Fragment Linking and Optimization of Inhibitors of the Aspartic Protease Endothiapepsin: Fragment‐Based Drug Design Facilitated by Dynamic Combinatorial Chemistry

Abstract Fragment‐based drug design (FBDD) affords active compounds for biological targets. While there are numerous reports on FBDD by fragment growing/optimization, fragment linking has rarely been reported. Dynamic combinatorial chemistry (DCC) has become a powerful hit‐identification strategy for biological targets. We report the synergistic combination of fragment linking and DCC to identify inhibitors of the aspartic protease endothiapepsin. Based on X‐ray crystal structures of endothiapepsin in complex with fragments, we designed a library of bis‐acylhydrazones and used DCC to identify potent inhibitors. The most potent inhibitor exhibits an IC50 value of 54 nm, which represents a 240‐fold improvement in potency compared to the parent hits. Subsequent X‐ray crystallography validated the predicted binding mode, thus demonstrating the efficiency of the combination of fragment linking and DCC as a hit‐identification strategy. This approach could be applied to a range of biological targets, and holds the potential to facilitate hit‐to‐lead optimization.

Abstract: Fragment-based drug design (FBDD) affords active compounds for biological targets.W hile there are numerous reports on FBDD by fragment growing/optimization, fragment linking has rarely been reported. Dynamic combinatorial chemistry (DCC) has become ap owerfulh it-identification strategy for biological targets.W er eport the synergistic combination of fragment linking and DCC to identify inhibitors of the aspartic protease endothiapepsin. Based on X-ray crystal structures of endothiapepsin in complex with fragments, we designed alibrary of bis-acylhydrazones and used DCC to identify potent inhibitors.The most potent inhibitor exhibits an IC 50 value of 54 nm,which represents a240-fold improvement in potency compared to the parent hits.S ubsequent X-ray crystallography validated the predicted binding mode,t hus demonstrating the efficiency of the combination of fragment linking and DCC as ahit-identification strategy.This approach could be applied to arange of biological targets,and holds the potential to facilitate hit-to-lead optimization.
Over the past decade,fragment-based drug design (FBDD) has emerged as anovel paradigm in drug discovery and it has been applied to ag rowing number of biological targets. [1][2][3] FBDD has higher hit rates than high-throughput screening and enables coverage of the chemical space using smaller libraries. [2] Since its inception in the mid-1990s, [4] FBDD has expanded tremendously and various pharmaceutical compa-nies have used FBDD to develop more than 18 drug candidates that are now in clinical trials. [5] After the identification of fragment hits by various screening techniques,t he hits are optimized to lead compounds and drug candidates by fragment growing,l inking, and/or merging. Fragment growing, on the one hand, has become the favorite optimization strategy, [6,7] even though it involves cycles of iterative design, synthesis and validation of the binding mode of each derivative.T oo vercome this drawback, we have previously reported the combination of fragment growing and dynamic combinatorial chemistry (DCC) to accelerate drug discovery. [8] Fragment linking,o n the other hand, is attractive because of the potential for superadditivity (an improvement in ligand efficiency( LE) rather than mere maintenance of LE). Thefirst example of fragment linking was reported by Fesik and co-workers. [4,9] Since then, afew studies demonstrating the efficiency of fragment linking of low-affinity fragments to produce higher-affinity ligands have been reported. [10,11] Thec hallenge lies in preserving the binding modes of the fragments in adjacent pockets whilst identifying alinker featuring an optimal fit. [12,13] In addition to FBDD,D CC [14][15][16][17][18] and dynamic ligation screening (DLS) [19][20][21][22] are powerful strategies for identifying/ optimizing hit compounds for biological targets.Inadynamic combinatorial library (DCL), the bonds between the building blocks are reversible and are continuously being made and broken. Addition of the target protein leads to re-equilibration as one or more library components are bound to the protein, resulting in amplification of the strongest binder(s) from the DCL. In DLS,f ormation of ar eversible covalent bond between adirecting probe and anucleophilic fragment enables the detection of low-affinity ligands while measuring at micromolar concentrations.
