Discovery of BACE1 Inhibitors for the Treatment of Alzheimer’s Disease

Alzheimer’s disease is the most common cause of dementia. According to the amyloid hypothesis, β-secretase (BACE1) is a promising molecular target for the development of anti-Alzheimer’s disease drugs. BACE1 triggers the formation of the amyloid-β (Aβ) peptides that are the main component of senile plaques in the brain of patients with Alzheimer’s disease. As BACE1 cleaves the amyloid precursor protein at the N-terminus of the Aβ domain, BACE1 inhibitors reduce the Aβ level in the brain. Previously, we designed a series of peptidic inhibitors that possessed a substrate transition-state analogue, and the structure-activity relationship of our inhibitors was evaluated, based on docking and scoring, using the docking simulation software Molecular Operating Environment (MOE). However, there was no association between the scoring values and the inhibitory activities at the P 2 position. Hence, we hypothesized that the interaction of the P 2 position of the inhibitor with the S 2 site of BACE1 was critical for the mechanism of inhibition, and we proposed the novel concept of ‘electron donor bioisostere’ for drug discovery. Using this concept, we designed potent small molecule non-peptidic BACE1 inhibitors.


Introduction
Alzheimer's disease (AD), which is the most common cause of dementia, is characterized by progressive intellectual deterioration. In 1901, Alois Alzheimer, a psychiatrist and neuropathologist, observed a 51-year-old female patient at Frankfurt Asylum. The patient showed strange behavioural symptoms and the loss of short-term memory, which was later called 'AD'. Unfortunately, the cause of AD was unclear until recently, and there have been no treatment approaches since that first report by Dr. Alzheimer over 100 years ago. Recently, the development of many drug candidates based on the amyloid hypothesis has been reported. β-secretase (BACE1; β-site amyloid precursor protein (APP) cleaving enzyme 1) is a promising molecular target for the development of anti-Alzheimer's drugs. BACE1 triggers the formation of the amyloid-β (Aβ) peptide that is the main component of the senile plaques found in the brain of AD patients. Previously, we had designed a series of peptidic inhibitors that possessed a substrate transition-state analogue, and evaluated the structure-activity relationship of our inhibitors, based on docking and scoring, using the docking simulation software Molecular Operating Environment ((MOE), Chemical Computing Group Inc., Canada).

Pathology of Alzheimer's disease
Although the cause of AD was unclear until recently, a breakthrough was obtained from the genetic study of some patients with familial AD. Certain mutations of the amyloid precursor protein (APP) or presenilin gene increased amyloid-β peptides (Aβs) in the brain, which indicated their involvement in the pathogenesis of AD [1][2][3][4]. Aβ is produced from APP by two processing enzymes, β-secretase and γ-secretase, which are potential molecular targets for the development of anti-AD drugs [5][6][7]. The cleavage sites of APP are shown in Figure 1A. BACE1, one of the processing enzymes of APP, triggers Aβ formation in the rate-limiting first step by the cleavage of APP at the Aβ domain N-terminus (β-site). BACE1 is a type-I transmembrane γ -secretase cleavage site (γ-site) aspartic protease with 501 amino acids. BACE1 and APP are located in the same intracellular granules, the endoplasmic reticulum, Golgi, and trans-Golgi networks, which are acidic environments, which suggested that Aβs are produced in these locations [8]. Next, another aspartic protease, the γ-secretase complex, cleaved at the C-terminus of the Aβ domain and released Aβ peptides. The γ-secretase complex that contains a protein via the presenilin gene as a catalytic component cleaved at two cleavage sites, 'γ-sites', which mainly resulted in the formation of two species of Aβs: Aβ 1-40 and Aβ 1-42 ( Figure 1A). Aβ 1-42 shows greater neurotoxicity and aggregability than Aβ 1-40 and appears to be a key biomolecular marker of AD pathogenesis. According to the amyloid hypothesis, BACE1 and γ-secretase appear to be molecular targets for the development of anti-AD drugs. However, because γ-secretase can cleave other singlepass transmembrane proteins in vivo, such as Notch, which plays a critical role in cell differentiation, γ-secretase inhibitors appeared to lead to serious side effects. As BACE1 knockout transgenic mice demonstrated normal survival, this indicated a promising direction of study, in which BACE1 is a molecular target for the development of AD drugs [9]. At present, many BACE1 inhibitors have been revealed, including those in our study [10][11][12][13][14][15][16].

