Compounds that bind APP and inhibit Abeta processing in vitro suggest a novel approach to Alzheimer disease therapeutics.

Extracellular deposits of aggregated amyloid-beta (Abeta) peptides are a hallmark of Alzheimer disease; thus, inhibition of Abeta production and/or aggregation is an appealing strategy to thwart the onset and progression of this disease. The release of Abeta requires processing of the amyloid precursor protein (APP) by both beta- and gamma-secretase. Using an assay that incorporates full-length recombinant APP as a substrate for beta-secretase (BACE), we have identified a series of compounds that inhibit APP processing, but do not affect the cleavage of peptide substrates by BACE1. These molecules also inhibit the processing of APP and Abeta by BACE2 and selectively inhibit the production of Abeta(42) species by gamma-secretase in assays using CTF99. The compounds bind directly to APP, likely within the Abeta domain, and therefore, unlike previously described inhibitors of the secretase enzymes, their mechanism of action is mediated through APP. These studies demonstrate that APP binding agents can affect its processing through multiple pathways, providing proof of concept for novel strategies aimed at selectively modulating Abeta production.

loid precursor protein (APP) (4). Production of A␤ has been linked genetically to AD. Recent studies (5)(6)(7) have shown that A␤ multimerizes into neurotoxic oligomers that target the synapse, leading to disruption of long term potentiaton, cell death, and memory loss. These oligomers then coalesce into fibrils, ultimately forming the plaques originally used to characterize the disease (5)(6)(7).
To produce A␤, APP, a type I single-transmembrane glycoprotein, is sequentially cleaved by two aspartyl proteases, ␤and ␥-secretase. The initial cleavage is mediated by ␤-secretase (BACE1) on the luminal side of the endosomal membrane or at the cell surface (8). BACE1 cleavage generates a secreted Nterminal product (sAPP␤) and a transmembrane N-terminal product (CTF99). Processing of CTF99 within the transmembrane domain by the ␥-secretase enzyme complex generates the C-terminal end of A␤ and results in its release (9). The ␥-secretase enzyme is known to cleave CTF99 at any of several discrete positions, generating A␤ peptides that range in length from 38 to 42 amino acids. Although A␤ 40 is the predominant species produced both in vitro and in vivo, A␤ 42 is the more aggregenic species and the major component of amyloid plaques. The release of A␤ 42 and subsequent aggregation into insoluble fibrils are believed to lead to the formation of the dense plaques characteristic of AD (10).
Drug discovery efforts aimed at preventing the accumulation of A␤ have been directed toward inhibition of A␤ production and prevention of A␤ aggregation (11,12) as well as enhancing A␤ clearance (13)(14)(15). Preventing A␤ production has focused on targeting the enzymes involved in processing APP. Transition state-based inhibitors of BACE1 have been described with in vitro activity, as well as activity in cell culture (16 -25), but data on in vivo efficacy is limited. Although numerous ␥-secretase inhibitors have been shown to block A␤ processing in vitro and in vivo (26 -30), ␥-secretase plays an essential role in the processing of other, disparate targets, and mechanism-based toxicity is a potential concern (31-34). In contrast, molecules known as ␥-secretase modulators selectively inhibit ␥-secretase processing of APP at the 42 cleavage site and may have a reduced effect on processing other substrates; however, this has not been extensively studied to date.
Herein we describe a novel series of benzofuran-containing compounds that inhibit APP processing by a previously undescribed mechanism of action. Although the compounds were initially identified in a search for BACE1 inhibitors, they inhibit BACE1, BACE2, and ␥-secretase-mediated cleavage of APP by binding APP within the A␤ domain of the protein. Some of these inhibitors preferentially affect ␥-secretase processing at the 42 cleavage site; however, their mechanism of action is distinct from ␥-secretase modulators. These studies provide evidence that targeting APP may present opportunities to identify small molecules that affect APP processing and provide a novel strategy to develop chemotherapeutic agents for AD.

EXPERIMENTAL PROCEDURES
Materials-BACE1 protein was produced from a recombinant baculovirus expression system, and BACE2 protein was produced from HEK293T cells as described in Refs. 35 and 36. ␥-Secretase was purified from HeLa S3 cells as described by Li et al. (37).
