Ligand-Based and Docking-Based Virtual Screening of MDM2 Inhibitors as Potent Anticancer Agents

A ligand-based and docking-based virtual screening was carried out to identify novel MDM2 inhibitors. A pharmacophore model with four features was used for virtual screening, followed by molecular docking. Seventeen compounds were selected for an in vitro MDM2 inhibition assay, and compounds AO-476/43250177, AG-690/37072075, AK-968/15254441, AO-022/43452814, and AF-399/25108021 showed promising MDM2 inhibition activities with Ki values of 9.5, 8.5, 23.4, 3.2, and 23.1 μM, respectively. Four compounds also showed antiproliferative activity, and compound AO-022/43452814 was the most potent hit with IC50 values of 19.35, 26.73, 12.63, and 24.14 μM against MCF7 (p53 +/+), MCF7 (p53 -/-), HCT116 (p53 +/+), and HCT116 (p53 -/-) cell lines, respectively. Compound AO-022/43452814 could be used as a scaffold for the development of anticancer agents targeting MDM2.


Introduction
The tumor suppressor protein p53 plays a critical role in the regulation of cell proliferation by induction of growth arrest or apoptosis [1]. Overexpression of MDM2 has been found in many cancer cells. The regulatory protein MDM2 binds to the p53 transactivation domain to suppress its transcriptional activity [2]. Therefore, the p53-MDM2 interface has emerged as an important target for the development of novel chemotherapeutic agents.
Thus far, drug resistance and toxicity are still the main challenges of MDM2 inhibitors [20]. Therefore, it is necessary to continue to discover novel inhibitors for the development of anticancer drugs. In the present study, novel MDM2 inhibitors were identified among commercially available compounds using a combination of virtual screening and biological tests. A MDM2 pharmacophore model was first established based on common features of known MDM2 inhibitors, and secondly, the virtual screening was carried out through the pharmacophore model and molecular docking. From a list of retrieved compounds, seventeen were selected for enzymatic assays and five were tested for antiproliferative activity in vitro.
the training set ( Figure 1). All seven ligands were extracted from their crystal structures of MDM2 (Table 1) and checked for bond orders. Hydrogen-bond acceptor (A), hydrogenbond donor (D), and hydrophobic (H) were specified as the pharmacophore features based on the chemical features of the compounds in the training set. The Principal and MaxO-mitFeat values were set to 2 and 0, respectively. The maximum number of hypotheses was set to ten. The maximum feature option was set to 6, and the minimum feature option was set to 4. The minimum distance between features was set to 2.0. All other parameters were at their default.
2.2. Pharmacophore Validation. Ten pharmacophore hypotheses were successfully generated. These models were validated according to the rank scores, fit values, experience, and GH score methods. To obtain the fit values, the ligands of the training set were aligned to the ten pharmacophore models using the Ligand Profiler protocol in DS 4.5. Default values were used for all parameters.
The validity of the models was evaluated using the GH score method [21], which is used to evaluate the performance of the pharmacophore model in virtual screening. A decoy set containing 1200 molecules was constructed, among which 50 molecules were known MDM2 inhibitors and the other 1150 molecules were obtained from the Specs database using the Find Similar protocol in DS 4.5. The pharmacophore model was valid with the GH scores ranging from 0.6 to 1. The GH score was calculated by the following formula: where T is the total number of molecules in the decoy set, Ta is the total number of actives in the decoy set, Ht is the number of total hits including actives and decoy molecules, Ha is the number of active hits, and GH is the Güner-Henry score.

Ligand-Based Virtual
Screening. The ligand-based virtual screening was performed using the Ligand Profiler protocol in DS 4.5. A database containing 216,347 compounds (from Specs library, http://www.specs.net) was firstly filtered by the "Rule of five." The "fitting method" was set to "flexible," and the "maximum omitted features" was set to "-1." One pharmacophore feature was allowed to be missed when the virtual screening was carried out, and other parameters were set as default. The optimal pharmacophore model (hypo 02) was used as the 3D query. The obtained ligands were then subjected to molecular docking studies.
2.4. Docking-Based Virtual Screening. The docking-based virtual screening was carried out using Sybyl-X2.0 (Tri-I Biotech, Shanghai Inc). The crystal structure of MDM2 with PDB code 3LBL was processed through the "Structure Preparation Tool" of Sybyl-X2.0. The ligand and water molecules were firstly removed, and the hydrogen atoms were added.  30 nM preincubated (30 min) MDM2 and 10 nM PMDM-F peptide in reaction buffer (100 μL, potassium phosphate, 100 mM; bovine gamma globulin, 100 mg/mL; sodium azide, 0.02%) were added into a 96-well plate containing tested compounds at pH 7.5. The plate was incubated at 37°C for 30 minutes, and the polarization values were measured (485 nm excitation, 528 nm static and polarized filter). The K i values were calculated according to a reported method [23].
The cells were seeded into a 96-well plate and incubated at 37°C in 5% CO 2 for 24 h, after which the test compounds were added to the culture medium and the cells were incubated for 48 h. 20 μL fresh MTT was added to each well, and the cells were incubated at 37°C in 5% CO 2 for 4 h. After removing the culture medium, 150 μL DMSO was added to dissolve the formazan crystals. The absorbance at 540 nm was measured using a microplate reader. The results, expressed as IC 50 values, were the average of three determinations and were calculated using nonlinear regression analysis.

