Presenilin 1 Controls γ-Secretase Processing of Amyloid Precursor Protein in Pre-Golgi Compartments of Hippocampal Neurons

Materials and Methods

Results

Mouse brain and hippocampal neurons derived from wild-type mouse brain were used to localize endogenously expressed PS1 using subcellular fractionation and confocal immunofluorescence microscopy. In addition, transgenic mice expressing human PS1 under the control of the prion promoter, which drives expression of the transgene in neurons, were used for confocal microscopy and to obtain fractionation data on the distribution of PS1 in neurons in situ.

Because of their relatively high buoyant density, nuclei can be easily purified by differential centrifugation in high sucrose media. As is clear from Fig. 1, only a little PS1 staining is observed in purified nuclei, which contain mainly histones in the protein fraction (not shown). However, additional DNA digestion combined with ultracentrifugation yields a fraction of highly purified nuclear envelopes, as demonstrated by the abundant lamin B1 immunoreactivity (Fig. 1 a, NE). Interestingly, this final purification step results in a strong enrichment for PS1-CTF and -NTF immunoreactive bands (Fig. 1 a), indicating that the little amount of PS1 present in the nuclei is almost exclusively associated with the nuclear envelope. It should be noticed that this fraction also stains with calnexin antibodies, in agreement with the continuity between nuclear envelope and RER. Remarkably, in this nuclear envelope fraction, a prominent immunoreactive band representing full-length PS1 is detected with both NTF- and CTF-specific antibodies (Fig. 1 a and not shown) (see below). The presence of presenilin in the nuclear envelope was confirmed in fixed polarized hippocampal neurons using confocal microscopy. PS1 immunoreactivity colocalized with lamin B1 in confocal sections through the middle of the nucleus (Fig. 1 b, arrows). Together, these results indicate that PS1 fragments and PS1 holoprotein are present in the nuclear envelope.

Next, the distribution of PS1 was investigated in pre-Golgi compartments of hippocampal neurons derived from human PS1wt transgenic mice (Fig. 2). rSEC61αp is an essential component of the mammalian ER protein translocation complex and is tightly associated with membrane-bound ribosomes in RER (Gorlich et al. 1992). PS1 immunoreactivity was found to codistribute only partially with the reticular staining pattern observed for rSEC61αp (Fig. 2, a–c, arrowheads). In contrast, a considerable fraction of PS1 staining, especially obvious in the perinuclear region, did not localize with rSEC61αp (Fig. 2, a–c, arrow). Antibodies to calnexin (Fig. 2, d–f) and BAP31 (Fig. 2, g–i), two proteins that are abundantly distributed throughout the ER, displayed a reticular staining that overlapped very well with most of the immunolabeling for PS1-CTF. However, it should be noticed that several distinct PS1-positive spots did not colocalize with calnexin immunoreactivity. Transport from the ER to the IC and Golgi apparatus occurs in specific subdomains of the ER where transported proteins are concentrated in budding transport vesicles coated by COPII coat proteins (Barlowe et al. 1994). SEC23p is one of the cytosolic components of this COPII coat complex. As shown in Fig. 2j–l, this COPII protein displayed an overall colocalization with PS1-CTF (Fig. 2 l, inset, yellow). At a higher magnification, all PS1-CTF–positive patches were also immunolabeled for SEC23p (see detailed area in Fig. 2, j–l). These export sites are often found juxtaposed to vesiculotubular clusters (VTCs) of the IC (Bannykh and Balch 1998). Indeed, immunolabeling of ERGIC-53/p58, so far the best characterized IC resident protein (Schweizer et al. 1991; Klumperman et al. 1998), showed a close colocalization with PS1-CTF (Fig. 2, m–o). Therefore, these results indicate an abundant localization of presenilin in the ER.

