Presenilin 1 regulates pharmacologically distinct gamma -secretase activities. Implications for the role of presenilin in gamma -secretase cleavage.

Presenilins (PSs) are polytopic membrane proteins that have been implicated as potential therapeutic targets in Alzheimer's disease because of their role in regulating the gamma-secretase cleavage that generates the amyloid beta protein (Abeta). It is not clear how PSs regulate gamma-secretase cleavage, but there is evidence that PSs could be either essential cofactors in the gamma-secretase cleavage, gamma-secretase themselves, or regulators of intracellular trafficking that indirectly influence gamma-secretase cleavage. Using presenilin 1 (PS1) mutants that inhibit Abeta production in conjunction with transmembrane domain mutants of the amyloid protein precursor that are cleaved by pharmacologically distinct gamma-secretases, we show that PS1 regulates multiple pharmacologically distinct gamma-secretase activities as well as inducible alpha-secretase activity. It is likely that PS1 acts indirectly to regulate these activities (as in a trafficking or chaperone role), because these data indicate that for PS1 to be gamma-secretase it must either have multiple active sites or exist in a variety of catalytically active forms that are altered to an equivalent extent by the mutations we have studied.

Presenilins (PSs) are polytopic membrane proteins that have been implicated as potential therapeutic targets in Alzheimer's disease because of their role in regulating the ␥-secretase cleavage that generates the amyloid ␤ protein (A␤). It is not clear how PSs regulate ␥-secretase cleavage, but there is evidence that PSs could be either essential cofactors in the ␥-secretase cleavage, ␥-secretase themselves, or regulators of intracellular trafficking that indirectly influence ␥-secretase cleavage. Using presenilin 1 (PS1) mutants that inhibit A␤ production in conjunction with transmembrane domain mutants of the amyloid protein precursor that are cleaved by pharmacologically distinct ␥-secretases, we show that PS1 regulates multiple pharmacologically distinct ␥-secretase activities as well as inducible ␣-secretase activity. It is likely that PS1 acts indirectly to regulate these activities (as in a trafficking or chaperone role), because these data indicate that for PS1 to be ␥-secretase it must either have multiple active sites or exist in a variety of catalytically active forms that are altered to an equivalent extent by the mutations we have studied.
The 4-kDa amyloid ␤ protein (A␤) 1 deposited in Alzheimer's disease (AD) is a normally secreted proteolytic product of the amyloid ␤ protein precursor (APP) (1)(2)(3). Generation of A␤ from APP requires two sequential proteolytic events: an initial cleavage at the amino terminus of the A␤ sequence referred to as ␤-secretase (4) and a subsequent cleavage at the carboxyl terminus known as ␥-secretase. Recently, a membrane-bound aspartic protease has been implicated as a ␤-secretase (5)(6)(7)(8). However, the protease(s) responsible for ␥-secretase cleavage have not been identified. In addition, a third proteolytic activity referred to as ␣-secretase cleaves within the A␤ sequence to release a large secreted derivative (sAPP), thus precluding formation of full-length A␤. In mammalian cells, at least two members of the ADAM family (a disintegrin and metalloprotease) can contribute to the ␣-secretase activity (9,10). Although full-length APP is not cleaved by ␥-secretase, APP carboxyl-terminal fragments (CTF) generated through cleavage of APP by either ␣or ␤-secretase are both substrates for ␥-secretase, with cleavage of CTF␣ releasing a peptide referred to as p3 (A␤17-40 or A␤17-42) (1-3).
