Molecular Modeling, Synthesis and Activity Studies of Novel Biaryl and Fused-Ring BACE1 Inhibitors

Abstract

A series of transition-state analogues of beta-secretases 1 and 2 (BACE1, 2) inhibitors containing fused-ring or biaryl moieties were designed computationally to probe the S2 pocket, synthesized, and tested for BACE1 and BACE2 inhibitory activity. It has been shown that unlike the biaryl analogs, the fused-ring moiety is successfully accommodated in the BACE1 binding site resulting in the ligands with excellent inhibitory activity. Ligand 5b reduced 65% of Aβ40 production in N2a cells stably transfected with Swedish human APP.

Memory loss is the most common characteristic in the clinical manifestation of Alzheimer’s disease (AD) which affects the learning of both recent as well as recalled stored information. ¹ It has been hypothesized that the therapeutic targeting of Amyloid Precursor Protein (APP) processing may mitigate the formation of toxic fragments and prevents plaque formation. This large transmembrane protein can be cleaved at three distinct sites by proteolytic enzymes collectively referred to as “secretases”. ^2–5 The key enzyme crucial for the release of amyloidogenic fragments during APP processing is called β-secretase (BACE1). ² BACE1 enzyme cleaves APP at the N-terminal side of the Aβ sequence to secrete sAPPβ, producing the cell-bound, carboxyl-terminal fragment C99 (also termed C100 or CTF). ³ The C99 fragment is then cleaved by the γ-secretase enzyme in the proteolytic cascade of APP processing. The resulting amyloid-beta peptides, called Aβ40 and Aβ42 because of the number of amino acids they are comprised of, form the extracellular neuritic amyloid plaques – one of the key factors in the pathogenesis of AD. ³

It has been shown that BACE1 levels are significantly elevated in vivo in rapidly autopsied brains of sporadic AD patients (< 3 hours) compared with age-matched non-AD patients. ^6–8 There has also been reports of a correlation (R² = 0.54) between BACE1 activity and amyloid plaques count.⁹ It was, therefore, suggested that lowering amyloid plaques in vivo may be achieved by a decrease in production of Aβ40 and Aβ42 through the inhibition of BACE1, thus opening up a new avenue for AD drug discovery. ^{10, 11} The product of β-secretase cleavage, C99, also has toxic and pathogenic effects in transgenic mouse cells. ^{12, 13} Another APP processing player, α-Secretase, cleaves within the Aβ sequence to generate a large extracellular soluble fragment (sAPPα) and a smaller intracellular fragment (C83). These fragments appear to have no pathological significance, although sAPPα may have neuroprotective characteristics. A close-homolog of BACE1, BACE2, is thought to function as α-secretase by cleaving APP in the vicinity of the APP α-secretase cleavage site. ^{14, 15} Due to its low concentration in the brain, it appears unlikely that BACE2 plays a significant role in the generation of Aβ40 or Aβ42,¹⁶ however, it may play an important role in APP processing on the periphery affecting the muscle degeneration in AD patients.¹⁷ Most of the known BACE1 inhibitors are relatively polar, non-selective and target other aspartic proteases including the highly homologous BACE2. Since BACE2 may serve as an alternative α-secretase, its inhibition is highly undesirable and needs to be addressed during the development of therapeutics that target BACE1. Multiple studies suggest that the physiologically relevant cleavage of APP by BACE1 occurs in endosomes where pH is optimal for BACE1 enzymatic machinery^18–22 or in cholesterol rich lipid rafts. ²³ Therefore, further improvement of the cell membrane permeation properties of BACE1 inhibitors would be one of the most important tasks for lead optimization. A proof of principal for an alternative to the passive permeation approach, a BACE1 inhibitor conjugated with a lipid fragment targeting endosomes, was recently reported by Simons and co-authors. ²⁴

The BACE1 binding site and a common binding pose of one of the recently developed BACE1 inhibitors is shown in Fig 1 (PDB code: 2B8L).²⁵ The in vitro and cell-based potencies, as reported by Stachel et al., are 15 and 29 nM, respectively.²⁵ In the course of our preliminary modeling efforts, we identified the S2 site (Fig. 1) as a potential target for further improvement of BACE1 activity and selectivity vs. BACE2. First, the BACE1 S2 site is formed by highly polar residues Asn233, Arg235 and Ser325, and is solvent-exposed. Many of the known ligands target S2 with polar groups that impose strict geometric requirements for their optimal binding. Replacing these polar substituents with either non-polar ones or groups that require only one or two hydrogen bonds would lead to less polar ligands in general. Second, the BACE1 S2 site contains a highly flexible residue Arg235, which adopts various conformations depending on the nature of P2 substituents. The extreme displacements of the Arg235 side-chains are observed in the X-ray structures of 2B8L and 1W51 and are shown in Fig. 2. This observation suggests that the S2 site may accommodate novel scaffolds that would not otherwise be considered in the structure-based drug design efforts for BACE1 because they would poorly fit into the binding site. Third, despite the very large homology between BACE1 and BACE2 proteins the conformational changes that are induced by the ligands would not be identical and, thus, an improvement in selectivity may be observed.