We therefore envisaged the potentially synergistic combination of fragment linking and DCC as an efficient hitidentification/optimization strategy.I nt his work, we combined fragment linking and bis-acylhydrazone-based DCC to identify inhibitors for endothiapepsin, which belongs to the notoriously challenging family of pepsin-like aspartic proteases. [23] Aspartic proteases are found in fungi, vertebrates,plants, and retroviruses such as HIV.T his class of enzymes play ac ausative role in important diseases such as malaria, Alzheimers disease,h ypertension, and AIDS. [23] Owing to its high similarity with these drug targets,endothiapepsin has been used as amodel enzyme for mechanistic studies [24][25][26] and for the discovery of inhibitors of renin [27] and b-secretase. [28] Endothiapepsin is arobust enzyme,which remains active for more than 20 days at room temperature,isreadily available in large quantities,and crystallizes easily,thus making it auseful representative for aspartic proteases. [18] Pepsin-like aspartic proteases are active as monomers and consist of two structurally similar domains,e ach of which donates an aspartic acid residue to the catalytic dyad (D35 and D219 in endothiapepsin), which hydrolyzes the peptide bond of the substrate through nucleophilic attack by ac atalytic water molecule.
Although bis-acylhydrazone-based DCC has been reported, [29] there are no reports of fragment linking using DCC.H erein, we describe the combination of fragment linking/optimization and DCC to efficiently afford ligands for inhibiting the aspartic protease endothiapepsin.
We chose X-ray crystal structures of endothiapepsin in complex with acylhydrazones 1 and 2 (PDB IDs:4 KUP and 3T7P,r espectively) as as tarting point for fragment linking ( Figure 1). We had previously identified 1 and 2 as hits from an acylhydrazone-based DCL using the synergistic combination of de novo SBDD and DCC. [18] Our hits 1 and 2 displayed IC 50 values of 12.8 mm and 14.5 mm and ligand efficiencies (LEs) of 0.27 and 0.29, respectively,a gainst endothiapepsin. Both hits displayed alternative binding modes with the catalytic dyad:e ither through aw ater-mediated interaction or through direct interaction, with displacement of the lytic water molecule.Fragments 1 and 2 occupy the S1 and S2 or S1 and S2' pockets,respectively (Figure 2a).
We envisaged the linking of 1 and 2 to afford an inhibitor that should occupy the S1, S1',S 2, and S2' pockets of endothiapepsin and benefit from numerous protein-ligand interactions,w hile maintaining/improving the LE. With the help of the molecular-modeling software Moloc, [30] we linked the mesityl moiety of 1 to the naphthyl moiety of 2 through an acylhydrazone linker,w hich resides at the junction of the S2 and S2' pockets and appeared to be as uitable linker. Acylhydrazone-based DCC is attractive for medicinal chemistry-based projects because the resulting products feature both H-bond donors and H-bond acceptors and are stable enough as drug candidates under acidic and physiological conditions.I no ur previous studies,w ed emonstrated that acylhydrazone-based DCC is compatible with endothiapepsin. [8,18] Inspection of known co-crystal structures of endothiapepsin [31] and hotspot analysis [32] of the active site suggested that both aromatic and aliphatic moieties can be hosted in the S2' pocket, since they benefit from hydrophobic contacts with residues G37, L133, and F194. Based on our molecular-modeling studies and evaluation of synthetic accessibility,w ed esigned and optimized as eries of bis-acylhydrazone-based inhibitors of endothiapepsin. As uperimposition of am odeled potential bis-acylhydrazone-based inhibitor and the parent fragments is shown in Figure 2. All of the bis-acylhydrazones form H-bonding interactions with the catalytic dyad and most of them occupy the S1, S2, S1',and S2' pockets,and maintain the binding mode of fragments 1 and 2.
We set up aD CL with bis-aldehyde 3 and the nine hydrazides 4-12,w hich has the potential to produce 78 bisacylhydrazones (excluding E/Z isomers) and 12 mono-acylhydrazones.T of acilitate the analysis,w ed ivided the library into two sub-libraries.W eu sed reversed-phase HPLC and LC-MS to analyze and identify the best binders from the DCLs and we employed aniline as an ucleophilic catalyst to ensure that the equilibrium is established faster than in the absence of acatalyst.
Thefirst library,DCL-1, consisted of the four hydrazides 5, 6, 10,a nd 12 (100 mm each), and bis-aldehyde 3 (50 mm)i n presence of 10 mm aniline and 2% DMSO in 0.1m sodium acetate buffer at pH 4.6, thus resulting in the formation of 15 potential homo-and hetero-bis-acylhydrazones (excluding E/ Z isomers) and five mono-acylhydrazones in equilibrium with the initial building blocks.W ew ere able to detect all of the homo-and hetero-bis-acylhydrazones by LC-MS analysis. Upon the addition of endothiapepsin, we observed amplifi-   Figure 2a. [33] Angewandte Chemie Communications cation of the bis-acylhydrazones 13 and 14 by more than three times compared to the blank reaction (Figure 3and Figure S1 in the Supporting Information). We set up the second library, DCL-2, using the five hydrazides 4, 7, 8, 9,a nd 11 (100 mm each), and bis-aldehyde 3 (50 mm)under the same conditions, giving rise to the formation of 28 potential homo-and hetero-bis-acylhydrazones (excluding E/Z isomers) and seven monoacylhydrazones in equilibrium with the initial building blocks. Upon addition of the protein, bis-acylhydrazones 15 and 16 were amplified by afactor of more than two compared to the blank reaction (Figure 3a nd Figure S2 in the Supporting Information). We also constructed al arge library,D CL-3, using all nine hydrazides (4-12)a nd bis-aldehyde 3 and observed amplification of the previously observed bis-acylhydrazones 13, 14,and 16 along with bis-acylhydrazones 17 and 18 (Figure 3a nd S3 in the Supporting Information). We identified at otal of two homo-(13 and 16)a nd four hetero-(14, 15, 17 and 18)bis-acylhydrazones from the three libraries DCL-1-3 ( Figure 3).
