In-silico identification of novel Cis-aconitate decarboxylase inhibitors as potential anti-inflammatory agents using molecular docking and dynamics

Abstract

Cis-aconitate decarboxylase (CAD), also known as ACOD1 or IRG1, catalyzes the conversion of cis-aconitate to itaconate, playing a pivotal role in innate immunity and inflammatory diseases. Although CAD is a key enzyme in the pathophysiology of inflammatory diseases, no specific inhibitors for CAD currently exist. In this study, we screened 86,326 compounds similar to CAD ligand from the Zinc database to identify potential CAD inhibitors using molecular docking with Molegro Virtual Docker and AutoDock-Vina, followed by molecular dynamics (MD) simulations. The top candidates were further assessed through molecular dynamics (MD) simulations, density functional theory (DFT) calculations, and free energy estimation using the MM/GBSA method. Pharmacokinetics, drug-likeness, and toxicity were evaluated using SWISSADME and Discovery Studio. Among the tested ligands, four compounds demonstrated strong binding affinity, stable interactions, and favorable pharmacokinetic properties, including high gastrointestinal absorption, solubility, and non-toxicity. These compounds demonstrated promising pharmacokinetics, good gastrointestinal absorption, solubility, non-toxicity, and compliance with Lipinski’s rules for drug-like properties. MD simulations further confirmed the stability of ligand-CAD complexes, with cumulative deviations and fluctuations under 2 Å. These findings suggest novel CAD inhibitors with potential as anti-inflammatory agents, paving the way for CAD-targeted drug discovery.

Introduction

Computer-aided drug design (CADD) is frequently used to find the probable enzyme inhibitors, substantially reducing the time and cost required for new drug development. Molecular docking as a CADD technique which docked small molecules into the macromolecular structures for scoring their complementary values at the binding sites has received much attention as an attractive tool and vibrant research area for structure-based drug design^1,2. Various computational methods such as molecular docking, molecular dynamics simulations, and quantum mechanics-based approaches have been extensively applied to understand the binding mechanisms of enzymes, receptors, and other biological targets. These techniques have proven valuable in assessing ligand-binding efficiency and guiding the design of lead compounds³. Moreover, the integration of machine learning with molecular dynamics has significantly improved the accuracy of drug-receptor interaction predictions, providing deeper insights into the drug design process. The application of pharmacophore modeling and structure-based drug design has also been pivotal in understanding how computational models can assist in identifying and optimizing drug candidates⁴. Additionally, molecular docking combined with ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) predictions has played a crucial role in the discovery of novel anti-inflammatory agents, demonstrating the importance of in-silico methods for reducing the time and cost of drug development⁵. These recent advancements highlight the essential role of computational approaches in the early-phase identification of target-specific inhibitors, reinforcing their value in accelerating drug discovery processes.

Inflammation is a highly complex response that underlies a variety of pathological and physiological processes, including protection against pathogen resistance as well as involvement in metabolic disorders and diseases⁶. A successful acute inflammatory response leads to the elimination of infectious agents and is followed by a repair phase⁷. Excessive and aberrant immune reactions, concurrent with the production of inflammatory cytokines and chemokines, are linked to the progression and development of several neurodegenerative and metabolic diseases^8–11. Taken together, identifying compounds or interventions that can effectively modulate inflammation may contribute to the development of novel treatments or approaches aimed at reducing the progression of neurodegenerative and metabolic diseases^8–11.

The immune response gene 1 (IRG1), which is highly expressed in inflammatory cells including macrophages, plays an essential role in regulating inflammation in several mammalian tissues. The IRG1 gene encodes the enzyme cis-aconitate decarboxylase (CAD), which catalyzes cis-aconitic acid to itaconic acid during the Krebs cycle^12,13. Studies on animals and humans have shown that itaconic acid and its derivatives, like dimethyl itaconate (DI) and 4-octyl itaconate (OI), can have a pivotal effect on the modulation of immune and inflammatory mechanisms and reduce the production of inflammatory cytokines such as tumor necrosis factor-alpha (TNFα), interleukin 6 (IL-6), and interferon beta (IFN-β) in LPS-stimulated immune cells by activating the nuclear factor erythroid 2-related factor 2 (NRF2) and heme oxygenase-1 (HO-1) pathway, inhibiting the NOD-, LRR- and pyrin domain-containing 3 (NLRP3) inflammasome/pyroptosis pathway, and facilitating the expression of Activating transcription factor 3 (ATF3)^14–17. It has been reported that silencing of CAD expression enhances the inflammatory response in bone marrow-derived macrophages (BMDM) of LPS-stimulated mice and myeloid cells infected with Mycobacterium tuberculosis^18,19. Contrary to the above findings, some studies have reported that the increased expression of the IRG1 protein in syncytial viral infection (RSV) in children stimulates the expression of pro-inflammatory cytokines and ROS production. It has also been observed that suppression of IRG1 prevents damage to rat lungs, as well as inhibits the expression of RSV-induced pro-inflammatory genes¹⁹. Evidence shows that IRG1 plays a crucial role in suppressing antitumor activity and may cause tumor metastasis by decreasing the penetration of CD8⁺ T cells into the ovarian tumor microenvironment²⁰. Also, monocytes isolated from ascites fluid of ovarian cancer patients have high levels of IRG1, and suppression of IRG1 using shRNA, significantly alleviates peritoneal tumors. Therefore, targeting IRG1 can be a potential therapeutic target for treating peritoneal tumors²¹. Increased expression of IRG1 in U251 glioma cells in vitro as well as in vivo leads to growth, invasion, and tumorigenesis of glioma cells, and suppression of its expression prevents nuclear factor kappa B (NF-κB) and signal transducer and activator of transcription 3 (STAT3) expression contributing to reduced proliferation and metastasis of glioma cells²². Furthermore, the NOTCH4-GATA4-IRG1axis provides a gene network that has an effect on cancer progression and patient survival through the PI3K/AKT pathway NFκB- and PPARγ-related mechanisms²³.

