The Presenilin 1 Protein Is a Component of a High Molecular Weight Intracellular Complex That Contains β-Catenin*

The presenilin (PS) genes associated with Alzheimer disease encode polytopic transmembrane proteins which undergo physiologic endoproteolytic cleavage to generate stable NH2- and COOH-terminal fragments (NTF or CTF) which co-localize in intracellular membranes, but are tightly regulated in their stoichiometry and abundance. We have used linear glycerol velocity and discontinuous sucrose gradient analysis to investigate the distribution and native conformation of PS1 and PS2 during this regulated processing in cultured cells and in brain. The PS1 NTF and CTF co-localize in the endoplasmic reticulum (ER) and in the Golgi apparatus, where they are components of a ∼250-kDa complex. This complex also contains β-catenin but not β-amyloid precursor protein (APP). In contrast, the PS1 holoprotein precursor is predominantly localized to the rough ER and smooth ER, where it is a component of a ∼180-kDa native complex. PS2 forms similar but independent complexes. Restricted incorporation of the presenilin NTF and CTF along with a potentially functional ligand (β-catenin) into a multimeric complex in the ER and Golgi apparatus may provide an explanation for the regulated accumulation of the NTF and CTF.

Mutations in the genes encoding the presenilin (PS) 1 1 (PS1) and presenilin 2 (PS2) account for the majority of early-onset familial Alzheimer's disease (1)(2)(3). Both genes encode polytopic transmembrane proteins that are predominantly localized in intracellular membranes, including the nuclear envelope, the endoplasmic reticulum, and the Golgi apparatus (4,5). Structural studies suggested that these proteins contain either six or eight transmembrane (TM) domains, and that the amino and carboxyl termini as well as a large hydrophilic loop following the sixth TM domain are located in the cytoplasm (5)(6)(7). The presenilins undergo physiological endoproteolytic processing within the large cytoplasmic loop following TM6 by an unknown protease that produces heterogeneous ϳ29-kDa aminoterminal and ϳ18 -20-kDa carboxyl-terminal fragments (8,9). The presenilin holoproteins are maintained at low steady levels both in brain and in other peripheral tissues, probably by proteasome-mediated degradation (10,11). As a result, the presenilin species most readily detected are the stable NH 2and COOH-terminal endoproteolytic fragments, the stoichiometry of which appears to be tightly regulated (8).
The functional role of the presenilins is still unknown, although roles in cellular differentiation, in signal transduction, in apoptosis, or in intracellular protein trafficking have been proposed. To better understand the presenilin proteins and their biological functions, we have investigated both the native state of these proteins and their biochemical subcellular localization in membrane fractions derived from cultured cells and from the human brain. Our data suggest that the presenilins form detergent-sensitive high molecular mass complexes. The PS1 holoprotein is a component of a complex of ϳ180 kDa, while the endoproteolytic fragments are components of a ϳ250-kDa complex that contains ␤-catenin (a member of the armadillo protein family which may have both structural and signal transduction roles) but not ␤APP. In addition to forming complexes of differing sizes, the intracellular distribution of the holoprotein and the endoproteolytic fragments differ.

EXPERIMENTAL PROCEDURES
Protein Extraction from Cells-Untransfected HEK293 cells (10-cm dishes) or the transfected HEK293 cell line stably expressing wild type human PS1 and APP 695 (gift from Drs. D. Selkoe and M. Citron) were grown to confluence. Cells were washed twice with ice-cold phosphatebuffered saline and then lysed at 4°C for 30 min with 1.0 ml of lysis buffer containing either 1.0% digitonin (Sigma), 0.5% Triton X-100 (Boehringer Mannheim), 0.5% Nonidet P-40 (Boehringer Mannheim), or 0.5% SDS (Sigma) plus 25 mM Hepes, pH 7.2, 150 mM NaCl, 2 mM dithiothreitol, 2 mM EDTA, 5 g/ml each of chymostatin, pepstatin, leupeptin, and antipain. Other protease inhibitor mixtures have been tested but do not alter the results reported here. Insoluble material was removed from the lysates by centrifugation at 17,000 ϫ g for 15 min.
