The Amyloid-β Rise and γ-Secretase Inhibitor Potency Depend on the Level of Substrate Expression*

The amyloid-β (Aβ) peptide, which likely plays a key role in Alzheimer disease, is derived from the amyloid-β precursor protein (APP) through consecutive proteolytic cleavages by β-site APP-cleaving enzyme and γ-secretase. Unexpectedly γ-secretase inhibitors can increase the secretion of Aβ peptides under some circumstances. This “Aβ rise” phenomenon, the same inhibitor causing an increase in Aβ at low concentrations but inhibition at higher concentrations, has been widely observed. Here we show that the Aβ rise depends on the β-secretase-derived C-terminal fragment of APP (βCTF) or C99 levels with low levels causing rises. In contrast, the N-terminally truncated form of Aβ, known as “p3,” formed by α-secretase cleavage, did not exhibit a rise. In addition to the Aβ rise, low βCTF or C99 expression decreased γ-secretase inhibitor potency. This “potency shift” may be explained by the relatively high enzyme to substrate ratio under conditions of low substrate because increased concentrations of inhibitor would be necessary to affect substrate turnover. Consistent with this hypothesis, γ-secretase inhibitor radioligand occupancy studies showed that a high level of occupancy was correlated with inhibition of Aβ under conditions of low substrate expression. The Aβ rise was also observed in rat brain after dosing with the γ-secretase inhibitor BMS-299897. The Aβ rise and potency shift are therefore relevant factors in the development of γ-secretase inhibitors and can be evaluated using appropriate choices of animal and cell culture models. Hypothetical mechanisms for the Aβ rise, including the “incomplete processing” and endocytic models, are discussed.

The amyloid-␤ (A␤) peptide, which likely plays a key role in Alzheimer disease, is derived from the amyloid-␤ precursor protein (APP) through consecutive proteolytic cleavages by ␤-site APP-cleaving enzyme and ␥-secretase. Unexpectedly ␥-secretase inhibitors can increase the secretion of A␤ peptides under some circumstances. This "A␤ rise" phenomenon, the same inhibitor causing an increase in A␤ at low concentrations but inhibition at higher concentrations, has been widely observed. Here we show that the A␤ rise depends on the ␤-secretase-derived C-terminal fragment of APP (␤CTF) or C99 levels with low levels causing rises. In contrast, the N-terminally truncated form of A␤, known as "p3," formed by ␣-secretase cleavage, did not exhibit a rise. In addition to the A␤ rise, low ␤CTF or C99 expression decreased ␥-secretase inhibitor potency. This "potency shift" may be explained by the relatively high enzyme to substrate ratio under conditions of low substrate because increased concentrations of inhibitor would be necessary to affect substrate turnover. Consistent with this hypothesis, ␥-secretase inhibitor radioligand occupancy studies showed that a high level of occupancy was correlated with inhibition of A␤ under conditions of low substrate expression. The A␤ rise was also observed in rat brain after dosing with the ␥-secretase inhibitor BMS-299897. The A␤ rise and potency shift are therefore relevant factors in the development of ␥-secretase inhibitors and can be evaluated using appropriate choices of animal and cell culture models. Hypothetical mechanisms for the A␤ rise, including the "incomplete processing" and endocytic models, are discussed.
Evidence suggests that the amyloid-␤ (A␤) 9 peptide plays a key role in Alzheimer disease. A␤ is generated by proteolytic processing of the amyloid-␤ precursor protein (APP) through consecutive cleavages by the ␤-site APP-cleaving enzyme (BACE) and ␥-secretase. APP is cleaved by BACE to form a ␤-secretase-derived C-terminal fragment of APP (␤CTF), which undergoes further cleavage by ␥-secretase to form A␤. In addition, APP is cleaved by ␣-secretase to form ␣CTF, which undergoes ␥-secretase cleavage to produce an N-terminally truncated form of A␤ known as "p3." Using the conventional amino acid numbering of A␤, BACE cleavage leads to A␤ peptides with N-terminal ends at positions 1 and 11, whereas ␣-secretase leads to p3 peptides with an N-terminal end at position 17. Cleavage of ␤CTF and ␣CTF by ␥-secretase produces a mixture of different C termini in the resulting A␤ and p3 peptides. For example, the predominant ␥-secretase cleavage of ␤CTFs at position 40 produces A␤-  and A␤- , whereas other ␥-secretase cleavage sites produce a range of less abundant A␤ peptides, such as the disease-associated A␤-(1-42) (1, 2).
Although ␥-secretase cleavage can be fully inhibited in cellbased assays, some inhibitors cause an increase in the amount of A␤ at subinhibitory concentrations. This "A␤ rise" phenomenon, the same inhibitor causing an increase in A␤ at low concentrations but inhibition at higher concentrations, has been observed frequently (3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15). For peptide aldehyde inhibitors, some studies reported a rise that was specific for A␤42 (3)(4)(5)(6), whereas other studies reported a rise also for A␤40 in addition to A␤42 (7)(8)(9)(10). However, these studies also differed as to the pharmacological target proposed to mediate the effects on A␤; some authors considered only the protease calpain via an indirect effect on ␥-secretase (7,9,10), whereas others proposed a direct effect on ␥-secretase (3)(4)(5)(6)8). In one study, the A␤ rise was reported in isolated membrane preparations, suggesting a direct effect of peptide aldehydes on ␥-secretase (6). Further evidence that ␥-secretase can mediate the A␤ rise comes from studies with difluoroketone-based inhibitors, which are selective for ␥-secretase and which cause a robust rise in A␤42 both in cell culture (11)(12)(13) and in isolated membrane-based assays (6). Furthermore a rise in total A␤, as well as A␤42, in response to highly selective ␥-secretase inhibitors has been observed in vivo in the plasma of guinea pigs (14) and in humans (15).
