Chemical Cross-linking Provides a Model of the γ-Secretase Complex Subunit Architecture and Evidence for Close Proximity of the C-terminal Fragment of Presenilin with APH-1*

γ-Secretase is an intramembrane cleaving aspartyl protease complex intimately implicated in Alzheimer disease pathogenesis. The protease is composed of the catalytic subunit presenilin (PS1 or PS2), the substrate receptor nicastrin (NCT), and two additional subunits, APH-1 (APH-1a, as long and short splice forms (APH-1aL, APH-1aS), or APH-1b) and PEN-2. Apart from the Alzheimer disease-associated β-amyloid precursor protein, γ-secretase has been shown to cleave a large number of other type I membrane proteins. Despite the progress in elucidating γ-secretase function, basic questions concerning the precise organization of its subunits, their molecular interactions, and their exact stoichiometry in the complex are largely unresolved. Here we isolated endogenous human γ-secretase from human embryonic kidney 293 cells and investigated the subunit architecture of the γ-secretase complex formed by PS1, NCT, APH-1aL, and PEN-2 by chemical cross-linking. Using this approach, we provide evidence for the close neighborhood of the PS1 N- and C-terminal fragments (NTF and CTF, respectively), the PS1 NTF and PEN-2, the PS1 CTF and APH-1aL, and NCT and APH-1aL. We thus identify a previously unrecognized PS1 CTF/APH-1aL interaction, verify subunit interactions deduced previously from indirect approaches, and provide a model of the γ-secretase complex subunit architecture. Finally, we further show that, like the PS1 CTF, the PS2 CTF also interacts with APH-1aL, and we provide evidence that these interactions also occur with the other APH-1 variants, suggesting similar subunit architectures of all γ-secretase complexes.

␥-Secretase is an intramembrane cleaving aspartyl protease complex that cleaves numerous type I membrane proteins after they have undergone shedding of the bulk of their ectodomains (1,2). ␥-Secretase cleavage of the membrane protein stubs, which remain after ectodomain shedding, results in the liberation of small peptides into the extracellular space as well as in the release of intracellular domains into the cytosol. Although ␥-secretase cleavage may serve to turn over type I membrane protein stubs (3), it also can mediate signal transduction via the intracellular domain that is released by ␥-secretase. This novel mode of signal transduction has been firmly established for the intracellular domain of the Notch1 cell surface receptor, which, following its release by ␥-secretase cleavage, translocates to the nucleus to act as a transcriptional regulator of target genes essential for cell differentiation during development and in adulthood (4).
Most interest in the study of ␥-secretase, however, stems from its relevance as a molecular drug target for the treatment and prevention of Alzheimer disease. Here, ␥-secretase cleaves the ␤-amyloid precursor protein (APP), 3 following initial ectodomain shedding by ␤-secretase, to liberate the amyloid ␤-peptide (A␤) (5). A␤ is heterogeneous in its length, and the long A␤ variant A␤42, although much less produced than A␤40, is highly aggregation-prone, neurotoxic, and believed to initiate the disease-causing amyloid cascade (6). Mutations in presenilin (PS), which represents the catalytic subunit of ␥-secretase (7)(8)(9)(10)(11) are associated with familial forms of Alzheimer disease. These mutations, as well as most of the less frequent mutations in APP, increase the A␤42/A␤40 ratio (6).
