Active γ-Secretase Complexes Contain Only One of Each Component*

γ-Secretase is an intramembrane aspartyl protease complex that cleaves type I integral membrane proteins, including the amyloid β-protein precursor and the Notch receptor, and is composed of presenilin, Pen-2, nicastrin, and Aph-1. Although all four of these membrane proteins are essential for assembly and activity, the stoichiometry of the complex is unknown, with the number of presenilin molecules present being especially controversial. Here we analyze functional γ-secretase complexes, isolated by immunoprecipitation from solubilized membrane fractions and able to produce amyloid β-peptides and amyloid β-protein precursor intracellular domain. We show that the active isolated protease contains only one presenilin per complex, which excludes certain models of the active site that require aspartate dyads formed between two presenilin molecules. We also quantified components in the isolated complexes by Western blot using protein standards and found that the amounts of Pen-2 and nicastrin were the same as that of presenilin. Moreover, we found that one Aph-1 was not co-immunoprecipitated with another in active complexes, evidence that Aph-1 is likewise present as a monomer. Taken together, these results demonstrate that the stoichiometry of γ-components presenilin:Pen-2:nicastrin:Aph-1 is 1:1:1:1.

The ␥-secretase complex hydrolyzes within the transmembrane region of type I integral membrane proteins, including the ␤-amyloid precursor protein (APP) 2 and the Notch recep-tor, and plays critical roles in biology and human disease. This enzyme is composed of four membrane proteins, presenilin (PS), Pen-2, nicastrin (NCT), and Aph-1 (1), which are necessary and sufficient for protease activity. PS is the catalytic component of ␥-secretase, with two conserved transmembrane aspartates that form the active site (2). More than 100 missense mutations in PS are associated with early onset, familial Alzheimer disease (FAD), and these mutants shift the cleavage sites within APP to alter the lengths of the products, the amyloid ␤-peptide (A␤) (3), and the APP intracellular domain (AICD) (4). NCT is involved in the recognition of substrates, interacting with their luminal/extracellular N termini (5), whereas Aph-1 is thought to be a scaffold for assembly of the complex (6). Pen-2 triggers the endoproteolysis of PS into N-terminal and C-terminal fragments (NTF and CTF) as part of the maturation and activation of the protease complex (7) and stabilizes the catalytic subunit PS NTF/CTF after endoproteolysis (8).
Although the essential components of ␥-secretase were all known by 2003, the stoichiometry of the complex remains unknown. Some evidence suggests that two PS molecules reside in a single protease complex, leading to the suggestion that a PS dimer is at the catalytic core of ␥-secretase and/or that substrates are cleaved at the dimer interface (9 -11). Recently reported low resolution electron microscopy structures of ␥-secretase (at 15 and 48 Å (Refs. 12 and 13, respectively)) could not distinguish between the one-PS and two-PS models. Native gel electrophoresis, glycerol velocity gradient centrifugation, and size exclusion chromatography experiments have given variable results, none of which have clarified this issue (14 -18). The distant presenilin homolog signal peptide peptidase (19) can apparently form stable dimers (20), although in our hands, purified, active, bacterially expressed signal peptide peptidase is not observed as a dimer (48). Although the question of PS representation in the complex is unclear, that of other members of the complex has been studied even less. Here we address the stoichiometry of active ␥-secretase by co-immunoprecipitation (co-IP) and Western blot methods and show that the complex contains only one of each component. How these find-ings relate to results from native gel electrophoresis analysis and the critical importance of detergents is also discussed.