Despite the crucial role of CAD in inflammatory diseases and cancer, no selective inhibitor for this enzyme has been identified to date^19–23. Given its therapeutic significance, developing CAD inhibitors could provide new opportunities for treating inflammatory-associated diseases and malignancies. However, the lack of known inhibitors poses a major research gap, limiting therapeutic advancements in this area. Unlike previous studies, which have primarily focused on the biological functions of CAD, this study represents the first attempt to identify and characterize potential CAD inhibitors using an integrated computational approach. Therefore, this study was performed to identify novel compounds with high binding affinity against CAD enzyme and appropriate physicochemical properties by computational molecular docking and structural dynamics studies.

Materials and methods

Analysis of IRG1 protein sequence, structure, and interaction network

The sequences of human IRG1 protein were retrieved from UniProtKB with the accession number A6NK06. The BLASTP was used for identification of the status of homologs in Homo sapiens. The protein-protein interaction (PPIs) network of IRG1 with other proteins has been created on STRING database (version 11.0; July 2019; https://string-db.org/) The protein interaction network generated by the String platform helped in identifying potential targets for novel inhibitors²⁴.

Ligand data set selection

The crystallographic structure of human cis-aconitate decarboxylase (CAD/ACOD1) bound in the active-site pocket (PDB 6R6U) was downloaded from the RCSB PDB, and the co-crystallographic ligand was extracted with PyMOL (v1.7.4.5). This ligand was used as the query in ZINC15 (https://zinc15.docking.org) to retrieve commercially available small molecules with similar topology/shape²⁵. A total of 86,326 ligands similar to the main ligand were suggested in the zinc server. We then applied a multi-stage medicinal-chemistry funnel to obtain a proper set for docking. Drug-likeness/developability filters^26,27 were imposed to retain compounds with molecular weight 150–500 Da, HBD ≤ 5, HBA ≤ 10, cLogP between − 1 and 5, topological polar surface area (TPSA) ≤ 140 Ų, and rotatable bonds (RB) ≤ 10; only purchasable/in-stock ZINC entries were kept. The reactivity/toxicophore exclusion was performed by removing structures flagged by PAINS and Brenk/NIH alerts (e.g., catechols/quinone-like motifs, Michael acceptors, nitroso/azo groups, aldehydes, highly strained rings), and salt forms/duplicates were collapsed by InChIKey. The protonation and tautomer states were standardized to pH 7.4 ± 0.5, followed by 3D conformer generation (ETKDG-like) and MMFF94s energy minimization to ensure consistent geometry prior to docking. Finally, to preserve structural breadth, the remaining pool was clustered by ECFP4 fingerprints with Butina clustering (Tanimoto = 0.60), and representatives were selected from each cluster by balancing drug-likeness and pharmacophore fit. This funnel reduced the initial set to 29 chemically diverse, purchasable ligands, which were then subjected to docking (AutoDock Vina and Molegro Virtual Docker) and pose refinement/rescoring (LeDock) before molecular dynamics and MM/GBSA analyses.

Evaluation of protein stability

The IUPred2 server (https://iupred2a.elte.hu) was used to predict the presence of disorder in the CAD protein structure following the binding of selected ligands.

Ligand and receptor interaction analysis

Ligand-active site docking

The CAD protein, along with 29 ligands obtained from virtual screening of the Zinc15 database, was imported into MOLEGRO VIRTUAL DOCKER (MVD) version 6.0 (http://molexus.io/molegro-virtual-docker/) and Auto Dock-Vina version 1-1-2 software²⁸. Each ligand was then docked into the enzyme’s active site. Prior to docking, both the proteins and ligands underwent energy minimization using MVD or Chimera software. Additionally, residues that were found to be in an unfit site in the protein structure were structurally modified using MVD software, based on default settings. The best possible case was considered to have the undermost binding energy (BE) in kcal/mol. BE values less than − 6 kcal/mol indicate a strong binding affinity. The amino acids involved in this interaction were identified and images of these interactions were obtained using Discovery Studio software (https://discover.3ds.com/discovery-studio-visualizer). The binding energy of the ligands was evaluated using the CB-DOCK server (http://clab.labshare.cn/cb-dock/php/blinddock.php) which works using Auto Dock-Vina version 1-1-2²⁸. The LeDock program was further adapted to perform the re-docking for the above compounds, which was found to give a good performance on the ligand-protein systems. To ensure the robustness of our docking predictions, we compared the binding poses of the top-scoring ligands with their best docking results from AutoDock-Vina by performing a superimposition analysis.

Density functional theory

To gain deeper insights into the electronic properties and stability of the top-ranked compounds identified from MD. We selected the five top compounds based on their binding affinity and stability in MD simulations. The B3LYP gradient-corrected exchange-correlation functional was employed, incorporating Becke’s three-parameter exchange potential and the 6–31 + G(d, p) basis set for accurate geometry optimization. The ligand structures were optimized to their lowest energy configurations to ensure that the geometry used in subsequent analyses accurately reflects the most stable state of each compound. Electronic properties, such as the Highest Occupied Molecular Orbital (HOMO), Lowest Unoccupied Molecular Orbital (LUMO), and the HOMO-LUMO energy gap, were computed to predict the reactivity of the ligands. A smaller energy gap indicates higher reactivity, which is relevant to their ability to interact with CAD’s active site²⁹.