Early peptidic BACE1 inhibitors
An early inhibitor of BACE1, an aspartic protease, was designed on the basis of a substrate transition-state concept, as well as that of other aspartic proteases, such as renin and HIV protease, which have a substrate transition-state analogue at the P 1 position [10][11][12][13][14][15][16]. It is wellknown that the Swedish mutant APP (K670N and M671L double mutation, Figure 1A) is cleaved faster than wild-type APP by BACE1, which results in increased Aβ 1-42 and Aβ 1-40 levels. Early BACE1 inhibitors were designed based on the Swedish-mutant APP amino acid sequence. In 1999, Sinha et al. from Elan Pharmaceuticals purified the BACE1 enzyme from the human brain using a transition-state analogue based on the Swedish-mutant sequence, and succeeded in cloning the BACE1 enzyme [17]. Ghosh et al. reported the potent inhibitors OM99-2 (Ki = 1.6 nM) and OM00-3 (Ki = 0.3 nM) with a hydroxyethylene unit as a substrate transition-state analogue ( Figure 1B) and the first X-ray crystal structure (PDB ID: 1FKN) of a complex between recombinant BACE1 and OM99-2 [18][19][20][21].
We have reported a series of peptidic BACE1 inhibitors that possessed a norstatine-type transition-state analogue [22][23][24][25][26][27][28][29][30]. Our early inhibitors are shown in Table 1. Octapeptide 1 with an Asn residue and (2R, 3S)-3 amino-2-hydroxy-5-methylhexanoic acid (Nst, Leu-type transitionstate analogue) at the P 2 and P 1 positions, respectively, corresponding to the Swedish-mutant APP sequence showed no inhibitory activity. Octapeptide 2 with (2R, 3S)-3-amino-2-hydroxy-4-phenylbutyric acid (Pns) as a transition-state analogue at the P 1 position, showed weak inhibitory activity. Compound 3 with an Asp residue similar to OM00-3, and compound 4, with a Met residue at the P 2 position, also showed weak inhibitory activity. Although compound 5 with the P 2 -Lys residue that corresponded to the wild-type APP sequence showed no inhibitory activity, octapeptide 6 that possessed a Leu residue at the P 2 position exhibited potent inhibitory activity (>90% at 2 μM). We synthesized some truncated peptides on the N-or C-terminus in order to confirm the essential moiety for the inhibitory effect. N-truncation of peptides eliminated their inhibitory activity (peptides 7-8). Although C-truncated peptides 10-13 showed a weaker inhibitory activity than octapeptide 6, pentapeptide 12 replicated the inhibitory activity of heptapeptide 10. Hence, we designed a series of pentapeptidic BACE1 inhibitors based on compound 12 using a computational approach.

Docking simulation and design of pentapeptidic BACE1 inhibitors
Early BACE1 inhibitors were designed using the coordinates of the first reported X-ray crystal structure (1FKN) of a complex between BACE1 and OM99-2. OM99-2 has an Asn residue, which corresponded to the P 2 residue of Swedish mutant sequence. As OM99-2 in 1FKN appeared to interact with the BACE1-Arg235 side chain via hydrogen bonding, many researchers have designed BACE1 inhibitors that possessed a hydrogen bond receptor, such as an Asn residue, at the P 2 position, using the 1FKN coordinates. However, our peptidic BACE1 inhibitors with an Asn at the P 2 position showed no inhibitory activity, and, as shown in Table 1, the P 2 residue that showed potent BACE1 inhibitory activity was a hydrophobic amino acid residue, Leu. Thus, our design strategy seemed to require a fundamental review.