Especially the close overlap with SEC23p and ERGIC-53 prompted us to further document the interesting localization of PS1 in the transit zone between ER and Golgi. We applied a cell fractionation scheme described by Schweizer et al. 1991 that allows one to obtain highly enriched fractions of the IC in VERO cells. This method starts with a Percoll gradient centrifugation step, resulting in a rough separation of RER and Golgi membranes. When applied on the postnuclear fraction of mouse brain, most of the PS1 immunoreactivity was detected in two peaks (Fig. 3 a). One small peak (15.1 ± 0.5% of total PS1-CTF or 19.4 ± 3.3%, n = 3, of total PS1–NTF immunoreactivity, respectively) was associated with the bottom fractions of the gradient coinciding with the distribution of the RER marker protein rSEC61αp (Fig. 2 a), confirming that a relative low amount of PS1 is associated with the RER compartment. A second, major peak (pool II: 42.7 ± 2.4% for PS1-CTF and 48.1 ± 0.5%, n = 3, for PS1–NTF) was located in the low density region coinciding with the density of Golgi membranes (Fig. 3 a and Schweizer et al. 1991). Typically, IC membranes equilibrate as a broad peak in the midportion of this self-formed gradient. Fractions 5–14, containing 61.6 ± 7.3% (n = 3) of the total ERGIC-53/p58 immunoreactivity contained 25.1 ± 2.1% of PS1-CTF and 17.9 ± 1.5% of PS1-NTF (n = 3). This fraction was pooled (Fig. 3 a, Pool I) and further purified and concentrated by an additional centrifugation step in a discontinuous Nycodenz gradient (Fig. 3 b). Three interfaces, F1, F2, and F3, are recovered. The interface F3 at 18.5–27% Nycodenz is highly enriched in IC membranes, as demonstrated by the strong immunoreactivity for p58/ERGIC-53 (Fig. 3 b). Both PS1-NTF and -CTF coenrich to a similar extent in this final F3 fraction. Unfortunately, and in contrast to the original method (Schweizer et al. 1991), immunoblotting for rSEC61αp and TRAPα, a protein associated with the translocon site (Hartmann et al. 1993), revealed contamination of F3 with RER membranes (Fig. 3 b). This discrepancy is probably a consequence of the difference in starting material, i.e., VERO cells versus mouse brain used here. Therefore, we purified the F3 fraction in an additional continuous sucrose gradient (0.8–1.6 M sucrose). Whereas the marker proteins calnexin and ERGIC-53/p58 displayed together with PS1 a bimodal distribution (Fig. 3 c), the rSEC61αp and TRAPα immunoreactivity was solely recovered in the high density region of the gradient. Thus, the PS1 immunoreactivity in the low density region (fractions 4 and 5) of the gradient cannot be explained by contaminating RER membranes. Furthermore, we quantified all marker proteins by densitometric scanning. When the results were expressed as density units per milligram of protein (Fig. 3 c, bottom), it became obvious that PS1 fragments specifically coenriched with ERGIC-53/p58 rather than with calnexin, suggesting their localization in the IC. Moreover, the enrichment factor for ERGIC-53/p58 obtained in the low density peak fraction was 35 compared with total homogenate, reaching a similar degree of enrichment as originally described for VERO cells (Schweizer et al. 1991).

From the initial Percoll gradient (Fig. 3 a), a major peak of PS1 immunoreactivity (Pool II) can be observed corresponding to less dense regions where Golgi-derived membranes tend to equilibrate (Schweizer et al. 1991). To analyze whether this reflected an association of PS1 with Golgi membranes, we further purified the pooled fractions 15–19 using discontinuous Nycodenz gradient centrifugation consisting of three layers of different Nycodenz concentrations. Most of the PS1 immunoreactivity was recovered in the 14–21% interface (IF3) together with marker proteins of the Golgi apparatus (such as GM130). This interface IF3 was loaded on a continuous sucrose gradient (0.3–1.7 M sucrose) and the distribution of PS1 was analyzed after ultracentrifugation (Fig. 4 A). Established Golgi marker proteins such as GM130 (Fig. 4 A) and β-cop (not shown) equilibrated at high sucrose densities in fractions 4 and 5, whereas the bulk amount of PS1-CTF and -NTF immunoreactivity was recovered at a lower buoyant density with a peak in fractions 7 and 8. The striking similar distributions of the unglycosylated form of TGN38 (Fig. 4 A, M _r 45,000) and RAP with the PS1-CTF and -NTF indicate that the PS1 in this pool is mainly associated with early cis-Golgi membranes. Indeed, unglycosylated TGN38 is located in the early cis-Golgi, whereas RAP is a molecular chaperone assisting membrane proteins such as the LDL receptor in their passage through the early secretory pathway between the ER and Golgi (Bu and Schwartz 1998). Since the IC resident protein p58/ERGIC-53 did not codistribute with fractions 7 and 8, these fractions may identify a membrane-bound intermediate between the IC and cis-Golgi. Mature (glycosylated) TGN38, which is a marker for the TGN tubules, is mainly found in the bottom fraction (Fig. 4 A, fraction 1). Since this fraction is completely devoid of PS1 immunoreactivity, we can assume that PS1 does not reach the TGN in significant amounts.

