In order to more accurately and sensitively quantify caspase-cleaved APP in cell and tissue lysates, we partnered with Enzo Life Sciences to develop both a cleavage site-specific (neo epitope) polyclonal antibody that recognizes the C-terminus of APPΔC31 resulting from the caspase cleavage and an ELISA (ENZ-ABS445-0100, ADI-900-227, respectively). The immunizing antigen comprised a short peptide sequence on the C-terminus of APPΔC31 (red circle, Figure 1).

The specificity of the antibody for the neo epitope, rather than full length APP, was validated by immunoblot (Figures 2A,B, Supplementary Figure S1A) and ELISA (Supplementary Figure S1B). Human embryonic kidney (HEK 293T) cells were grown in high glucose DMEM with 10% heat-inactivated fetal bovine serum and 1X antibiotic/antimycotic and transfected with full-length pcDNA3-APP₆₉₅. pcDNA3-APPΔC31, or empty pcDNA3 vector using Lipofectamine 2000 (Life Technologies). Two days later, RIPA buffer lysates complemented with complete protease inhibitors (PI, Roche) were prepared. Samples were run on 12% Bio-Rad Criterion gels, transferred onto 0.2 μm PVDF membrane and probed with the anti-APPΔC31 antibody. Anti-rabbit secondary antibody conjugated to HRP (horseradish peroxidase, Amersham) was then incubated with the membrane, signal was generated using chemiluminescence (Thermo) and film exposed. In addition, samples were analyzed by APPΔC31 ELISA.

Similarly, Chinese hamster ovary cells stably transfected with human wildtype (wt) APP₇₅₁ (CHO- 7W, a kind gift from Dr. Edward Koo) were cultured in DMEM, 10% FBS, 1X antibiotic/antimycotic. They were treated with activated simvastatin or staurosporine both at 10 μM for 24 h. RIPA lysates were prepared using complete protease inhibitors (Roche).

For immunoprecipitation, approximately 300 μg cell lysate supernatants were brought to 250 μL with cold RIPA + PI buffer in microfuge tubes. One μL of 6E10 APP antibody (Covance) was added and the mixture was rotated overnight in a cold room. The next day, 25 μL of protein A/G beads (Santa Cruz Biotech) were added and the samples were rotated for 1.5 h in the cold room. The beads were then centrifuged down and washed four times with cold PBS. Forty μL 1x LDS sample buffer + DTT was added to the pellets, which were heated to 70°C for 10 min, vortexed and centrifuged. Supernatants were loaded onto Nupage 4–12% Bis-Tris gels (Life Technologies) for electrophoresis and were then transferred to 0.2 μm PVDF membrane for immunoblotting with the anti-APPΔC31 antibody.

A custom AlphaLISA (Perkin-Elmer) was developed to quantitate APPΔC31 in the CHO-7W screening studies described herein. As shown in Figures 2C–E, biotinylated N-terminal APP antibody AF1168 (R&D) is bound to streptavidin-conjugated donor beads, and the Enzo anti-APPΔC31 antibody is used to coat acceptor beads. In the assay, 0.003 μL N-terminal anti-APP antibody/μL and 0.002 μL acceptor-bead-bound anti-APPΔC31 antibody/μL are added to samples in 1X HiBlock AlphaLISA buffer (Perkin-Elmer) on the assay plates. Plates are sealed and pulse centrifuged at 1000 rpm after each addition or mix step, shaken for 1 hr, and then incubated an additional hour without shaking in the dark. Streptavidin donor beads (PerkinElmer) are then added (1.5 μL, to give 30 μg/mL final), followed by 1 hr incubation with plate shaking. When streptavidin-donor beads are added to the reaction, they bind the biotinylated N-terminal APP antibody. When both antibodies bind with APPΔC31, they are in close enough proximity for excitation at 680 nm to result in the production of ¹O₂, interaction with the acceptor bead and emittance at 580 nm that is then detected using an Envision plate reader (Perkin-Elmer).

