Development of a Novel β-Secretase Binding Assay Using the AlphaScreen Platform

Abstract

Alzheimer’s disease (AD) is a devastating neurodegenerative disease affecting millions of people. β-secretase-1 (BACE1), an enzyme involved in the processing of the amyloid precursor protein (APP) to form Aβ is a validated target for AD. Herein, the authors develop and validate a novel binding assay for BACE1 using the AlphaScreen platform that is amenable for high-throughput screening (HTS). Small-molecule BACE1 inhibitors of the hydroxyethylamine, hydantoin, and sulfamide classes were functionalized by biotin PEG linkers of varying lengths forming probes that were bound to streptavidin donor beads. BACE1 was coupled to nickel-chelate acceptor beads. Upon mixing, probes designed from all three classes registered high signal-to-background values in the AlphaScreen binding assay, where the interaction between probe and BACE1 was completely blocked by free parent compound. A probe from the hydantoin class was chosen for further optimization, where the final assay conditions of 50 nM BACE and 250 nM probe were used and Z^′ values >0.75 were commonly observed. IC₅₀ values determined by the AlphaScreen assay format exhibited ~10-fold greater sensitivity when compared with a fluorescence polarization–based activity assay. The assay was miniaturized to a 1536-well format for HTS, in which 525 000 compounds were screened.

Introduction

Alzheimer’s disease (AD) is a chronic and progressive neurodegenerative disease characterized by irreversible loss of memory, overall cognitive decline, and ultimately death. It is estimated that more than 20 million people worldwide have AD, where it is the leading cause of dementia in the elderly population. ¹ Several defining hallmarks of AD include the presence of senile plaques consisting of insoluble Aβ, neurofibrillary tangles composed of hyperphosphorylated tau, and dystrophic neurites. ² Biochemical, genetic, and animal model data suggest that the pathogenesis of AD is related to the accumulation of Aβ_1-40 and Aβ_1-42, more formally known as the amyloid hypothesis. ³ Aβ_1-40 and Aβ_1-42 are formed by the sequential proteolytic cleavage of the amyloid precursor protein (APP) by β-secretase-1 (BACE1) and γ-secretase.^2–6

BACE1 is a type I integral membrane glycoprotein consisting of a large ectodomain of ~434 amino acids, a single transmembrane domain of ~22 amino acids, and a short cytoplasmic tail of 24 residues. BACE is an aspartyl protease in which the active site contains two conserved aspartic acid residues at Asp32 and Asp228 within the ectodomain. BACE has two isoforms (BACE1 and BACE2), where BACE2 has 68% similarity to BACE-1.^5,6 Despite the high homology to BACE1, BACE2 has a different tissue distribution, different subcellular localization, and substrate specificity compared with BACE1.^5–7 In a series of genetic knockout experiments in mice, it was shown that APP does not get processed to Aβ in the absence of BACE1, indicating that BACE1 is the isoform of relevance for AD.^5,6,8

In the opinion of many investigators, BACE1 is a validated target for AD. Mutations in APP that cause familial AD are all located near the secretase cleavage sites. ⁶ For example, the Swedish mutation of APP is located immediately adjacent to the BACE1 cleavage site of APP.^9,10 Synthetic peptides that contain the Swedish sequence are cleaved with 60-fold higher catalytic efficiency by BACE compared with APP peptide substrates based on the wild-type sequence. ¹¹ In accordance with this result, cells expressing the Swedish mutant of APP accumulate higher levels of Aβ than cells expressing wild-type APP.^6,9,10 Interestingly, individuals with Down’s syndrome, caused by an extra copy of chromosome 21, which contains the APP gene, invariably develop AD-like pathology usually by the fourth decade of life. ¹² This clinical observation is consistent with the amyloid hypothesis and putative role of BACE1 in AD as higher substrate levels of APP that occurs as a result of the extra copy of the APP gene would lead to higher levels of Aβ accumulation. Recently, a mutation in APP was shown to protect against AD and age-related cognitive decline in data obtained from a cohort of 1795 Icelanders. ¹³