Figure 1.

The binding site of BACE1, ligand 1, and a crystal water molecule in S2. The key residues in S2 site are rendered as “ball and stick”, whereas the ligand as “capped sticks”. Water 43 is shown as blue sphere. The solvent accessible area of the binding site is rendered by red surface.

Figure 2.

Two different conformations of Arg235 in BACE1 PDB structures 2B8L (left) and 1W51 (right) are rendered as capped sticks and are color coded (green-carbon, blue-nitrogen, red-oxygen, yellow- sulfur).

Collectively, these observations prompted us to hypothesize that by lowering the number of polar groups we might gain more freedom in the positioning of the ligand since in this case the ligand has to form fewer highly directional hydrogen bonds. The loss in hydrogen bonding potential would, thus, be compensated for by non-directional van der Waals (vdW) interactions. In addition, compounds with fewer polar atoms are expected to have a more favorable drug-like profile. We also hypothesized that the increased flexibility of the residues in the S2 pocket makes it a perfect candidate for exploring different chemical scaffolds for further improvement activity, selectivity, and druglike profile. To validate our hypothesis, we designed two series of compounds using computer-aided drug design, synthesized them, tested their activities against BACE1 and BACE2, and analyzed the results using molecular modeling methods. The results of these studies are described in this paper.

First, we established a strategy shown in Fig. 3 to identify novel scaffolds which would allow us to verify our initial hypotheses. A software package for automated de novo design available in Sybyl,²⁶ Real-time Automated Combinatorial Heuristic Enhancement of Lead compounds (RACHEL),²⁷ was used to generate the new scaffolds. We elected to maintain the X and Y substituents in ligand 1, while allowing changes in portions A, B, and C (Fig. 3). The bridging group C was designed to maintain the orientation of the polar groups (if any) in positions A and B, to decrease the number of rotatable bonds, and to form an additional interaction with the binding site. The A, B, C fragments were allowed to be either “nothing” (in case of the fragment C only), H, or a group selected from an in-house database of small molecule fragments. The RATOMS (number of ring atoms) descriptor of the chemical component corresponding to A, B, and C was set to minimum 5 and maximum 10. This database was generated by RACHEL from a set of drugs and druglike small molecules found on the DrugBank web-site.²⁸ The RACHEL scoring function was amended to promote selection of cyclic structures by setting the maximum number of rotatable bonds to 4 and increasing the weight of the function from 1 to 4. We selected Asn233 as a target site for growing substituents and the molecular weight was limited to 200. The newly generated ligands by RACHEL were re-docked to the binding site using FlexX docking software. Among all potential ligands generated by RACHEL bi-cycles and biaryls stood out as groups of ligands having the binding poses similar to those found in BACE1 X-rays, targeting the S2 site, and having fewer or no hydrogen bond donors or acceptors. Upon visual inspection and evaluation of synthetic accessibility two scaffolds were selected for further synthetic elaboration: saturated bicycles with fused 6 or 7 ring systems containing a heteroatom and biaryls with 5- and 6-membered terminal aryl (Fig. 3). It was further decided not to use the actual FlexX docking score because (1) docking scores are typically less reliable compared with the accuracy of the binding poses prediction, and (2) we expected that the binding site residues in S2 may change their conformation to accommodate the new substituents. To explore if these new chemotypes are indeed compatible with the binding site of BACE1 two diverse series of biaryl and 5,6,7-membered bicyclic ligands shown in Table 1 were advanced for further medicinal chemistry efforts and biological testing.

Figure 3.

De novo design strategy utilized to find P2 fragments.

Table 1.

BACE1 and BACE2 inhibition profile of P2 substituents.

Ligands	R1	BACE1 IC50 (nM)	BACE2 IC50 (nM)	log BBa
1		281 ± 42	132 ± 17	−0.9
2a	OH	1581 ± 269	> 100000	−0.8
2b	OMe	2506 ± 576	> 100000	−0.5
2c		889 ± 108	> 100000	−0.3
3a		2233 ± 384	> 100000	−0.6
3b		811 ± 102	> 100000	−0.1
3c		1119 ± 173	> 100000	−0.1
3d		1256 ± 140	> 100000	−0.5
4a		1774 ± 218	141 ± 16	−0.9
4b		158 ± 19	251 ± 33	−0.8
Ligands	-R1-	R2	BACE1 IC50 (nM)	BACE2 IC50 (nM)	log BBa
5a		H	524 ± 150	> 100000	−0.3
5b		H	63 ± 8	> 100000	−0.3
5c		H	56 ± 7	> 100000	−0.3
5d		F	740 ± 95	2089 ± 311	−0.3
5e		F	262 ± 36	1949 ± 207	−0.2
6a		H	998 ± 78	> 100000	−0.3
6b		H	112 ± 12	> 100000	−0.4
6c		H	89 ± 13	> 100000	−0.2

