Inhibitors of γ-secretase stabilize the complex and differentially affect processing of amyloid precursor protein and other substrates

Abstract

γ-Secretase inhibitors (GSIs) are drugs used in research to inhibit production of Aβ and in clinical trials to treat Alzheimer's disease (AD). They inhibit proteolytic activities of γ-secretase noncompetitively by unknown mechanisms. Here, we used cortical neuronal cultures expressing endogenous levels of enzymes and substrates to study the effects of GSIs on the structure and function of γ-secretase. We show that GSIs stabilize the interactions between the C-terminal fragment of presenilin (PS-CTF), the central component of the γ-secretase complex, and its partners the APH-1/nicastrin and PS1-NTF/PEN-2 subcomplexes. This stabilization dose-dependently correlates with inhibition of N-cadherin cleavage, a process limited by enzyme availability. In contrast, production of amyloid precursor protein (APP) intracellular domain (AICD) is insensitive to low concentrations of GSIs and is limited by substrate availability. Interestingly, APP is processed by both PS1- and PS2-containing γ-secretase complexes, while N-cadherin and ephrinB1 are processed only by PS1-containing complexes. Paradoxically, low concentrations of GSIs specifically increased the levels of Aβ without affecting its catabolism, indicating increased Aβ production. Our data reveal a mechanism of γ-secretase inhibition by GSIs and provide evidence that distinct γ-secretase complexes process specific substrates. Furthermore, our observations have implications for GSIs as therapeutics because processing of functionally important substrates may be inhibited at lower concentrations than Aβ.—Barthet, G., Shioi, J., Shao, Z., Ren, Y., Georgakopoulos, A., Robakis, N. K. Inhibitors of γ-secretase stabilize the complex and differentially affect processing of amyloid precursor protein and other substrates.

Presenilin (PS) is the catalytic component of the γ-secretase complex that processes the amyloid precursor protein (APP), producing the Aβ peptides, the structural components of the amyloid depositions of Alzheimer's disease (AD). This complex also promotes the cleavage of a large number of cell surface proteins, including APP, Notch1, cadherins, and ephrinB1, producing peptides that have been shown to regulate signal transduction and gene expression (1, 2). In addition to PS, the γ-secretase complex contains at least 3 other partners, including the anterior pharynx-defective 1 (APH-1), nicastrin (NCT), and presenilin enhancer 2 (PEN-2). The latter protein has been reported to stimulate the endoproteolysis of full-length PS zymogen into catalytically active fragments PS-NTF and PS-CTF (3). Genetic studies showed that both PS and its substrate APP play causative roles in the development of the forms of familial AD (FAD). In addition, brain amyloid depositions of Aβ are used in the diagnosis of AD and have also been proposed to play causative roles in its development. Thus, γ-secretase inhibitors (GSIs) have been used to lower the in vivo levels of Aβ and treat the disease (4). A number of groups however, reported that prolonged treatment of mice or humans with micromolar concentrations of GSIs resulted, after an initial decrease, in levels of Aβ exceeding the starting levels (4–6). Furthermore, low (nanomolar) concentrations of GSIs increased the in vivo levels of Aβ without an initial inhibitory effect (4, 7), although it was unclear whether this effect resulted from increased production or decreased degradation of Aβ. The inhibitory mechanisms of GSIs are under investigation, and recent data indicate that they inhibit catalysis noncompetitively, consistent with a model where substrates bind a docking site before migrating to the catalytic site (8–10). To examine whether GSIs modify the conformation of the γ-secretase, we studied their effects on the interactions between components of the γ-secretase complex and on substrate proteolysis. Our data show that GSIs increase the interactions between PS1-CTF and its binding partners, APH-1/NCT and PS1-NTF/PEN-2 heterodimers, and differentially affect processing of substrates. In addition, we obtained evidence supporting an increased production of Aβ42 at low concentrations of GSIs.

MATERIALS AND METHODS

Materials and antibodies

Mouse monoclonal antibody 33B10 against residues 331–350 of PS1, polyclonal antibody R222 against PS1 N-terminal fragment, and R57 antibody against C-terminal domain of APP were described previously (11). Mouse anti-N-cadherin (cat. no. 610920) was from Becton Dickenson Transduction Laboratory (Franklin Lakes, NJ, USA). Anti-APH-1 specific of APH-1aL isoform (PA1–2010) was from Affinity BioReagents (Golden, CO, USA); anti-NCT (N1660) was from Sigma (St. Louis, MO, USA). Anti-PS2-NTF (7861) and ephrinB1-Cter (c18) were from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Anti-PEN-2 (NE1008), anti-PS2-CTF (PC235), and GSIs L665,458 and DAPT were from Calbiochem (San Diego, CA, USA).