Historically, the most compelling data for validation of BACE1 as a target for AD comes from a number of genetic knockdown studies of BACE1 performed in AD transgenic mice. In Tg2576 and Tg6799 mice, genetic ablation of BACE1 rescued Aβ−dependent hippocampal memory deficits.^14,15 In addition, Aβ-mediated cholinergic dysfunction was also rescued in Tg2576 BACE knockout mice. Although initial reports demonstrated that BACE1 knockout mice were healthy and normal, later reports showed that genetic knockout of BACE1 in mice did result in mechanism-based toxicity. ¹⁶ In a landmark study by McConlogue et al., ¹⁷ this group demonstrated that partial knockdown of BACE1 had dramatic effects on Alzheimer plaque and synaptic pathology in PDAPP mice. Furthermore, the heterozygous mice were determined to be healthy and normal in numerous cohorts. Taken together, the partial reductions of BACE1 enzyme activity that lead to a marked reduction of Aβ-mediated pathology in AD transgenic mice suggest that small-molecule inhibition of BACE1 could be disease modifying in the absence of a mechanism-based toxicity in patients.

There has been significant interest in BACE from the pharmaceutical industry, which has led to the development of a number of assays amenable to HTS. One specific example would be a fluorescence resonance energy transfer assay in which donor and acceptor fluorophores can be covalently attached to a BACE1 substrate peptide. Upon BACE1-mediated cleavage of the peptide, donor quenching is relieved with a concomitant emergence of acceptor fluorescence. ¹⁸ This type of assay can suffer from fluorescence interference from compounds commonly found in HTS libraries ¹⁹ and often times may not have a high enough signal-to-background (S/B) ratio for high-throughput screening (HTS). Fluorescence polarization (FP) assays have been used for other aspartic proteases ²⁰ but also may suffer from compound-based fluorescent interference ²¹ as well as sensitivity issues. BACE1 inhibitors initially identified by conventional HTS using the BACE1 ectodomain and assay conditions of pH 4.5 have not been successful in generating compounds suitable for clinical testing. ²² Subsequently, structure-guided drug design approaches initiated by a fragment-screening effort have yielded more tractable compounds and are in clinical phase testing.^23–26 This new wave of small-molecule BACE1 inhibitors may bring hope to AD patients in the future, yet it remains to be determined if they will have potential toxicity or pharmacokinetic (PK) liabilities in clinical phase testing.

More innovative assay approaches for BACE1 offering robust sensitivity have been described. One such example is a β-galactosidase (β-Gal) enzyme fragment complementation assay. ²⁷ This assay uses a cyclic β-Gal enzyme donor peptide containing a protease-selective cleavage sequence. ²⁷ After cleavage by BACE, the peptide is linearized and can now complement with the β-Gal enzyme acceptor forming an active β-Gal enzyme. However, it remains to be determined whether these other assay approaches will lead to compounds that are successful in clinical phase testing. Thus, new and better scaffolds for BACE1 inhibitors should be useful in the future. Herein, we describe a novel binding assay that is amenable to HTS to identify small-molecule inhibitors of BACE1. This assay uses the AlphaScreen technology and shows enhanced sensitivity compared with traditional assay formats.

Materials and Methods

Reagents

The recombinant BACE1 protein containing a C-terminal 6xHis tag was purchased from Proteos (Kalamazoo, MI) and stored at −80 °C in small aliquots until use. All small-molecule BACE inhibitors were synthesized in house, and NHS-PEG_(n)-biotin was purchased from Thermo Scientific (Rockford, IL). The NHS ester (within NHS-PEG_(n)-biotin) reacted with primary amine groups engineered in the BACE inhibitors to generate biotinylated probes, which were purified by high-performance liquid chromatography to >98% purity. The AlphaScreen histidine detection kit containing nickel-chelate acceptor and streptavidin-conjugate donor beads was acquired from PerkinElmer (Waltham, MA). The peptide substrate (Biotin-GLTNIKTEEISEISYEVEFRC [Oregon Green] KK-OH) used in the FP-BACE assay was custom made by Anaspec (Fremont, CA). Bovine serum albumin (BSA), Tween-20, streptavidin, and all other reagents were obtained from Sigma-Aldrich (St. Louis, MO). Polypropylene 96-well and polystyrene 384- and 1536-well microplates were purchased from Corning (Corning, NY).

