Ligand-based design of GLUT inhibitors as potential antitumor agents

Abstract

Glucose transporters (GLUTs) regulate glucose uptake and are often overexpressed in several human tumors. To identify new chemotypes targeting GLUT1, we built a pharmacophore model and searched against a NCI compound database. Sixteen hit molecules with good docking scores were screened for GLUT1 inhibition and antiproliferative activities. From these, we identified that compounds 2, 5, 6 and 13 inhibited the cell viability in a dose-dependent manner and that the IC₅₀s of 2 and 6 are <10 μM concentration in the HCT116 colon cancer cell line. Lead compound 13 (NSC295720) was a GLUT1 inhibitor. Docking studies show that GLUT1 residues Phe291, Phe379, Glu380, Trp388, and Trp412 were important for inhibitor binding.

1. Introduction

Glucose is an important energy source for several biological functions, including cell proliferation, and motility.^1–2 Glucose transporters (GLUTs) regulate cellular glucose uptake and help maintain proper glucose concentration in various tissues.³ The GLUTs are membrane proteins encoded by the solute carrier transporter family, SLC2, and SLC5 genes.⁴ Cancer cells transport more glucose than normal cells due to their rapid growth and high rate of aerobic glycolysis (Warburg effect).^5–7 Glucose transport 1 (GLUT1) is upregulated in many types of cancers such as breast, lung, prostate, and colon.^8–11 In addition, stimulation of oncogenes such as KRAS, BRAF, c-Myc, and p53, and transcription factors such as hypoxia-inducible factor-1a (HIF-1) upregulate the GLUT1 expression.^9,12–15

A specific antibody targeting GLUT1 reduced glucose uptake, induced apoptosis in lung and breast cancer cells, and improved the activity of anticancer drugs such as cisplatin, paclitaxel and gefitinib.¹⁶ Glucose transport in the brain is facilitated by both GLUT1 and GLUT3.^17–18 In the brain, GLUT3 has greater affinity and higher capacity than GLUT1.^17–20 Therefore, GLUT-1-selective ligands would minimize potential neurotoxicity. Several small molecular GLUT inhibitors including resveratrol, naringenin, phloretin, cytochalasin B, WZB117, STF-31, pyrazolopyrimidines, phenylalanine amides, and (1H-pyrazol-4-yl)quinolones inhibit cell proliferation and induce apoptosis of cancer cells.^21–28

GLUT1 transports substrates through alternating mechanisms, which involves substantial conformational changes across the cell membrane.^29–30 GLUTs conformation changes from an outward-facing conformation (open to the extracellular) to an inward-facing conformation (open to the intracellular) to deliver glucose through the cell membrane passing through an intermediate state in which the GLUTs conformation is occluded.³⁰ The substrate-free GLUT protein favors the outward-open conformation.³⁰ Once the substrate binds to the C domain of the GLUT1, the GLUT1 shifts to the inward-open conformation to release glucose.³⁰ Our docking study on different conformations of GLUT1 also confirmed that the inward-open conformation is the most favorable ligand binding site.³¹

In this paper, we explored a lead generation technique for identification of GLUT1 inhibitors by ligand-based pharmacophore modeling. We developed a 3D pharmacophore model that was formed from known GLUT1 inhibitors reported in the literature (Figure 1).^24,27–28 To identify potential hit molecules, we searched the generated pharmacophore model against a National Cancer Institute (NCI) database. Our results led to the identification of four compounds (hit compounds 2, 5, 6, and 13) that inhibited glucose uptake and decreased growth of colon cancer cells in vitro.

Fig. 1.

GLUTs Selective Inhibitors Used to build a Pharmacophore Model. The IC₅₀s are provided in parentheses with units of μM.

