Development and exploration of novel substituted thiosemicarbazones as inhibitors of aldose reductase via in vitro analysis and computational study

The role of aldose reductase (ALR2) in causing diabetic complications is well-studied, with overactivity of ALR2 in the hyperglycemic state leading to an accumulation of intracellular sorbitol, depletion of cytoplasmic NADPH and oxidative stress and causing a variety of different conditions including retinopathy, nephropathy, neuropathy and cardiovascular disorders. While previous efforts have sought to develop inhibitors of this enzyme in order to combat diabetic complications, non-selective inhibition of both ALR2 and the homologous enzyme aldehyde reductase (ALR1) has led to poor toxicity profiles, with no drugs targeting ALR2 currently approved for therapeutic use in the Western world. In the current study, we have synthesized a series of N-substituted thiosemicarbazones with added phenolic moieties, of which compound 3m displayed strong and selective ALR2 inhibitory activity in vitro (IC50 1.18 µM) as well as promising antioxidant activity (75.95% free radical scavenging activity). The target binding modes of 3m were studied via molecular docking studies and stable interactions with ALR2 were inferred through molecular dynamics simulations. We thus report the N-substituted thiosemicarbazones as promising drug candidates for selective inhibition of ALR2 and possible treatment of diabetic complications.


Results and discussion
Chemistry. The synthesis of thiosemicarbazone hybrids 3a-o was carried out as detailed in Scheme 1. The starting thiosemicarbazide compounds 1a-o were prepared by reaction of the corresponding isothiocyanate with hydrazine in an ethanol/water solution as reported previously by da Silva et al. 15 . The thiosemicarbazone derivatives 3a-o were synthesized through condensation of the appropriate N 4 -substituted thiosemicarbazide (1a-o) with 3,5-di-tert-butyl-2-hydroxybenzaldehyde (2) in methanol with a catalytic quantity of glacial acetic acid. Reaction conditions were optimized by treating phenyl thiosemicarbazide (1a) (1 mmol) with 3,5-di-tertbutyl-2-hydroxybenzaldehyde (2) (1 mmol) using different solvents such as ethanol, methanol, DCM, THF and DMSO. Methanol was established as the best solvent for the reaction while glacial acetic acid (1-2 drops) was identified as an effective catalyst. A variety of target N 4 -substituted thiosemicarbazones were obtained in pure form in good to excellent yields (79-90%) via recrystallization from ethanol.
The structures of compounds 3a-o were confirmed by CHN analysis and various spectroscopic techniques. The infrared (IR) spectra showed absorption bands in the range of 1547-1580 cm −1 due to the new azomethine linkage (C=N) while C=S stretching was observed in the 1189-1245 cm −1 range. NH bands appeared in the 3218-3328 cm −1 region while absorption in the 3398-3450 cm −1 range indicated the presence of an OH functional group. In 1 H NMR, the methyl protons of two tertiary butyl groups resonated at 1.27-1.34 ppm and 1.40-1.48 ppm, respectively, while the proton of the azomethine moiety (N=CH) appeared as a singlet in the 8.06-8.39 ppm range. Similarly, the NH-CS proton appeared as a singlet in the range of 7.42-10.15 ppm, whereas the NH-N proton was also observed as a singlet in the 9.75-11.18 ppm region. In thiosemicarbazone 3h, a derivative bearing a cyclohexyl group, a doublet was observed for NH-CS at 7.42 ppm. The most downfield signal observed was consistently attributed to the OH functional group, appearing in the region of 9.87-11.91 ppm. The structures of compounds 3a-o were also confirmed by 13 C NMR spectroscopy. The methyl carbons of the tertiary butyl moiety were found to resonate in the region of 29.40-31.74 ppm while methyl carbons directly attached to the In LC-MS spectra, molecular ion peaks of the compounds appeared as [M+H] + which were consistent with the molecular weights of the synthesized derivatives.

Enzyme inhibition and structure-activity relationship (SAR).