␥-Secretase-catalyzed cleavages are of particular interest for a number of reasons. First, they are unusual in that the cleavage site of the substrate is predicted to lie within the transmembrane domain (TMD). Rather than primary amino acid sequence, position of the ␥-cleavage site with respect to the membrane appears to be the prime determinant of cleavage, with the length of the luminal TMD determining that position (11). Whether ␥-secretase actually cleaves residues within the membrane is a controversial topic. To date, there is no definitive evidence showing that any protease can cleave bonds when they are buried within a TMD. Second, altered ␥-secretase cleavage is implicated in the development of AD (reviewed in Ref. 12). FAD-linked mutations in APP, presenilin 1 (PS1), and PS2 alter ␥-secretase activity by increasing the amount of a minor A␤ species, the more amyloidogenic A␤ 42 , without significantly altering total A␤ production. Significantly, both PS1 knockout and presenilin aspartate mutants decrease ␥-secretase cleavage, but it is not known if this is a direct or indirect effect (13)(14)(15). Third, APP CTF are not the only substrate for ␥-secretase activity: CTF generated after ligand-induced cleavage of the extracellular domain of Notch are cleaved at residues near the cytoplasm/membrane junction by a ␥-secretase-like activity (16 -18). Significantly, this ␥-secretase activity appears necessary for Notch signaling and also appears to be regulated by PSs. Finally, because ␥-secretase cleavage is the final step in the generation of A␤, it remains a major therapeutic target for strategies designed to lower A␤ production. Thus, ␥-secretase is not only an unusual proteolytic activity, but its activity has important biological consequences both with respect to the pathogenesis and treatment of AD and for cell biology in general. Many  the observed effects of PSs on ␥-secretase cleavage. Recent studies of ␥-secretase activity in cells derived from PS1 knockout mice have implicated PS1 in the regulation of intracellular trafficking (19), as cofactors for ␥-secretase activity (13), or as ␥-secretases themselves (14). The latter notion that PSs may be ␥-secretase gained further support from studies demonstrating that mutation of either of two aspartate residues potentially lying in opposing transmembrane domains in both PS1 and PS2 decrease ␥-secretase activity, presumably through a dominant negative mechanism (14,15). It was thus proposed that PSs may be novel intramembranous proteases with the aspartates functioning as the catalytic residues; alternatively, PSs may function as di-aspartyl cofactors for ␥-secretase activity (14). Consistent with this hypothesis, treatment of cultured cells with either pepstatin (11), a prototypic aspartyl protease inhibitor, or a difluoroketone compound, which inhibits aspartyl proteases (20), reduces A␤ production to a similar extent as seen in PS1 knockout cell lines. However, in all of these cases A␤ production is not completely abolished, indicating that more than one protease likely contributes to ␥-secretase-catalyzed cleavages, a notion suggested by numerous studies showing differential inhibition of the ␥-40 and ␥-42 activities (Refs. 11, 21-24, and this study).
Because ␥-secretase is one of the major therapeutic targets in AD, it is essential to determine whether PSs are ␥-secretases or whether they alter ␥-secretase activity in some other fashion. It has been pointed out that one of the major problems with the "PSs-as-␥-secretases" hypothesis is the discrepancy between the predominantly ER localization of PSs and the subcellular sites of the majority of ␥-secretase activity in the more distal secretory pathway (Golgi) and endosomal system (25). In this study, we illustrate another aspect of presenilin regulation of ␥-secretase whereby PS1 appears to regulate multiple pharmacologically and spatially distinct ␥-secretase activities as well as inducible ␣-secretase activity. The major implication of these observations is that PS1 either functions indirectly to regulate these activities (as in a trafficking or chaperone role), or, if PS1 is a ␥-secretase, it must have multiple active sites or exist in a variety of catalytically active forms that are altered to an equivalent extent by the mutations we have studied.