This observation was entirely corroborated by confocal laser microscopy of control hippocampal neurons. Antibodies against the peripherally associated 58-kD Golgi protein (Fig. 4 B, p58/Golgi), detected Golgi staining restricted to the cell body and proximal dendrites of neurons (Fig. 4 B, a–c) in agreement with Krijnse-Locker et al. 1995. At first glance (Fig. 4 B, c), a high degree of overlap with the immunostaining for PS1-CTF seemed to be observed. However, when a more careful analysis was performed at a higher magnification (Fig. 4 B, d–f), only a limited overlap could be demonstrated (Fig. 4 B, d–f, asterisk and arrowhead). This becomes even more clear in the vertical plane section of a proximal dendrite (Fig. 4 B, g–i; the arrow in d–f indicates the site of the section). PS1-CTF staining was concentrated in defined stacks often capped by or juxtaposed to staining for p58/Golgi (Fig. 4 B, g–i, arrowheads). Both proteins colocalized only partially in discrete areas. This is in line with the fractionation data.

α- and β-secretase cleavage of APP occurs in the late Golgi, in transport vesicles, at the cell surface, and in endocytic compartments (Selkoe 1998), generating the α- and β-APP carboxy-terminal stubs associated with the membranes. These stubs are the direct substrates for γ-secretase. Given the restricted distribution of PS1 in the ER, IC, and early cis-Golgi compartments, the question arises how the γ-secretase cleavage of these APP stubs can be controlled by PS1.

Therefore, we analyzed the distribution of endogenous APP in the same fractions analyzed before for presenilin immunoreactivity. As shown in Fig. 5 a, both the gradient distributions of the APP holoprotein and the APP carboxy-terminal fragments were bimodal, which is similar to what was found for PS1 (Fig. 3 a). The larger amounts are found in the fractions of Pool II. Further purification resulted in the concomitant enrichment for the APP carboxy-terminal fragments in the F3 interface (Fig. 5 b), where PS1 is also recovered. Remarkably, this enrichment was much less obvious for the APP holoprotein, suggesting a selective accumulation of the carboxy-terminal APP fragments in this fraction. Further analysis of the Pool II–associated APP immunoreactivity revealed the enrichment of the APP carboxy-terminal fragments in exactly the same fractions of the sucrose gradient where PS1 fragments (together with RAP) are recovered (compare Fig. 5 c with Fig. 4 A). Interestingly, therefore, the carboxy-terminal fragments of APP are apparently at least partially codistributing with PS1-NTF and -CTF in the same subcellular membrane compartments (see Discussion). Given the existing controversy whether or not PS1 can be found in compartments beyond the ER/IC and cis-Golgi, we extended our study to some well-defined post-Golgi compartments such as lysosomes, endosomes, synaptic vesicles, the cell surface, and clathrin-coated vesicles. Both PS1 fragments clearly purify away from mature and intact lysosomes (Fig. 6, compare with the enrichment for cathepsin D), and were observed only after prolonged exposure times. The weak immunoreactivity in the pure lysosomal fractions is most likely a consequence of minor contamination of the isolated fraction.

We also investigated the possible colocalization of CTF-PS1 with marker proteins of post-Golgi compartments such as the transferrin receptor and synaptobrevin II in hippocampal neurons of control, nontransgenic mice (Fig. 7). In fully polarized neurons, the transferrin receptor is localized somatodendritically and, accordingly, numerous transferrin receptor–positive punctae are noticed in the dendritic arbor (Fig. 7, a–c, dendrites, and d–f, growth cone). Virtually no overlap is observed for PS1-CTF and transferrin receptor staining, indicating that PS1 is not present in recycling endosomes. Synaptobrevin II is often used as a marker for synaptic vesicles that cluster in specialized axonal regions of mature neurons referred to as synaptic boutons. Again no immunostaining for PS1 could be detected within these boutons (Fig. 7, g–l, with emphasis on the detailed area in g–i), suggesting that no PS1 is associated with the membranes of synaptic vesicles. Further analysis of highly purified clathrin-coated vesicles from nerve terminals, or double immunostaining with specific cell surface marker such as ICAM-5, and surface biotinylation experiments (not shown) did not yield any conclusive evidence that PS1 could be associated with clathrin-coated vesicles or with the cell membrane. (data not shown). It is clear that the amounts of presenilin present in the membranous compartments beyond the ER–cis-Golgi in hippocampal neurons are very limited and, in any event, below the detection limits of our current available technology. In conclusion, the data obtained from both cell fractionation and immunofluorescence studies on brain and neurons, demonstrate the abundant association of PS1 with the ER, IC, and to a lesser extent, the early cis-Golgi.

