Differential Effects of Inhibitors on the γ-Secretase Complex MECHANISTIC IMPLICATIONS

γ-Secretase is a protease complex of four integral membrane proteins, with presenilin (PS) as the apparent catalytic component, and this enzyme processes the transmembrane domains of a variety of substrates, including the amyloid β-protein precursor and the Notch receptor. Here we explore the mechanisms of structurally diverse γ-secretase inhibitors by examining their ability to displace an active site-directed photoprobe from PS heterodimers. Most γ-secretase inhibitors, including a potent inhibitor of the PS-like signal peptide peptidase, blocked the photoprobe from binding to PS1, indicating that these compounds either bind directly to the active site or alter it through an allosteric interaction. Conversely, some reported inhibitors failed to displace this interaction, demonstrating that these compounds do not interfere with the protease by affecting its active site. Differential effects of the inhibitors with respect to photoprobe displacement and in cell-based and cell-free assays suggest that these compounds are important mechanistic tools for deciphering the workings of this intramembrane-cleaving protease complex and its similarity to other polytopic aspartyl proteases.

␥-Secretase is a protease complex of four integral membrane proteins, with presenilin (PS) as the apparent catalytic component, and this enzyme processes the transmembrane domains of a variety of substrates, including the amyloid ␤-protein precursor and the Notch receptor. Here we explore the mechanisms of structurally diverse ␥-secretase inhibitors by examining their ability to displace an active site-directed photoprobe from PS heterodimers. Most ␥-secretase inhibitors, including a potent inhibitor of the PS-like signal peptide peptidase, blocked the photoprobe from binding to PS1, indicating that these compounds either bind directly to the active site or alter it through an allosteric interaction. Conversely, some reported inhibitors failed to displace this interaction, demonstrating that these compounds do not interfere with the protease by affecting its active site. Differential effects of the inhibitors with respect to photoprobe displacement and in cell-based and cell-free assays suggest that these compounds are important mechanistic tools for deciphering the workings of this intramembrane-cleaving protease complex and its similarity to other polytopic aspartyl proteases.
Cerebral accumulation of the amyloid-␤ protein (A␤) 1 is considered a central event in the pathogenesis of Alzheimer's disease (AD). A␤ is produced via ␤and ␥-secretases, proteases that have become important therapeutic targets for AD (1). ␥-Secretase plays a crucial role in determining the proportion of two forms of A␤, A␤ 40 and A␤ 42 . The 42-residue A␤ 42 is more prone to fibril formation and is disproportionately present in the plaques characteristic of the AD brain. Accumulating evi-dence (2)(3)(4) strongly suggests that ␥-secretase is an intramembrane-cleaving aspartyl protease with presenilin (PS) as the catalytic component. Three other multipass membrane proteins, nicastrin, Aph-1, and Pen-2, are genetically linked to ␥-secretase activity (5)(6)(7), and biochemical isolation has provided evidence that these proteins are indeed necessary members of the protease complex (8 -10). Despite the remarkable progress in uncovering the identity of ␥-secretase, its mechanism of action remains unclear.
A body of work (11) suggests that ␥-secretase cleaves amide bonds within the transmembrane regions of its substrates, a poorly understood process of hydrolysis within a hydrophobic environment. Elucidating the molecular interaction between an inhibitor and its enzyme target can help identify the enzyme and provide insight into the catalytic mechanism. The study of peptidomimetic inhibitors of ␥-secretase that contain classic aspartyl protease transition state-mimicking moieties led to the suggestion that ␥-secretase is an aspartyl protease and that two conserved aspartates in presenilins are catalytic residues (12). PS is processed into N-terminal (NTF) and C-terminal (CTF) fragments. These fragments are metabolically stable, remain associated, and their formation is tightly regulated, suggesting that together they are the bioactive form of PS (12). The direct binding of transition state analog ␥-secretase inhibitors to these fragments strongly suggests that the active site is at the NTF/CTF heterodimeric interface (3,4), consistent with the fact that each subunit contributes one of the two critical aspartates (2).