Mutant PS Constructs-Each PS mutant was constructed by generating two polymerase chain reaction fragments from wild type (wt) PS1 cDNA template. The 5Ј-segment was produced using a forward primer to generate an HindIII site at the 5Ј-end of PS1 and a specific reverse primer from the actual site of mutation, which also created a class IIa restriction site (BsmBI). The BsmBI site is removed by the restriction enzyme leaving the final sequence unaltered except for the specific desired mutation. A similar strategy was used to generate the 3Ј-end of the cDNA, terminating in a BamHI site. Following the appropriate restriction digests, the pieces were assembled by triple ligation in the pAG3hyg vector (11). All mutations were verified by sequencing (primer sequences are available on request). In this manner we generated four PS1 aspartate mutants (D257A or E, D385A or E; D 3 A/E), the PS ins254 -6ϩins386 -8 (PS "out") mutant, which contains amino acid insertions of SVY at position 256 and FIF at position 388 (duplicating the amino-and carboxyl-terminally adjacent 3 residues, respectively), and ⌬TM1-2, which contains a deletion from amino acids 81-154 (inclusive).
Generation of Pooled Stable Lines-70% confluent 6-well plates of either Chinese hamster ovary (CHO), human embryonic kidney 293 (HEK), or human neuroglioma (H4) cells were transfected with 1 g of plasmid DNA preincubated with 3 l of FuGene 6 transfection reagent (Roche Molecular Biochemicals) in serum-free OptiMEM (Life Technologies, Inc.) overnight. Media were then replaced with either Ham's F-12 (CHO cells), Dulbecco's modified Eagle's (HEK cells), or OptiMEM (H4 cells) media supplemented with 10% fetal bovine serum (Hyclone) and 800 g/ml Hygromycin B (Calbiochem). After 10 -14 days, selection was reduced to 200 g/ml. Expression levels were monitored periodically by immunoblotting throughout the course of the experiment, and have been maintained for multiple passages. Transient transfections were performed in an identical manner, albeit without hygromycin selection. All cell lines were maintained at 37°C under 5% CO 2 .
A␤ Analysis-A␤ was analyzed by sandwich enzyme-linked immunosorbent assay (ELISA) and immunoprecipitation/mass spectrometry (IP/MS) as described previously (11). Briefly, serum-free media samples were collected following overnight conditioning, complete protease inhibitor cocktail added (PIC; Roche Molecular Biochemicals), and total A␤ concentration was measured by 3160/4G8 sandwich ELISA. The amount of A␤ was standardized (to correct for transfection efficiency) to sAPP levels measured in the same sample using an Ab207 competitive ELISA with plates coated with 60 ng/well purified, recombinant sAPP (11). A␤ 40 and A␤ 42 were measured by BAN50/BA27 and BAN50/BC05 ELISA, respectively. All measurements were performed in duplicate. For mass spectrometry, conditioned serum-free media from APP-transfected CHO cells was immunoprecipitated with 4G8/protein-A/G-plusagarose beads and subjected to matrix-assisted laser desorption/timeof-flight mass spectrometry analysis as described (26).
Metabolic Labeling-Confluent 6-well plates of pooled stable HEK293 cells or HEKwt cells were labeled for 2 h with 200 Ci/well [ 35 S]methionine/cysteine. For inducible ␣-secretase experiments, cells were treated with 1 M phorbol 12,13-dibutyrate (PDBu), and serumfree conditioned media were collected after 6 h. Total sAPP was immunoprecipitated with Ab207/protein-G-agarose beads, separated on 10% Tris-Tricine gels, dried, and exposed to a low energy phosphor screen for 7 days. Data were analyzed using a Storm PhosphorImager (Molecular Dynamics) and ImageQuaNT software. For analysis of CTF, APP from Triton X-100-lysed cells was immunoprecipitated with rabbit anti-CT-20, separated on 10 -20% Tris-Tricine gels, and analyzed as above.