^acalculated log BB

The biaryl and fused-ring ligands were prepared using a common intermediate 12 according to a published procedure.³⁰ The synthesis of biaryl ligands is shown in Scheme 1. Commercially available substituted bromoisophthalic acid 7 was converted into 9 by 1) hydrolysis of 7 with 1N NaOH which afforded the monoacid 8 and 2) subsequent coupling of 8 with (R)-α-methylbenzylamine in the presence of EDCA and HOBt resulting in the formation of 9. Intermediates 10 required for final products 2c and 3c were prepared by a Sonogashira coupling of bromide 9 with the mono TMS derivative of acetylene and phenylacetylene, respectively. Suzuki coupling of 3-thiopheneboronic acid or 4-chlorophenylboronic acid with 9 resulted in intermediates 10 required for the final products 3a and 3b. Sultam intermediates 10 required for the final products 4a and 4b were prepared by coupling 10 with the corresponding ring sultams. The final compounds 2c, 3a–3d, 4a–4b were obtained by hydrolysis of 10 to form the acid intermediate 11 followed by subsequent coupling with the TFA salt of (2R, 3S)-N-1-2-hydroxy-4-phenylbutane-1, 3-diamine.^31–33 Compounds 2a and 2b were prepared using hydroxy and methoxy derivatives of diethyl esters of isophthalic acid in four steps: hydrolysis of one of the ester group with 1N NaOH, coupling with the left side fragment (R)-α-methylbenzylamine, hydrolysis of the remaining ester group using 1N NaOH, and finally coupling with the TFA salt of (2R, 3S)-N-1-2-hydroxy-4-phenylbutane-1, 3-diamine.

Scheme 1.

Reagents and conditions: (a) 1 N NaOH; THF: MeOH (50:50) (90%); (b) (R)-α-methylbenzylamine, EDCA/HOBt, 0°C to r.t (92%); (c) 8, EDCA/HOBt, 0°C to r.t (90%); (c) BOP, DIPEA.(90%). In this and further schemes the typical yields of the reactions are provided in parentheses.

The bis-esters 15, 17 and 19 were prepared according to the reported procedures starting from 14 (Scheme 2).^{34 35} Subsequent selective hydrolysis of 17 yielded monoester 18a as a major product whereas 19 resulted in monoester 20a as a major product. Further treatment of 16a, 18a and 20b with (R)-α-methylbenzylamine led to the corresponding intermediates 21–23 that upon hydrolysis and subsequent reaction with 12 resulted in the final products 5a, 5b and 6c (Scheme 3). Reaction of 16b, 18b and 20a with 12 resulted in intermediates 24–26 which upon hydrolysis and subsequent coupling with (R)-α-methylbenzylamine led to the final ligands 5c, 6a, and 6b (Scheme 4). The chemical characterization of compounds 5b³⁶ and 5c³⁷ is attached as a footnote.

Scheme 2.

Reagents and conditions: (a) PdCl₂, Cu(OAc)₂, LiCl, MeOH:H₂O (2mL : 0.1mL) (85%); (b) 1 N NaOH; THF: MeOH (50:50), 0°C to r.t (80%); (c) Tetravinyltin, Cu(OAc)₂, O₂, CH₃CN, 2 days (82%); (d) 2^nd generation Grubbs catalyst, DCM, r.t (80%)²⁹; (e) H₂, Pd/C, MeOH (90%); (f) allyl bromide, K₂CO₃, acetone, reflux (96%).

Scheme 3.

Reagents and conditions: (a) (R)-α-methylbenzylamine, EDCA/HOBt, 0°C-r.t (92%) (b) 1 N NaOH; THF: MeOH (50:50), 0 °C to r.t (90%); (c) EDCA/HOBt, 0°C to r.t (85%).

Scheme 4.

Reagents and conditions: (a) EDCA/HOBt, 0°C to r.t (90%); (b) 1 N NaOH; THF: MeOH (50:50), 0°C to r.t (95%); (c) (R)-α-methylbenzylamine, EDCA/HOBt, 0°C to (90%).