The recent discovery of structurally diverse inhibitors suggests that these new compounds may also be important molecular probes for the protease complex. Several non-transition state analog inhibitors are very potent; however, unlike transition state analogs, their site(s) of interaction within the ␥-secretase complex is unclear. To probe the mechanism of ␥-secretase inhibitor action, we studied the ability of structurally diverse compounds to displace a transition state-based photoactive molecule from its target, the ␥-secretase active site at the PS1 NTF/CTF interface. Differential effects of these compounds suggest that they inhibit ␥-secretase by distinct mechanisms and are thus important new probes to elucidate the workings of this complex protease. Moreover, the ability of a transition state analog inhibitor of signal peptide peptidase (SPP), a multipass membrane aspartyl protease with presenilin-like motifs (13,14), to displace the ␥-secretase photoprobe suggests that the active site topographies of these two proteases are similar.
Cell-based and Cell-free A␤ Production-Inhibition of A␤ production in cells and measurement by sandwich ELISA were performed as described previously (19). The cell-free assay was performed as reported previously (8). IC 50 values were estimated by plotting the ELISA data * This work was supported by National Institutes of Health Grants NS41355 and AG17574 and Alzheimer's Association Grant IIRG-02-4047 (to M. S. W.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
‡ To whom correspondence should be addressed. E-mail: mwolfe@ rics.bwh.harvard.edu. 1 The abbreviations used are: A␤, amyloid-␤ protein; AD, Alzheimer's disease; PS, presenilin; NTF, N-terminal fragment; CTF, C-terminal fragment; SPP, signal peptide peptidase; DAPT, N-[N- (3,5- on SigmaPlot and fitting it to a sigmoidal function. Preparation of Cell Lysates and Photoaffinity Labeling-HeLa cells were lysed in buffer containing 50 mM MES, pH 6.0, 150 mM NaCl, 5 mM MgCl 2 , 5 mM CaCl 2 , 1% CHAPSO, and protease inhibitors. The cell homogenate was centrifuged at 20,000 ϫ g, and the supernatant was spun at 100,000 ϫ g for 1 h. The final supernatant was diluted with PIPES buffer (50 mM PIPES, pH 7.0, 150 mM NaCl, 5 mM MgCl 2 , 5 mM CaCl 2 ) to a final 0.25% CHAPSO solution. The photolabeling was performed essentially as described previously (4), only the labeled proteins were released from streptavidin beads with 8 M guanidinium hydrochloride, pH 2.3. The solvent was then exchanged to phosphate-buffered saline by using Microcone centrifugation tubes. The labeled proteins were detected by SDS-PAGE/immunoblotting using antibody AB14 to the PS1 N terminus and antibody 4627 to the PS1 C terminus and with anti-biotin antibody to the biotinylated species. The concentration of the competitor was 25 times the cell-free IC 50 for active compounds or at a maximum of 200 M for compounds inactive in the cell-free assay.

RESULTS
To fully characterize the inhibitors used in this study and to provide consistency in the data analysis, we evaluated the inhibitory potency of each compound in our cell-based and cell-free assays. The ability of inhibitors to decrease A␤ production in cell culture was tested in Chinese hamster ovary cells stably transfected with human APP. The IC 50 values are reported below for each compound, and all values are summarized in Table I. In cell-free assays CHAPSO-solubilized HeLa cell membrane preparations were used to evaluate the ability of inhibitors to reduce ␥-secretase proteolysis of recombinant APP-based substrate (Fig. 1).
Photoactivatable Transition State Analog Inhibitor III-63 Covalently Labels PS1-We have developed and optimized the labeling of PS1 with a derivative of a highly potent aspartyl protease transition state analog inhibitor of ␥-secretase, called III-31-C, a (hydroxyethyl)urea peptidomimetic (8). III-31-C inhibits A␤ production with an IC 50 of 10 nM in the cell-free ␥-secretase assay and 200 nM in APP-transfected cells. We modified this compound with a photoactivatable benzophenone and with biotin at the C terminus of the molecule (20). The derivatized molecule, called III-63, exhibited an in vitro IC 50 value similar to the parent compound, indicating that the modifications did not affect the ability of the compound to bind the ␥-secretase active site.
