5-Arylidene(chromenyl-methylene)-thiazolidinediones: Potential New Agents against Mutant Oncoproteins K-Ras, N-Ras and B-Raf in Colorectal Cancer and Melanoma

Background and objectives: Cancer represents the miscommunication between and within the body cells. The mutations of the oncogenes encoding the MAPK pathways play an important role in the development of tumoral diseases. The mutations of KRAS and BRAF oncogenes are involved in colorectal cancer and melanoma, while the NRAS mutations are associated with melanoma. Thiazolidine-2,4-dione is a versatile scaffold in medicinal chemistry and a useful tool in the development of new antitumoral compounds. The aim of our study was to predict the pharmacokinetic/pharmacodynamic properties, the drug-likeness and lead-likeness of two series of synthetic 5-arylidene(chromenyl-methylene)-thiazolidinediones, the molecular docking on the oncoproteins K-Ras, N-Ras and B-Raf, and to investigate the cytotoxicity of the compounds, in order to select the best structural profile for potential anticancer agents. Materials and Methods: In our paper we studied the cytotoxicity of two series of thiazolidine-2,4-dione derivatives, their ADME-Tox properties and the molecular docking on a mutant protein of K-Ras, two isoforms of N-Ras and an isoform of B-Raf with 16 mutations. Results: The heterocyclic compounds strongly interact with K-Ras and N-Ras right after their posttranslational processing and/or compete with GDP for the nucleotide-binding site of the two GTPases. They are less active against the GDP-bound states of the two targets. All derivatives have a similar binding pattern in the active site of B-Raf. Conclusions: The data obtained encourage the further investigation of the 5-arylidene(chromenyl-methylene)-thiazolidinediones as potential new agents against the oncoproteins K-Ras, N-Ras and B-Raf.


Introduction
Cancers are a group of diseases which can be perceived as miscommunications between the cells and within the cells. One of the most dangerous triggers that leads to cancers is the mutation of genes. Oncogenes are genes that encode proteins able to induce cancer via various metabolic pathways. Somatic mutations in genes encoding the mitogen-activated protein kinase (MAPK) pathways components occur frequently in various tumors, making them critical turning points in the development of human cancer [1]. Oncogenic mutations in MAPK signaling pathways frequently

Virtual Screening
VS carried out in this paper used two cheminformatics tools to accomplish its aim: an ADME-Tox predictor and a docking software.
The Lead-Like Soft filter uses descriptors for lead-likeness [58][59][60][61], meanwhile the Drug-Like Soft filter is built on physico-chemical, molecular properties and bioavailability data, commonly encountered in the development of new drugs [58,[62][63][64][65]. These soft filters use a build-in statistical analysis of drugs [36] from the e-Drugs3D library [65] for the threshold values of the computed parameters.
PPIs are required in a normal, healthy life, while the abnormal PPIs may lead to diseases. In consequence, they may be considered important targets in medicinal chemistry [66]. For this, FAF-Drugs3 uses a decision tree [39,67] built on two trained Dragon descriptors, Ui and RDF070m [68].
The detection of UMSs involved in toxicity was possible due to FAF-Drugs3 build-in filters, based on the literature data. Also, for identifying the covalent inhibitors (CIs), FAF-Drugs3 build-in filters were applied. Three filters (A, B and C) [36,53,69] were used in order to discover PAINS [45,46].
We investigated the safety profiling according to the GSK 4/400 rule [57], the Pfizer 3/75 rule [56], the estimation of phospholipidosis induction (PhI) [55], the MedChem rules [44] and the golden triangle (GT) rule [70]. The MedChem rules allow the identification of the molecules that may disturb the biological assays, allowing their removal from the screening. We used the MedChem rules in our screening process with the regular settings, involving a 100-demerit cutoff.

Molecular Docking
Targets: docking demands the 3D structure of a target, which contains its spatial coordinates, and for the target identification process it was necessary to cross-reference 3 on-line databases to identify the most adequate targets with a high-resolution 3D structure-a resolution higher than 2.0 Å being advised for the docking computations [71]: The Human Gene Database-GeneCards ® (http://www.genecards.org) [72], The Universal Protein Resource-UniProt (http://www.uniprot. org) [73] and RCSB Protein Data Bank-RCSB-PDB (http://www.rcsb.org) [74]. The targets selected were: one of the most common mutants of K-Ras [6], the two isoforms of N-Ras (the canonical form and the short isoform) and an oncogenic B-Raf isoform with 16 mutations (Table 1). Docking set-up: all ligands were docked against the selected targets (Table 1), in separate runs, with PyRx-Python Prescription 0.9.5 [77] AutoDock Vina was used as the docking algorithm [78]. AutoDock Vina automatically calculates the grid maps [78] based on a scoring function, inspired by an X-score [79] and tuned with PDBbind data set [80,81], to predict noncovalent binding and clusters the results. The renderings of resulting 3D images were performed with the help of The Visualization ToolKit (VTK -Kitware, Inc. USA) (http://www.kitware.com) [82]-an embedded module of PyRx 0.9.5. Supplementary, Molegro Molecular Viewer 2.5 (Molegro, A CLC bio company, Aarhus N, Denmark) was performed for more advanced extraction of data and high resolution renderings of the individual poses. Docking runs against the corresponding PDB IDs (4DSU and 3CON) were carried out twice: with and without the GDP molecule bound in its pocket. Separate single docking runs were carried out for the N-Ras isoform 5 (2N9C) and for B-Raf (5ITA). The docking runs were realized setting an extended search space to completely cover the target, with a volume higher than 27.0 Å3, meanwhile the exhaustiveness was manually expanded 10 times from the default value (8), to increase the precision of all predictions [78,83].