Compounds 3a-o were tested against ALR2 to evaluate their inhibitory potential against aldose reductase whereas selectivity was determined via performing ALR1 inhibition assay. For the enzyme inhibition assay, ALR2 and ALR1 were extracted from bovine sources (ALR2 from eyes and ALR1 from kidneys). Moreover, inhibitions studies were also carried out on the expressed ALR2 (human AKR1B1) and inhibition data for ALR2-bovine source and expressed enzyme was compared and analyzed. The protocols for enzyme extraction and expression in the bacterial system are included in the section of "Supporting information". The compounds which demonstrated greater than 50% target inhibition at 100 µM were further investigated to determine their IC 50 values, summarized in Table 1. For the purpose of comparison and validation of enzyme inhibition data, these compounds were tested against human AKR1B1 (hAKR1B1) expressed in E. coli BL21 (DE3). We found a similar pattern of enzyme inhibition for the synthesized compounds against extracted enzyme (ALR2) as well as expressed enzyme (hAKR1B1). Enzyme inhibition data (against ALR2, hAKR1B1 and ALR1) is summarized in and 3o all showed low to moderate inhibition of both ALR1 and ALR2. From these data, a structure-activity relationship (SAR) for the N-substituted thiosemicarbazones with respect to both ALR1 and ALR2 inhibition was established. For all synthesized compounds, the NH group of the thiosemicarbazone was functionalized with either a phenyl or benzyl group. These aromatic rings were substituted with different electron withdrawing or electron donating groups at various positions. The phenyl ring of 3a was associated with non-selective inhibition of both ALR1 and ALR2, with the addition of methyl substituents at the ortho and/or para positions (3b, 3d, 3k) resulting in a decrease in inhibition of both enzymes (≤44.63% for ALR2, ≤31.62% for ALR1). A meta methoxy group (3c) resulted in strong inhibition of both enzymes (ALR2 IC 50 4.49 µM, ALR1 IC 50 4.21 µM) while incorporation of electron withdrawing fluorine (3g) and chlorine (3l) atoms at the para position of the phenyl ring, as well as the benzyl group (3f), enhanced inhibition of ALR2 and caused selective inhibition (3f ALR2 IC 50 3.12 µM, 4.27% inhibition of ALR1). Other compounds possessing fluorine and chlorine substitutions at ortho and meta positions on the phenyl ring (3j, 3m) also showed selective inhibition of ALR2 (3m ALR2 IC 50 1.18 µM, 5.98% inhibition of ALR1). Bromine (3i) and isopropyl (3n) substitutions at the phenyl ring para position, as well as cyclohexyl (3h), benzyl (3e) and phenethyl (3o) analogs of 3a, all resulted in significant loss of both ALR1 and ALR2 inhibition (≤39.13% for ALR2, ≤29.91% for ALR1).
DPPH radical scavenging activity. DPPH (2,2-diphenyl-1-picrylhydrazyl) was used to explore the free radical quenching ability of 3a-o so as to provide a gauge of their antioxidant potential. The experiment was conducted according to a previously reported protocol with slight modifications 16 . Percent free radical scavenging activities (% FRSA) of 3a-o were determined by spectrophotometric analysis at 517 nm using a homogenous mixture of methanolic DPPH (0.025 mg/mL) and a 100 µM solution of the tested compound, with ascorbic acid used as a positive control (Fig. 3). The majority of the compounds showed strong antioxidant potential (60-90% FRSA), though 3n was notably weak (<10% FRSA). 3j exhibited 89.56% FRSA, a value greater than that of the positive control (Table 2).

Molecular docking results.
In vitro testing identified 3l and 3m as promising selective inhibitors of ALR2, whereas 3a and 3c showed non-selective inhibition of both ALR2 and ALR1. Therefore, molecular docking studies were carried out for compound 3m and their interactions with the amino acid residues of the AKR1B1 active site analyzed. To perform this docking analysis, the crystal structures of AKR1B1 (1US0) 17 and AKR1A1 (3FX4) 18 were downloaded from the Protein Data Bank and docking protocols from a recent study conducted within the group were used 13 .