In Vitro Secretase Assays-BACE1 and BACE2 cleavage of peptide substrates and A␤ was carried out as described in Refs. 23, 36, and 38. Mass spectrometric detection of BACE2 cleavage of A␤ by surfaceenhanced laser desorption/ionization time-of-flight mass spectrometry (Ciphergen Biosystems) was done as described in Ref. 38. Peptides assayed for BACE1 and BACE2 cleavage are described in the legend to Fig. 3. To assay cleavage of APP FL , purified biotinylated maltose-binding protein (MBP)-fused, bacterially expressed, full-length APP containing an enhanced BACE1 cleavage sequence was incubated with compound and BACE1 protein for 1 h at 37°C or BACE2 protein for 30 min at 37°C. The cleavage product was detected using streptavidinconjugated donor AlphaScreen beads (20 g/ml, PerkinElmer Life Sciences) and polyclonal anti-sAPP␤ NF (39) bound to protein A-coated acceptor AlphaScreen beads (20 g/ml, PerkinElmer Life Sciences) in 0.1% bovine serum albumin in phosphate-buffered saline. Plates were incubated overnight in the dark and then read using an Alpha-Fusion HT instrument (PerkinElmer Life Sciences).
Cell-based Assay for ␥-Secretase Activity-H4 neuroglioma cells were stably transfected with the SP␤A4CTF expression vector (27). The cells were treated with compound for 16 -20 h, and conditioned media were harvested. A␤ X-40 and X-42 in conditioned media were assayed via electrochemiluminescence using an Origen 1.5 Analyzer (Igen), essentially as described in Ref. 37, except using biotinylated 6E10 (1 g/ml) (Signet Laboratories) and ruthenylated G2-10 (0.375 g/ml) (41) for detection of A␤ X-40 and biotinylated 6E10 (1 g/ml) and ruthenylated G2-11 (0.375 g/ml) (41) for detection of A␤ X-42. Surface Plasmon Resonance Binding Assay-Sensorgrams for binding of compounds 1-3 onto MBP-APP FL were generated using a Biacore S51 instrument. Equal amounts of recombinant MBP-APP FL and MBP-sAPP␤ NF proteins were immobilized on Spot 1 or Spot 2 in Flow cell 1 of a Sensor Chip SA (Biacore), respectively. The compounds were dissolved in 5% Me 2 SO in phosphate-buffered saline. The sensorgrams were recorded at a flow rate of 30 l/min at 30°C. The compound was injected for 30 s. Following compound injection, 5% Me 2 SO/phosphatebuffered saline buffer was injected over the chip for another 30 s. A solvent correction cycle was run for each sample. Specific binding to MBP-APP FL was calculated as binding to MBP-APP FL Ϫ MBP-sAPP␤ NF binding.

Identification of a Novel Group of Inhibitors That Inhibit BACE1 Cleavage of Full-length APP but Not Peptide Sub-
strates-In the process of performing a high throughput screen for BACE1 inhibitors using a purified recombinant full-length APP substrate, we identified the novel benzofuran-containing compound 1 shown in Table I. The activity of compound 1 and a panel of related analogs (compounds 2-5, Table I) was evaluated using an AlphaScreen assay for BACE1 cleavage of APP (see "Experimental Procedures"). This assay is a homogeneous proximity-based method that uses donor beads binding both cleaved and uncleaved APP and acceptor beads coated with an antibody that binds only the cleaved APP product. Excitation of the donor bead triggers the release of singlet oxygen, which then stimulates the emission of light in the acceptor. With the   Table I). The effect of these compounds on BACE1 cleavage of APP was confirmed by Western blot analysis of the reaction products. For compounds 1-4 the relative potencies for APP processing inhibition did not vary significantly between the AlphaScreen and Western assays (Fig. 1, A and C, and data not shown). However, compound 5 did not inhibit APP cleavage when assayed via Western blot, and subsequent analyses suggested compound 5 may have disrupted detection in the Al-phaScreen assay by triggering the precipitation of substrate (data not shown). The specificity of this class of compounds for inhibition of APP processing was established by reviewing data from over 70 assays run at Ն5 M that included compound 2 (the most potent compound). Compound 2 was not active in a wide variety of binding, cell-based, or enzymatic activity screens (data not shown).