Primary neuronal cultures

Cortices from embryonic day 17 (rat) or 15 (mouse) embryos were dissected and dissociated in trypsin. Neuronal progenitors were plated in serum-free Neurobasal + B27 medium. Cultures were maintained at 37°C in a humidified atmosphere in 5% CO₂ (10⁶ cells/well in 6-well plate). All experiments were performed with neurons cultured for 8 days in vitro (DIV).

Analysis of γ-secretase complexes

Neuronal cultures were treated or not with inhibitors before lysis in a dodecylmaltoside-based lysis buffer (50 mM HEPES, pH 7.4; 100 mM NaCl; 10% glycerol; and 0.5% DDM). Samples were immunoprecipitated (IPed) with APH-1, NCT, or PS1-NTF antibodies. Obtained proteins were separated by WBs using Tris-tricine gels.

In vitro γ-secretase activity assay

Cortical neurons of 8 DIV were treated or not overnight (O/N) with DAPT or L685,458, and then scraped in hypotonic buffer (10 mM MOPS and 10 mM KCl). Membranes purified from postnuclear fraction were either incubated at 37°C in a citrate buffer (150 mM, pH 6.4) to allow γ-secretase enzymatic activity or kept at 4°C. In some experiments, DAPT or L685,458 was added to the membrane suspension. After 16 h of incubation, the reactions were stopped by the addition of Laemmli buffer, and proteins in samples were separated by Western blot (WB) analysis using 10–20% gradient Tris-tricine gels. Membranes were probed for the analysis of APP with R1 antiserum specific to cytoplasmic APP (12). N-cadherin full-length and C-terminal fragments were detected with anti-N-cadherin monoclonal antibodies (BD Transduction Laboratories). In experiments reported in Fig. 4A, PS1/2 double-knockout (PS1/2 dKO) mouse embryonic fibroblasts (MEFs) were lysed in hypotonic buffer (10 mM MOPS; 10 mM KCl) and membrane purified by ultracentrifugation. Membrane proteins were extracted with CHAPSO 1% lysis buffer and used as a source of APP-CTFs (which accumulate in fibroblasts lacking PS).

Figure 4.

Production of AICD is limited by APP-CTF substrates, while production of Ncad- and ephrinB1-CTF2 is limited by enzyme availability. A) Production of AICD is proportional to the amount of substrate. Membrane proteins from PS1/2 dKO MEFs were used as a source of APP-CTFs and added to membranes from APP^−/− cortical neuronal cultures. Neuronal membranes were then incubated at 37°C in a citrate buffer (run) or directly suspended in Laemmli buffer (start). Reactions were stopped by Laemmli buffer, and proteins were analyzed by WBs. B) Fifty percent reduction of PS1 in PS1^+/− has no effect on AICD production. Membranes from mouse cortical neurons of PS1^+/+, PS1^+/−, and PS1^−/− embryos were prepared as above, and γ-secretase activities were analyzed by incubating at 37°C. C) Reduction of PS1 proportionally decreases production of N-cad/CTF2 and ephrinB1/CTF2. Experiment was performed as in B, except membranes were probed for N-cadherin and ephrinB1.

Surface biotinylation

Neuronal cultures were incubated with 1 mg/ml sulfo-NHS-SS-biotin in PBS at 4°C for 1 h, washed with 0.1 M glycine, and lysed in a dodecylmaltoside-based lysis buffer (see above). The cell extracts were incubated with streptavidin-agarose beads O/N. The bound proteins were washed 3 times with PBS and eluted with Laemmli.

Medium preparation and Aβ analysis using sandwich ELISA assay

Growth medium from neuronal culture of 8 DIV was changed to fresh Neurobasal medium without B27 (to eliminate insulin that may interfere with both Aβ production and degradation), and cultures were treated with GSIs at various concentrations, as indicated in figures, for 16 h. Medium was then collected, supplemented with 1 mM pefablock, and centrifuged at 14,000 g for 15 min to remove any membrane contaminants. In experiments reported in Fig. 5C, medium from a wild-type neuronal culture was transferred to an APP^−/− neuronal culture. Aliquots were analyzed after 16 h of incubation to access Aβ42 degradation. Collected medium was analyzed using the Covance Aβ42 murine kit (Covance, Madison, WI, USA) following the manufacturer's instructions.