Results and Discussion
3.1. Pharmacophore Modeling and Validation. Ligand-based pharmacophore models were established through DS 4.5 software. Seven structurally diverse MDM2 inhibitors were selected as the training set ( Figure 1), and the activities of the inhibitors targeting MDM2 are listed in Table 1.
The pharmacophore model was established based on the common features of the training set compounds. As suggested by the Feature Mapping protocol, hydrogen-bond acceptor (A), hydrogen-bond donor (D), and hydrophobic (H) were specified as the pharmacophore features. The top ten hypotheses were generated to identify the necessary common features to inhibit MDM2. Two groups emerged according to the hierarchical clustering analysis: AHHH and DAHH. Statistical parameters of the generated models are listed in Table S1. Rank scores of these models range from 195.19 to 218.23 kcal·mol -1 . Direct and partial hit values of "1" and "0" show that the training set ligands are mapped onto all features of the model. The max fit of all hypotheses is 4.
To select the best model from the ten pharmacophore hypotheses, three strategies were employed. (i) There is an unreasonable hypothesis (hypo 03) that the specified chemical characteristics of a small number were deleted. (ii) Four hypotheses (hypo 07, 08, 09, and 10) were not considered because of low rank scores. (iii) The remaining hypotheses were evaluated according to the calculated fit values (Table S2). Hypo 02 and hypo 04 that exhibited high fit values were identified. (iv) Hypo 02 and hypo 04 were further evaluated using the GH scoring method. A database containing 1150 inactive decoy ligands and 50 active MDM2 inhibitors was constructed to evaluate the screening ability of the pharmacophore. The results are demonstrated in Table 2. For hypo 02, among the total hits, 47 active hits were obtained, giving a yield of actives 67.1%. The enrichment factor and GH score were 16.1 and 0.72, respectively. For hypo 04, 42 active hits were obtained, giving a yield of 51.2%. The enrichment factor and GH score were 12.3 and 0.57, respectively. Considering that the hit rate and GH score are the significant parameters in the virtual screening, hypo 02 was chosen as the best pharmacophore model, including one hydrogen bond acceptor group (A) and three hydrophobic groups (H1, H2, and H3) ( Figure 2).

