described protein kinase C/protein kinase A-phosphorylated form of the PS-1 20-kDa CTF (29,36). The majority of the CTF accumulated at an apparent molecular mass of approximately 100 -150 kDa in fractions 7-11 of the glycerol velocity gradient (Fig. 2, A-C). It should be noted that the sedimentation velocity not only depends on the molecular weight but also on the density of the complex. Therefore, we cannot rule out that the PS-1 complex could be of much higher molecular weight in vivo. In addition to a major peak at 100 -150 kDa, we also detected slightly variable minor amounts of the CTF in the first fractions of the low molecular weight range (Fig. 2, B and C). It should be noted that in all experiments no PS-1 fragments were detected in the pellet of the glycerol velocity gradients.
Aliquots from the same gradient were then analyzed using the monoclonal antibody PS1N to the N terminus of PS-1 (33). Like the CTF, the majority of the NTF accumulated in fractions 7-11, corresponding to a molecular mass of approximately 100 -150 kDa (Fig. 2, B-D). Much smaller amounts of the NTF were also detected in the first (low molecular weight) fractions of the gradient (Fig. 2, B and D). The detection of multiple NTF bands is consistent with data reported previously (28) and is most likely due to the heterogeneous cleavage of PS-1 (28). Taken together, these results indicate that both the CTF and the NTF of PS-1 occur predominantly as a 100 -150-kDa complex.
Similar experiments to those in Fig. 2 were also carried out using Triton X-100 as the detergent to extract membrane proteins. Again, co-sedimentation of both fragments was observed in glycerol velocity gradients (data not shown). However, extraction with 1% Triton X-100 led to a disassembly of the 100 -150-kDa complex, as indicated by large amounts of the CTF within the low molecular weight range and very small amounts in the high molecular weight fraction (Fig. 3, A and  B). We further treated membrane preparations with 0.5% SDS and 1% Triton X-100 and separated the extracted proteins on a glycerol velocity gradient in the presence of 0.05% SDS. This resulted in a complete disassembly of the high molecular weight PS complex (Fig. 3, C and D), as indicated by the almost exclusive detection of the CTF within the low molecular weight fractions. These experiments therefore demonstrate that the molecular interactions necessary for complex formation are sensitive to SDS and Triton X-100.
To assess whether intermolecular covalent binding by disulfide bridges is involved in the formation of the 100 -150-kDa complex, we extracted membrane proteins in 2% CHAPS and increased the concentration of DTT to 10 mM, followed by sedimentation on glycerol velocity gradients containing 10 mM DTT as well. Under reducing conditions, no change in the sedimentation profile of both PS species was observed. Again, the CTF was predominantly observed as a 100 -150-kDa complex, with a second smaller peak in the low molecular weight range (Fig. 4, A and B). These data indicate that disulfide bridges are not involved in PS complex formation.
Co-immunoprecipitation of the NTF and CTF of PS-1-The co-sedimentation of both fragments within the same fractions raised the possibility that the PS fragments interact with each other. To prove this is indeed the case, we performed co-immu- The sedimentation of molecular mass markers is shown below the concentration profiles. B, sedimentation profile of the PS-1 NTF and CTF in a linear 5-25% glycerol velocity gradient. Shorter exposures of the immunoblots shown in C and D were scanned, and the relative amounts of the corresponding protein are shown. C, sedimentation of the PS-1 CTF in a 5-25% glycerol velocity gradient. The endogenous PS-1 CTF was detected with antibody 3027. *, CTF phosphorylated by protein kinase C/protein kinase A (29,36). D, sedimentation of the PS-1 NTF in a 5-25% glycerol velocity gradient. The endogenous PS-1 NTF was detected with antibody PS1N. noprecipitation studies. Isolated membranes from K293 cells were extracted with 2% CHAPS. Identical aliquots of the extract were immunoprecipitated either with antibody 2953 to the N terminus of PS-1 or antibody 3027 to the C terminus of PS-1. Immunoprecipitates were separated on 11% SDS-urea gels and immunoblotted with the monoclonal antibody, APS18, to the large loop of PS-1. Interestingly, the monoclonal antibody detected the CTF not only after immunoprecipitation with antibody 3027 but also after immunoprecipitation with antibody 2953 to the N terminus (Fig. 5A). This finding clearly indicates that the NTF interacts with the CTF under in vivo conditions in untransfected cells. To confirm this more rigorously, the converse experiment was performed. CHAPS-extracted membrane preparations were immunoprecipitated with the same antibodies but immunoblotted with the monoclonal antibody PS1N to the N-terminal domain of PS-1. Again, co-immunoprecipitation of both PS fragments was observed, as indicated by the detection of the NTF after immunoprecipitation with antibody 3027 to the large loop of PS-1 (Fig. 5B). Taken together, these data indicate that the NTF and CTF of PS-1 interact in untransfected cells.
