Repurposing strategies on pyridazinone-based series by pharmacophore- and structure-driven screening

Abstract

We report here in silico repurposing studies on 52 new pyridazinone-based small-molecules through inverse virtual screening (iVS) methodologies. These analogues were originally designed as formyl peptide receptor (FPR) ligands. As it is sometimes the case in drug discovery programmes, subsequent biological screening demonstrated the inefficacy of the molecules in binding FPRs, failing in the identification of new hits. Through a focussed drug-repurposing approach we have defined a variety of potential targets that are suitable to interact with this library of pyridazinone-based analogues. A two-step approach has been conducted for computational analysis. Specifically, the molecules were initially processed through a pharmacophore-based screening. Secondly, the resulting features of binding were investigated by docking studies and following molecular dynamic simulations, in order to univocally confirm “pyridazinone-based ligand-target protein” interactions. Our findings propose aspartate aminotransferase as the most favourable repurposed target for this small-molecule series, worth of additional medicinal chemistry investigations in the field.

Graphical Abstract

Keywords:

• Pyridazinones
• pharmacophore model
• structure-based screening
• drug repurposing
• aspartate aminotransferase

1. Introduction

Heterocyclic-based small molecules possess the ability to mimic the structure of endogenous ligands and reversibly bind various biologically relevant targets, determining a wide range of possible biological activities^1–3. In medicinal chemistry, one important benefit associated to heterocycles is the possibility to create such a desired library of analogues based on a specific core, allowing rapid screening protocols and structure-activity relationships (SARs) for a particular target of interest⁴. However, very frequently the molecular design does not correlate with positive outcomes in biological tests. With the increase of new chemical entities discovered, there is a proportional boost of the number of molecules that fail during the twisted path of drug development. Clearly, this raises concerns with regard to the optimisation of the use of economic resources and time. As an example, by taking into accounts common failures in the process, it has been estimated that the average cost of developing a new drug ranges from two to three billion dollars and not less than 13–15 years are required to introduce the medicine into the market⁵. In line with this, only 10% of the drugs that enter into clinical trials are approved by regulatory agencies, while the remaining 90% fails due to toxicity issues or inefficacy linked to limited predictive value of preclinical studies. Specifically, more the 60% of molecules fail in phase II and 45% fail in phase III⁶. There is consensus in the medicinal chemistry community that the high percentage of failures in drug development can be a direct consequence of the R&D efficacy in identifying the appropriate drug response on the target of interest, due to limited availability of suitable preclinical disease models⁷.

Lately explored strategies of “drug repurposing” (or “drug repositioning”) have demonstrated a useful tool at the different stages of drug development, by assessing and validating new possible therapeutic effects for existing drug candidates, as well as by identifying alternative targets for abandoned compound series^8,9. Interestingly, repositioned drugs currently represent almost the 30% of all the new medicines that reach their first market^10,11.

In this context, in silico methods, such as machine learning, data-mining, and network-based approaches, offer an unprecedented and cost friendly opportunity to predict possible drug repositioning candidates, by employing large and heterogeneous data sources during the entire process of drug development^12,13. Thus, computer-aided molecular screening is a crucial tool in drug design/discovery and computational techniques represent an important resource for prompt evaluation of possible links between compounds and biological/pharmacological effects¹⁴. Various approaches, such as structure-based, ligand-based, virtual screening and inverse virtual screening (iVS), are constantly adopted in various drug discovery settings, spanning from the hit identification to the lead optimisation stages^15–17. Pyridazinone-based small-molecules are a validated scaffold for the development of enzyme and G-protein coupled receptor (GPCR) binders^18,19. All of the 52 new molecules reported here were designed as agonists of formyl peptide receptors (FPRs)^20–22, a small family of GPCRs. As the biological evaluation on this series demonstrated the lack of efficacy for all the terms to establish interactions with FPRs, we have focussed our attention on the possible repurposing of these same molecules towards alternative biological targets. To this end, we report here the in silico investigation carried out on the pyridazinone-based library through a two-step iVS analysis. The compounds were firstly screened in a pharmacophore-based iVS against 23236 proteins covering 16159 pharmacophore models formerly predicted as drug able binding sites. Subsequently, the binding capabilities of the 52 terms have been processed through docking studies. Further exploratory analyses on the binding poses of selected hits were performed with molecular dynamic simulations, to validate the outcomes of the docking analysis and confirm effective ligand/protein interactions. Our findings define the aspartate aminotransferase as a valid biological target for efficient repurposing opportunities of this heterocyclic small-molecule series.