Thus, the biochemical mechanism of the A␤ rise has not been elucidated, and the experimental conditions required for this phenomenon have not been defined. Here we show that a rise in multiple A␤ species can be readily observed in cell cultures treated with ␥-secretase inhibitors and that the key experimental requirement is a low level of ␤CTF or C99 expression. In addition, low substrate expression caused a shift in inhibitor potency that was independent of the A␤ rise. We also show that increased A␤ can occur in the brain following ␥-secretase inhibitor dosing in rats, demonstrating the potential of ␥-secretase inhibitors to cause the opposite of the intended effect in the target organ. Thus, the A␤ rise is a relevant issue in the development of ␥-secretase inhibitors for A␤-lowering therapy, and experimental conditions that exhibit the A␤ rise can be readily applied in cell culture models.
Cell Culture, DNA Constructs, and Transfection-HEK293 cells stably transfected with APP wild type (HEKwt) or Swedish variant (HEKsw) were derived essentially as described previously (26,27). Growth media and supplements were obtained from Invitrogen. For HEKwt cells, growth medium was Dulbecco's modified Eagle's medium (DMEM) containing 10% heat inactivated fetal bovine serum, 2 mM L-glutamine, 1 mM sodium pyruvate, 400 g/ml G418, and 5 g/ml blasticidin. Growth medium for HEKsw cells was the same as for HEKwt cells but containing 200 g/ml hygromycin B and lacking blasticidin. THP-1 cells were grown in roller bottles in RPMI 1640 medium containing L-glutamine and 10 M ␤-mercaptoethanol to a density of 1 ϫ 10 6 /ml. Cells were harvested by centrifugation, and cell pellets were quick frozen in dry ice/ethanol and stored at Ϫ80°C prior to use. Mouse embryonic fibroblasts were passaged twice per week in a 1:1 mixture of DMEM and F-12 nutrient mixture supplemented with 10% fetal bovine serum, penicillin, and streptomycin. HeLa cells were maintained in DMEM containing 10% fetal bovine serum, penicillin, streptomycin, and 2 mM L-glutamine. For inhibitor treatments, cell cultures were grown for 24 h, and the medium was replaced with DMEM containing high glucose, 0.0125% bovine serum albumin, nonessential amino acids, 2 mM L-glutamine, and 100 units/ml penicillin and streptomycin. Inhibitors were added in DMSO to a final concentration of 0.1% DMSO. Treated cells were incubated overnight at 37°C in a humidified atmosphere containing 5% CO 2 . Inhibitor treatments were carried out in 96-well format for Figs. 1, 3, and 4, in cell culture flasks for Fig. 2, and in 24-well format for Fig. 6. For transient expression of C99, the ␤CTF sequence (99 C-terminal residues of APP) was preceded by a signal peptide containing the amino acid sequence MLP-GLALLLLAAYTARADA and followed by the c-Myc epitope tag, EQKLISEEDL, at the C-terminal end. The DNA construct encoding this sequence was cloned into the expression vector pCDNA3.1 (Invitrogen). HeLa cell cultures were transiently transfected in T 162-cm 2 flasks with the C99 expression construct using TransIT-HelaMONSTER (Mirus) according to the manufacturer's directions. A total of 36 g of DNA was used in each transfection consisting of a mixture of vector DNA and either 0.9, 1.8, or 36 g of the C99 expression construct DNA. After incubation overnight, transfected cells were detached using trypsin/EDTA for 2 min at ambient temperature, collected by centrifugation, and resuspended in growth medium. Cells were again collected by centrifugation; resuspended in defined medium consisting of DMEM, 2 mM L-glutamine, penicillin, streptomycin, 0.0125% bovine serum albumen, and nonessential amino acids; and then replated in 96-well plates at 200 l/well (Packard View Plate, 96-well, black).
Cell-free ␥-Secretase Assay-Assays were carried out based on a procedure described previously using C99 substrate expressed and purified from Escherichia coli (28). To prepare concentrated ␥-secretase enzyme, lipid rafts were isolated based on methods described previously (29). Briefly THP-1 cell pellets were lysed in 4 volumes of lysis buffer (125 mM NaCl, 1% CHAPSO, 25 mM Na-MES, pH 6.5) containing a protease inhibitor mixture of 104 M 4-(2-aminoethyl)benzenesulfonyl fluoride, 80 nM aprotinin, 2 M leupeptin, 4 M bestatin, 1.5 M pepstatin A, and 1.4 M E-64 (0.1% protease inhibitor mixture P8340, Sigma-Aldrich) using five passages through a 25-gauge needle, and the lipid raft fraction was isolated by discontinuous sucrose density gradient centrifugation (29). This procedure yielded a stock preparation of CHAPSO-solubilized proteins containing ␥-secretase at a concentration of 3 nM as determined by saturation radioligand binding. ␥-Secretase activity assays were performed in assay buffer (100 nM NaCl, 0.25% CHAPSO, 50 mM HEPES, pH 7.0). The lipid raft preparation was mixed with C99 substrate at molar ratios of 1:1 and 1:100, corresponding to absolute concentrations of 0.3 nM:0.3 nM and 0.01 nM:1 nM, respectively, at a volume of 200 l/reaction in 96-well polypropylene plates. After incubation for 3 h at 37°C, A␤-(1-40) was quantified by enzyme-linked immunosorbent assay (ELISA). Extensive experimentation to optimize assay conditions showed that the maximum concentration of ␥-secretase that could be used in the assay was 0.3 nM because of inhibition of A␤ production at higher concentrations presumably due to inhibitory activities present in the raft preparation. Likewise C99 substrate at concentrations less than 0.03 nM yielded insufficient signal for reliable quantification of A␤- . Thus, the highest enzyme to substrate ratio that could be utilized using this method was 1:1. A␤-(1-40) was quantified by ELISA using the antibodies TSD9S3.2 and 26D6 described below.