Cytotoxicity
The murine cancer cell lines B16 (mouse melanoma) and CT26 (colorectal carcinoma) (Thermo Fisher Scientific, Waltham, MA, USA) were grown in Dulbecco's modified essential medium (DMEM). This was supplemented with 10% fetal bovine serum, 2 mM L-glutamine, 100 U/mL penicillin and 100 µg/mL streptomycin. The growing cancer cells were plated onto 96-well plates, at 5000 cells/well, in 200 µL DMEM. After 24 h, the cells were exposed, for 48 h, to dimethylsulfoxide (DMSO) and to the compounds' solutions, respectively. Stock solutions of compounds (10 mm) were diluted, in order to obtain solutions of different concentrations (100 µm, 50 µm, 25 µm, 12.5 µm, 6.25 µm and 3.125 µm) [34]. The MTT (1-(4,5-dimethylthiazol-2-yl)-3,5-diphenyltetrazolium) test was used to assess the viability of the cells. The number of the living cells, after 72 h of culture, directly proportional to the intensity of the blue color, was spectrophotometrically measured, at 562 nm, by a microplate reader (BioKinetics Reader EL340, Fisher Bioblock Scientific, Illkirch, France) [84]. Control cells were used and they were exposed to 1% DMSO. The experiments were repeated three times. The results obtained were quantified as the inhibitory concentrations for 50% of cells (IC50), for a 48 h exposure time.

Virtual screening (VS)-ADME-Tox predictions
A potential drug candidate has to correspond to some important drug features, such as: oral absorption, body distribution, metabolism, excretion, low toxicity, beside its pharmacological activity. This is why the virtual screening is extremely important in drug development, allowing the prediction of these parameters (ADME-Tox predictions), with the help of specialized software, before the effective lab synthesis. In our case, a license of MarvinSketch was involved in drawing and generating the 2D structures, 3D optimization of all ligands, and also for creating the input SDF files for the ADMET profiling and Tripos MOL2 files for docking (MarvinSketch 17.6.0, 2017, ChemAxon, Budapest, Hungary) [35]. Table 2 summarizes the results of the ADME-Tox screening carried out with FAF-Drugs3, for the lead-likeness and drug-likeness criteria.  For a drug, a good oral bioavailability is a desired characteristic. The predictors used for this property are: a good intestinal absorption, a reduced molecular flexibility, low polar surface area and the hydrogen-bounding ability.
Thiazolidinediones 1-3 and 7-11 have a higher value of LogP, in terms of the drug-likeness filters (LogP > 5), and only compounds 16-25 respect the lead-like criteria (LogP < 4). Thiazolidinedione derivatives 1-3 and 26-28 also have bad predictions for the molecular weight (MW). All molecules have values of tPSA inferior to 160 Å 2 , passing the criteria requested for the gastro-intestinal absorption, after an oral administration. The studied substances have less than 9 rotatable bonds (RtB) and no chirality center (SC); in consequence, exhibiting low conformational flexibility. All compounds validate all other filters for drug-likeness and also, for lead-likeness ( Table 2).
The ADME-Tox profiling also provides a helpful guidance on acute and later toxicity (Table 3). Table 3. ADME-Tox profiling-risks and safety concerns. The risk and safety concern profiling for the studied molecules revealed that these are not phospholipidosis non-inducers. They are free of PAINS. All the lead-like compounds (16)(17)(18)(19)(20)(21)(22)(23)(24)(25) were detected as not being PPIs friendly and, with the exception of 16, the rest of all the investigated compounds were flagged as possible covalent inhibitors, due to the presence of the α, β-unsaturated carbonyl [40,86,87].