2. Results and discussion

2.1. Heterocyclic small-molecule dataset

The dataset of compounds is composed by 52 terms (Tables 1–3) which have been obtained through standard synthetic methodologies (see Section 1, Supporting Information). The experimental procedures and characterisation data of all new intermediates and final compounds are reported in Supporting Information (Section 2)^18,23–34.

Table 1. Structures of analogues 1–9.

Compd	R1	R2	R3	R4
1a	C6H4-OCH3 (p)	C6H4-O(CH2)3CH3 (p)	OCH3	H
1b	C6H4-OCH3 (m)	C6H4-O(CH2)3CH3 (p)	OCH3	H
2a	C6H4-OCH3 (p)	Cl	C6H4-O(CH2)3CH3 (p)	H
2b	C6H4-OCH3 (m)	Cl	C6H4-O(CH2)3CH3 (p)	H
3a	CO-NH-C6H4-F (p)	H	H	CH3
3b	CO-NH-C6H4-3,4-methlyenedioxy	H	H	CH3
4a	C6H5	NO2	COCH3	C6H5
4b	C6H4-CN (m)	NO2	COCH3	C6H5
5a	C6H5	Cl	COCH3	C6H5
5b	C6H4-CN (m)	Cl	COCH3	C6H5
6a	C6H5	Br	COCH3	C6H5
6b	C6H4-CN (m)	Br	COCH3	C6H5
7	C6H5	COCH3	NH2	C6H5
8	CO-NH-C6H4-Br (p)	CH2-C6H4-(m)CO-NH-C6H4-Br (p)	H	CH3
9	CO-NH-C6H4-Br (p)	N-(C6H4-OCH3)2 (p)	H	CH3

Table 2. Structures of analogues 10–12.

Compd	R	R1
10a	Ph
11	Ph
10b	Ph
10c	Ph
12	Ph
10d	Ph
10e	Ph
10f	CH3
10g	CH3

Table 3. Structures of analogues 13–21.

Compd	R	R1	R2	R3
13a	Ph	Ph	NH2	CONHC3H7
13c	Ph	Ph	NH2	COOcC6H11
13b	Ph	Ph	NH2
17c	4-F-Ph	Ph	NHCOiC3H7	H
18	Ph	CH2Ph	NH2	CONHtC4H9
19	Ph	H	NH2	COOCH2Ph
20	Ph	CH2Ph	NH2	COOCH2Ph
14a	Ph	CH3	NH2	COCH = CHPh
15a	Ph	CH3	NH2	COCH2CH2Ph
14b	Ph	C3H7	NH2	COCH = CHPh
15b	Ph	C3H7	NH2	COCH2CH2Ph
14c	4-F-Ph	CH3	NH2	COCH = CHPh
15g	Ph	C2H5	NH2
15i	Ph	C3H7	NH2
15h	Ph	C3H7	NH2
15l	Ph	Ph	NH2
15f	Ph	Ph	NH2	COCH2CH2Ph
17b	Ph	Ph	NHCOiC3H7	H
17a	Ph	Ph	NHCOC4H9	H
15e	cC6H11	C3H7	NH2	COCH2CH2Ph
15c	Ph	C4H9	NH2	COCH2CH2Ph
15d	Ph	iC3H7	NH2	COCH2CH2Ph
16b	4-F-Ph	Ph	NH2	H
21a	Ph	CH2CONHnC3H7	H	Ph
21b	Ph	CH2CONHiC3H7	H	Ph
21c	Ph	CH2CONHcC5H9	H	Ph
21d	Ph	CH2CONHnC3H7	Ph	H
21e	Ph	CH2CONHiC3H7	Ph	H