Radioligand Binding in Cell Homogenates-HEKsw, HEKwt, and mouse embryonic fibroblast cell pellets were homogenized in 10 ml of 50 mM HEPES with 0.1% protease inhibitor mixture (Sigma-Aldrich P8340) at pH 7.0 and 4°C using a Dounce homogenizer. The homogenate was centrifuged at 48,000 ϫ g for 20 min. Protein determinations were carried out using a Bradford based assay (Bio-Rad). The final pellet was resuspended in buffer to yield a protein concentration of 5 mg/ml. Binding assays were performed in polypropylene 96-deepwell plates (Beckman Coulter, Fullerton, CA) in a final volume of 0.25 ml containing 5% (v/v) DMSO. Assays were initiated by the addition of 25 l of assay buffer containing radioligand to 12.5 l of dimethyl sulfoxide containing various concentrations of unlabeled compounds followed by 212 l of cell homogenate. Nonspecific binding was defined in the presence of 1 M BMS-433796. After incubating at 25°C for 1.5 h, bound radioligand was separated from free by filtration over GF/B glass fiber filters (Brandel, Gaithersburg, MD) presoaked in phosphate-buffered saline (PBS), pH 7.0, using a cell harvester (Brandel). Filters were washed four times with 1.0 ml of ice-cold PBS, pH 7.0, dried, and then assessed for radioactivity by liquid scintillation counting using a Wallac Microbeta Trilux (PerkinElmer Life Sciences). Equilibrium saturation data were analyzed using the Kell software package (Biosoft, Cambridge, UK).
␥-Secretase Inhibitor Binding Site Occupancy in Intact Cells-HEKwt and HEKsw cells were seeded at a density of 125,000 cells in 500 l of growth medium/well in Biocoat 24-well plates treated with poly-D-lysine and incubated for 24 h. After overnight incubation in assay medium in the presence or absence of compound, as described above, 275 l was removed for assay of A␤. To the remaining 225 l, 25 l of PBS, pH 7.0, containing [ 3 H]BMS-570479 at 7.5 nM was added. Nonspecific binding was determined in the presence of BMS-433796 at a concentration of 1 M. After incubation for 1 h at 37°C in 5% CO 2 , assay medium was removed, and the cells were rinsed gently three times in 1 ml of ice-cold PBS, pH 7.0. Cells were then incubated at room temperature for 30 min with rotary mixing in 200 l of 0.5 N sodium hydroxide. The sodium hydroxide was then mixed into 3 ml of scintillation fluid and counted using a Microbeta scintillation counter (Wallac). The percentage of displacement of radioligand was then calculated, and an IC 50 value was determined by best fit to a four-parameter dose-response curve using GraphPad Prism (GraphPad Software, San Diego, CA). Because of the low concentration of radioligand present in these assays, we would expect a slight underestimate of occupancy for compounds with a competitive mechanism of radioligand displacement.
Ex Vivo Inhibitor Occupancy-Inhibitor occupancy of ␥-secretase was quantified as described previously (30) except that a different radioligand, [ 3 H]BMS-570479, was used. Briefly 20-m coronal brain sections were incubated for 10 min in 50 mM HEPES buffer, pH 7.4, containing 5 nM [ 3 H]BMS-570479. Nonspecific binding was determined by incubating adjacent sections in the presence of 500 nM BMS-433796. The sections were washed in PBS and dried, and bound radioligand was quantified by phosphorimage analysis.
Handling of Rats-Female Sprague-Dawley rats were obtained from Charles River Laboratories (Wilmington, MA) and orally dosed at 30 mg/kg, 4 ml/kg in polyethylene glycol 400, 1% Tween 80. Half of the forebrain was quick frozen on dry ice. All experimental procedures with rodents were authorized by and in compliance with the Bristol-Myers Squibb Animal Care and Use Committee.
Quantification of A␤ Peptides by ELISA-For Figs. 1, A and B, 3, and 5, ELISAs using an A␤ position 40 or position 42 neoepitope-specific antibody and a monoclonal antibody directed to A␤ residues 10 -20 were used (16). For the remaining figures, A␤-(1-40) was assayed using the monoclonal antibody TSD9S3.2, which is specific for the free C terminus of A␤- , in combination with a peroxidase conjugate of 26D6, which is specific for an epitope within the 12 N-terminal amino acids of human A␤-(1-40) (19), or with a peroxidase conjugate of 252Q6 (BIOSOURCE/Invitrogen), which is specific for an epitope within the 12 N-terminal amino acids of rodent A␤- . Human A␤-(1-x) was assayed using monoclonal antibody 26D6 in combination with a biotin conjugate of 4G8 (Signet/ Covance, Berkeley, CA) and streptavidin-peroxidase conjugate (Zymed Laboratories Inc./Invitrogen). This antibody combination detects A␤ peptides containing the 1-24 amino acid region. Data from inhibitor-treated cell cultures were evaluated by non-linear regression using a four-parameter sigmoidal dose-response curve, and the derived IC 50 values represent the concentration of compound required to inhibit A␤ to 50% of vehicle-treated control.
