Contribution of Presenilin Transmembrane Domains 6 and 7 to a Water-containing Cavity in the γ-Secretase Complex*

γ-Secretase is a multiprotein complex responsible for the intramembranous cleavage of the amyloid precursor protein and other type I transmembrane proteins. Mutations in Presenilin, the catalytic core of this complex, cause Alzheimer disease. Little is known about the structure of the protein and even less about the catalytic mechanism, which involves proteolytic cleavage in the hydrophobic environment of the cell membrane. It is basically unclear how water, needed to perform hydrolysis, is provided to this reaction. Presenilin transmembrane domains 6 and 7 seem critical in this regard, as each bears a critical aspartate contributing to catalytic activity. Current models imply that both aspartyl groups should closely oppose each other and have access to water. This is, however, still to be experimentally verified. Here, we have performed cysteine-scanning mutagenesis of both domains and have demonstrated that several of the introduced residues are exposed to water, providing experimental evidence for the existence of a water-filled cavity in the catalytic core of Presenilin. In addition, we have demonstrated that the two aspartates reside within this cavity and are opposed to each other in the native complex. We have also identified the conserved tyrosine 389 as a critical partner in the catalytic mechanism. Several additional amino acid substitutions affect differentially the processing of γ-secretase substrates, implying that they contribute to enzyme specificity. Our data suggest the possibility that more selective γ-secretase inhibitors could be designed.

␥-Secretase is a multiprotein complex responsible for the intramembranous cleavage of the amyloid precursor protein and other type I transmembrane proteins. Mutations in Presenilin, the catalytic core of this complex, cause Alzheimer disease. Little is known about the structure of the protein and even less about the catalytic mechanism, which involves proteolytic cleavage in the hydrophobic environment of the cell membrane. It is basically unclear how water, needed to perform hydrolysis, is provided to this reaction. Presenilin transmembrane domains 6 and 7 seem critical in this regard, as each bears a critical aspartate contributing to catalytic activity. Current models imply that both aspartyl groups should closely oppose each other and have access to water. This is, however, still to be experimentally verified. Here, we have performed cysteine-scanning mutagenesis of both domains and have demonstrated that several of the introduced residues are exposed to water, providing experimental evidence for the existence of a water-filled cavity in the catalytic core of Presenilin. In addition, we have demonstrated that the two aspartates reside within this cavity and are opposed to each other in the native complex. We have also identified the conserved tyrosine 389 as a critical partner in the catalytic mechanism. Several additional amino acid substitutions affect differentially the processing of ␥-secretase substrates, implying that they contribute to enzyme specificity. Our data suggest the possibility that more selective ␥-secretase inhibitors could be designed.
The Presenilin (PS) 2 proteins are the prototypic members of a group of aspartic proteases involved in regulated intramembrane proteolysis, a mechanism responsible for cleavage of peptide bonds within the lipid bilayer (1,2). More than 150 muta-tions in Presenilin 1 and 10 mutations in Presenilin 2 have been associated with Alzheimer disease (for a list of the mutations, see molgen.ua.ac.be/ADMutations), demonstrating their pivotal role in the pathogenesis of the disease. Presenilins are critical for the ␥-secretase cleavage of the amyloid precursor protein (APP) that generates the amyloid ␤ peptide (A␤) (3). They are also responsible for the intramembrane proteolysis of several other type I transmembrane proteins (reviewed in Ref. 4), including the S3 cleavage of Notch that releases the Notch intracellular domain (NICD), a major regulator of gene transcription (5). Together with Presenilin, three other membrane proteins are necessary and sufficient for processing by ␥-secretase (6 -8), i.e. Nicastrin, APH-1, and PEN-2. Nicastrin appears to recognize the free N terminus of potential substrates (9), whereas the catalytic core resides in Presenilin (reviewed in Ref. 10).
Most recent topological studies propose a nine-transmembrane domain model for Presenilins, with the N terminus oriented toward the cytosol and the C terminus toward the extracellular space (11)(12)(13). Mutation of two conserved aspartates, Asp-257 and Asp-385, located in transmembrane domains (TMs) 6 and 7 respectively, abolishes activity as well as binding to the transition state inhibitors of ␥-secretase (14 -16), supporting the hypothesis that these residues constitute the catalytic site of the protein. Furthermore, inhibitor profiling studies have provided evidence for at least one additional substrate binding site on Presenilin, distinct from, but in close proximity to, the catalytic site (17)(18)(19).