2.2. Computational repurposing

As representative compounds, 17 compounds (1–3a, 5a, 7–9, 10b,g, 11, 13a, 15a,f, 17a, 20, 21a, 21d) were selected for following molecular modelling studies. The compound library was firstly screened in a pharmacophore mapping strategy based on a PharmMapper (http://lilab-ecust.cn/pharmmapper/) to identify potential molecular targets for the synthesised compounds^35,36. PharmMapper is backed up by a large, in-house repertoire of pharmacophore database extracted from all the targets in different database such as TargetBank, DrugBank, BindingDB and PDTD. 16159 receptor-based pharmacophore models were used during our PharmMapper screening. The results of the pharmacophore map iVS are reported in Tables S1–S17 (Supporting Information). With the aim of finding a common target for our dataset of compounds, only the best 50 targets identified for each molecule are included here and only the best 10 results for each molecule were further analysed (see results in Table S18, Supporting Information). Interestingly, various targets were highlighted as common targets for different molecules. In particular, cytoplasmic aspartate aminotransferase, protein MG296 homologue, sensor protein fixL, ATP-dependent protease hslV, dehydrogenase and superoxide dismutase, demonstrated common targets for 4–6 molecules between the 10 best fitting targets of the pharmacophore-based iVS (Table S18, Supporting Information). The cytoplasmic aspartate aminotransferase (AST) resulted a shared target by 6 different compounds (i.e. 35% of the molecular dataset). This protein is a liver pyridoxal phosphate dependent enzyme involved in gluconeogenesis and amino acid metabolism. Increased levels of activity of AST has been observed in a number of conditions, such as liver metabolic syndrome, atherosclerosis and diabetes (type I and II), and the use of small molecule that can inhibit the protein activity is under evaluation for the treatment of diabetes³⁷. Furthermore, AST has been also proposed as a promising biological target for the development of anti-neoplastic agents^38,39.

We have used docking calculations to further characterise the activity of our set of molecules on AST. To this end, calculations were performed with Autodock Vina, a validated software for iVS structure-based applications^40,41. Docking analysis of crystallised ligands, with an established binding mode, were carried out in order to obtain a minimum energy level which has been used as the cut-off for the assessment of binding energies of the new ligands. For the docking calculations, we adopted a previously reported methodology based on the crystal structures of AST (PDB ID: 7AAT) as the enzyme input files³⁷. To validate our docking procedure, we have firstly docked two know inhibitors in the active site and, subsequently, we have compared the results with former data from the literature. Hesperetin and hesperidin (K_i of 51 and 682 µM, respectively)³⁷ were docked in the AST active site (see Table 4 for calculated energies and binding efficiency; see Figure 1 for the best poses). Among the two calculated results – i.e. ligand efficiency (LE) and binding energy (BE) – the first value demonstrates linearity with the experimentally obtained data for both the hesperetin and hesperidin. Interestingly, all the 17 molecules analysed for repurposing have a predicted BE lower than that of hesperidin and analogue 3a has predicted LE higher than hesperetin.

Figure 1. Binding poses of the full molecular series (dark pink) in the active site of 7AAT. The cofactor pyridoxal phosphate is reported in yellow.

Table 4. Docking results for 7AAT.

Lig.	LE	BE[kcal/mol]
1a	0.2162	6.271
2a	0.2274	6.368
3a	0.3363	6.390
5a	0.2813	6.752
7	0.2791	6.699
8	0.2203	7.931
9	0.2156	7.760
10b	0.2583	7.233
10g	0.2792	6.421
11	0.2887	6.930
13a	0.2501	6.503
15a	0.2536	6.339
15f	0.2280	6.840
17a	0.2613	6.795
20	0.2311	7.164
21a	0.2718	7.068
21d	0.2539	6.602
hesperetin	0.3054	6.718
hesperidin	0.1831	7.875

The active site of AST is formed by Ser-107, Gly-108, Thr-109, Trp-140, His-143, His-189, Asp-194, Asp-222, Arg-224, Tyr-225, Ser-255, Phe-256, Ser-257, Lys-258, Phe-360, and Arg-386. The coenzyme PLP is covalently bonded to Lys-258 and interacts by hydrogen bonds with different residues (Trp-140, His-143, Asp-222, Tyr-225). Previous experimental and modelling studies indicate that the guanidinium groups of Arg-386 interact through hydrogen bonds and ionic interactions with the carboxylate groups of the substrate⁴². The best poses of each analysed molecule are reported in Figure S1-S17 (Supporting Information), indicating that most of the terms interact with the above-mentioned amino acids in the AST active site.