Quantification of A␤ Peptides by Liquid Chromatography Mass Spectroscopy (LC/MS)-A␤ peptides were concentrated from the cell medium by immunoprecipitation. 14 ml of cell medium was mixed for 1 h at 4°C with 15 g of 4G8 (Covance), 30 g of 26D6 (Bristol-Myers Squibb), protease inhibitor mixture (Hoffmann-La Roche Ltd.), and 3.5 ng of synthetic [ 15 N]A␤-(1-40) peptide (rPeptide, Athens, GA) followed by the addition of 80 l of protein G-agarose beads (Pierce) and continued incubation with mixing overnight at 4°C. Samples were then centrifuged at 1000 ϫ g for 3 min. The beads were washed three times by centrifugation with 1 ml of PBS and washed a final time with 1 ml of 10 mM Tris-HCl, pH 8.0. The A␤ peptides were eluted from the beads with 30 l of 70% acetonitrile, 0.1% formic acid. After a final 10,000 ϫ g centrifugation for 5 min, the supernatant was removed and stored on dry ice before LC/MS. The LC/MS system was comprised of a Leap Technologies (Carrboro, NC) CTC HTS PAL autosampler, an Agilent Technologies (Wilmington, DE) 1100 Capillary LC pump, and a Thermo Fisher (San Jose, CA) linear ion trap LTQ mass spectrometer. Sample injection carryover concerns were addressed with an injection port cleaning procedure based on the Clean LC TM macro as part of the Leap Technologies autosampler operating software, Cycle Composer. The injection loop is taken off line during an LC run, and the loop, the injector needle, and the injector syringe barrel are then washed consecutively with 10 mM EDTA in 2% acetonitrile and 10 mM ammonium hydroxide in acetonitrile. For HPLC/MS analysis, 3 l of sample was injected onto an Agilent Technologies Zorbax Extend C 18 reversed-phase chromatography column (50 ϫ 1 mm, 3.5 m) with a Zorbax Extend guard column (1 ϫ 1.7 mm, 3.5 m). Mobile phase A was 100% water (HPLC grade, J. T. Baker) containing 20 mM ammonium hydroxide (99.99% purity, Sigma-Aldrich), and mobile phase B was 80:20 acetonitrile (HPLC grade, J. T. Baker)/water (v/v) also containing 20 mM ammonium hydroxide. The HPLC was performed at a flow rate of 50 l/min with the following gradient: 10 -40% B at 10 min, 70% B at 11 min, and a hold at 70% B for 1 min. The mass spectrometer was operated in positive ion full-scan mode with the following ionization source conditions: spray voltage, 5 kV; capillary voltage, 30 V; tube lens voltage, 150 V; capillary temperature, 300°C; sheath gas flow, 40 arbs; auxiliary gas flow, 5 arbs; and sweep gas flow, 2 arbs. Quantitative data were acquired in centroid mode with three microscans and 500-ms ion time across the 600 -2000 m/z range. For relative quantitation and to monitor injection reproducibility, isotopically labeled [ 15 N]A␤-(1-40) (rPeptide) was added to the samples prior to analysis as an internal standard. The internal standard (1 ng/l) was prepared in 20 mM ammonium hydroxide and stored at Ϫ20°C. The internal standard was used to dilute analytical samples before analysis. There was 1 ng of [ 15 N]A␤-(1-40) absolute on the column per analysis. Polypropylene limited volume inserts (National Scientific, Rockwood, TN) were used to prevent losses occurring from nonspecific interactions of the A␤ peptides with the sample vials. Data were processed using the Thermo Fisher mass spectrometry data acquisition analysis suite, Xcalibur. For quantitative purposes, m/z ratios corresponding to the most abundant signal in the multiply charged isotopic envelope were extracted from the chromatographic data. The summed ϩ3 and ϩ4 charge state ion current was extracted for each of the peptides A␤-  . The ϩ2 and ϩ3 charge state was used for A␤- . Areas under the extracted peaks were integrated, and their ratios with the integrated internal standard peak area were calculated.