To prepare membrane extracts, cells were washed twice with ice-cold phosphate-buffered saline and resuspended in 10 mM Hepes, pH 7.2, containing protease inhibitors. After being swollen for 60 min, cells were subjected to a freeze-thaw cycle. The homogenate was clarified by centrifugation at 1000 ϫ g for 15 min at 4°C; the resulting supernatant was then centrifuged at 107,000 ϫ g for 1 h at 4°C, and the pellets were washed once with 10 mM Hepes, pH 7.2, containing protease inhibitors, followed by extraction with lysis buffer as above.
Human brain extracts were prepared by homogenizing 1.0 g of normal cerebral cortex in a total volume of 5.0 ml of 20 mM Hepes, pH 7.2, at 4°C with protease inhibitors using a Polytron tissue shredder and a Teflon homogenizer followed by rehomogenization in the presence of 1.0% digitonin or 0.5% Nonidet P-40. Samples were centrifuged at 11,000 ϫ g for 30 min, and the supernatant was used for glycerol gradient centrifugation.
Glycerol Velocity Gradient Centrifugation-The glycerol gradient centrifugation procedure was as described previously (12,13). Briefly, 0.5 ml of total protein extracts or membrane extracts was applied to the top of a 11.5-ml 10 -40% (w/v) linear glycerol gradient containing 25 mM Hepes, pH 7.2, 150 mM NaCl, and 0.4% appropriate detergent. Gradients were centrifuged for 15 h at 35,000 rpm and 4°C using an SW41 rotor and collected by upward displacement into 1.0-ml fractions using an Isco model 640 density gradient fractionator. To quantitatively recover proteins, 10 g of bovine serum albumin were added to 150-l aliquots of each fraction, followed by precipitation with 5 volumes of Ϫ20°C acetone.

Identification of a Stable 250-kDa Complex of PS1 Endoproteolytic Fragments-
To determine the native state of PS1 protein, untransfected HEK293 cells were solubilized in nondenaturing digitonin buffers, and the extracted proteins were fractionated on a linear glycerol velocity gradient. Both NTF and CTF of PS1 were found in identical fractions (fractions 3-7) with a peak in fractions 5 and 6 corresponding to an apparent molecular mass of 250 kDa (note that because the apparent molecular mass is also dependent upon relative buoyancy, it is likely that the size of a multimeric complex estimated from the glycerol gradient will be underestimated) (Fig. 1A). There was virtually no PS1 immunoreactivity in the fractions corresponding to molecular masses lower than 60 kDa even after prolonged exposure, implying that very little PS1 exists as stable monomers in the digitonin lysate. The NTF and CTF of PS1 may therefore be components of the same high molecular mass oligomeric complex in vivo.
To explore this hypothesis further, we examined the sensitivity of the complexes to different detergents. When total proteins from the untransfected HEK293 cells were extracted in 0.5% Triton X-100 and then subjected to the same gradient centrifugation, the majority of the NTFs and a portion of the CTFs were detected in the low molecular mass range (less than 100 kDa) (Fig. 1B). However, the bulk of the CTFs still appeared in the high molecular mass fractions at 150 -200 kDa (Fig. 1B) indicating either a stronger CTF self-association or high affinity interactions with other proteins. In the presence of 0.5% SDS, the high molecular mass complex was completely disassembled, so that both the NTF and CTF were found within the low molecular mass range (less than 69 kDa) (Fig. 1C).
In agreement with the observations from HEK293 cells, the majority of PS1 NTFs and CTFs from digitonin extracts of human brain were present in fractions 3 through 7, with a peak at an apparent molecular mass of approximately 250 -300 kDa ( Fig. 2A). As with the HEK293 cells, the NTF and CTF complexes from human brain were detergent-sensitive. Thus, when extracted in 0.5% Nonidet P-40, the brain NTFs were present in fractions 1-5, while the majority of the CTF were found in fractions 5 and 6 ( Fig. 2B). When brain tissue was extracted with Triton X-100 (data not shown), there was slightly more disassembly of the complex than was observed in the Nonidet P-40 treated brain extracts. Extraction of brain tissue with SDS resulted in complete dissociation of the PS1 complex (Fig.  2C). The differential sensitivity of the complex to different detergents is therefore similar in brain and in cultured HEK293 cells.