Theoretically, two regioisomeric monoesters could have been produced from each diester 15, 17, and 19 by hydrolysis of the right-side ester or the left-side ester. The structure of the hydrolysis products produced from 17 (Figures 4b and 4c) and 19 were deduced in a self-consistent manner from the interpreted results of extensive NMR studies (1-D ¹H and ¹³C, 2-D gCOSY, gHSQC and gHMBC). The assignment strategy employed for the determination of the regioisomeric structure of the monohydrolysis products represents a straight forward approach involving: 1) unambiguous assignment the ¹H and ¹³C NMR spectra for each of the hydrolysis products and 2) the careful examination of the patterns of the ³J_C,H correlations present in the gHMBC 2-D spectrum of each of the hydrolysis products, specifically the correlations pertaining to the respective carbonyl carbons. In the generalized structure (n = 1 or 2; figure 4a), the aromatic ring proton located between both carboxylate carbons will exhibit ³J_C,H correlations in the gHMBC to both carbonyl carbon signals (red and blue). The remaining aromatic ring proton (right side) will show the indicated ³J_C,H correlation only to the carbonyl carbon on the right side (in red). If a carbonyl exists as an ester moiety, i.e., with a methoxy group bonded to the carbonyl carbon, there will be an additional correlation cross peak in the gHMBC 2-D spectrum arising from a ³J_C,H coupling to the protons of the methoxy group to that carbonyl carbon. In this way, unambiguous assignment of the structure of the hydrolysis products can be made with confidence. This concept is illustrated for the starting dimethyl ester 17 (n = 1, R = R′ = Me), for which all of the ¹H and ¹³C resonances were assigned.

Figure 4.

(a) General structure of ligands 17 and 19; (b) ¹H NMR assignments for 17; (c) ¹³C NMR assignments for 17; (d) ¹³C NMR assignments for 18a; (e) ¹³C NMR assignments for 20a.

In the case of the hydrolysis product 18a derived from 17, the 1-D proton NMR spectrum (400 MHz, DMSO-d₆) confirmed the presence of three chemically distinct methylene groups with resonances centered at δ 1.921 (m), 3.027 (triplet, J = 5.2), and 4.175 (triplet, J = 6.6), each integrating for two (2) protons and all associated with the 6-membered ring present in the monohydrolyzed product. The gCOSY spectrum confirmed that the methylene groups constituted an isolated mutually coupled proton spin system. A singlet (3H) appearing at δ3.837 was assigned to the methyl group of the monoester moiety. Two doublets were observed centered at δ 7.979 and δ 7.428 (J = 1.5 Hz), which were shown to be spin coupled (gCOSY) and were assigned to the two aromatic protons present on the aromatic ring. The magnitude of the J-coupling between the two aromatic protons confirmed the indicated meta–relationship. The protonated carbons could be assigned from the gHSQC spectrum (¹J_C,H) and gHMBC spectrum (²J_C,H and ³J_C,H). A broad resonance (1H) appeared at ~ δ13 ppm and was assigned to the proton of the free acid. The gHMBC of hydrolysis product 18a from 17 showed correlations from the proton at δ δ7.979 to both of the carbonyl carbons at δ 166.2 and 166.3 ppm. There was a correlation observed from both the proton signal at δ7.428 to the carbonyl carbon appearing at δ166.3 as well as a correlation cross peak to the ester methyl group at δ 3.837 indicating that the hydrolysis of 17 proceeded at the left-side carbonyl in the generalized structure. In contrast, the 1-D proton NMR spectrum (400 MHz, DMSO-d₆) of the hydrolysis product 20a from 19 confirmed the presence of four chemically distinct methylene groups with slightly broadened resonances centered at (δ 1.673, 1.899, 2.992–3.018 m, and 4.012 t (J = 6.6 Hz) each integrating for two (2) protons and all associated with the 7-membered ring present in the mono-hydrolyzed product from 19. As in the case of 20a, the gCOSY spectrum confirmed that these resonances also constituted an isolated mutually coupled proton spin system. A singlet (3H) at δ 3.854 was assigned to the methyl group of the monoester moiety. Two signals were observed at δ 7.937 and δ7.582, which were shown to be spin coupled (gCOSY) (J = 1.7 Hz) and which were assigned to the two aromatic protons present on the aromatic ring. The magnitude of the J-coupling between the two aromatic protons is also consistent with the indicated meta –relationship. Similar to the NMR of 18a, the hydrolysis product of 17, a broad resonance (1H) at ~ δ 13 ppm was also observed and was assigned to the proton of the expected free acid functionality. The protonated carbons could be easily assigned from the gHSQC spectrum (¹J_C,H). The gHMBC of hydrolysis product 20a from 19 showed correlations from the proton at δ 7.937 to the both of the carbonyl carbons appearing at δ 166.0 and 167.2 ppm. There was a correlation observed from the proton signal at δ 7.582 to the carbonyl carbon at δ 166.0. The carbonyl at δ 167.2 also showed a correlation in the gHMBC spectrum to a methyl singlet (δ 3.854) indicating that the hydrolysis of 19 proceeded at the right-side carbonyl (red, in the generalized structure). Analysis of the 2-D gCOSY, gHSQC, and gHMBC data together with the 1-D ¹H and ¹³CNMR data obtained for both hydrolysis products permitted a self-consistent assignment of the ¹H chemical shifts and of the ¹³C chemical shifts for both regioisomers. The ¹H and ¹³C NMR assignments of intermediates 17 and 19 and their hydrolysis products 18a and 20a are shown in Figures 4d and 4e.

