To perform the in silico research, a high-end Dell workstation (Precision 7865) equipped with a dual NVIDIA RTX A2000 graphics card was used. Molecular docking was performed using Maestro V14.5 (Schrödinger Suite). Cytoscape V3.10.4 was used to construct the network.

PPINs help to reveal, how proteins cooperate in biological systems and identify key interaction points that serve as potential drug targets [19, 20]. In this study, a PPIN associated with the IMPDH gene (guaB) of C. jejuni was constructed to explore its functional interaction partners. Initially, draft PPINs were constructed using the STRING v12.0 [21] server with a high confidence score of 0.7 (i.e., interaction confidence > 70%). The final network was designed and validated using Cytoscape V3.10.4 [22]. Network features were calculated using the Network Analyzer module in Cytoscape.

The phytochemicals included in this study were selected through a comprehensive literature survey of plant-derived compounds with reported antimicrobial and pharmacological activities. The selection was based on the following criteria: (i) evidence of antimicrobial, anti-inflammatory, or antioxidant properties reported in previous studies; (ii) structural diversity to represent different classes of natural products; (iii) availability of chemical structure information for computational analysis; and (iv) limited or no prior reports investigating their inhibitory activity against C. jejuni inosine monophosphate dehydrogenase.

Based on these criteria, six structurally diverse phytochemicals—afzelechin, amarogentin, amaroswerin, eriocitrin, naringenin, and peganine—were selected for further investigation (Figure 1 and Table 2). These compounds span multiple phytochemical classes, including flavonoids, secoiridoid glycosides, and alkaloids, and have been reported to exhibit promising biological activities. However, their potential as inhibitors of C. jejuni IMPDH has not been systematically explored. Therefore, these compounds were selected to evaluate their binding affinity, interaction stability, and drug-like properties against IMPDH using an integrated computational approach.

The crystal structure of the catalytic domain of inosine monophosphate dehydrogenase (IMPDH) from C. jejuni in complex with the inhibitor p200 was obtained from the Protein Data Bank (PDB) [28]. The deposition code for this protein is 5UQG. It contains 406 amino acid residues and is a key enzyme involved in guanine nucleotide biosynthesis, which is essential for bacterial growth, virulence, and pathogenicity [28]. The protein retrieved from the database was prepared using the Protein Preparation Wizard in Maestro V14.5 (Schrödinger, LLC). During preparation, the water molecules and heteroatoms were removed from the structure. Following optimization, minimization was continued until the average root mean square deviation (RMSD) of the non-hydrogen atoms reached 0.3 Å. This procedure alleviated steric clashes and achieved a stable protein conformation suitable for subsequent computational analyses [29].

The protein retrieved from the databank has a co-crystallized ligand (p200). Hence, the co-crystallized ligand site was selected as the active binding pocket for docking the phytochemicals [30]. Subsequently, a grid was constructed around the co-crystallized ligand to stabilize the protein binding site [31].

The physical complexity of the plant metabolites was reduced using the LigPrep module of Maestro V14.5. This module employs two phases: tautomerism and stereoisomerism. The ligand molecules were energy-minimized using the Optimized Potentials for Liquid Simulations (OPLS) 5 force field to ensure optimal shape and stability. To preserve the specified chirality of the ligand molecules, a maximum of 32 stereoisomers were generated for each ligand molecule. According to Lipinski’s rule of five, ligands containing more than 500 Da were excluded from the ligand preparation. Furthermore, during the preparation process, the ionization states were consistently maintained at pH 7.0 ± 2.0 using Epik to ensure accurate molecular configuration [32].

To assess the antimicrobial potential of the selected phytochemicals, the prepared ligands and grid-generated protein models were subjected to molecular docking using the ligand docking module in Maestro V14.5. The binding affinities of the ligands and protein molecules were measured using various parameters, including Glide docking score, Glide energy, Glide eModel, hydrogen bond contacts, pi-pi stacking, and pi-cation. The docked complex was explored using the Maestro interface to measure hydrogen-bond contacts and their distances. The interaction plot module was used for each phytochemical complex to screen the types of contacts that existed at the IMPDH binding site [29].