To investigate the distribution of the low abundance PS1 holoprotein, we examined glycerol velocity gradient fractions derived from digitonin extracts of an HEK293 cell line stably transfected with human PS1 and ␤APP 695 , which expresses moderate levels of full-length PS1. As with the untransfected HEK293 cells, the NTF and CTF were components of a ϳ250 kDa complex (fractions 5 and 6; Fig. 3). The PS1 holoprotein appeared in fractions 2-7 (molecular mass of 69 -250 kDa), which partially overlapped with the fractions containing the NTF and CTF (Fig. 3). However, the strongest holoprotein signal was obtained in lower molecular mass fractions (Fig. 3, fractions 4 and 5). This suggests that PS1 holoprotein is either a digitonin-sensitive part of the high molecular mass fragment complex or, more likely, that it forms a distinct complex.
The PS1 NH 2 -and COOH-terminal Fragments Associate within the 250-kDa Complex-The presence of both the NTF and CTF of PS1 in the same density gradient fractions does not preclude the possibility that they form independent complexes after endoproteolytic cleavage. To address this, fractions con- taining the 250-kDa complex were pooled (fractions 3-8) and used for reciprocal immunoprecipitation studies with anti-CTF antibody (Fig. 4), or with anti-NTF antibody (not shown). As with co-immunoprecipitation experiments performed on whole cell lysates ( Fig. 5C and D; lanes 2 and 3), the CTF and NTF co-precipitate from glycerol gradient fractions containing the ϳ250-kDa complex (Fig. 4). The specificity of the NTF-CTF association was confirmed by the absence of co-immunoprecipitation of holoprotein, NTF or CTF with: 1) preimmune serum ( Fig. 4; lanes 4 -6), or 2) with antibodies to irrelevant ERresident proteins such as calnexin (data not shown). However, as would be expected from the detergent sensitivity of the ϳ250-kDa complex, the NTF and CTF did not co-precipitate when the HEK293 cells were lysed with Triton X-100 (Fig. 5, C   and D; lanes 5 and 6). Taken together, these results indicate that the NTF and CTF of PS1 are components of the same ϳ250-kDa complex. The NH 2 -and COOH-terminal fragments of PS2 are also present in a similar high molecular mass complex in digitonin lysates of HEK293 cells (data not shown). However, PS2 NTF does not co-precipitate with either the NTF or the CTF of PS1 (Fig. 5B). This suggests that the two presenilins form distinct protein complexes of similar size.
␤-Catenin Is a Component of the 250-kDa PS1 Complex-While it is possible that the NTFs and CTFs of PS1 are the sole components of a high molecular mass complex (15), we consider this to be unlikely. In the Triton X-100 lysates of HEK293 cells (Fig. 1), and in the Nonidet P-40 lysates of human brain (Fig.  2), most of the CTF immunoreactivity is contained in high molecular mass fractions (150 -250 kDa), whereas the majority FIG. 5. The PS1 fragments and ␤-catenin are members of a ϳ ϳ250-kDa complex and interact in a detergent-sensitive manner. Untransfected HEK293 cells were solubilized in either 1% digitonin (lanes 2-4) or 0.5% Triton X-100 (lanes 5-7) and immunoprecipitated using the antibodies as indicated at the top; PS1 NH 2 -terminalspecific antisera (Anti-PS1-N), COOH-terminal-specific antisera (Anti-PS1-C), and preimmune sera. The immunoprecipitation products resolved on SDS-polyacrylamide gel electrophoresis and the immunoblots investigated with antibodies to: ␤-catenin (A); the presenilin-2 NH 2 -terminal fragment (B), PS1-NTF (C), or the PS1-CTF (D). The column labeled Lysate (lane 1) represents about 30% of the starting digitonin lysate utilized for immunoprecipitation, which was used as a positive control. of NTF is contained in lower molecular mass fractions (Ͻ100 kDa). This suggests that under slightly stronger detergent conditions, the CTF and NTF dissociate leaving the CTF still associated with partners other than the NTF. Using yeast two hybrid screens, we and others have found that members of the armadillo family such as ␤-catenin and hNPRAP can interact with presenilins (22). 2 Considering this, we examined the 250-kDa PS1 complex from HEK293 cells to determine whether it also contained endogenous ␤-catenin.