2. Results and discussion

We created a 3D pharmacophore model in MOE³² based on structures of known GLUT1 inhibitors (Figure 1). The selection of these compounds to build a pharmacophore for potential GLUT1 inhibitors was based on their high potency and structural diversity. The MOE software Pharmacophore modeling module³² allowed the superposition of compounds b-d to compound a and thus generated a 3D pharmacophore model composed of one aromatic group (F1), one hydrophobic group (F2), an aromatic ring/hydrophobic (F3), and one hydrogen bond donor (F4, Don) (Figure 2). Using this 3D pharmacophore model, we performed a virtual screening of the NCI compound database. 1,469 compounds were found to satisfy the 3D pharmacophore and were reported as hits to identify novel GLUT inhibitors. Figure 2 showed the pharmacophore that was built based on compounds a-d (Fig. 2). Figure 2 shows that hit molecules 6 and 13 were able to fit into three out of four pharmacophoric points.

Fig. 2.

Pharmacophore model for GLUTs inhibitors. Color code: Cytochalasin B (a), magenta; 5RF (b), red; Compound c, green; Compound d, cyan; hit molecules NSC328095 (6), orange; and NSC295720 (13), blue. Aro: aromatic rings; Don: H-bond donor; Hyd: hydrophobic groups.

We then carried out docking studies of all 1,469 molecules that satisfied the pharmacophore model. Those with good docking scores and availability from the NCI were considered as potential hit molecules. Our previous study showed that the inward-open conformation of GLUT1 proteins is the most favorable conformation for ligand binding.³¹ The inward-open conformation crystal structures for GLUT1 are PDB ID: 4PYP, 5EQG, 5EQH, and 5EQI.^24,30 We used 5EQH as a model protein because the bound ligand was one of the potent molecules used to build the pharmacophore model. From this docking of 1,469 NCI compounds, 16 top-ranking compounds were chosen as lead molecules with GLUT1 inhibition potential based on their Glide docking scores and the availability from NCI. The Glide docking scores against the inward-open GLUT1 are listed in Table 1. Based on our previously study, this inward-open conformation is more likely for ligand binding because it identified more active molecules as tight binders than those outward-open or inward-occluded conformations.³¹ All 16 hit molecules also had better Glide docking scores than glucose (either α- or β-conformation), the natural GLUT1 substrate. The structures of the 16 compounds are given in Figure 3.

Table 1.

Glide docking scores (kcal/mol) of hit molecules with NSC numbers, logP, and interacting residues of 5EQH.

NSC# (Comp#)	logP	XP GScore	Interacting residues
657377 (1)	4.45	−7.54	Phe291, Phe379, Trp388, Trp412
657996 (2)	4.03	−8.78	Ser80, Phe291, Phe379, Trp388, Trp413
649554 (3)	4.53	−7.85	Glu380, Trp388, Trp412
281301 (4)	0.75	−8.47	Phe291, Phe379, Trp388, Trp412
32458 (5)	3.82	−9.48	Ser80, Thr137, Trp388, Asn411, Trp412
328095 (6)	4.64	−8.14	Thr137, Gln282, Trp388, Trp412
730702 (7)	3.37	−7.00	Phe379, Glu380, Trp388, Trp412
657374 (8)	4.7	−7.85	Asn288, Glu380,Trp388, Trp412
295714 (9)	3.04	−7.63	Phe291, Phe379, Trp388, Trp412
295715 (10)	1.54	−6.66	Phe379, Trp388, Trp412
641409 (11)	1.18	−7.46	Trp388, Trp412
659680 (12)	2.31	−8.12	Phe379, Trp388, Trp412
295720 (13)	3.71	−6.30	Phe291, Phe379, Trp388, Trp412
314887 (14)	1.08	−7.08	Phe379, Trp388, Trp412
641422 (15)	3.85	−7.14	Glu380, Trp388
380503 (16)	4.27	−8.59	Thr137, Trp388,Asn411, Trp412
β-d-glucose		−5.96
α-d-glucose		−5.91

Fig. 3.

Chemical structures of sixteen hit molecules.