Molecular Docking
The results of the molecular docking runs on the mutant K-Ras, N-Ras or B-Raf isoforms are presented in Table 4 as binding affinity (BA) for the best poses, at root-mean-square deviation (RMSD)  Supplementary Table S1. The graphical depiction of the docking results is illustrated in Figures 1-5. The strongest interaction with the targets chosen was that of compound 12, bearing a nitro moiety: the binding affinity for the mutant isoform of K-Ras was −10 kcal/mol, respectively for N-Ras isoforms, −6.70 kcal/mol and −6.40 kcal/mol. Compound 26 proved to be a strong binder of B-Raf (−10.40 kcal/mol) and compound 27, of N-Ras (−10 kcal/mol). The substances which passed the lead-like filter in the ADME-Tox predictions (16)(17)(18)(19)(20)(21)(22)(23)(24)(25) have good binding affinities to the targets, but do not display the strongest interaction. From Table 4 it could be observed that the safer lead-like compound, 16 (the non-covalent binder- Figure 3) is a weak binder of K-Ras and B-Raf, and slightly more potent that 19 against the canonical isoform of N-Ras. All compounds are less active against the N-Ras isoform 5 (which doesn't have GTPase activity and is responsible for an aggressive phenotype of melanoma) and interact with the alpha-helix (Supplementary Table S1 and Figure 4).
The thiazolidinedione derivatives studied bound more weakly to the GDP-bound state of the GTPases (Figures 1 and 2).     backbone; the ligand is drawn as sticks; the H-bonds are in dashed blue lines-image rendered with Molegro Molecular Viewer 2.5). (a) General view and detail: ligand is presented buried in the nucleotide-binding pocket (molecular surface of target is depicted as electrostatic potential molecular surface); (b) Detail: hydrogen bonds formation with Ala18 and Tyr32; (c) Detail: energy grid depiction (green-steric favorable; light blue-hydrogen acceptor favorable; yellow-hydrogen donor favorable; orange to red and dark blue-electrostatic interactions).   Table S1).

Cytotoxicity
Genetic alterations in the MAPK pathway (including the mutation of KRAS, NRAS and BRAF genes) can cooperate in the development of B16 melanomas [88,89] and CT26 colorectal carcinoma [90,91]. Considering this aspect, we investigated the cytotoxicity of the compounds on B16 and CT26 murine cell lines. The results obtained are presented in Table 5.  The strongest interaction with the targets chosen was that of compound 12, bearing a nitro moiety: the binding affinity for the mutant isoform of K-Ras was −10 kcal/mol, respectively for N-Ras isoforms, −6.70 kcal/mol and −6.40 kcal/mol. Compound 26 proved to be a strong binder of B-Raf (−10.40 kcal/mol) and compound 27, of N-Ras (−10 kcal/mol). The substances which passed the lead-like filter in the ADME-Tox predictions (16)(17)(18)(19)(20)(21)(22)(23)(24)(25) have good binding affinities to the targets, but do not display the strongest interaction. From Table 4 it could be observed that the safer lead-like compound, 16 (the non-covalent binder- Figure 3) is a weak binder of K-Ras and B-Raf, and slightly more potent that 19 against the canonical isoform of N-Ras. All compounds are less active against the N-Ras isoform 5 (which doesn't have GTPase activity and is responsible for an aggressive phenotype of melanoma) and interact with the alpha-helix (Supplementary Table S1 and Figure 4).
The thiazolidinedione derivatives studied bound more weakly to the GDP-bound state of the GTPases (Figures 1 and 2).
The thiazolidinediones studied have a similar binding pattern in the active site of B-Raf ( Figure 5) from the protein kinase domain, between Ile463 and Gly596 (Supplementary Table S1).

