Discovering novel carbonic anhydrase type IX (CA IX) inhibitors from seven million compounds using virtual screening and in vitro analysis

Abstract

Carbonic anhydrase type IX (CA IX) enzyme is mostly over expressed in different cancer cell lines and tumor tissues. Potent CA IX inhibitors can be effective for adjusting the pH imbalance in tumor cells. In the present work, we represented the successful application of high throughput virtual screening (HTVS) of large dataset from ZINC database included of ∼7 million compounds to discover novel inhibitors of CA IX. HTVS and molecular docking were performed using consequence Glide/standard precision (SP), extra precision (XP) and induced fit docking (IFD) molecular docking protocols. For each compound, docking code calculates a set of low-energy poses and then exhaustively scans the binding pocket of the target with small compounds. Novel CA IX inhibitor candidates were suggested based on molecular modeling studies and a few of them were tested using in vitro analysis. These compounds were determined as good inhibitors against human CA IX target with K_i in the range of 0.85–1.58 μM. In order to predict the pharmaceutical properties of the selected compounds, ADME (absorption, distribution, metabolism and excretion) analysis was also carried out.

• ADME
• carbonic anhydrase
• CA IX inhibitors
• molecular docking
• virtual screening

Introduction

Carbonic anhydrase (CA) family includes different isomers that display different level of enzyme activity and kinetic characteristics1–11. Only two of the CA isomers are distinguished as targets that are related to cancer: CA IX and CA XII12. While CA IX is commonly represented in a narrow number of ordinary tissues, its over expression is identified on solid tumors, where it is always connected with the hypoxic phenotype, mediated by the transcription factor HIF-113. Over expression of CA IX is mostly affiliated with a weak responsiveness to classical chemotherapy14. CA IX has been the subject of several studies as an enzyme responsible for the pH adjustment of tumor cells, as well as for the cell reproduction and cell adhesion15^,16. It is an important target for cancer diagnostics and therapy17. CA IX has an N-terminal domain, an extracellular CA catalytic domain, a trans-membrane (TM) helical domain and a short intra-cytoplasmic tail. The first domain plays an important role in the protein-mediated cell adhesion events. The catalytic domain catalyzes the reversible hydration of CO₂, thus it contributes to pH regulation18. In this work, a large ligand database (∼7 million) from ZINC database is used to detect novel CA IX inhibitors using high throughput virtual screening (HTVS) approach. A few proposed compounds based on virtual screening analysis are then tested by the in vitro and experimental data confirmed simulation results. Absorption, distribution, metabolism and excretion (ADME) profiles of studied compounds are also predicted using molecular modeling techniques.

Methods

Structural preparation of enzyme

The crystal structure of carbonic anhydrase19 IX (CA IX) (PDB ID: 3IAI) has been taken from the Protein Data Bank20. This structure is a dimer, which involves two identical inhibitors (Figure 1) inside the binding sites. In this work only one chain is used for docking simulations. Hydrogen atoms were added using Protein Preparation Wizard21 integrated in the Maestro molecular modeling Suite. The disulfide bond between Cys23 and Cys203 was constrained utilizing the patch DISU command during preparation22. The PROPKA23–25 program is used to locate protonation states at the physiological pH of 7.4. Hydrogen atoms were then relaxed using the energy minimization.

Figure 1. Chemical structure of acetazolamide (AZM) exists in each monomer.

MD simulations protocol

In order to relax the side chains of the CA IX complex, 10 ns MD simulations was performed before the docking simulations. Ligand–protein complex was hydrated with explicit water molecules using TIP3P water models, and neutralized by 0.15 M KCI using VMD (Visual Molecular Dynamics)26 program. MD simulations were then carried out using the parallel MD software27 NAMD 2.9 with CHARMM36 force-field parameters28. SHAKE algorithm was used to restrain the hydrogen atoms. Langevin piston method with Langevin thermostat was used to maintain constant temperature at 310 K and at a constant pressure of 1 atm during simulations. Periodic Boundary Conditions (PBC) and Particle Mesh Ewalds (PME) methods were utilized for treating the long-range electrostatic interactions. The non-bonded cut-off radius was set as 10 Å for the nonbonding (Lennard-Jones (LJ 6-12) and Coulomb) interactions. Water molecules around the CA IX were equilibrated with 500 ps MD simulations at fixed coordinates of the atoms of the CA IX. The production MD run was then carried out for 10 ns for all the studied system (Figure 2) without any constraints.