We researched the inhibitory mechanism of our peptides using a computational approach. As we found that pentapeptide 21, which possessed an aminobenzoic acid residue as a bioisostere of the Asp residue at the P 1 ' position, showed higher inhibitory activity than pentapeptide 12, we evaluated the series of pentapeptides 14-24 with an aminobenzoic acid residue by using a docking simulation, and then synthesized the compounds ( Table 2). The docking Compound P 4 P 3 P 2 P 1 P 1 ' simulation was performed using MOE software under the MMFF94x force field. The calculated active sites of BACE1 were depicted in Figure 2A using the Alpha SiteFinder application in MOE software and the coordinate set of X-ray crystal structure, 1FKN. The 3D structure of OM99-2 after a docking simulation is shown in Figure 2B as a magenta-coloured stick model. Because the moieties from the P 1 to P 4 positions of OM99-2 almost coincided with that of the X-ray crystal structure (aqua coloured stick model) of OM99-2, we performed the docking simulation study using this calculation model. Although the moieties from the P 2 ' to P 4 ' positions of OM99-2 assumed a different pose from that of the X-ray crystal structure of OM99-2, their moieties were placed outside the BACE1 enzyme. It is likely that the difference between the X-ray crystal structure and the docking calculation might occur by a packing at the crystallization of BACE1 complex. The results of the docking simulation are shown in Table 2. Peptides 14-24 and OM99-2 were scored using the scoring function in the MOE software. 2 van der Waals energy between BACE1 and inhibitor (kcal/mol). 3 Strain energy of inhibitor (kcal/mol). 4 Docking score (kcal/mol); U_dock = U_ele + U_vdw + U_str. 5 BACE1 inhibition % at 2 μM. 6 Cyclohexylalanine (Cha). U_ele, U_vdw, and U_str indicated the potential energies of electrostatic interaction, van der Waals interactions between enzyme and ligand, and strain energy of the ligand, respectively, and their sum is a docking score, U_dock. OM99-2 and peptide 14, which possess an Asn residue at the P 2 position, showed good scores. However, peptides 14 and 17, that possess a hydrophilic amino acid, such as Asn and Asp, showed a low inhibitory activity. The residues of peptides 21 and 24 that exhibited a high inhibitory activity were hydrophobic amino acids such as Leu and cyclohexylalanine (Cha), as well as the results in Table 1. Interestingly, peptides 21 and 24 showed low scores. Thus, there was no association between the scoring values and the inhibitory activity at the P 2 position.
As the docking model using the coordinate set of 1FKN appeared to give an unfavourable score for the BACE1 inhibitor, we compared the publicly available X-ray crystal structures of BACE1-inhibitor complexes. Surprisingly, the guanidine group of BACE1-Arg235 in most crystal structures, except 1FKN, showed similar figures flopping over the P 2 region of the inhibitors, and the nearest distances between the guanidino-plane of Arg235 side chain and the P 2 region of the inhibitor showed similar values of approximately 3Å [31]. The P 2 moieties in most of the crystal structures found to interact with the BACE1-Arg235 side chain were a methyl group, carbonyl oxygen atom, or aromatic ring, which were bound to the guanidineplane of Arg235 side chain by CH-π, O-π, or π-π stacking interactions. This suggested that the π-orbital on the guanidino-plane interacted with the P 2 region by a weak quantum force such as stacking or σ-π interaction. The only exception was the interaction in the first reported X-ray crystal structure, 1FKN. The P 2 moiety of OM99-2 in the crystal structure 1FKN appeared to interact with the BACE1-Arg235 side chain via hydrogen bonding ( Figure 3A). OM00-3, which was reported by the same researchers, was