Compound III-63 was used for photolabeling of PS1 under in vitro conditions that preserve ␥-secretase activity. We used lysates as well as microsomes isolated from HeLa cells by lysing and/or solubilizing with the detergent CHAPSO (21). The labeled species were precipitated with streptavidin beads and analyzed by Western blot, detecting with anti-PS1 antibodies. The observed biotinylated proteins, a major band at about 21 kDa and a minor band at about 31 kDa, were identified as PS1 CTF and PS1 NTF, respectively (Fig. 2, lane 1). These results strongly suggest that the III-63 photolabel directly binds to the interface of PS1 heterodimers. This observation is essentially the same as seen by Li et al. (4), who used a nearly identical photoprobe and labeled the CTF of PS1 exclusively. Apparently, our closely related compound can also label small amounts of NTF. When the unbiotinylated parent compound III-31-C (250 nM) was used as a competitor, no labeled PS1 NTF and PS1 CTF species were detected (Fig. 2, lane 2). No labeled proteins were detected when irradiation was not applied (data not shown). These observations demonstrated specific labeling of PS1 heterodimers by III-63. This direct photoprobe binding provided a means of assessing the mechanism of structurally diverse ␥-secretase inhibitors by testing the ability of these inhibitors to displace the photoprobe from its molecular target. All subsequent competition experiments were performed in cell lysates, because this method requires smaller amounts of initial cellular material, but yields similar results. We used competitor concentrations equal to 25 times their IC 50 value in vitro. The latter was chosen to compare the degree of displacement with the positive control of displacement by III-31-C at such a concentration (250 nM, 25 times above the 10 nM IC 50 value).
Effects of DAPT and Compound E-Several reported inhibitors were identified from small molecule library screening and subsequent optimization, but the means by which they inhibit the protease are unclear. For example, the dipeptide DAPT (Table I) is a non-transition state analog, but nevertheless a very potent inhibitor, that substantially decreased A␤ levels in the brains of APP transgenic mice (16). In our hands, this compound inhibited A␤ production with an IC 50 of 10 nM in the cell-free ␥-secretase assay and with an IC 50 of 20 nM in APPtransfected cells. After addition of DAPT as a competitor to photoprobe III-63, somewhat less biotinylated species were observed at a competitor concentration of 250 nM (25 times the cell-free IC 50 ), and much less labeled species were detected TABLE I Chemical structures, inhibition properties, and the displacement ability of the ␥-secretase inhibitors used in the study In cells IC 50 values for inhibition of A␤ production were determined using Chinese hamster ovary cells stably expressing human APP 751 . Cell-free IC 50 values were measured using solubilized ␥-secretase prepared from HeLa cells and a recombinant APP-based (C100FLAG) substrate (see Fig. 1). Displacement indicates the ability of the inhibitor to prevent the transition state-based photolabel from binding to PS1 heterodimer (see Fig. 3).  (Fig. 3A). This observation that DAPT prevents labeling with a transition state analog affinity reagent suggests that DAPT either directly binds to the active site between PS1 heterodimers or alters it through an allosteric interaction (20). Interestingly, DAPT does not displace the photoprobe as well as III-31-C does at the same relative concentration (25 times the IC 50 ), despite being essentially equivalent inhibitors in the solubilized ␥-secretase assay. This intriguing observation suggests that transition state analog III-31-C and non-transition state analog DAPT act at partially, but not completely, overlapping sites. Similar results were obtained with Compound E, a non-transition state analog inhibitor containing a benzodiazepine moiety. In our assays, this compound inhibited A␤ production with an IC 50 of 3 nM in microsomes and with an IC 50 of 0.3 nM in APP-transfected cells. After addition of Compound E as a competitor to photoprobe, the level of labeled species was reduced somewhat (Fig. 3A) at a competitor concentration of 75 nM (25 times the IC 50 ), and almost no labeled species were detected when 600 nM (200 times the IC 50 ) Compound E was used.