To validate the molecular docking protocol, the co-crystallized ligand was extracted from the crystal structure of C. jejuni inosine monophosphate dehydrogenase, and all water molecules and other heteroatoms were removed. The native ligand was then redocked into the original binding pocket using the same docking parameters as for all phytochemical ligands. In addition, the interactions of the redocked ligand with the active-site residues were compared with those of the crystal ligand to confirm conservation of the binding pattern.

To further assess the binding stability of the amaroswerin-IMPDH complex, MD simulations were performed using the Desmond module (Schrödinger). It was run for 200 ns. The molecular simulation system was built by placing the protein-ligand complex inside an orthorhombic periodic boundary box solvated with Transferable Intermolecular Potential with 3 Points (TIP3P) water molecules. This widely used explicit solvent model accurately captures hydrogen bonding and solvent dynamics in biomolecular simulations. A minimum buffer distance of 10 Å was maintained between the protein surface and the box boundary to prevent artificial interactions across periodic images. The system was neutralized by adding suitable counterions and supplemented with 0.15 M NaCl to mimic the physiological ionic strength. The OPLS_2005 force field was applied to model the molecular interactions of both protein and ligand atoms, offering reliable parameters for biomolecular simulations and it has been extensively validated in protein-ligand interaction studies. Before the production run, the system was energy-minimized using the steepest-descent algorithm to eliminate unfavorable steric contacts and relax the structures. Equilibration and production simulations were performed in the NPT ensemble [constant number of particles, pressure, and temperature (ensemble)]. The Nosé-Hoover chain thermostat maintained the system temperature at 300 K, and the Martyna-Tobias-Klein barostat maintained a pressure of 1 atm, ensuring accurate thermodynamic sampling [33]. Covalent bonds involving hydrogen atoms were constrained using the SHAKE algorithm, which enabled a 2-fs integration time step to improve computational efficiency while preserving stability.

The pharmacokinetic profiles of the hit phytochemicals were evaluated using two online databases, pkCSM and ProTox-v3.0. The SMILES representations of the phytochemicals were used to evaluate their potential for ADMET in the body using the pkCSM server. Furthermore, organ toxicity, including hepatotoxicity, neurotoxicity, nephrotoxicity, respiratory toxicity, and cardiotoxicity, as well as toxicity endpoints such as carcinogenicity, immunotoxicity, mutagenicity, cytotoxicity, blood-brain barrier (BBB) permeability, ecotoxicity, and clinical toxicity, were assessed using the ProTox-v3.0 server.

PPINs revealed that the target receptor guaB interacts with several key proteins, including purA, purF, purL, purQ, guaA, pheT, proS, ribB, rpoB, rpoC, and htpG. The connectivity among these proteins highlights the central role of guaB as a hub in metabolic and regulatory functions in purine biosynthetic metabolism (Figure 2a).

The predicted network features and functional descriptions of the proteins are presented in Tables 3 and 4. Gene Ontology (GO) enrichment analysis was performed for both biological processes and metabolic pathways using the constructed PPINs.

Proteins were predominantly involved in purine-containing compound biosynthetic processes, nucleoside monophosphate biosynthesis, and ribonucleotide biosynthesis, as shown in Figure 2b. The enrichment based on Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway showed that most proteins are involved in ligase activity (1.4e-04) and magnesium ion binding (6.9e-04), as depicted in Figure S1. GO enrichment analysis of cellular compartments revealed that the network proteins were primarily associated with the cytoplasm (15 proteins) and intracellular compartments (17 proteins), as shown in Figure S2. The KEGG pathway enrichment analysis showed that the network proteins were primarily involved in purine metabolism, followed by riboflavin metabolism and RNA polymerase-related pathways, as depicted in Figure 2c. These results suggest that guaB and its partners are crucial for maintaining nucleotide synthesis, transcriptional activity, and cofactor metabolism (Tables 3 and 4).

Amaroswerin had a docking score of –8.612 kcal/mol (Table 5), with six interactions comprising five hydrogen bond contacts and one pi-cation contact (Table 6). In this case, ASN 273 (1.69 Å), GLU 411 (1.61 Å), LEU 408 (2.17 Å), MET 384 (2.20 Å), and LYS 71 (2.41 Å) were the key residues that formed hydrogen-bond contacts with amaroswerin. Notably, LYS 71 had two contacts: a π-cation interaction and a hydrogen bond. The two-dimensional interaction plot shows that, except for LEU 408, all hydrogen-bond contacts were established in the hydroxyl region of amaroswerin (Figure 3a and 3b).