an inhibitor that was structurally similar to OM99-2; surprisingly, the P 2 -Asp side chain of OM00-3 docked in BACE1 (PDB ID: 1M4H) interacted with the π-orbital on the guanidine-plane of the BACE1-Arg235 side chain via O-π interaction ( Figure 3B). Many early BACE1 inhibitors that possess a hydrogen bond receptor at the P 2 position were designed using the first reported crystal structure 1FKN. However, the hydrogen bonding interaction between most of the inhibitors and the BACE1-Arg235 side chain was not shown in their crystal structures. For instance, the inhibitor from Merck (MSD), crystal structure (PDB ID: 2B8L), interacted with the BACE1-Arg235 side chain via a CH-π interaction ( Figure 3C). The researchers at MSD most likely based their inhibitor on a structure that possessed an N-methyl-N-methanesulfonyl group at the P 2 position in anticipation of the hydrogen-bonding interaction between the sulfonyl oxygen atom and the BACE1-Arg235 side chain. However, the N-methyl group of the MSD inhibitor interacted with the π-orbital on the guanidine-plane of the BACE1-Arg235 side chain at a distance of 2.8 Å. The inhibitor reported by Pfizer (PDB ID: 2P83) appeared to interact with the BACE1-Arg235 side chain via O-π interactions, as shown in Figure 3D. As seen above, most of the BACE1 inhibitors, except OM99-2 in the crystal structure 1FKN, interacted with the BACE1-Arg235 side chain by a weak quantum force such as stacking or σ-π interaction. The Arg235 side chain of the BACE1-OM99-2 complex (1FKN) assumed an exceptionally different pose to the other crystal structures because the BACE1 complex appears to be stabilized by intramolecular hydrogen-bonding interaction between the P 4 -Glu and P 2 -Asn side chains of OM99-2 ( Figure 3A). Because OM00-3 does not form such intramolecular hydrogen-bonding, the P 2 residue of OM00-3 appears to interact with the BACE1-Arg235 side chain by a quantum chemical interaction. As many researchers have designed BACE1 inhibitors with a hydrogen bond receptor on the basis of the first reported crystal structure 1FKN, docking models using 1FKN will require further review. Furthermore, we found that the side chain of BACE1-Arg235 could move in concert with the inhibitor's size. The superimposed figure of four crystal structures (PDB ID: 2B8L, 1M4H, 1W51, and 2IQG) of the complex between BACE1 and the inhibitors is depicted in Figure 4. The guanidino-planes of BACE1-Arg235 in the crystal structures of most BACE1 complexes showed similar distances from the P 2 regions of the inhibitors regardless of their molecular size [31]. This fact suggested a serious issue for a docking simulation for the drug discovery of BACE1 inhibitors. However, the BACE1-Arg235 side chain seems to have a restricted range of motion: the BACE1-Arg235 side chain slides sideways, not up and down, along the wall of the β-sheet structure that consists of four peptide strands behind the flap domain of BACE1; therefore, the location of the BACE1-Arg235 side chain could be predicted by the inhibitor's size. As shown in Figure 4, the  orbital follows the same pattern along the wall of the β-sheet structure of BACE1 and might be predicted according to the inhibitor's size. We hypothesized that the role of the BACE1-Arg235 side chain was important for the BACE1 inhibitory mechanism. The guanidine-plane of Arg235 that can move in concert with the inhibitor's size appears to push down on the P 2 region of the inhibitor, which caused them to be affixed to the active site of BACE1 because of this 'flop-over' mechanism by the BACE1-Arg235 side chain. Although a quantum chemical force, such as σ-π interaction, has a weaker binding energy than a hydrogen bonding interaction, this 'flop-over' mechanism permits a strong binding mode with the active site of BACE1.