Despite the considerable progress in the elucidation of the cellular function of ␥-secretase and its mode of action, basic questions regarding the subunit organization, the presence of additional components in ␥-secretase complexes, and their molecular interactions and stoichiometry are largely unresolved. To begin to address the above questions, we isolated endogenous human ␥-secretase by a multistep purification procedure and analyzed its subunit architecture by chemical crosslinking. This approach has been used successfully to demonstrate a direct interaction of the PS1 NTF and CTF in intact cells (23). In addition, evidence for homodimerization of the PS1 NTF was obtained with a related approach using a photocross-linkable ␥-secretase inhibitor (24), and in agreement with this study, close proximity of two PS1 molecules was also observed in intact cells by a microscopy assay (25). Apart from these studies, ␥-secretase subunit interactions were mostly deduced from experiments performed under conditions where either one subunit was lacking (13, 18 -20), where mutant subunits defective in assembly were expressed (18, 26 -31), or where ␥-secretase was dissociated (32). None of these latter studies, however, demonstrated direct contacts between the subunits, and in many cases they could be indirect, being mediated by a third partner (18). By identifying specific intersubunit cross-links, including a previously unrecognized PS CTF/ APH-1 interaction, we now provide a model of the molecular architecture of ␥-secretase based on direct subunit interactions. Furthermore, our results provide implications for ␥-secretase complex subunit stoichiometry.

MATERIALS AND METHODS
Cell Lines and Antibodies-Human embryonic kidney (HEK) 293 cells and HEK 293 cells stably co-expressing APPsw and PS1 WT or PS1 M292D were cultured as described before (33). Monoclonal and polyclonal antibodies, respectively, against the PS1 N terminus (PS1N) and against the PS1 large loop (APS18 and 3027) as well as against the PS2 N terminus (BI.5D3) and the PS2 large loop (HF5C) were described previously (34 -36). Affinity-purified polyclonal antibodies against the N terminus of PEN-2 (antibody 1638), against the loop three domain of APH-1a (antibody 2021) and against the C termini of APH-1aL (antibody 433) and APH-1b (antibody 435) have been described before (16,37,38). Polyclonal antibody N1660 against the C terminus of NCT was obtained from Sigma. Monoclonal antibody 6E10 to A␤ was obtained from Signet Laboratories.
Cross-linking-The ␥-secretase preparation was cross-linked with dithiobis(succinimidylpropionate) (DSP) or 1,4-bis-maleimidyl-2,3-dihydroxybutane (BMDB) (both obtained from Pierce) for 15-30 min at 30°C according to the instructions of the supplier. Cross-linked ␥-secretase complex subunits were analyzed by immunoblotting with the indicated antibodies. Alternatively, ␥-secretase was dissociated after cross-linking in 1% SDS, and following 10-fold dilution, it was subjected to immunoprecipitation prior to immunoblot analysis. Nonreducing conditions were used during cross-linking and subsequent SDS-PAGE to avoid dissociation of cross-link products generated by DSP or to avoid interference of immunoprecipitated cross-links with antibody chains. To detect PEN-2 cross-links, the Western blots were heated for 5 min in PBS containing 2% ␤-mercaptoethanol prior to immunostaining. This procedure enhanced the accessibility of epitopes of antibody 1638 in cross-link adducts of PEN-2. Where indicated, cross-links containing NCT were deglycosylated with N-glycosidase F according to the instructions of the supplier (Roche Applied Science).
Cross-link Analysis by Two-dimensional SDS-PAGE-To identify the respective subunits in a cross-link band, lanes from the first dimension nonreducing SDS-PAGE were cut out and incubated in sample buffer containing 10% ␤-mercaptoethanol for 20 min at room temperature with shaking to cleave the disulfide bond of the cross-linker. The lanes containing cleaved cross-link bands were then subjected to a second dimension SDS-PAGE, and the released subunits were analyzed by immunoblotting using the indicated antibodies.