Virtual
Screening through the Pharmacophore Model and Molecular Docking. A subset of 216,347 commercially available compounds from Specs was used for virtual screening. "Rule of five" was applied to remove the compounds with unwanted physical and chemical properties, resulting in 147,192 compounds kept in the database. The  Figure S2). The compounds which could interact with the amino acids surrounding the active site of MDM2 were chosen. Seventeen compounds were finally selected ( Figure 3) and purchased for in vitro biological activity. The chemical names and physicochemical properties of the seventeen hit compounds are listed in Tables S3 and S4. 3.3. In Vitro Enzymatic Assays. The MDM2 binding affinity of all selected compounds was tested, and nutlin-3a was used as a positive control. The results are exhibited in Table 3.
The proposed binding mode of compound AO-476/43250177 is shown in Figures 4(a 1 ) and 4(a 2 ). Two nonclassical H-bonds and fifteen hydrophobic interactions were observed between compound AO-476/43250177 and MDM2. The nitrogen atom of the pyridine ring and the connected carbon atom serve as an accepter to form nonclassical H-bonds with Gln24 and Tyr100, respectively. The pyridine ring also makes σ-π hydrophobic interactions with Ile19 and Leu54. The carbon-carbon double bond makes π-π hydrophobic interactions with Tyr100 and His96 and σ-π hydrophobic interactions with Ile99. The triazole ring forms a σ-π hydrophobic interaction with Leu54. The dihydropyridine ring forms σ-π hydrophobic interactions with Val 75, Val93, and Ile99 and forms π-π hydrophobic interactions with Phe91. The furan ring connected to the dihydropyridine makes σ-π hydrophobic interactions with Met62 and Val93, and the other furan ring makes σ-π hydrophobic interactions with Leu57, Ile99, and Ile103.
The proposed binding mode of compound AG-690/37072075 is shown in Figures 4(b 1 ) and 4(b 2 ). One classical H-bond and eleven hydrophobic interactions were observed between compound AG-690/37072075 and MDM2. The hydroxyl group donates a classical H-bond to Gly58. The hydroxyl phenyl forms σ-π hydrophobic interactions with Leu57, Ile61, and Ile99. Chlorophenyl forms σ-π hydrophobic interactions with Leu54 and Ile99 and forms π-π hydrophobic interactions with His96. The chlorine atom     Computational and Mathematical Methods in Medicine makes σ-σ hydrophobic interactions with Met17 and Ile19. The quinoxaline ring makes σ-π hydrophobic interactions with Ile61, Met62, and Val93. The proposed binding mode of compound AK-968/15254441 is shown in Figures 4(c 1 ) and 4(c 2 ). One classical H-bond, one nonclassical H-bond, and eight hydrophobic interactions were observed between compound AK-968/15254441 and MDM2. The fluorine atom acts as a classical H-bond acceptor to Try100. The carbonyl group serves as a H-bond donor to form a nonclassical interaction with His96. The fluorophenyl makes σ-π hydrophobic interactions with Lys51 and Leu54, and the other phenyl makes a σ-π hydrophobic interaction with Met62. The methyl forms σ-π hydrophobic interactions with His96 and Tyr100 and a σ-σ hydrophobic interaction with Ile99. The triazole ring forms σ-π hydrophobic interactions with Val93 and Ile99.
The proposed binding mode of compound AO-022/43452814 is shown in Figures 4(d 1 ) and 4(d 2 ). Two nonclassical H-bonds and ten hydrophobic interactions were observed between compound AO-022/43452814 and MDM2. The carbonyl group serves as a donor to form a nonclassical H-bond interaction with Gly58. The methyl was observed to form a nonclassical H-bond interaction with Tyr67. The phenyl connected to the oxadiazole ring makes σ-π hydrophobic interactions with Leu54 and Ile99 and a π-π hydrophobic interaction with His96. The oxadiazole ring forms σ-π and π-π hydrophobic interactions with Val93 and His96, respectively. The cyclohexyl ring forms σ-σ hydro-phobic interactions with Leu57, Ile61, and Ile99. The methoxy phenyl forms π-π and σ-π hydrophobic interactions with Tyr67 and Val93.
The proposed binding mode of compound AF-399/25108021 is shown in Figures 4(e 1 ) and 4(e 2 ). One classical H-bond and eight hydrophobic interactions were observed between compound AF-399/25108021 and MDM2. The hydroxyl of carboxyl serves as a donor to form a classical H-bond interaction with Gly58. The phenyl ring connected with sulfonyl forms σ-π hydrophobic interactions with Leu54 and Ile99 and a π-π hydrophobic interaction with His96. The other phenyl ring makes π-π and σ-π hydrophobic interactions with Tyr67 and Val93, respectively. The piperidine ring makes σ-σ hydrophobic interactions with Ile61, Val75, and Val93.

Conclusions
In summary, new potent MDM2 inhibitors were identified through the integrative application of virtual screening and biological tests. In this paper, a pharmacophore model was established and successfully used for virtual screening. Five inhibitors (compounds AO-476/43250177, AG-690/37072075, AK-968/15254441, AO-022/43452814, and AF-399/25108021) were identified in enzymatic assays through the biological tests of seventeen hit compounds, with K i values 9.5, 8.5, 23.4, 3.2, and 23.1 μM, respectively. In the following antiproliferative tests, compounds AO-022/43452814 showed more potent than nutlin-3a, with IC 50 values of 19.35, 26.73, 12.63, and 24.14 μM against MCF7 (p53 +/+), MCF7 (p53 -/-), HCT116 (p53 +/+), and HCT116 (p53 -/-) cell lines, respectively. Compound AG-690/37072075 showed comparable antiproliferative activity to nutlin-3a. As the same with nutlin-3a, the compounds showed a great difference in activity in the pairing of wild type p53 cells and p53 null cells and more superior activity against wild type p53 cells. In conclusion, we consider that compounds AG-690/37072075 and AO-022/43452814 are promising lead compounds for the development of anticancer drugs targeting p53-MDM2 interactions. Further studies are ongoing to optimize the structure of the compounds and verify the mechanics.

Data Availability
All used data is within the paper.

Conflicts of Interest
The authors declare no conflict of interests.