For IC₅₀ measurements, 293T cells stably transfected with pcDNA-BACE1 were maintained in 200ug zeocin/ml DMEM, 10% FBS. Cells were lysed by lysis buffer (PBS with 1%TritonX100 and 0.1%SDS), and the lysate was adjusted to 4ug/ul. The BACE1 inhibitors were diluted to the desired concentration with reaction buffer (100mM Tris-HCl, 100mM NaCl, pH4.5). The consistent expression levels of BACE1 protein and enzyme activity were confirmed by BACE1 ELISA and BACE1 activity assay. The BACE1 inhibitors were diluted to the desired concentration with reaction buffer (100mM Tris-HCl, 100mM NaCl, pH4.5). We notice that pH4.5 is not ideal for Tris buffer, however we found that Tris buffer does not interfere with the activity and stability of the BACE1 (and BACE2) lysates at pH 4.5 necessary for their enzymatic activity. ^{2, 9, 38}

The BACE1 substrate (EDANS-SEVNLDAEFR-DABCYL) was dissolved in DMSO as stock and then diluted to 10μM working solution. We then mixed 20ug of 293T/pcDNA-BACE1 of cell lysate with BACE1 inhibitor and the substrate, and the final substrate concentration was 5μM. Fluorescence was detected in microplate reader (Bio-tek) every 5min at emission length 500 nm as well as excitation at about 430 nm. Again, the consistent expression levels of BACE1 protein and enzyme activity were confirmed by BACE1 ELISA and BACE1 activity assay. Maximal velocities were calculated by the time point within 20minutes. Similarly, 293T cells stably transfected with pcDNA-BACE2 were maintained in 200ug G418/ml DMEM, 10% FBS. Cells were lysed by lysis buffer (PBS with 1%TritonX100 and 0.1%SDS), 5ug protein were used in the assay. The BACE2 inhibitors were diluted to desired concentration with a reaction buffer (100mM Tris-HCl, 100mM NaCl, pH4.5). The BACE2 substrate (MCA-ERHADGLALALEPA(K-Dnp) was dissolved in DMSO as stock. Inhibitors were further diluted to the desired concentration with a reaction buffer (100mM Tris-HCl, 100mM NaCl, pH4.5). The cell lysate was mixed with BACE2 inhibitors and the substrate (final concentration 5μM). Fluorescence was detected every 5minutes at an emission length of 430 nm as well as excitation at about 340 nm. Maximal velocities were calculated by the time point within 20min. The IC₅₀ of BACE1 and BACE2 inhibition was then calculated and the results are shown in Table 1.

The following procedure was implemented to determine the efficacy of our best BACE1 inhibitors in reducing the Aβ40 production. Mouse neuroblastoma cells (N2a) stably transfected with Myc-epitope tagged Swedish mutant APP 695 cDNA (a kind gift from Dr. Gopal Thinakaran) were cultured in 1:1 Opti-MEM/Dulbecco’s modified Eagle’s medium (high glucose) containing 5% fetal bovine serum, 1% penicillin/streptomycin and 0.2 mg/ml Geneticin (G418) in a humified air incubator at 37°C (5% CO₂ – 95% O₂). Cells were plated in 24 well plates for 24h at a concentration of 15×10⁴ cell/well. After 24h, the growth medium was discarded and replaced with 0.5 ml of serum reduced medium consisting of Dulbecco’s modified Eagle’s medium (high glucose) and 0.2% fetal bovine serum. Fifteen minutes after medium replacement, the cells were treated with drugs dissolved in DMSO and incubated for 24hours. At the end of the incubation period, 100 μl of conditioned media were collected and the protease inhibitor AEBSF (4-(2-aminoethyl) benzenesulfonyl fluoride hydrochloride) was immediately added at a final concentration of 1mM. The concentration of Aβ40 was determined using human beta amyloid [1–40] colorimetric ELISA kit from Invitrogen. The procedure was followed as instructed in the supplied protocol.

The subsequent molecular modeling analysis was performed in the following way. The X-ray coordinates of BACE1 (PDB code: 2B8L) were downloaded from the Protein Databank. The catalytic residue Asp32 was protonated according to the published theoretical studies. ^{39, 40} The active site was defined as a sphere enclosing residues within 10Å of the bound ligand. The 3D structures of ligands shown in Table 1 were built using Sybyl 7.3⁴¹ and were energy minimized using BFGS algorithm, 5000 iterations, and gradient of 0.005 kcal/mol/Å as the termination criterion. The resulting minimized molecules were docked to the binding site of BACE1 using the GOLD software.⁴² All poses outputted by the docking program were visualized; however only the pose with the best fitness score was used for the following SAR analysis. Wherever appropriate, docking under induced-fit mode was applied to accommodate the side-chain flexibility of Arg235. Prior to the SAR analysis, the ability of the docking program to successfully reproduce the binding mode of the co-crystallized ligand 1 was evaluated. It was found that GOLD was able to reproduce the X-ray binding mode of cocrystallized 1 with a RMSD of 0.36 Å. We also calculated the logBB values to confirm that the new ligands exhibit improved blood-brain permeation properties. The log BB values of all inhibitors 1, 2a–2c, 3a–3d, 4a–4b, 5a–5e, 6a–6c were calculated using Clark’s equation.⁴³ The actual coefficients in the Clark equation were refitted using the original data in the Clark paper to account for the difference between the polar surface area and ClogP used by Clark and those generated by the Tripos software. The contact van der Waals surface area (cvdWSA) was calculated using MOLCAD separated surfaces between R¹ and the residues Thr231, Thr232, Asn233, Ser325, Arg235, Gln73 and Thr72 proximal to R¹. The figures were generated either with Vida 3.0 or with Sybyl8.0.