Molecular simulation analysis

MD simulation is a popular tool for studying the behavior of biomolecules due to the high speed, affordability, and flexibility of this approach³⁰. In this study, the GROMACS package (version 2019.1) and employing the CHARMM27 force field were used to simulate the structural and dynamic changes of the CAD protein with PDB ID 6R6U with the ligands (inhibitor) compounds³¹. The structure was placed in a triclinic unit cell 0.1 nm from the box edge. The system was first solvated using a simple point (SPC) water model before being ionized and neutralized by Na⁺ and Cl⁻ ions with a salt concentration of 0.15 M. Following that, a conjugate gradient algorithm with steepest descent (SD) minimization was used to minimize energy on a model that had already been predicted. In the equilibration step, an NVT (constant number, volume, and temperature) ensemble with a constant temperature of 300 K at 100 ps and an NPT (constant number, pressure, and temperature) equilibrated to the 1 bar pressure at 100 ps. Finally, MD was performed for a period of 60 ns with coupled temperature (300 K) and pressure (1 bar) to analyze the stability of each system. Graphpad prism (version 8) was used to plot graphs and all structural parameters such as root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), solvent accessible surface area (SASA), and the number of hydrogen bonds were recorded for unbound and ligand-bound protein structures.

Free energy calculations

The binding free energies of CAD-ligand complexes were calculated using the gmx_MMPBSA tool by analyzing the final 10 ns of molecular dynamics simulation trajectories³². To model solvation effects, we employed the MM/GBSA method, which combines molecular mechanics (MM) with generalized Born (GB) surface area (SASA) to estimate solvation free energies. The internal dielectric constant was set to 1.0, and the external dielectric constant was set to 80.0, reflecting the solvation environment. A salt concentration of 0.154 M was used to model the physiological ionic strength. For the binding free energy calculation, we included the gas-phase energy (electrostatic + van der Waals) and solvation energy components (polar and nonpolar solvation). The total binding free energy (ΔG_Total) was calculated using the equation:

This equation measures the difference between the free energy of the bound complex and the sum of the free energies of the receptor and ligand in their unbound states. The energy decomposition analysis was performed using the MM/GBSA method based on the final 10 ns of the MD simulation trajectories. The active site residues of CAD were defined based on crystallographic data (PDB ID: 6R6U), which indicated that key residues such as Ala151, Phe158, Pro160, Val163, Phe245, and Tyr246 were involved in ligand binding. These residues were selected as the region of interest for the energy decomposition analysis. Contributions of these residues to the binding free energy were calculated, focusing on their hydrophobic, electrostatic, and hydrogen bonding interactions with the ligand.

We also computed changes in SASA upon binding using:

Furthermore, we divided the overall binding free energy into enthalpic and entropic components:

Here, the enthalpy change (ΔH) combines the gas-phase energy (ΔG_GAS) and the solvation free energy (ΔG_SOLV). The gas-phase energy includes both bonded and non-bonded interactions, while the solvation energy is split into polar and non-polar contributions.

Evaluation of toxicity, pharmacokinetics, and drug-likeness prediction of ligands

Toxicity prediction

Toxic prediction of small target molecules is critical to predict the tolerance of those molecules before consumption in human models. To assess the toxicity risk, all predicted inhibitor compounds were analyzed using the pkCSM database (http://structure.bioc.cam.ac.uk/pkcsm). This website can detail the toxicological effects in the areas of AMES toxicity, maximum human tolerance dose, The human ether-a-go-go-related gene (hERG) 1 inhibitor, inhibition of hERG-II, hepatotoxicity, dermatology, T. pyriformis³³.

The molecular and genetic toxicity of selected inhibitory compounds was determined using Lazar Toxicity Prediction software (https://lazar.in-silico.ch/predict).

ADME predictions

The absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of the potential inhibitors were predicted in silico using the SWISS-ADME website (https://www.swissadme.ch) that allows determining the parameters such as lipophilicity (Log P), water solubility (Log S), the laws of drug similarity (Lipinski) and the pharmaceutical³⁴.

Drug-likeness prediction

The degree of drug similarity of predicted inhibitory compounds was determined using the Molsoft server (https://www.molsoft.com/).

A review of lipinski’s rules

The selected ligands were examined using Server (http://www.scfbio-iitd.res.in/software/drugdesign/lipinski. jsp) to predict the suitability of selected inhibitors as potential therapeutic agents based on Lipinski’s rules²⁶.

Results and discussion

The computer-aided drug discovery has become a crucial tool in the drug discovery and designing process. The successful identification of compounds that bind to and inhibit the normal function of enzymes involved in the inflammatory process has been a prominent strategy in discovering direct-acting anti-inflammatory drugs. The IRG1 gene which encodes the CAD enzyme plays a pivotal role in regulating inflammation in several inflammatory diseases such as sepsis, psoriasis, glioma, peritoneal tumors, abdominal aortic aneurysm, hepatic ischemia-reperfusion injury, and pulmonary fibrotic conditions. To date, no computational analysis has been applied to identify the CAD inhibitors. In this study, we employed a molecular docking strategy to identify novel molecules capable of binding to CAD, potentially inhibiting the function of this enzyme.

Analyses of IRG1 gene interactions using the STRING server (Fig. 1) revealed that this gene is involved in the regulation of several inflammatory pathways. These include the pathways of pro-inflammatory cytokines such as interleukin 1 beta (IL-1β), Chemokine (C-C motif) ligand 4 (CCL4), and C-X-C Motif Chemokine Ligand 10 (CXCL10), which are implicated in several inflammation-associated diseases^17–20. Figure 2 displays the crystallographic structure of the CAD enzyme and its main ligand that the active site of CAD can probably be Ala151, Phe158, Pro160, Val163, Phe245, and Tyr246.