RESULTS
To identify subunits within ␥-secretase, which are in close contact with each other, a cross-linking approach was chosen. To minimize potential background of nonspecific cross-linking reactions, endogenous ␥-secretase of HEK 293 cells was enriched in functional form by several purification steps outlined in Fig. 1A. The intactness of ␥-secretase was confirmed by co-purification of the complex subunits NCT, PS1, APH-1aL, and PEN-2 in the affinity purification steps (Fig. 1B) and by demonstrating that the purified ␥-secretase was active, as evident from the specific generation of A␤ using recombinant C100-His 6 substrate (Fig. 1C). Having confirmed that the purification procedure yielded functional ␥-secretase, we subjected the protease preparation to chemical cross-linking, focusing on subunit interactions between PS1, NCT, APH-1aL, and PEN-2. To establish cross-link conditions, we probed the molecular environment of the PS1 NTF using the amine-reactive, homobifunctional, and thiol-cleavable cross-linker DSP, which has a spacer arm length of 12 Å. Following cross-linking, samples were subjected to SDS-PAGE under nonreducing conditions to keep the cross-link products intact. DSP efficiently cross-linked the PS1 NTF in a dose-dependent manner, and several discrete cross-link products were obtained ( Fig. 2A). At a 0.1 mM concentration of cross-linker, robust cross-link products of ϳ40, ϳ50, and ϳ60 kDa were observed. When the cross-linker was used at 10-fold higher concentration, additional higher molecular weight cross-link products migrating at ϳ150 kDa were detected. Because NCT is the only known ␥-secretase subunit with a molecular mass above 100 kDa, we investigated whether NCT was also cross-linked. As shown in Fig. 2B, this was indeed the case, suggesting that the higher molecular weight cross-link products could possibly represent cross-links with NCT. The signal intensity of the NCT cross-links decreased with increasing molecular weight of the cross-links, probably due to impaired antibody accessibility to cross-links of possibly higher order (i.e. cross-links containing more than two partners). Because the most complete spectrum of cross-link products was obtained with 1 mM DSP, this cross-linker concentration was used in the subsequent experiments, in which we first analyzed the crosslink patterns of PS1 NTF, PS1 CTF, PEN-2, APH-1aL, and NCT (Fig.  2C). A cross-link pattern closely resembling that of PS1 NTF was observed for the PS1 CTF yielding cross-link products with very similar if not identical molecular weight in the lower molecular weight range in addition to higher molecular weight cross-link products. The cross-link product pattern of PEN-2 contained cross-link products of ϳ40 and ϳ60 kDa and higher order cross-links migrating at ϳ150 kDa. For APH-1aL, a cross-link product of ϳ40 kDa was observed in addition to several cross-link products in the higher molecular weight range, which had the strongest band intensity. This indicated that APH-1aL was highly efficiently cross-linked to NCT (the only ␥-secretase complex subunit of higher molecular weight) and contained in higher order cross-links thereof. Consistent with this view, several high molecular weight cross-link products were observed for NCT, as shown before in Fig. 1B. Thus, the most robust cross-links occurred in two principal molecular weight ranges, between ϳ40 and ϳ60 kDa, suggesting that these cross-links probably represented interactions among the small subunits PS1 NTF (ϳ30 kDa), PS1 CTF (ϳ20 kDa), APH-1aL (ϳ20 kDa), and PEN-2 (ϳ10 kDa) and above ϳ100 kDa, thus possibly containing NCT as interaction partner.
Because the individual subunits differ in their molecular weight, cross-links between them could be identified with reasonable certainty by co-migration analysis. Side by side comparison of the co-migration behavior of the cross-link products of the individual subunits shown in Fig. 2C suggested for the lower molecular weight range of cross-links the following subunit interactions as the most likely ones (Fig. 2, D-F). The occurrence of the only cross-link products that co-migrated at ϳ50 kDa for the PS1 NTF and CTF ( The presence of co-migrating ϳ40-kDa cross-link products for the PS1 NTF and PEN-2 indicated an interaction of these subunits (Fig. 2E), again in good agreement with the sum of the apparent molecular weights of the monomers. The co-migrating ϳ60-kDa cross-link products of the PS1 NTF and PEN-2 ( Fig. 2E) most likely represented the ternary interaction of the PS1 NTF⅐CTF heterodimer with PEN-2, because the ϳ60-kDa PS1 NTF cross-link also co-migrated with the ϳ60-kDa cross-link product of the PS1 CTF (Fig. 2D). Finally, the presence of co-migrating ϳ40-kDa cross-  link products for PS1 CTF and APH-1aL suggested a direct interaction of these subunits (Fig. 2F).