The best docking poses were visualized in the binding site (Fig. 5a), and the SAR of the synthesized compounds is discussed below. Hydroxyethylamine (HE) motif is well known to mimic the transition state of the peptides cleaved by aspartic proteases, and it has been successfully employed in the BACE1 inhibitor design.⁴⁴ Multiple transition state analogs containing HE motif were crystallized. Our docking confirms that the binding modes of our ligands are similar to those found for the transition state analogs.

Figure 5.

(a) Gold docked conformations of all ligands in BACE1 binding site. The P2 moieties are circled yellow; (b) Biaryl compounds 3a–3d in BACE1 binding site; (c) Cyclic ring sultam compounds 4a–4b in BACE1 binding site rendered as ribbon and tube representation. Residues Asn233 and Arg235 are rendered as ball-and-stick representations; (d) Overlay of docked poses of ligands 5a–5c when side-chain flexibility of Arg235 is considered; (e) Overlay of docked poses of ligands 6a–6c when side-chain flexibility of Arg235 is considered; (f) Ligand 5e in BACE1 binding site. Different orientations of methyl group in Ala335 observed in X-ray crystal structure is rendered as a ball-and-stick representation and circled in red for clarity.

Substitution of R¹ by a simple hydroxyl or a methoxy group in 2a and 2b (Table 1) resulted in no activity at BACE2 and approximately 6- and 9-fold loss in BACE1 activity compared to ligand 1, respectively, a relatively small tradeoff in exchange of improvement of the druglike profile and BACE1/BACE2 selectivity. The docking shows that the solvent-exposed hydroxyl or methyl group in 2a and 2b do not form favorable contacts with the binding site. In the induced-fit mode docking, the side-chain guanidine group of Arg235 is able to interact with the hydroxyl or methoxy groups in 2a and 2b, respectively. Substitution of R¹ in 1 with an acetylene moiety in 2c led to a mere 3–fold decrease in BACE1 activity. As well as in the case of ligands 2a and 2b this is a small trade-off for replacing methyl sulfonamide group of 1 with a small acetylene group in 2c. This observation supports our initial hypothesis that the loss of hydrogen- bonding contribution may be compensated by non-directional vdW interactions of the P2 substituents as a result of possible induced fit changes. Indeed, the docking shows that the P2 acetylene moiety forms extensive vdW contact with the alkyl side-chains of Arg235 and Thr231 located in S2. None of 2a–c ligands showed BACE2 inhibition below 100 μM concentration. As expected, the relatively less polar 2a–c are predicted to have better log BB of −0.3 compared to −0.9 for 1.

Encouraged by these initial findings, we synthesized and tested ligands 3a–3d to explore whether the S2 pocket can accommodate larger biaryl substituents. The docking poses of these compounds are shown in Fig. 5b. The 3-thiophene moiety in 3a did not significantly influence the activity compared to 2b. A replacement of R¹ with the bulkier p-chlorophenyl moiety in ligand 3b led to a 3-fold increase in potency compared to 3a and an almost equal activity to 2c, a ligand with a smaller R¹ acetylene substituent. An analysis of the docking pose of 3b shows that binding of its p-chlorophenyl substituent requires a displacement of the conserved water molecule WAT43 (in 2B8L) typically mediating hydrogen bonding between the sidechains of Ser325, Arg235 and Gln326 (discussed below). It is likely that an induced fit disorder caused by the removal of the conserved water molecule may contribute to improved binding energy of 3b. ⁴⁵ When the GOLD program was allowed to determine whether WAT43 should be bound or displaced upon binding of 3b, the resulting ligand-protein complex did not contain WAT43. Probing of the binding site with larger, extended substituents as in the phenylacetylene- and N-benzyltriazole-containing ligands 3c and 3d resulted in a 1.4- and 1.6-fold loss of activity compared to 3b, respectively. The loss was more pronounced in 3d where the benzyl appendage appears to be completely solvent-exposed. Unlike 3c and 3d, ligands 3a and 3b fit well within the cleft of the S2 pocket. The predicted log BB values for the biaryl series range from modest −0.6 and −0.5 for compounds 3a and 3d, respectively, to −0.1 for both 3b and 3c. Similarly to ligands 2a–c ligands 3a–d are not active against BACE2 at concentrations up to 100 μM.