An additional molecular dynamic (MD) simulation study (Figures S18–S26) was then conducted to verify the effectiveness and stability of the poses of the most promising compound (i.e. 3a, with high LE). Specifically, we performed 50 ns MD simulations starting with the result of the docking calculations (see results in Figures 2 and 3). BE of the molecule result stable over 50 ns and the root-mean-square deviation of atomic positions (RMSD) is less than 2.0 Å. This suggests that after the first 20 ns 3a presence is stabilised inside the binding pocket and confirms the finding of the docking calculations^43,44. Moreover, the total potential energy of the system and the structure of the protein results stable over the time of simulation (Figure S18–S26, Supporting Information).

Figure 2. Binding of analogue 3a after 50 ns of MD.

Figure 3. Up, binding energy (BE) of analogue 3a during the MD simulation. Down, root-mean-square deviation of atomic positions (RMSD) of 3a during the MD simulation. Time is expressed in ps.

2.3. Admet assessment

The in silico assessment has been expanded through the evaluation of pharmacokinetic profiles and possible adverse side effects for the representative 17 molecules. The ability to reach targets in bioactive form was assessed using the SwissADME (http://swissadme.ch)⁴⁵ and pkCSM (http://biosig.unimelb.edu.au/pkcsm/)⁴⁶ web platforms. SwissADME results are reported in Figures S27-S43 and pkCSM results are reported in Tables S19–S23. Most of the compounds resulted orally available and from moderately to soluble in water (with the exception of 8 and 9 that resulted instead poorly soluble). None of the compounds resulted a substrate of glycoprotein G (Pgp). Only compounds 8 and 9 violate the Lipinski rule of 5. Additional four drug-likeness rules, namely Ghose, Egan, Veber, and Muegee, were contextually satisfied by the majority of the analogues (with the exception of 8 and 9, that do not satisfy the Ghose and Muegee rules). The result of the pan assay interference structures (designed to exclude molecules that are most likely to show false positives in biological assays) did not point out relevant issues. In parallel, the calculated absorption and distribution have been graphically represented by the Edan–Egg model reported in Figure 4 (Brain or IntestinaL EstimateD, BOILED-Egg). The visual analysis the Edan–Egg model highlights that most of the molecules are predicted to passively permeate the BBB, whereas compounds 8, 9, 13a, 15f and 20 can only be passively absorbed by the gastrointestinal tract.

Figure 4. BOILED-Egg plot. Points located in the BOILED-Egg’s yellow represent the analogues predicted to passively permeate the BBB. Points in the egg white are relative to the analogues predicted to face passive absorption by the gastrointestinal tract. Red dots indicate that the molecules are predicted not to be affected by P-glycoprotein mediated extrusion from the CNS.

With regard to the absorption parameters, the analogues present a promising oral availability due to the optimal Caco-2 cell permeability and intestinal absorption (>0.89 and >87%, respectively) (Table S19, Supporting Information). The calculated values of steady state volume of distribution are relatively low for 3a, 7a, 8, 13a, 15a, 15f, 17a, 20a, 21a and 21d. The whole data set shows a significant unbound fraction in the plasma, thus proposing availability to interact with the pharmacological target (Table S20). The calculated values of the total clearance (Table S22, Supporting Information) indicate the majority of the compounds have a good renal elimination and are not substrates of the renal organic cation transporter 2, with the exception of compounds 5a and 10g. Lastly, compounds 10g, 15f, 17a, 20a, 21a and 21d did not pass the AMES toxicity test, whereas all others did not present any relevant toxicity problem that could limit the use as drugs (Table S23, Supporting Information).

3. Materials and methods

3.1. Structure preparation and minimisation

The molecular structures of this study were built using Marvin Sketch (18.24, ChemAxon Ltd., Budapest, Hungary). A first molecular mechanics energy minimisation was used for 3 D structures created from the SMLES, and the Merck molecular force field (MMFF94) present in Marvin Sketch was used. The protonation states were calculated assuming a neutral pH. The PM3 Hamiltonian, as implemented in the MOPAC package (MOPAC2016, Stewart Computational Chemistry) was then used to further optimise the 3 D structures.