MATERIALS AND METHODS
DNA Constructs and Cell Lines-The cDNA of His-FLAG-⌬E9 was inserted at the restriction enzyme sites between KpnI and NotI into pcDNA6/V5-His (Invitrogen). This vector was transfected into ␥-30 cells (17) with Lipofectamine 2000 (Invitrogen), and the transfected cells were selected with blasticidin S (Invitrogen). The cDNA of N-terminally Myc-or HAtagged PS1 was inserted at the restriction enzyme sites between HindIII and XhoI into pMX-IZ or pMX-IB, in which the green fluorescent protein gene of pMX-GFP (21) is replaced with a resistance gene of zeocin or blasticidin, respectively. These vectors were transfected into Plat-E cells (22) with Effectene (Qiagen), and the resultant retroviruses in the condition media were collected by filtration and centrifugation and frozen as a viral stock. PS double knock-out (DKO) cells (23,24) were infected with these viruses supplemented with 4 g/ml of polybrene (Sigma), and the infected cells were selected with zeocin and blasticidin. The cDNA of C-terminally HA-or FLAG-tagged Aph-1 was inserted at the restriction enzyme sites between HindIII and EcoRI into pcDNA3.1(ϩ) or pcDNA6/V5-His, respectively. These vectors were transfected into Chinese hamster ovary (CHO) cells with Lipofectamine LTX (Invitrogen), and the transfected cells were selected with zeocin and blasticidin. All of the cell lines were maintained in Dulbecco's modified Eagle's medium supplemented with fetal bovine serum and penicillin-streptomycin.
Preparation of Detergent-solubilized Membrane Fractions and ␥-Secretase Assays-The cells in 15 cm dishes were collected and resuspended in homogenization buffer (50 mM Hepes, pH 7.0, 250 mM sucrose, 5 mM EDTA) containing complete protease inhibitor mixture (Roche Applied Science). The resuspended cells were collapsed by passing once through a French press at 1,000 p.s.i., and the cell debris and nuclei were removed at 3,000 ϫ g for 10 min. The supernatants were further centrifuged at 100,000 ϫ g for 1 h, and the resultant pellets were washed with 0.1 M sodium bicarbonate (pH 11.4) and then centrifuged again. The membrane pellets were solubilized with 1% digitonin or CHAPSO in homogenization buffer for 90 min on ice and then centrifuged at 100,000 ϫ g for 1 h. The protein concentration of the supernatant was determined with BCA protein assay reagent (Pierce). CHAPSO-solubilized membrane fractions were diluted to 0.25% and incubated with C100FLAG (25) at 37°C for 2 h. The reaction was stopped by adding 2ϫ SDS sample buffer containing 5% 2-mercaptoethanol. The samples were normalized for protein levels in the solubilized membrane and subjected to SDS-PAGE (16% Tricine gel (Invitrogen)) and Western blot, with A␤ and AICD detected with mouse monoclonal antibody 6E10 (Covance Research Products) and M2 (Sigma) antibodies, respectively.
IP of Active ␥-Secretase-1% CHAPSO-solubilized membrane fractions of DKO cells expressing Myc-wt PS1 and HA-wt PS1 were prepared to 0.25% CHAPSO and 0.02% Nonidet P-40. The prepared membrane fractions were incubated at 37°C for 30 min in the presence of Me 2 SO as a control vehicle, C100FLAG, 10 M of 31C (25), 10 M of compound 11 (34), or all of them. After incubation, ␥-secretase was immunoprecipitated overnight with anti-Myc or anti-HA antibody. The resultant pellets were washed with 0.25% CHAPSO and 0.02% Nonidet P-40 three times and subjected to Western blot as described above. To label ␥-secretase with a photoaffinity reagent, solubilized membrane fractions of 0.25% CHAPSO and 0.02% Nonidet P-40 were incubated with 500 nM of biotinylated inhibitor photoprobe III-63 (35) in the presence of Me 2 SO or 10 M of parent inhibitor 31C and then irradiated for 30 min at 350 nm. Irradiated samples were immunoprecipitated overnight with anti-Myc or anti-HA antibody, and the resultant pellets were washed and subjected to Western blot. The labeled PS was detected with rabbit polyclonal anti-biotin antibody (Bethyl).
