Identification of small-molecule urea derivatives as novel NAMPT inhibitors via pharmacophore-based virtual screening

Identification of small-molecule urea derivatives as novel NAMPT inhibitors via pharmacophore-based virtual screening

Journal Pre-proofs Identification of small-molecule urea derivatives as novel NAMPT inhibitors via pharmacophore-based virtual screening Fikriye Ozgen...

4MB Sizes 2 Downloads 110 Views

Journal Pre-proofs Identification of small-molecule urea derivatives as novel NAMPT inhibitors via pharmacophore-based virtual screening Fikriye Ozgencil, Gokcen Eren, Yesim Ozkan, Sezen Guntekin-Ergun, Rengul Cetin-Atalay PII: DOI: Reference:

S0968-0896(19)31512-3 https://doi.org/10.1016/j.bmc.2019.115217 BMC 115217

To appear in:

Bioorganic & Medicinal Chemistry

Received Date: Revised Date: Accepted Date:

5 September 2019 22 October 2019 13 November 2019

Please cite this article as: F. Ozgencil, G. Eren, Y. Ozkan, S. Guntekin-Ergun, R. Cetin-Atalay, Identification of small-molecule urea derivatives as novel NAMPT inhibitors via pharmacophore-based virtual screening, Bioorganic & Medicinal Chemistry (2019), doi: https://doi.org/10.1016/j.bmc.2019.115217

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier Ltd.

Identification of small-molecule urea derivatives as novel NAMPT inhibitors via pharmacophore-based virtual screening Fikriye Ozgencil1,a, Gokcen Eren1,a,*, Yesim Ozkan2, Sezen Guntekin-Ergun3, Rengul Cetin-Atalay4 1Department

of Pharmaceutical Chemistry, Faculty of Pharmacy, Gazi University, 06330 Ankara, Turkey of Biochemistry, Faculty of Pharmacy, Gazi University, 06330 Ankara, Turkey 3Department of Medical Biology, Faculty of Medicine, Hacettepe University, 06100 Ankara, Turkey 4Cancer System Biology Laboratory (CanSyL), Graduate School of Informatics, Middle East Technical University, 06800 Ankara, Turkey aThese authors contributed equally. 2Department

*Correspondence Gokcen Eren, Department of Pharmaceutical Chemistry, Faculty of Pharmacy, Gazi University, 06330, Ankara, Turkey +903122023235 [email protected] Abstract Nicotinamide phosphoribosyltransferase (NAMPT) catalyzes the condensation of nicotinamide (NAM) with 5-phosphoribosyl-1-prophosphate (PRPP) to yield nicotinamide mononucleotide (NMN), a rate limiting enzyme in a mammalian salvage pathway of nicotinamide adenine dinucleotide (NAD+) synthesis. Recently, intracellular NAD+ has received substantial attention due to the recent discovery that several enzymes including poly(ADP-ribose) polymerases (PARPs), mono(ADP-ribose) transferases (ARTs), and sirtuins (SIRTs), use NAD+ as a substrate, suggesting that intracellular NAD+ level may regulate cytokine production, metabolism, and aging through these enzymes. NAMPT is found to be upregulated in various types of cancer, and given its importance in the NAD+ salvage pathway, NAMPT is considered as an attractive target for the development of new cancer therapies. In this study, the reported NAMPT inhibitors bearing amide, cyanoguanidine, and urea scaffolds were used to generate pharmacophore models and pharmacophore-based virtual screening studies were performed against ZINC database. Following the filtering steps, ten hits were identified and evaluated for their in vitro NAMPT inhibitory effects. Compounds GF4 (NAMPT IC50=2.15±0.22 μM) and GF8 (NAMPT IC50=7.31±1.59 μM) were identified as new urea-typed inhibitors of NAMPT which also displayed cytotoxic activities against human HepG2 hepatocellular carcinoma cell line with IC50 values of 15.20±1.28 and 24.28±6.74 μM, respectively. Keywords NAMPT, Pharmacophore modeling, Virtual screening, HepG2 1. Introduction Nicotinamide adenine dinucleotide (NAD+) is a critical redox cofactor in a wide range of cellular reactions and has emerged as an important regulator of many types of disease condition, most notably, cancers [1]. Poly(ADP-ribose) polymerases (PARPs),

mono(ADP-ribose) transferases (ARTs), and sirtuins (SIRTs) use NAD+ as substrate in their catalytic activity and recycle nicotinamide (NAM) produced during NAD+ consumption [2]. In cancer cells, NAD+ is rapidly consumed because of increased demand for ATP and high activity of NAD+-consuming enzymes [3]. Therefore, malignant cells are more sensitive to NAD+ as compared to normal cells and requires the constant resynthesis of NAD+ in order to maintain sufficient levels for cell survival [4]. Biochemically tryptophan, nicotinic acid (NA) and NAM are three major precursors for NAD+ synthesis in mammalian cells [5] and among these salvage pathways, the pathway which recycles NAM back to NAD+ is the main and efficient source of NAD+ [6]. This biosynthetic process is a 2-step conversion. The first step which is the ratedetermining step relies on nicotinamide phosphoribosyltransferase (NAMPT) to facilitate the condensation reaction between NAM and 5-phosphoribosyl-1pyrophosphate (PRPP) to generate nicotinamide mononucleotide (NMN). Subsequently NMN is converted into NAD+ catalyzed by nicotinate/nicotinamide mononucleotide adenyltransferase (NMNAT) [7,8]. Due to the high NAD+ requirement for survival and proliferation, various cancer cells are highly dependent on NAMPT activity. Given that NAMPT is the rate-limiting enzyme in a key NAD+ recycling pathway and observed overexpression of NAMPT across many cancers including colorectal, ovarian, breast, gastric, prostate, thyroid cancers, endometrial carcinomas, melanoma, gliomas and astrocytomas [3,9], reduction of intracellular NAD+ levels via inhibition of NAMPT activity may be an appropriate mechanism to target the malignant cells [7]. Thus, NAMPT has emerged as a promising target for the development of anticancer agents. To date a number of NAMPT inhibitors with different chemical scaffolds have been described (Fig. 1) and three of them have reached human clinical trials [10–12]. However, the clinical development of FK866 and CHS828 has been hampered by dose-limiting thrombocytopenia and gastrointestinal toxicity [13,14] while a dual PAK4/NAMPT inhibitor KPT-9274 [15] is currently being evaluated. For this reason, discovering more potent NAMPT inhibitors with less side effects are still needed.