␥-40 and ␥-42 Secretase Activities Are Pharmacologically and
Spatially Distinct-To further establish the existence of pharmacologically distinct ␥-secretase cleavage sites, we examined the effects of a substrate-based (␥40-site) aldehyde inhibitor (Boc-glycine-valine-valinal; GVV) on A␤ production and APP CTF. When tested on transiently transfected 293T cells overexpressing APP695NL, GVV increased APP CTF in a dosedependant fashion (data not shown) and selectively inhibited A␤ cleavage at sites other than 42, as measured both by ELISA (Fig. 1A) and (IP/MS) (Fig. 1B). The selective inhibition of cleavages other than A␤ 42 by GVV is qualitatively similar to the effects on A␤ production observed with pepstatin (11), other peptide aldehydes (23), and some difluoroketone compounds (21); however, GVV appears to be even more selective than those previously reported with doses that inhibit A␤ 40 by Ͼ90% while not inhibiting A␤ 42 production. Significantly, we have recently observed selective inhibition of A␤ 40 production by GVV, pepstatin, and other peptide aldehyde inhibitors in an in vitro assay of ␥-secretase activity, 2 indicating that the selectivity is not simply an issue of differential cell penetrance of the inhibitor. Thus, the elevations in A␤ 42 production seen at low concentrations are best explained by an increased availability of APP CTF to a distinct ␥-42 activity that is less sensitive to inhibition by GVV.
Subsequent analysis of the effects of GVV treatment on ␥-secretase cleavage in APP TMD mutants by ELISA (data not shown) and IP/MS (Fig. 1B) indicates that these mutants pos-sess a variety of GVV-sensitive and -insensitive sites, further indicating that multiple proteolytic activities are involved in ␥-secretase cleavage (11). Interestingly, these data also revealed that those GVV-sensitive sites were closer to the luminal side of the APP TMD, whereas those GVV-insensitive sites were more distal. Thus, it appears that we can monitor at least two ␥-secretase activities, a "␥-40 activity," which cleaves residues more proximal to the luminal side and a ␥-42 activity, which cleaves residues more distal to the luminal side. A, inhibitor profile for GVV. CHO cells stably expressing human wt ␤APP were treated overnight with the indicated concentrations of GVV, and serum-free-conditioned media samples were collected with protease inhibitors. Total A␤ production as assayed by 3160/4G8 ELISA is inhibited in a dose-dependent manner, whereas A␤ 42 is slightly increased at lower concentrations and relatively unaffected at concentrations that nearly eliminate production of shorter A␤ species. No obvious toxicity was apparent at these concentrations. Total sAPP secretion was unaffected by GVV (data not shown). Results shown are averaged from three experiments, and all ELISA measurements were performed in duplicate. B, schematic of ␥-secretase cleavage site utilization and GVV sensitivity of APP TMD mutants. The amino acid sequence of the APP TMD is shown. Mutations (including the insertion) are shown in blue; the position of the GAII deletion is indicated with the arrowhead. Mass spectrometric analyses of A␤ peptide secreted by the TMD mutants are schematically depicted. The carboxyl terminus of the A␤ peptides detected in each mutant are indicated by arrows above the cleavage sites (all A␤ species detected by the methods used in this study begin at A␤1), whereas the height of the arrows indicates relative peptide amount. Black arrows indicate a cleavage inhibited Ͼ50% by 250 M GVV, red arrows indicate cleavages increased by GVV, and the orange arrow indicates a cleavage inhibited Ͻ50%. These effects are similar to those obtained with pepstatin in these same APP TMD mutants (11).
identified two additional PS1 mutants that inhibit ␥-secretase (Fig. 2). The first mutant, PS1 ins254 -6ϩins386 -8 (PS1 out), was generated to assess whether location of the aspartates relative to the membrane was an important functional element. In this mutant, the aspartates are not mutated, but their positions shifted by two 3-amino acid insertions within the transmembrane domain, potentially moving both Asp-257 and Asp-385 toward the cytoplasm and out of phase with the potential catalytic site. In the second mutant, ⌬TM1-2, the first two transmembrane domains of PS1 are deleted, theoretically preserving the topology of the remaining protein. The inhibitory effect of this mutant on ␥-secretase was identified during a screen for PS1 deletion mutants that altered A␤ production.