Figure 5.

Low concentrations of GSIs increase Aβ without affecting its degradation. A, B) GSIs increase Aβ42 level at nanomolar concentrations in neuronal culture. Rat cortical neurons were treated with different concentrations of DAPT for 16 h in fresh medium. Aβ42 levels in the medium were determined using an ELISA assay. B) Same as in A but using GSI L685,458. C) Degradation of exogenous Aβ42 in APP^−/− is inhibited by insulin but not by GSIs. WT mouse neuronal culture medium was replaced with fresh Neurobasal and then incubated for 16 h. This 16-h-conditioned medium was used as a source of Aβ42 and transferred to APP^−/− neuronal cultures treated either with DAPT or insulin, which inhibits Aβ degradation by insulin-degrading enzyme. Aβ42 in medium was measured as above. D) Degradation of Aβ and p3 peptides is inhibited by insulin but not DAPT. WT mouse cortical neurons were radio-labeled with S³⁵-methionine/cysteine. Medium from these cultures was used as a source of exogenous Aβ and added to sister cultures in the presence or absence of DAPT or insulin (100 μg/ml). Cultures were incubated O/N, and supernatants were then IPed with anti-Aβ and anti-p3 Abs. Proteins were analyzed on 10–20% gradient Tris-tricine gels and by autoradiography. E) Proposed model of allosteric effects of GSIs containing two γ-secretase complexes associated in trans. PS1-CTF fragment of one complex associates with the NTF fragment of the adjacent γ-secretase complex. To help visualize the enzymatic sites, APH-1 and NCT of one complex have been omitted in the right middle panel. The two catalytic aspartates on TM domains VI and VII of each PS are depicted as yellow-red sparks. Right panel illustrates a positive allosteric effect on Aβ42 production at low concentrations of inhibitors when only one half of active sites are occupied. The two aspartates of the PS on the left are blocked by inhibitor and are inactive (shown in black); in the countercomplex, the two aspartates of PS are more active (shown in brighter yellow).

Study of Aβ degradation

Medium from 8-DIV neuronal cultures was changed to fresh medium containing S³⁵-labeled methionine/cysteine. After O/N incubation, medium was purified and concentrated by dialysis and added to sister neuronal cultures grown in the absence of labeled methionine/cysteine. DAPT or insulin was added for 16 h to evaluate their effects on Aβ catalysis. Medium was then IPed with anti-p3/Aβ antibodies, and obtained proteins were separated using 10–20% Tris-Tricine gels. Gels were dried and used for autoradiography.

Statistical analysis

Four independent experiments were performed for statistical analysis of Aβ secreted by neuronal cultures and measured by ELISA, as described earlier. Likewise, 3 independent experiments were performed for the in vitro γ-secretase assay with neuronal membranes. To evaluate statistical significance of the pharmacological treatments, paired t tests were performed against the value of the untreated basal condition. Values of P < 0.05 were considered significant.

RESULTS

γ-Secretase inhibitors enhance interactions between PS1-CTF and PS1-NTF and stabilize the γ-secretase complex

The mechanism by which GSIs block substrate cleavage is under intense investigation, and reports indicate that both transition- and nontransition-state analogs are noncompetitive inhibitors (8, 9). Functional γ-secretase complexes contain at least 4 subunits, including NCT, APH-1, PS-CTF/PS-NTF heterodimer, and PEN-2. APH-1 and NCT form a subcomplex that binds the PS-CTF component of the PS heterodimer, while PEN-2 binds PS-NTF, in an arrangement that places PS-CTF in the center of the proteolytic complex (13–15). We postulated that GSIs affect catalysis by changing the interactions between γ-secretase complex subunits and tested this hypothesis using coimmunoprecipitation experiments of detergent-extracted γ-secretase components. O/N treatment of primary neuronal cultures with DAPT, a potent GSI (16), increased the amounts of PS1-CTF and PEN-2 associated with either APH-1 or NCT, while the ratio of APH-1 to NCT remained constant (Fig. 1A), in agreement with reports of an APH-1/NCT subcomplex. Interestingly, the amounts of PS1-CTF and PEN-2 associated with APH-1/NCT were increased by 2- and 6-fold, respectively. Since PEN-2 binds PS-NTF, this observation suggests that in addition to the interaction between APH-1/NCT and PS1-CTF, DAPT may also increase the interaction between PS-CTF and PS-NTF and possibly between PS-NTF and PEN-2. To test these possibilities, we IPed PS1-NTF and observed that DAPT increases the interaction between PS1-NTF and PS1-CTF (Fig. 1B), but not the interaction between PS1-NTF and PEN-2, which remained constant (Fig. 1B).