The APPΔC31 ELISA developed in partnership with Enzo Life Sciences (ADI-900-227) follows a standard sandwich ELISA design wherein a monoclonal N-terminal APP-specific antibody is used to coat the wells of the microtiter plate (capture antibody) and the anti-APPΔC31 cleavage site-specific polyclonal antibody serves as the detection antibody. An HRP-conjugated secondary antibody is then added and a colorimetric signal generated by enzymatic conversion of the tetramethylbenzidine (TMB) substrate. Protein levels were first determined using a colorimetric coomassie assay (Coomassie Plus, Thermo).

Since APPΔC31 is typically produced at very low levels and below the level of detection in the in vitro systems used here, it was necessary to induce this caspase cleavage in order to then knock it down. We previously reported the ability of statins, including simvastatin and cerivastatin, to stimulate the caspase cleavage of APP (Descamps et al., 2011). Here, we utilized this ability to discover APP C-terminal caspase cleavage inhibitors using CHO-7W cells. Cells were treated with activated simvastatin at 6 doses ranging from 100 nM to 10 μM, for 7.5 and 24 h, at which time APPΔC31 was measured in cell lysates using the AlphaLISA. Simvastatin was activated by opening the lactone ring before use in cell culture as previously described Dong et al. (2009); 8 mg simvastatin was dissolved in 0.2 mL of 100% ethanol (0.019 mM), with subsequent addition of 0.3 mL of 0.1 N NaOH and the solution was heated at 50°C for 2 h and then neutralized with HCl to pH 7.2. The resulting solution was brought to 1 mL final with distilled water, and aliquots were stored at -80°C.

In a separate experiment, CHO-7W cells were treated with: simvastatin, cerivastatin, and atorvastatin at 1, 5, and 10 μM, 5 μM simvastatin plus 30 μM pan-caspase inhibitor Q-VD-OPh (QVD, Sigma-Aldrich), and no statins, all for 24 h. APPΔC31 was determined in cell lysates by ELISA.

For HTS, CHO-7W cells were plated into 384-well Proxiplate SW plates (Perkin-Elmer) at 4K cells/well/10 μL in the growth media/conditions described above but also containing 1X Glutamax, 20 mM HEPES, 1X NEAA and 500 μg/ml G418. The following day, simvastatin at a final concentration of 10 μM and test compounds at 30 μM were added to assay plates via ECHO^® acoustic transfer (Labcyte). The next day, media was removed from the cells by “Flick and Slam” and centrifuging upside down. Cells were lysed in 3 μL of AlphaLISA buffer complemented with protease inhibitors on a plate shaker for 15 min.

The APPΔC31 AlphaLISA assay (Figures 2C–E) was run directly in the plate with the cell lysate as described above in “AlphaLISA to detect APPΔC31.” APPΔC31 levels were measured using the Envision plate reader and data were normalized to DMSO (100% APPΔC31 production) and Q-VD-OPh (MP Biomedicals) pan-caspase inhibitor at 30 μM (0%); these controls were present on each plate. In addition, some compounds of interest were further analyzed in a dose-response curve to determine IC₅₀ values.

Compound libraries screened and confirmed included: (1) the US (1040 compounds) and International (240 compounds) Drug collections (Microsource); (2) the Enzo protease inhibitor library (60 compounds); and (3) a 903 compound natural products screen including a 502 compound Enzo natural products library, a 64 compound EMD inhibitors library, a 97 compound Enzo autophagy library, and a 240 compound GreenPharma natural products library.