Quantification of ␤CTF-␤CTF levels in cell cultures were determined by quantitative Western blotting in comparison with C100 (C99 with an additional N-terminal methionine) expressed and purified from E. coli (28). HEKsw and HEKwt cells were seeded at 900,000 cells/well in 6-well dishes and incubated with increasing concentrations of test compound as described above. Medium was aspirated; and cells were washed with PBS; lysed with buffer containing 0.3% SDS, 0.1 mM sodium orthovanadate, 1 mM sodium fluoride, 10 mM Tris (pH7.5), 1:200 dilution of protease mixture (catalog number 5931, Calbiochem); and subjected to SDS-polyacrylamide (12% bis-tris) gel electrophoresis. Gel loading was normalized for total protein as determined using the detergent compatible protein analysis kit (Bio-Rad), and 10, 15, 20, 25, or 30 g of protein was loaded per lane. A standard curve of C100 peptide was generated on each gel. Gels were blotted onto nitrocellulose membranes and probed with rabbit polyclonal anti-␤APP (CT695, Zymed Laboratories Inc.) followed by the anti-rabbit Vectastain ABC Peroxidase kit (Vector Laboratories). Bound antibody was detected using a Western Lightening Chemiluminescence Detection kit (NEL 102, PerkinElmer Life Sciences). Membranes were exposed to Eastman Kodak Co. X-Omat Blue autoradiography film and developed. The films were scanned, and the bands of interest were quantified with a Kodak Image Station using ImagePro software. To determine the quantity of C100 in the calibration standard, the C100 was digested to completion using trypsin, generating the surrogate peptide A␤-(17-28) (31), which was quantified using a calibration curve of synthetic A␤- (17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28) peptide. An internal synthetic standard consisting of 0.5 M [ 15 N]A␤-(1-40) was added to the C100 samples and to serial dilutions of A␤-(17-28) (American Peptide Co., Sunnyvale, CA) in water, acetonitrile, 0.1% ammonium hydroxide. Low protein binding tubes (Protein LoBind Tube, Eppendorf AG, Hamburg, Germany) were utilized to prevent loss of analyte. The samples were concentrated to dryness in a nitrogen stream at 30°C. To each tube 90 l of 20 mM ammonium bicarbonate and 5 g of trypsin (Trypsin Gold, MS grade, Promega, Madison, WI) were added. The samples were incubated at 37°C, and after 3 h an additional 5 g of trypsin was added following a 14-h incubation at 37°C. The protein/ peptide digestion was quenched with 1 l of formic acid, and samples were analyzed by LC/MS. A Leap Technologies CTC HTS PAL autosampler, an Agilent Technologies 1100 Capillary LC pump, and a Thermo Fisher linear ion trap LTQ mass spectrometer was used as the LC/MS system. For HPLC/MS analysis, 10 l of sample was injected onto a Phenomenex (Torrance, CA) Jupiter Proteo column (50 ϫ 1 mm, 4 m, 90 Å). The mobile phases consisted of water/acetonitrile (98:2, v/v, mobile phase A) and acetonitrile/water (98:2, v/v, mobile phase B). The water in each of the solvent mixtures contained 0.1% formic acid. The HPLC was performed at a flow rate of 75 l/min and with the following gradient: a hold at 95% A for 2 min, linear gradient to 30% B over 8 min, linear gradient to 95% B over 2 min, hold at 95% B for 3 min, step gradient to 95% A, and hold at 95% A for 6 min. The mass spectrometer was operated in positive ion full-scan mode with the following ionization source

RESULTS
The A␤ Rise Occurs in Cell Lines That Express Low Levels of A␤-During initial studies with ␥-secretase inhibitors, we found that subinhibitory concentrations of inhibitors caused a rise in A␤ in cell cultures expressing low levels of A␤, such as the neuroblastoma cell line IMR-32 and cultured rat brain slices (not shown). To evaluate this effect in more detail, ␥-secretase inhibitors were compared in HEKwt or HEKsw. The Swedish mutation of APP enhances ␤-secretase cleavage at amino acid position 1, thereby increasing production of the ␤CTF intermediate and A␤-(1-x) where "x" denotes the range of different C-terminal end positions resulting from ␥-secretase cleavage. Accordingly untreated control cultures of HEKwt typically accumulated 190 pg/ml of (44 pM) A␤-  in the medium, whereas HEKsw accumulated 2,400 pg/ml of (550 pM) A␤-(1-40) after overnight incubation in 96-well culture format. Cell cultures were treated with the ␥-secretase inhibitor DAPT or the benzodiazepinone DPH-111122 at a range of concentrations, and the A␤-(1-40) and A␤-(1-42) secreted into the cell culture media were assayed by ELISA. Rises in both A␤-(1-40) and A␤-(1-42) were detected in the HEKwt cultures, whereas no rises were observed in the HEKsw cultures (Fig. 1, A and B). In addition, inhibitor potency was up to 10-fold greater in the HEKsw than in the HEKwt cell line. IC 50 values are summarized in Table 1. In further experiments, we included an ELISA antibody combination, 26D6/4G8, which detects A␤-(1-x), to evaluate whether the A␤ rise was unique to A␤-(1-42) and A␤- . A robust rise in A␤-(1-x) was observed for DAPT and L-685,458 treatment in HEKwt but not HEKsw cell cultures (Fig. 1, C and D).
The A␤ Rise Occurs for ␤-Secretase-derived, but Not ␣-Secretase-derived, Peptides-To determine which species of A␤ peptides were involved in the A␤ rise, we carried out an analysis of medium from DAPT-treated HEKwt cultures by liquid chromatography/mass spectroscopy. Both HEKwt and HEKsw cells secrete a range of A␤ peptides but in different relative concentrations. For example, in HEKsw medium the A␤-(1-x) peptides predominate, whereas in HEKwt medium the A␤-(17-x) peptides are more abundant (Fig. 2, A and B). When HEKwt cultures were treated with a range of concentrations of DAPT, a robust A␤ rise was exhibited by the BACE-derived A␤ peptides 1-37, 1-38, 1-40, 11-40, and 1-19 but not by the ␣-secretase-derived A␤ peptides 17-37, 17-38, 17-40, and 17-28 (Fig. 2, C-F). For the BACE-derived peptides, IC 50 values were all in the 100 -300 nM range, whereas the ␣-secretasederived peptide IC 50 values were in the 20 -30 nM range consistent with the IC 50 values for DAPT determined by ELISA (Table 1). A␤-(1-42) was detected at low levels limiting its quantitation by this method.
Low Substrate Level, not BACE Cleavage, Is Required for the A␤ Rise-To address both the role of substrate expression level and the potential need for BACE cleavage, experiments were performed by transient expression of ␤CTF, here referred to as "C99" because its formation does not require BACE cleavage. HeLa cell cultures were transfected with different quantities of C99 cDNA and then treated with a range of concentrations of DAPT, BMS-267593, or compound E for 16 h after which the A␤-(1-40) secreted into the culture medium was determined (Fig. 3). All cultures exhibited a rise in A␤-(1-40) at the lowest level of C99 transfection. For BMS-267593 and compound E, the two higher levels of C99 substrate exhibited no A␤ rise, but there was a potency shift despite the absence of an A␤ rise. Thus, the A␤ rise and potency shift were dependent on the low level of C99 or ␤CTF expression but independent of BACE and the Swedish mutation.
