Discovery of Thiopyrimidinone Derivatives as a New Class of Human Aldose Reductase Inhibitors

Diabetes is a chronic metabolic disorder characterized by insufficient insulin production, the cells’ inability to use this insulin, or a combination of both, leading to secondary complications such as diabetic neuropathy and retinopathy. One way to prevent or control such complications is the use of aldose reductase (AR) inhibitors. In this work, we synthesized and tested new candidates for human AR inhibition containing a 2-thiopyrimidin-4-one heterocycle as a central ring. The fifteen derivatives were tested in vitro and their binding modes were evaluated via molecular docking simulations. AR inhibition assays showed that all synthesized compounds were able to inhibit the AR enzyme at 50 μM. From these results, seven compounds were noteworthy and had their half maximal inhibitory concentration (IC50) values estimated, ranging from 2.0 to 14.5 μM. Molecular docking simulations showed that these compounds bind specifically to the catalytic subpocket and the results indicate a good association between in vitro and in silico studies.


Introduction
Aldose reductase (AR) (EC 1.1.1.21) is a ubiquitous aldo-keto reductase, 1-3 whose primary function is to catalyze the reduction of toxic aldehydes produced by reactive oxygen species into non-reactive alcohols. 3 AR activity in hyperglycemic patients is frequently associated with diabetes complications. [1][2][3] In order to perform its enzymatic activity, AR uses nicotinamide adenine dinucleotide phosphate (NADPH) as a cofactor in high demand. This overuse results in a low availability of NADPH for other enzymes, such as glutathione reductase (GR), which, in turn, may result in cellular oxidative stress. [1][2][3] Another feature related to an AR increased activity is linked to sorbitol, the main AR reaction product. As any highly hydrophilic substance, sorbitol does not diffuse rapidly through cell membranes and can be accumulated in tissues, causing ionic imbalance and cellular osmotic stress. 1,2 Sorbitol accumulation results in retinopathy, neuropathy and nephropathy; conditions that are closely associated with diabetes. 2 Diabetic retinopathy is one of the most common microvascular complications in this scenario, it can lead to cases of vision loss in adults 2 and affects almost 3 million people a year. 4 In addition to controlling blood glucose, treatment for diabetic retinopathy may be complemented with drugs such as acetylsalicylic acid, 5 vitamin E, 6 angiotensin-converting enzyme inhibitors (ACE), 7 vascular endothelial growth factor (VEGF) receptor inhibitors, 8 and, of course, AR inhibitors. 2 AR inhibitors present a wide structural diversity, bearing carboxylic acid moieties, cyclic imides, esters derivatives, and heterocyclic derivatives. 2 Among these, heterocyclic systems containing a carbonyl portion as part of the scaffold appear to be the most promising drug candidates [9][10][11] (Figure 1). As stated before, AR is a pervasive enzyme and some problems concerning the use of its inhibitors consist of undesirable effects that include increased levels of liver enzymes and gastrointestinal events, such as nausea and vomiting. 12 For this reason, there is an unceasing demand for more potent and selective inhibitors for diabetic retinopathy treatment.
Pyrimidines and their oxo-derivatives are widely reported due to their many biological activities, such as antiviral, 13 antimicrobial, 14,15 anti-inflammatory, 16 antioxidant, 14 anticancer agent, 17 and AR inhibition. 10,18 The pyrimidine ring can be synthesized by several methodologies. 19 Usually, a Michael reaction intermediate and a uronium derivative are condensed in basic medium. This reaction results in an alicyclic derivative that can be converted into pyrimidine or the pyrimidinone ring, depending on the nature of the uronium derivative. 20 This reaction can be carried out via multicomponent synthesis (MCRs). 21 In these procedures, three or more reagents are mixed together to obtain a single product, leading to better overall yields and shorter reaction times. 22 Thiopyrimidinones are an intriguing class of pyrimidinones. These compounds present a particular tautomeric equilibrium due to double bonds conjugated to heteroatoms inside and outside of the heterocycle. The most notorious compound in this family is thiouracil, a bioactive thiopyrimidinone closely related to uracil. 23,24 Thiouracil derivatives are well reported in the literature because of their pharmacological properties. 21,25,26 We have decided to synthesize thiopyrimidinone derivatives, and to conduct in vitro studies to assess AR inhibition as part of our research program aiming to develop new drug candidates for aldose reductase inhibition. These efforts were combined with molecular docking simulations to better understand ligand-enzyme binding modes.