Fig. 1. Chemical structures of some reported NAMPT inhibitors

In effort to identify novel inhibitors targeting NAMPT, we reported herein generation and validation of pharmacophore models based on a set of known active NAMPT inhibitors with amide, urea and cyanoguanidine skeleton and application of these models in the virtual screening of ZINC database. Retrieved hit compounds were subjected to biological evaluation in order to identify their in vitro NAMPT inhibitory activity. The compounds which exhibited NAMPT inhibitory potency were further tested for their in vitro cytotoxic effects on human HepG2 hepatocellular carcinoma cell line. It is aimed that the results obtained from this study may represent enzyme-inhibitor recognition at molecular level and provide new insights that can be used to discover novel therapeutic agents targeting cancer and metabolic diseases. 2. Material and methods 2.1. Computational Studies The pharmacophore modeling and virtual screening studies were carried out using Schrodinger Small-Molecule Drug Discovery Suite (Small-Molecule Drug Discovery Suite 2018-1, Schrödinger, LLC, New York, NY, 2018). 2.1.1. Dataset generation and alignment The dataset used to generate the pharmacophore model comprises of 27 reported NAMPT inhibitors [2,16–26] and a publicly available x-ray coligand structure of NAMPT. The molecules which were built via builder panel in Maestro were then

subjected to ligand preparation by LigPrep (Schrödinger Release 2018-1: LigPrep, Schrödinger, LLC, New York, NY, 2018) using default conditions. After converting to 3D, conformation search was carried out to generate conformers and search for low energy structures using OPLS3 force field and other default parameters by ConfGen tool (Schrödinger Release 2018-1: ConfGen, Schrödinger, LLC, New York, NY, 2018). The collection of these conformers was then flexibly aligned separately onto bioactive conformation of publicly available x-ray coligand structures (PDBs: 4LTS, 4JNM, 4M6P) using Flexible Ligand Alignment tool prior to model generation yielding three different aligned datasets (A1-3) comprising 28 molecules. 2.1.2. Protein preparation and receptor grid generation The crystal structures of NAMPT (PDBs: 4M6P [19], 4LTS [21], 4JNM [24]) was retrieved from the Protein Data Bank. The proteins were prepared using the Protein Preparation Wizard tool. Water molecules except HOH709, HOH715, HOH722, HOH731, HOH737, HOH743, HOH769, HOH826, HOH873, HOH894, HOH988, HOH1005, HOH1006, HOH1013, HOH1014 from chain A and HOH710 from chain B were removed from the crystallographic structures followed by addition of hydrogen atoms. All atom charges and atom types were assigned. Finally, energy minimization and refinement of the structures were done up to 0.3 Å RMSD by applying OPLS-3 force field. Centroid of the residues, predicted by x-ray coligand was defined as the grid box. Van der Waals scaling factor 1.00, charge partial cutoff 0.25 and OPLS-3 force filed were used for receptor grid generation. 2.1.3. Pharmacophore model generation Pharmacophore feature sites for the molecules were assigned using a set of features defined in Phase as hydrogen bond acceptor (A), hydrogen bond donor (D), hydrophobic group (H), negatively charged group (N), positively charged group (P), and aromatic ring (R). Common pharmacophore hypothesis: Three pharmacophore models were generated for mutually aligned conformations of 28 compounds including 27 known active NAMPT inhibitors and a x-ray coligand using Develop Pharmacophore Hypothesis protocol of Phase [27]. The in vitro NAMPT inhibitory data of the compounds used for pharmacophore models was presented as IC50 values and these values were then converted to pIC50. Compounds were categorized into an active and inactive set based on a pIC50 value which has greater than 5.5 and less than 5.5, respectively, which resulted in 19 active and 9 inactive compounds. Common pharmacophore hypotheses were generated, scored and ranked based on active and inactive survival scores. E-pharmacophore hypothesis: Three e-pharmacophore models depicting protein residues important for ligand binding were developed from three NAMPT inhibitorprotein complexes (PDBs: 4M6P [19], 4LTS [21], 4JNM [24]) using Phase [28]. Protein binding sites were defined using the centroids of residues involved in inhibitor binding in all complexes. A Van der Waals scaling factor of 0.5 was used for excluded volumes in each hypothesis to account for the shape of the protein binding site. 2.1.4. Validation of hypothesis The selected pharmacophore hypotheses were further evaluated for the classification performance by reputable methods. A database consists of 50 structurally diverse NAMPT inhibitors [2,16–26] and 1240 NAMPT decoys which were generated from