This prompted us to further investigate the relationship between APP processing and presenilin at a more functional level. Therefore, we generated a series of APP trafficking mutants that limit APP processing to specific subcellular compartments. A dilysine motif was added to the carboxy terminus of APP to retain APP/KK in pre-Golgi compartments, thereby emphasizing the βA4(1-42) production (Chyung et al. 1997). Reinternalization of cell surface APP was blocked by deleting the cytoplasmic tail (APPΔCT), limiting APP trafficking to the biosynthetic pathway (Tienari et al. 1996b), and severely impairing βA4 secretion (Perez et al. 1999). A chimerical APP/LDLR was generated by replacing the cytoplasmic domain of APP by that of the LDL receptor. This mutation increases the recycling of APP in the endosomal compartments and promotes βA4(1-40) production (see below). We also investigated the effects of the clinical APP/London mutation in combination with the PS1 mutations, to see whether these mutations operate additively or not. The APP constructs were expressed in neurons using the SFV expression system. Similar expression levels for all constructs were obtained in hippocampal cultures, except for APPΔCT (Fig. 8, CELL). For comparisons between the different experiments, the obtained phosphorimaging results were normalized to the expression levels of APP holoprotein in the cell extracts. Expression of APPΔCT led to decreased βA4 secretion concomitantly with the production of a short cell-associated β-stub. In the case of APP/KK infection, the decrease of secreted βA4 was even more pronounced (also see below). The APP/KK mutant is actively retrieved from the cis-Golgi to the ER as a consequence of the dilysine motif. This was confirmed by its complete endo-H glycosidase sensitivity (data not shown) and its codistribution with calnexin (Fig. 9 b) and endogenous PS1 (Fig. 9 c). The striking colocalization of endogenous PS1 with virally expressed APP/KK (Fig. 9 c, inset) should be noticed. As a control, SFV-APPwt is shown, which confirms the axonal delivery of this protein (Fig. 9 a) (Simons et al. 1995; Yamazaki et al. 1995; Tienari et al. 1996b). We analyzed systematically the metabolism of the expressed APP mutants (Table ). Quantitative immunoprecipitation and phosphorimaging determined the total amounts of cell-associated α- and β-stubs and secreted βA4. To standardize and to allow comparisons between different experiments, we normalized all values to cell-associated APP holoprotein. Similarly, normalization of the numbers obtained in the ELISA experiments was performed because of the relative large variations in the absolute amounts of peptide produced in separate independent experiments. Therefore, only ratios of secreted and intracellular βA4(1-42)/(1-40) are given. Except for APP/LDLR, all constructs were completely analyzed in the three types of neuronal culture, i.e., expressing human PS1 wt or the clinical mutants PS1_L286V or PS1_M146L. The results were compared with those obtained in cells derived from control and nontransgenic littermate embryos. Table summarizes the data obtained for APPwt (A), the London mutant (B), APPΔCT (C), and APP/KK (D).

With respect to α- and β-cleaved carboxy-terminal fragments, no significant differences were noticed related to PS1 wild-type or clinical mutants except for the London mutant. The clinical mutations in PS1 do not affect α- and β-secretase activity, complementing the data obtained previously in PS1-deficient neurons (De Strooper et al. 1998). However, in the case of the London mutant, levels of α- and β-stubs were surprisingly decreased in the FAD mutant PS1 neurons. This reflects probably the increased turnover of both stubs by γ-secretase cleavage, in accordance with the strong additive effects of the London- and FAD-linked PS1 mutations (especially PS1_M146L) on γ-secretase processing (see below).

To evaluate γ-secretase activities, we measured the total pool of secreted βA4 by immune precipitation as well as the ratio of the individual peptides by ELISA. A modest increase in the total amount of secreted βA4 was observed for the PS1_M146L mutant. This effect was not very dramatic (except for the APP/KK, see below). In contrast, PS1 mutations caused dramatic increases in the βA4(1-42) to βA4(1-40) peptide ratio, and this effect was even more pronounced when the effect of the mild overexpression of PS1wt alone was directly compared with that of PS1 containing clinical mutations. The PS1_M146L mutation was overall more effective than the PS1_L286V mutation.