Antibodies against either the NTF or the CTF of PS1 both co-precipitated ␤-catenin from digitonin lysates of native HEK293 cells (Fig. 5A, lanes 2-4). Reciprocal co-immunoprecipitation experiments confirm that anti-␤-catenin antibodies co-precipitate the PS1 NTF and CTF (Fig. 6, lane 2). We have shown that the armadillo-protein binding site of PS1 is located near residues 372-399 within the CTF. 2 As would be expected from this, dissociation of the NTF and CTF by solubilization in Triton X-100 resulted in co-precipitation of ␤-catenin with the PS1 CTF but not with the NTF (Fig. 5A, lanes 5-7). These interactions are specific because the NTF and the CTF did not co-immunoprecipitate with either ␥-catenin (Fig. 6, lane 3) or calnexin (data not shown).
Finally, endogenous ␤-catenin was present in fractions 2-10 of digitonin-extracted HEK293 cells (Fig. 7A). While the size distribution of ␤-catenin complexes overlaps that of PS1, it is obviously much broader than that of PS1. The broad size distribution of ␤-catenin complexes reflects the fact that ␤-catenin exists as free cytoplasmic molecules, and as complexes with proteins such as ␣-catenin, ␥-catenin, cadherin, adenomatous polyposis coli, and glycogen-synthase-kinase-3␤ (23)(24)(25). A direct interaction between PS1 and ␤-catenin within the same ϳ250-kDa complex was confirmed by showing that antibodies to PS1-CTF could co-immunoprecipitate ␤-catenin and PS1-CTF only from the subset of fractions containing both proteins, i.e. fractions 3-7 (Fig. 7B). The profile of PS1 CTF-precipitated ␤-catenin is very similar to the migration profile of the fragments of PS1 in the glycerol gradient (compare Figs. 1A and 7). ␤-Catenin, NTF, and CTF are therefore tightly associated members of a ϳ250-kDa complex.
␤APP has also been proposed to interact with PS1 and PS2 (26,27). However, in the double stable PS1/␤APP HEK293 cell line, ␤APP immunoreactivity was found predominantly at ϳ150 kDa, close to its expected migration for a monomeric molecular mass of ϳ110 kDa (Fig. 7C). There was partial overlap with the migration of PS1 holoprotein although the peak immunoreactivities for each occur in different regions of the gradient (␤APP in fraction 3, Fig. 7C; PS1 holoprotein in fraction 4, Fig. 3). In contrast to the co-immunoprecipitation of PS1 and ␤-catenin, ␤APP and PS1 could not be co-immunoprecipitated from fractions containing either the ϳ180-kDa or the ϳ250-kDa complexes (data not shown).
Subcellular Localization of PS1 Fragments and Holoprotein-The apparent difference between the sizes of complexes containing PS1 holoprotein and those containing the NTF/CTF implies that they may exist in different intracellular compartments. We therefore examined the distribution of PS1 holoprotein and NTF/CTF complexes by subcellular discontinuous sucrose density fractionation of homogenates from human brain, untransfected HEK293 cells, and HEK293 cells overexpressing both wild-type human PS1 and ␤APP. The organelle distribution within the gradient was confirmed using the Golgi-specific marker protein, p58 (19,20), and the ER-resident protein, calnexin (12,21). In brain, as well as in transfected and untransfected HEK293 cells, PS1-NTF and PS1-CTF were both present in fractions from the Golgi apparatus and sER (Fig. 8). Slightly lower amounts were present in the rER (Fig. 8). This topographic co-localization of the NTF and CTF is in good agreement with the results described above showing that the NTF and CTF of PS1 form stable high molecular mass complexes.