Before docking of the synthesized inhibitor, redocking was done with inhibitor LDT320 (co-crystallized with ALR2) for the purpose of validation. To achieve reproducible docking results, the root-mean-square deviation (RMSD) value of co-crystallized inhibitor was found 0.69Å while using docking software (LeadIT). The most active inhibitor of ALR2 identified, 3m, was selected for docking and it was observed that 3m showed a similar binding orientation and conformation within the active pocket of ALR2 as the co-crystallized inhibitor 17  www.nature.com/scientificreports/ was also docked against the active pocket of ALR1 and its interactions compared to those formed with ALR2. Two-and three-dimensional views of the interactions of 3m within the active site of ALR2 are shown in Fig. 4; notable interactions included hydrogen bonds between Val47/Tyr48 and the hydrogen atoms of the two thiosemicarbazone -NH moieties, a π-alkyl interaction with Trp111 and van der Waals interactions with several hydrophobic residues including Trp20, Lys21, Phe121, Trp219 and Leu300.

Molecular dynamics simulations.
To revalidate the aforementioned docking results, molecular dynamic simulations of 3m in the ALR2 active site in the presence of cofactor (NADPH) were carried out. The enzymecofactor-inhibitor system was solvated in a cubic PBC water box and the overall charge was neutralized using the counterions. The system was observed up to 50 ns.
The RMSD values of the protein backbone, cofactor and inhibitor (3m) were observed to determine any drastic change; there was no considerable change in the RMSD of the protein backbone and cofactor. However, the RMSD of 3m showed drastic fluctuations from the start of the simulation up until 20 ns, after which it remained constant. Upon visual inspection of the 50 ns trajectory, it was clear that the selected docked conformation was not very stable, drifting from its initial position within the anionic pocket ( Fig. 5a) to the specificity pocket (Fig. 5b). The average structure from the trajectory after 20 ns of MD simulations was observed to occupy the specificity defining pocket. 3m formed several interactions with residues within the pocket, including a hydrogen bond between the hydrogen atom of one of the -NH moieties and Ala299, an arene-H interaction with Leu301 and hydrophobic interactions with Trp20, Phe122, Pro218, Trp219 and Leu300 (Fig. 5c). These results provide a molecular-level rationale for the specificity of 3m as observed in the in vitro enzyme inhibition assay.
To further investigate the drastic fluctuations in RMSD values seen for 3m, the short-range coulombic and Lennard-Jones (LJ) interactions were plotted. Fluctuations in RMSD values coincided with fluctuations in both Table 1. IC 50 values for ALR1 and ALR2 inhibition. a The IC 50 value (the half-maximal inhibitory concentration) is the concentration of drug required to decrease enzyme activity by 50%. b The percentage inhibition for enzymes measured at 100 µM of inhibitor concentration. c Sorbinil is a standard inhibitor of ALR2. d Valproic acid is a standard inhibitor of ALR1. Results are mean values ±SEM based on three measurements. Synthesized compounds' codes were written in bold text whereas the most significant and selective inhibitor's values were made bold upon suggestion of reviewer's comments.  www.nature.com/scientificreports/ the short-range coulombic and LJ interactions, occurring until around 20 ns. The higher values of short-range coulombic and LJ interaction energy mean 3m is not as stable as after 20 ns and therefore the most probable pose is regarded as the one obtained after running the 20 ns simulations (see Fig. 6).