The London mutation on its own caused a threefold increase (3.44±0.43; n = 6) in the ratio of secreted βA4(1-42)/(1-40) versus APP wild-type in control neurons (not shown). Interestingly, an additional rise in this ratio was observed in combination with PS1 mutations (Table B). This rise ranged from 1.38±0.07 (n = 8) for the PS1_L286V mutation to 2.97±0.65 (n = 3) for the PS1_M146L mutation. When compared with APPwt expressed in wild-type neurons, the total raise ranged up to 12.3 ± 3.1 (n = 3). The increase appears to be solely due to an increased secretion of the β4A(1-42) peptide and not a consequence of a decreased βA4(1-40) production, explaining the decrease in α- and β-carboxy-terminal fragments as mentioned above.

APPΔCT in control neurons yielded low levels of secreted βA4. However, the ratio between the two peptides was slightly raised compared with SFV-APPwt (1.30±0.14, n = 8, versus SFV-APPwt). This ratio increased even further when this mutant was expressed in PS1_M146L neurons (Table C). For APP/KK, the total secretion of βA4(1-40) was also lowered dramatically to 5.0±1.6% (n = 6) when compared with APPwt (see also Fig. 8). The βA4(1-42) secretion was not affected as strongly as the βA4(1-40) secretion and only dropped to 18.0±4.7% (n = 6) when compared with APPwt. The introduction of the ER-retrieving dilysine motif on its own, therefore, caused an appreciable increase in the 42:40 ratio of the secreted peptides to 3.98±0.70 (n = 6) times relative to APPwt. Upon expression in PS1_M146L neurons, an additional raise in this ratio of 2.83±0.57 (n = 6, Table D) was observed. The increase could again solely be contributed to an increased secretion of the βA4(1-42) peptide. Interestingly, only for this construct, a sharp rise in total secreted βA4 was noticed in PS1_M146L neurons. It should be noticed that the main peptide produced by APP/KK is βA4(1-42). This observation indicates that some of the presenilin mutations do not only affect the ratio of the secreted peptides, but can also increase the total production of βA4(1-42). The results obtained with this ER-retained APP/KK mutant also indicate that the clinical PS mutations are operating in the ER. This conclusion was further corroborated in an experiment comparing the amyloid peptide production from APP/LDLR in control and PS1_M146L hippocampal neurons. As expected for a mutant trafficking mainly in the endocytic compartments, a relative increase in secreted βA4(1-40) of 2.36±0.18 (n = 3, SEM) fold was observed, whereas secretion of βA4(1-42) remained largely unaffected. Interestingly, the PS1_M146L mutation did not cause a statistically significant effect on the ratio of secreted βA4 peptides in these neurons. Finally, we note that in our experimental system, and for all APP constructs studied, PS1 clinical mutations do not cause a significant increase in the βA4(1-42)/(1-40) ratio of intracellular peptides.

Discussion

In this study, we provide conclusive evidence that the bulk of endogenously expressed PS1 in neurons is localized in the early compartments of the secretory pathway, i.e., in the ER, the IC, and the early Golgi. Relatively limited levels of PS1 are furthermore detected in the nuclear envelope and the RER. The ER, in contrast to the Golgi apparatus, is widely distributed in the soma and the dendrites (including the most distal regions) of neurons (Krijnse-Locker et al. 1995). Therefore, our data explain the previously documented somatodendritic localization of the presenilins (Cook et al. 1996; Busciglio et al. 1997; Capell et al. 1997). Within these early compartments of the biosynthetic pathway, a clear gradient of increasing amounts of the supposedly functional PS1 fragments is observed. The PS1 immunoreactivity previously detected in the nuclear membrane (Li et al. 1997) is, therefore, a mixture of (predominantly) PS1 fragments and full-length PS1, indicating that cleavage of PS1 by the unknown presenilinase occurs very early after biosynthesis (Xia et al. 1998). Importantly, these results definitively resolve the controversy whether and where full-length presenilin can be detected, namely in nuclear envelope fractions.