AlphaScreen Binding Assay Development and Optimization

For initial assay characterization, BACE-6xHis (50 nM) and biotinylated probes (titrated from 10 nM to 10 µM) were mixed together in the assay buffer, phosphate-buffered saline (PBS) containing 150 mM NaCl, pH 7.4, 0.01% Tween-20, and 0.05% BSA in white 384-well low-volume microplates. Reactions were incubated for 1 h at 25 °C, nickel-chelate acceptor and streptavidin donor beads were then added simultaneously to a final concentration of 20 µg/mL each and incubated for 2 h. The plates were subsequently read on the EnVision multilabel reader (PerkinElmer) using the default settings (a 640 nm dichroic mirror for the excitation light and a 570 ± 50 nm cutoff filter for emission). To confirm the specificity of positive assay signals generated by the interactions between BACE-6xHis and biotinylated probes, we performed competition experiments with a high concentration of parent inhibitors. BACE-6xHis (50 nM) was preincubated with the parent inhibitors (100 µM) for 15 min, followed by the addition of respective biotinylated probes (1 µM). After incubation at 25 °C for 1 h, acceptor and donor beads were then added to a final concentration of 20 µg/mL and read on the EnVision using the AlphaScreen protocol after a 2 h incubation.

The assay buffer was further optimized to increase the S/B ratio. The concentrations of BACE-6xHis and the chosen biotinylated probe were maintained at 50 nM and 1 µM, respectively, and a series of experiments were performed on three different reagents that could affect assay signal: pH (6.5–7.4), NaCl (25–150 mM), and Tween-20 (0.001%–0.05%). In addition, DMSO was separately titrated in the optimized buffer (PBS with 150 mM NaCl, pH 7.0, 0.01% Tween-20, and 0.05% BSA) to examine its effect on assay signal. To confirm that the optimized assay buffer delivered a robust signal and low background, we performed further titration of BACE-6xHis and the chosen biotinylated probe. Selected for final assay validation and inhibitor screening were 50 nM BACE-6xHis and a 250 nM biotinylated probe.

Assay Automation and Miniaturization

The 384-well assay was performed using Matrix electronic multichannel pipettes and/or a Multidrop Combi reagent dispenser (Thermo Scientific, Hudson, NH). Four microliters of test compounds in 20% DMSO, equivalent DMSO controls (maximum signal), or 2 mM EDTA controls (maximum signal) were first added to the plates. Four microliters of BACE-6xHis was then added and followed by 4 µL of the biotinylated probe. The plates were incubated for 1 h at 25 °C, then 4 µL of a mixture of donor and acceptor beads was added to a final concentration of 20 µg/mL each and a final volume of 16 µL. The plates were incubated for 2 h at 25 °C before reading. The interday and interplate variation of the automated assay were examined over the course of 3 d by running two identical 384-well plates per day, in which the minimum and maximum signal control wells were positioned in alternating columns. We further miniaturized the assay to a 1536-well format by proportionally reducing the total assay volume in half to 8 µL.

Quantitative HTS

The library used in these studies represents an internal collection of approximately 525 000 compounds, consisting of both commercially available and Elan proprietary entities.

For the screen, the screening collection was transferred to 1536-well plates (Corning, cat No. 3725). Columns 5 to 44 contained compounds to be tested, and columns 1 to 4 and 45 to 48 contained negative (DMSO) and positive (5 µM compound) controls. Compounds were screened at a final concentration of 500 µM in 5% DMSO. Primary screening consisted of single-point screening, followed by hit confirmation in triplicate, as well as artifact screening to remove AlphaScreen-specific false-positives. Relative fluorescent units were normalized to the control compound to calculate a percentage of control (%POC) for each compound, using standard analysis calculations. Compounds were considered hits when the %POC was greater than the mean + 3 median absolute deviation (MAD).

The primary HTS hits were counterscreened in a 1536-well assay to identify false-positive hits that interfere with the AlphaScreen detection method. The compounds were plated as described above, and biotin-6xHis peptide (PerkinElmer) was then added to a concentration of 5 nM and incubated for 30 min at 25 °C prior to the addition of detection beads at 20 µg/mL. The reactions were incubated for 2 h at 25 °C before being read on the EnVision. The maximum control wells contained an equivalent volume of DMSO, whereas the minimum control wells had 0.5 mM EDTA.