Pharmacokinetic profile and ADME evaluation. Absorption, distribution, metabolism and excretion (ADME) studies for 3a-o were carried out using an in silico method (SwissADME) that uses various algorithms to predict ADME parameters. The results of this analysis (Table 3) were used to draw a brain or intestinal estimated permeation (BOILED-Egg) plot (Fig. 7), a plot of topological polar surface area (TPSA) against Wildman-Crippen partition coefficient (WLOGP) that predicts gastrointestinal absorption (white area) and blood-brain barrier permeation (yellow area) 18 . While the majority of the synthesized compounds were predicted to show good gastrointestinal absorption, none were predicted to cross the blood brain barrier. The results of the SwissADME analysis further indicated that all compounds satisfy Lipinski's rule of five 19 . Druglikeness was also evaluated through applying a pan-assay interference compound (PAINS) filter 20 , with all

Conclusion
A series of thiosemicarbazones with phenolic moieties appended to the thiosemicarbazone backbone were synthesized with the aim of generating a chemical scaffold with potent, selective aldose reductase (ALR2) inhibitory activity as well as antioxidant activity. Such a scaffold could potentially possess a synergistic ability to treat diabetic complications through dual aldose reductase inhibition and oxidative stress suppression. In the current study, an in vitro ALR2 inhibition assay demonstrated that compounds 3f, 3g, 3j, 3l and 3m are strong and selective ALR2 inhibitors with IC 50 values in the low micromolar range (3.12, 2.38, 4.1, 1.46 and 1.18 µM, respectively). 3m, the most potent inhibitor of the set, showed strong antioxidant properties with a percent free radical scavenging activity of 75.95%. Molecular docking and molecular dynamics simulation studies were used to suggest a molecular-scale rationale for the selective ALR2 inhibitory activity of 3m. This compound therefore represents a potential drug candidate for treatment of diabetic complications.  RRID: Addgene_82928). Biological assay substrates (D,L-glyceraldehyde and sodium-D-glucoronate) and nicotinamide adenine dinucleotide phosphate (NADPH) were purchased from Sigma Aldrich (Merck KGaA, USA). Synthetic building blocks, reagents, solvents and thin layer chromatography plates were purchased from Sigma Aldrich. Fourier-transform infrared spectroscopy (FTIR) analysis in the range of 4000-500 cm −1 was performed using a Bruker Vector-22 spectrometer. All NMR spectra were obtained at room temperature using a Bruker Ascend 400 MHz NMR spectrometer and interpreted using ACD/NMR Processor Academic Edition software; chemical shifts (δ H) are expressed in parts per million (ppm) relative to deuterated chloroform (CDCl 3 ; residual signal 1 H δ = 7.26, 13 C δ = 77.2) or deuterated dimethyl sulfoxide (DMSO-d 6 ; residual signal 1 H δ = 2.50, 13 C δ = 39.5), coupling constants are expressed in Hz and multiplicities in 1 H NMR spectra are quoted as follows: s = singlet, d = doublet, t = triplet q = quartet, dd = doublet of doublets, m = multiplet. Analytical liquid chromatography-mass spectrometry (LC-MS) was employed to monitor reaction progression and for compound identification. LC-MS analysis was performed on an Agilent InfinityLab LC/MSD System consisting of an Agilent 1290 Infinity II Analytical-Scale LC Purification System coupled to a 6120 Quadrupole mass spectrometer. High-performance liquid chromatography was carried out using an Onyx™ Monolithic C18 column (50 x 4.6 mm) with water (A) and acetonitrile (B) as the mobile phases, with formic acid (0.1%) added to both to ensure acidic conditions throughout the analysis. Gradient conditions used were as follows: Method A (5 min), flow rate 1.0 mL/min, 100 μL was split via a zero dead volume T piece which passed into the mass spectrometer. The wavelength range of the UV detector was 220-500 nm. Gradient progressed from 95% A/5% B to 10% A/90% B over three minutes, then to 5% A/95% B over a further 30 seconds, was held constant at 5% A/95% B for a further minute and finally returned to 95% A/5% B over a final 30 seconds. Method B (10 min), flow rate 0.5 mL/min, 200 μL was split via a zero dead volume T piece which passed into the mass spectrometer. The wavelength range of the UV detector was 220-400 nm. Gradient progressed from 95% A/5% B to 50% A/50% B over three minutes, then to 20% A/80% B over two further minutes, to 5% A/95% B over a further 1.5 minutes, was held constant at 5% A/95% B for a further 1.5 minutes, returned to 95% A/5% B over a further 0.2 minutes and remained at 95% A/5% B for a further 1.8 minutes. Analytical liquid chromatography was carried out using the following parameters: injection volume 10 μL; draw speed 100 μL/min; ejection speed 400 μL/min; wait time after drawing 1.2 s. Mass spectrometry data (both ESI+ and ESI-modes) were collected using the following parameters: capil-   6,21 . The reaction mixture without cofactor was incubated at 32°C for 10 min, then the enzymatic reaction was initiated with the addition of NADPH and monitored for 5 minutes. Similar protocols were followed for ALR1 and ALR2, however the substrate was different for each enzyme, with sodium-D-glucoronate and DL-glyceraldehyde used as the substrates for the ALR1 and ALR2 assays, respectively. Sorbinil was employed as standard inhibitor of ALR2 while valproic acid was used as a standard inhibitor for ALR1. In addition, similar protocols were adopted for human AKR1B1 that was expressed in E. coli BL21 (DE3), though the determined protein concentration was 12 µg/mL for the expressed enzyme. Detailed protocols for the preparation of ALR1, ALR2 and expressed human AKR1B1 can be found in the supporting information.