A subset of hits from our HTS was validated in a secondary assay. In addition, since the HTS hit spiperone has been reported to be a Wnt-signaling pathway inhibitor, we included in our testing another compound known to affect this pathway, thapsigargin, along with a spiperone analog 3′-fluorobenzylspiperone maleate (3′-fluorobenzylspiperone). Thapsigargin also acts as a sarco/endoplasmic reticulum calcium ATPase (SERCA) inhibitor. Therefore, another SERCA inhibitor, 2,5-Di-tert-butylhydroquinone, was also added to the testing. In this secondary assay, CHO-7W cells were cultured as described above and plated into 24-well dishes at ∼200K/well. The next day, cells were treated for 24 h in duplicate with 200 μL fresh media containing 10 μM cerivastatin (Sigma) to induce APPΔC31 production as well as select hits at 3, 10, and 30 μM, although the more potent thapsigargin was used at 0.033, 0.1, and 0.3 μM, and 2,5-Di-tert- butylhydroquinone at 1 and 10 μM. These hits were selected for their potency in primary screen, their structural features, their known activities, and availability of compound for additional testing. There were cerivastatin-only (negative control), + QVD-OPh (positive control), and for viability, vehicle-only (negative) controls. Cells were harvested by removing the media to a tube, adding 100 mL trypsin to detach cells then neutralizing with media. Cell viability (% live) was determined by trypan blue exclusion using BioRad’s TC-20 cell counter as it was important to ensure that any observed APPΔC31 inhibition was not due to inhibitor toxicity. Cells were centrifuged at 5000 rpm (2300 × g) for 5 min. Cell pellets were frozen until use. Forty μL cold RIPA buffer plus complete protease inhibitors (Roche) was added to each pellet and the pellets were reconstituted by pipetting up and down, incubating on ice for 30 min, vortexing, and centrifugation as above. Cell lysate supernatants were then assayed by APPΔC31 ELISA as described above. For most compounds, multiple/repeat studies were performed and two wells used for each condition. Statistical comparison was performed using one-way ANOVA for multiple comparisons, with a Dunnett post hoc test to compare each condition to the control. Further dose-response studies were performed on select compounds.

Our lead APPΔC31 inhibitors were further tested ex vivo in I5 mouse organotypic hippocampal slice cultures in pilot experiments (Supplementary Figure S4). I5 mice express wt hAPP under the control of the platelet-derived growth factor promoter (Mucke et al., 2000). I5 mice originally purchased from Jackson labs (JAX B6.Cg-Tg(PDGFB-APP)5Lms/J) were bred and maintained in the Buck Institute vivarium facility, and used under the guidelines of an approved institutional IACUC protocol. P5–7 day mice were sacrificed using CO2 gas and cervical dislocation according to established IACUC protocols. Brains were removed, and under sterile conditions, hippocampi were dissected out on ice in Hybernate media + 1:400 Glutamax, 1:100 Pen/Strep and 1:50 B27 supplement (all Life Technologies). Hippocampi were laid parallel on a tissue chopper (McIlwain) in a small amount of media and sliced at 400 μm thick. Millicell-CM 6-well cell culture inserts (Millipore, 0.4 μm) were prehydrated by pipetting 750 μL culture media under each well and placing in a 35°C, 5% CO2 incubator for at least 30 min. Culture media: 100 mL MEM-HEPES, 50 mL HBSS, 50 mL HI horse serum, 1 mL 100x Pen/Strep (all Gibco), 1.3 g (6.5 mg/mL) glucose, and 2 mL 200 μM (2 mM) Glutamax, sterile filtered. Approximately equal portions of sliced hippocampi (4–5 pups/well) were placed onto inserts in minimal media (as stated above) and cultured for 5 days before treatment, changing the media under the insert every 2–3 days. After 5 days, treatment with vehicle (DMSO), cerivastatin (CS) at 30 μM, or CS plus 30 μM Q-VD-OPh, 10 μM spiperone, or 30 μM 3′-fluorobenzylspiperone for 72 h, with treatment being replaced under the insert each day. At the end of treatment, tissue was rinsed gently with PBS and harvested into a tube by gently scraping off the membrane with a scalpel using ∼100 μL/well cold RIPA buffer + protease inhibitors. Tissue was sonicated on ice for ∼5 s at “60” (Sonics Vibra Cell). Tissue sat on ice 30 min, was vortexed and centrifuged 5 min at 5000 RPM (2300 × g) before assaying the supernatant. Protein was quantitated for each sample using Coomassie Plus reagent (Thermo). An AMP’d Signal Amplification kit (Enzo) was used with the ELISA to detect APPΔC31 in these cultures as the signal was very low.

In an additional study (Supplementary Figure S5A), measurement of the α secretase fragment of full length (FL) APP – soluble APP alpha (sAPPα) – was performed using cell media by ELISA (IBL). Furthermore, in two studies (Supplementary Figures S5B,C), total protein levels in CHO-7W cell lysates were determined by Coomassie Plus protein assay (Pierce).