In fractions from the PS1/␤APP stable HEK293 cell line, PS1 holoprotein partially overlapped the distribution of PS1 NTF and CTF (Fig. 8D). Thus, the majority of the holoprotein immunoreactivity was present in fractions corresponding to rER and sER, while the NTF and CTF predominated in fractions corresponding to Golgi as well as both the rER and the sER.  6. The PS1 proteolytic fragments co-immunoprecipitate with ␤-catenin but not ␥-catenin. HEK293 crude cell membrane extracts were prepared and solubilized in 1% digitonin and used for immunoprecipitation studies with anti-␤-catenin (lane 2) or anti-␥catenin (lane 3) antibodies. The immunoprecipitation products were separated by SDS-polyacrylamide gel electrophoresis and the immunoblots probed with a mixture of PS1 NH 2 -(Ab14) and COOH-terminalspecific (antibody 1143) antisera. This indicated that the NH 2 -terminal and COOH-terminal bound and specifically co-precipitated with ␤-catenin. The column labeled Extract (lane 1) represents about 10% of the starting membrane extract used for immunoprecipitation.

FIG. 7. Analysis of PS1 fragments/␤-catenin association by velocity gradient centrifugation and co-immunoprecipitation.
Digitonin extracted total proteins from native HEK293 cells (A and B) or from ␤APP stable-transfected HEK293 cells were fractionated by centrifugation through 10 -40% linear glycerol density gradients. Onequarter of each fraction was analyzed by immunoblotting to reveal the distribution of total ␤-catenin (A) or the full-length amyloid precursor protein (FL-APP) (C) along the gradient. The remainder of each gradient fraction was immunoprecipitated with an anti-PS1 COOH-terminal antibody 520. The immunoprecipitated products were immunoblotted and probed with anti-␤-catenin antibody to reveal the distribution of PS1-associated ␤-catenin along the gradient (B). Arrows at the top indicate the mobilities of protein molecular mass markers. Numbered fractions collected from the glycerol gradient are indicated at the bottom.
the holoprotein than might have occurred under physiologic conditions. Nevertheless, the disparity in the biochemical localization of the holoprotein versus the fragments is consistent with the fact that the holoprotein and the fragments associate into complexes of different sizes on glycerol velocity gradients (Fig. 3). Taken together, these data suggest that the PS1 holoprotein and its endoproteolytic fragments may exist within biochemically distinct complexes during the transportation and processing of PS1 from ER to Golgi apparatus.
The sucrose gradient blots from the PS1/␤APP HEK293 stable cells and from human brain were investigated with anti-␤catenin antibodies (Fig. 8E). Although ␤-catenin is a soluble protein, it co-purifies with PS1 holoprotein and PS1 CTF in the rER, sER, and Golgi membrane fractions. This would be expected from its interaction with a domain in the large cytoplasmic loop of PS1. DISCUSSION We have identified a ϳ250-kDa native protein complex which contains at least ␤-catenin, PS1-NTF and PS1-CTF. This complex is present in human brain, in native HEK293 cells, and in PS1 transfected HEK293 cells. It is apparent that the molecular masses of PS1-NTF (29 kDa), PS1-CTF (18 kDa), and ␤-catenin (92 kDa) on a 1:1:1 stoichiometry do not account for all of the estimated mass of the ϳ250-kDa complex. It is unclear whether the additional mass reflects a different ratio of PS1 and ␤-catenin (e.g. 3:3:1 or 1:1:2) or, more probably, whether other unidentified PS1-associated proteins are present. Further studies will be required to address this question and identify these other components if they exist.
Like many membrane proteins, presenilins have a tendency to aggregate under non-native conditions (28). Several lines of evidence indicate that the ϳ250-kDa protein complex we have identified is authentic and is not caused by nonspecific aggregation. First, we analyzed the native state of endogenous PS1 by linear glycerol velocity gradients. Second, the same complex exists in both human brain and in the HEK293 cell line. Third, when PS1 is overexpressed the holoprotein complex appears to be smaller than the native fragment complex, an observation which contradicts the intuitive expectation that nonspecific aggregation should be more severe when PS1 is overexpressed. Fourth, the fact that the PS1 complex does not contain the highly homologous PS2 protein also contradicts the expectations for a nonspecific aggregation. Fifth, three components of the complex (PS1 NTF, PS1 CTF, and ␤-catenin) can be specifically co-immunoprecipitated from both whole cell lysates and from the appropriate glycerol gradient fractions. The identification of ␤-catenin as a member of the high molecular mass complex, but not other proteins such as ␥-catenin, calnexin and ␤APP (discussed below), also argues against a nonspecific aggregate. Finally, in agreement with previously published immunocytochemical studies on the intracellular distribution of presenilin proteins (4, 5), we can show that the three identified components of the ϳ250 kDa presenilin complex (␤-catenin and NTF and CTF of PS1) are located in the same biochemically defined subcellular compartment.