Quantification of ␥-Components-␥-Secretase was immunoprecipitated from solubilized membrane fractions of DKO cells expressing Myc-and HA-tagged wt PS1 or HEK293 cells expressing endogenous PS1 with mouse monoclonal anti-Myc antibody or rat monoclonal PS1 antibody (Millipore). The immunoprecipitated Myc-PS1 was subjected to SDS-PAGE (10 -20% Tricine gel (Invitrogen)) and Western blot and then quantified with rabbit polyclonal anti-Myc antibody (Sigma) by using N-terminally Myc-tagged ubiquitin (Boston Biochem) to generate a standard curve. To quantify the immunoprecipitated wt PS1 from HEK293 cells, the different amounts of quantified Myc-PS1 were further used as a standard curve and detected with AB14. To quantify co-immunoprecipitated endogenous Pen-2, different amounts of FLAG-Pen-2 immunoprecipitated from digitonin solubilized membrane fractions of S-1 cells (38) were quantified by using FLAG-tagged ubiquitin (Boston Biochem) as a standard curve with M2 antibody. The various quantified FLAG-Pen-2 samples were next used as standards to quantify endogenous Pen-2 with rabbit polyclonal ab24743 (Abcam). To quantify co-immunoprecipitated endogenous NCT, different amounts of NCT-GST immunoprecipitated from solubilized membrane fractions of S-1 cells (38) were quantified by using GST-E1 (Boston Biochem) to generate a standard curve with rabbit polyclonal GST antibody (Invitrogen). The quantified NCT-GST samples were next used as standards to quantify endogenous NCT with rabbit polyclonal anti-NCT. Software, ImageJ, was used to quantify the intensity of immunoblot signals.

RESULTS
To determine whether two PS molecules are present in a single ␥-secretase complex, we stably expressed two different and distinquishable forms of PS1 in cells. One FAD mutant form of PS1, ⌬E9 (deletion of exon 9), does not undergo endoproteolysis into NTF and CTF and can be readily discriminated against wt PS1 by Western blot (39). We transfected His-FLAG-⌬E9 (or its aspartyl mutants TM6DC and TM7DC, in which the active site aspartates of transmembrane domains 6 and 7, respectively, are mutated to cysteine) into ␥-30 cells. ␥-30 cells are CHO cells that stably express human proteins APP, wt PS1, FLAG-Pen-2, and Aph-1a-HA (17). Thus, the transfected ␥-30 cells stably express both His-FLAG-⌬E9 and untagged wt PS1 (␥-30 ϩ ⌬E9 cells).
The membrane fractions of these cells were solubilized with the zwitterionic detergent CHAPSO, and IPs were performed with anti-His antibodies. After washing the resultant pellets, we analyzed PS1 and other members of the complexes by Western blot. Although the parent cell line ␥-30 does not express any His-tagged proteins, wt PS1 and NCT were pulled down with anti-His antibody and protein G beads (supplemental Fig. S1A), as was active ␥-secretase (supplemental Fig. S1B), suggesting that ␥-secretase in CHAPSO-solubilized membrane fractions strongly binds to beads nonspecifically and that CHAPSO, commonly used in the study of ␥-secretase, is not an appropriate detergent for evaluating the stoichiometry of the complex.
We then identified digitonin as a detergent that eliminates the nonspecific binding but conserves activity. When HAtagged Aph-1a of ␥-30 cells was immunoprecipitated with anti-HA antibody, all components of the complex were coimmunoprecipitated, whereas anti-His antibody did not pull down any (Fig. 1A, lanes 2 and 3; a small amount of Aph-1a-HA is brought down, probably because of the highly hydrophobic nature of this protein). On the other hand, when His-tagged ⌬E9 was immunoprecipitated from ␥-30 ϩ ⌬E9 cell membranes with anti-His antibody, all of the other components co-immunoprecipitated, but wt PS1 (as NTF and CTF) did not (Fig. 1A, lane 5). Catalytically inactive PS ⌬E9 mutants TM6DC and TM7DC likewise assembled into full, mature complexes pulled down by anti-His antibodies but without wt PS1 NTF or CTF co-immunoprecipitating ( Fig. 1A, lanes 7 and 9).
The A␤ species so produced were immunoprecipitated with 6E10 and analyzed by MALDI-TOF MS. Whereas the A␤ species produced by wt PS1 ranged from A␤38 to A␤45, ⌬E9 produced mainly A␤43 with small portions of A␤45 and A␤46 (Fig.  1D), indicating that the upper band contains species of A␤43 and longer. ⌬E9 can produce small amounts of shorter A␤ species (see Fig. 3B, lane 8); however, overall activity was much lower than wt PS. These observations are consistent with previous reports that FAD PS mutants, including ⌬E9, cleave substrates less efficiently and generate longer forms of A␤ (40,41). Importantly, ⌬E9-containing ␥-secretase complexes were proteolytically active upon co-IP, with no wt PS1 NTF and CTF present.