The A␤ Rise Is Not Readily Observed in Solubilized Cell-free Assays-Two observations suggested that it might be possible to recreate the A␤ rise in a ␥-secretase enzyme assay: first, the  report by Zhang et al. (6) in which an A␤-(1-42) rise occurred in non-solubilized cell membrane preparations, and second, our observation that low C99 substrate levels were required. We therefore carried out cell-free ␥-secretase assays using defined enzyme to substrate (E/S) ratios. Concentration-response curves for A␤-  in the presence of DAPT and L-685,458 were determined at E/S ratios of 1:1 and 1:100. Although inhibition was observed in each case, an A␤ rise did not occur under any of the conditions tested (Fig. 4). Unexpectedly L-685,458 exhibited a 40-fold shift in potency with IC 50 ϭ 0.1 M at an E/S ratio of 1:100 and IC 50 ϭ 4.1 M at an E/S ratio of 1:1. In contrast, there was no significant difference in the IC 50 values for DAPT under the same conditions that were 8.6 and 7.9 M, respectively. A possible explanation for the IC 50 shift for L-685,458 would be the presence of abundant quantities of signal peptide peptidases in the cell-free extracts that bind with high affinity to L-685,458 but with low affinity to DAPT (32). Under the conditions used at the E/S ratio of 1:1, L-685,458 at concentrations in the low nanomolar range would be mostly titrated by signal peptide peptidases (32) and so unavailable to inhibit ␥-secretase. In conclusion, we could not reproduce the A␤ rise under solubilized enzyme assay conditions. Many ␥-Secretase Inhibitors Exhibit the A␤ Rise and Potency Shift-To evaluate the generality of the A␤ rise and potency shift, concentration-response curves for A␤-(1-42) and A␤-(1-40) in both HEKwt and HEKsw cell cultures were determined for 65 different compounds mostly with structures related to DPH-111122 and covering nearly a 1000-fold range in potency. The A␤ rise was observed only in the HEKwt cell line. In addition, the extent of the A␤ rise was compound-dependent and more pronounced for A␤-(1-42) than for A␤-(1-40) (Fig. 5A). The extent of the A␤-(1-42) rise correlated with the extent of the A␤-(1-40) rise for a given compound (Fig. 5B). Compound potency correlated with the A␤ rise, and a significant decrease (p ϭ 0.0003) in the extent of the rise was observed among compounds of higher potency (Fig. 5C). Comparison of the IC 50 values showed that A␤-(1-42) and A␤-(1-40) potencies were strongly correlated (Fig. 5D), but on average there was a 10-fold decrease in potency in the HEKwt relative to the HEKsw cultures (Fig. 5E). Surprisingly the A␤ rise and the extent of the potency shift were not significantly correlated (Fig. 5F), again suggesting that a significant part of the potency shift was independent of the A␤ rise and therefore not solely a secondary consequence of the rise in A␤.
Determination of ␥-Secretase and Cellular ␤CTF Concentrations-One hypothesis for the potency shift is that HEKwt cells contain an excess of ␥-secretase enzyme relative to  substrate such that the amount of enzyme would not be ratelimiting for A␤ production, thus requiring considerable inhibition of enzyme before substrate turnover was affected. To quantify the level of ␥-secretase enzyme in the HEKwt and HEKsw cells, saturation binding isotherms were determined in cell homogenates using the radiolabeled ␥-secretase inhibitor [ 3 H]BMS-570479. Analysis of the binding isotherms indicated single high affinity binding sites in both cultures with identical dissociation constants (K d ) of 0.6 nM and B max values of approximately 400 fmol/mg of protein (Fig. 6A). Based on the number of cells used to make the homogenate, the B max value of 400 fmol/mg is equivalent to 15,000 ␥-secretase binding sites per cell for both HEKsw and HEKwt cells.
To quantify the amount of ␤CTF in HEKwt and HEKsw cells, quantitative Western blotting was carried out using recombinant C100 (equivalent to ␤CTF with an additional N-terminal methionine; purified from E. coli) as a calibration standard. In this method, the intensity of the Western blot was proportional to the amount of C100 loaded (Fig. 6B). The concentration of recombinant C100 used for Western blot calibration was determined by tandem mass spectroscopy of the A␤-(17-28) fragment in tryptic digests of the recombinant C100 calibrated against known amounts of synthetic A␤- (17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28). The linear response of synthetic A␤-(17-28) peptide by mass spectros-copy is shown (Fig. 6C). See "Experimental Procedures" for further details. In HEKwt cells, the concentration of ␤CTF was found to be 30 fmol/mg of homogenate protein, whereas in HEKsw cells there was 140 fmol/mg. Thus, subject to additional factors, ␥-secretase has the potential to be present in molar excess with respect to ␤CTF substrate with the result that the enzyme may not be rate-limiting for A␤ production.