Aldose reductase in vitro assays
The experiments were performed using the aldose reductase inhibitor screening kit by Biovision Incorporated © (Milpitas, USA). The series of compounds 5a-5e, 6a-6e and 7a-7e were tested as AR inhibitors. The assays were carried out, as per the kit manual, in a 96-round bottom transparent microplate (Thermo Scientific, Massachusetts, USA) and data were acquired using BIOTEK © (Vermont, USA) ELx808 absorbance spectrophotometer at 340 nm.
Stock solutions were prepared as described by the kit manual for each component. Dithiothreitol (DTT) stock solution (10 mM) was prepared using 2 μL of 1 M DTT diluted in 198 μL of ultrapure water. Then, the assay buffer solution containing DTT was prepared using 2 μL of DTT stock solution in 1998 μL of assay buffer provided in the kit. Aldose reductase was firstly reconstituted with 100 μL of assay buffer containing 10 μM DTT. Then, aliquots of 6 μL were made and stored at -80 °C. To prepare the AR working solution, each 6 μL of AR stock solution aliquots were mixed with 534 μL of assay buffer containing 10 μM DTT. The AR cofactor NADPH (nicotinamide adenine dinucleotide phosphate) stock solution (20 mM) was obtained after reconstitution with 440 μL of ultrapure water. Then, aliquots of 20 μL were made and stocked at -20 °C. To prepare the NADPH working solution, each 20 μL of NADPH stock solution aliquots were mixed with 340 μL of assay buffer.
All fifteen compounds (5a-5e; 6a-6e and 7a-7e) were at first tested at the same concentration of 50 μM. Then, some of these compounds were tested at different concentrations (i.e., 0.05, 0.5, 2.5, 5.0, 25, 40, and 60 μM) to provide a dose-response inhibition curve and to measure the half maximal inhibitory concentration (IC 50 ). In this case, in order to prepare test samples (S) for each compound, 50 mM stock solutions were prepared in DMSO. Then 100× and 20× dilutions were prepared for each concentration to be tested. For example, to prepare the test sample for 50 μM, 100× concentrated solutions (5 mM) were made by dilution of 5 μL of respective 50 mM stock solution with 45 μL of DMSO. Then, the dilution of 10 μL of 100× solutions with 40 μL of assay buffer was used to prepare the 20× working solution.
As indicated by the kit manual, control samples were prepared as follows: the background control (BC) is deprived of the AR enzyme and inhibitor, and we replaced its volume (100 μL) with AR buffer. Enzyme control (EC) is deprived of the inhibitor, and we replaced its volume (10 μL) with AR buffer. The inhibitor control (IC) was prepared by a 100× dilution of 10 mM Epalrestat (72 nM) in assay buffer to a final concentration of 5 μM. The solvent control (SC) was prepared by the same 20× dilution in assay buffer as described for test samples to certify that samples' solvent (DMSO) will not interfere in AR activity. With all working solutions prepared, the reaction components were added to a 96-well plate in the following order: 100 μL of assay buffer in the BC well, 10 μL of assay buffer in the EC well, 10 μL of 100× diluted Epalrestat in the IC well, 10 μL of 20× diluted DMSO in the SC well, and 10 μL of each 20× working solution for each compound to be tested (5a-5e; 6a-6e and 7a-7e). After that, 60 μL of NADPH working solution was added to all wells, whereas 90 μL of AR working solution was added to all wells except the BC well. The 96-well plate was then incubated at 37 °C for 20 min. Finally, the substrate solution was prepared by mixing 4 μL of AR substrate with 36 μL of assay buffer, with the final volume (40 μL) being added to each well after incubation time, resulting in a final total volume of 200 μL in each well for all compounds and control samples. Tests were done in duplicate and absorbances were immediately measured in kinetic mode for 60 min with 1 min interval at 37 °C. As described by the kit manual, two periods of time (t 1 and t 2 ) were chosen from absorbance over time curves and each slope was calculated using the respective absorbances (A 1 and A 2 ). Then, the following equation was used to calculate the relative inhibition percentage (equation 1). (1) Relative inhibition percentage calculation of the tested compounds as aldose reductase inhibitors, where EC: enzymatic control; S: sample.