ZINC drug-like database by Decoy finder [29] was used to calculate several indices including accuracy, precision, recall, enrichment factor (EF) and goodness of hit score (GH) to find out the efficiency of the selected pharmacophore hypotheses. Finally, ROC analysis was performed to determine of the most appropriate cutoff value and additional performance indicators such as robust initial enhancement (RIE), receiver operating characteristic curve value (ROC), area under the accumulation curve (AUAC), and Boltzmann-enhanced discrimination of receiver operating characteristic (BEDROC) were calculated [30,31]. 2.1.5. Virtual screening The validated pharmacophore models were used as a query and ZINC drug-like database (13 million compounds) was used for the virtual screening campaign. Fast Flexible Search method was applied to screen the databases, the compounds were selected only if it maps all the chemical features present in validated model and ranked in order to pharmacophore fitness score. Considering the model exhaustion point determined from ROC curve, the best scoring percentage of selected compounds were further sorted by applying the Lipinski’s rule of five, ADMET and visual inspection. The compounds which passed all filters were subjected to Virtual Screening Workflow (VSW) [32] which uses Glide docking protocol to rank the compounds which utilizes the multi-step workflow; (i) Glide high throughput virtual screening (HTVS), (ii) standard precision (SP), (iii) extra precision (XP) and (iv) postprocess with MM-GBSA to find the suitable orientation in the active site of NAMPT. 2.1.6. Molecular docking Molecular docking studies were carried out using Glide docking protocol granting full flexibility to the ligands. The HTVS step collected 5 poses with keeping 100% best configuration and subjected to SP scoring function. SP was performed on the output complexes in order to reduce the initially collected 5 poses per compound to 3. A rescoring of the top-ranking SP pose of each compound was then performed with the XP scoring function of Glide and output poses were resorted by postprocess with Prime MM-GBSA calculation on the final poses to obtain ligand binding energies. The compounds which are docked most favorably were ranked based on XP Gscore. 2.2. Biological Studies 2.2.1. In vitro NAMPT inhibition assay NAMPT inhibition assay was performed using a NAMPT Inhibitor Screening Assay Kit (BPS-71276), according to the supplier’s protocol [33]. The compounds were tested at concentrations starting from 300 μM to 10 μM in duplicates. The compounds which showed inhibitory activity below 10 μM, were tested starting from 10 μM to 0.1 μM. Briefly, the master mixture (30 µl/well) consisting of 12.5 μl NAMPT assay buffer + 2.5 μl ATP + 2.5 μl NAM + 2.5 μl PRPP + 2.5 μl 30% ethanol + 12.5 μl water was added to every well. Then, 10 μl of inhibitor solution in water:DMSO (DMSO concentration is 1% in the final reaction mixture) of each well was added. The reaction was started by adding 10 μl of diluted NAMPT enzyme to the wells. The mixture was incubated at 30 °C for 2 h. After the reaction, the fluorescence of the wells was measured on a fluorometric plate reader with excitation at 340 nm and emission at 460 nm. The value of inhibition% was calculated from the fluorescence readings of inhibited wells relative to those of control wells.

2.2.2. Cell culture and NCI-60 sulforhodamine B assay Human HepG2 hepatocellular carcinoma cells were grown in Dulbecco's Modified Eagles Medium (DMEM) supplemented with 10% fetal bovine serum (GIBCO, Invitrogen), 1% non-essential amino acids (GIBCO, Invitrogen). All media contained 100 units/ml penicillin and streptomycin and cells were maintained at 37 °C in a humidified incubator under 5% CO2. Cells were plated in 96-well plates at the density of 3500 cells/well and grown for 24 h in an incubator. The compounds were dissolved in DMSO (0.1%) and were prepared 20 mM stock solution. The compounds were tested starting from 100 μM to 6.25 μM in triplicates. The reference compound FK866, which was below 6.25 μM, was tested starting from 3.12 μM to 0.03 μM. After 72 h of incubation, cells were fixed using 10% (v/v) trichloroacetic acid for an hour. The fixed plates were dried and samples were stained with sulforhodamine B (SRB) solution (50 μl of a 0.4% (m/v) of SRB in 1% acetic acid solution) for 10 min. The excess amount of SRB dye was discarded by washing samples with 1% acetic acid and left for airdrying. The protein bound SRB dye was dissolved in 10 mM Tris-base and its absorbance was measured with 96-well plate reader at 515 nm [34]. The IC50 values were calculated based on DMSO control normalization. 3. Results and discussion 3.1. Pharmacophore modeling A pharmacophore model represents the chemical features and their corresponding locations which are essential for a molecule to interact with a target protein. Ligandbased approach reveals the significant and common chemical features from 3D structures of a set of known ligands while the structure-based pharmacophore modeling works directly with the 3D structure of a protein or a protein-ligand complex analyzing the chemical features of the active site and key interaction points between the protein and its co-crystallized ligand. Although various ligand-based and structurebased approaches have been successfully applied in virtual screening, there are still challenging problems in many cases [35]. In order to minimize these problems and further improve the effect, we have implemented a strategy that combines ligandbased with structure-based pharmacophore methods and also with molecular docking [36,37]. In order to construct the pharmacophore models, 27 amide, cyanoguanidine and urea derivatives reported as NAMPT inhibitor and publicly available x-ray coligand structures in complex with NAMPT (PDBs: 4M6P, 4LTS, 4JNM) were gathered from published data [2,16–26] and three dataset consisting of 28 compounds (27 NAMPT inhibitors and a x-ray coligand) in each set were created. Compounds in each dataset were categorized into an active and inactive set based on a pIC50 value which has greater than 5.5 and less than 5.5, respectively, resulted in 19 active and 9 inactive compounds (Table S1 in Supplementary Material). With the aim of sampling all feasible conformations of the dataset compounds except available co-crystallized ligands, the conformational search was performed as an initial stage using ConfGen tool implemented in Maestro. After obtaining a large number of conformers, the next stage in building the pharmacophore model was to align the obtained conformers at the manner of finding common groups overlap. Since the

bioactive conformation of three co-crystallized ligands in dataset had already been obtained, these coligands were used as reference ligands during the Flexible Ligand Alignment process, and the resulting mutual alignments (A1-3) were directly used to define the pharmacophoric features in generation of common pharmacophore hypotheses (Fig. 2).

Fig. 2. Overlay of the dataset compounds flexibly aligned to bioactive conformations of x-ray coligands in complex with NAMPT. (A) Alignment 1, A1: x-ray coligand is represented as green ball and stick (PDB: 4M6P), (B) Alignment 2, A2: x-ray coligand is represented as magenta ball and stick (PDB: 4LTS), and (C) Alignment 3, A3: x-ray coligand is represented as cyan ball and stick (PDB: 4JNM). The remaining dataset compounds are shown in gray sticks.