Isocoumarin-based Compounds-Isocoumarin "JLK" compounds (Table I) reported by Petit et al. (15) were also examined in the displacement assay. These compounds were reported to inhibit A␤ production in cells without affecting proteolysis of the Notch receptor. The Notch signaling pathway is involved in cell fate decisions, and presenilins are linked to proteolysis of the transmembrane region of the Notch receptor as a key step in the signaling mechanism (22). As we reported previously (23), we confirmed that JLK compounds lower A␤ production in cell culture with an IC 50 of 80 M; however, the isocoumarins failed to block ␥-secretase activity in our cell-free assays. Moreover, we observed no change in the degree of PS1 heterodimers labeling by the photoprobe (Fig. 3B) when the most active analogs, JLK2 and JLK6, were used as the competitors at concentrations as high as 0.5 mM. This observation indicates that JLKs fail to block the interaction of III-63 with its molecular target, suggesting that these compounds are not likely interacting with or otherwise affecting the active site of ␥-secretase. Taken together, our evidence implies that the isocoumarins probably work upstream of ␥-secretase.
Aib-containing Peptides-We also examined APP-based peptides designed to assume a helical conformation and mimic the APP transmembrane domain upon initial interaction with ␥-secretase. Thus, these compounds were designed to inhibit the protease by a different mechanism than transition-state analogs do. Specifically, these peptides are based on the sequence of the APP transmembrane domain, modified with the helix promoting residue Aib. One of the most potent helical peptides, D-294 (Table I), displays IC 50 values of 3 M in cells and 100 nM in the in vitro assay (Fig. 1). When this peptide was run as a competitor to the photoprobe at concentrations as high as 6 M (60 times IC 50 ), it did not affect the photoprobe binding to PS1 heterodimers (Fig. 3B). When D-294 was used at 100 M (1000 times IC 50 ), little or no change in the degree of labeling was observed (data not shown). Therefore, despite the fact that this helical peptide directly inhibits ␥-secretase activity, it does not interact with or otherwise affect the active site. These observations are consistent with the helical peptide inhibitor competing with the substrate for a separate initial docking site (8).
Epoxide Peptidomimetic-Another class of ␥-secretase inhibitor with a distinct structure are epoxide-containing molecules. A number of small epoxide molecules have been identified as irreversible inhibitors of aspartyl proteases that act by alkylating the catalytic aspartates (24). One such molecule (18) was reported by Golde et al. to be an inhibitor of ␥-secretase, an unconventional aspartyl protease. We synthesized a similar epoxide (Table I) and analyzed its inhibitory properties toward ␥-secretase. This epoxide inhibited ␥-secretase activity in the cell-based assay (IC 50 of 20 ⌴) and in the cell-free assay (IC 50 of 20 M) (Fig. 1). When tested in the displacement assay under standard conditions, it did not prevent the photoprobe from binding to the active site (Fig. 3C). However, when the epoxide was pre-incubated with lysate for 2 h before the addition of photoprobe and irradiation, much less photolabeling was observed. Such time-dependent displacement was not observed with compound III-31-C. This result suggests that the epoxide is a mechanism-based inhibitor, inactivating ␥-secretase by irreversible binding to the active site.