Fig. 1.

Predicted protein network between human IRG1 (also called CAD) and other proteins by STRING server. Each node represents a protein and each line represents an interaction. Proteins were clustered based on the STRING Evidence Score. IRG1: mmune-responsive gene 1; CAD: Cis-aconitate decarboxylase; IL-1β: Interleukin-1 beta; CXCL10: C-X-C motif chemokine ligand 10; CCL4: Chemokine (C-C motif) ligands 4; IRF1: Interferon regulatory factor 1. The red-colored circle indicates the proteins that participate in inflammatory pathways.

Fig. 2.

Three-dimensional structure of CAD enzyme with PDB code: 6R6U. The Figure is generated using Discovery Studio software Version 3.5.0.12158 (Accelrys Software Inc, San Diego, CA, USA). CAD: Cis-aconitate decarboxylase.

In this study, an in silico-based virtual screening approach was employed to identify novel compounds targeting the CAD enzyme. Subsequently, their physicochemical potency, toxicity, and pharmacokinetic properties were investigated. Upon assessing the stability of the CAD enzyme’s sequence structure and creating a related stability diagram using the IUPred web server, it was shown that the amino acids scores of the CAD enzyme’s active site, which plays a role in binding with the predicted ligands, were under the threshold of IUPred diagram, especially at the junction of ligands with the enzyme (Fig. 3).

Fig. 3.

Prediction of protein stability and disorder using the IUPred web server for CAD enzyme (PDB ID: 6R6U) for docking analysis. CAD: Cis-aconitate decarboxylase.

Molecular docking helps identify ligand binding sites on proteins and the interactive roles of protein amino acids and ligand atoms³⁵. For this purpose, the PDB file of the main ligand of CAD enzyme with access code 6R6U was used as a template, and after structural analysis, four new ligands were selected, whose two-dimensional and three-dimensional structures are shown in Fig. 4. Notably, all predicted ligands exhibited a strong binding affinity to the enzyme active site. Analysis of ligand-enzyme interactions demonstrated that the main ligand can establish a hydrogen bond with Phe158. Data on molecular binding also indicated that, aside from binding with Phe158 at the active site, the predicted ligands could form hydrogen bonds with other amino acids including, Arg157, Glu150, Asp153, Met154, and Phe245 without compromising the enzyme’s stability. Consistent with these contacts, visual inspection of bound vs. unbound overlays shows side-chain reorientation of pocket residues (Arg157/Phe158/Tyr246) toward the ligand, forming a tighter packing of the binding cleft while preserving the global fold. Based on our docking data, binding energies of ligand 1, ligand 2, ligand 3, and ligand 4 were − 6.8, −5.8, −6.4, and − 6.3 kcal/mol (measured by AutoDock Vina), and − 98.18, −96.24, −93.01, and − 91.44 kcal/mol (measured by MVD software), respectively (Table 1). The lower the binding energy level (more negative), indicates the stronger interactions between the enzyme and the inhibitor. Our results showed that ligand No. 1 with the C7 H12 N2 O4 chemical formula exhibited the lowest binding energy indicating the strongest bond at the active site compared to the binding energy of the main ligand (Table 1). While both AutoDock Vina and MVD were used to predict the binding affinity of the selected ligands, we observed some discrepancies in the binding energy values between the two tools. For instance, ligand 1 showed a binding energy of −6.8 kcal/mol in AutoDock Vina and − 98.18 kcal/mol in MVD, with similar discrepancies observed for other ligands (e.g., ligand 2: −5.8 kcal/mol in Vina vs. −96.245 kcal/mol in MVD, and ligand 3: −6.4 kcal/mol in Vina vs. −93.010 kcal/mol in MVD). Despite these numerical differences, the overall trend remained consistent across both software tools, with ligand 1 consistently showing the strongest binding affinity in both cases. These discrepancies can be attributed to the different scoring functions employed by the tools: AutoDock Vina uses a grid-based docking approach with a flexible scoring function, while MVD utilizes MolDock with a separate empirical scoring function. Such differences often result in slightly different binding energy calculations, but they do not significantly alter the predicted interaction patterns or the ranking of the ligands. Interestingly, both software tools predicted similar interaction patterns for the ligands, with key residues such as Phe158, Arg157, Asp153, and Phe245 consistently identified as important for ligand binding. The formation of hydrogen bonds and hydrophobic interactions in these regions was similarly observed in both docking protocols. This suggests that despite differences in the calculated binding energies, both tools accurately predicted the ligand–protein interaction modes.

Fig. 4.

Two-dimensional (left images) and three-dimensional (right images) views of the interaction between CAD enzyme (PDB ID: 6R6U) and ligands using Discovery Studio software. (A) main ligand (C4 H10 O2); (B) Ligand 1 (C7 H12 N2 O4); (C) Ligand 2 (C6 H12 O6); (D) Ligand 3 (C6 H10 O6); (E) Ligand 4 (C6 H12 O6).

Table 1.

In Silico Docking results of the ligands and CAD enzyme using Molegro virtual docker (MVD) and auto dock Vinna software.