Figure 1.

GSIs increase interactions between PS1-CTF and other components of γ-secretase. A) DAPT treatment increases interactions between APH-1/NCT heterodimer and PEN-2. Extract in dodecylmaltoside-based lysis buffer was prepared from rat cortical neurons treated O/N with 2 μM DAPT, and samples were IPed with either anti-APH-1 or anti-NCT antibodies (Abs). IPed proteins were analyzed on WBs using 10–20% gradient Tris-tricine gels. Experiments were repeated 4 times. B) DAPT increases the interaction between PS1-NTF and PS1-CTF but not between PS1-NTF and PEN-2. Samples were IPed with anti-PS1-NTF Abs, and obtained proteins were analyzed on WBs as above. Duplicates are shown. C) DAPT stabilizes NCT to PEN-2 interactions in a concentration-dependent manner. Rat neurons were treated O/N with increasing concentrations of DAPT (molar), and prepared extract was IPed with anti-NCT Abs. Top panel: precipitates were blotted for NCT and PEN-2. Bottom panel: quantification of NCT to PEN-2 interactions, expressed as a percentage of interaction at the highest concentration of inhibitor. EC₅₀ of DAPT effect is ∼30 nM. D) DAPT increases interactions between PS1-CTF and other γ-secretase subunits in a concentration-dependent manner. Samples prepared as in C were IPed with anti-PS1-CTF Abs. E) Transition-state analog L685,458 also increases interaction between NCT and other γ-secretase components. Rat neuronal cultures were treated O/N with 2 μM DAPT or 1 μM L685,458 (L685), and extract was IPed with anti-NCT Abs. F) Model of γ-secretase complex assembly and inhibition. Left and middle panels depict the γ-secretase assembly, with the PS1-CTF fragment of PS1 heterodimer binding the APH-1/nicastrin subcomplex, while the PS1-NTF fragment of PS1 heterodimer binds PEN-2, as described previously (13–15). The two catalytic aspartates on TMs VI and VII of PS are depicted as yellow-red sparks. Right panel illustrates proposed effects of GSIs on structure of the complex. By enhancing component interactions (see text), GSIs induce a closed conformation of the enzyme complex, characteristic of inhibited enzymes (inactive aspartates shown in black).

We then evaluated the sensitivity of the interaction between γ-secretase subunits to DAPT concentration. Figure 1C shows that 5 nM DAPT is sufficient to increase the interactions between NCT and PEN-2. A maximal effect was observed at 100 nM DAPT, with an EC₅₀ of ∼30 nM. A comparable sensitivity of the interactions between PS1-CTF and the other γ-secretase components to DAPT was seen when PS1-CTF was IPed (Fig. 1D). In this case, all complexes between PS1-CTF and other subunits increased, in agreement with reports that DAPT acts on PS-CTF (17), the central component of the γ-secretase complex. Since DAPT is routinely used in inhibition assays at concentrations > 100 nM, our data indicate that at these concentrations, the drug acts as an effective stabilizer of the γ-secretase complex, increasing the interactions between PS1-CTF and its partners. Moreover, this stabilization is not associated with a defect in γ-secretase trafficking, because DAPT does not affect cell surface expression of the γ-secretase subunits (Supplemental Fig. S1). To examine whether this stabilization effect was restricted to nontransition-state analog inhibitors like DAPT, we analyzed the effect of γ-secretase inhibitor L685,458, a transition-state analog. As shown in Fig. 1E, L685,458 also increases the interaction between NCT/APH-1 and other components of the γ-secretase complex. Together, our results show that GSIs increase the interactions between PS1-CTF and its partners, including the APH-1/NCT and the PS1-NTF/PEN-2 subcomplexes. Furthermore, our data conform with enzymology paradigms predicting that enzymes exhibit a tensed or closed conformation when inhibited and a relaxed or opened conformation when active (18). We propose that GSIs inactivate γ-secretase by closing the PS1-CTF/NTF interface where the catalytic site forms and by stabilizing the complex in a tensed conformation (Fig. 1F).