Although compounds 1-4 reproducibly blocked BACE1 cleavage of APP, the benzofuran analogs did not inhibit BACE1 processing of peptide substrates, such as P5-P5Ј (Table I, Fig.  1A), and extended substrates encompassing P21-P10Ј (see Fig.  3). In contrast, a well characterized transition state mimic inhibitor of BACE1, Statine-Val, blocked BACE1 cleavage of the peptide and APP FL substrates ( Fig. 1B) with comparable potencies, suggesting the inability of the benzofurans to block cleavage of the P5-P5Ј peptide may be related to a functionally distinct mechanism of action.
Compounds 1-4 Exhibit Substrate-dependent Activity on BACE1 and -2 as Well as ␥-Secretase-To further examine the substrate-dependent activity of these inhibitors, we evaluated their effect on the closely related protease BACE2. Although BACE2 also processes APP at the ␤-site, it preferentially cleaves APP at two sites positioned at 19 and 20 amino acids within A␤ (42), as well as at position 34 (38).
Inhibition of BACE2 cleavage at the ␤-site was first tested using full-length APP and the P5-P5Ј peptide as described for BACE1 (Table II). Compounds 1-3 inhibited BACE2 cleavage of full-length APP with IC 50 values that were similar to inhibition of BACE1 cleavage. Compound 4 weakly inhibited fulllength APP cleavage, with an IC 50 value of ϳ100 M. Again, these compounds did not block BACE2 cleavage of the P5-P5Ј peptide, although the transition state analog inhibitor was effective.
The effect of compounds 1-4 on BACE2 cleavage of 〈␤ was tested by incubating 〈␤ (1-40) with BACE2 in the presence or absence of a concentration of 100 M for each compound and resolving the cleavage products by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry. Cleavage of 〈␤ (1-40) at position 34 was reduced by all four compounds ( Fig. 2 and Table II). Surprisingly, the effect on BACE2 processing at positions 19 and 20 in A␤ was compound-specific. Compound 3 had little effect on BACE2 cleavage at positions 19 and 20, whereas compounds 2 and 4 blocked cleavage by ϳ50%. In contrast, compound 1 substantially increased BACE2 cleav-age at these positions ( Fig. 2 and Table II). The differential effects of the benzofurans on BACE1 and BACE2 activity with respect to peptide versus protein substrates and BACE2 activity with regard to processing distinct sites within A␤ is novel for compounds that inhibit BACE1/2 activity.
A potential explanation for the distinctive substrate dependence exhibited by benzofurans 1-4 is that the compounds block cleavage of APP and 〈␤ by binding to and interfering with the substrate. In this case, the compounds might also block ␥-secretase cleavage of substrates containing 〈␤. Compounds 1-5 were therefore evaluated for inhibition of ␥-secretase cleavage of both CTF99 and a 28-mer peptide encompassing the Cterminal domain of 〈␤ extending from residues 28 through 55 of CTF99 (Table III and Fig. 3). Compounds 1-3 inhibited ␥-secretase cleavage of CTF99 at the X-40 position with IC 50 values similar to those observed for inhibition of BACE1 and BACE2 cleavage of APP. (Cleavage of CTF99 at the X-42 position was too low to be detected). Compounds 1-3 also inhibited ␥-secretase cleavage of the 28-mer peptide substrate, and the IC 50 values obtained in assays using the 28-mer peptide were comparable with those obtained with CTF99. Compound 4 inhibited ␥-secretase cleavage of the CTF99 but did not inhibit  cleavage of the peptide substrate. Compound 5 did not have an effect on cleavage of CTF99 or the peptide substrate.

Compounds 1-3 Bind APP within the A␤ Domain-
The benzofurans 1-4 inhibited processing of a variety of related APP substrates by structurally distinct aspartyl proteases. Substrates in which cleavage was affected had in common the C-terminal domain of 〈␤. Taken together, the data suggest these compounds may interfere with processing by interacting with the substrates rather than the enzymes, and this interaction is likely mediated through the C-terminal portion of the 〈␤ domain (Fig. 3). Although we attempted to explore the potential for interaction with APP using a variety of biophysical methods, such as NMR and nitrocellulose filter binding, the limited solubility of these compounds was problematic for many solution-based methods (data not shown).