␤APP has been reported to be capable of interacting with PS1 and PS2 (26,27). However, ␤APP is unlikely to be a significant component of the ϳ250-kDa PS1 complex described here for the following reasons. First, the expression level of endogenous ␤APP is very low in HEK293 cells. Despite this, the glycerol gradient migration pattern of PS1 fragments in untransfected HEK293 cell lysates was very similar to that in lysates of both PS1/␤APP double-transfected HEK293 cells and of human brain. If ␤APP is a major component of the complex, the PS1 complex would be expected to differ in size between the untransfected and the PS1/␤APP-transfected HEK293 cells. Second, the glycerol gradient fractionation profiles of ␤APP and PS1 holoprotein only partly overlap even in digitonin extracts, and do not overlap at all with the PS1 fragment complex. Thus, only a small amount of ␤APP and PS1 holoprotein would be available to interact. Finally, ␤APP and PS1 do not co-precipitate from the glycerol gradient fractions containing PS1 and ␤APP in digitonin lysates of ␤APP/PS1 stable cells. However, we can not exclude the possibility of a greater detergent sensitivity of ␤APP binding or that ␤APP is a component of a minor PS1-containing complex. This would be especially true if this complex was the result of a weak or transient interaction with the PS1 holoprotein (27).
Whether the complex functions simply to fold, transport and process PS1, or whether it has other intrinsic functions remains to be determined. However, regulated insertion of the PS1 NTF and CTF into this complex does provide a potential explanation for the observation that the stoichiometry and level of these fragments appears to be tightly controlled in most cells (8). Previous investigations have demonstrated that the majority of the PS1 and PS2 holoproteins are rapidly degraded by the proteasome with half-lives of ϳ15 min (10,11). It is likely that this represents a separate pathway for the removal FIG. 8. Biochemical subcellular localization of PS1 holoprotein and endoproteolytic fragments within the rER and sER and Golgi apparatus. Immunoblots are sequential fractions from discontinuous sucrose gradients containing cellular proteins with the organelle distribution shown by the localization of ER-resident calnexin and Golgi-specific p58 (A). Distribution within these compartments was investigated for full-length PS1 (PS1-FL as well as NH 2 -terminal and COOH-terminal fragments from: human brain (B), native HEK293 cells expressing endogenous PS1 levels (C), and HEK293 cells stably overexpressing PS1 and APP 695 (D). The location of ␤-catenin immunoreactivity relative to the various PS1 proteins is also indicated (E). of full-length presenilins, and is consistent with our observation that most PS1 holoprotein exists as a separate complex which may be more susceptible to proteasome degradation. In contrast, PS1 CTF and NTF are incorporated into a stable ϳ250-kDa complex which contains at least one other potentially functional ligand (␤-catenin). The regulated incorporation into this complex (e.g. by limited availability of one or more other ligands) would provide an explanation for the regulated abundance of the CTF and NTF. Furthermore, inclusion of the CTF and NTF into a stable complex would also account for the long turnover times of the cleavage fragments. It is unclear whether endoproteolysis occurs before or following insertion into the ϳ250-kDa complex. The transition from a lower molecular mass holoprotein enriched complex to the ϳ250-kDa fragment enriched complex could be brought about by an enhanced PS1 NTF/CTF oligomerization following endoproteolytic cleavage. Alternatively, the endoproteolytic cleavage might occur beforehand, but could expose previously internalized domains which then facilitates binding of the other ligands. Understanding the mechanism and the composition of this complex will likely lead to a better understanding of PS1 function.