Structure-function studies of the ␥-secretase complex are not easy to perform because of its hydrophobic nature and its sensitivity to membrane lipid composition and detergent extraction procedures. This explains why most of our knowledge of the catalytic activity of the complex is based on indirect evidence and assumptions. For instance, hydrolysis of peptide bonds requires that the active catalytic site of the protease have access to water within the lipid bilayer, but no formal proof for this assumption has been provided in the case of Presenilin. To probe experimentally the microenvironment of the catalytic site of Presenilin, we employed cysteine-scanning mutagenesis, a method widely used to investigate the structural features of polytopic membrane proteins (reviewed in Refs. 20 and 21). The principle involves substitution of amino acid residues of interest with cysteine, which is average in size and thus normally quite well tolerated and amenable to highly specific modification with sulfhydryl-directed reagents. Combination of membrane-permeable and -impermeable reagents can provide valuable information about the extracellular or cytosolic posi-tion of a cysteine (22,23), whereas cysteines embedded in the membrane, unless exposed to a water-containing cavity, are not reactive with these reagents (24 -26). This makes the technique a valuable tool for topological studies and even allows detection of conformational changes (27,28). In addition, in combination with disulfide cross-linking strategies, domains can be identified that are remote in the primary structure but in close proximity in the tertiary structure of the protein (29 -31).
Taking advantage of these unique properties of cysteinescanning mutagenesis, we have studied here the contribution of TMs 6 and 7 to a potential hydrophilic pocket in the ␥-secretase complex.

EXPERIMENTAL PROCEDURES
Site-directed Mutagenesis and Generation of Stable Cell Lines-All mouse PS1 mutants were constructed using the multisitedirected mutagenesis kit (Stratagene). Immortalized mouse embryonic fibroblasts (MEFs) derived from PS1/PS2-deficient mice were cultured in Dulbecco's modified Eagle's medium/ F-12 containing 10% fetal bovine serum (Sigma). At 30 -40% confluency, the MEFs were transduced using a replication-defective recombinant retroviral expression system (Clontech) with either wild-type or mutant PS1. Cell lines stably expressing the desired proteins were selected based on their acquired resistance to 5 g/ml puromycin.
Preparation of Cell Lysates and Immunoblotting-Total cell extracts were prepared in lysis buffer containing 250 mM sucrose, 5 mM Tris-HCl (pH 7.4), 1 mM EGTA, 1% Triton X-100, and Complete protease inhibitors (Roche Applied Science). After centrifugation at 13000 ϫ g for 15 min at 4°C, 20 g of protein from the post-nuclear extracts were separated on 4 -12% BisTris gels (Invitrogen). Proteins were transferred to nitrocellulose membranes and then blocked and probed with antibodies as indicated. For detection, horseradish peroxidasecoupled secondary antibodies (Bio-Rad) were used followed by chemiluminescence detection with Renaissance (PerkinElmer Life Sciences). Quantifications were performed by means of densitometry.
Statistical Analysis-Data from three independent experiments were used for calculations (S.E. values of the mean are indicated), which were then subjected to one-way analysis of variance with a Bonferroni correction to determine their significance.
Transduction with APP Adenovirus-Urea Gel Electrophoresis-Subconfluent stable MEF cell lines were transduced with the recombinant adenovirus Ad5/cytomegalovirus-APP bearing human APP-695 (33) for 7 h at 37°C, after which they were kept overnight at 37°C in Dulbecco's modified Eagle's medium supplemented with 0.2% fetal bovine serum. 24 h post-infection, the conditioned medium (1000 l total volume) was collected and cleared by centrifugation, and 10 l were assayed immediately for the production of A␤40 and A␤42 by specific ELISA (see below). A similar volume sample was used for the determination of the amount of secreted APP fragments (APPs) by SDS-PAGE and direct Western blotting with the polyclonal antibody 22C11 (Chemicon). The remaining conditioned medium (equal volumes) was immunoprecipitated overnight with mAb B7/8 raised against the N terminus of the A␤ sequence (34) and 30 l of protein G-Sepharose (Amersham Biosciences). After extensive washing, the bound material was eluted from the beads by boiling for 10 min in sample buffer (0.74 M BisTris, 0.32 M Bicine, 0.88 M sucrose, 2% SDS, and 0.015% bromphenol blue) and was loaded on a 12%T/5%C (%T, percentage (w/v) of total acrylamide monomer; %C, percentage (w/v) of bisacrylamide/total acrylamide monomer) Bicine/Tris SDS-PAGE gel containing 8 M urea (35). Separation was allowed to proceed at room temperature for 2 h at a constant current of 24 mA/gel and was followed by Western blotting and visualization of the A␤ species with mAb WO2.