FP-Based BACE Assay

An FP-based BACE was employed to determine the functional potency of BACE inhibitors in black 384-well microplates. Briefly, BACE-6xHis (0.1 nM final concentration) was preincubated with serially diluted inhibitors or DMSO control (5% final) in 20 µL assay buffer (100 mM sodium acetate, pH 4.5, and 0.001% Tween-20) for 30 min at 25 °C. The reaction was initiated with the addition of 10 µL of a biotinylated peptide substrate (450 nM in assay buffer) and incubated at 37 °C for 3 h. The reaction was subsequently quenched with the addition of 30 µL streptavidin (1.5 µM in a pH-neutral assay buffer). After incubation for 15 min at 25 °C, the plates were read on the EnVision reader.

Data Analysis

All experimental data were analyzed and plotted using Prism software (GraphPad Software, San Diego, CA). Curve fitting for IC₅₀ values was done using four-parameter fits with variable slopes. The quality and robustness of the assay were determined by analysis of the Z′ factor. ²⁸

Results

Characterization of BACE-Probe Interaction

Historically, small-molecule BACE inhibitors have been discovered primarily through HTS campaigns using functional biochemical assays and more recently through fragment screening and structure-guided drug design approaches.^23–26 However, PK and safety issues as well as appropriate patient stratification have presented challenges toward advancing BACE inhibitors to late-stage clinical testing. Here, we investigated the use of an AlphaScreen binding assay to discover novel low-affinity BACE inhibitors. AlphaScreen is designed to measure the proximity of donor and acceptor beads conjugated to biomolecules of interest. When excited at 680 nm, the donor beads convert ambient oxygen to singlet oxygen, which diffuses out and comes into contact with the acceptor beads within 200 nm distance. The singlet oxygen then initiates a cascade of chemiluminescent reactions in the acceptor beads that emits light in the 520 to 620 nm range. Each donor bead can produce ~60 000 singlet-oxygen molecules per second, resulting in tremendous amplification of the binding signals. Because of the avidity of the AlphaScreen beads, binding partners can appear to have a higher affinity for each other, which is not uncommon with bead-based assay technologies. This feature makes direct binding kinetics measurement unsuitable; however, it renders AlphaScreen an ideal assay platform for capturing lower-affinity (i.e., in the micromolar range) molecular interactions.

Taking advantage of this unique feature of the AlphaScreen assays, we investigated the binding of the BACE1 protein to a panel of small-molecule probes derived from known BACE1 inhibitors ( Table 1 ). Probes were synthesized as described in Supplementary Methods, in which small- molecule BACE inhibitors of the hydroxyethylamine (HEA), hydantoin (S1), and sulfamide (S2–S3) classes were functionalized with a biotin PEG linker of varying lengths ( Fig. 1A ). BACE was obtained in >95% purity by nickel affinity and size exclusion chromatography. A C-terminal 6x His tag was engineered and retained for binding to the nickel-chelate acceptor beads. The streptavidin-conjugated donor beads interact with the biotin moiety of the probes. The formation of a complete detection complex, which contains BACE1, probe, and the donor and acceptor beads, should generate an AlphaScreen assay signal ( Fig. 1B ).

Table 1.

Profile summary of the 11 BACE probes generated and tested.

Probe	Scaffold	Linker	Signal-to-Background Ratio
1	HEA	PEG4	13
2	HEA	PEG12	122
3	HEA	PEG3400	189
4	S1	PEG4	1
5	S1	PEG12	72
6	S2	PEG4	1
7	S2	PEG12	8
8	S2	PEG3400	78
9	S3	PEG4	20
10	S3	PEG12	143
11	S3	PEG3400	278

HEA = hydroxyethylamine.

Figure 1.