GSIs stabilize both PS1- and PS2-containing γ-secretase complexes independently of APP

We then asked whether the stabilization of γ-secretase complexes by GSIs is specific to PS1-containing complexes or also occurs with PS2-containing complexes. To avoid interference from PS1, we analyzed the effect of DAPT on PS2-containing complexes in PS1-KO (PS1^−/−) mouse cortical neurons. Consistent with data obtained in cell lines (19), absence of PS1 results in decreased levels of both NCT and PEN-2 but increased levels of neuronal PS2 (Fig. 2A, left panels). IP of APH-1 revealed increased amounts of PS2 associated with APH-1 in PS1^−/− neurons compared to WT neurons (Fig. 2A, right panels, lanes 1, 5), while IP of PS2-CTFs revealed increased amounts of both APH-1 and NCT bound to PS2, even though total levels of NCT decreased in these neurons (Fig. 2B). Furthermore, DAPT stabilized the interaction of APH-1 with both PEN-2 and PS2-CTF (Fig. 2A, right panel), as well as the interaction between the NTF and CTF fragments of PS2 (Fig. 2C), indicating that similar to PS1-containing complexes, DAPT also stabilizes interactions between components of the PS2-containing γ-secretase complexes. To examine whether the GSI-induced stabilization of γ-secretase complexes involves APP-CTF substrates that accumulate during enzyme inhibition, we measured inhibitor effects on γ-secretase complexes in the presence or absence of APP. Figure 2D shows that absence of APP has no effect on the DAPT-induced stabilization of these complexes. These data support the conclusion that stabilization of γ-secretase complexes by GSIs is due to PS binding of these drugs.

Figure 2.

GSIs stabilize both PS1- and PS2-containing γ-secretase complexes independently of APP. A) GSIs also affect PS2-containing γ-secretase complexes. Cortical neurons from PS1^+/+, PS1^+/−, and PS1^−/− mouse embryos were treated O/N with 2 μM DAPT. Protein samples were IPed with anti-APH-1 Abs, and precipitates were analyzed on WBs as in Fig. 1A. B) PS2-containing γ-secretase complexes increase in absence of PS1. Samples as in A were IPed with anti-PS2-CTF Abs. C) DAPT increases the interaction between PS2-NTF and PS2-CTF. Neuronal cultures were treated or not with 2 μM DAPT, and extracts were IPed with anti-PS2-NTF or anti-PS2-CTF Abs. D) DAPT effect on γ-secretase complex is independent of APP. Neurons from APP^+/+ or APP^−/− cultures were lysed, and extracts were IPed with anti-APH-1 or anti-NCT Abs.

GSIs differentially affect the ε-cleavage of APP and N-cadherin

We then asked whether modulation of the interactions of γ-secretase components induced by GSIs correlates with the inhibitory activity of the drugs. Figure 1D indicates that at 10⁻⁷ M, DAPT inhibitor has a significant effect on the interaction between PS1-CTF and PS1-NTF. To examine whether γ-secretase activity is significantly changed at this DAPT concentration, we used an in vitro assay based on cleavage of endogenous substrates that copurify with neuronal membranes (20). We first tested N-cadherin, an abundant neuronal protein that is processed by γ-secretase at the epsilon site to yield C-terminal fragment N-Cad/CTF2 (20). Figure 3A (top panels) shows that incubation of neuronal membranes resulted in the production of N-Cad/CTF2 (lanes 1, 2), and this process was inhibited by increasing DAPT concentrations, with an IC₅₀ of ∼30 nM. At 100 nM DAPT, γ-secretase processing of N-cadherin was mostly inhibited. Surprisingly, however, APP processing was less sensitive to DAPT inhibition. Indeed, in vitro production of AICD was insensitive to DAPT concentrations as high as 75 nM, but decreased at concentrations above 100 nM (Fig. 3A, bottom panel; B). These differences in dose-response inhibition of N-Cad/CTF2 and AICD suggest that processing of APP-CTFα/β substrates is much less sensitive to GSIs than processing of N-Cad/CTF1 substrate.

Figure 3.