ELISA-For the measurement of secreted A␤40 and A␤42, specific ELISA kits (The Genetics Company) were used according to the manufacturer's protocol.
Analysis of Notch Processing-Subconfluent MEF cell lines were infected with the Ad5/dE1dE2a/cytomegalovirus Myctagged Notch ⌬E adenovirus (36) for 24 h, and after treatment with the proteasomal inhibitor lactacystin (Calbiochem) for 4 h at 37°C, cell extracts were prepared. Samples of 10 g of total protein were separated on a 7% Tris acetate gel, transferred to nitrocellulose membranes, and probed with the appropriate antibodies to assess the levels of Notch ⌬E infection and NICD production, respectively.
Blue Native Gel Electrophoresis-This method was performed as described previously (37), with the exception that the samples (20 g) were separated on a 5-16% polyacrylamide gradient at a constant voltage of 200 for 3.5 h.

RESULTS
Generation of a Cysteine-less PS1-Mouse PS1 contains five endogenous cysteines (Fig. 1A), which we replaced with alanine using site-directed mutagenesis. The resulting "cysteine-less" (Cys-less) PS1 was able to rescue the maturation of Nicastrin and the stabilization of PEN-2, as well as the accumulation of unprocessed APP C-terminal fragments in PS1 Ϫ/Ϫ PS2 Ϫ/Ϫ fibroblasts (Fig. 1B). Further functional characterization of the Cys-less PS1 revealed that it has a similar rescuing activity to the wild-type PS protein in PS1 Ϫ/Ϫ PS2 Ϫ/Ϫ fibroblasts as far as it concerns the in vivo processing of three major ␥-secretase substrates, APP (Fig. 1C), N-cadherin, and Notch (data not shown).
Transmembrane domains 6 and 7 contain the putative catalytically active aspartates and therefore likely contribute structurally also to the catalytic site of PS1. Alignment of these domains from various species (supplemental Fig. S1) revealed, apart from the two catalytic aspartates, several other highly conserved amino acids that could be involved in substrate binding, in catalysis, or in delineating a hypothetical interior watercontaining chamber in the complex (38). We, therefore, substituted these residues one by one with cysteines and generated stably transfected PS1 Ϫ/Ϫ PS2 Ϫ/Ϫ fibroblast cell lines with the mutants in a similar manner as the cysteine-less PS1 (Fig. 1C). Two mutations (G382C and K395C) resulted in impaired PS1 endoproteolysis, but all 22 mutants were nevertheless able to rescue the stabilization of PEN-2 and the maturation of Nicastrin. In addition, we also substituted the catalytic aspartic residues, both individually and in combination, without again perturbing the ability of PS1 to stabilize the other ␥-secretase components (Fig. 1D).
Water Accessibility of Introduced Cysteines and Disulfide Cross-linking-Recent electron microscopy studies have shown that an electrolucent channel is present in the interior of the ␥-secretase complex. Our primary aim was to investigate biochemically whether such a postulated "hydrophilic pocket" (38) indeed exists and to delineate the amino acid residues exposed to it. We used the membrane-permeable reagent EZ-link biotin-HPDP, which readily reacts via its pyridyldithiol moiety with free sulfhydryl (ϪSH) groups exposed to water.
As shown in Fig. 2, A and B, of 10 different residues in TM6 tested, only W247C was labeled. In TM7, in contrast, 6 of 11 introduced cysteines showed reactivity with biotin-HPDP. In addition, PS1 D385C in TM7 was modified, whereas the D257C mutant in TM6, which represents the putative second catalytic aspartate of PS1, was not modified in our assay. However, in order for the aspartic residues to perform substrate cleavage, they should be in close proximity to each other and be exposed to a hydrophilic environment. To investigate this further, we used a specific disulfide cross-linking strategy in the cell line expressing the double mutant PS1 D257C/D385C. We chose two homobifunctional alkylthiosulfonates of the type depicted in Fig. 2C (3,6-dioxaoctane-1,8-diyl bismethanethiosulfonate and 1,2-ethanediyl bismethanethiosulfonate with spacer arm lengths 13 and 5.2 Å, respectively), which react selectively with ϪSH groups, resulting in the formation of disulfide bridges between two cysteines and the spacer arm of the cross-linker. This reaction can only take place when the cysteines have free sulfhydryl groups accessible to water and are located at a maximum distance from each other equal to the length of the spacer arm of the cross-linker.