We therefore used surface plasmon resonance to assess binding to APP using the Biacore S51 instrument. Biotinylated APP FL and biotinylated sAPP␤ were immobilized onto a streptavidin-coated sensor chip. Compounds 1-3 were analyzed at various concentrations, and the binding response at each concentration was monitored for both proteins. Representative data for binding to APP FL are shown for compound 2 in Fig. 4. For compound 2, binding to APP FL was observed to occur in a dose-dependent manner (Fig. 4A), whereas no interaction was observed between the BACE inhibitor, Statine-Val, and APP (Fig. 4B). Compound 2 did not bind to sAPP␤, consistent with the results from enzyme assays, which suggest binding to APP FL occurs within the A␤ domain (data not shown). Finally, the K d determined in this analysis (43 M) was comparable with the IC 50 values obtained for this compound for inhibition of APP processing by BACE1 and ␥-secretase (9 -29 M) (Fig.  4B). APP binding was also observed with compounds 1 and 3 (data not shown).
The Benzofuran Analogs Inhibit CTF99 Cleavage in Cell Culture-Compounds 1-4 were tested in cell culture for their effects on APP processing by BACE1 and ␥-secretase (39). Although none of the compounds inhibited BACE1 cleavage of APP, all four analogs inhibited ␥-secretase cleavage of exogenously expressed CTF99. Interestingly, compounds 2 and 3, similar to a number of NSAIDs, had position-specific effects on ␥-secretase cleavage of CTF99 (Fig. 5B). As with the ␥-secretase modulators, production of the A␤ X-42 species of amyloid was more sensitive to treatment with the benzofuran compounds than was the A␤ X-40 species (Fig. 5). DISCUSSION We have characterized a series of benzofuran analogs that inhibit secretase-mediated APP processing by a novel mechanism. These molecules inhibit BACE1 and BACE2 processing of full-length APP but not ␤-site peptide substrates. In addition, they inhibit BACE2 processing of A␤ and ␥-secretase processing of CTF99 with comparable efficacy. Although BACE1 and ␥-secretase are both aspartyl proteases, they do not share structural similarity, and to date, there have been no other reports of compounds that block both ␤and ␥-secretasemediated proteolysis in vitro. Inhibition of both BACE1/2 and ␥-secretase cleavage, along with the substrate specificity associated with compound activity and the demonstration of binding to APP by surface plasmon resonance, suggest these molecules block APP processing, not by interacting with the processing enzymes, but by binding to the substrate. These results validate a new mechanism to interfere with APP processing and suggest a novel approach to AD therapeutics.
These compounds were able to disrupt ␥-secretase processing of both APP and CTF99 in vitro, but in cells, they prevented only ␥-secretase cleavage of CTF99, an apparent contradiction. The physical properties of the compounds used in this study made it unlikely that they would cross the plasma membrane, so one possible explanation is that ␥-secretase cleavage of CTF99 occurs at the plasma membrane, whereas full-length APP is internalized upon interaction with BACE1 and is cleaved by both BACE1 and ␥-secretase within the endosome (43). ␥-Secretase activity at the plasma membrane, including interactions between ␥-secretase and CTF99, has been described previously, supporting this assertion (8,34,45,47,48). Although other studies carried out with cell impenetrant ␥-secretase inhibitors have suggested that ␥-secretase cleavage of Notch occurs at the cell surface, while cleavage of CTF99 occurs in the endosomal compartment; however, differences in cell lines and in expression conditions between these assays and those used in this study may have resulted in differences in CTF99 trafficking that could account for these differences (48).
Indirect evidence implicating binding to APP through the C-terminal portion of the 〈␤ includes the following observations: (i) compounds 1-4 blocked BACE1 and BACE2 cleavage of APP at the ␤-site but had no effect on the processing of ␤-site peptide substrates, (ii) the compounds inhibited BACE2 cleavage of A␤ at position 34 but exhibited variable efficacy at the 19/20 site, and finally, (iii) ␥-secretase cleavage of a peptide beginning at A␤ position 28 was inhibited. Although these data allow one to infer that binding occurs within the A␤ domain proximal to the C-terminal end, it is interesting to note that these compounds inhibit cleavage of APP by BACE1 and -2 at the N terminus of A␤. Therefore, although these effects on substrate processing may be the result of direct steric interference, it is also possible that binding may induce conformational changes and prevent processing indirectly. The latter could explain the enhancement of processing by BACE2 observed at specific cleavage sites.