The newly synthesized N-substitute thiosemicarbazones (3a-o) were dissolved in 100% DMSO and diluted with deionized water, keeping the DMSO concentration equal to 0.1% in the assay. Compounds were initially tested for percent inhibition at a concentration of 100 µM and IC 50 values were determined for various dilutions up to 10 nM. Where compounds showed percent inhibition greater than 50%, their IC 50 values were calculated through non-linear regression analysis using GraphPad Prism version 8.

Methodology for docking and simulation studies.
To investigate the probable binding mode of the specific inhibitor 3m, molecular docking and molecular dynamics simulation studies were performed. The FlexX utility of BioSolveIT's LeadIT software package was used to perform the docking studies 22 . The x-ray crystallographic structure of ALR2 (PDB ID 3FX4) was downloaded and prepared using the default docking parameters of the software 23 . Docking was performed in the presence of cofactor NADP. Initially the docking protocol was revalidated by redocking the co-crystallized ligand and comparing its RMSD value. Enthalpy entropy hybrid approach of FlexX utility was used for scoring and ranking of the conformational poses. The highest scoring poses were further subjected to HYDE assessment in order to assess their binding affinities 24,25 .
Molecular dynamic simulation of 3m was carried out using GROMACS 26,27 . The latest CHARMM36 forcefield was used with TIP3P as an explicit water model 28 . The docked pose of 3m was used as an initial coordinate and the topology and parameter files were obtained using CHARMM General Force Field (CGENFF) web-based server (https:// cgenff. umary land. edu). The protein-cofactor-inhibitor complex was prepared and wrapped in the TIP3P water box and neutralized with Na + and Clcounterions. The complex system was minimized using steepest decent and conjugate gradient methods until the maximum force experienced by the system was less than 10 3 KJ mol −1 nm −1 . The system was allowed to equilibrate for 100 ps using NVT (isothermal-isochoric) and NPT (isothermal-isobaric) ensemble. The complex system was observed to reach 300 K temperature and the pressure was observed to be around 1 atmosphere prior to running the production run. An MD simulation of about 50 ns was performed. Twin-range van der Waals and coulomb interactions were used to determine the non-bonded interactions with a cutoff of 1.0 nm. VMD v9.13 and XMGRACE v5.1.19 were used for visualization and plotting of graphs 29,30 . General procedure for synthesis of thiosemicarbazone derivatives (3a-o). The thiosemicarbazone derivatives (3a-o) were synthesized by adding equimolar quantities (1 mmol) of the appropriate N 4 -substituted thiosemicarbazide (1) and 3,5-di-tert-butyl-2-hydroxybenzaldehyde (2) to an oven dried flask, dissolving in 10 mL methanol and adding a few drops of glacial acetic acid as a catalyst 7,31,32 . The reaction mixture was then heated under reflux for 2-3 hours until the reaction was complete as shown by TLC. The mixture was then cooled to room temperature, allowing the product thiosemicarbazone to precipitate. The crude product was filtered under vacuum, washed with hot methanol followed by ether and then oven dried. Finally, the crude product was recrystallized from ethanol to afford the target thiosemicarbazone in good to excellent yield (79-90%).Characterization data for synthesized thiosemicarbazone derivatives is provided below.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.