A High Level of Inhibitor Binding Site Occupancy Is Required for ␥-Secretase Inhibition in HEKwt Cells-A functional consequence of non-rate-limiting levels of enzyme would be the requirement for a high level of inhibitor binding site occupancy for inhibition to occur. To quantify inhibitor binding under conditions that produced the A␤ rise and potency shift, [ 3 H]BMS-570479 binding was applied to living cells. HEKsw and HEKwt cultures were treated with the aryl sulfonamide inhibitor BMS-299897 and DAPT at a range of concentrations. Samples of the cell culture supernatants were then taken to measure A␤-(1-40), and [ 3 H]BMS-570479 was added to the remaining cell cultures to determine binding site occupancy. As expected, the HEKwt cultures exhibited a robust rise in A␤ for both compounds, whereas the HEKsw cultures showed little or no A␤ rise. In contrast, binding site occupancy was not significantly different between the HEKwt and HEKsw cultures (Fig.  7). Thus, consistent with a functional excess of enzyme, a high level of occupancy was required for inhibition of A␤ in the HEKwt culture. In addition, the A␤ rise correlated with partial occupancy of inhibitor binding sites for both compounds consistent with a direct role for ␥-secretase in the A␤ rise.
The A␤ Rise Occurs at Partial Binding Site Occupancy in Rat Brain-To examine whether rises in brain A␤ can occur in vivo, rats were dosed with BMS-299897, and brains were harvested between 1 and 12 h after dosing for determination of inhibitor binding site occupancy and A␤. Brains from control animals were harvested 5 h after dosing with vehicle alone. Mean values of A␤ and radioligand-specific binding were then calculated relative to mean vehicle control. A rise in A␤ was observed starting at the 1-h time point and reached a maximum at 7 h (Fig. 8). Because the A␤ rise did not follow a period of A␤ inhibition in this experiment, this rules out the possibility that increased A␤ resulted from processing of accumulated APP ␤CTF intermediates. In contrast, in Tg2576 mice, A␤ lowering correlated with inhibitor occupancy, and there was no A␤ rise (28).

DISCUSSION
Although ␥-secretase inhibitors can fully inhibit A␤ production, they often increase the production of A␤ at subinhibitory concentrations. Thus, a given inhibitor causes an A␤ rise at low concentrations but inhibition at higher concentrations. In this report, we describe a variety of novel observations about the A␤ rise. First, we confirmed that both the allosteric class and the active site-directed class of ␥-secretase inhibitors can cause the A␤ rise, ruling out any allosteric activation mechanism. Second, the A␤ rise did not occur for p3 peptides or in the context of the APP Swedish variant. Third, the A␤ rise occurred only under conditions of low levels of C99 or ␤CTF expression. Fourth, cleavage of APP by BACE was not necessary for the A␤ rise because a rise could be observed even when ␤CTF was expressed ectopically. Fifth, A␤ exhibited a rise in the brains of rats dosed with a ␥-secretase inhibitor, showing that a rise can occur in the therapeutically relevant organ.
The Potency Shift-In addition to the A␤ rise, low level C99 or ␤CTF expression caused a decrease in ␥-secretase inhibitor potency. Two observations indicated that a significant part of this "potency shift" was independent of the A␤ rise. First, ␥-secretase inhibitor potency varied as a function of C99 expression even when an A␤ rise did not occur (Fig. 3). Second, evaluation of a group of ␥-secretase inhibitors showed that the extent of the A␤ rise was not correlated with the extent of the potency shift (Fig. 5), suggesting that the underlying mecha-nisms of the rise and the potency shift were not identical. Thus, there were apparently two phenomena dependent upon the level of C99 or ␤CTF expression in the presence of ␥-secretase inhibitors: the A␤ rise and a potency shift.
A simple hypothesis for the potency shift is based on the molar ratio between enzyme and substrate. To calculate this ratio in HEKwt and HEKsw cells, the quantity of ␥-secretase enzyme was determined from saturation binding isotherms, and the quantity of cellular ␤CTF was determined by quantitative Western blotting calibrated by mass spectroscopy (see "Experimental Procedures"). In HEKwt and HEKsw cells, the total substrate to enzyme ratios were 0.075 and 0.36, respectively. Additional factors are expected to affect the functionally relevant ratio in cells, including the differential subcellular localization of enzyme and substrate molecules, the presence of other proteins that are ␥-secretase substrates, and the separation of ␥-secretase into lipid raft microdomains (33). Therefore, ␥-secretase is potentially not ratelimiting for A␤ production such that a relatively high level of enzyme inhibition would be required before substrate turnover and A␤ production was affected. Indeed we observed that a high level of inhibitor binding site occupancy was required for inhibition of A␤ production in the HEKwt cells (Fig. 7). In addition, the observation that ␥-secretase inhibitor binding was not affected by differences in the level of substrate expression suggested that an uncompetitive inhibition mechanism based on preferential binding to the enzyme-substrate complex was not involved. Thus the difference in substrate/enzyme ratio between the HEKwt and HEKsw cell lines could contribute to the difference in response of A␤ to ␥-secretase in these cell lines. In support of this hypothesis, the A␤ rise and potency shift could be controlled by using transient transfection to vary ␤CTF expression levels in cells (Fig. 3).
The A␤ Rise as an Intrinsic Characteristic of ␥-Secretase-A key question is whether or not the A␤ rise is mediated directly by inhibitor binding to ␥-secretase itself or indirectly by inhibitor binding to a secondary target involved in A␤ turnover as suggested previously (5). Pharmacological evidence favors the direct mechanism. First, a wide structural variety of ␥-secretase inhibitors cause the A␤ rise, including both the allosteric and active site-directed classes of inhibitors. These diverse compounds have one obvious property in common: their inhibition of ␥-secretase. Second, there are many inhibitors that cause an A␤ rise with potencies in the subnanomolar range. The indirect mechanism would require commensurate potency for the putative A␤ turnover target. Thus, in agreement with previous suggestions (6,14), the A␤ rise is likely mediated through ␥-secretase itself either by increased production of A␤ or decreased turnover of A␤.