Several compounds emerged from testing as possible leads. Since the brain is the target tissue, brain-penetrance is a key feature of any potential new AD therapeutic. Immobilized artificial membrane (IAM) chromatography such as PAMPA (parallel artificial membrane permeability assay) is used to predict a compound’s passive diffusion through membranes and may predict blood-brain barrier (BBB) penetrance.

To minimize variability between PAMPA assays for different compounds, the well-characterized, reliable IAM.PC.DD2 Regis Technologies analytical columns were used. In these 10 cm × 4.6 mm columns, the phospholipid is bonded on 10 μm, 300 Å, spherical aminopropyl silica and end capped with C10 and C3 amides. It has been reported that, while column aging depends upon the column used as well as the test analyte, the intra-batch variability for IAM.PC.DD2 phases was small (Taillardat-Ertschinger et al., 2002). In addition, Ong et al. (1996) and others (Yoon et al., 2006; Shin et al., 2009) have shown that the IAM.PC.DD2 could be used as a rapid screening method for the prediction for drug absorption. In our hands, this PAMPA method has shown good correlation with brain-penetrance determined in vivo by pharmacokinetic (PK) analysis, and allows us to prioritize compounds for PK study.

Columns were connected to an Agilent high performance liquid chromatography system. Compounds (spiperone, pimozide, and 3′-fluorobenzylspiperone) were prepared at 10 μM in DMSO and diluted to 500 μM in 50:50 water:methanol. The mobile phase was a mixture of water and methanol and a gradient was used for the elution of the compounds. The detector was set at 250 and 280 nm. After a compound was eluted, the retention time and void volume times were obtained from the chromatogram.

The IAM capacity factor (KIAM) was calculated using the equation: KIAM=tr−t0t0; where tr is the

retention time of a compounds and t0 is the void volume time of the column. It has been shown that the membrane permeability of a drug following passive diffusion is directly proportional to the K_IAM and inversely proportional to the molecular weight of a compound. To predict the likelihood a compound would be BBB penetrant, we used the correlation described by Yoon et al. (2006):

As PAMPA is merely a predictive analysis of brain permeability, a pharmacokinetic (PK) study with spiperone was performed. Five adult (>3 months of age) male mice (C57B6, Jackson Labs) were injected subcutaneously (SQ) with 50 μL of spiperone at 5 mg/mL in 100% DMSO. Another 5 mice were injected intraperitoneally (IP) at the same dose. One mouse was euthanized from each dosing route group at 1, 2, 4, 6, and 8 h. This is done by ketamine/xylazine over- anesthesia followed by blood collection for plasma isolation by cardiac puncture, and transcardial perfusion of tissue with saline to remove residual blood from the brain. Both plasma and brain tissue were sent to Integrated Analytical Solutions (IAS, Berkeley, CA, USA) for compound level analysis using an LC-MS-MS method.

The animal testing was carried out in accordance with Public Health Service Policy on Humane Care and Use of Laboratory Animals, Assurance # A3196-01 and in the Buck Institute vivarium that has received AAALAC accreditation.

Immunoblot analysis shows that the anti-APPΔC31 antibody is specific for caspase-cleaved APP, and does not react with full-length APP. As seen in Figure 2A, the band just below 100 kDa expressed after transfection of HEK 293T cells with pcDNA3-APPΔC31, but not for pcDNA3-FL APP is revealed by the antibody. In Figure 2B, again only APPΔC31 induced by treatment of CHO-7W cells with either staurosporine or simvastatin produced a band detectable with the antibody; FL APP present in lysates of untreated CHO-7W cells was not detected. In both cell types, two non-specific bands appeared (the lower is a triplet) and while the proteins in these bands have not been identified, they could be metabolites of APPΔC31 or other non-APP caspase-cleaved proteins with a similar antigenic site. It is important to note that in both the AlphaLISA and ELISA used for the screening and validation studies, a second N-terminal APP-specific antibody is used so that only APPΔC31 is detected and not shorter APP fragments.