ε-Cleavage of APP is less sensitive to inhibitors compared to other substrates. A, B) In vitro dose-response of γ-secretase inhibition. Rat cortical neurons were lysed in hypotonic buffer. Purified membranes were either incubated at 37°C in citrate buffer to follow γ-secretase activity (run) or directly suspended in Laemmli buffer (start; starting material). Under run conditions, membranes were incubated with indicated concentrations of DAPT for 16 h, and reactions were stopped with Laemmli buffer. Proteins were analyzed on WBs as described in Materials and Methods. Membranes were probed for γ-secretase processing of APP and N-cadherin by following AICD and N-cad/CTF2. A) Results from a representative experiment. B) Quantifications of 3 experiments. C) O/N treatment of neurons with GSIs induces a rebound effect on APP but not N-cadherin cleavages. Neuronal cultures were either left untreated (Ctrl) or treated O/N (L685,458) with 1 μM L685,458, and then lysed in hypotonic buffer. Purified membranes from both samples were then incubated at 37°C in citrate buffer (run) or directly suspended in Laemmli buffer (start). A duplicate of run in 1 μM L685,458 was included to assess in vitro inhibition of γ-secretase, and samples were analyzed on WBs. D) APP-CTFα/β substrates of γ-secretase do not inhibit N-cadherin cleavage. WT or APP^−/− mouse cortical neurons were treated O/N or not with 2 μM DAPT; γ-secretase activity was analyzed by incubating membranes at 37°C in citrate buffer, and obtained samples were analyzed on WBs.

The lower sensitivity of APP processing to GSIs is further illustrated by experiments involving O/N treatment of primary neuronal cultures with GSIs. Production of N-Cad/CTF2, but not of AICD, was inhibited in membranes prepared from such cultures incubated O/N with inhibitor L685,458 (Fig. 3C, lanes 1–4). Since AICD production is still mediated by γ-secretase (Fig. 3C, lane 5), the most probable explanation for this differential effect is that drug copurified with the membranes inhibits the γ-secretase cleavage of N-Cad/CTF1 but not cleavage of APP-CTFα/β, in agreement with the lower sensitivity of the APP cleavage to GSIs compared to cadherin cleavage. Interestingly, ephrinB1, another substrate of γ-secretase (2) behaves like N-cadherins (Supplemental Fig. S2). This differential sensitivity of N-cadherin and APP cleavage to GSIs is consistent with the hypothesis that distinct γ-secretase complexes cleave N-cadherin and APP, a hypothesis strongly supported by reports that distinct docking proteins recruit cadherins and APP to γ-secretase complexes. Specifically, p120ctn recruits cadherins (11), and GSAP (γ-secretase activating protein) recruits APP-CTFs (21) to γ-secretase. To further test this hypothesis, we used neuronal cultures from WT and APP^−/−-transgenic mice to ask whether there is substrate competition for γ-secretase enzymes. Figure 3D shows that cadherin processing and production of N-Cad/CTF2 is independent of APP (compare lanes 1, 2 to 5, 6). Furthermore, DAPT inhibition of N-Cad/CTF2 does not depend on accumulated APP-CTFs (Fig. 3D; compare lanes 3, 4 to 7, 8). Together, these observations indicate the existence of distinct γ-secretase complexes that cleave N-cadherin or APP and suggest that distinct cellular γ-secretase complexes may process specific substrates.

Substrate quantity limits AICD production, whereas enzyme quantity limits production of N-Cad/CTF2 and ephrinB1/CTF2

Since enzymatic reactions can be limited by substrate or enzyme availability, we asked whether APP substrate is the limiting factor in our assays by examining whether increasing levels of APP-CTFα/β would increase production of AICD. To this end, we used MEFs derived from PS1/2 dKO mice to prepare an extract with high concentrations of APP substrates. The addition of increased amounts of this extract as a source of APP-CTF substrates to membranes prepared from APP^−/− neurons resulted in a proportional increase in AICD production (Fig. 4A). Moreover, production of AICD was not reduced by a 50% decrease of PS1 in membranes prepared from PS1^+/− neurons (Fig. 4B), indicating that enzyme availability is not limiting production of neuronal AICD. A substantial decrease in the production of AICD was only observed in the total absence of PS1 (Fig. 4B). Although remaining production of AICD in the absence of PS1 is attributed to PS2, our data suggest that this PS cannot fully support AICD production in the absence of PS1. In contrast to AICD, which derives from APP, production of Ncad/CTF2 and ephrinB1/CTF2, the products of the γ-secretase processing of N-cadherin and ephrinB1, respectively, decreased by ∼50% in membranes from PS1^+/− neurons (Fig. 4C), suggesting that cleavage of these substrates is limited by enzyme availability. Interestingly, despite the presence of PS2 (Fig. 4C, bottom panel), no cadherin or ephrinB1 cleavage products were detected in the absence of PS1, indicating that only PS1-containing γ-secretase complexes process cadherins and ephrinB1, as opposed to APP, which can be processed by both PS1- and PS2-containing complexes, although at different efficiencies. Thus, γ-secretase processing of N-cadherin and ephrinB1 is limited by the quantity of γ-secretase enzymatic sites and cannot be performed by PS2-containing γ-secretase.