3.2. Molecular docking

Flexible ligand docking experiments were performed by employing AutoDock Vina software implemented in YASARA (v. 14.7.17, YASARA Biosciences, licenced to King’s College London), using the three-dimensional crystal structure of the Aspartate Aminotransferase in complex with the co-factor pyridoxal-5′-phosphate (PDB ID: 7AAT) obtained from the Protein Data Bank (PDB, http://www.rcsb.org/pdb). All the parameters were used at their default settings. The sequence in the structure, differently to the PDB ID: 5AX8, has the cofactor co-crystalized. The two structures have an overall RMSD: 0.817 Å, and an RMSD of 0.334 Å considering only the active site.

3.3. Molecular dynamics simulations

The molecular dynamics simulations of the aspartate aminotransferase/3a complex was performed with the YASARA Structure package using the three-dimensional crystal structure of the aspartate aminotransferase in complex with the co-factor pyridoxal-5′-phosphate (PDB ID: 7AAT). A periodic simulation cell with boundaries extending 5 Å from the surface of the complex was employed. The box was filled with water, with a maximum sum of all water bumps of 1.0 Å, and a density of 0.997 g/mL with explicit solvent. YASARA’s pKa utility was used to assign pKa values at pH 7.4, and the cell was neutralised with NaCl (0.9%) in these conditions. Water molecules were deleted to readjust the solvent density to 0.997 g/mL. The final system dimensions were approximately 71 × 79 × 69 Å. The ligand force field parameters were generated with the AutoSMILES utility, which employs semiempirical AM1 geometry optimisation and assignment of charges, followed by the assignment of the AM1BCC atom and bond types with refinement using the RESP charges and, lastly, the assignments of general AMBER force field atom types. Optimisation of the hydrogen bond network of the various enzyme–ligand complexes was obtained using the method established by Hooft et al. (see YASARA manual), to address ambiguities arising from multiple side-chain conformations and protonation states that are not well resolved in the electron density. A short MD was run on the solvent only. The entire system was then energy minimised using first a steepest descent minimisation to remove conformational stress, followed by a simulated annealing minimisation until convergence (<0.01 kcal/mol Å). The MD simulation was then initiated, using the NVT ensemble at 298 K and integration time steps for intramolecular and intermolecular forces every 1.25 fs and 2.5 fs, respectively. Finally, 50 ns MD simulations without any restrictions were conducted and the conformations of each system were recorded every 200 ps. The binding energy (BE; Figure 3) is calculated from the energy at an infinite distance by subtracting the energy of the system. More positive is the binding energy, more favourable is the interaction. The obtained binding energy values include solvation events according to the Poisson-Boltzmann approach (PBS). PBS uses a customised version of the APBS programme (see YASARA manual) to solve the Poisson-Boltzmann equation, yielding the electrostatic potential with implicit solvent and counter ions. Accordingly, the PBS binding energy is calculated through the following equation: $$\text{BE}={\mathrm{E}}_{\text{potRecept}}+{\mathrm{ }\mathrm{E}}_{\text{solvRecept}}+{\mathrm{ }\mathrm{E}}_{\text{potLigand}}+{\mathrm{ }\mathrm{E}}_{\text{solvLigand}}–{\mathrm{ }\mathrm{E}}_{\text{potComplex}}–{\mathrm{ }\mathrm{E}}_{\text{solvComplex}}$$

4. Conclusion

We have described here computational repurposing approaches for a newly synthesised series of pyridazinone-based small-molecules. The in silico screening has been performed by pharmacophore mapping and structure-based models. Our pharmacophore iVS adopted 16159 receptor-based pharmacophore models and the results highlighted aspartate aminotransferase as a valid target for the compound dataset. Furthermore, docking calculations and molecular dynamic simulations were also used to fully establish the suitability of aspartate aminotransferase as target for this molecular series. AST catalyses the reversible endogenous interchange of aspartate and α-ketoglutaric acid to glutamic acid and oxaloacetic acid. The inhibition of this enzyme has been also investigated for selective interference for the survival of breast cancer and other malignant cells over normal mammalian cells^39,47. Clearly, novel inhibitors for this enzyme can be useful in anti- cancer studies and therapies. Biological experiments are ongoing to experimentally confirm the findings reported here and propose a number of these pyridazine-based analogues as new hit compounds for AST inhibition and linked pharmacological opportunities.

Disclosure statement

The authors declare no conflicts of interest.