For the reasons mentioned above, we performed a docking calculation using the X-ray crystal structure 1M4H, in which the P 2 moiety of the inhibitor (OM00-3) had a similar size to that of our inhibitor ( Table 3). OM00-3 appears to show a high docking score value owing to its large molecular size: OM00-3 has many more amide bonds than our peptapeptidic inhibitors, and 2 van der Waals energy between BACE1 and inhibitor (kcal/mol). 3 Strain energy of inhibitor (kcal/mol). 4 Docking score (kcal/mol); U_dock = U_ele + U_vdw + U_str. 5 BACE1 inhibition % at 2 μM. 6 Cyclohexylalanine (Cha). can closely interact with the BACE1 active site via electrostatic energy. The correlation chart between BACE1 inhibitory activities and the docking score values is shown in Figure 5, and it indicated a good correlation coefficient (r = −0.717). Peptides that possess a hydrophilic amino acid residue at the P 2 position showed low docking score values, which indicated that these P 2 residues cannot interact with the BACE1-Arg235 side chain via electrostatic energy in the X-ray crystal structure 1M4H, and the docking score showed a good correlation with BACE1 inhibitory activity as a result. However, the plot of peptide 24 was outside of the correlation line. Peptide 24, with the bulky amino acid Cha at the P 2 position, might show van der Waals repulsion against the BACE1-Arg235 side chain.
Furthermore, we designed a series of BACE1 inhibitors that possessed one or more bioisosteres of carboxylic acid from pentapeptide 21 that conferred excellent values to both docking score and BACE1 inhibitory activity, in order to develop practical BACE1 inhibitors as drug candidates (Figure 6). A tetrazole ring is known as a carboxylic bioisostere. Because it is well known that 5-aminotetrazole was highly explosive, peptides 25 and 26, which possessed a carboxylic acid bioisostere at the P 4 position were designed and synthesized using tetrazole-5-carboxylic acid. Peptides 25 and 26 showed potent BACE1 inhibitory activities (IC 50 = 8.2 nM and 3.9 nM, respectively) [24,25]. Moreover, peptide 27, which possessed two tetrazole rings on the P 1 ' ring, was synthesized. Peptide 27 showed the most potent BACE1 inhibitory activity (IC 50 = 1.2 nM) [26].

Design of small-sized non-peptidic BACE1 inhibitors
At present, many non-peptidic BACE1 inhibitors have been discovered. The research of Elan pharmaceuticals, Merck (MSD), Pfizer, and Schering-Plough, and Ghosh et al. reported the BACE1 inhibitors 28-32 (IC 50 = 15, 15, 5, 3, and 1 nM, respectively) with an isophthalic scaffold at the P 2 position, as shown in Figure 7 [14,16]. Because the distance between the flap domain and the cleft domain that forms the S 2 pocket of BACE1 was narrow, a planar aromatic ring, such as an isophthalic scaffold, might dock closely in the S 2 pocket of BACE1. Hence, we designed a series of BACE1 inhibitors from the virtual inhibitor 28 (Figure 8), in which the P 2 moiety of our peptidic inhibitors was replaced with an isophthalic scaffold [31][32][33][34][35]. First, we focused on the sterically hindered interaction between the P 3 amide and a proton on the P 2 -isophthalic ring of the virtual inhibitor, which restricts the configuration. Using the approach 'in-silico conformational structure-based design' based on a conformer of the docked inhibitor in BACE1, we adopted a pyridinedicarboxylic scaffold as a P 2 moiety, which lacked the 2-proton from the isophthalic ring, and designed inhibitor 29 [32]. Next, we optimized the P 3 -region of inhibitor 29   Quantitative Structure-activity Relationship with a P 2 -pyridinedicarboxylic scaffold. There is a S 3 sub-pocket behind the active site of BACE1, and the P 3 phenyl group of 29 appears to interact with the S 3 sub-pocket. We envisioned that inhibitors with a P 3 benzylamide group assumed a folding pose between the P 2 aromatic scaffold and the P 3 benzylamide, and that the α-methyl group on the P 3 benzylamide of 29 stabilized this folding structure. Hence, we designed inhibitor 30 by the introduction of a five-membered ring, oxazolidine, at the P 3 region to fix the folding structure [32]. The oxazolidine ring fixes the direction of the phenyl ring at the P 3 position, so that the P 3 phenyl ring may be able to bind closely to the S 3 sub-pocket of BACE1. Inhibitor 30 showed moderate BACE1 inhibitory activity (IC 50 = 140 nM).