Figure 2. A snapshot picture taken from the simulation box: Protein (CA IX-dimer in yellow, inhibitor (inside the protein), water (as quick surface) and K⁺, Cl⁻ ions (cyan and brown colored spheres).

Co-crystallized ligand docking

In an initial step, we tested the capability of Glide/induced fit docking (IFD)29, Glide/SP and Glide/XP30 protocols to predict the ligand position of reference ligand (Figure 1) in the CA IX X-ray structure. Co-crystallized inhibitor is used at the ligand preparation module LigPrep31 and then it is docked into the active site of the protein using the Glide docking protocols with different approaches. The binding cavity was assigned as an outer box of (30 × 30 × 30) Å3 which is determined by the position of the reference ligand. We employed the SP, XP and IFD methods of the Glide docking. The conformation of co-crystallized X-ray structure is compared with docking poses. RMSD values for all used docking methods were found as less than 1.5 Å (Figure 3).

Figure 3. Cognate docking poses alignment with the co-crystallized ligand conformation using different methods.

Virtual docking screening

It has become possible to perform HTVS on large libraries (i.e. ZINC database) with the enormous increase in computer facilities. Chemical structures of around 7 million ligands are obtained from the ZINC database in the present study32. The 3D structures were prepared using LigPrep module of Maestro. The physiological medium and ionization states were determined for all the ligands and the pH was set to 7.4. The docking screening process was carried out in two steps: (i) Glide/HTVS is performed for accelerated docking simulations of 7 million ligands, (ii) the ligands with top- docking scores from HTVS were selected for further detailed analyses by Glide/XP docking.

ADME analysis

The absorption, distribution, metabolism and excretion (ADME) characteristics of all selected compounds were predicted using QikProp33 software, which calculates physicochemical properties of the compounds. This software may validate the drug-likeliness of the ligands based on Lipinski's criteria34.

In vitro analysis

Adherent epithelium renal cell adenocarcinoma cell line (ACHN, ATCC CRL-1611) was grown at Dulbecco's modified Eagle's medium (Sigma-Aldrich, Germany) containing 10% (v/v) FBS, 100 units/mL streptomycin and 100 µg/mL penicillin. Cells were maintained 80% confluency at 5% CO₂ at 37 °C into T75 flasks and then transferred into four T125 flask and maintained 90% confluency. Cells were quickly washed three times with 50 mM K_xPO₄ buffer. Cells were scratched from T125 flasks and transferred into falcon tubes (4 °C) and centrifuged 1500 rpm, 4 °C, 5 min. The cell pellet was maintained at −80 °C for 4 h and then lyophilized. About 1 mL dry ACHN cell pellet was obtained. CA IX activity was assayed by following the change in absorbance at 348 nm of 4-nitrophenylacetate (NPA) to 4-nitrophenylate ion over a period of 3 min at 25 ºC using a spectrophotometer (Shimadzu UV-VIS) according to the method described by Verpoorte et al.35 The enzymatic reaction, in a total volume of 3.0 mL, contained 1.4 mL 0.05 M tris-SO4 buffer (pH 7.4), 1.0 mL 3 mM NPA, 0.5 mL H2O and 0.1 mL enzyme solution. A reference measurement was obtained by preparing the same cuvette without enzyme solution. The Lineweaver–Burk curves were used to determine kinetic parameters and inhibition constants36.

Results and discussions

MD simulations

It is necessary to prepare the used X-ray structure and to relax the target before starting the drug screening in order to get successful results. For this aim, we performed 10 ns MD simulations to relax the CA IX complex. Figure S1 shows the RMSD of the Cα atoms of the target during the MD stimulations. It is clear that system reached to the acceptable stability after the 3.0 ns simulations. Average structure from generated trajectories is constructed and representative structure selected from the derived trajectories, which is the smallest RMSD with the average structure, for docking simulations.