Molecular docking
In order to assess the binding modes of the described compounds (5a-5e, 6a-6e and 7a-7e) at the aldose reductase active site, molecular docking simulations were performed using AutoGrid v.4.0 34 and AutoDock 35 v.4.0 36,37 as described in our previous works. 38,39 Crystallographic structures with different conformations at AR active site were obtained from PDB 40 (3DN5, 2FZD, 2NVD and 1US0). Enzymes were treated as rigid, while the respective ligands were treated as flexible. To prepare the enzymes for simulations, polar hydrogens were added and partial charges for protein atoms were assigned according to AMBER86 force field parameters, 41 which were assigned to all atoms. To prepare ligands (cocrystallized ligands or target compounds), charges were assigned by Gasteiger method to certify that all residues presented integer charges. In order to evaluate the reproducibility of our simulation procedure, a validation method was arranged (commonly called redocking) to prove that the procedure can reproduce the structural conformation of the cocrystallized ligand at the active site of the respective enzymes tested. These experiments were conducted with the ligands slightly outside the active site but still placed in a grid box, using 200 simulation steps. Redocking results were compared to the initial coordinates of the cocrystallized ligands to validate the methodology and it was successful. A grid with dimensions of 28 Å × 28 Å × 28 Å and spacing of 0.22 Å was centered at coordinates 16.691, -7.611, 13.869. All structures were superimposed to 2FZD to simplify the analysis of the results.
The binding modes for all the ligands were predicted based on affinity maps generated by AutoGrid v4.0, all affinity maps were calculated using the same settings. The Lamarckian genetic algorithm 42,43 was employed for molecular docking simulations using the following parameters: an initial population of 150 individuals/ generation, a maximum number of 27000 generations, and a total of 2.5 × 10 6 energy evaluation/generation. Thus, a total of 4.05 × 10 6 conformations for each of the ligands were generated in each of the 200 steps of the simulation. An elitism rate of 1 was applied to ensure that the top individual always survives into the next generation in conjunction with mutation and crossover rates of 0.02 and 0.08, respectively. In the local search, 300 steps were applied with a 0.06 probability of searching for one individual. At the end of the calculation at each simulation step, the ligand conformations that have the most favorable binding energy were selected, so that the atomic coordinates of the 200 conformers that better fit the binding site were obtained. All obtained conformers were structurally compared through their root mean square deviations (RMSDs), and clustered into groups of similar conformations with a RMSD tolerance of 2 Å for each cluster. Structures and figures were examined and generated by PyMOL. 44

Chemistry
Our strategy began with the selection of the 2-thiopyrimidin-4-one as a scaffold for AR inhibitor candidates. A literature review showed that good AR inhibitors contain multiple aromatic systems, fused or conjugated, and at least one carbonyl portion. 2,[9][10][11] Thus, we decided to perform modifications on the uronium moiety of the pyrimidinone ring, and then study how the S-substitution pattern can influence AR inhibition in vitro.
The multicomponent reaction mechanism for the formation of the pyrimidine ring has been described by de Andrade et al. 21 The synthesis begins with a Knoevenagel reaction between the methylene-active compound (ethyl cyanoacetate) and the aldehyde, producing the Michael reaction intermediate. This step is base-mediated and a water molecule is lost. The Michael reaction intermediate then reacts with thiourea, forming a uronium intermediate. Then, an intramolecular attack of a nitrogen electron pair occurs, thus closing the ring; in this step, an ethanol Scheme 1. Reagents and conditions: (a) K 2 CO 3 , EtOH/H 2 O, reflux; (b) ethyl bromoacetate (5a-5e), bromoacetic acid (6a-6e) or bromopropionic acid (7a-7e); triethylamine, acetone, room temperature. molecule is eliminated. Finally, the oxidation of the pyrimidine ring is carried out by atmospheric oxygen leading to the final product. 48 For the functionalization of 2-thiopyrimidin-4-ones 4a-4e, S-alkylation reactions were performed using a combination of triethylamine and acetone. This system was chosen from a series of experiments seeking the best synthetic strategy with different classical reaction conditions (Table 1) using 4a and ethyl bromoacetate. This method led to the exclusive formation of the product of interest (5a). The approach was then replicated for functionalization on the other series: one containing an acetic acid portion (6a-6e) and the other containing a propionic acid moiety (7a-7e). This shows that the same reaction conditions proved to be effective for the heterocycle functionalization with other alkylating agents.