Ligand name	Formula	Molecular weight	Log p	Log s	binding energy (MVD(	binding energy (Auto dock vinna)	Ligand structure
Main ligand	C4 H10 O2	90.07	−0.45	0.62	−39.1006	−4.3
1	C7 H12 N2 O4	188.08	−1.73	0.27 (in Log (moles/L)	−98.18	−6.8
2	C6 H12 O6	180.06	−3.14	0.04 (in Log (moles/L)	−96.245	−5.8
3	C6 H10 O6	178.05	−2.36	0.03 (in Log (moles/L)	−93.010	−6.4
4	C6 H12 O6	180.06	−2.25	0.14 (in Log (moles/L)	−91.449	−6.3

To further validate the docking results, we refined the binding modes of the compounds using the LeDock program. The superimposition of the best docking poses, which were selected based on the highest docking scores, showed low RMSD values (< 2.5 Å) for the heavy atoms of the compounds. These low RMSD values indicate that the docking results are reliable and suitable for subsequent post-docking analysis. Additionally, the LeDock program was utilized for re-docking the compounds, and it demonstrated good performance in accurately reproducing the ligand-protein interactions, further supporting the validity of the computational predictions. Using the Molsoft server, the drug-likeness of the predicted ligands was evaluated. The drug-likeness scores for ligand 1, ligand 2, ligand 3, and ligand 4 were 0.29, −1.26, −0.35, and − 0.97, respectively (Fig. 5). Our results indicated that ligand 1 exhibited a positive drug-likeness score, indicating its potential as a drug-like compound, while ligands 2, 3, and 4 had negative scores, but still fell within the range typically associated with drug-like compounds. Upon evaluation the structural features of ligand 1 and comparing them to the others, we observed that ligand 1 had a relatively higher molecular weight (188.08 g/mol) compared to the other ligands, yet it still meets the Lipinski criteria for drug-likeness. Specifically, ligand 1 possessed a logP value of −1.73, which was in the acceptable range for Lipinski’s rule (logP < 5), and it had high water solubility, which was a favorable characteristic for drug-like compounds. Additionally, the total polar surface area (TPSA) of ligand 1 was 63.697 Ų, which, although slightly lower than some of the others, still falls within the range associated with favorable absorption and bioavailability. On the other hand, ligands 2, 3, and 4 had lower TPSA values, lower molecular weights, and negative logP values, indicating that they might not meet all the ideal pharmacokinetic requirements to be classified as drug-like based on Lipinski’s rules. These structural features can contribute to the negative scores for these ligands, potentially affecting their ability to permeate biological membranes or bind to targets effectively..

Fig. 5.

Plotting of Drug-likeness score of ligands using MolSoft server. Non-drug like behavior (green-colored curve), drug-like behavior (blue-colored curve), and the predicted ligands (red-colored line). (A) Ligand 1; (B) Ligand 2; (C) Ligand 3; (D) Ligand 4.

Poor pharmacokinetic properties are the main reason for the failure of many ligands in drug applications³⁵. Thus, after substantiating the optimal interaction of predicted ligands with the target protein, we assessed the most important pharmacokinetic parameters of selected ligand, including absorption, distribution, metabolism, excretion, and toxicity (ADMET)^36–38. Compounds’ hydrophobicity, derived from the logarithm of their n-octanol/water partition coefficient (log P), should be less than 5, and compounds with log S values over − 10 indicate good solubility^39,40. In this study, the Log P values of ligand 1, ligand 2, ligand 3, and ligand 4 were − 0.45, −1.73, −3.14, −2.36, and − 2.35, respectively. In addition, the Log S values of ligand 1, ligand 2, ligand 3, and ligand 4 were 0.62, 0.27. 0.04, 0.03, and 0.14, which were matched to Lopinski’s rules (Table 1). It has been shown that the total polar surface area (TPSA) has a direct role in the permeability of the compounds. According to studies, permeability augments with increasing mass and decreasing TPSA, and compounds with TPSA less than 140 angstroms have good permeability⁴¹, which TPSA values of ligand 1, ligand 2, ligand 3, and ligand 4 were 68.69, 68.01, 68.64, and 75.15 that they indicate good permeability (Table 2). In addition, xenobiotic compounds entering cells are generally metabolized by cytochrome P450 enzymes (CYP450), which determine the drug concentrations within cells. The analysis showed that the four chosen chemical structures are either non-substrates or non-inhibitors of CYP1A2, CYP2C19, CYP2C9, CYP2D6, and CYP3A4, indicating that these compounds could be metabolized slowly and can stay within cells long enough to inhibit CAD activity (Table 2). Furthermore, the molecular and genetic toxicity of inhibitory ligands were evaluated using the Lazar Toxicity Predictions software. The analysis concluded that none of the compounds exhibited hepatic, renal, cardiovascular, or neurological toxicity (Table 2).

Table 2.

The physicochemical properties of ligands according to lipinski’s rules.

Ligand name	Total polar surface area (TPSA)	Drug likeness	Pharmacokinetic	Water solubility	Toxicity
Main ligand	75.153	LIPINSKI = YES	-High absorption of the gastrointestinal tract -Without BBB absorption -Lack of inhibition of cytochromes	SOLUBLE	No toxicity and inhibition in: AMES toxicity hERGI inhibitor hERGII inhibitor Hepatotoxicity Skin Sensitisation Cardiovascular toxicity Nephrotoxicity Neurological toxicity
1	68.697	LIPINSKI = YES	-High absorption of the gastrointestinal tract -Without BBB absorption -Lack of inhibition of cytochromes	SOLUBLE	No toxicity and inhibition in: AMES toxicity hERGI inhibitor hERGII inhibitor Hepatotoxicity Skin Sensitisation Cardiovascular toxicity Nephrotoxicity Neurological toxicity
2	68.010	LIPINSKI = YES	-Low absorption of the gastrointestinal tract -Without BBB absorption -Lack of inhibition of cytochromes	SOLUBLE	No toxicity and inhibition in: AMES toxicit hERGI inhibitor hERGII inhibitor Hepatotoxicity Skin Sensitisation Cardiovascular toxicity Nephrotoxicity Neurological toxicity
3	68.643	LIPINSKI = YES	-Low absorption of the gastrointestinal tract -Without BBB absorption -Lack of inhibition of cytochromes	SOLUBLE	No toxicity and inhibition in: AMES toxicit hERGI inhibitor hERGII inhibitor Hepatotoxicity Skin Sensitisation Cardiovascular toxicity Nephrotoxicity Neurological toxicity
4	75.153	LIPINSKI = YES	-Low absorption of the gastrointestinal tract -Without BBB absorption -Lack of inhibition of cytochromes	SOLUBLE	No toxicity and inhibition in: AMES toxicit hERGI inhibitor hERGII inhibitor Hepatotoxicity Skin Sensitisation Cardiovascular toxicity Nephrotoxicity Neurological toxicity