The validation of the glide docking method was performed by the pose selection method.³³ This is a standard method used by comparing the docked pose of a ligand with the cocrystal structure of the ligand. A pose with an RMSD < 2.0 Å is considered to be good.³³ The superposition of the Glide-generated docked pose, and the native conformation in the co-crystal structure (PDB ID: 5EQH) for compound 5RF (the ligand bound to 5EQH, Figure 4) showed that the RMSD between these two poses is 1.34 Å. This indicates that the glide docking can successfully predict ligand-binding conformations. We also investigated ligand–protein interactions of the 16 top-ranking compounds to analyze their binding modes. Phe379, Trp388, and Trp412 play essential roles in GLUT1 function and glucose uptake.^24,34–36 Gln282, Gln283, Asn288, Phe291, Asn317, and Glu380 are also important residues for glucose binding.^31,30 Notably, our analysis showed that the 16 top-ranking compounds were able to bind to these reported residues. Our Glide docking results of the 16 top-ranking compounds against the inward-open GLUT1 conformation revealed that all compounds bound to the glucose binding site of GLUT1, and the most frequently observed interacting residues were Thr137, Gln282, Phe291, Phe379, Glu380, Trp388, and Trp412 for their role in providing H-bonds or π – π stacking interactions (Table 1). Hit compound 2, 6, 13, and 5 were able to create π - π stacking interactions with Phe291, and Phe379, Trp388, and Trp412 and H-bonds with Thr137 of GLUT1 (Figure 5 shows interactions between GLUT1 and compound 2). This suggests that Thr137 may be a significant residue for ligand binding. Taken together, the docking scores and ligand–protein interactions of the 16 top-ranking compounds suggest that these compounds may be potential GLUT1 inhibitors. However, we note that docking scores do not always correspond to activity order of hit molecules. For example, compounds with the best docking scores were 2, 4, 5, 6, 12, and 16. Among them, 4 and 12 were tested as inactive. On the other hand, hit molecule 13 which had a low docking score turned out to be quite active. One should not over-interpret the docking scores when it comes to ranking, because compounds with ± 1 kcal/mol are considered comparable. The purpose of docking in this paper is to narrow the number of hit molecules for biological evaluation, and the docking-based protein–ligand interactions will help identify important binding residues that show high frequency for ligand binding.

Fig. 4.

The superposition of the Glide generated pose and the native ligand of the crystal structure (5EQH). The RMSD between these two conformations is 1.34 Å.

Fig. 5.

Ligand interactions between NSC657996 (2) and GLUT1: The H-bond interactions are shown as green dotted lines.

In order to evaluate the antiproliferative activities of the 16 potential lead compounds, we assessed the growth inhibition of these candidates in the highly malignant HCT116 colon cancer cell line.³⁷ Compounds 2 and 6 were the most potent with IC₅₀ values < 10 μM followed by 13 with an IC₅₀ value of 56 μM. Compounds 1,3, 5, 10, and 16 had IC₅₀ values ranging from 250 and 660 μM whereas the remaining compounds were inactive (Table 2). Compounds 2, 5, 6 and 13 inhibited cell viability of the HCT116 colon cancer cell lines in a dose-dependent manner.

Table 2.

IC₅₀s of 16 tested compounds against the HCT116 colon cancer cell lines.

NSC#	IC50 (μM)	NSC#	IC50 (μM)
657377 (1)	450	295714 (9)	inactive
657996 (2)	4.9	295715 (10)	660
649554 (3)	560	641409 (11)	inactive
281301 (4)	inactive	659680 (12)	inactive
32458 (5)	250	295720 (13)	56
328095 (6)	9.1	314887 (14)	inactive
730702 (7)	inactive	641422 (15)	inactive
657374 (8)	inactive	380503 (16)	260

We next determined whether the cell viability inhibition by lead compounds 2, 5, 6, and 13 was mediated through inhibition of GLUT1. We tested GLUT1 specificity in the presence of mitochondrial electron transport inhibitor rotenone using previously published methods.^38–41 Rotenone inhibits ATP production through mitochondrial respiration; therefore, in co-incubations of the compounds with rotenone, cells would produce ATP only through glycolysis which is linked to glucose uptake. These data showed that 13 significantly decreased ATP levels in the presence of rotenone (Figure 6), suggesting that this compound specifically inhibits glucose uptake, leading to reduced glycolysis. Taken together, these results indicate that 13 is a GLUT inhibitor.