AlphaScreen binding assay platform for the discovery of novel BACE inhibitors. (A) Schematic of a biotinylated small-molecule inhibitor probe. The NHS-(PEG)n-biotin molecule used to generate assay probes is shown at the bottom. The (PEG)n linker is used to reduce potential steric hindrance upon binding, increase flexibility, and thus facilitate the interactions of the small molecule in the active site. (B) Principle of the AlphaScreen binding assay. Interaction of BACE-6xHis and a biotinylated probe brings the donor and acceptor beads in close proximity. The donor beads generate singlet-oxygen molecules upon activation, which travel up to 200 nm and interact with the acceptor beads, eliciting a cascade of chemiluminescent reactions to generate an AlphaScreen signal. The presence of BACE inhibitors can lead to reduction in the AlphaScreen signal by competing with the biotinylated probe.

In our initial proof-of-concept study, we examined the binding profile of three probes (i.e., with viable PEG linkers) derived from a single high-affinity (100 nM) BACE1 inhibitor that belongs to the HEA class ( Fig. 2A ). We performed a titration for all three probes with concentrations ranging from 10 nM to 10 µM, while maintaining the BACE1 concentration constant at 50 nM. As shown in Figure 2A , positive binding events were detected for all three probes, with maximal signals observed at 1 µM for probe 1 and 2 and 0.1 µM for probe 3. The latter also generated the highest signal with an average S/B ratio of 190. In addition, a prominent hook effect was observed ( Fig. 2A ). This phenomenon is fairly common in homogeneous assays, in which an excess of individual components can result in unproductive binding that significantly decreases assay signal and/or increases background. To rule out the possibility that the observed binding events might be due to nonspecific interaction between the probes and BACE protein, we conducted a competition experiment in the presence of a saturating concentration of parent inhibitor or DMSO control. The robust signals generated from both HEA-derived probes were reduced to background level by the parent inhibitor, thus confirming the specificity of the BACE-probe interaction ( Fig. 2B ). Next, we expanded the same experimental paradigm to additional probes derived from two different classes of low-affinity (2–20 µM potency) BACE inhibitors (hydantoin S1 and sulfamides S2–S3; Table 1 ). In a titration study similar to the one described above, we observed positive signals generated by one or more probes belonging to each of these three chemical classes ( Fig. 2C ). It is important to point out that robust interaction was observed at probe concentrations significantly lower than the reported affinity values. This is another demonstration of the ability of AlphaScreen, due to a bead-mediated avidity effect, to capture low-affinity molecular interactions difficult to detect in other assay platforms. The specificity of these interactions was subsequently confirmed in a competition study with their respective parent inhibitors ( Fig. 2B ). In addition, to exclude the possibility of nonspecific probe binding to BACE1 accounting for the signal, full displacement curves of free parent to probes 1, 2, 3, 5, 7, and 8 are shown in Supplementary Figure 1A–C. Of relevance, the IC₅₀ values for displacement of the probe by the free parent compound were similar to the IC₅₀ values of the free parent compounds in the FP-BACE activity assay (Supplementary Fig. 1D and E). Taken together, these experiments help rule out the possibility that the signal that we observed was a result of nonspecific binding. Through these studies, we also noted a significant impact of the linker size on assay signal. We observed that within each of the four probe sets examined, a longer linker size consistently resulted in higher assay signal (i.e., PEG-3400 > PEG-12 > PEG-4) as shown in Figure 3 . In the absence of such linkers, the binding signals were negligible for all probes (data not shown), further highlighting the importance of appropriately sized linker for the optimal probe design. It is unlikely that the trend of longer linker size leading to higher S/B is a result of nonspecific interactions of probe to BACE1, as free parent compound exhibited a similar IC₅₀ value in displacing probes linked to PEG-4, PEG-12, and PEG-3400 (Supplementary Fig. 1A and C).

Figure 2.

Evaluating probes from four different structure classes in the AlphaScreen binding assay. (A) The profile of three hydroxyethylamine (HEA)–derived probes. Each probe was tested at four different concentrations 0.01 µM (dotted), 0.1 µM (solid), 1 µM (hatched), and 10 µM clear, with the best signal obtained at 0.1 or 1 µM. Shown on the right is a structure of the parent HEA BACE inhibitor. (B) Competition study to assess specificity of the AlphaScreen binding assay. Six representative probes from four structure classes were tested at optimal concentrations, either in the presence (black) or absence (gray) of their respective parent inhibitors. The addition of 100 µM parent inhibitors reduced the signal-to-background level. (C) The profile of eight additional probes derived from three classes of low-affinity BACE inhibitors. Each probe was tested at the four indicated concentrations, 0.01 µM (dotted), 0.1 µM (solid), 1 µM (hatched), and 10 µM clear, with the best signal obtained at 1 µM.