The common pharmacophore hypotheses were scored and ranked and the variants that achieved the best scores were selected from each aligned set. The seven-featured variant (Model 1) with a survival score 4.38, site score 0.57, vector score 0.94, volume score 0.68, and phase inactive score 2.09 was the selected hypothesis generated from A1. The six-featured hypothesis (Model 2) based on A2 had the best survival score of 4.20, site score of 0.42, vector score of 0.90, volume score of 0.58, and phase inactive score of 2.13. The variant (Model 3) that comprises of six features with a survival score 4.75, site score 0.98, vector score 0.99, volume score 0.66, and phase inactive score 2.32 was selected as the best common pharmacophore model derived from A3 (Table 1). In addition to common pharmacophore models, e-pharmacophore method [28] that combines pharmacophore perception with protein-ligand energetic terms to rank the importance of pharmacophore features in ligand binding was also used and applied to three NAMPT x-ray crystal structures (PDBs: 4M6P, 4LTS, 4JNM). The results were a

four-featured pharmacophore model derived from 4M6P (Model 4) and five-featured pharmacophore models so-called Model 5 and 6 derived from 4LTS and 4JNM, respectively (Table 1). The e-pharmacophore models were together with a set of excluded volume spheres denoting the steric boundaries of the NAMPT active site. Table 1. The pharmacophore models generated Pharmacophore model*

Feature

Provenance

Model 1

AAADRRR

Common pharmacophore (derived from A1)

Model 2

AADDRR

Common pharmacophore (derived from A2)

Model 3

AAADRR

Common pharmacophore (derived from A3)

Model 4

DRRR

E-pharmacophore (derived from 4M6P)

Model 5

ADRRR

E-pharmacophore (derived from 4LTS)

Model 6

ADRRR

E-pharmacophore (derived from 4JNM)

*The pharmacophore features are represented as follows: hydrogen-bond acceptor (A; red sphere with vectors), aromatic group (R; orange ring), hydrogen-bond donor (D; blue sphere with vector). The shared pharmacophore features were shown as the superposition of the compound with the highest fitness score.

The developed pharmacophore models were further challenged based on common features matching against a decoy set containing 50 structurally diverse NAMPT inhibitors apart from dataset [2,16–26] (Table S2 in Supplementary Material) and 1240 inactive compounds and the ranking of actives was statistically analyzed in terms of ROC sensitivity and a variety of well-established metrics. The decoy set was screened with models using the Phase Ligand Screening tool with flexible fitting method and the results were presented in Table 2.

Table 2. Results of pharmacophore models validation Statistical parameter* D A H TP FN FP TN % Yield of actives % Ratio of actives Se Sp Accuracy Precision GH ROC BEDROC (α=20) RIE AUAC EF (% 1)

Model 1 1290 50 23 20 30 3 1237 87 40 0.40 0.99 0.97 0.87 0.75 0.40 0.48 6.75 0.69 22

Model 2 1290 50 27 21 29 6 1234 78 42 0.42 0.99 0.97 0.78 0.69 0.42 0.50 7.01 0.70 24

Value Model 3 Model 4 1290 1290 50 50 35 17 28 15 22 35 7 2 1233 1238 80 88 56 30 0.56 0.30 0.99 0.99 0.98 0.97 0.80 0.88 0.74 0.74 0.56 0.30 0.63 0.38 8.92 5.30 0.77 0.64 26 22

Model 5 1290 50 58 22 28 36 1204 38 44 0.44 0.97 0.95 0.38 0.38 0.44 0.51 7.11 0.70 26

Model 6 1290 50 49 24 26 25 1215 49 48 0.48 0.98 0.96 0.49 0.48 0.48 0.55 7.79 0.73 26

*D: Total number of molecules in database; A: Total number of actives in database; H: Total number of hit compounds from database; TP: Total number of active compounds in the hit list, true positives; FN: False negatives; FP: False positives; TN: True negatives; % Yield of actives: [(TP/H)x100]; % Ratio of actives: [(TP/A)x100]; Se: Sensitivity or Recall, [TP/(TP+FN)]; Sp: Specificity, [TN/(TN+FP)]; Accuracy: [(TP+TN)/D]; Precision: [TP/(TP+FP)]; GH: Goodness of hit score, [(TP/4HA)(3A+H)(1-((H-TP)/(D-A)))]; ROC: Receiver Operating Characteristic; BEDROC: Boltzmann-Enhanced Discrimination of ROC; AUAC: Area under the accumulation curve; EF: Enrichment factor, [(TP/H)/(A/D)].

According to the results of pharmacophore model validation step, Model 1 recognized 20 of 50 actives, predicted 30 active compounds to be inactive compounds (false negative), and 3 inactives to be active compounds (false positive). While Model 2 has shown similar results compared to that of Model 1, Model 3 had a better efficiency of the screening the active from the inactive molecules. In case of Models 4-6, % yield of actives was found to be lower than the others. The GH values indicated that Models 5 and 6 had poor classification performance for estimating the actives over a variety of compounds. The calculated ROC and BEDROC values for Model 4 suggested that this model could not be successfully employed in virtual screening studies. As a result, all calculated parameters for assessing the quality of pharmacophore models suggested that Models 1-3 have high predictive power for estimating the activities over a variety of compounds and this makes them optimal pharmacophore models for the virtual screening. The obtained matched hit list for each database screening using Models 1-3 was also statistically analyzed in terms of ROC curve which is a graphic representation of the relation existing between the sensibility and the specificity of a test and generated by plotting the fraction of true positives out of the total actual positives versus the fraction of false positives out of the total actual negatives [38]. As shown in Fig. 3, the calculated ROC curve patterns showed that database screening based on Model 1 and 2 give good results if a reduced part (4%) of the database is sampled. When the pharmacophore model search was enlarged above the level of 4% of the database, the classification performance decreased defining the “model exhaustion point” [31]. In case of Model 3, the obtained pattern indicated better results with an enlarged threshold close to 6% compared to that of Models 1 and 2 suggesting Model 3 was the most selective pharmacophore model in virtual screening process.

Fig. 3. The ROC curves obtained from the total hit list (active compounds and decoys) ranked based on Models 1-3.