The drugs’ uptake predictability and similarity were evaluated using molecular weight (M.W), lipophilicity, and the number of hydrogen bond donors (HBD) and acceptors (HBA), guided by Lipinski’s rules³³. The parameters indicated that the predicted compounds offer drug-like properties with molecular weights under 500 Dalton, less than 5 HBD, and less than 10 HBA, alongside fewer than three rotatable bonds (RB)³⁸. Based on our findings, it seems that ligand 1 (C7 H12 N2 O4), with four HBD bonds, four HBA bonds, two RB bonds (Table 1), good permeability and solubility (Table 3) had a better physicochemical and pharmacokinetic properties as inhibitor of CAD enzyme compared to other ligands.

Table 3.

The number of rotatable bond (RB) and hydrogen bond donor (HBD) and acceptor (HBA) between ligands and CAD enzyme.

Ligand name	Formula	Ligand structure	Hydrogen bond	Number of RB, HBD and HBA
Main ligand	C4 H10 O2		Phe158	RB: 1 HBD: 2 HBA: 2
1	C7 H12 N2 O4		Phe158, Arg157, Glu150, Asp153, Met154, Phe245	RB: 2 HBD: 4 HBA: 4
2	C6 H12 O6		Phe158, Arg157, Phe245	RB: 5 HBD: 5 HBA: 6
3	C6 H10 O6		Phe158, Arg157, Asp153, Lys156, Phe245	RB: 2 HBD: 4 HBA: 6
4	C6 H12 O6		Phe158, Arg157, Glu150, Lys156	RB: 2 HBD: 5 HBA: 6

Molecular dynamics is used to analyze the binding efficiency of the ligands with the CAD protein at an atomistic level. Based on docking results, we conducted a molecular dynamics simulation to obtain more information about the structural dynamics and interactions of CAD enzyme with selected ligands. Several parameters such as RMSD, RMSF, Rg, SASA, and intermolecular hydrogen bonds help understand the binding pattern of ligands to the active site⁴². RMSD is a parameter that provides detailed structural information that indicates the protein’s conformational stability⁴³. RMSD was calculated based on the atomic distances between the simulated structure and a reference structure (typically the initial or crystal structure) over the simulation time. A sudden increase in RMSD indicates instability or large conformational changes (e.g., protein denaturation), low RMSD Values (< 0.2 nm) indicates high structural stability. The average RMSD of ligand 1, ligand 2, ligand 3, ligand 4, and unbound protein was 0.1308, 0.1512, 0.1328, 0.1417, and 0.1747 nm, respectively (Fig. 6A), which they were less than 2 Å. This low RMSD values suggest that the docking results are in good agreement, further supporting the reliability and accuracy of our predictions^44,45. Superposition of apo vs. bound average structures confirmed that global deviations remained small (sub-2 Å), whereas differences were concentrated locally within the binding pocket. RMSF is an important parameter to analyze the fluctuation of protein atom positions, reflecting the impact of ligand binding on the targeted protein (residue) flexibility over time. Unlike RMSD which shows global deviation, RMSF reveals local flexibility of different molecular regions Residues with lower RMSF values are more rigid or stable^46,47. In this study, the obtained values of RMSF for ligand 1, ligand 2, ligand 3, ligand 4, and unbound protein were 0.0927, 0.1013, 0.0971, 0.0981, and 0.1003, respectively (Fig. 6B). Results indicated that ligand 1 could decrease the flexibility of the protein residues and increase the stability of the protein structure. Importantly, relative to apo CAD (0.1003 nm), ligand 1 reduced pocket RMSF by ~ 7–8%, indicating ligand-stabilized local dynamics; in contrast, ligand 2 showed a modest RMSF increase consistent with its higher Rg. Rg provides data on the overall compactness of the enzyme and folding of standard secondary structure into the tertiary or functional structure of the targeted protein before and after binding of the ligand. Our results indicated no significant changes between unbound protein and selected ligand-bound protein complexes, demonstrating the compact and stabilized folding within the ligand-bound complexes. However, the Rg value increased after binding of ligand 2 to the CAD protein, indicating that ligand 2 could decrease the stability of the protein (Fig. 6C). This pattern agrees with the bound–unbound overlays, where ligand 2 fails to induce the compact side-chain packing observed for ligand (1) Calculation of SASA value of ligand-bound and unbound protein gives important information about the potential ligand binding. Based on our data, the SASA value of the ligand 1-CAD protein complex decreased in comparison to unbound and other ligands, indicating the stabilized protein structure (Fig. 6D) in the presence of ligand1. Consistently, ligand–pocket buried surface area increased for ligand 1 versus apo, reflecting deeper burial and improved shape complementarity in the bound state. Hydrogen bonds are weak electrostatic interactions between a hydrogen atom (donor) and an electronegative atom (acceptor like oxygen and nitrogen) and they play a fundamental role in stabilizing biological structures like proteins, DNA, and molecular complexes. In MD studies, analyzing hydrogen bond dynamics provides valuable insights into molecular stability and function. The complex is stable if the number of hydrogen bonds remains relatively constant over time^46,47. In the present study, the increased hydrogen formation potential of the ligand 1, contributed more stability to the protein structure than the other ligand complexes (Fig. 6E). Time-resolved H-bond analysis showed persistent (> 30% occupancy) contacts for ligand 1 within the crystallographic pocket, predominantly involving Arg157/Glu150/Asp153, with hydrophobic packing against Phe158/Phe245; such contacts were intermittent for ligand (2) Together, the bound–unbound comparisons (reduced pocket RMSF, decreased SASA/increased BSA, and persistent H-bonding) support a mechanism in which subtle side-chain reorientation of Arg157/Phe158/Tyr246 tightens the pocket around ligand 1 while preserving the global fold. Collectively, Ligand 1 consistently reduced RMSF of pocket residues relative to apo CAD, indicating ligand-stabilized local dynamics, whereas ligand 2 showed a modest increase consistent with its higher Rg. We also quantified hydrogen-bond dynamics as time-occupancy (%) over the 60-ns trajectory: ligand 1 formed persistent H-bonds (> 30% occupancy) within the crystallographic pocket, supporting sustained engagement rather than transient contacts. SASA/BSA behavior (decreased protein SASA and increased ligand–pocket BSA for ligand 1) further indicated greater burial/compactness upon binding. Taken together, the combination of sub-2 Å RMSD, pocket-level RMSF damping, and high-occupancy H-bonds supports a stable, native-like binding mode for ligand 1.