Figure 3.

The effect of probe linker size on assay signal. Data generated from the probe-profiling study were further analyzed to illustrate the impact of linker size on assay performance. The data were collected at 1 µM probe concentration, with the exception of probe 3 (0.1 µM). Data are expressed as signal-to-background ratio, and the background is defined by negative controls treated with 0.5 mM of EDTA. Within each of the four structure classes tested, a longer linker size resulted in a higher assay signal in which PEG-3400 (black) > PEG-12 (hatched) > PEG-4 (white). Because of technical difficulties, a PEG3400-linked S1 probe could not be generated and purified for testing.

Assay Buffer Optimization for HTS

Among the three classes of low-affinity probes, we were most interested in exploring the pocket occupied by the S1 probes and discovering additional novel chemical entities that target the same binding site. To achieve that, a robust binding assay using probe 5 (S1 class with a PEG-12 linker) needs to be further optimized and validated before an HTS campaign can take place. First, we examined the impact of various buffer components on assay signals, including pH, Tween-20, NaCl, and BSA. As shown in Supplementary Figure 2A, assay signal was significantly diminished after reducing pH to 6.5. This was caused at least partially by the reduced affinity between the 6xHis tag and the nickel- chelate acceptor beads at acidic pH, and similar signal decreases were also observed for other probes (data not shown). Furthermore, pH alteration may also result in changes in the BACE binding pocket and/or the protonation status of the probe, which in turn could lead to reduced binding and assay signal. In contrast, minimal changes in S/B ratio were observed when the concentrations of Tween-20 (0.001%~0.05%) and NaCl (25~150 mM) were altered (Supplementary Fig. 2B and C). Variation in BSA concentration (0.01%~0.1%) also caused no significant change in assay signal (data not shown). Although these three buffer components did not appear to have a major impact on assay signal, we still selected 150 mM NaCl, 0.01% Tween-20, and 0.05% BSA to reduce potential nonspecific bindings and compound aggregation. In a subsequent study, DMSO was titrated from 0% to 10%, and no substantial reduction of signal was observed in the presence of up to 5% DMSO (Supplementary Fig. 2D). The excellent DMSO tolerability renders this assay suitable for high-concentration (i.e., 100 µM) screening campaigns where 5% DMSO is needed to help minimize solubility issues of library compounds.

Optimizing BACE and Probe Concentrations

The optimized assay buffer was PBS pH 7.0, containing 150 mM NaCl, 0.01% Tween-20, and 0.05% BSA. To achieve the highest S/B ratio, we performed additional titrations of BACE1 and probe 5 in the presence of 5% DMSO. We first held the probe constant at 1 µM and tested a wide range of BACE1 concentrations. A rapid signal decline was detected when the BACE1 concentration was reduced from 100 nM to 0.05 nM, resulting in a complete loss of assay window with less than 10 nM BACE ( Fig. 4A ). We therefore chose a moderate BACE concentration of 50 nM for the subsequent probe titration. As shown in Figure 4B , a dose-dependent decrease of signal was also observed, with probe concentration ranging from 1 µM to 31 nM. Based on these titration data, we chose the combination of 50 nM BACE and 250 nM probe 5 as our final assay condition, balancing the need to conserve reagents while maintaining a robust assay window (~20-fold). To further evaluate the robustness and reliability of the assay for HTS, we studied interplate and interday variations with replicate plates containing both the minimum and maximum signal controls. The Z′ values were 0.81, 0.78, and 0.77 for each respective run, with interday variability in the maximum controls less than 10% (Supplementary Fig. 3). All together, these data suggest a high-quality assay suitable for HTS. Furthermore, smooth and reproducible dose-response curves were obtained for three model inhibitors of the HEA, benzamide, and sulfamide classes across a wide spectrum of potencies ( Fig. 5A ).

Figure 4.