Low concentrations of GSIs stimulate production of Aβ42

Paradoxically, an increase, rather than a decrease, of Aβ42 levels has been reported in cell lines and animal models treated with low concentrations of GSIs (4, 7). We verified these data in primary neuronal cultures treated with either DAPT (Fig. 5A) or L685,458 (Fig. 5B). Both drugs at concentrations < 50 nM increased the amounts of Aβ42. Since DAPT is a protease inhibitor, we examined whether it stabilizes Aβ42 by inhibiting its degradation. Figure 5C, D shows that at these concentrations, DAPT has no effect on the degradation of this peptide, suggesting that low DAPT concentrations stimulate production of Aβ42. AICD and Aβ are thought to derive from different substrates, as the former is mostly produced from APP-CTFα (22), while the latter derives from APP-CTFβ. Combined with reports that treatment of cell cultures with low DAPT concentrations also fails to change the levels of p3 peptides, another product of APP-CTFα (7), our data suggest that low DAPT concentrations have different effects on the γ-secretase processing of different APP substrates.

DISCUSSION

Aβ peptides, the structural subunits of the amyloid depositions of AD, have been proposed to play crucial roles in the development of the disease, and recent therapeutic strategies use inhibitors of γ-secretase to block their production. The mechanism of GSIs is still under investigation, but it was reported that transition-state analogs predicted to bind the catalytic site of γ-secretase display noncompetitive inhibition toward substrate binding (9), suggesting that substrates first bind a docking site before they migrate to the catalytic site (10). To understand the mechanisms by which GSIs regulate γ-secretase activity, we asked whether these drugs affect the interactions between components of the γ-secretase complex. Because brain Aβ is mostly produced by neurons, we used primary neuronal cultures expressing endogenous levels of enzymes and substrates to avoid problems associated with overexpression systems, such as changes in the enzyme to substrate ratios. We discovered that both transition- and nontransition-state analogs of GSIs stabilize interactions between PS-CTF, the central component of the γ-secretase proteolytic complex, and its partners, the APH-1/nicastrin and PS1-NTF/PEN-2 subcomplexes. These stabilization effects suggest a mechanism by which these drugs inhibit substrate processing. Indeed, a common paradigm of enzymology is that enzymes exhibit a relaxed conformation when active and a restrained or “tensed” conformation when inhibited by drugs (18). We propose that both transition- and nontransition-state analogs of GSIs inactivate γ-secretase by closing the PS CTF/NTF interface that forms the active site of the enzyme. This mechanism is also consistent with observations that GSIs inhibit substrate proteolysis without preventing substrate binding (8, 9).

The stabilizing effect of GSIs on γ-secretase complexes correlates with the inhibitory effects of these drugs on N-cadherin processing (Fig. 3A, B). However, higher concentrations of GSIs are required to inhibit production of AICD compared to N-Cad/CTF2, suggesting that γ-secretase complexes processing APP-CTFα/β are less sensitive to inhibitors than complexes processing N-Cad/CTF1. This lower sensitivity of APP processing to GSIs is also supported by our data showing that, in contrast to N-cadherin, APP processing is not inhibited in membranes prepared from neuronal cultures pretreated with GSI (Fig. 3C). Although the molecular basis of this differential sensitivity of substrates to GSIs needs further examination, recent evidence suggests that the composition of γ-secretase complexes may be tailored to the processing of specific substrates. Indeed, distinct cellular factors, termed γ-secretase docking proteins (GSDPs), recruit N-cadherin or APP to γ-secretase complexes (11, 21). PS1 binds p120ctn, which links cadherin substrates to γ-secretase (11), and recent reports show that APP is recruited to γ-secretase by the GSAP (21). We previously reported that by recruiting γ-secretase to cadherin substrates, p120ctn promotes cleavage of cadherins, while it suppresses cleavage of APP, indicating that p120 recruits γ-secretase complexes to cadherin processing, thus reducing the number of complexes available for APP processing (11). The existence of different pools of γ-secretase complexes specific to distinct substrates is also supported by our data showing that APP and N-cadherin substrates do not compete for γ-secretase (Fig. 3D). Furthermore, we obtained evidence that while APP substrates are processed by both PS1- and PS2-containing γ-secretase complexes, N-cadherin is processed only by PS1-containing complexes, a result in agreement with reports that sequence 330–360 of PS1-CTF is critical for N-cadherin recruitment (11). Indeed, this sequence is very different in PS2, suggesting that PS2-containing γ-secretase complexes do not bind p120ctn and are thus unable to recruit cadherin substrates. Thus, GSDPs may control access of γ-secretase to substrates. Further support for the existence of distinct γ-secretase complexes comes from our data indicating that neuronal γ-secretase complexes contain mostly the low-molecular-weight NCT, in contrast to cell line γ-secretase complexes that incorporate mostly the high-molecular-weight NCT (23). Because the two NCT forms differ in their degree of glycosylation (24, 25), our observation suggests that NCT glycosylation is not required for γ-secretase activity, in agreement with previous reports (24). Together, our data provide strong evidence that distinct γ-secretase complexes process specific substrates and suggest a dynamic steady-state model, according to which γ-secretase components assemble into functional complexes in response to cellular needs for specific substrate processing.