We examined the effects of PS1 D3 A/E, ⌬TM1-2, and PS out mutants on the ␥-secretase cleavages in three different cell lines: H4, HEK, and CHO. In each of these, we established pooled stable lines expressing wild type PS1 or the PS1 mutants, and then assessed APP processing and A␤ production after transient transfection with various APP expression constructs. To validate this system of transient transfection of APP into pooled stable PS1 lines, we performed extensive experiments on pooled stable lines expressing the FAD-linked PS1 mutants M139V and E280G. An example of this validation for H4 cells is shown in Fig. 3. In these experiments, both M139V and E280G are overexpressed (Fig. 3A) and increase A␤ 42 production without significantly altering total A␤ production (Fig. 3B). Fig. 3 also shows the results obtained for PS1 mutants that inhibit ␥-secretase activity in H4 cells. In lines expressing the ⌬TM1-2 mutant, limited endoproteolysis of the smaller holoprotein (Fig. 3A) is indicated by the presence of a truncated amino-terminal fragment migrating at ϳ23 kDa. The use of pooled stable lines prevents us from precisely determining both the extent of cleavage and replacement of endogenous PS1. However, it is clear that overexpressed PS1 to some extent replaces endogenous PS1 (as indicated by decreased levels of endogenous PS1 NTF in the case of ⌬TM1-2), and aspartate mutants are cleaved less efficiently (D257E shown). This was confirmed by blotting with a human-specific PS1 antibody (Chemicon; not shown). Significantly, following transient transfection with APP695NL, A␤ production (total, A␤ 40 and A␤ 42 ) is greatly curtailed in these cells (Fig. 3B), and both the ⌬TM1-2 and PS1 out mutants (not shown) were at least as effective at inhibiting A␤ production as were the aspartate mutations.
To further investigate the effect of these PS1 mutants on ␥-40 and ␥-42 activities, pooled stable CHO and HEK cells overexpressing either PS1wt or one of the mutant constructs (Fig. 4) were transiently transfected with either APP695NL or the APP TMD mutants I637P, T639K, ins625-628, or del625-628. PS1 expression, total A␤, sAPP, and APP CTF were analyzed in these transfected cells. As in H4 cells, overexpression of these PS1 mutants resulted in varying degrees of replacement of endogenous PS1 and inefficient endoproteolysis of the PS1 holoprotein. APP CTF accumulated both at steady state (Fig. 5A) and during a 2-h chase period after metabolic labeling, whereas no effect was observed on constitutive sAPP secretion or turnover of APP holoproteins (data not shown). A␤ production from the APP695NL and APP TMD mutants was reduced dramatically when these constructs were expressed in CHO or HEK cells coexpressing these PS1 mutants (Fig. 5B), and IP/MS analyses indicated that all ␥-secretase cleavages were inhibited equally irrespective of inhibitor sensitivity (Fig.  5C). These data indicate that these PS1 mutations regulate production of A␤ produced by both the pepstatin/GVV-sensitive and pepstatin/GVV-insensitive ␥-secretases.
PS1 Mutants That Inhibit ␥-Secretase Are Incorporated into High Molecular Weight Complexes-One possible explanation for the inhibition of ␥-secretase by these mutants is their ability to replace endogenous PS1 but not be incorporated into

FIG. 3. Overexpression of PS1 mutants in H4 cells.