Antibody specificity was also confirmed by analysis of purified MBP (maltose binding protein)-APPC125 (C-terminal 125 amino acids of APP including the β-, α-, γ-, and caspase-cleavage sites) fusion protein reactivity to anti-APPΔC31 as compared to anti-APP (6E10) by immunoblot where only anti-APP and not anti-APPΔC31 recognized this uncleaved fusion protein (Supplementary Figure S1A). In addition, overexpressed APPΔC31 and full-length APP₆₉₅ from HEK 293T cell lysates were tested by APPΔC31 ELISA and showed similar results with the full-length APP lysate being almost undetectable with less than 1% cross-reactivity (Supplementary Figure S1B).

As shown in Figure 3A, an initial time- and dose-response study showed simvastatin induction of APPΔC31 was both time and dose-dependent. Simvastatin increased APPΔC31 proportionally from 1 to 10 μM, and this was seen at 24, but not 7.5 h. Therefore, a 24-h treatment time was used for further studies. Simvastatin was then compared to cerivastatin and atorvastatin. Of the statins tested, cerivastatin was found to be the most effective at inducing the caspase cleavage of APP (Figure 3B). These results led us to use cerivastatin in the validation studies.

Induction of cerivastatin-induced APPΔC31 production was also seen by immunoblot. 3′-fluorobenzylspiperone decreased 10 μM CS-induced APPΔC31 at 30 μM here (boxed area). QVD at 30 μM eliminated the cerivastatin-induced APPΔC31 increase (Supplementary Figure S2B) and inhibited APPΔC31 production in a dose-dependent manner (Supplementary Figure S2A).

A scatterplot (Figure 4) shows that a total of 91 hits out of 2243 compounds screened (4% hit rate) were found, using a hit criterion of at least 70% inhibition of APPΔC31 production. The red-filled circle represents the inhibition of APPΔC31 by the control pan-caspase inhibitor QVD-OPh, and hits initially selected for further evaluation that have undergone hit confirmation are represented by blue-filled circles. Screening was performed in duplicate and the average coefficient of variation (CV) across all assays was 17 ± 4%.

These hits comprised many different categories of compounds including histone acetyltransferase inhibitors, antibacterials, antifungals, cathepsin inhibitors, protease inhibitors, kinase inhibitors, mTOR inhibitors, natural products, protein synthesis inhibitors, phosphodiesterase inhibitors, Ca2⁺ channel blockers, Ca2⁺ signaling modulators, GPCR agonists and antagonists, and anti-inflammatory agents. Identities and APPΔC31 inhibition data for all confirmed hits from our HTS are shown in Supplementary Table S1. Potential therapeutic hits based on structure were cherry-picked for further validation and study.

Confirmed screening hits were validated for inhibition of APPΔC31 induction by ELISA using larger format cultures of CHO-7W cells, and several additional compounds were added in these secondary assays based on their known mechanisms of action. Specifically, since the hit spiperone is a known Wnt signaling inhibitor, another known inhibitor of Wnt signaling was analyzed as well, thapsigargin. Thapsigargin is also a SERCA inhibitor, and so another SERCA inhibitor, 2,5-Di-tert-butylhydroquinone was also analyzed. Dose-response analyses were performed on hits chosen for their potency and efficacy of APPΔC31 inhibition as well as some for structural features. Some of these compounds were run in 4- to 8-point dose-response curves as shown in Figures 5B,C.

Most of these compounds reduced cerivastatin-stimulated APPΔC31 in a dose-responsive manner after 24 h (Figures 5A–C) with Q-VD-OPh serving as the positive control. Thapsigargin (Figure 5A, Supplementary Figure S3A) proved to be a remarkably potent inhibitor of APPΔC31 production and another SERCA inhibitor, 2,5-Di-tert-butylhydroquinone, was also very effective. HTS hits spiperone, 3′-fluorobenzylspiperone maleate and pimozide were also potent inhibitors. The IC₅₀ values (μM) of the lead compounds were: spiperone 4.6, pimozide 3.6, 3′-fluorobenzylspiperone maleate 2.2, and thapsigargin 0.077.

Cerivastatin induction of APPΔC31 and inhibition by Q-VD-OPh and 3′-fluorobenzylspiperone in a dose-dependent manner was also demonstrated by APPΔC31 immunoblots from CHO-7W cell lysates (Supplementary Figure S2).