Structure-based virtual screening

The ZINC ligand library consists of commercially available ligand structures for docking screening. It involves almost 20 million ligands that can be used for virtual screening. In order to determine novel CA-IX inhibitors from the ZINC database, Glide/HTVS docking simulations for 7 million drug-like ligands were performed. Out of these ligands, 70 compounds were selected with the high-binding scores at the target site (see supplementary Figure S2). Glide/XP protocol is then utilized to identify the final CA-IX inhibitor candidates. Their chemical structures and binding energies are shown in Table 1. There are two criteria for choosing the potent ligands. The screened ligands were ranked according to their docking scores and ligands forming critical hydrogen bonding interactions with key-residues such as Thr199 and Thr200 at active sites are also considered. All the reported selected compounds have docking scores of less than −8 kcal/mole (Table 1). In order to better understand the molecular mechanisms of inhibition profiles of the selected CA IX inhibitors, their docking poses at the binding site are analyzed (Figure S3). Figure S3 shows the ligand interaction diagram for each selected compound at the binding pocket of CA IX. All selected ligands were superimposed in the active pocket in order to show the population of the ligand poses (Figure 4). As it is shown, functional groups of similar chemical properties were positioned in similar modes with analogous bindings with the CA IX residues and encourage for an acceptable protein–ligand interactions, that was indicated the same inhibitory treatment of these ligands (Figure 4). The docked poses of selected ligands display reasonable hydrogen-bonding network including the hydroxyl group in most of the selected compounds and polar residue of the active cavity.

Figure 4. Superimposition of selected ligands from docking simulations in the active site.

Table 1. 2D chemical structures and their corresponding docking scores for selected top-docking scored compounds.

Compound	Chemical structures	Glide/XP scores (kcal/mole)	Compound	Chemical structures	Glide/XP scores (kcal/mole)
ZINC20464003		−9.80	ZINC67956012		−9.42
ZINC72421916		−9.66	ZINC11935360		−9.41
ZINC67805921		−9.47	ZINC67433231		−9.30
ZINC01592824		−9.45	ZINC67433435		−9.10
ZINC67714195		−9.43	ZINC67458868		−8.69
ZINC12291144		−8.67	ZINC58328414		−8.30
ZINC58327836		−8.61	ZINC32533143		−8.22
ZINC67850016		−8.38	ZINC72438236		−8.22
ZINC40528234		−8.38	ZINC19376149		−8.11

The top docking pose of compound from the ZINC database (ZINC20464003) was shown in Figure 5. This ligand fits into the CA IX active cavity. The ligand–protein complex was stabilized by establishing four polar interactions via residues Asn62, Gln67, Thr199 and Thr200 at the target. The other main residues that participated in hydrophobic interactions were represented as 2D ligand interaction diagram using docking poses of ZINC20464003 and ZINC72421916 compounds in Figure 5. Thr200, His94 and Thr199 were suggested as key amino acid residues in CA IX inhibitory mechanism19. All these amino acids formed strong interactions with docked compounds in our study.

Figure 5. 3D and 2D ligand interaction diagrams for the ligands, ZINC20464003 (upper panel) and ZINC72421916 (lower panel).

Physicochemical predictions of docked compounds

ADME analysis was also performed for the selected compounds. Physicochemical and pharmaceutical properties of selected ligands which involve molecular weight, octanol/water partition coefficient, apparent Caco-2 cell permeability in nm/s, blockage of hERG K channel, percentage of human oral absorption and Lipinski's rule of five were listed in Table 2. As shown, all of the predicted values are in the acceptable ranges. Lipinski's rule of 5 is a rule of thumb to predict drug like compounds and, selected compounds were positioned in good range of Lipinski's rule.

Table 2. Selected compounds with their ADME properties.