3.2. Virtual screening A pharmacophore-based virtual screening campaign which helped in identification of novel potential inhibitors of NAMPT was performed. The validated pharmacophore models were used as a 3D query for screening the “ZINC clean drug-like” database [39] comprised of 13 million compounds to retrieve new scaffolds. Virtual screening workflow is illustrated in Fig. 4.

Fig. 4. Virtual screening workflow

The hits with the best pharmacophore fitness score were obtained from initial screening. The ROC curve which used to check whether the pharmacophore model is able to select active compounds with respect to decoys, presented that active compounds of the screening database were ranked in the top 4% for Models 1 and 2 while close to 6% for Model 3. Considering the high number of hits, the threshold was kept under the obtained cutoff values; the top 3% and 2% of the hits obtained from initial screening were extracted for further refinement in case of Models 1-2 and Model 3, respectively. In order to separate the compounds with drug-like properties, the hits

were filtered using a Lipinski’s rule of five, ADMET and visual inspection. Based on these properties, the selected compounds were further subjected to Virtual Screening Workflow (VSW) which uses Glide docking protocol to rank the compounds which utilizes the multi-step workflow; (i) Glide high-throughput virtual screening (HTVS), (ii) standard precision (SP), (iii) extra precision (XP) and (iv) postprocess with MM-GBSA to find the suitable orientation in the active site of NAMPT structures. The compounds ranked by XP Gscore were retained for visual inspection to retrieve hits with better binding affinity in a good binding orientation in NAMPT active site as well as possessing high pharmacophore fitness score values. This step reduced the number of remaining compounds to 31 for overall process. Finally, 10 hit compounds (Table 3) which satisfied all the drug-like properties and formed crucial interactions in NAMPT active site with well fitness for the pharmacophore model were selected and purchased. Table 3. Selected hits identified by pharmacophore-based virtual screening XP Compd ZINC code Provenance Gscore

Fitness score

O

GF1

F

N H N

O

ZINC36391579

-7.00

Model 1

1.66

ZINC32747605

-6.76

Model 1

1.42

ZINC58210520

-6.69

Model 1

1.41

ZINC48020409

-8.66

Model 3

1.74

ZINC32743681

-6.83

Model 2

1.70

ZINC21559194

-7.80

Model 3

1.72

ZINC32930787

-6.44

Model 1

1.38

ZINC48354269

-8.00

Model 3

1.68

F

O N

GF2

O N

O

N

N H O

N

GF3

O N

N H

N H

O

O O

GF4

O N H

N H

N

O O

GF5

H N O

N H

O O

GF6

N

GF7

O

N S

N H

O

F

O

N

N H

F

O

GF8

O

O N H

N H

N N

N

GF9

O N S O

O

ZINC13022964

-7.59

Model 3

1.66

ZINC40089377

-6.84

Model 1

1.51

N H

O

GF10

N H

O

O

3.3. Biological evaluation The purchased hit compounds were initially screened for their in vitro NAMPT inhibitory activity starting from 300 μM to 10 μM. The compounds which showed inhibitory activity below 10 μM, were tested starting from 10 μM to 0.1 μM. FK866 was used as the reference compound. The results are presented separately in graphical form (Fig. 5).

Fig. 5. The inhibition% of NAMPT activity in the presence of non-fluorogenic hit compounds and FK866 tested at 300-10 µM.

Among the hit compounds GF3, showed no inhibition against NAMPT at concentrations tested. On the other hand, compounds GF6 and GF9 were identified as fluorogenic compounds and were not appropriate for inhibition profile determination using our biochemical assay for NAMPT activity. This will be investigated as a subject of further studies due to the importance of the development of fluorescent probe targeting NAMPT [40]. Comparison of the inhibition% values relative to FK866, the reference compound defining the effects of the hit compounds on NAMPT activity suggested that compounds GF4 and GF8 displayed NAMPT inhibitory potency, but GF8 showed lower than that of FK866 at concentrations tested. In case of remaining compounds, the NAMPT activity was also reduced but inhibition% values were found to be below 50%. Further, using three concentrations (3, 1 and 0.1 μM) GF4 and GF8 were again tested in vitro against NAMPT to make logarithmic-dose vs response curves for determining IC50 values using GraphPad Prism 8 (demo version), and the IC50 values were calculated as 2.15±0.22 μM and 7.31±1.59 μM for the NAMPT inhibitory effect of compounds GF4 and GF8, respectively (Fig. 6).

Fig. 6. Logarithmic-dose vs response curves used to generate IC50 values of NAMPT inhibition for compounds GF4 and GF8 presented as blue and red lines, respectively. For the curve of compound GF4: Hill slope is 2.15; R2 is 0.9993; DF is 3. For the curve of compound GF8: Hill slope is 2.847; R2 is 0.9964; DF is 4.

Given the fact that NAMPT was predominantly localized to hepatocytes and especially in HepG2 hepatocytes strong NAMPT immunostaining was shown [41], the inhibition on HepG2 cell viability was evaluated for two compounds GF4 and GF8 in a dose dependent manner for 72 h (Fig. 6).

Fig. 7. The Inhibition% of cell viability in HepG2 cell line treated with different concentrations of compounds GF4 and GF8 for 72 h.

According to obtained results against HepG2 cell line, compound GF4 exhibited a good dose-dependent inhibitory effect on cell viability with an IC50 value of 15.20±1.28 μM (Fig. 8). However, compound GF8 showed much lower inhibitory activity with an IC50 value of 24.28±6.74 μM for the assayed time period suggesting the correlation between their in vitro NAMPT inhibitory effects and the degree of cytotoxicity toward the cancer cell line tested. The IC50 value was calculated as 18.72±5.18 nM for the reference compound FK866 which was found to be in a good agreement with reported data [10].

Fig. 8. Logarithmic-dose vs response curves used to generate IC50 values of HepG2 cell viability inhibition for compounds GF4, GF8 and FK866 presented as blue, red and purple lines, respectively. For the curve of compound GF4: Hill slope is 1.20; R2 is 0.9981; DF is 3. For the curve of compound GF8: Hill slope is 1.13; R2 is 0.9743; DF is 3. For the curve of FK866: Hill slope is 1.152; R2 is 0.9938; DF is 9.