Interestingly, structurally distinct benzofuran analogs that dissociate fibrillar A␤ have been described previously (49,50), and the benzofuran motif in both series may play a critical role in binding to A␤. The C terminus of A␤ is known to be involved in aggregation, thus the observation that benzofuran analogs can affect A␤ aggregation is consistent with our results, suggesting that benzofurans 1-4 interact with A␤ within this region. For compound 2, the K d determined for binding to  full-length APP by surface plasmon resonance was similar to the IC 50 value obtained for BACE1 and ␥-secretase cleavage of APP, as well as BACE2 cleavage of A␤. Taken together, these results suggest the structure of the A␤ domain may not differ significantly between APP and the processed peptide.
Intriguingly, the benzofuran-containing compounds, although inhibiting cleavage at the ␤-site and at A␤ position 34, also behaved like cleavage modulators. At positions 19 and 20 of A␤, compound 1 enhanced BACE2 cleavage, although reducing BACE2 cleavage at position 34. In addition, when assayed for their effects on ␥-secretase cleavage of CTF99 in cells, compounds 2 and 3 inhibited cleavage at position 42, although having a decreased effect on cleavage at position 40. This differential effect on ␥-secretase cleavage of CTF99 in cells is similar to the reported effects of NSAIDs on ␥-secretase cleavage of APP, both in mice and in cells.
Several mechanisms for the effect of NSAIDs on 〈␤ secretion have been proposed. The ␥-secretase modulatory effects of NSAIDs have been ascribed as an indirect effect because of their inhibition of Rho (51). However, NSAIDs directly block ␥-secretase activity in cell extracts and compete with transition state inhibitors for binding to ␥-secretase, suggesting a noncompetitive allosteric binding mode (46,52,53). Therefore, it appears unlikely that the benzofurans described in this manuscript modulate APP cleavage via the same mechanism as the NSAIDs. However, NSAIDs have also been shown to bind 〈␤ fibrils (44). Although it is possible that an additional mechanism for the ␥-secretase modulatory effects of NSAIDs is mediated by the APP CTF, the structure of A␤ fibrils is distinct from the APP FL , CTF99, and soluble A␤ substrates of ␤and ␥-secretase.
Our results demonstrate that compounds that bind APP can inhibit its processing by each of the three different enzymes responsible for its metabolism in vivo. This represents a novel mechanism for inhibition of A␤ generation. Although the higher molar concentration of substrate relative to enzyme suggests that small molecule inhibitors are more effective when directed to the enzyme component of the pathway, given the potential for mechanism-based toxicity associated with ␥-secretase inhibitors and the overall difficulties in generating BACE1 inhibitors with in vivo efficacy, targeting APP may provide an alternative approach for AD drug discovery efforts. FIG. 4. Surface plasmon resonance of compound 2 binding to APP FL . A, sensorgram showing a titration of compound 2 flowing over APP FL and sAPP␤ NF . Binding to sAPP␤ NF was substantially lower than binding to APP FL and was considered to represent background binding. Data shown is binding to APP FL with sAPP␤ NF binding subtracted. The sensorgram plot depicts resonance units (RU) versus time for each concentration of compound 2. B, titrations of compound 2 and the BACE1 transition-state inhibitor Statine-Val were applied to a sensor chip spotted with APP FL and sAPP␤ NF . The specific binding of compound 2 or Statine-Val in resonance units was calculated and plotted for each concentration of compound tested.
FIG. 5. Inhibition of ␥-secretase processing of CTF99 in cells. H4 neuroglioma cells stably expressing CTF99 were treated with a titration of compounds 1-4. Conditioned media from each treatment were sampled and assayed for levels of A␤ X-40 and X-42 (where X can be any of the first five amino acids in A␤). A␤ levels were compared with vehicle-treated cells and graphed as a percentage of control. A, compound 2 was titrated in four separate experiments. Relative A␤ X-40 and X-42 levels from all four experiments were graphed together and a curve fit was generated. B, comparison of EC 50 values for inhibition of A␤ X-40 and X-42 secretion from H4 CTF99 cells. The average and standard deviation of four experiments is shown for compounds 1 and 2, two experiments for compound 4, and 1 experiment for compound 3.