Interestingly, both the PS1-NTF and -CTF codistributed and coenriched in all fractionation schemes applied, confirming at the endogenous level of protein expression that both fragments remain in a 1:1 stoichiometric relationship along the whole early biosynthetic pathway (Thinakaran et al. 1996). Our data not only settle the debate on the subcellular localization of presenilin, but also document for the first time that PS1 immunoreactivity in the ER is concentrated in discrete patches that are sec23p-positive. Sec23p is one out of the five cytosolic proteins that constitute the COPII coat complex required for vesicle budding from the ER (Barlowe et al. 1994). These COPII components bind specifically with cargo molecules destined for sorting into transport vesicles and separate them from the ER resident proteins (Campbell and Schekman 1997; Aridor et al. 1998; Kuehn et al. 1998). The formation of COPII transport vesicles occurs nonrandomly in the ER at tubular protrusions often juxtaposed to VTC, referred to as the IC (Schweizer et al. 1990, Schweizer et al. 1991; Bannykh and Balch 1998). PS1 codistributes and coenriches with markers of this intriguing compartment such as p58/ERGIC-53 (Saraste and Svensson 1991; Schweizer et al. 1991), suggesting that PS1 is operating in these compartments. A similar conclusion was recently made by Culvenor et al. 1997 who, based on temperature block experiments, suggested that PS1 is localized in the IC and cis-Golgi in stably transfected SY5Y and P19 cell lines. On the other hand, and in contrast with many earlier studies investigating overexpressed PS1 (for instance De Strooper et al. 1997), we could not confirm that PS1 at the endogenous level of expression distributes to a significant extent into the Golgi apparatus. This limits the subcellular compartment where PS1 is operating to the early compartments of the biosynthetic pathway. The earlier reported abundant Golgi localization of the presenilins is probably a consequence of overexpression and missorting of the protein. As illustrated in Fig. 4 B, f and i, colocalization studies should indeed be carefully interpreted. As we suggested before (De Strooper et al. 1997), experiments based on overexpression of presenilins may lead to erroneous conclusions since normal proteolytic maturation and also normal subcellular ER localization is apparently significantly disturbed.

This conclusion is further corroborated by recent findings that overexpression of presenilins results in their redistribution into aggresomes (Johnston et al. 1998). These structures are located at the perinuclear microtubule organizing center, where also the Golgi apparatus is located and are the signature of abnormal biosynthetic stress in the cell. In view of our new data, it is conceivable that PS1 passes from the ER, via VTCs, to cis-Golgi cisternae. The close colocalization of PS1 with sec23p further indicates that at least part of this transport is associated with the trafficking of COPII vesicles. (Schekman and Orci 1996). Since PS1 fragments exist as a stable complex with a long half-life (Ratovitski et al. 1997) and no accumulation in the Golgi apparatus was observed, PS1 is probably also retrogradely transported. Electron microscopical observations demonstrating abundant PS1 immunoreactivity in vesiculotubular structures and coated vesicles in close vicinity or in continuity with Golgi cisternae support the view that presenilins are actively transported back and forward between the ER and cis-Golgi (Lah et al. 1997). The lack of PS1 immunoreactivity in TGN-enriched fractions (Fig. 4 A) suggests that only minimal amounts of PS1, if any, exit the Golgi area. Indeed, no significant amounts of PS1 were found in recycling endosomes, lysosomes, synaptic vesicles, or clathrin-coated vesicles (not shown and Efthimiopoulos et al. 1998). Finally, all published data that support the localization of PS1 at the plasma membrane (Dewji and Singer 1997; Schwarzman et al. 1999) imply a topology of PS1 with the amino-terminal domain at the luminal side of the ER. This is not in agreement with several studies (Doan et al. 1996; Li and Greenwald 1996; De Strooper et al. 1997; Lehmann et al. 1997) that addressed this question more specifically, although it could be speculated that cell type–specific differences exist. In any event, our data in hippocampal neurons clearly indicate that the bulk of PS1 fragments are localized in compartments of the early biosynthetic pathway and that only limited amounts (if any at all) traffic beyond the Golgi apparatus, a finding that needs to be taken into account when considering the association of presenilin with γ- secretase activities.

Accumulating evidence indicate that presenilins are either part of the catalytic γ-secretase activity (Wolfe et al. 1999) or at least controlling its activity (De Strooper et al. 1998, De Strooper et al. 1999). Wolfe et al. 1999 presented evidence that presenilins are aspartate proteases that become inactivated by mutating one of the two aspartic acid residues in the sixth or seventh transmembrane region respectively. Overexpression of this mutant PS1 results in dominant negative effects on γ-secretase activity (Steiner et al. 1999; Wolfe et al. 1999). The most simple explanation is that PS1 is γ-secretase itself. This interpretation is consistent with the observation that deletion of the PS1 gene also results in the inhibition of γ-secretase (De Strooper et al. 1998, De Strooper et al. 1999). In both experiments, secretion of βA4(1-40) as well as βA4(1-42) peptides was equally affected (De Strooper et al. 1998; Wolfe et al. 1999). The subcellular localization of PS1 in the ER as we demonstrate here, is, however, difficult to reconcile with current concepts that γ-secretase processing of APP mainly occurs in the endocytic limb of the subcellular trafficking pathways.