The reaction with ethyl bromoacetate generated five products alkylated exclusively on the sulfur atom (5a-5e); in other words, no N or O-alkylated products were observed. Although the thiol form is not the predominant one in equilibrium, this reaction is possible due to the thiol/thione tautomeric equilibrium in 2-thiopyrimidin-4-ones, very similar to the keto/enol tautomerism. 49 The particular tautomeric equilibrium present in 2-thiopyrimidin-4-ones is due to the proton mobility in the uronium moiety of the heterocycle. Martos-Calvente et al. 24 showed that this equilibrium can be disturbed, with one tautomer being favored over another, and it can be affected by several factors, such as: physical state, solvent, temperature, pH and concentration. Several techniques, such as infrared, NMR and UV-Vis have shown this type of tautomerism. It is known that in the solid state, for example, the thione form is favored. In the liquid phase, the solvent plays an important role in the thiol/thione formation rate. Solvents with a high dielectric constant favor the thione form, while solvents with low dielectric constant favor the thiol tautomer. 23 The preference for nucleophilic attack in equilibrium prevails in the sulfur region over nitrogen atoms. Sulfur is a larger and softer atom, which makes it more polarizable. 50 These particularities explain why alkylations in 2-thiopyrimidin-4-ones occur preferentially in the sulfur atom, as reported herein (Scheme 2). Thiopyrimidinones do not solubilize in most organic solvents, but they do solubilize well in acetone. This feature, combined with the aforementioned factors, explains why acetone was the solvent of choice, even though it has intermediate polarity and a medium dielectric constant. 51 In vitro AR inhibition assays At first, the compounds 5a-5e, 6a-6e and 7a-7e were submitted to an initial trial at a fixed concentration of 50 µM in duplicate. 48 All compounds showed AR inhibition percentages that varied between 53 to 100% ( Table 2). The IC 50 values were estimated for the canonical compounds 5a, 6a and 7a (Table 3), tested at concentrations ranging from 0.05 to 50 µM. After these results, we observed that the compound 7a (IC 50 = 14 µM) showed an IC 50 greater than the compounds 5a (IC 50 = 6 µM) and 6a (IC 50 = 8 µM). This reveals that the elongation of the lateral chain does not favor AR inhibition in these compounds. Thus, the series 5 and 6 were chosen for a more in-depth investigation regarding IC 50 values.
After this preliminary analysis (Table 2), compounds 5b, 5d, 6b and 6d were selected to have their IC 50 values estimated. In series 5, the compounds 5a (IC 50 = 5.9 µM) and 5d (IC 50 = 6.4 µM) showed lower IC 50 values than the compound 5b (IC 50 = 8.7 µM), which means that a p-methyl substitution in the aromatic ring is not as favorable for AR inhibition as a p-F substitution. In other words, in series 5, which contains an ester derivative, the presence of an electron donating group on the aromatic ring decreased activity. Regarding series 6, compound 6b showed the lowest IC 50 value (2.0 µM), followed by compound 6d (3.3 µM), which is lower than the canonical 6a (IC 50 = 8.2 µM). Therefore, for compounds containing a -CH 2 COOH moiety, substitutions were more beneficial to AR inhibitory activity. These data are also presented comparatively in Figure 2.  acetone  55  3  pyridine  pyridine  45  4 AcONa ethanol a a Since many by-products were observed in TLC, the product was not purified and/or isolated. DMF: dimethylformamide.