Membrane extracts from PS1 D257C/D385C cells treated with either of the two cross-linkers at 4°C (to reduce protein molecular motions) and separated in SDS-PAGE under nonreducing conditions displayed a band running at an apparent molecular weight close to full-length PS1 (Fig. 2D). Because this band could be stained with antibodies specific for both PS1 NTF and CTF, we propose that it reflects an intramolecular cross-linking product between the two cysteines replacing the aspartates in TMs 6 and 7. The disulfide bond causes an expected shift in mobility compared with the full-length unprocessed protein, because it prevents complete unfolding under non-reducing conditions. When the free sulfhydryls are blocked with the alkylating agent NEM prior to the cross-linking reaction (Fig. 2D, lanes labeled with ϩ NEM) or when reducing conditions are applied (data not shown), this band is not observed confirming its specificity. Furthermore, no crosslinked products are observed with Cys-less PS1, single D257C and D385C PS1 mutants, or a control mutant with two cysteines at remote positions in TMs 1 and 9. Note also that no band derived from intermolecular cross-linking of two different PS1 molecules was observed in any case under our experimental conditions. This experiment demonstrates that, in the tertiary structure of PS1, cysteines introduced at the positions of the catalytic aspartic residues are both accessible to water, facing each other with a maximal distance of ϳ5.2 Å.
Activity of the Mutants on APP Processing-We next investigated whether any of the cysteine substitutions influences the ␥-secretase processing of APP. As shown in Fig. 3A, APP-CTF fragments generated by ␣-secretase from endogenously expressed APP (the direct substrate for ␥-secretase) accumulate in PS knock-out fibroblasts. This phenotype could be completely rescued by reintroduction of wild-type or Cys-less PS1 in these cells, but not, as expected, by PS1 bearing either the D257C and D385C mutations or each of the substitutions G382C, G384C, F388C, Y389C, or K395C. Therefore, these residues seem to be particularly important for the activity of the protease.
After transduction of the fibroblasts with full-length APP695Swe and direct quantification of the A␤ produced by ELISA (Fig. 3B), we observed that, indeed, three of the mutants (G382C, G384C, and K395C) were causing a total loss of function with regard to A␤ generation. Interestingly, the Y389C mutation displayed residual activity (ϳ10% A␤40 production compared with the Cys-less PS1) but no detectable A␤42 production. The remaining mutants can be divided into three categories: 1) the ones that reduce the production of both A␤40 and A␤42 (S254C, V261C), 2) those that cause a significant decrease in the levels of A␤40 with minor effects on A␤42 (T245C, Y256C, F388C), and 3) those that produce amounts of A␤ similar to the wild-type or Cys-less PS1 (all of the rest). Strikingly, none of these mutations seemed to severely affect the ratio of A␤42/40 produced (Fig. 3B), with the exception of Y256C and F388C, which, due to a dramatic reduction in A␤40, behaved like extreme "clinical" familial Alzheimer disease mutations (with 4.5-and 6-fold increase in the ratio, respectively, compared with Cys-less PS1). The effect observed for some of the mutations in the individual production of A␤40 and A␤42 was also independently confirmed by urea gel electrophoresis (supplemental Fig. S2).
Effects of the Mutations on the Processing of Other Substrates-Next, we analyzed the behavior of the cysteine mutants in the processing of N-cadherin. Similar to APP, the C-terminal fragment of N-cadherin generated by metalloprotease cleavage accumulated in PS-deficient cells. This knock-out phenotype was not rescued by the T245C, S254C, Y256C, V261C, G382C, G384C, F388C, Y389C, K395C, and the catalytic D257C and D385C mutants (Fig. 4A).