To determine whether the NTF/CTF interaction is Triton X-100-and SDS-stable, we extracted membrane preparations with 1% Triton X-100 or 0.5% SDS. The extracted proteins were then immunoprecipitated using antibody 2953 to the N terminus and 3027 to the C terminus of PS-1. Precipitated PS fragments were detected by immunoblotting with the monoclonal antibodies APS18 (to the large loop) or PS1N (to the N-terminal domain). By using either SDS or Triton X-100 as a detergent, no co-precipitation of PS fragments was detected (Fig. 5C). This result indicates that the interactions between the CTF and NTF are SDS-and Triton X-100-labile. It also confirms the previous experiments that showed that SDS and Triton X-100 obviate the formation of the 100 -150-kDa complex (Fig. 3).
We next performed similar co-immunoprecipitations on the low molecular weight fraction (fractions 1-3 in Fig. 2) of 5-25% glycerol velocity gradients. Immunoprecipitation of the CTF allowed the detection of the co-precipitating NTF by immunoblotting with the monoclonal antibody PS1N to the N terminus of PS-1 (Fig. 5D). Therefore, these results suggest that PS-1 fragments in the low molecular weight fraction exist as heterodimers. In view of the evidence that PS-1 fragments form high molecular weight complexes (Fig. 2), we also co-immunoprecipitated PS-1 fragments from the high molecular weight gradient fraction described above (fractions 7-11 in Fig. 2). Aliquots of the peak fractions were immunoprecipitated with antibodies 2953 and 3027 as described. After separating the immunoprecipitates on 11% SDS-urea gels, PS fragments were detected using the monoclonal antibody PS1N to the N terminus of PS-1. Again, immunoprecipitation with a polyclonal antibody to the C terminus of PS-1 (3027) allowed the detection

FIG. 3. The 100 -150-kDa complex of the PS-1 N-terminal and C-terminal fragment is Triton X-100-and SDS-sensitive.
A, isolated membranes from K293 cells were extracted with 1% Triton X-100 and centrifuged on a 5-25% glycerol velocity gradient. The relative distribution of the PS-1 CTF upon Triton X-100 or CHAPS extraction is shown (similar data were obtained for the NTF; data not shown). For scanning, shorter exposures of the immunoblot in B were used. The sedimentation of molecular mass markers is shown below. B, sedimentation of the PS-1 CTF in a 5-25% glycerol velocity gradient upon extraction of membrane proteins with Triton X-100. The endogenous PS-1 CTF was detected with antibody 3027. *, CTF phosphorylated by protein kinase C/protein kinase A (29,36). C, isolated membranes from K293 cells were extracted with 0.5% SDS and centrifuged on a 5-25% glycerol velocity gradient. The sedimentation profile of the PS-1 CTF upon SDS or CHAPS extraction is shown (similar data were obtained for the NTF; data not shown). For scanning, shorter exposures of the immunoblot in D were used. D, sedimentation of the PS-1 CTF in a 5-25% glycerol velocity gradient upon extraction of membrane proteins with SDS. The endogenous PS-1 CTF was detected with antibody 3027. *, CTF phosphorylated by protein kinase C/protein kinase A (29,36). 4. The 100 -150-kDa complex is stable under reducing conditions. A, CHAPS-extracted membrane proteins were separated on a 5-25% glycerol velocity gradient in the presence of 10 mM DTT. The sedimentation profile of the CTF in control gradients and DTT containing gradients is shown (similar results were obtained for the NTF; data not shown). The amount of the CTF in the low molecular fractions is slightly variable; however, the majority of the CTF is detected within the high molecular weight fraction. For scanning, shorter exposures of the immunoblot in B were used. The sedimentation of molecular mass markers is shown below the sedimentation profile. B, sedimentation of the PS-1 CTF in a 5-25% glycerol velocity gradient in the presence of DTT. The endogenous PS-1 CTF was detected with antibody 3027. *, CTF phosphorylated by protein kinase C/protein kinase A (29,36). of co-precipitating NTFs by immunoblotting with the monoclonal antibody PS1N (Fig. 5E), showing directly that PS fragments within the 100 -150-kDa complex interact with each other. Therefore, these data suggest that the NTF/CTF heterodimers observed in the low molecular weight fractions could either form oligomers and/or interact with so far unknown binding proteins, thus forming a high molecular weight complex.