To further substantiate the identity and specificity of the cross-link adducts observed, we next asked whether these could be cleaved by reducing agent. To this end, we subjected the cross-links to two-dimensional SDS-PAGE analysis. Following separation of the cross-link products by SDS-PAGE under nonreducing conditions in a first dimension and cleavage of the disulfide bridge of the cross-linker present in the cross-link adducts, samples were subjected to SDS-PAGE in a second dimension under reducing conditions to separate the cleaved cross-linked products. In this analysis, the cleaved cross-link products are expected to migrate on the same vertical line below one another in the second dimension. As shown in Fig.  3A, cleavage of the cross-link adducts of PS1 NTF and PEN-2 that migrated at ϳ40 kDa in the first dimension yielded the PS1 NTF and PEN-2 migrating on the same vertical line in the sec-ond dimension. Likewise, the ϳ50-kDa cross-link adducts observed for PS1 NTF and CTF in the first dimension were found to migrate on the same vertical line in the second dimension after cross-linker cleavage. In addition, migration on the same vertical line in the second dimension after cross-linker cleavage was observed for PS1 NTF, PS1 CTF, and PEN-2 at the position of the respective ϳ60-kDa cross-link adduct observed for these subunits in the first dimension. As shown further in Fig. 3B, the ϳ40-kDa cross-link products of the PS1 CTF and APH-1aL co-migrating in the first dimension were also cleaved under reducing conditions and migrated on the same vertical line in the second dimension after cross-linker cleavage, thus providing further evidence that these two subunits were specifically cross-linked to each other in the ϳ40 kDa band. Taken together, these data are consistent with the data presented in Fig. 2 and further verify the close subunit interactions of the PS1 NTF with PEN-2, of the PS1 NTF with the PS1 CTF, of the PS1  PS1N. B, samples shown in A were analyzed for the generation of cross-link products of NCT by immunoblotting using antibody N1660. C, ␥-secretase was subjected to cross-linking with 1 mM DSP. Cross-link products of the PS1 NTF and CTF, PEN-2, APH-1aL, and NCT were detected by immunoblotting with antibodies to the PS1 NTF (PS1N), PS1 CTF (3027), PEN-2 (1638), APH-1aL (433), and NCT (N1660). In A-C, the arrows mark specific cross-links. The asterisk denotes a PS1 CTF aggregate that is occasionally observed. D-F, individual lanes from the immunoblots of the samples shown in C that received cross-linker were arranged side by side and analyzed for co-migration of cross-link products. Co-migration analysis of cross-link products immunopositive for PS1 NTF and PS1 CTF identifies a PS1 NTF/PS1 CTF interaction (D). Co-migration analysis of cross-link products immunopositive for PS1 NTF and PEN-2 identifies PEN-2/PS1 NTF as well as, when additionally compared with A, PEN-2/PS1 NTF/PS1 CTF interactions (E), and co-migration analysis of cross-link products immunopositive for PS1 CTF and APH-1aL identifies a PS1 CTF/APH-1aL interaction (F). Molecular mass markers are shown on the left in kDa.
NTF⅐CTF heterodimer with PEN-2, and of the PS1 CTF with APH-1aL.