We confirmed the above findings by carrying out the reverse experiment, pulling down wt PS1 using a different antibody, B17, which recognizes wt PS1 but not ⌬E9 (26). Although wt PS1 CTF was detected with B17, ⌬E9 was not, even when concentrated by IP ( Fig. 2A, bottom panel). IP of ␥-complexes con-taining wt PS1 with this antibody did not bring down any ⌬E9 (Fig. 2B), indicating that none of the isolated complexes contained PS-PS dimers of wt PS1 and ⌬E9. In digitonin-solubilized membrane fractions, blue native gel showed full ␥-complexes containing PS1 and NCT with mobilities of ϳ500 kDa (Fig. 2C, lanes 1-4), although note that ␥-complexes in CHAPSO-solubilized membrane fractions have slower mobilities (lanes 5-6). These observations are consistent with previous reports (16,18); however, the apparent large size is not due to the presence of two or more PS molecules/complex. One is sufficient for an active ␥-secretase complex.
These findings were elicited by using wt PS1 and ⌬E9; however, these two PS variants may not interact because of their differences. wt PS1 undergoes endoproteolysis, and ⌬E9 does not. To further investigate the possibility of a PS-PS interaction within ␥-complexes, we co-expressed Myc-tagged PS1 and HAtagged PS1 in PS1/2 DKO cells and performed co-IP experiments (Fig. 3A). Digitonin-solubilized membranes from these PS-expressing cells showed the same size of ␥-secretase by blue native PAGE as shown in Fig. 2C (supplemental Fig. S2A). IP with anti-Myc antibodies brought down ␥-secretase activity (Fig. 3B). A small amount of background activity was brought down with beads alone (i.e. without antibody; e.g. lane 3). Immunoprecipitated ␥-secretase containing wt PS1 produced A␤ and AICD (lanes 4 and 6), whereas isolated complexes containing ⌬E9 again possessed low activity and produced longer A␤ species (lane 8), similar to that shown in Fig. 1B, indicating that these immunoprecipitated ␥-complexes maintain their integrity and protease activity. Co-IPs with anti-HA antibody brought down activity as well (data not shown).
We next examined the co-IPs isolated with each antibody. When Myc-tagged wt PS1 or ⌬E9 was immunoprecipitated with anti-Myc antibody (Fig. 3C), HA-tagged wtPS1 or ⌬E9 did not co-IP (cf. lanes 5 versus 6, lanes 8 versus 9, and lanes 11 versus 12, middle panel), whereas mature, glycosylated NCT (the form found in active protease complexes) did (bottom panel). Likewise, when HA-tagged wt PS1 or ⌬E9 was immunoprecipitated with anti-HA antibody (Fig. 3D), Myc-tagged wt PS1 or ⌬E9 did not co-IP (cf. lanes 5 versus 6, lanes 8 versus 9, and lanes 11 versus 12, top panel), although mature NCT did (bottom panel). These results demonstrate that active ␥-secretase can be isolated in the absence of interactions between two wt PS molecules or between two ⌬E9 molecules, each with different N-terminal tags.
binding of ␥-secretase and that ␥-secretase was intact, of similar size by BN-PAGE (data not shown), and active under these conditions (Fig. 4A, lanes 4 and 8). We incubated the solubilized membranes for 30 min with substrates (C100FLAG) or inhibitors (31C, a transition state analog inhibitor (25), or Compound 11, a helical peptide inhibitor (34)), and then immunoprecipitated ␥-complexes with anti-Myc antibody. After IP, although Myc-tagged PS1 and NCT were co-immunoprecipitated (Fig. 4B, top and bottom panels), HA-tagged PS1 showed as background levels (middle panel, background lane 7, beads alone), indicating that HA-tagged PS was not co-immunoprecipitated with Myc-tagged PS1 in the presence of substrates or inhibitors. Likewise, Myc-tagged PS1 was not co-immunoprecipitated with HA-tagged PS1 (data not shown). Substrates or inhibitors might have been washed out from the co-immunoprecipitated ␥-secretase in the above experiments. To avoid this, a photoreactive transition state analog inhibitor III-63, a biotinylated probe that covalently binds to the active site of ␥-secretase (35), was immobilized to ␥-secretase prior to co-IP. When ␥-secretase bound to III-63 was immunoprecipitated with anti-Myc antibody, NCT was co-immunoprecipitated as well (Fig. 4C, bottom panel); however, HA-tagged PS1 was not co-immunoprecipitated (second panel). PS was photolabeled (third panel, lane 3), and this labeling was blocked by transition state analog inhibitor 31C, the parent compound of III-63 (lane 2), suggesting that ␥-secretase is active and specifically binds to the inhibitor. Similar results were obtained upon co-IP with anti-HA antibodies (data not shown). These results further confirm that the isolated, active ␥-secretase contains only one PS molecule/complex.