Finally, the production of NICD (Notch intracellular domain) from a membrane-tethered form of Notch (Myctagged Notch ⌬E) was investigated (Fig. 4B). Similar to APP and N-cadherin, PS1 bearing the G382C, G384C, Y389C, or K395C mutations were unable to support any NICD production. Minimal activity was, however, also seen with the T245C, S254C, Y256C, and V261C mutants in contrast to what we observed for APP. In addition, three mutants appeared to increase Notch processing significantly (W247C, L250C, and L258C). Finally, and important in the context of the question of whether APP and Notch processing can be specifically modulated by ␥-secretase, the F388C substitution, which caused a remarkable decrease in the levels of A␤40, did not affect significantly the production of NICD. To verify that the observed effects on the processing of different substrates were not a consequence of deficient ␥-secretase complex formation, we confirmed the integrity of the complexes by blue native gel electrophoresis (supplemental Fig. S3).

DISCUSSION
In the present study, we have reported our efforts to investigate the catalytic site of PS1 by cysteine-scanning mutagenesis. FIGURE 2. Accessibility of unique PS1 cysteines to sulfhydryl-specific reagents. A, cysteines introduced in TM6 were exposed to biotin-HPDP and precipitated with neutravidin beads in the presence of Triton X-100. Biotinylated material was revealed by Western blotting using the indicated antibodies. PEN-2 (one endogenous extracellular cysteine) and wild-type (WT) PS1 NTF (with three endogenous cysteines) serve as positive controls. PS1 CTF, which does not contain cysteines, and Nicastrin, with all its 12 cysteines clustered in disulfide bridges as predicted with the DISULFIND program (56) are negative controls. B, reaction of cysteines in TM7, as well as cysteines substituting the catalytic aspartates (D257C and D385C) with biotin-HPDP. C, chemical structure of the cross-linker 3,6-dioxaoctane-1,8-diyl bismethanethiosulfonate. D, specific cross-linking of D257C with D385C in the PS1 D257C/D385C mutant. Membrane fractions were treated with cross-linker with (ϩ) and without (Ϫ) preblocking the free sulfhydryls with NEM, and the proteins were separated under nonreducing conditions on a 7% Tris acetate gel. FL, full-length.
A primary prerequisite for the use of this technique is the generation of a protein lacking all endogenous cysteines but retaining the structure and functional properties of the wild-type molecule. Although clinical mutations have been identified at three of the five endogenous cysteines in PS1 (C92S, C263F/R, C410Y) (39 -41), we have found that, under our experimental conditions, their substitution to alanine does not affect ␥-secretase cleavage of the three substrates (APP, Notch, and N-cadherin) examined here. Thus, these mutations seem to be well tolerated and do not affect significantly the structure of PS1, as assessed in our cell biological experiments. Recently, Kornilova et al. (42) have suggested that cysteines Cys-92, -410, and -419 may contribute to the active site of PS1. However, they did not check to what extent their replacement with serine (or alanine) interferes with this function. This together with significant methodological differences might explain the discrepancies between their findings and the findings in the current study. Although we cannot exclude entirely at this moment that the cysteine substitutions in PS1 might indeed affect the activity of the complex in other assays (e.g. cell-free assays), our experiments have clearly demonstrated that the Cys-less PS1 protein is able to rescue efficiently many of the normal functions of PS1 in the cell-based assays used in the current work.
When we assessed the accessibility of cysteines introduced in TM7 to a sulfhydryl-specific reagent, several positions, including the catalytic Asp-385, were reactive, implying that these residues are accessible to water. Although this domain does not seem to be a classically amphipathic helix, most of the accessible residues cluster on one side (Fig. 5). Recently, Lazarov et al. (2006) have reported the presence of an interior translucent chamber in purified ␥-secretase complex as visualized by electron microscopic tomography. In accordance with this morphological evidence, our experiments prove biochemically the existence of a water-filled cavity in the intramembranous part of ␥-secretase. Surprisingly, of all the mutants analyzed in TM6, including the catalytic Asp-257, only one was readily accessible to our reagent (W247C). This negative data can mean that either these residues are all facing the lipid bilayer or that they are buried in the protein interior and thus inaccessible to the bulky biotin-HPDP. Because the second aspartic residue (Asp-257) of the catalytic dyad of PS1 should be accessible to water during hydrolysis of the substrates, we reasoned that the second hypothesis was more likely. To discriminate between the two possibilities, we replaced simultaneously both catalytic aspartates with cysteines and showed that they could be cross-linked. This experiment confirms for the first time experimentally that the proposed catalytic aspartates (Asp-257 and -385) are facing each other in an aqueous environment with a maximum distance of 5.2 Å. Thus, we conclude that Asp-257 is also accessible to water, at least in the active conformation when both aspartyl residues should be closely opposed to each other. The fact that the D257C residue did not react with more bulky chemical reagents such as biotin-HPDP might indicate that either TM6 is densely packed (but water-exposed) and not accessible to these reagents because of steric hindrance or that we probed, with these chemicals, mainly the inactive ␥-secretase complex. Indeed, several reports suggest that only part of wild-type PS1 in cells is actually associated with the active complex (43)(44)(45). Thus, our data are compatible with the possibility that TM6 changes conformation upon binding of the other components of the complex or after binding of the substrates, bringing the active Asp-257 residue in line with the other active Asp-385 residue in the catalytic site of the complex. This possibility is under further investigation and will require discrimination between active and inactive ␥-secretase complex.