Because a previous study provided evidence for PS1 NTF homodimer formation using a related approach that employed photocross-linking with a transition state analog ␥-secretase inhibitor derivate (24), the ϳ60-kDa cross-link product immunopositive for the PS1 N-terminal antibody (Fig. 2, D and E) could additionally contain a PS1 NTF⅐NTF adduct apart from the proposed PEN-2/PS1 NTF/CTF interaction identified above. Likewise, the ϳ40-kDa cross-link product immunopositive for the PS1 C-terminal antibody could potentially include a corresponding PS1 CTF⅐CTF homodimer. To investigate these possibilities, we took advantage of the PS1 M292D variant, which is not cleaved into PS1 NTF and CTF due to a mutation at the site of PS endoproteolysis (33). If DSP cross-links putative PS1 NTF or CTF homodimers, then ϳ100-kDa crosslink products would be expected for the uncleavable PS1 M292D mutant. To prove whether that was the case, ␥-secretase was purified from HEK 293 cells stably expressing PS1 M292D and subjected to cross-linking with DSP. As shown in Fig. 4, no cross-link products with a molecular mass of ϳ100 kDa that could correspond to PS1 homodimers were observed. Only cross-link products in the ϳ150 kDa high molecular weight range and a ϳ60-kDa cross-link product consistent with the molecular weight of the cross-link adduct of PS1 M292D and PEN-2 were observed. We therefore conclude that it is very unlikely that the observed ϳ60and ϳ40-kDa cross-link adducts of PS1 NTF and CTF identified above represented PS1 NTF⅐NTF and PS1 CTF⅐CTF homodimers, respectively.
We next sought to further verify the novel PS1 CTF/APH-1aL interaction by additional and independent experimental  approaches. Further evidence for a close physical association of the PS1 CTF with APH-1aL was obtained when BMDB was used as cross-linker. This cross-linker has a similar spacer arm length (10 Å) as DSP but is thiol-reactive. As shown in Fig. 5A, cross-linking with BMDB also yielded a co-migrating cross-link product of ϳ40 kDa for both subunits. In contrast to crosslinking with DSP, this cross-link product represented the major cross-link product for the PS1 CTF for this cross-linker. To further confirm the identity of the PS1 CTF/APH-1aL interaction, we dissociated the ␥-secretase subunits after cross-linking with BMDB by treatment with SDS. APH-1aL was then immunoprecipitated, and the presence of the PS1 CTF in the ϳ40-kDa cross-link product was examined by immunoblotting. As shown in Fig. 5B, the immunoprecipitated ϳ40-kDa APH-1aL cross-link product was immunopositive for PS1 CTF, demonstrating that the PS1 CTF and APH-1aL were cross-linked with each other. Finally, we investigated whether these cross-links to APH-1aL can also be observed with the uncleavable PS1 M292D variant (Fig. 5C). ␥-Secretase was therefore purified from HEK 293 cells stably expressing either PS1 WT or the PS1 M292D mutant and subjected to cross-linking with BMDB. As expected from the experiments above, a ϳ40-kDa PS1 CTF crosslink product (indicated by a triangle) was observed for the WT PS1 ␥-secretase complex, which co-migrated with the ϳ40-kDa cross-link product of APH-1aL. Consistent with a cross-link adduct of PS1 with APH-1aL, a ϳ70-kDa cross-link product of the uncleaved PS1 holoprotein (indicated by a square) was observed for the PS1 M292D mutant ␥-secretase complex. Furthermore, the ϳ40-kDa APH-1aL cross-link product observed for the PS1 WT complex also shifted to a ϳ70-kDa APH-1aL cross-link product in the PS1 M292D mutant ␥-secretase complex. This ϳ70-kDa APH-1aL cross-link product co-migrated with the ϳ70-kDa cross-link product positive for PS1. In addition, no PS1-positive ϳ100-kDa band was observed for PS1 M292D, arguing again against the possibility that the ϳ40-kDa cross-link product observed for WT PS1 represented a dimeric PS1 CTF. Taken together, these data unambiguously demonstrate that the PS1 CTF is in close contact with APH-1aL in the ␥-secretase complex.