3.4. Predicted binding mode of compounds GF4 and GF8 in NAMPT active site Compounds GF4 and GF8 showed the best inhibitory activity against NAMPT among the hit compounds, their binding modes in NAMPT active site were thus analyzed. The human NAMPT in complex with a urea-containing NAMPT inhibitor (PDB: 4JNM) was selected as receptor. The compounds bound to a site formed by the interface of two NAMPT protein monomers and stabilized by π-π stacking contacts, direct/watermediated hydrogen bonds suggesting good agreement with x-ray coligand conformation in NAMPT active site. The amino pyridine ring of the x-ray coligand was stabilized by forming π-π stacking contacts with Phe193 and Tyr18’ (from chain B) and π-cation interaction with Phe193. The amino group of pyridine ring was found to be interacted with Asp16’ (from chain B) via hydrogen bonding. The urea NH groups participated in two hydrogen bonds with a bound water molecule (HOH737), which in turn made hydrogen bonds to the side chain of Asp219 and the backbone carbonyl oxygen of Val242. The urea carbonyl oxygen atom of the coligand was placed along the side chain of Ser275 forming a hydrogen bond. The aniline-phenylsulfone moiety extended through a hydrophobic and aromatic tunnel and formed π-π stacking contact with His191, while the terminal phenyl moiety contacted the solvent exposed surface of the protein. The sulfone oxygen was found to be interacted with Val350 and Lys189 via two water molecules (HOH873, HOH715) [24] (Fig. 9).

Fig. 9. The predicted binding mode for urea-containing x-ray coligand (orange ball and sticks) in NAMPT active site (PDB: 4JNM). Two NAMPT protein monomers A and B and their residues were depicted as green and cyan, respectively. Hydrogen bonds, π-π stacking and π-cation contacts were shown as dotted yellow, cyan and green lines, respectively.

The compound GF4 occupied a relatively extended binding site that is ordinarily occupied by phosphorylated ribose forming π-π stacking contacts with Phe193 and Tyr18’ (from chain B) by pyridine ring. The urea carbonyl oxygen atom of compound GF4 was positioned near the side chain of Ser275 forming a hydrogen bond. The urea NH groups participated in two hydrogen bonds with a bound water molecule (HOH737), whichin turn made hydrogen bonds to the side chain of Asp219 and the backbone carbonyl oxygen of Val242. The phenyl ring extended through a hydrophobic and aromatic region interacting with His191 by π-π stacking contact. The ester chain substituted to phenyl ring contacted the cavity of the protein exposed to the solvent surface (Fig.10).

Fig. 10. The predicted binding mode for compound GF4 (purple ball and sticks) in NAMPT active site (PDB: 4JNM). Two NAMPT protein monomers A and B and their residues were depicted as green and cyan, respectively. Hydrogen bonds and π-π stacking contacts were shown as dotted yellow and cyan lines, respectively.

Compound GF8 showed a similar interaction pattern compared to compound GF4 forming π-π stacking interactions between phenyl ring and Phe193 and Tyr18’ (from chain B). A water-mediated hydrogen bond was formed between nitrogen atom of urea moiety providing two hydrogen bonds with the side chains of Asp219, Val242 while the carbonyl oxygen was interacting with Ser275 by a hydrogen bond. The contacts that formed by the ester group in the side chain of compound GF4, were missing for compound GF8 due to reversed orientation in active site compared to that of compound GF4. Since the terminal diazole bound pyridine moiety occupied the solvent exposed area, it was expected to interact with HOH731 likely carbonyl oxygen atom in the structure of compound GF4. However, the distance between nitrogen of pyridine and water molecule (3.85 Å) was not close enough that an optimal hydrogen bond interaction may not be formed between the two moieties. (Fig. 11).

Fig. 11. The predicted binding mode for compound GF8 (yellow ball and sticks) in NAMPT active site (PDB: 4JNM). Two NAMPT protein monomers A and B and their residues were depicted as green and cyan, respectively. Hydrogen bonds and π-π stacking contacts were shown as dotted yellow and cyan lines, respectively.

4. Conclusions In this study, a virtual screening protocol was developed taking advantages of a strategy which combines both ligand- and structure-based approaches and molecular docking as well. The protocol provided to identify 10 potential hits which formed crucial interactions in NAMPT active site and presented drug-like properties. The identified compounds were evaluated for their ability to inhibit NAMPT activity in vitro. Compounds GF4 and GF8 which exhibited inhibitory effect on NAMPT activity, were also tested for cell growth inhibition potency against human HepG2 hepatocellular carcinoma cell line. Compound GF4 displayed an inhibitory potency to NAMPT with an IC50 value of 2.15±0.22 μM while compound GF8 was with lower potency (IC50=7.31±1.59 μM). The cytotoxic activities of compounds were found to be correlated with NAMPT inhibitory profiles suggesting that compounds were able to inhibit cell growth in cancer cells by modulating NAMPT activity. Additionally, these compounds were identified as hits by screening with Model 3 which was expected due to being the best pharmacophore model generated among the all. Based on the results, compounds GF4 and GF8 as a new chemical scaffold were chosen as lead