Mutant PS-1 with the ⌬Exon 9 Deletion but Not Wild Type Full-length PS-1 Forms a Complex Similar to the Endogenous
Fragments-Consistent with previously reported results (1, 29), we could not detect the endogenous, unclipped full-length form of PS-1 in untransfected K293 cells (data not shown). To determine if unclipped PS-1 can participate in PS complex formation, we analyzed membrane preparations from K293 cells stably overexpressing wt PS-1 (13) on the 5-25% glycerol velocity gradients. Although the PS-1 fragments were predominantly detected in fractions 7-11 (see above), the unclipped PS-1 holoprotein sedimented at a lower molecular weight range in fractions 3-8 (Fig. 6, A and B).
We next expressed the naturally occurring FAD-associated ⌬exon 9 mutation (26) and analyzed the sedimentation of this truncated, unclipped mutant PS-1 in the glycerol velocity gradients. Surprisingly, we found that this uncleavable form of PS-1 accumulated in high molecular fractions (fractions 7-12 in Fig. 6, A and C). It therefore appears that PS-1 with the exon 9 deletion is able to form a 100 -150-kDa complex similar to the endogenous PS-1 fragments.

FIG. 5. The CTF and the NTF of PS-1 interact in a Triton X-100and SDS-sensitive manner.
A, co-immunoprecipitation of the PS-1 CTF. CHAPS-extracted membranes were immunoprecipitated with antibody 2953 (to the N terminus of PS-1) or antibody 3027 (to the large loop of PS-1). The precipitates were immunoblotted with the monoclonal antibody APS18 (to the large loop of PS-1). Note that APS18 detects the PS-1 CTF after immunoprecipitation with antibody 2953 (to the N-terminal domain of PS-1). *, IgG. B, co-immunoprecipitation of the PS-1 NTF. CHAPS-extracted membranes were immunoprecipitated with antibody 2953 (to the N terminus of PS-1) or antibody 3027 (to the large loop of PS-1). The precipitates were immunoblotted with the monoclonal antibody PS1N (to the N-terminal domain of PS-1). Note that antibody PS1N detects the PS-1 NTF after immunoprecipitation with antibody 3027 (to the large loop of PS-1). *, IgG. C, binding of PS fragments is Triton X-100-and SDS-sensitive. Isolated cell membranes were extracted either with 1% Triton X-100 or 0.5% SDS. The extracted proteins were immunoprecipitated with antibody 2953 (to the N-terminal domain of PS-1) or antibody 3027 (to the large loop of PS-1). PS-1 fragments were detected by immunoblotting with antibodies APS18 (to the large loop of PS-1) or antibody PS1N (to the N-terminal domain of PS-1). Note that antibody 2953 and 3027 did not co-immunoprecipitate the CTF or NTF of PS-1, respectively. D, PS-1 fragments co-immunoprecipitate from the low molecular weight fractions of 5-25% glycerol velocity gradients. Fractions 1-3 (low molecular weight) of a 5-25% glycerol velocity gradient were pooled and immunoprecipitated with antibodies 2953 and 3027. Pooled fractions from three gradients (3x) were used since the low molecular weight fractions contain very small amounts of PS-1 fragments. PS-1 fragments were detected by immunoblotting using the monoclonal antibody PS1N. Note that antibody 3027 to the large loop of PS-1 co-immunoprecipitated the NTF of PS-1 from the low molecular weight fraction. E, PS-1 fragments co-immunoprecipitate from the high molecular weight fractions of 5-25% glycerol velocity gradients. Fractions 7-11 (high molecular weight) of one (1x) 5-25% glycerol velocity gradient were pooled and immunoprecipitated with antibodies 2953 and 3027. PS-1 fragments were detected by immunoblotting using the monoclonal antibody PS1N. Note that antibody 3027 to the large loop of PS-1 co-immunoprecipitated the NTF of PS-1 from the high molecular weight fraction.