SPP Inhibitor-Finally, we tested (Z-LL) 2 -ketone (Table I), an aspartyl protease transition state analog inhibitor of SPP, which is a recently discovered presenilin-like aspartyl protease (13). We found this compound to be only moderately active in the ␥-secretase cell-free assay, with an IC 50 value of 30 M (Fig.  1). The (Z-LL) 2 -ketone similarly inhibited the formation of the other (FLAG-tagged) proteolytic product generated in this assay (data not shown). When tested in the displacement assay, 200 M (Z-LL) 2 -ketone (only seven times IC 50 ) completely pre- Distinct Mechanisms of ␥-Secretase Inhibitors 16472 vented the photoprobe from binding (Fig. 3D), suggesting that it successfully competes for the active site. Thus, this SPP inhibitor directly affects ␥-secretase activity by binding to or allosterically altering the catalytic site. This finding indicates that the topographies of the SPP and presenilin active sites share similarities. DISCUSSION The use of a photoactivatable probe directed to the active site of ␥-secretase provides a convenient means to probe the inhibitory mechanism of action of structurally diverse molecules. Such studies are particularly important in this case, because ␥-secretase is a multicomponent membrane protease that has eluded purification and is unlikely to be crystallized in the near future. Certain small organic inhibitors (e.g. transition state analogs) of this enzyme have served as important mechanistic probes, and the more recently identified agents will likely do so as well. By analyzing the ability of structurally different ␥-secretase inhibitors to displace the active site-directed photoprobe from its target, we show here that not all inhibitors act by means of binding or altering the protease active site. Indeed, different ␥-secretase inhibitors fell into different mechanistic classes, summarized in Table I. Most ␥-secretase inhibitors interfered with the ability of the photoprobe to bind PS1, indicating that these compounds either bind directly to the active site or alter the active site through an allosteric interaction. Importantly, some non-transition state analogs, namely DAPT and Compound E, are very potent in vitro and successfully compete for the active site in the displacement assay. However, these non-transition state analogs do not displace the photoprobe as well as the transition state analog III-31-C does at a comparable concentration. Such behavior suggests that these non-transition state analogs bind at a site overlapping with that of transition state mimics.
Some reported ␥-secretase inhibitors, the isocoumarins and the Aib-containing helical peptides, failed to displace the photoprobe from the active site, demonstrating that these compounds do not block A␤ production by affecting the active site of the protease. However, these two types of compounds apparently do not share the same mechanism. Unlike the Aib-containing helical peptide D-294, the isocoumarins are not active in the solubilized enzyme assay, indicating that these inhibitors probably do not target ␥-secretase directly. In contrast, helical peptide D-294 does inhibit A␤ production in the cell-free assay but does not compete with the photoaffinity probe, suggesting that D-294 acts directly on ␥-secretase, but not at the active site. Overall, these findings show that there are apparently several distinct mechanisms for inhibiting ␥-secretase.
These studies do not distinguish between compounds that bind to the active site and those that allosterically alter it. To gain insight into interaction specifics, each inhibitor class should be modified with a photoactive or chemically reactive moiety to identify its direct target(s). Such studies should also provide additional characterization of the initial substrate binding site of ␥-secretase. Recently, Greenberg and co-workers (25) reported that a number of ␥-secretase inhibitors of diverse structures show non-competitive inhibitory kinetics. This finding suggested the existence of another substrate binding site (i.e. for initial docking) besides the active site, which is consistent with other evidence from our laboratory (8). It is possible that the initial binding site may be a target of the helical peptide inhibitors (e.g. D-294), which do not affect the active site. Inhibitors that target the initial substrate docking site may be selective for different ␥-secretase substrates, if different substrates do not share the same initial binding site. These future mechanistic studies may uncover the possibility of finding ␥-secretase inhibitors selective for APP over other substrates such as Notch.
We also found that an SPP inhibitor displays pharmacological crossover, blocking ␥-secretase activity as well. Sequence homology between PS and an entire family of so-called "PS homologs," which includes SPP, is found primarily at the transmembrane motifs YD and LGLGD that contain the critical and conserved aspartates (14). (Z-LL) 2 -ketone is a transition state analog that directly interacts with SPP (13). By showing that this compound can inhibit ␥-secretase and compete for the active site on PS, we provide evidence that ␥-secretase and SPP have similar active sites and likely share the same proteolytic mechanism. Other than the short, conserved aspartate-containing motifs, SPP and PS share very little homology, suggesting that the two proteins arrived at their aspartyl protease mechanisms via independent evolutionary paths.
Finally, analyzing an epoxide inhibitor in our displacement assay revealed that this molecule affects the active site of ␥-secretase in a time-dependent manner. The epoxide may inactivate ␥-secretase by irreversible binding to the active site aspartates due to its chemical properties. Further work will focus on studying this interaction in detail, as epoxides might be good chemical probes for the active site of ␥-secretase or other unknown intramembrane aspartate proteases awaiting a discovery.