compounds for the design of novel potent NAMPT inhibitors since they had good potential for further optimization to improve their activities. Supplementary Material Table S1. Dataset compounds used for generating the pharmacophore model Table S2. The chemical structures of known SIRT2 inhibitors used in database screening of ARRR.22 for pharmacophore model validation Acknowledgements This work was financially supported by the Scientific and Technological Research Council of Turkey (TUBITAK SBAG 118S726). References [1] Sharif T, Martell E, Dai C, Ghassemi-Rad MS, Kennedy BE, Lee PWK, et al. Regulation of Cancer and Cancer-Related Genes via NAD+. Antioxidants & Redox Signaling 2018;30:906-23. [2] Chen W, Dong G, He S, Xu T, Wang X, Liu N, et al. Identification of benzothiophene amides as potent inhibitors of human nicotinamide phosphoribosyltransferase. Bioorganic & Medicinal Chemistry Letters 2016;26:765-8. [3] Sampath D, Zabka TS, Misner DL, O’Brien T, Dragovich PS. Inhibition of nicotinamide phosphoribosyltransferase (NAMPT) as a therapeutic strategy in cancer. Pharmacology & Therapeutics 2015;151:16-31. [4] Chen H, Wang S, Zhang H, Nice EC, Huang C. Nicotinamide phosphoribosyltransferase (Nampt) in carcinogenesis: new clinical opportunities. Expert Review of Anticancer Therapy 2016;16:827-38. [5] Revollo JR, Grimm AA, Imai S. The regulation of nicotinamide adenine dinucleotide biosynthesis by Nampt/PBEF/visfatin in mammals: Current Opinion in Gastroenterology 2007;23:164-70. [6] Rongvaux A, Andris F, Van Gool F, Leo O. Reconstructing eukaryotic NAD metabolism. Bioessays 2003;25:683-90. [7] Zak M, Yuen P, Liu X, Patel S, Sampath D, Oeh J, et al. Minimizing CYP2C9 Inhibition of Exposed-Pyridine NAMPT (Nicotinamide Phosphoribosyltransferase) Inhibitors. Journal of Medicinal Chemistry 2016;59:8345-68. [8] Eren G. Homology Modeling of Human Nicotinamide/Nicotinic Acid Mononucleotide Adenylyltransferase 2: Insights into Isoenzyme-Specific Differences Using Molecular Docking Simulatons. Letters in Drug Design & Discovery 2017; 14: 727-36. [9] Shackelford RE, Mayhall K, Maxwell NM, Kandil E, Coppola D. Nicotinamide Phosphoribosyltransferase in Malignancy. Genes Cancer 2013;4:447-56. [10] Hasmann M, Schemainda I. FK866, a Highly Specific Noncompetitive Inhibitor of Nicotinamide Phosphoribosyltransferase, Represents a Novel Mechanism for Induction of Tumor Cell Apoptosis. Cancer Research 2003;63:7436-42.

[11] Beauparlant P, Bédard D, Bernier C, Chan H, Gilbert K, Goulet D, et al. Preclinical development of the nicotinamide phosphoribosyl transferase inhibitor prodrug GMX1777. Anti-Cancer Drugs 2009;20:346-54. [12] Watson M, Roulston A, Bélec L, Billot X, Marcellus R, Bédard D, et al. The Small Molecule GMX1778 Is a Potent Inhibitor of NAD+ Biosynthesis: Strategy for Enhanced Therapy in Nicotinic Acid Phosphoribosyltransferase 1-Deficient Tumors. Molecular and Cellular Biology 2009;29:5872-88. [13] Holen K, Saltz LB, Hollywood E, Burk K, Hanauske A-R. The pharmacokinetics, toxicities, and biologic effects of FK866, a nicotinamide adenine dinucleotide biosynthesis inhibitor. Invest New Drugs 2008;26:45-51. [14] von Heideman A, Berglund A, Larsson R, Nygren P. Safety and efficacy of NAD depleting cancer drugs: results of a phase I clinical trial of CHS 828 and overview of published data. Cancer Chemother Pharmacol 2010;65:1165-72. [15] Neggers JE, Kwanten B, Dierckx T, Noguchi H, Voet A, Bral L, et al. Target identification of small molecules using large-scale CRISPR-Cas mutagenesis scanning of essential genes. Nature Communications 2018;9:502. [16] Zheng X, Bauer P, Baumeister T, Buckmelter AJ, Caligiuri M, Clodfelter KH, et al. Structure-Based Identification of Ureas as Novel Nicotinamide Phosphoribosyltransferase (Nampt) Inhibitors. Journal of Medicinal Chemistry 2013;56:4921-37. [17] Zheng X, Bauer P, Baumeister T, Buckmelter AJ, Caligiuri M, Clodfelter KH, et al. Structure-Based Discovery of Novel Amide-Containing Nicotinamide Phosphoribosyltransferase (Nampt) Inhibitors. Journal of Medicinal Chemistry 2013;56:6413-33. [18] Dragovich PS, Bair KW, Baumeister T, Ho Y-C, Liederer BM, Liu X, et al. Identification of 2,3-dihydro-1H-pyrrolo[3,4-c]pyridine-derived ureas as potent inhibitors of human nicotinamide phosphoribosyltransferase (NAMPT). Bioorganic & Medicinal Chemistry Letters 2013;23:4875-85. [19] Zheng X, Bair KW, Bauer P, Baumeister T, Bowman KK, Buckmelter AJ, et al. Identification of amides derived from 1H-pyrazolo[3,4-b]pyridine-5-carboxylic acid as potent inhibitors of human nicotinamide phosphoribosyltransferase (NAMPT). Bioorganic & Medicinal Chemistry Letters 2013;23:5488-97. [20] Dragovich PS, Zhao G, Baumeister T, Bravo B, Giannetti AM, Ho Y-C, et al. Fragment-based design of 3-aminopyridine-derived amides as potent inhibitors of human nicotinamide phosphoribosyltransferase (NAMPT). Bioorganic & Medicinal Chemistry Letters 2014;24:954-62. [21] Zheng X, Baumeister T, Buckmelter AJ, Caligiuri M, Clodfelter KH, Han B, et al. Discovery of potent and efficacious cyanoguanidine-containing nicotinamide phosphoribosyltransferase (Nampt) inhibitors. Bioorganic & Medicinal Chemistry Letters 2014;24:337-43. [22] Clark DE, Waszkowycz B, Wong M, Lockey PM, Adalbert R, Gilley J, et al. Application of virtual screening to the discovery of novel nicotinamide phosphoribosyltransferase (NAMPT) inhibitors with potential for the treatment of cancer and axonopathies. Bioorganic & Medicinal Chemistry Letters 2016;26:2920-6. [23] Bai J, Liao C, Liu Y, Qin X, Chen J, Qiu Y, et al. Structure-Based Design of Potent Nicotinamide Phosphoribosyltransferase Inhibitors with Promising in Vitro and in Vivo Antitumor Activities. Journal of Medicinal Chemistry 2016;59:5766-79. [24] Gunzner-Toste J, Zhao G, Bauer P, Baumeister T, Buckmelter AJ, Caligiuri M, et al. Discovery of potent and efficacious urea-containing nicotinamide