Compound	mol MW^a	QplogP o/w^b	QPPCaco^c	QploghERG^d	Percent human oral absorption^e	Rule of five
ZINC20464003	385.46	0.61	68.54	−6.02	63.39	+
ZINC72421916	352.44	1.60	128.29	−4.21	74.04	+
ZINC67805921	371.44	1.89	318.28	−4.73	82.83	+
ZINC01592824	215.18	−2.30	3.43	−2.04	23.04	+
ZINC67714195	341.36	0.58	44.33	−2.64	59.83	+
ZINC67956012	399.44	4.13	678.67	−4.83	100.00	+
ZINC11935360	364.39	2.82	302.50	−4.54	87.86	+
ZINC67433231	383.49	2.68	354.23	−4.51	88.25	+
ZINC67433435	377.49	3.58	263.38	−4.95	91.20	+
ZINC67458868	387.50	4.11	873.75	−4.35	100.00	+
ZINC12291144	316.44	2.37	398.16	−5.12	87.34	+
ZINC58327836	302.38	2.08	290.27	−5.62	83.22	+
ZINC67850016	341.40	2.15	227.63	−4.79	81.70	+
ZINC40528234	282.77	0.40	74.90	−3.57	62.86	+
ZINC58328414	314.43	2.69	263.47	−5.18	86.00	+
ZINC32533143	315.86	3.03	671.66	−4.84	95.31	+
ZINC72438236	342.44	2.53	351.71	−5.20	87.35	+

^aMolecular weight (reasonable value < 500).

^bPredicted octanol/water partition coefficient (reasonable value from −2.0 to 6.5).

^cPredicted apparent Caco-2 cell permeability in nm/s (reasonable value > 25).

^dPredicted blockage of hERG K + channel (reasonable value < −7).

^ePercentage of human oral absorption (<25% is weak and >80% is strong).

In addition, a similarity search toward standard drugs was carried out. All the discovered ligands displayed more than 80% similarity towards known drugs are listed in Table S1. As shown in Table S1, beta-adrenergic antagonists (Propranolol, Timolol, Nebivolol, Tretoquinol, Carazolol, Bucumolol, Tertatolol, Penbutolol and Xamoterol) have more similarity with proposed selected compounds in this study. It is known that beta-adrenergic inhibitors block the carbonic anhydrase receptors37. These experimental reports can validate the results of present virtual screening study. In order to compare docking scores of known CA-IX inhibitors with selected compounds from our study, we docked 14 known CA-IX inhibitors38–44 at the active site of the target. Docking scores and chemical structures of these compounds are found in the range of −5.66 to −6.79 kcal/mole (Table S2).

In vitro analysis

In order to test and validate the predicted results of molecular simulations three compounds are ordered from available compound databanks (ChemBridge Corporation, San Diego, CA) and tested by in vitro analysis. Results showed that all compounds are inhibited CA IX in the range of 0.85–1.58 μM K_i values. Interestingly docking scores of these compounds are in correct order with their inhibition data (Table 3). 3D docking poses and 2D ligand interaction diagrams of these selected compounds are represented in Figures 6 and 7.

Figure 6. 3D docking poses of selected compounds at the active site of CA-IX.

Figure 7. 2D ligand interaction diagrams of selected compounds into the active site of CA-IX.

Table 3. hCA IX inhibition data of selected compounds.

		Determined Ki (μM)*	Predicted Ki (μM)**
Compound	Chemical structures	hCA IX	hCA IX
ZINC67714195		0.85	0.227
ZINC11935360		1.03	0.231
ZINC67458868		1.58	0.750

*Mean from at least three determinations. Errors in the range of 1–3% of the reported value (data not shown).

**Converted from top-docking scores.

Conclusions

We performed a low-cost and accelerated structure-based virtual screening method45–48 to discover novel inhibitor candidates of hCA IX. In summary, ∼7 million compounds from the ZINC database were virtually screened against CA IX to discover novel ligands with high affinity toward this target. 70 ligands showed a high-docking score using Glide/HTVS method. These compounds are then docked with Glide/XP method and 19 of them showed high-docking scores at the binding cavities of the CA IX target. All selected compounds displayed drug-like properties and their properties are similar to standard drugs. A few of these ligands are ordered from available provider of the ligands and tested by in vitro analysis in order to validate the molecular simulations results. In vitro analysis showed that ZINC67714195, ZINC11935360, and ZINC67458868 compounds inhibited hCA IX by K_i values of 0.85, 1.03, and 1.58 μM, respectively. These discovered novel compounds may be guided as candidate inhibitors of CA IX.

Supplementary material available online

Supplementary Figures S1–S3 and Tables S1–S2

Acknowledgments

Part of computations for the work described in this paper was supported by Turkish Scientific and Technical Research Council (TUBITAK) ULAKBIM High Performance Computing Center. SD acknowledges support from Bilim Akademisi - The Science Academy, Turkey, under the BAGEP program

Declaration of interest

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of this article.