We next set out to characterize and identify the subunit interactions in the higher molecular weight range, including those represented by NCT. Because NCT is the only glycoprotein of the four ␥-secretase subunits, we reasoned that cross-links containing NCT should shift to lower molecular weight upon deglycosylation, whereas cross-links, which were composed solely by the other much smaller subunits, would not. In addition, deglycosylation should facilitate analysis of cross-links containing NCT, because these cross-link products could be better separated at lower molecular weight. Following cross-linking with DSP, the samples were therefore treated with N-glycosidase F to deglycosylate NCT prior to SDS-PAGE. Strikingly, upon deglycosylation, not only the cross-link products of NCT but also those of the other subunits shifted to lower molecular weight, strongly suggesting that the cross-link products in the higher molecular weight range represented cross-links with NCT (Fig. 6). Two clearly separated major bands were observed for NCT, APH-1aL and the PS1 CTF, whereas one major band . Identification of a PS1 CTF/APH-1aL subunit interaction. A, ␥-secretase was subjected to crosslinking with 2 mM BMDB. Samples were analyzed for PS1 CTF and APH-1aL cross-links by immunoblotting using antibodies 3027 (PS1 CTF) and 433 (APH-1aL). The asterisk denotes a PS1 CTF aggregate that is occasionally observed. B, ␥-secretase was subjected to cross-linking with BMDB as in A. Following cross-linking, ␥-secretase subunits were dissociated with SDS followed by immunoprecipitation (IP) with antibody 433. The presence of the PS1 CTF in the APH-1aL cross-link product was analyzed by immunoblotting with antibody 3027. C, ␥-secretase was isolated from PS1 WT-or PS1 M292D-expressing cells. Aliquots of the experiment shown in Fig. 3 were subjected to cross-linking with BMDB as in A, and cross-links were analyzed by immunoblotting as in A. The PS1 CTF⅐APH-1aL cross-link in PS1 WT-expressing cells is indicated by a triangle, and the PS1 FL⅐APH-1aL cross-link in PS1 M292D expressing cells is shown with a square. An arrow marks a specific cross-link of PS1 M292D, most likely with PEN-2. Molecular mass markers are shown on the left in kDa.
was observed for the PS1 NTF and PEN-2. As shown by careful co-migration analysis, the cross-link product with the lowest molecular weight (indicated by a square) was consistent with an NCT⅐APH-1aL adduct. This cross-link product was followed in molecular weight by a cross-link between NCT⅐APH-1aL⅐PS1 CTF (indicated by a triangle). Based on its molecular weight higher than that of the NCT⅐APH-1aL⅐PS1 CTF cross-link, the cross-link product(s) with the highest molecular weight (indicated by a circle) represented an adduct that additionally contained the PS1 NTF (i.e. the NCT⅐APH-1aL⅐PS1 CTF⅐PS1 NTF cross-link) and/or the PS1 NTF cross-linked to PEN-2 (i.e. the NCT⅐APH-1aL⅐PS1 CTF⅐PS1 NTF⅐PEN-2 cross-link between all four subunits). Taken together, these data show that the observed cross-links occurring between the small subunits (i.e. PS1 NTF⅐PEN-2, PS1 CTF⅐APH-1aL, and PS1 NTF⅐PS1 CTF) are recovered in high molecular weight cross-links that all additionally contain NCT. Furthermore, the characteristic molecular weight pattern of these cross-links strongly suggests that PS1 NTF⅐PEN-2 are linked to NCT in these adducts via APH-1aL⅐PS1 CTF.
Finally, we investigated the subunit interactions of the respective ␥-secretase complexes containing the other APH-1 variants or PS2. As shown in Fig. 7, A and B, APH-1aL could also be cross-linked with BMDB to the PS2 CTF. Moreover, all three APH-1 variants, APH-1aL, APH-1aS, and APH-1b, showed the same characteristic cross-link pattern with the ϳ40 kDa cross-link band, suggesting that all APH-1 variants interact with the PS1 and PS2 CTFs (Fig. 7C). Although cross-linking of APH-1aS (which is more highly expressed than APH-1aL in HEK 293 cells (16) (Fig. 7C) could not directly be analyzed due to the unavailability of a monospecific anti-APH-1aS antibody, the relative abundance of the ϳ40 kDa cross-link band compared with that of the APH-1aL monomer indicated also crosslinking of APH-1aS (Fig. 7C). Last, we found that the cross-link pattern of PS2 NTF closely resembled that of the PS1 NTF ( Fig.  2A), again indicating that PS2 has the same subunit interactions as PS1 (Fig. 7D). Taken together, these data suggest that all ␥-secretase complexes have a very similar subunit organization.