To determine the biochemical activity of the amplified bisacylhydrazones,w esynthesized the two homo-bis-acylhydrazones 13 and 16 from their corresponding hydrazides 5 and 8 and the bis-aldehyde 3 (see Schemes S2 and S3 in the Supporting Information). We determined their inhibitory potencybyapplying afluorescence-based assay adapted from an assay for HIV protease. [34] Biochemical evaluation confirmed the results of our DCC experiments,w hich were analyzed by LC-MS.B is-acylhydrazones 13 and 16 indeed

Angewandte Chemie
Communications inhibit the enzyme with IC 50 values of 0.054 mm and 2.1 mm, respectively (see Figure 4, and Figures S4 and S5 in the Supporting Information). Thep otency of the best inhibitor was increased 240-fold compared to the parent hits.T he experimental Gibbs free energies of binding (DG)a nd LEs, derived from the experimental IC 50 values using the Cheng-Prusoff equation, [35] are DG(13) = À49 kJ mol À1 , DG(16) = À34 kJ mol À1 ,a nd LE(13) = 0.29, LE(16) = 0.25, which represents an improvement in DG values while preserving the LEs compared to the parent fragments (Table 1).
To validate the predicted binding mode of the linked fragments,w es oaked crystals of endothiapepsin with the most potent inhibitor (13)and determined its crystal structure (PDB ID:5 HCT) in complex with endothiapepsin at 1.36 resolution. 13 binds to the S1, S1',a nd S2 pockets and addresses the catalytic dyad through its a-C amino group (Figure 5a). Ap art of this bis-acylhydrazone is not visible in the electron-density map,t hus implying disorder of this substituent across multiple conformational states,which is in line with our modeling studies.I nt wo plausible poses,t he unresolved portion of bis-acylhydrazone 13 would be oriented towards the S2' and S6 pockets of the enzyme or remain solvent-exposed (Figure 5b).
Thed etailed binding mode of bis-acylhydrazone 13 is shown in Figure 5a.T he visible portion of 13 preserves the binding mode of the initial hit 1 and forms four charged Hbonds with the catalytic dyad through its a-C amino group, as well as an Hbond with the carboxylate group of D81 through the NH of the indolyl moiety,w hich is accommodated in the S1 pocket and engaged in offset p-p stacking and CH-p interactions with F116 and L125, respectively.T he phenyl group of 13 binds in the S2 pocket and is involved in hydrophobic interactions with I300 and I304. Like the mesityl group of 1,the phenyl group of 13 also engages in an amide-p interaction with the peptide bond connecting residues G80 and D81. Thephenyl moiety in 13 is connected with two imine functionalities,t hus making the aromatic ring electrondeficient, which presumably strengthens the amide-p interaction compared to the electron-rich mesityl group in 1. [36] In this study,wehave demonstrated for the first time that the synergistic combination of fragment linking and DCC is apowerful and efficient strategy for accelerating hit identification and optimization to afford inhibitors of the aspartic protease endothiapepsin. We exploited LC-MS analysis to identify the best binders directly from the DCLs.T he best inhibitor exhibits an IC 50 value of 54 nm,r epresenting a2 40fold improvement in potency. Subsequent soaking experiments validated our in silico fragment linking.O ur strategic combination of computational and analytical methods holds great promise for accelerating drug development for this challenging class of proteases,a nd it could afford useful new lead compounds.T his approach could be also extended to alarge number of other protein targets. [a] The Gibbs free energies of binding (DG)a nd the ligand efficiencies (LEs) were derived from the experimentally determined IC 50 values. Figure 5. a) X-ray crystal structure of endothiapepsin co-crystallized with bis-acylhydrazone 13 (PDB ID:5HCT). b) Superimposition of the crystal structure (violet) and modeled structures (yellow and cyan) of 13. [33]