To explore the impact of rigidification/cyclization of the sulfonamide group of ligand 1 on BACE1 and BACE2, compare the resulting ligands with biaryls without sulfonamide group, and to find a future pharmacophore for the hot spot Asn233 (Leu246 in BACE2), we have tested two ring sultams 4a and 4b. The six-membered sultam 4b is found to be about 11-fold more active than five-membered sultam 4, 1.8-fold more active than ligand 1 at BACE1, and 1.9-fold less active at BACE2 than ligand 1. The five-membered sultam 4a, on the other hand, was 12.6-fold more active at BACE2 than at BACE1. Sultam 4a is exclusively BACE2 selective among all the ligands that were tested. Although it is not sufficient to talk about a general trend, it appears that a sulfonamide group has to be present in S2 to maintain the activity of the ligands at BACE2. An analysis of the binding modes of ligands shows that the SO₂ group in sultams points toward the S2 site whereas the alicyclic ring is solvent-exposed (Fig. 5c). This predicted mode of binding is very similar to the one that is observed in the X-ray crystal structure of the 6-membered sultam published during the preparation of this manuscript (PDB code: 2VNM). In 4b, one of the SO₂ oxygen atoms forms a hydrogen-bond with backbone NH of Asn233 (N-O distance = 3.3 Å, N…H…O angle = 170°), whereas the other oxygen atom forms ion-dipole interactions with epsilon nitrogen (NE) of Arg235 (distance = 2.8 Å). Ligand 4a, owing to its constrained 5-membered ring, forms a weak hydrogen bond with Asn233 (N-O distance = 3.7 Å, N…H…O angle = 142°) and weak ion-dipole interactions with NE of Arg235 (distance= 3.7 Å) and hence less active compared to 4b. Further, the cvdWSA in Ų of 4a and 4b are 278 and 292 respectively. The calculated logBB values of compounds 4a and 4b are similar to that of ligand 1.

Next, we synthesized a series of fused-ring compounds which were subsequently tested for the inhibition of BACE1 and BACE2 (Table 1). We synthesized and tested three sets of regioisomers that differ by the size and position of the ring fused to the central phenyl moiety. We found that the degree of lipophilicity and flexibility incorporated at the P2 position in 5a–5c and their regioisomers 6a–6c correlates with the BACE1 inhibitory activity of the ligands: (less lipophilic, less active) 5a < 5b < 5c (more lipophilic, more active) and similarly 6a < 6b < 6c. Upon docking using GOLD under induced-fit mode, it was found that ligands 5c and 6c, containing the more flexible 7-membered ring, form better vdW contact with the enzyme as compared with the ligands 5b and 6b, which contain only a moderately flexible 6-membered ring. Ligands 5b and 6b possess more efficient vdW interactions with the binding site than their 5-membered ring analogs 5a and 6a (Figs. 5d and 5e). The cvdWSA in Ų calculated under the induced-fit mode for 5a–5c are 183, 258, 266, whereas for compounds 6a–6c it is 209, 197, and 227, respectively. The cvdWSA correlates with activity of compounds 5a–5c but not 6a–6c suggesting that this parameter alone is not sufficient to explain the trend in activity.

To find a plausible explanation for the difference in activity profiles of 5a–5c and their regioisomers 6a–6c, we analyzed the X-ray co-crystal structures of BACE1 protein available in Protein Data Bank. Two interesting observations resulted from this analysis. First, the group at the P2 position appears to control the position of Arg235 through an induced-fit effect. Depending on the steric bulk and electronic properties of the P2 substituent, Arg235 is forced to adopt different conformations. The maximum change in the location of Arg235 was observed between 1W51 and 2B8L with a deviation of 4.4Å between the two guanidine carbon atoms in the Arg235 sidechain. Second, in those cases where the P2 has N-methylsulfonamide moiety (PDB codes: 1TQF, 2IRZ, 2B8L, 2IS0, 2NTR, 2OAH, 2P4J, 2P8H, and 2PH6), a conserved water molecule mediates hydrogen bond between the sidechains of Arg235, Ser325 and Gln326. On the other hand, if Arg235 changes its position as in 1W51 the water molecule is not observed in the crystal structures. It appears that both observations are interconnected – those ligands responsible for the movement of Arg235 also contribute to the displacement of the water molecule and loss of the hydrogen bonds between this water molecule and residues Ser325, Arg235 and Gln326. To account for the induced fit effect Arg235 was allowed to change conformation during docking with Gold. Fused-ring compounds 5a–5c has an oxygen atom locked in a position amenable for hydrogen bonding with side-chain NH₂ of Arg235. Our docking experiments suggest that in order to facilitate this hydrogen bond, Arg235 must lose its contact with water (WAT43 in 2B8L). Similar hydrogen-bonding interaction is not possible in regioisomers 6a–6c because the ring precludes their contact with Arg235 and hence a near 2-fold drop in potency compared to 5a–5c is observed. The calculated logBB is −0.3 for ligands 5a–5c and −0.3, −0.4 and −0.2 for 6a–6c, respectively. This shows improvement when compared to the ligand with the sulfonamide moiety bound to S2 pocket. As with the other ligands in the biaryl and fused ring series that according to the docking experiments require an induced fit, ligands 5a–c and 6a–c did not show inhibition of BACE2.