The comparison of the two homologous series shows that compounds from series 6 are more promising drug candidates. This fact corroborates the findings from Singh Grewal et al., 2 who collected data from more than 70 carboxylic acid derivatives as AR inhibitors, demonstrating the AR enzyme affinity for this moiety. Among all these derivatives, over 55 compounds have a heterocycle moiety, and 33 compounds contain at least one sulfur atom, thus endorsing our hypothesis of testing thiopyrimidinones as AR inhibitors.
Singh Grewal et al. 2 also described two fused pyrimidinones as AR inhibitors (Figure 1). The first one, an aminopyrimidinone, showed an IC 50 of 34 µM. 52 The decreased degree of rotation in comparison to the thiopyrimidinones described herein could be one of the countless factors responsible for reducing AR inhibitory activity. The second molecule does not contain a carboxylic acid moiety, but three hydroxyl substituents, which probably counterbalanced the absence of that group and allowed for increased AR inhibitory activity (IC 50 = 0.1 µM). 18 Although the thiopyrimidinones described here displayed IC 50 at a low micromolar range, modifications on aromatic ring substituents (e.g., -NO 2 groups or fused rings) will be further investigated to improve the   thiopyrimidinones AR inhibitory activity to a nanomolar range.

Molecular docking
As previously stated, the AR enzyme has been studied as a target for possible inhibitors to prevent diabetic complications. Over the years, some authors [53][54][55][56][57] have analyzed AR crystallographic structures, and investigated the main residues involved in the binding pocket. AR has been described as composed of two subpockets, one of these, known as the catalytic subpocket, comprises the residues Tyr48, Lys77, His110 and Trp111, as well as its cofactor NADPH. 11,53,55 The second one, known as the specificity pocket, is formed by residues Trp111, Thr113, Phe122, Ala299 and Leu300, 11,53-55 these last residues adopt different conformations that depend on ligand size and properties. [55][56][57] Computer simulations can be used to investigate how these changes at the enzyme active site affect the multiple conformations of the crystallized structure. For this reason, we chose to perform molecular docking simulations in the following crystallographic structures: PDB ID 2FZD, 3DN5, 2NVD and 1US0. AR crystallographic structures were chosen based on works of Wang et al. 57 and Eisenmann et al. 11 In Wang et al. 57 work, the structures of PDB code 1US0 and 2FZD were simulated, in addition to a mutant 2PDK (L301M). This latter structure was replaced by 2NVD because they have similar conformations among residues of catalytic site (RMSD = 0.101, Figure S1, Supplementary Information (SI) section), but in absence of mutations. Finally, 3DN5 structure was chosen due to different conformations of the residues on active site and the chemical similarity between the cocrystallized ligand and the inhibitors proposed here, like sulfur atoms and carboxylic acid portions.
Firstly, we performed the validation of the chosen molecular docking method by means of redocking experiments. In these assays, we implemented the molecular docking with cocrystallized ligands in their respective crystallographic enzyme structures to evaluate if the same conformation can be achieved after the simulations. After the redocking simulations, we observed that the ligands showed high conformational similarity (RMSD < 0.5 Å), in comparison with the original position of the cocrystallized ligands (Figures 3a-3d). It was also observed that the cocrystallized ligands interact with the Tyr48, His110 and Trp111, important residues for catalysis.
By performing redocking experiments in the 3DN5 structure, we found that ligand 53N also interacts with Leu300, a residue located at the specificity pocket. 11 Conformational changes were also observed in others residues from the specificity pocket (Leu301, Ser 302 and Cys303) among all structures studied herein, as reported in previous literature. [55][56][57] In addition, the change in the specificity pocket conformation alters the opening of the binding site, which switches between open ( Figure 4a) and closed (Figure 4b) depending on the ligand, contributing to distinct docking results in different crystallographic structures. 58 After the redocking studies, we began to investigate how series 5, 6 and 7 would interact in the available crystallographic structures. The results obtained after simulations for all ligands in each AR structure showed different binding modes varying with the structure employed, which would be expected due to conformational changes at the binding site. The structures 1US0 and 2NVD showed the fewest interactions of strong character (hydrogen bond). The structures 2FZD and 3DN5 presented more interactions at the catalytic subpocket, and the resulting conformations showed interactions with residues Tyr48, His110 and Trp111, which are the same as described in the redocking experiments. Between these two structures, 2FZD was the one with the greater number of strong interactions for most of the tested ligands.