FIG. 6. Mutant PS-1 lacking exon 9 but not wt full-length PS-1 forms a complex similar to endogenous PS-1 fragments.
A, sedimentation profile of wt full-length PS-1 and mutant PS-1 lacking exon 9 in a linear 5-25% glycerol velocity gradient. Shorter exposures of the immunoblots shown in B and C were scanned, and the relative amounts of the corresponding protein are shown. The sedimentation of molecular mass markers is shown below the sedimentation profile. B, sedimentation of wt full-length PS-1 in a 5-25% glycerol velocity gradient. Unclipped PS-1 was detected with antibody PS1N. C, sedimentation of mutant PS-1 lacking exon 9 in a 5-25% glycerol velocity gradient. Unclipped mutant PS-1 ⌬exon 9 was detected with antibody PS1N.

DISCUSSION
The data presented here strongly indicate that endogenous PS-1 fragments interact with each other and form a 100 -150-kDa complex in untransfected cells. Binding of PS-1 fragments is supported by multiple co-immunoprecipitation experiments using C-terminal antibodies for the detection of the NTF and N-terminal antibodies for the detection of the CTF. Co-migration of the NTF and CTF in glycerol velocity gradients supports these results. Moreover, both fragments were co-immunoprecipitated from the peak fractions of the glycerol velocity gradients. Binding of the PS fragments was also independently observed. 2 The separation of ␤APP from PS-1 fragments within the same glycerol velocity gradient (data not shown) rules against nonspecific aggregation of the endogenous PS-1 fragments. Furthermore, the overexpressed highly hydrophobic wt PS-1 holoprotein sediments at lower molecular weight than the endogenous fragments, whereas overexpression of ⌬exon 9 PS-1 resulted in a sedimentation more similar to that of the endogenous fragments. These results make it highly unlikely that the observed PS complexes are due to artifactual aggregation, because that would be expected to be more likely after overexpression of PS proteins (32). Moreover, our data are also consistent with the previous finding that the CTF of PS-1 may oligomerize (36). PS-1 fragments are bound most likely by non-covalent interactions. The sensitivity of the interaction to SDS and Triton but not DTT suggests that PS-1 fragments bind via hydrophobic interactions, making it likely that the transmembrane domains are involved.
PS proteins predominantly occur as proteolytic fragments in all tissues and cell lines analyzed so far (1,27,28). These fragments appear to play an important role in causing early onset FAD. The natural occurring ⌬exon 9 mutation (26), which abolished proteolytic cleavage (1), results in early onset FAD probably related to the accumulation of unclipped PS-1 molecules (1,4). Moreover, in transgenic mice, the expression of two mutations within the N-terminal domain of PS-1 can result in the hyperaccumulation of PS-1 fragments (4). It therefore appears that subtle changes in a normally highly balanced fragment formation could result in early onset FAD. To prevent excess fragment formation, a natural occurring control mechanism appears to inhibit any marked overproduction or accumulation of PS fragments. Upon overexpression of PS-1 or PS-2 in transfected cells or transgenic mice, only a very small increase of fragment amounts is observed (1). Interestingly, it was found that overexpression of PS-1 results in a replacement of the endogenous PS-1 fragments by the exogenous fragments (1). These findings indicate a highly regulated biological mechanism that leads to balanced fragment formation.