Several publications have reported that fluorinated BACE1 inhibitors exhibited better activity than the non-fluorinated inhibitors due to increased lipophilicity. ^{30, 46} Unlike the compounds described in these publications, both fluorinated compounds 5d and 5e are found to be less active than non-fluorinated analogs 5b and 5c by c.a. 5- and 12-fold, respectively. Ligand 5e exhibits 2.8-fold better activity than 5d perhaps for the same reasons 5c is more active than 5a and 5b – larger area of contact with the binding site due to flexibility of the seven-membered ring. To understand what could be the reason for overall lower activity of 5d and 5e compared to their non-fluorinated analogs we visually inspected the crystal structures of BACE1 with different P2 substituents and found that the induced-fit effect of P2 affects the orientation of the methyl group in Ala335 (Fig. 5f). Of particular interest, the RMSD between CB atoms for crystal structures 2P8H (P3 = p-fluoro-α-methylbenzylamino, P2 = N-methylsulfonamide) and 2IQG (P3 = N, N-dipropanoyl and P2 = Me) is 1.2 Å. Because of the proximity of the methyl group in Ala335 and the fluorine atom in the BACE1 inhibitors, any conformational changes in S2 may result in mutual steric clashing. If an induced fit caused by the fused rings in series 5 and 6 takes place it may, indeed, affect the position of Ala 335 and explain the overall lower activity of the fluorinated compounds 5d and 5e compared to their non-fluorinated analogs. The predicted log BB values of 5d and 5e are almost the same as those of the non-fluorinated compounds. Unlike non-fluorinated fused ring ligands in series 5 and 6, both fluorinated ligands 5d and 5e showed a reasonable, c.a. 2 uM, activity against BACE2.

To find a possible explanation for BACE1 selectivity of the ligands we tried to compare qualitatively the flexibility of the BACE1 and BACE2 proteins using their b-factors. A quantitative comparison of the b-factors is not possible since there is large number of the same residues in different X-ray structures of BACE1 proteins that differ in their b-factors depending on the co-crystallized ligand. It is clear, however, that the flexibility patterns for the residues in the S2 pocket and adjacent to it areas of the BACE1 and BACE2 proteins are different, suggesting that induced fit may play a role in an improved selectivity toward BACE1.

To investigate the effect of BACE1 inhibitors on APP processing the most active ligands 5b and 5c were tested for reduction in Aβ40 production in mouse neuroblastoma cells (N2a) transfected with Swedish mutant APP695 (Fig. 6). Compounds 5b and 5c reduce c.a. 65% and 35% of Aβ production, respectively, compared to the control. At the moment it is unclear why almost equally active 5b and 5c exhibit 1.9-fold difference in reduction of Aβ40. It may be related to their metabolic stability or delivery to the site of action. This observation is being actively investigated at this time.

Figure 6.

Reduction of Ab40 levels by BACE1 inhibitors detected in N2a cells stably transfected with Swedish human APP

In summary, by using computer-aided drug design methods, we have designed, synthesized, tested the activity of the fused-ring and biaryl compounds against BACE1 and BACE2, and analyzed the resulting SAR using docking protocols. The fused-ring compounds are in general more active than the biaryl-based ligands with an activity range from 56 nM to 998 nM. This is comparable to the activity of the ligands with polar substituents occupying the S2 binding pocket, which lends support for our initial hypothesis that the S2 site may be targeted with less polar substituents. Most of the ligands displayed more favorable calculated logBB. The side-chain flexibility of Arg235 and, perhaps, adjacent residues, and possibly the presence of a water molecule mediating hydrogen bond interactions between Arg235, Ser325 and Gln326 in S2 appears to play an important role in accommodating the fused-ring and biaryl-based ligands in the binding site. Most of the fused-ring and biaryl-based ligands are BACE1 selective, and thus support our initial hypothesis that differences in the flexibility of BACE1 and BACE2 proteins and the ability to accommodate the ligands through induced fit can be utilized to design BACE1/2 selective ligands. In addition, it appears that the sulfonamide group is required to maintain BACE2 activity. The fused-ring ligands 5b and 5c combine the best attributes of acceptable logBB, BACE1 inhibitory activity, BACE2 selectivity, and ability to reduce the Aβ production. A similar approach may be used to explore other binding pockets of BACE1 (and BACE2) to create new chemotypes for further development of therapeutics to treat AD and to serve as general probes for drug discovery.