The A␤ Rise Is Not Consistent with the "Incomplete Processing" Hypothesis-Inhibition of ␥-secretase-mediated turnover of A␤ appears to explain at least some cases of rises in A␤, for example the increase in A␤-(1-42) caused by presenilin familial Alzheimer disease (FAD) mutants. Ihara and co-workers (34 -38) and Xu and co-workers (10, 39 -41) have reported a stepwise mechanism of substrate cleavages by ␥-secretase in which the shorter forms of A␤ are derived from longer forms of A␤ by successive proteolytic cleavages of the same substrate molecule. Some ␥-secretase inhibitors cause the accumulation of longer forms of A␤, which are ␥-secretase processing intermediates that remain bound to the enzyme. Taken together, these studies show that longer forms, such as A␤-(1-46) (40), accumulate because inhibitors cause incomplete processing to the shorter secreted forms of A␤, such as A␤- . Taking this idea further, it was proposed that the increase in A␤-  caused by some presenilin FAD mutants results from incomplete processing of A␤-(1-42) to shorter forms (42,43). We therefore considered the possibility that ␥-secretase inhibitors might cause incomplete processing of A␤ in the HEKwt cells. Using an ELISA that quantifies the combined A␤-(1-x) species, we detected robust A␤-(1-x) rises in response to DAPT and L-685,458 (Fig. 1), indicating that incomplete processing would have to involve A␤ peptides shorter than A␤-(1-24) in this case. Mass spectrometry confirmed the rises in A␤ peptides with C termini at positions 37, 38, and 40 (Fig. 2). In principle, inhibition of shorter peptides that can be derived from C99 or ␤CTF but not from the N-terminally shorter ␣CTF could explain why rises were seen for A␤ but not for p3. However, shorter A␤ peptides appeared insufficiently abundant to account for an A␤ rise through incomplete processing. One shorter peptide itself, A␤- (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19), exhibited a rise in response to DAPT (Fig. 2F). Thus, we found no evidence for significant ␥-secretase-mediated turnover of A␤, although this does not rule out the possibility that turnover could result in abundant very short peptide fragments that we did not detect in our assays. Furthermore there are several noteworthy differences between increased A␤42 in presenilin FAD mutants and the inhibitor-mediated A␤ rise. First, the A␤42 increase in FAD mutants does not require low substrate expression. Second, FAD mutants decrease total A␤ levels, whereas low concentrations of ␥-secretase inhibitors can increase total A␤ (Fig. 1). Third, FAD mutants caused a change in presenilin conformation, whereas ␥-secretase inhibitors did not (44,45). Thus, it seems possible that the inhibitor-mediated A␤ rise may involve a mechanism different than that involved in the A␤42 increase observed in presenilin FAD mutants. The A␤ Rise May Depend on Dual Roles of ␥-Secretase in the Endocytic Pathway-Production of p3 and Swedish APP-derived A␤ both occur predominantly in the late secretory pathway (46 -49), and in both cases we found no rise in these peptides upon ␥-secretase inhibitor treatment. In contrast, A␤ derived from wild type APP occurs predominantly in the endocytic pathway (47,50,51), suggesting that subcellular trafficking pathways determine the A␤ rise. This may be significant because inhibition of ␥-secretase leads to accumulation of ␤CTF in the endocytic pathway (52). Potentially this could lead to increased exposure of ␤CTF to ␥-secretase, which could increase A␤ production if the increased ␤CTF exposure exceeded the inhibition of ␥ cleavage. This circumstance might occur under conditions of low substrate expression where the enzyme would be present in relative excess consistent with our ␤CTF transfection studies (Fig. 3). A mechanism of this type would require intact endocytic membrane trafficking, hindering reproduction of the A␤ rise in solubilized ␥-secretase assays (Fig. 4). Thus, the A␤ rise would be readily apparent only under conditions in which A␤ peptides are produced predominantly in the endosomal trafficking pathway where ␥-secretase plays dual antagonistic roles in A␤ production.
Implications for Drug Discovery-We showed that a robust A␤-(1-40) rise can occur in rat brain under conditions of partial occupancy of inhibitor binding sites, demonstrating that ␥-secretase inhibitors have the potential to cause the opposite of the intended inhibitory effect in the target organ. In contrast to the rat, dosing of the Tg2576 mouse with the same ␥-secretase inhibitor did not result in an A␤ rise (30) consistent with the results we obtained in cell cultures that overexpress the ␤CTF or APP Swedish mutant. This suggests that the Tg2576 mouse can exhibit a misleading degree of inhibitory activity for ␥-secretase inhibitors. The potential influence of the A␤ rise  and potency shift in transgenic mice could therefore lead to misleading estimates of the exposure necessary for A␤ lowering in non-transgenic animals. This is important for a target such as ␥-secretase where the therapeutic index is narrow and must be carefully monitored (53). As a consequence, ␥-secretase inhibitors that do not cause an A␤ rise would be advantageous as drug candidates. The findings described here demonstrate that the A␤ rise and potency shift are relevant issues for the development of ␥-secretase inhibitors and enable these phenomena to be studied under defined conditions.
In conclusion, the A␤ rise and potency shift are intrinsic characteristics of inhibitor binding to ␥-secretase and depend on a low level of C99 or ␤CTF substrate expression perhaps because of a limitation in the amount of substrate trafficking within the endocytic pathway. The mechanistic details remain unresolved; however, our results show how the A␤ rise and potency shift phenomena can be taken into account for the purpose of drug discovery by making appropriate choices of animal and cell culture models.