There is clearly a spatial paradox. The first question to be addressed in this regard is whether the real substrates of γ-secretase, namely the α- and β-secretase cleaved APP carboxy-terminal stubs generally believed to be generated in the late biosynthetic compartments and the endocytic pathways, can indeed reach the subcellular compartments where PS1 resides. Unexpectedly, we could demonstrate the enrichment of the APP carboxy-terminal stubs in the ER/IC subcellular fractions where PS1 resides. From a cell biological point of view, it remains to be explained how this retrograde transport of the APP carboxy-terminal fragments to the ER/IC can occur and which sorting signals are involved, but this finding is at least consistent with the PS1–γ-secretase hypothesis (Wolfe et al. 1999). The second question is whether and how PS1 can affect the generation of amyloid peptide in the endocytic limb of the protein trafficking pathways in the cell. This problem was addressed via a functional approach by expressing a series of APP trafficking mutants in hippocampal neurons using the recombinant SFV system. In previous work, it has been extensively demonstrated that the processing and subcellular trafficking of APP expressed with this system does not differ from that observed with endogenously expressed APP (De Strooper et al. 1995; Simons et al. 1995, Simons et al. 1996; Tienari et al. 1996a,Tienari et al. 1996b; Chyung et al. 1997). Therefore, and in contrast with presenilins, the results obtained after transfection of APP with SFV under the experimental conditions used, can be considered as biologically relevant.

The results (see Table ) obtained with APP retained in the ER (APP/KK), with APP processed in the biosynthetic pathway (APPΔCT) or with APP retrieved in the endocytic pathway (APP/LDLR, not shown), corroborated the conclusion that presenilins operate in the ER/IC. Significant increases in βA4(1-42) secretion were observed with the APP/KK and to a lesser extent with the APPΔCT mutant in neurons expressing PS1_M146L.. Such an effect was almost undetectable with the APP/LDL mutant. The latter most likely reflects the inability of the APP/LDLR carboxy-terminal fragments generated by α- or β-secretase to be recycled to compartments where γ-secretase resides. The APP/KK mutant was particular relevant in this regard. The dilysine Golgi–ER retrieval signal is recognized by COPI subunits in proteins such as p58/ERGIC-53 and results in their retrograde retrieval from VTCs and cis-Golgi cisternae to the ER (Cosson and Letourneur 1994; Arar et al. 1995; Lahtinen et al. 1996; Bannykh and Balch 1997). The KK motif limits the trafficking of APP/KK to those compartments where also PS1 resides. Indeed, the subcellular distribution of APP/KK was strikingly similar to that of endogenous PS1 (Fig. 9). As previously demonstrated, only very low amounts of amyloid peptide are recovered in the media with this construct (5% compared with wild-type APP). Since β-secretase is mainly operating beyond the ER/IC (Higaki et al. 1995; Chyung et al. 1997), it is not surprising that only very small amounts of the immediate precursor of βA4, i.e., the β-secretase cleaved carboxy-terminal APP fragment (Paganetti et al. 1996; De Strooper et al. 1998), are generated with APP/KK (Fig. 9). In contrast to the interpretation of Chyung et al. 1997, we do not consider the residual β-secretase activity as evidence for a novel, ER-linked β-secretase activity. Since >95% of the normal β-secretase activity is precluded with this construct, this indicates in fact that β-secretase is operating mainly beyond the cis-Golgi, and that APP/KK is only marginally processed by enzyme trailing in the cis-Golgi. This interpretation is certainly more in line with general concepts of β-secretase activity in other cell types since shown to depend on the activity of Rab1B (Dugan et al. 1995) and on O-glycosylation (Tomita et al. 1998). In any event, the residual generation of β-secretase cleaved fragments obtained with APP/KK was sufficient to demonstrate that mainly 1-42 peptide is generated in ER, IC, and early cis-Golgi, where the APP/KK carboxy-terminal stub is recycling. More importantly, the presenilin mutants have a pronounced effect on this process, which is clearly consistent with their subcellular localization.