phosphoribosyltransferase (NAMPT) inhibitors with reduced CYP2C9 inhibition properties. Bioorganic & Medicinal Chemistry Letters 2013;23:3531-8. [25] Chen W, Dong G, Wu Y, Zhang W, Miao C, Sheng C. Dual NAMPT/HDAC Inhibitors as a New Strategy for Multitargeting Antitumor Drug Discovery. ACS Medicinal Chemistry Letters 2018;9:34-8. [26] Dong G, Chen W, Wang X, Yang X, Xu T, Wang P, et al. Small Molecule Inhibitors Simultaneously Targeting Cancer Metabolism and Epigenetics: Discovery of Novel Nicotinamide Phosphoribosyltransferase (NAMPT) and Histone Deacetylase (HDAC) Dual Inhibitors. Journal of Medicinal Chemistry 2017;60:7965-83. [27] Dixon SL, Smondyrev AM, Rao SN. PHASE: A Novel Approach to Pharmacophore Modeling and 3D Database Searching. Chemical Biology & Drug Design 2006;67:370-2. [28] Salam NK, Nuti R, Sherman W. Novel Method for Generating Structure-Based Pharmacophores Using Energetic Analysis. J Chem Inf Model 2009;49:2356-68. [29] Cereto-Massagué A, Guasch L, Valls C, Mulero M, Pujadas G, Garcia-Vallvé S. DecoyFinder: an easy-to-use python GUI application for building target-specific decoy sets. Bioinformatics 2012;28:1661-2. [30] Truchon J-F, Bayly CI. Evaluating Virtual Screening Methods: Good and Bad Metrics for the “Early Recognition” Problem. Journal of Chemical Information and Modeling 2007;47:488-508. [31] Rizzi A, Fioni A. Virtual Screening Using PLS Discriminant Analysis and ROC Curve Approach: An Application Study on PDE4 Inhibitors. J Chem Inf Model 2008;48:1686-92. [32] Friesner RA, Murphy RB, Repasky MP, Frye LL, Greenwood JR, Halgren TA, et al. Extra Precision Glide: Docking and Scoring Incorporating a Model of Hydrophobic Enclosure for Protein−Ligand Complexes. J Med Chem 2006;49:6177-96. [33] Ramsey KM, Yoshino J, Brace CS, Abrassart D, Kobayashi Y, Marcheva B, et al. Circadian Clock Feedback Cycle Through NAMPT-Mediated NAD+ Biosynthesis. Science 2009;324:651-4. [34] Skehan P, Storeng R, Scudiero D, Monks A, McMahon J, Vistica D, et al. New colorimetric cytotoxicity assay for anticancer-drug screening. J Natl Cancer Inst 1990;82:1107-12. [35] Yang S-Y. Pharmacophore modeling and applications in drug discovery: challenges and recent advances. Drug Discovery Today 2010;15:444-50. [36] Banoglu E, Çalışkan B, Luderer S, Eren G, Özkan Y, Altenhofen W, et al. Identification of novel benzimidazole derivatives as inhibitors of leukotriene biosynthesis by virtual screening targeting 5-lipoxygenase-activating protein (FLAP). Bioorganic & Medicinal Chemistry 2012;20:3728-41. [37] Eren G, Bruno A, Guntekin-Ergun S, Cetin-Atalay R, Ozgencil F, Ozkan Y, et al. Pharmacophore modeling and virtual screening studies to identify novel selective SIRT2 inhibitors. J Mol Graph Model 2019;89:60-73. [38] Florkowski CM. Sensitivity, specificity, receiver-operating characteristic (ROC) curves and likelihood ratios: communicating the performance of diagnostic tests. Clin Biochem Rev 2008;29 Suppl 1:S83-87. [39] Irwin JJ, Shoichet BK. ZINC--a free database of commercially available compounds for virtual screening. J Chem Inf Model 2005;45:177-82. [40] Wang X, Xu T-Y, Liu X-Z, Zhang S-L, Wang P, Li Z-Y, et al. Discovery of Novel Inhibitors and Fluorescent Probe Targeting NAMPT. Scientific Reports 2015;5. [41] Dahl TB, Haukeland JW, Yndestad A, Ranheim T, Gladhaug IP, Damås JK, et al. Intracellular Nicotinamide Phosphoribosyltransferase Protects against Hepatocyte

Apoptosis and Is Down-Regulated in Nonalcoholic Fatty Liver Disease. The Journal of Clinical Endocrinology & Metabolism 2010;95:3039-47.

Graphical abstract

Identification of small-molecule urea derivatives as novel NAMPT inhibitors via pharmacophore-based virtual screening Fikriye Ozgencil1,a, Gokcen Eren1,a,*, Yesim Ozkan2, Sezen Guntekin-Ergun3, Rengul Cetin-Atalay4 1Department

of Pharmaceutical Chemistry, Faculty of Pharmacy, Gazi University, 06330 Ankara, Turkey of Biochemistry, Faculty of Pharmacy, Gazi University, 06330 Ankara, Turkey 3Department of Medical Biology, Faculty of Medicine, Hacettepe University, 06100 Ankara, Turkey 4Cancer System Biology Laboratory (CanSyL), Graduate School of Informatics, Middle East Technical University, 06800 Ankara, Turkey aThese authors contributed equally. 2Department

*Correspondence Gokcen Eren, Department of Pharmaceutical Chemistry, Faculty of Pharmacy, Gazi University, 06330, Ankara, Turkey +903122023235 [email protected] Conflict of interest The authors have no conflicts of interest to declare.