Although the isolated active ␥-secretase complex contains only one PS molecule, it nevertheless appears relatively large by native gel analysis (supplemental Fig. S2A), raising the question about the stoichiometry of the other components, NCT, Pen-2, and Aph-1. To address this issue, we set out to quantify complex members by Western blot analysis. Endogenous NCT, Pen-2, and Aph-1 in DKO cells transfected with wt PS1 could be detected with antibodies for each of these components (Fig.  5A, lane 6). Because one of the wt PS1 molecules transfected in DKO cells is N-terminally Myc-tagged, we first quantified wt PS1 by generating a standard curve with known amounts of commercially available Myc-tagged ubiquitin (Fig. 5B, lanes  4 -8). PS1 NTF and holoprotein that had been immunoprecipitated with active ␥-secretase using anti-Myc antibody were quantified (Fig. 5, B and E, and Table 1), with 0.24 pmol of PS1 NTF and 0.12 pmol of holo-PS brought down. Closely similar results were obtained with a different anti-Myc antibody (data not shown). The possibility remains that epitope accessibility differs between the Myc-tagged ubiquitin and Myc-tagged PS1. Nevertheless, The N terminus of PS1 is not needed for ␥-secretase assembly and activity, which is the reason that epitope tags are typically fused there. Thus, the N-terminal epitope is expected to be freely available for antibody binding. An additional concern is unequal transfer efficiencies between the small Myc-ubiquitin and the larger PS1 NTF and holoprotein. Care must be taken in the interpretation of these quantification results, although the quantification of other members of the complex (see below) suggest that our results are reasonably accurate.
We next set out to quantify co-immunoprecipitated Pen-2; however, endogenous Pen-2 of DKO cells has no epitope tag. Thus, we first quantified varied amounts of FLAG-tagged Pen-2 (immunoprecipitated with M2 anti-FLAG antibodies) from membrane fractions of S-1 cells, CHO cells that stably express wt PS1, FLAG-Pen-2, NCT-GST, and Aph-1a-HA (38), using FLAG-tagged ubiquitin to generate the standard curve (Fig. 5C,  top panel). These FLAG-Pen-2 standards were then used to quantify endogenous Pen-2 of DKO cells that had co-immunoprecipitated with PS (Fig. 5C, bottom panel). The amount of Pen-2 was virtually identical to that of PS-NTF (Fig. 5E), suggesting that Pen-2 is the last member added to the complex and promotes the endoproteolysis of PS as previously reported (7).
We also quantified endogenous NCT of DKO cells that coimmunoprecipitated with PS using the same method. Various amounts of NCT-GST, co-immunoprecipitated with FLAG-Pen-2 from S-1 cells using M2 antibody, were quantified using recombinant GST-E1 to generate the standard curve (Fig. 5D,  top panel). The NCT-GST standards were then used to quantify the total amount of endogenous NCT (mature and immature) that co-immunoprecipitated with PS (Fig. 5D,  bottom panel). The amount of endogenous NCT, 0.34 pmol, was virtually identical to the sum total of PS-NTF and holoprotein (Fig. 5E) FIGURE 2. wt PS1 is not co-immunoprecipitated with ⌬E9. A, digitoninsolubilized membrane fractions were immunoprecipitated with anti-His antibody and subjected to Western blot. PS1 was detected with AB14, and only wt PS1 was detected with B17. Lysate indicates solubilized membrane fractions, and Ϫ or ϩ indicates with protein G beads alone (no antibody) or with both antibody and protein G beads, respectively. B, wt PS was immunoprecipitated with B17 and subjected to Western blot. ⌬E9 was not co-immunoprecipitated with wt PS1. C, digitonin-and CHAPSO-solubilized membrane fractions were subjected to 3-12% blue native gel analysis followed by Western blot. PS1 and NCT were detected with AB14 and anti-NCT antibodies, respectively.