Fig. 6.

Molecular dynamics (MD) simulation trajectory plot of cis-aconitate decarboxylase (CAD) protein in unbound and inhibitor ligand compound bound complex using GROMACS. (A) Root mean square deviations (RMSD), (B) Root mean square fluctuation (RMSF), (C) the radius of gyration (Rg), (D) solvent accessible surface area (SASA), and (E) number of intermolecular hydrogen bonds correspond to 60 ns MD simulation. Black color refers to unbound CAD protein; red color indicates ligand 1-CAD complex; green color indicates ligand 2-CAD complex; blue color indicates ligand 3-CAD complex; orange color indicates ligand 4-CAD complex.

Following the MD simulations, the final frames of the ligand-receptor complexes were selected as the input file for the DFT calculation. The electronic molecular properties, including electron density, frontier molecular orbitals (such as LUMO and HOMO), and the molecular electrostatic potential map, were analyzed to illustrate the biological activity and molecular characteristics. The HOMO and LUMO levels play a critical role in charge transfer during chemical interactions. Molecules with higher HOMO levels are typically excellent nucleophiles, while those with lower LUMO energies are generally strong electrophiles. Therefore, the values of HOMO and LUMO, along with their energy gap (ΔE = E_HOMO - E_LUMO), are indicative of the molecule’s biological activity^48,49. Our results indicated that, the energy levels of HOMOs of ligands were between − 0.27027 eV to − 0.2917 eV, however, the LUMOs were between − 0.02062 and − 0.05529 eV (Fig. 7). Furthermore, the energy gaps of ligand 1, ligand 2, ligand 3 and ligand 4 were 0.25046, 0.2518, 0.2487 and 0.21498, respectively (Fig. 7). In the present study, the HOMO–LUMO gap was used as a qualitative descriptor of electronic polarizability/responsiveness rather than a direct predictor of binding efficiency. In biological contexts, binding is primarily governed by shape complementarity, persistent non-covalent networks, desolvation energetics, and pocket dynamics. In our experiment, ligand 4 shows the smallest gap (0.21498 eV) yet does not exhibit the most stabilizing pocket behavior compared with ligand 1, which has an intermediate gap (~ 0.25 eV) but demonstrates more persistent H-bond occupancy (> 30%), increased ligand–pocket BSA, and reduced pocket-level RMSF. We know that the flip side of smaller gaps can sometimes correlate with chemical/photo-instability or faster metabolic turnover. To address this, in-silico ADMET showed no major CYP/toxicity alerts, and 60-ns MD indicated stable complexes; nonetheless, we propose follow-up experiments (microsomal stability, GSH-trapping for soft-spot reactivity, and redox/photostability assays) to distinguish beneficial polarization for binding from potential metabolic liability⁴⁹. In addition, the Hartree energies of ligand 1, ligand 2, ligand 3 and ligand 4 were − 684.469, −686.23, −686.22, and − 686.19, respectively (Table 4). All the designed compounds exhibited negative free energies, indicating greater stability. Dipole moment is also an important parameter for predicting the solubility in polar solvents. With the help of DFT calculations, the dipole moment can be calculated and this would be an important parameter for predicting the solubility in polar solvents. If the dipole moment of one molecule is high then it is considered to be polar, means it could be more water-soluble^49,50. The dipole moments of ligand 1, ligand 2, ligand 3, and ligand 4 were 7.498, 1.33, 6.173, and 6.351, respectively, suggesting that these ligands may be suitable candidates for oral delivery (Table 4).

Fig. 7.

HOMO, LUMO and energy gap (E gap) of cis-aconitate decarboxylase (CAD) inhibitor compounds. (A) Ligand 1 (C7 H12 N2 O4); (B) Ligand 2 (C6 H12 O6); (C) Ligand 3 (C6 H10 O6); (D) Ligand 4 (C6 H12 O6). HOMO: highest occupied molecular orbital; LUMO: lowest unoccupied molecular orbital.

Table 4.