A, lysates from H4 human neuroglioma cells (10 g of total protein) were separated, transferred, and probed with a combination of anti-PS1-loop/ anti-PS1-N. As expected, PS1 holoprotein from ⌬TM1-2 is noticeably smaller than wild type PS1. Replacement of endogenous PS1 is not complete in this cell line. Also, ⌬TM1-2 is cleaved in H4 cells (smaller ϳ23 kDa ⌬TM1-2 derived NTF marked with an asterisk), whereas D257E is not cleaved efficiently. B, ELISA analysis of secreted A␤ from H4 cells transfected with APP. The reduction in A␤ for total, A␤ 40 , and A␤ 42 is equivalent for both D257E and ⌬TM1-2 (t test versus PS1wt: * ϭ p Ͻ 0.05). Two PS1 FAD mutants, E280G and M139V, are shown as positive controls for the pooled stable approach and ELISA measurement of A␤ 42 . functional high molecular weight complexes. To explore this possibility, we examined digitonin lysates from PS1wt, ⌬TM1-2, and D385A cell lines by glycerol density gradient centrifugation. To facilitate analysis, lysates from cells expressing the PS1 wild-type protein were used as an internal control and combined with an equal amount of lysate from the ⌬TM1-2 cells. Analysis of these gradients revealed that the full-length ⌬TM1-2 was incorporated into a defined complex with an apparent molecular mass of ϳ150 -250 kDa (Fig. 6). Similar results were obtained for the D385A cell lines (data not shown). Both mutant proteins were distributed in a comparable manner to that of the PS1wt holoprotein. In these gradients, the endoproteolytic fragments overlapped with the holoproteins, making it difficult to discern if the mutant complex was completely matured. Although additional studies may be required to ascertain whether this represents a fully functional complex, the mutants were not abnormally distributed and appeared to be handled similar to the wild-type protein.
PS1 Aspartate Mutants do Not Augment Inducible ␣-Secretase Cleavage-Although a role of PSs in regulating ␥-secretase activity is well established, several studies also suggest that PS1 may regulate the phorbol ester-inducible ␣-secretase cleavage that appears to be carried out by members of the ADAM family of metalloproteases (30 -32). Two studies showed that overexpression of PS1 augments inducible ␣-secretase cleavage of APP, whereas overexpression of FAD-linked mutant PS1 does not (31,32). Additionally, PS1 knockout abolishes inducible ␣-secretase cleavage of APP (30). To determine whether PS1 aspartate mutants also inhibit inducible ␣-secretase cleavage, wild-type CHO and HEK lines were compared with those overexpressing PS1wt, M139V, Asp-257, or D385E treated with 1 M of the phorbol ester compounds phorbol FIG. 4. Overexpression of PS1 mutants in CHO and HEK cells. Western blot analysis of PS1 in CHO-and HEK293-pooled stable cell lines used in this study (anti-PS1-N antibody). Overexpression of PS1 holoprotein is clearly visible as an increase in full-length PS1 band intensity. Although the ⌬TM1-2 is not endoproteolyzed efficiently, the presence of the 23-kDa ⌬TM1-2-derived NTF (asterisk) indicates that it is cleaved to some degree.
FIG. 5. Effects of PS1 mutants on A␤ production. A, steady-state accumulation of APP CTF generated by ␤-secretase (C99) and ␣-secretase cleavage (C83) in PS1 aspartate mutants (blotted with . B, ELISA analysis of total A␤. Pooled stable CHO and HEK lines stably expressing PS1wt, PS1 aspartate mutants, PS out, and ⌬TM1-2 were transiently transfected with the indicated APP TMD mutant or APP695NL control, and the total A␤ and the sAPP were assayed by ELISA as described under "Experimental Procedures." The amount of secreted A␤ relative to the corresponding PS1wt control was markedly reduced in every case (averaged from three independent experiments; error bars indicate S.D.). C, IP/MS analysis of A␤ produced from APP TMD mutants in PS1 aspartate mutant cells. No differences were seen among the MS profiles from the different PS1 mutant lines. Schematic data from APP695NL and I637P are shown. The MS profile of A␤ peptides is essentially identical in both APP695NL and I637P for both PS1wt and PS1 mutant-expressing lines, indicating that the quantitative reduction of A␤ by the PS1 aspartate mutants alters both pepstatin/GVV-sensitive and pepstatin/GVV-insensitive ␥-secretase cleavages equivalently.