Titration of BACE and probe 5 in optimized assay buffer. (A) BACE titration in the presence of 1 µM probe 5. A rapid signal decline was detected when the BACE concentration was reduced from 100 nM to 0.41 nM. (B) Probe 5 titration in the presence of 50 nM BACE. A dose-dependent decrease of signal was also observed when lowering probe concentration from 1 µM to 31 nM. Based on these titration data, we chose the combination of 50 nM BACE and 250 nM probe 5 as our final assay conditions.

Figure 5.

Characterization and validation of the AlphaScreen binding assay with BACE1 inhibitors. (A) Eight-point dose-response titration was conducted for BACE inhibitors of the hydroxyethylamine (HEA; chemical structure shown in Fig. 2A ), benzamide, and sulfamide classes, in the presence of 5% DMSO. Curve fitting was done with a four-parameter algorithm, and the IC₅₀ values were determined to be 0.02 µM, 1.83 µM, and 17.6 µM, respectively. (B) Correlation between potency values generated from AlphaScreen binding assay and fluorescent polarization BACE activity assay. A panel of 11 internal BACE inhibitors covering five different structure classes was tested in both assays. The respective IC₅₀ values were plotted on a log scale and analyzed by linear regression. A robust correlation between two assays was observed, with an excellent R² value of 0.94.

Assay Validation

Finally, to validate this novel binding assay pharmacologically, we tested a panel of 11 internal BACE inhibitors covering five different structure classes and compared their IC₅₀ values to those obtained from an FP-based activity assay. As shown in Supplementary Table 1, the potency rank orders determined by the two independent assays were largely consistent with each other. However, on average, the IC₅₀ values derived from the AlphaScreen binding assay were about one order of magnitude more potent. This is also likely attributed to the avidity effect inherent to this bead-based assay platform as described previously. In addition, quantitative analysis confirmed a robust correlation between two assays, with an excellent R² of 0.94 ( Fig. 5B ).

HTS of Internal Library and Hit Follow-up

We screened more than 525 000 compounds using a validated AlphaScreen assay miniaturized to a 1536-well format to identify BACE1 inhibitors. Because of the high screening concentration (500 µM), a high hit rate was predicted. Initial screening was followed in parallel by primary hit confirmation in triplicate and artifact counterscreening using percentage inhibition >24% (mean + 3 MAD) for hit calling, which identified 26 144 hits, yielding a final hit rate of 4.97% (Supplementary Fig. 4). The artifact assay used standard Biotin-His bead coupling, which can infer the presence of both biotin mimetics and singlet oxygen quenchers. The Z′ factor was calculated per plate for the screen (481 plates), where the overall (averaged) Z′ value was 0.58 ± 0.06. The 26 144 confirmed primary hits were run in dose-response curve format where 3477 compounds were selected for further analysis (solubility and Biacore) based on potency, quality of the dose-response curve, chemical novelty, and good calculated Lipinski properties. The HTS and hit confirmation workflow is highlighted in Supplementary Figure 5.

Biacore positive hits were then shuttled through high-throughput crystallography. Compound 12 (Supplementary Fig. 6) registered a compelling Biacore sensorgram trace with a calculated K_d of 62 µM and formed a co-crystal structure with BACE. ²⁹ Lead optimization efforts on compound 1 are highlighted elsewhere and are beyond the scope of this report. ²⁹ In addition, as expected, a number of compounds based on the HEA scaffold emerged as hits from the HTS campaign, which served to further validate the assay and screen (data not shown).

Discussion

The binding assay and subsequent screen described in this article differ in many respects from that of previous assays used to screen and identify BACE1 inhibitors. In this assay, a BACE construct consisting of the BACE1 extracellular domain fused with a C-terminal 6x-histidine tag was coupled to nickel-chelate–coated acceptor beads. BACE1 inhibitors linked to biotin by PEG linkers of varying lengths (probes) were bound to streptavidin-coated donor beads. The interaction of the probe with the active site of BACE1 will bring the donor and acceptor beads into close proximity and thus yield a strong signal. Free compounds that will bind in the active site of BACE1 and displace the probe will abrogate the signal, forming the basis of our assay. Probes were based on three different classes, an HEA, a hydantoin, and two variants of sulfamides. For our actual HTS, probe 5, which was based on the hydantoin scaffold, was chosen as it does not displace water from the catalytic aspartic acids of BACE. This was important as energetically it is unfavorable to displace water from the catalytic aspartic acids, and this type of binding mode will usually increase the number of hydrogen bond donors of the small molecules, which can be unfavorable from a PK standpoint as well. HEA-based compounds will displace the water from the catalytic aspartic acid compounds, which could contribute in part to the poor PK characteristics that have been observed with some compounds in this class. However, it should be emphasized that this assay itself will identify compounds of different binding modes regardless of whether they displace water from the active site or not.