On the other hand, γ₄₀-secretase processing in the endocytic pathways is not at all influenced by the PS1 mutations as was most dramatically illustrated with the APP/LDLR mutant. The processing of this mutant, which selectively increases the production and secretion of the β4A(1-40) peptide (this study), was not altered by coexpression of the PS1_M146L. It is generally accepted that the bulk of the βA4(1-40) peptide is generated beyond the TGN (Cook et al. 1997; Hartmann et al. 1997) and in the endocytic pathways (Koo and Squazzo 1994; Perez et al. 1996, Perez et al. 1999; Peraus et al. 1997; Wild-Bode et al. 1997; Greenfield et al. 1999) compartments where no PS1 is found (this study). This explains why PS1 mutations have no major effects on the secretion of βA4(1-40). Further evidence for the selective effect of the presenilin mutations on γ-secretase cleavage at position 42 was obtained when coexpressing APP-London together with mutant PS1 (Borchelt et al. 1997; Citron et al. 1998; this study). Both clinical mutations increase the βA4(1-42) secretion and act additively, which suggests independent or at least synergistic mechanisms operating with the two mutations. The experimental evidence presented here together with many reports in the literature suggests strongly an exclusive role for the presenilins in γ₄₂-cleavage (as opposed to γ₄₀ cleavage). The effect of the inactivation of PS1 (De Strooper et al. 1998) or the expression of dominant negative forms of PS1 (Wolfe et al. 1999) on γ₄₀-secretase activity must, therefore, be indirect. One possibility is a precursor–product relationship between βA4(1-42) and βA4(1-40). In this view, only one γ-secretase is needed that generates βA4(1-42), whereas in compartments distal to the ER, carboxypeptidases convert this peptide to βA4(1-40). This would explain the above-mentioned spatial paradox, but not completely the selective increased secretion of βA4(1-42) linked to PS1 clinical mutations. In a precursor–product relationship, one should also expect a rise in the secretion of βA4(1-40), and this has not been really observed yet. Along similar lines, the fact that the APP/LDLR or APPΔCT mutations that affect APP trafficking in the endocytic compartments have a direct effect on βA4(1-40) production and secretion, support the general importance of this pathway in APP processing, and imply the existence of a γ₄₀-secretase activity in the endocytic limb. Therefore, it is clear that further investigations are needed to explain the relationship between PS1 and γ₄₀-secretase activity.

Unexpectedly, blocking the processing of APP in the endocytic pathway also decreases secretion of βA4(1-42) (this study; Perez et al. 1999), whereas stimulating the endocytic recycling (APP/LDLR) selectively promotes βA4(1-40) secretion. A recent study provided evidence that the cytoplasmic tail of APP contains, in addition to endocytic signals, information that influences its metabolism in the exocytotic limb (Perez et al. 1999).

Apart from the spatial paradox, it remains to be explained why mainly the immature, N-glycosylated form of APP holoprotein has been found to interact physically with PS1 (Weidemann et al. 1997; Xia et al. 1997). One would predict a preferential association with the carboxy-terminal APP fragments if PS1 was indeed γ-secretase. It could be speculated that PS1 is involved in a posttranslational modification of the newly synthesized APP, tagging it for later recognition by the γ-secretases, although any direct evidence supporting such a possibility is lacking. Other hypotheses have been proposed as well. PS1 might control trafficking of APP to its processing compartments, or alternatively, control trafficking of the γ-secretases to APP-bearing compartments. For instance, the subtle effects on the maturation of TrkB and on BDNF-inducible TrkB autophosphorylation caused by PS1 deficiency in neurons was interpreted to reflect such a function (Naruse et al. 1998). Our findings that endogenous PS1 colocalizes with marker proteins of the IC and ER, would agree with a recruiting or activating function for PS1, either as a chaperone or an adaptor molecule bringing γ-secretase, APP carboxy-terminal fragments, and maybe other substrates (De Strooper et al. 1999) together in the right subcellular microenvironment needed for the controlled and limited proteolytic cleavage of the transmembrane domain.

In conclusion, our data indicate clearly a direct role for PS1 in γ₄₂-secretase activity in the early compartments of the biosynthetic pathway. They are compatible with the hypothesis that PS1 is γ₄₂-secretase, and imply that γ₄₂-secretase exerts its activity in close association with transport processes between ER and the cis-Golgi. However, some caution with this straightforward interpretation remains indicated, since several other observations remain difficult to integrate with this concept, most notoriously the relationship between PS1 deficiency and γ₄₀-secretase processing. Therefore, the alternative possibilities discussed above have to be further explored, and further research should clarify the exact molecular relationship between PS1 and γ₄₀- and γ₄₂-secretase activity.