DISCUSSION
Using a chemical cross-linking approach, we obtained insight into the subunit arrangement of ␥-secretase. Probing the molecular environment of the catalytic subunit PS1 with DSP suggested close proximity of the PS1 NTF with PEN-2, of the PS1 CTF with APH-1aL, and of the PS1 NTF with the PS1 CTF (i.e. the catalytic subunit of ␥-secretase itself). The last interaction had also been observed earlier when DSP was used for cross-linking in intact cells (23). Cross-link products consistent with interactions of the PS1 NTF with APH-1aL and of the PS1 CTF with PEN-2 and PEN-2 with APH-1aL were not observed. Besides these two-partner interactions, a corresponding three-partner cross-link of PEN-2⅐PS1 NTF⅐PS1 CTF was also observed. The basic two-subunit interactions were recovered in the higher molecular mass (Ͼ100 kDa) cross-links. These cross-links all contained NCT, as could be demonstrated by deglycosylation of the cross-link products that for all four subunits shifted to lower molecular weight. No direct crosslinks of NCT with PEN-2, the PS1 NTF, and surprisingly also the PS1 CTF (28) were observed. The failure to detect the latter interaction may be due to an improper cross-link spacer length of DSP or due to the lack of reactive groups of the interaction partner(s) at the correct distance. It should be noted here, however, that this is a general drawback of a cross-linking approach. Some subunit interactions may thus escape detection.
The proposed subunit interactions determined by crosslinking are in line with some of the interactions observed in different experimental settings. For example, PEN-2⅐PS1 NTF, NCT⅐APH-1aL, and NCT⅐APH-1aL⅐PS1 CTF complexes could be preserved when ␥-secretase was dissociated by the use of excess detergent (32). Likewise, the PS1 NTF⅐PEN-2 cross-link observed in our study is consistent with two reports, which showed that TMD4 of PS1, which is part of the PS1 NTF, is required for interaction with and stabilization of PEN-2 (39,40). An interaction of the PS1 CTF with APH-1aL and NCT was suggested by co-immunoprecipitation (27) could not be addressed by this approach. Our study now identifies a close contact of the PS1 CTF with APH-1aL, an interaction that has not been recognized before. This novel subunit interaction was identified with two different cross-linkers, by using co-immunoprecipitation and by using the uncleavable PS1 M292D mutant as an experimental tool. We further demonstrated that APH-1aL could also be cross-linked to the PS2 CTF and provided evidence that all APH-1 variants interact with the CTF of PS1 and PS2 in the respective ␥-secretase complexes (16,17). The close interaction of these subunits may be initiated during assembly of the ␥-secretase complex, which requires a GXXXG TMD interaction motif present in TMD4 of APH-1 (29 -31). Interestingly, this motif is also found in TMD7 of PS, although it is not fully conserved across species. Whether the close neighborhood of the CTF of PS with APH-1 may play a role for the catalytic reaction of ␥-secretase remains to be further investigated. Previously, evidence was provided that PS exists as a dimer in the ␥-secretase complex (24,25,41,42). Although our data do not exclude two PS molecules per complex, we did not obtain evidence for a PS dimer in the purified endogenous complex by cross-linking, since neither NTF⅐NTF nor CTF⅐CTF cross-links of PS1 were observed. Consistent with this observation, the higher molecular weight cross-links containing PS1 NTF and PS1 CTF contained NCT and did not represent higher order oligomeric forms of PS. Likewise, no evidence for the expected 100-kDa PS dimer was obtained when the uncleavable PS1 M292D mutant was subjected to cross-linking. Furthermore, no evidence was found for dimerization of the other subunits. The cross-linking data presented