Since the 2FZD structure showed the best binding modes for most compounds, we decided to use this enzyme as a model to further correlate the results obtained from in vitro assays (Figure 4). The most populous cluster was chosen for each compound (out of 200 runs). Experimental results with the canonical compounds 5a, 6a and 7a showed that ligand 5a was the most promising inhibitor, with the lowest IC 50 value (5.9 µM). In the docking simulations, all three ligands occupied the catalytic subpocket, interacting specifically with the residues Tyr48, His110 and Trp111. However, we noted that the ligand 5a displayed a different direction in accommodating its aromatic ring, thus interacting with Trp111 through π-stacking interactions ( Figure 4c). This extra interaction should be the reason for its best performance as an inhibitor. In the case of acidic derivatives 6a and 7a, there were no significant conformational differences at the active site, even though 6a is the most active among the two compounds.
Ligands 5b and 6b (Figure 4d), and 5d and 6d (Figure 4e) showed different conformations among them. Ligands 5b and 6b both interact specifically with the specificity pocket residues. Compound 6b interacts with Ser302, while 5b does not bind via a stronger force, which explains the lowest IC 50 for 6b (2.0 µM). On the other hand, compounds 5d and 6d interact mostly with the catalytic subpocket, as seen for canonical compounds. While ligand 6d interacts with Tyr48, His110 and Trp111 residues, compound 5d interacts only with Tyr48. Interestingly, this approach is mediated by the fluorine atom and not via the carbonyl group, as observed for other derivatives. Compound 6d (3.3 µM) is more active and this is possibly due to its multiple interactions with the AR active site.

Conclusions
Fifteen thiopyrimidinones were synthesized and obtained in yields ranging from 45 to 90% and have emerged as new aldose reductase inhibitors. These compounds were evaluated in vitro as AR inhibitors using a colorimetric method. In the first round, all synthesized compounds were screened at a fixed concentration of 50 micromolar. From these results, some derivatives stood out as promising drug candidates. Next, the canonical compounds 5a, 6a and 7a were evaluated individually and their respective IC 50 were 5.9, 8.2 and 14.5 µM. We then performed molecular docking simulations with these ligands in four AR crystallographic structures to assess the enzyme-inhibitor interactions. All compounds showed interactions with the AR catalytic subpocket, especially compound 5a, which showed more interactions with the AR binding site and, consequently, lower IC 50 value. In the acid derivatives 6a and 7a, we could observe that the elongation of the lateral chain does not favor AR inhibition. Regarding the compounds 5b and 5d, results showed that p-methyl substitution in the aromatic ring is not as favorable as p-F substitution with respect to AR inhibition, thus increasing IC 50 values (5b-8.7 µM and 5d-6.4 µM). Compounds 6b and 6d showed the lowest IC 50 values (2.0 and 3.3 µM, respectively), and these results indicate that even though compound 5a happens to be the most promising among the canonical compounds, serie 6 (containing a -CH 2 COOH moiety), especially compounds 6b and 6d, were the two most promising thiopyrimidinones as AR inhibitors. Molecular docking simulations were helpful to better understand the enzyme-inhibitor interactions that could explain the differences in IC 50 values among all compounds tested herein. Redocking experiments were successful and all thiopyrimidinones showed interactions in the catalytic subpocket (Tyr48, Lys77, His110, Trp111 and the cofactor NADPH). Among the canonical compounds 5a, 6a and 7a, the compound 5a displays a different conformation, allowing the π-stacking interaction with Trp111. However, in an in-depth investigation about the influence of aromatic ring substitutions, the compounds 6b and 6d, which showed the best IC 50 values, showed more and stronger interactions than the compounds 5b and 5d. Even though the thiopyrimidinones described herein displayed IC 50 values at a low micromolar range, modifications on aromatic ring substituents (e.g., -NO 2 groups or fused rings) will be further investigated to improve their AR inhibitory activity to the nanomolar range.

Supplementary Information
Supplementary information regarding NMR spectra and ligands and enzyme coordinates extracted from molecular docking simulations are available free of charge at http://jbcs.sbq.org.br as PDF file.