Furthermore, our scanning approach revealed several new residues within TMs 6 and 7 that are critical for the catalytic function of PS1. Amino acid substitutions G382C, G384C, and K395C rendered the complex completely inactive. The two gly-  cines are part of the GXGD motif, highly conserved not only in Presenilins, but also Presenilin-like proteins (46), signal peptide peptidase (47), and the type-4 prepilin peptidases (48). Such transmembrane domain glycines have an essential role in helix flexibility (49) and also often occur at helix-helix interfaces, facilitating closer packing of the TM helices (50). In particular, Gly-384 faces the water-filled channel, and its substitution with alanine constitutes one of the most severe and well characterized Alzheimer disease clinical mutations, whereas several artificial mutations of this residue render PS1 completely inactive (48). On the other hand, the position and water accessibility of Lys-K395 suggest a possible involve-ment in the catalytic mechanism, although further analysis is needed to confirm this hypothesis.
Interestingly, substitution of Phe-388, which lies on the same face of the helix as Lys-395 (Fig. 5B) results in a remarkable loss of both APP and N-cadherin processing but with minimal effect on NICD production. Four additional conserved residues were identified in TM6 (Thr-245, Ser-254, Tyr-256, and Val-261), which when substituted, cause an almost complete loss of Notch and N-cadherin cleavage, whereas A␤ production is only partially affected. Specifically, the Y256C and T245C mutants are situated on the same side of the helix and decrease only A␤40, whereas the other two (S254C and V261C), which align with the catalytic Asp-257, decrease both A␤40 and A␤42. Taking into account that the active and substrate binding sites on PS1 are distinct but close to each other, these observations indicate that the amino acids in question affect differentially the binding of the substrates per se or their accommodation in the catalytic site during cleavage.
Finally, particularly revealing was the substitution of the conserved Tyr-389 in TM7 with cysteine, causing a severe loss of function of PS1. According to the most widely accepted acid-base catalytic mechanism for aspartic proteases, one aspartate is deprotonated and acts as a general base, activating a water molecule and forming OH Ϫ in the transition state, and the other aspartate (general acid) donates a proton to the carbonyl of the substrate, facilitating the formation of a tetrahedral intermediate. Very often hydrogen bonds between the carboxylates of the aspartates and neighboring side chains are needed to keep the aspartates co-planar and to assist in the proton transfers during the catalysis (51,52). Because Tyr-389 is located very close to and aligns with the catalytic Asp-385 within the water-filled cavity, we propose that it is involved in the catalytic mechanism through the formation of a hydrogen bond with the side chain of Asp-385. The substitution of Tyr-389 with cysteine, which is shorter than tyrosine and not as good a proton donor, could then affect the catalysis by altering the protonation state of the catalytic residue. Similar interactions underlying the aspartyl-mediated catalytic mecha- nism have been reported for cathepsin E (A1 family of aspartic proteases), with Tyr-20 forming a hydrogen bond with Asp-43 (53) and Trichoderma reesei cellobiohydrolase II (family B of glycoside hydrolases), where Tyr-179 interacts with deprotonated Asp-175, maintaining it in a charged state (54,55).
In conclusion, our results provide experimental evidence for the presence of a water-filled cavity within the ␥-secretase complex, likely essential for substrate hydrolysis by the two closely opposed aspartic residues in the catalytic core of PS1. Further analysis identified crucial amino acids in TM6 and TM7 that differentially affected substrate proteolysis, providing experimental proof that substrates are differentially handled in the catalytic cleft of Presenilin. Further work is needed to investigate whether these fundamental properties can be translated into the development of more specific drugs to block the production of the A␤42 peptide without affecting other important ␥-secretase substrates in significant ways.