to Pen-2 and before PS undergoes endoproteolysis (42). Similar results were also obtained using different anti-FLAG and anti-GST antibodies (data not shown). IP of active endogenous ␥-secretase from membrane fractions of HEK293 cells gave very similar results ( Table 1), suggesting that these quantification methods do not vary with cell type. We next set out to quantify endogenous Aph-1a; however, we could not quantify Aph-1a-HA from S-1 cells by Western blot using HA-tagged ubiquitin as a standard because different anti-HA antibodies gave radically different results (data not shown). These discrepancies may be due to the position of the HA epitope tag; HA is fused to the N terminus of commercially available ubiquitin, whereas this epitope is fused to the Aph-1 C terminus. We therefore performed Aph-1 co-IP experiments, similar to those carried out for PS in Figs. 1-3. A series of HA-tagged Aph-1 isoforms (Aph-1aL-HA, Aph-1-aS-HA, and Aph-1b-HA) were each co-  expressed with Aph-1aL-FLAG in the CHO cells. All of these Aph-1 isoforms were incorporated into full ␥-complexes (supplemental Fig. S2B). HA-tagged Aph-1 was immunoprecipitated with anti-HA antibody-conjugated agarose beads (Fig. 6A, top panel, lanes 4,  6, and 8). Although PS1 NTF co-immunoprecipitated with each Aph-1-HA isoform (Fig. 6A, bottom panel, lanes 4, 6, and 8), the very small amount of Aph-1aL-FLAG pulled down by anti-HA beads (Fig. 6A, middle panel) was the same level as background seen with anti-His agarose beads (data not shown). The lack of Aph1-Aph1 interaction was confirmed using an Aph-1aL specific antibody: endogenous Aph-1aS did not co-IP with Aph-1aL from membrane fractions of MEFs that were wild type (32), PS DKO cells or NCT knock-out (5) (Fig. 6B, top right panel,  lanes 3, 6, and 9, respectively). Nevertheless, wt MEF showed co-IP of PS1 NTF (middle right panel, lane 3) and mature NCT (bottom right panel, lane 3) and displayed the high molecular size of full protease complexes on the blue native gel (Fig. 6B,   left panel). These observations were consistent with previous reports (33,43). We likewise confirmed that endogenous Pen-2 or endogenous NCT from membrane fractions of S-1 cells did not co-immunoprecipitate with FLAG-Pen-2 or NCT-GST using M2 antibody or GST antibody, respectively (data not shown). Taken together, these results demonstrate that the stoichiometry of the components of isolated, active ␥-secretase complexes is 1:1:1:1 for PS:Pen-2:NCT:Aph-1.

DISCUSSION
Although members of the ␥secretase complex were identified several years ago, the stoichiometry has remained unknown and, in the case of PS, controversial. Some reports provide evidence for two or more PS members per complex (9 -11, 44); however, co-IP from commonly used CHAPSO-solubilized membrane fractions may lead to experimental artifacts (supplemental Fig. S1), because the components of ␥-secretase are hydrophobic membrane proteins that can bind nonspecifically to agarose or Sepharose beads. Digitonin minimizes this nonspecific binding to negligible levels and conserves the integrity and activity of ␥-secretase. In our hands, active ␥-secretase contains only one molecule of PS per complex, a finding that rules out certain models for the protease active site that involve two pairs of aspartates (e.g. one pair responsible for the cutting at the ␥ site and the other at the ⑀ site) or one pair formed in trans at a PS dimer interface (10,11). Another group has recently come to a similar conclusion (49).
Our findings also do not support the suggestion that one PS can affect another PS, for instance, that one FAD mutant PS molecule can alter the site of proteolysis by wt PS (9). Pull-down of FAD mutant ␥-secretase complexes, containing only ⌬E9 PS1, shift the production of A␤ to longer forms, although the overall activity is substantially reduced. The presence of the FAD mutant PS1 did not cause the wt enzyme to produce longer A␤ peptides (Fig. 3B). How then do FAD mutations in PS1 both increase the ratio of A␤42 to A␤40 (gain of function) and decrease overall protease activity (loss of function)? Recent findings suggest that the protease cleaves from the ⑀ site in the APP transmembrane domain (to produce A␤48 or A␤49 peptides) and then trims every 3-4 residues (reviewed in Ref. 50). Thus, mutations that reduce the proteolytic efficiency of the enzyme would lead to less antibody but with an increase in the proportion of longer antibody peptides.