Since APPΔC31 is generated by caspase cleavage and we found some MEK inhibitors in our HTS, select caspase and MEK inhibitors were also tested for inhibition of cerivastatin-induced APPΔC31 in CHO-7W cells for 24 h using dose-response curves determined by AlphaLISA (Supplementary Figure S3B). They included Caspase 6 Inhibitor I and II, Caspase Inhibitor X (inhibits caspases 3, 7, and 8), AZ10417808 (inhibits caspase 3) and Ivachtin (inhibits caspase 3). Although none of these caspase inhibitors were found to be very effective here, Caspase Inhibitor X (potency for various caspases: caspase 3 IC₅₀ 4.3 μM; caspase 7 IC₅₀ 6.2 μM; caspase 8 IC₅₀ 2.7 μM) and Caspase 6 Inhibitor I at the highest concentration tested (30 μM) were somewhat effective. The broad-spectrum pan-caspase inhibitor Q-VD-OPh served as the positive control. The MEK1/2 (MAPK/ERK kinases) inhibitor U0126 reduced APPΔC31 ∼25% from 10 to 50 μM while the inactive isomer U0124 was ineffective. The MEK inhibitor SL327 was ineffective.

Lead APPΔC31 inhibitors were also tested in P5–7 day I5 wt mouse ex vivo organotypic hippocampal slice cultures (Supplementary Figure S4). These pilot studies were performed for lead compounds and it should be noted that the APPΔC31 signal was found to be low. Results within each experiment were protein normalized and are represented as % of cerivastatin control, which was set at 100% for each experiment, and thus there are no error bars for that treatment. Spiperone at 10 μM inhibited APPΔC31 production by ∼14% while 3′-fluorobenzylspiperone maleate at 30 μM inhibited it by ∼82%. Preliminary results suggest that these compounds are also efficacious ex vivo with, and some without, cerivastatin induction. Interestingly, pilot studies in primary I5 neuronal cultures treated with cerivastatin did not induce APPΔC31, additional investigation with neuronal cultures were not done as we found that the APPΔC31 could be induced by cerivastatin in slice cultures.

Cell viability was measured together with APPΔC31 since induction of APPΔC31 production (here stimulated by cerivastatin) increases cell death. It was also important to ensure that any observed APPΔC31 reduction was not due to compound toxicity. As seen in Figure 5D, increases in cell viability typically mirrored decreases in APPΔC31, confirming the toxic effect of the caspase cleavage of APP. Among hits from HTS, spiperone, pimozide, and rapamycin showed the greatest cell survival rescue effects at 30 μM. Of these, two compounds that were added to secondary screening due to their known mechanism(s) of action – thapsigargin and 2,5-Di- tert-butylhydroquinone – elicited the highest cell survival among compounds tested at the low concentrations. Complete APPΔC31 inhibition and viability data for all confirmed hits are shown in Table 1. Cell viability for hits tested represents rescue if greater than the level of cerivastatin treatment alone. DMSO represents maximal viability.

As shown in Supplementary Figure S5A, treatment of CHO-7W with QVD, spiperone and 3′fluorobenzylspiperone increased sAPPα in the presence 10 μM cerivastatin, did not significantly affect sAPPα indicating that these compounds do not affect APP expression. Furthermore, QVD alone at 30 μM and spiperone alone at 10 μM did not lower protein concentration in CHO-7W lysates as compared to vehicle as shown in Supplementary Figure S5B. In addition, and S5C QVD, spiperone and 3′fluorobenzylspiperone alone at 30, 10, and 10 μM, respectively, increased total protein concentration in the presence of 10 μM cerivastatin (Supplementary Figure S5C).

As shown in Figure 6A, of the three compounds tested in PAMPA, spiperone had the lowest tr and KIAM and the highest KIAMMW4x1010, and thus was predicted to be more brain-penetrant than pimozide or 3′-fluorobenzylspiperone maleate. This was not unexpected as spiperone also has the lowest MW of the three compounds. Pharmacokinetic analysis revealed spiperone did indeed enter the brain, peaking at 1 h for both the subcutaneous and intraperitoneal routes at ∼70 ng/g and ∼205 ng/g, respectively (Figure 6B). Plasma levels were higher at all time points and by both routes, with a brain:plasma ratio of ∼0.5.