in this study may therefore favor a 1:1:1:1 (PS/NCT/APH-1/ PEN-2) stoichiometry rather than a 2:1:1:1 (PS/NCT/APH-1/PEN-2) stoichiometry for the ␥-secretase complex. This interpretation would be in agreement with the recent quantitation of the ␥-secretase complex subunits using protein standards that revealed a 1:1: 1:1 ratio of PS/NCT/APH-1/ PEN-2 (43). The two major ␥-secretase cuts at the topologically distinct ␥and ⑀-site may therefore be executed by one rather than two catalytic PS subunits. Substrate cleavage may thus proceed from the ⑀to the ␥-site, involving in passing an additional cleavage at the -site (44,45), and not involve the alternatively possible simultaneous cleavage by two catalytic subunits. We want to point out, however, that at this stage it cannot be excluded that ␥-secretase complexes, whose four subunits in sum have a molecular mass of 200 -250 kDa (taking glycosylation of NCT into account), may dimerize, giving rise to an overall 2:2:2:2 stoichiometry of the four subunits. This, however, would be consistent with the apparent molecular weight of ␥-secretase in blue native gel electrophoresis (35,43,46,47) and could explain the previous observations of PS dimerization (24,25,41,42). Interestingly, such a dimeric structural organization of an integral membrane protein complex has recently been shown for the yeast oligosaccharyltransferase by electron microscopy (48).
In conclusion, our data give structural insight into the subunit organization of the ␥-secretase complex based on intersubunit interactions determined by chemical cross-linking. A schematic model of the subunit interactions between PS NTF, PS CTF, NCT, APH-1, and PEN-2 suggested by this study is shown in Fig. 8. In future research, intrasubunit interactions should be investigated using suitable short spacer arm and/or zero-length cross-linkers. Some of the first successful steps into this direction have been done by the recent demonstration of close proximity of the PS1 TMD1 and -8 (49) and TMD6 and -7  Fig. 5A were additionally analyzed for a PS2 CTF/APH-1aL interaction by immunoblotting using antibody HF5C (PS2 CTF). B, samples of the experiment shown in Fig. 5B were additionally analyzed for the presence of the PS2 CTF in the immunoprecipitated (IP) APH-1aL cross-link product by immunoblotting with antibody HF5C (PS2 CTF). C, ␥-secretase was subjected to cross-linking with 2 mM BMDB. Samples were analyzed for APH-1 cross-links by immunoblotting using antibodies 433 (APH-1aL), 2021 (APH-1aL and APH-1aS), and 435 (APH-1b). D, samples of DSP-cross-linked ␥-secretase of the experiment shown in Fig. 2C were additionally analyzed for PS2 NTF cross-links (arrows) by immunoblotting using antibody BI.5D3 (PS2 NTF). Molecular mass markers are shown on the left in kDa.
using disulfide cross-linking strategies (50,51). In addition, electron microscopy studies of ␥-secretase at higher resolution than obtained so far (52) will certainly provide further structural insight, and ultimately, the crystal structure of ␥-secretase, once available, will reveal the molecular and atomic details of this pivotal intramembrane cleaving protease complex. Schematic view of direct subunit interactions within the ␥-secretase complex as revealed by cross-link analysis with DSP and BMDB (i.e. in a 10 -12-Å distance). The NTF and CTF of the catalytic subunit PS are in direct contact. Beyond this interaction, the PS NTF has direct contact with PEN-2, whereas the PS CTF has direct contact with APH-1. The latter subunit also has direct contact with NCT. Note that additional subunit contacts other than those identified in this study are not excluded and may be observable with cross-linkers of different chemistry and/or spacer arm length. The small numbers denote the approximate apparent molecular weights of the subunits.