Although we find that the four essential components of ␥-secretase are represented only once per complex, we cannot fully rule out the possibility that in cells the complexes assemble into larger aggregates (e.g. two 1:1:1:1 complexes associated The results are representative of three independent experiments. C, co-immunoprecipitated endogenous Pen-2 was quantified with ab24743 by using various amounts of FLAG-Pen-2 (S1-M2-IP) and FLAG-tagged ubiquitin as standard curves. The band above the endogenous Pen-2 band in WT1-3 lanes is nonspecific. D, co-immunoprecipitated endogenous NCT was quantified with anti-NCT antibody by using various amounts of NCT-GST (S1-M2-IP) and GST-tagged E1 as standard curves. E, quantification results show the stoichiometry of PS1-NTF, PS1-NTF, holo-PS1, Pen-2, and NCT, respectively (see also Table 1). The error bars represent the standard deviation (n ϭ 3). with one another), because such assemblies might be interrupted by solubilization with detergent. Nevertheless, isolated ␥-secretase remains proteolytically active; any interaction between one ␥-secretase complex and another is apparently very weak and not necessary for recognition and cleavage of substrates. The size of the ␥-secretase complex has likewise been controversial. Our study shows that all four components are represented only once, and adding together the molecular weights of the individual subunits gives ϳ220 kDa. However, the size of the complex as estimated by blue native gel appears to be ϳ500 kDa; thus, we tested whether other proteins (e.g. substrates, modifiers) were isolated with the ␥-complexes. By native gel, APP appeared as a smear from ϳ300 to 800 kDa, but membrane fractions of PS DKO cells showed the same smear of APP in the absence of ␥-complexes (Fig. 7C). Also, Tmp21, recently reported as a regulator of ␥-secretase (45), did not co-migrate with ␥-complexes, appearing instead at ϳ200 kDa (Fig. 7D). Mobilities may differ with detergents (Fig. 2C, lanes 1  and 5, and Fig. 7), and comparisons of a hydrophobic membrane protein complex with soluble protein molecular weight standards may not be appropriate. Moreover, we found that different commercial standard marker kits showed different mobilities for certain marker proteins of comparable size (data not shown). Although we cannot rule out the presence of other possible modifier proteins and substrates that would bring the size up to ϳ500 kDa, we suggest that the interpretation of results from blue native gel analysis of membrane protein complexes can be fraught with difficulties. In contrast, glycerol velocity gradient centrifugation suggested a size of ϳ150 -250 kDa for solubilized endogenous ␥-secretase (i.e. not overexpressed) (14,46), which includes the size predicted by adding up the molecular weights of the four individual components. In our hands, glycerol velocity gradient analysis indeed showed a smaller size (ϳ200 -250 kDa) than observed by blue native gels (ϳ500 kDa) (supplemental Fig. 2C). Markers may also run differently in this method as well, and interpretations must be made with care, but our results combined with those in previous reports demonstrate the limitations of these methods of size determination for the ␥-secretase complex. Most recently, the size of the purified enzyme complex has been determined to be ϳ220 kDa by scanning transmission electron microscopy, or STEM. 3 STEM does not involve protein mobility and provides further support for our findings on the stoichiometry.
The present study shows the stoichiometry of ␥-components, with the amount of Pen-2 equal that of PS-NTF and the amount of NCT equal to that of the sum total of PS-NTF and holo-PS (Fig. 5). This quantification result is consistent with the current model for maturation of the ␥-secretase complex, in which an initially formed NCT⅐Aph-1 complex binds to holo-PS, followed by Pen-2 binding to this NCT⅐Aph-1⅐holo-PS complex to trigger PS endoproteolysis into NTF and CTF (7,47). In certain cells (wild type or transfected), Pen-2 is the limiting factor, with excess premature NCT⅐Aph-1⅐holo-PS complexes building up as a result. In conclusion, our results dem-3 H. Li   onstrate that having one of each of the four subunits is sufficient for ␥-secretase complexes to bind and cleave substrates. These observations exclude certain models for ␥-secretase assembly and active site composition and should aid interpretation of results from structural studies of the protease complex.