12-myristate 13-acetate or PDBu for 4 -6 h. CHO cells were transiently transfected with APP695wt, and sAPP was measured by ELISA (not shown), whereas endogenous levels of sAPP were evaluated in HEK cells following metabolic labeling with [ 35 S]methionine and IP with Ab207 (Fig. 7). As is the case with the PS1 FAD-linked M139V mutation, the PS aspartate mutants abolish the effect of PS overexpression on phorbol ester treatment.

DISCUSSION
The study of ␥-secretase activity using the substrate-based inhibitor GVV further extends and confirms our previous observations that there are at least two distinct ␥-secretase activities (11): a ␥-40 activity, which cleaves more proximal to the ectodomain of APP and is more sensitive to these inhibitors, and a ␥-42 activity, which cleaves more distal residues and is less sensitive. Based on these data we have explored the regulation of these distinct ␥-secretase activities by PS1 mutants.
As previously reported, we have found that expression of PS1 aspartate mutants inhibit ␥-secretase cleavage of APP CTF (33). In addition, we have identified two other PS1 mutants whose expression inhibits A␤ production. Both PS1 out and ⌬TM1-2 mutants inhibit ␥-secretase as effectively or more effectively than the PS1 D3 A/E mutants indicating (i) that the position of the aspartates within presumptive transmembrane domains is critical for proper PS1 function and (ii) that the alterations of PS1 structure at sites relatively distant from the hypothesized "active site" can result in an apparent dominantnegative effect on ␥-secretase activity. Though there is no definitive evidence that identifies the mechanisms by which these mutant PS1 proteins inhibit ␥-secretase activity, the finding that these proteins can replace endogenous PS1 and are incorporated into high molecular weight complexes is consistent with the proposed dominant-negative mechanism of action, although additional experiments will clearly be needed to support that claim. Furthermore, we did not observe a consistent absence of PS1 endoproteolysis accompanying the decrease in A␤ production (e.g. with ⌬TM1-2), indicating that this effect may not be related to PS1 holoprotein cleavage as was previously implied (14). This is consistent with other findings that the pathological effect of the PS1 ⌬exon 9 mutation is not related to the defect in endoproteolysis exhibited by this mutant (13,34).
Our finding that distinct ␥-40 and ␥-42 activities are equally inhibited by PS1 aspartate mutants is difficult to reconcile with the hypothesis that PS1 is a ␥-secretase with active site residues at Asp-257 and Asp-385. Based on data in this report and our previous studies of these APP TMD mutants, if PSs were novel intramembranous aspartic proteases responsible for ␥-secretase cleavage, we would predict that the effects of these mutations on A␤ production would be quite different from those observed. Instead of inhibiting all cleavages to a similar extent, they would only inhibit cleavage at pepstatin/GVV-sensitive sites. To illustrate this it is useful to focus on the I637P APP mutant. The PS1 mutants used in this study decrease both the pepstatin/GVV-sensitive cleavage at A␤37 and the pepstatin/ GVV-insensitive cleavage at A␤ 43 to an equal extent, a finding that contrasts dramatically with the effects of both pepstatin and GVV treatment, which slightly decreases A␤ 37 production and increased A␤ 43 production. It should be mentioned that it is not possible to easily account for these data by postulating that PS1 is one ␥-secretase and PS2 is the other, because knockout of either PS1 (which reduces both A␤ 40 and A␤ 42 equally) or PS2 (which has no effect on either A␤ species) does not specifically alter ␥-40 or ␥-42 activity (13,34).
Of course, it is still possible that PS1 is ␥-secretase, but for this to be the case then they must have either multiple active sites or multiple active conformations. In any case, one of these sites or conformations must then be responsible for the ␥-40 activity and the other for the ␥-42 activity. Given that PSs have no homology to any known protease, if PSs contain the active site(s) of ␥-secretase then they represent a truly unprecedented class of proteolytic enzyme. Furthermore, based on some of the same criteria that have lent credence to the PS-as-␥-secretase hypothesis (e.g. PS1 knockout and the effect of aspartate mu-FIG. 6. Mutant PS1 is incorporated into high molecular weight complexes. Glycerol velocity gradient centrifugation of ⌬TM1-2. Digitonin lysates were separated on a linear glycerol gradient and SDSpolyacrylamide gel electrophoresis as described (29). Immunoblots were probed with antibody NT1 to PS1 residues 41-49 (29). The incorporation of the ⌬TM1-2 PS1 mutant into high molecular weight complexes was essentially unchanged from that of PS1wt.