Based on our current results it appears possible that NTF⅐CTF complex formation could play a role in regulated fragment generation and accumulation. Binding of PS fragments could now explain the coordinate hyperaccumulation of PS fragments in mice expressing some mutant PS genes (4). The mutations used for the latter study occur within a TM domain far away from the recognition site of the PS cleaving enzyme in the large cytoplasmic loop. This makes it unlikely that the mutations influence the rate of precursor cleavage to cause a parallel accumulation of the CTF and the NTF. This apparent paradox may now be explained by our finding that the PS-1 fragments are bound to each other. If, for example, the turnover of a mutant NTF is slowed, the complementary wt CTF would also be stabilized. Binding of PS fragments to a high molecular weight complex might thus play a role in regulation of fragment formation. If one assumes a stoichiometrically defined PS complex containing PS fragments and perhaps other binding partners as well, only a limited number of PS heterodimers could bind. It might be possible that unbound PS fragments and full-length PS-1 are then rapidly removed by proteolytic degradation. This hypothesis is supported by recent findings demonstrating that multiple proteolytic pathways are involved in the degradation of full-length PS and its proteolytic fragments (Refs. 37-40; for review see Refs. 20 and 21).
Further purification of the PS-containing 100 -150-kDa complex might result in the identification of other binding proteins playing a functional role in the PS-1 complex. Attempts to identify metabolically labeled binding partners, which may co-immunoprecipitate with PS fragments, led to a very surprising result. Even after extended labeling periods, only very minor amounts of the de novo synthesized PS fragments assembled into the complex. Most of the de novo synthesized PS fragments were detected in the low molecular weight fractions of the glycerol velocity gradients. As described above the "free" fragments might be rapidly degraded, whereas PS fragments bound to the complex appear to be stable over very long periods. 3 Nevertheless, one of the most obvious binding proteins might be ␤APP. Indeed it was reported previously that PS-1 (41,42) and PS-2 (42) can bind to ␤APP. However, Thinakaran and Sisodia 2 could not confirm co-immunoprecipitation of PS-1/-2 with ␤APP. Separation of ␤APP from endogenous PS-1 fragments in glycerol velocity gradients (data not shown) might indicate that only very small amounts of ␤APP can bind to PS. However, this could still be in agreement with a transient binding of very small amounts of ␤APP to PS during its transport to the cell surface. Beside ␤APP, PS-2 itself might form a complex with PS-1. Other possible binding partners include members of the Notch signaling pathway, such as proteins of the Armadillo family (43). Due to the tissue-specific expression of proteins belonging to the Armadillo family (43), one might propose the presence of heterogeneous PS complexes containing a variety of different binding partners depending on the tissue analyzed. In that regard it is interesting to note that the Drosophila Notch protein undergoes proteolytic processing resulting in defined N-terminal and C-terminal fragments (44,45), and the fragments are tightly bound to each other. Moreover, only the proteolytic fragments appear to be biologically active in Notch signaling (44,45). It is therefore tempting to speculate that PS proteins also require proteolytic cleavage for their biological activity. Furthermore, PS proteins are believed to be involved in Notch signaling (2,3,(23)(24)(25), which could indicate a common mechanism of biological activation for at least some members of the Notch signaling pathway. However, how does one explain the surprising result that PS-1 with the ⌬exon 9 mutation rescues the mutant sel-12 phenotype well even though it cannot undergo proteolytic cleavage (2, 3)? One explanation could be that the unclipped ⌬exon 9 PS-1 can form a PS-1 complex similar to the endogenous fragments, whereas full-length wt PS-1 forms a lower molecular weight complex. The lack of the domain encoded by exon 9 might mimic a clipped PS-1 molecule, thus allowing complex formation. This is further supported by the finding that expression of PS-1 ⌬exon 9 causes a reduction of the endogenous PS fragment formation (1,4,29). This is also observed in the high molecular weight fraction of the glycerol velocity gradients (data not shown), thus indicating that PS-1 ⌬exon 9 behaves like the natural NTF⅐CTF complex.
Based on our data, we would therefore postulate a stoichiometrically defined PS complex to be required for the biological as well as the pathological functions of presenilins, a hypothe-sis supported by the recent finding that expression of a recombinant mutant NTF of PS-1 or PS-2 is not sufficient for overproduction of A␤42. 4,5