Our data (Fig. 4C) show that in neuronal systems, enzyme availability limits production of N-cad/CTF2 and ephrinB1/CTF2. In contrast, production of AICD is limited by the quantity of γ-secretase substrates APP-CTFα/β. We observed that these substrates accumulate dramatically in membranes treated with inhibitory concentrations of GSIs (Fig. 3C), in contrast to N-Cad/CTF1, which accumulates much less, suggesting a differential removal of these substrates by alternative proteolytic pathways. This accumulation of APP-CTFα/β substrates explains the “Aβ rebound effect” reported in patients treated with inhibitory concentrations of GSI (4–6, 26). As the in vivo concentrations of GSI decrease below inhibitory levels, accumulated APP-CTFβ substrates are processed by the enzyme, leading to increased amounts of Aβ, often higher than the initial levels.

However, recent publications report that treatment of animal models or cell cultures with low concentrations of GSIs increases the in vivo levels of Aβ peptides (4, 7). This increase could result from increased production or decreased degradation of Aβ. Our data showing that Aβ degradation is unaffected by GSIs (Fig. 5C, D) suggest that low concentrations of GSIs stimulate production of Aβ. Indeed, the hyperbolic shape of the dose-response curve of Aβ production (Fig. 5A, B), is consistent with a positive allosteric effect. Since L-685,458 is a transition-state analog, this biphasic response is incompatible with a model that incorporates 1 PS molecule/enzymatic unit. Instead, our data are consistent with a 2-site model, in which at low inhibitor concentrations, targeting of one site results in increasing activity of the other site through a positive allosteric effect induced by conformational changes (Fig. 5E). This explanation is consistent with recent evidence (27–29), including our own data presented here (Supplemental Fig. S3) that PS forms dimers. However, the increased level of Aβ42 secreted by neuronal cultures contrasts with the unchanged level of AICD produced by neuronal membranes treated with low concentrations of GSIs. The lack of effect on AICD is consistent with a recent report that low concentrations of GSIs do not increase the level of p3 peptide, which, like AICD, is also derived predominantly from the γ-secretase cleavage of APP-CTFα (7). Together, these observations suggest that the mechanism involved in the increase of Aβ42 may be specific to the processing of APP-CTFβ.

The finding that processing of APP substrates is less sensitive to GSIs than processing of cadherin substrates has important implications for the use of these inhibitors as therapeutic agents, because processing of substrates other than APP may be inhibited before Aβ. Thus, it would be interesting to determine the GSI sensitivity of other γ-secretase substrates, including Notch1, ErbB4, and EphB receptors (30–32). Nonetheless, our results raise the possibility that currently available inhibitors interfere with important cellular functions before production of Aβ is inhibited. This observation may explain reported clinical side effects in patients treated with GSIs (33) and indicates that a different pharmacological strategy is required to decrease Aβ. Indeed, recent efforts concentrate on the development of selective allosteric γ-secretase modulators that target PEN-2 and stimulate degradation of Aβ to shorter peptides (34). Alternatively, it may be possible to selectively target the GSDPs that recruit specific APP substrates to γ-secretase complexes (21).