Discussion
The thiazolidinedione derivatives belong to two structural profiles: 5-arylidene-N-(phenyl-thiazolyl-methylene)-2,4-thiazolidinediones 1-16 (Table 6) [33] and 5-chromenyl-methylene-2,4-thiazolidinediones 17-28 (Table 7) [34,85,92]. A good oral bioavailability represents a very important characteristic of a bioactive substance. In our case, all the investigated compounds complied with the drug-likeness filters. Even if molecules 1-3 and 7-11 had higher values of LogP, since there was only one violation of Lipinski's "Rule of 5" (RO5) [93], the estimation of the oral bioavalability was good. Regarding the lead-likeness predictions, only compounds 16-25 can be considered good lead-like molecules, meanwhile all the rest of the investigated structures fail in terms of the logarithm of the partition coefficient between n-octanol and water (LogP). All thiazolidinedione derivatives complied with Veber's rule [94] and with Egan's rule [95] on the molecular properties with impact on the oral bioavailability. Topological polar surface area (tPSA), a descriptor that correlates well with the transport through membranes, including the blood-brain barrier, had values inferior to 140 Å 2 , suggesting that the compounds pass the criteria required for a good gastro-intestinal absorption, after an oral administration. Supplementary, all derivatives were predicted to have a reduced blood-brain barrier transport (tPSA > 90 Å 2 . This aspect presents a high significance due to the reduced or totally absent possible side effects on the central nervous system. In drug development, the investigation of both acute and later toxicity is mandatory: a drug must be pharmacologically active, but also very well tolerated, without side effects. The software-based predictions of the safety and risks profiling are very useful tools in medicinal chemistry, providing many advantages, such as: the availability of the specialized software, the rapidity of the data obtention, the possibility of selecting multiple parameters and filters and the major ethical advantage of reducing the number of the lab animals sacrificed in the classical toxicity assays. The screening performed on our compounds showed that these are not inducers of phospholipidosis, which is a disorder manifested by the accumulation of phospholipids in tissues and a sign of molecules' toxicity. All substances seemed to be free of PAINS (structures or substructures predicted not to interfere with the biological assays) and PPIs friendly, therefore successfully complying with these safety criteria.
The ADME-Tox profiling identified some low risk problematic moieties in the structure of the compounds, like thiazolidinedione, halogenure, nitro or thioester. MedChem rules considered the thioester fragment as potentially reactive or promiscuous. Considering this observation, the lead-like compounds 16-20 should be placed on the short list of possible hits for further structural optimization in a drug development project. Applying the GSK 4/400 rule, the Pfizer 3/75 rule and the GT rule, we could select the derivatives 16, 17, 19, 20, 23 and 24 as the best candidates, with an optimal permeability (low clearance) and a good metabolic stability [70].
The ADME-Tox data revealed that the studied thiazolidinedione derivatives display good pharmacokinetic properties, but with some limitations. All compounds pass the drug-likeness criteria, while only compounds 16-25 pass the lead-likeness filter. Further structural optimization might be helpful in achieving better ADME-Tox properties.
Molecular docking represents a modern, useful tool in drug discovery. Based on computation methods, it aims to give a prediction of a ligand-receptor complex, therefore suggesting a mechanism of action for the compounds studied.
In our paper, the binding patterns and the binding affinities of the compounds suggest that the substances may strongly interact with K-Ras and N-Ras right after their posttranslational processing and/or compete with GDP for the nucleotide-binding site of the two GTPases. Moreover, all the investigated molecules are less active against the GDP-bound states of the two targets, being weak binders. The compounds interact with the G domain of K-Ras and N-Ras in the bordering region of the nucleotide-binding pocket, in the absence of GDP molecule, acting as a competitor of GDP for its binding pocket. In the presence of GDP in the binding pocket, the compounds not only are weaker binders, but the binding patterns also indicate less pharmacological relevance.
However, AutoDock Vina cannot evaluate the covalent binding capacity of which derivatives are capable (with the exception of 16), as resulted from the ADME-Tox predictions.
The binding pattern in the active site of B-Raf from the protein kinase domain, similar for all thiazolidinediones investigated, suggests that the tested compounds may also interact with the other members of TKL serine/threonine-protein kinase family, since the interaction region is highly conserved-especially for the RAF subfamily [7,8].
The evaluation of the cytotoxicity is a mandatory step in drug development. For this, we evaluated in vitro the viability of the B16 (mouse melanoma), respectively CT26 (colorectal carcinoma) murine cell lines growth, in an MTT assay. Analyzing the data obtained, we could say that compounds 2, 3, 7, 9, 11, 17, 21 and 22 manifested a good inhibitory effect (IC50 < 50 µm), against the B16 cells. From these, the derivative 3, with a 4-Br-benzylidene fragment in position 5 of the thiazolidinedione ring, displayed the lowest value of IC50. The proliferation of CT26 cells was effectively impeded by compounds 1, 2, 6, 13, 21. The strongest activity was manifested by the thiazolidinedione derivative 1, bearing the 2-Br-benzylidene fragment bound to the heterocycle. The good inhibitory effect expressed by the two molecules with bromine, confirm the impact of this halogen on the cytotoxic activity, in total agreement with the literature data [27].

Conclusions
The cytotoxicity assay realized on the two series of 5-arylidene-N-(phenyl-thiazolyl-methylene) -2,4-thiazolidinediones and 5-chromenyl-methylene-2,4-thiazolidinediones, respectively, revealed that the compounds expressed good inhibitory effects against the B16 and CT26 cell lines as well. The strongest activity was manifested by the derivatives with a Br-benzylidene fragment in the structure. All the investigated compounds complied with the drug-likeness filters, while substances 16-25 could be considered as lead-like molecules. The thiazolidine-2,4-diones were less active against the GDP-bound states of the K-Ras and N-Ras isoforms chosen as targets, acting as competitors of GDP for its binding pocket. The binding pattern in the active site of B-Raf was common to all the investigated compounds and was highly conserved. All data obtained encourage us to continue work in the field of the thiazolidinedione heterocyclic derivatives with therapeutic properties.

Conflicts of Interest:
The authors declare no conflict of interest.