FIG. 7. PS1 mutations attenuate inducible ␥-secretase activity. A, pooled stable HEK293 cell lines were labeled, treated with PDBu (ϩ) or vehicle alone (Ϫ), and serum-free-conditioned media was collected and endogenous sAPP was immunoprecipitated with Ab207. The experiment was performed twice, in triplicate; a representative 7-day phosphorimage is shown. B, sAPP bands from PDBu-treated cells were quantified, expressed as a percentage of control sAPP, and compared with PS1wt values by analysis of variance followed by Dunnett's test. There were no significant differences between the two experiments, and constitutive levels of sAPP did not differ between the various cell lines. Overexpression of PS1wt potentiated the effect of PDBu on sAPP secretion, whereas overexpression of mutant forms of PS1 (both aspartate mutant and the FAD-linked M139V) did not (* ϭ p Ͻ 0.05; ** ϭ p Ͻ 0.01). tants on ␥-cleavage), one could postulate that PSs are also the inducible ␣-secretase enzymes. This is obviously extremely unlikely, because two of the enzymes (TACE and ADAM-10) at least partially responsible for inducible ␣-secretase activity have been identified (9,10).
If PSs are not ␥-secretase, then how might they influence both the inducible ␣-secretase cleavage of APP and distinct ␥-secretase activities? We cannot rule out the possibility that PSs are proteases that cleave and activate multiple pharmacologically distinct ␥-secretases and/or inducible ␣-secretases in an upstream manner. However, in the absence of any direct evidence that PSs have proteolytic activity of any sort, we also find this explanation unlikely. PSs could play a role in "positioning" APP for inducible ␣-secretase cleavage and APP CTF for ␥-secretase cleavage, but evidence for an interaction of PS with mature APP (the substrate for ␣-secretase cleavage) and APP CTF (the substrate for ␥-cleavage) is inconsistent (19,35,36). It is possible that PSs are essential cofactors or chaperones for proteases involved in APP cleavage. If this is the case, then they must presumably interact not only with several ␥-secretases but also with the proteases responsible for phorbol esterinducible ␣-secretase activity. Yet another possibility is that PSs play a role in the normal trafficking of APP CTF after constitutive cleavage by ␣and ␤-secretase and are also required for appropriate trafficking of APP to the site where phorbol ester-inducible ␣-secretase cleavage occurs. Both of these latter possibilities are consistent with hypothesized roles of PSs in trafficking a select set of membrane proteins (19,37). Although no alterations in the subcellular distribution of APP have yet been observed in PS1 knockout cells, the amount of APP destined for processing either to A␤ or by inducible ␣-secretase activity is by most estimates only a relatively small fraction of total cellular APP. If PSs function in this manner, they are likely to regulate a specific pool of APP and APP CTF and alterations may not be readily apparent in relatively crude subcellular fractionation studies.
Ultimately, to settle the question of whether or not PSs are ␥-secretase, it will be necessary to either (i) prove that PSs are catalytic by showing that the purified protein has ␥-activity or (ii) show that another protease catalyzes ␥-cleavage. Nevertheless, given the evidence for complex interaction of PSs with a growing list of important cellular proteins, including several catenins (28, 38 -41), Bcl-X-L (42), Notch (16 -18, 43, 44) GSK3␤ (41,45), rab GDP dissociation inhibitor (46), E-cadherin (47), and a potential functional role in the unfolded protein response (48,49), specific targeting of the A␤-regulating activity of PSs may prove very difficult.