Although in-silico approaches, such as a docking simulation between an enzyme/receptor and drugs, have contributed greatly to drug discovery research, most docking simulation software programs adopt molecular mechanics/molecular dynamics (MM/MD) calculations based on classical Newtonian mechanics. However, docking simulations using these calculations do not appear to estimate weak quantum chemical interactions, such as stacking or σ-π interaction, between the BACE1-Arg235 side chain and inhibitors. Because the Arg residue is recognized as charged in these software programs, the quantum chemical interactions involving an Arg side chain are unlikely to receive a favourable score. The concept of 'bioisostere' is important for the development of practical drugs. However, in the case of BACE1 inhibitor design, the bioisostere of the P 2 moiety, according to the Swedish-mutant APP, is an Asn or an amide residue based on a classical bioisostere concept that does not assume quantum chemical interactions. Therefore, inhibitors that can interact with the Arg235 side chain on the basis of a quantum chemical interaction could never be designed using such a classical concept. The researchers at Bristol-Myers Squibb (BMS) reported a series of BACE1 inhibitors that can interact with the BACE1-Arg235 side chain by π-π stacking. According to their SAR study, the introduction of an electron-donating methoxy group to the p-position of the phenyl ring that interacts with the BACE1-Arg235 side chain enhanced BACE1 inhibitory activity. This indicated that an inhibitor that possessed a P 2 aromatic ring with higher electron density could strongly dock to the active site of BACE1 that has an electron-poor π-orbital on the guanidinoplane of the BACE1-Arg235 side chain. Hence, we proposed a new concept of the 'electrondonor bioisostere', which can interact with an electron-poor π-orbital, such as the guanidine group of Arg235, by quantum chemical interactions [14].
Based on the electron-donor bioisostere concept, we speculated that an electron-rich halogen atom could interact with the electron-poor guanidine π-orbital by Coulomb interaction. Using the ab initio molecular orbital approach, Imai et al. indicated that the calculated Cl-π interaction energy was slightly stronger than that of CH-π interaction, and its energy was affected by π-electron density [36]. Inhibitor 31, which possessed a halogen atom on the P 2 aromatic scaffold, exhibited potent inhibitory activity (IC 50 = 15 nM). Moreover, inhibitor 32, which possessed a fluorine atom on the p-position of the P 3 phenyl group exhibited the most potent inhibitory activity (IC 50 = 9 nM), and was available from Wako Pure Chemical Industries (Japan) as a reagent for biological research [31]. The drastic improvement in the inhibition of BACE1 following the introduction of a halogen atom into the P 2 position of our compounds appears to support our hypothesis; namely, the quantum chemical interactions between BACE1 and its inhibitors play a critical role in the mechanism of BACE1 inhibition.

Conclusion
Although we calculated the docking scoring values by a docking calculation between BACE1 and its inhibitors using the first reported X-ray crystal structure 1FKN, we found no association between the scoring values and BACE1 inhibition. We found that a specific interaction, a quantum chemical interaction between the Arg235 side chain and the P 2 region of the inhibitor, played a critical role in the inhibitory mechanism of BACE1. Whereas most BACE1 inhibitors, except OM99-2, interacted with BACE1-Arg235 by a quantum chemical interaction, such as stacking and σ-π interaction, many early BACE1 inhibitors were designed using the 1FKN coordinate set. As the crystal structure 1FKN showed a hydrogen bonding between the BACE1-Arg235 side chain and OM99-2, the early studies on BACE1 inhibitor design might have misdirected, as a docking simulation using 1FKN appears to be meaningless. In fact, unlike 1FKN, there is no hydrogen bonding interaction present in most of the X-ray crystal structures. We selected the peptide sequence that showed potent inhibitory activity by a docking simulation using the X-ray crystal structure 1M4H, and designed potent peptidic BACE1 inhibitors with one or more carboxylic acid bioisosteres. Moreover, we focused on a quantum chemical interaction, and designed the potent nonpeptidic BACE1 inhibitor 32 using the 'electron-donor bioisostere concept' that we have proposed. Our findings indicated the importance of the X-ray crystal structure in computational drug design.