The amount of Hartree energy and dipole moment of the selected ligands.

Ligand name	Hartree energy (eV)	Dipole moment (Debye)
No. 1	−684.469	7.498
No. 2	−686.23	1.33
No. 3	−686.22	6.173
No. 4	−686.19	6.351

It is widely recognized that docking algorithms have inherent limitations in accurately capturing the complex details of biomolecular interactions. To address this challenge and gain a comprehensive understanding of the interactions between the target protein and the ligand MM/GBSA were employed. This approach estimates detailed information from MD simulation trajectories to calculate the binding free energy of the simulated complex and highlights key residues within the protein that significantly contribute to these binding interactions^51,52. The negative total binding free energies (ΔTOTAL) of ligand 1 (− 22 ± 2.63 kcal/mol), ligand 2 (− 25.99 ± 2.79 kcal/mol), ligand 3 (− 30.48 ± 2.40 kcal/mol), and ligand 4 (− 30.31 ± 3.02 kcal/mol) indicate favorable binding affinities (Fig. 8), supporting the stability of the ligand-receptor complexes. Per-residue energy decomposition further supports the bound–unbound structural picture: residues that reorient upon binding (Phe158, Arg157, Pro160/Tyr246) contribute most favorably to ΔG_bind for ligand 1, mirroring the persistent contacts seen in MD. Thus, across independent analyses (structural overlays, dynamics, burial, and energetics), the bound state is consistently stabilized for ligand 1 compared with apo CAD. In addition, as demonstrated in Fig. 8, key residues with significant energetic contributions at the simulation time include Phe 158, Arg157, Pro155, Tyr246, and Pro160 (for ligand 1); Phe 158, Pro160, Val163, and Tyr246 (for ligand 2); Phe158, Pro155, Asp153, and Glu150 (for ligand 3); Phe158, Arg157, Lys156, and Pro155 (for ligand4) were the core functional amino acids with the highest binding energy. These core functional amino acids exhibit the highest binding energies and play crucial roles in the interaction between the CAD enzyme and the ligands (Fig. 8).

Fig. 8.

The binding free energies (ΔG) between the residues interacting in holo-structure calculated by the molecular mechanics-generalized born surface area (MM/GBSA). (A) Ligand 1 (C7 H12 N2 O4); (B) Ligand 2 (C6 H12 O6); (C) Ligand 3 (C6 H10 O6); (D) Ligand 4 (C6 H12 O6).

The selection of different tools was driven by the specific capabilities of each program, which complement one another to provide a comprehensive evaluation of the identified compounds. These tools, each utilizing distinct algorithms, allowed for accurate molecular docking, validation of binding affinities, drug-likeness prediction, and advanced visualization of protein-ligand interactions. Importantly, different compounds may exhibit superior characteristics in different programs. By using several tools, we were able to observe that ligand 1 consistently showed superior characteristics in comparison to the other compounds, further validating its potential as a CAD inhibitor.

In recent years, in-silico techniques have become an essential tool in drug discovery, enabling the identification of potential enzyme inhibitors. However, as of now, CAD has not been targeted by any computational studies, and there are no known in vitro CAD inhibitors. This gap in the literature highlights the novelty of our study, which is the first to explore CAD as a potential target for inhibitor development using computational methods.

Although molecular docking and dynamics simulations are well-established techniques, the novelty of this study lies in their application to the CAD enzyme, a target not previously investigated in this context. The identification of four novel compounds that could potentially inhibit CAD represents a significant advancement in enzyme inhibition research, expanding the pool of CAD inhibitors for further experimental validation. One major limitation of our study is that, currently, no known inhibitors for CAD exist in the literature, which makes it impossible to directly compare the efficacy of our identified compounds with existing molecules. This gap in the field highlights the novelty of our findings. However, the lack of existing CAD inhibitors presents a limitation in assessing the superiority of our compounds. Moreover, while molecular docking and dynamics simulations provide valuable insights into potential inhibitor binding, they are inherently limited by the accuracy of the computational models and the absence of experimental data for further validation. As such, the predictions made in this study should be considered as hypotheses that need to be validated through experimental approaches. These results must be interpreted cautiously, and further experimental research is needed to assess the true biological activity and therapeutic potential of the identified compounds.

Conclusion

Through molecular docking, molecular dynamics simulations, DFT, and MM-GBSA free energy calculations, we identified four inhibitors with strong binding affinity and stability against CAD, a key enzyme involved in itaconate biosynthesis and linked to inflammatory responses. These ligands complied with Lipinski’s rules and demonstrated favorable pharmacokinetic properties, including high bioavailability and no indications of cardiovascular, hepatic, or neurotoxicity. Their ability to effectively inhibit CAD enzyme activity suggests potential therapeutic applications in modulating inflammatory diseases. However, as these findings are based on computational models, further in vitro and in vivo studies are essential to validate their efficacy. Additionally, preclinical and clinical investigations will be crucial in assessing their potential as therapeutic agents targeting CAD-related inflammatory pathways (Fig. 9).

Fig. 9.

Graphical representation of our in-silico approach for the identification of novel Cis-aconitate decarboxylase inhibitors as potential anti-inflammatory drug using molecular docking and dynamics.

Author contributions

MRT, MDK and SP conceived and designed the study, performed the computational analyses, and drafted the manuscript. VZ and ZM contributed to data analysis, methodology refinement, and result interpretation. VZ assisted with molecular dynamics simulations, DFT and MM-GBSA free energy calculations. MRT supervised the research, reviewed the manuscript, and provided critical revisions. All authors contributed to the discussion, reviewed the final manuscript, and approved its submission.

Data availability

The datasets generated during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.