Fig. 6.

FET cells were treated with hit molecules 2, 5, 6 and 13 in the presence of rotenone (1 μM) for 1 h and ATP level was measured by Cell Titer-Glo® Luminescent Cell Viability Assay Kit. The experiment was repeated twice. * P < 0.1.

3. Conclusion

We used virtual screening and docking-assisted approaches in the design of GLUT1 inhibitors. Pharmacophore modeling and database searching identified 16 hit compounds. The ability to generate π - π stacking with Phe291, and Phe379, Trp388, and Trp412 and form H-bonds with Thr137 may be responsible for the binding of these hit molecules to GLUT1 protein. Two of these inhibited the HCT116 colon cancer cell line at low micromolar concentrations. One compound (13) was a GLUT inhibitor.

4. Materials and methods

4.1. Three-Dimensional (3D) pharmacophore model design and virtual screening

The computational pharmacophore generation was carried out using Molecular Operating Environment (MOE).³² The 3D structures of the known GLUTs inhibitors from literature (Figure 1) were built and energy minimized using MMFF94X force field⁴² in MOE software based on the 5RH ligand in the crystal structure 5EQH.²⁴ The 3D pharmacophore was generated by the flexible superimposition of the 3D model of GLUTs inhibitors, and identification of the 3D features that they shared. Then, the 3D pharmacophore model was applied in the design of novel GLUTs inhibitors according to the Pharmacophore Query module in MOE. A database of 260,071 molecules was downloaded from the National Cancer Institute (NCI),⁴³ then it was filtered according to the Lipinski’s rule of five by the logP (logP < 5), and molecular weight (MW < 500)⁴⁴, yielding 33,778 drug-like molecules.

A pharmacophore search against this database of 33,778 molecules resulted in 1,469 hit molecules that fit the pharmacophore model. To help narrow down the number of hit molecules that we will use for the biological test, we docked all these 1,469 molecules to the GLUT1 5EQH ligand binding site using the procedure established in our previous studies.^45,31 In this study, Glide Dock⁴⁶ in the Maestro 11.2 were performed for ligands docking against inward-open conformations of GLUT1 (PDB ID: 5EQH) which has been prepared using the Protein Preparation Wizard module followed by energy optimization using the MacroModel in the Schrödinger software suite with the OPLS force field. Then, a grid file was generated by the Glide Grid Generation panel with the bound ligand (5RF) as the centroid of the minimized protein. Then, all compounds were docked to the grid file with ligand sampling being set to extra-precision (XP) method, and all other parameters were used as defaults. The binding affinity of the various conformation of GLUT1/ligand complexes was evaluated by the Glide scores. The protein/ligand interactions were created by using PyMOL software.⁴⁷

4.2. Cell viability assay

Human colon cancer cell lines HCT116 and FET were maintained in McCoy’s 5A medium (Sigma, St Louis, MO, USA) with 10 ng/ml epidermal growth factor (EGF), 20 μg/ml insulin and 4 μg/ml transferrin. Cells were culture at 37 °C in a humidified incubator with 6% CO₂. HCT116 cells were seeded into 96-well plates at a density of 6000 cells per well and treated with compounds for 72 h. Cells were stained for 2 h with alamar blue reagent (Bio-rad). The OD at 570 nm and 630 nm were read on an ELx808 Absorbance Microplate Reader (BioTek, Winooski, VT, USA). Cell viability was calculated as a ratio of OD values of drug-treated samples to those of controls.

4.3. GLUT1 specificity assay

To test the specificity of the GLUT inhibitors, colon carcinoma cell line FET were seeded in 96 plates at a density of 20,000 cells per well. The cells were then cultured overnight in glucose free media. After 16 h, the cells were incubated with 0.1 M glucose with or without compounds in the presence of 1 μM rotenone for 1 h. The Cell TiterGlo® Luminescent Cell Viability Assay from Promega was then used to measure ATP levels. Statistical analyses were performed using Student’s t-test.