The optimal pH for measuring BACE1 activity using the extracellular domain construct is pH 4.5, whereas at a pH greater than 5.5, only minimal activity was observed (Supplementary Fig. 7). The pH in the binding assay was 7.0 and was chosen because below this pH, the interaction of 6x-histidine–tagged BACE1 with the nickel-chelate is disrupted. However, most BACE1 screens historically were performed at acidic pH values and usually at pH 4.5. This was due to a combination of factors. First, BACE1 activity is optimal in acidic pH, and the compartments where BACE resides are also thought to be acidic. Although the assay described here was performed at pH 7, the higher pH does allow for the possibility of identifying different scaffolds that may not have been identified if the screen were performed at acidic pH. Recently, an assay using full-length BACE1 screened at pH 6 was developed. ³⁰ BACE1 activity could be regulated by the transmembrane and cytoplasmic domains, and BACE1 does reside in Golgi and endocytic compartments that have a pH ~6. ³⁰ Thus, although the activity of the BACE1 extracellular domain is most active at pH 4.5, it may be that higher pH buffers better represent the physiological pH for BACE1.

A major advantage of the AlphaScreen assay described in this article compared with other assay formats is the enhanced sensitivity that is observed with this format, as shown in Supplementary Table 1. We observed that AlphaScreen IC₅₀ values were approximately 10-fold more potent than IC₅₀ values obtained using the FP-BACE assay. It should be emphasized that the enhanced sensitivity of the AlphaScreen binding assay compared with the FP assay was not a result of the difference in pH at which the two assays were performed (pH 4.5 versus pH 7.0, data not shown). The enhanced sensitivity (~10×) that we observed is consistent with what was reported using the AlphaScreen platform compared with other assay formats for other targets as well.^31,32 The donor and acceptor beads contain multiple binding sites, and thus the enhanced sensitivity of this platform is attributed to a bead avidity effect. The bead avidity effect allows for weak interactions (in the 50–100 µM range) to be successfully measured using this format. We also observed that increased linker length yielded higher S/B ratios for all probes tested ( Fig. 3 ). This could be a result of steric hindrance that occurs with shorter linkers, where the longer linker will allow greater flexibility and mobility for the small molecule, which will facilitate binding to the active site. In addition, longer linkers could allow for a larger volume of bead packing. For example, a donor bead can be bound/surrounded by a larger number of acceptor beads when longer linkers are used, allowing for the formation of a meshwork of bound beads and potentially an increased S/B. It is possible that both mechanisms contribute to the observation that longer linker size increases S/B. The high S/B values for this assay allowed for flexibility in adjusting assay parameters such as BACE1 and probe concentrations.

The AlphaScreen format can tolerate higher DMSO concentrations, thus enabling the screening concentration (500 µM) used in this HTS campaign. We observed a relatively high hit rate of 4.97% (Supplementary Fig. 4) due in part to the high screening concentration and the enhanced sensitivity of the assay compared with other formats. A high screening concentration was used to increase the chance of identifying weak binders that could be optimized for potency in the hit-to-lead phase of drug discovery.

The binding assay described for BACE1 can be extrapolated to other drug targets as well and could be considered a platform assay. A small-molecule binder can be functionalized with a biotin-PEG of varying lengths and coupled to streptavidin-coated beads. The target protein of interest can then be coupled to donor beads, and conditions of buffers and target concentrations can be optimized. Consistent with our results and those of others, this approach could enhance assay sensitivity by an order of magnitude, which could make screening possible for many targets that have nonoptimal Z′ values using more traditional assay approaches. Also, as this assay is a binding assay, it is amenable for targets in which functional activity cannot be measured or intrinsic enzyme activity is low in vitro.