Eriocitrin ranked second among metabolites for inhibiting IMPDH. This complex had six hydrogen bond contacts and a docking score of –8.488 kcal/mol (Table 5). In the binding site, GLY 296, GLU 411, VAL 409, LEU 408, and GLY 385 were key residues that established hydrogen bond contacts with eriocitrin at distances of 2.25, 2.19, 1.94, 1.82, and 2.03 Å, respectively (Figure 3c and 3d). Notably, the hydroxyl regions of eriocitrin were identified as key sites that formed hydrogen-bond contacts with residues in the target-binding site (Table 6).

Afzelechin had a docking score of –7.739 kcal/mol with three interactions with the target protein (Table 5). In this case, GLU 411, ALA 47, and ASN 273 were the key residues that interacted with afzelechin. The distances of the hydrogen bond contacts for GLU 411, ALA 47, and ASN 273 were 1.61, 1.93, and 2.69 Å, respectively, indicating that afzelechin has a good binding affinity for IMPDH (Table 6). The interaction plot indicated that the hydroxyl moieties of afzelechin were the primary sites of interaction with IMPDH (Figure 3e and 3f).

Amarogentin had a docking score of –7.635 kcal/mol (Table 5). It has a similar interaction profile to amaroswerin and eriocitrin, indicating strong binding affinity for IMPDH. The complex comprises five hydrogen-bond contacts and a π-cation contact. The active-site residues GLY 294, GLU 411, LEU 408, MET 384, and LYS 71 formed hydrogen-bond contacts at distances of 2.77 Å, 1.85 Å, 2.01 Å, 1.94 Å, and 2.39 Å, respectively (Figure 3g and 3h). Among these contacts, except for LEU 408, all residues formed hydrogen bonds with the hydroxyl moieties of amarogentin (Table 6).

Peganine had a docking score of –7.999 kcal/mol (Table 5). This complex has three interactions: GLU 411, THR 303, and GLY 296 were the active-site residues that formed hydrogen-bonding contacts with peganine (Figure 4a). In this case, GLU 411 established a hydrogen bond with the ammonium moiety of peganine, whereas the other two residues, THR 303 and GLY 296, established a hydrogen bond with the hydroxyl moiety of peganine. Naringenin had a docking score of –7.619 kcal/mol (Table 5). Despite this molecule having a lower docking score than peganine, it formed better interactions with IMPDH, including four hydrogen-bond contacts and a pi-cation contact (Table 6). The active site residues GLU 411, GLY 294, ALA 47, and LYS 71 interacted with naringenin (Figure 4b).

To assess the accuracy of the docking protocol, the co-crystallized ligand was redocked into its crystallized site (Figure 5). The redocked ligand occupied the native binding pocket and formed notable interactions with the active-site residues of IMPDH. These findings validate the docking procedure and increase confidence in the predicted binding poses of the selected phytochemicals (Figure 6).

To assess the dynamic behavior and binding stability of the amaroswerin–IMPDH complex, MD simulations were performed for 200 ns. During the initial phase of the trajectory (0–85 ns), the Cα RMSD remained below approximately 2.8 Å. The RMSD gradually increased to approximately 4.5 Å at 180 ns, then stabilized toward the end of the 200 ns simulation. This pattern suggests that the complex underwent conformational adjustment during the simulation while retaining its overall structural association. The most fluctuating residual areas in this protein were ASN 1, ALA 0, MET 1, LYS 392, LEU 408, SER 307, LYS 2, GLY 308, GLN 391, VAL 309, and VAL 409, which fluctuated with RMSF ranges of 9.259, 7.802, 6.321, 5.663, 5.414, 5.145, 4.999, 4.862, 4.753, 4.681, and 4.642 Å, respectively. ASP 334 had the highest interaction fraction of 2.03 Å, involving a water-bridge-assisted hydrogen bond. Notably, ASP 334 (94% and 80%), GLU 411 (72%), ALA 47 (54%), ASP 49 (53%), ASP 244 (52% and 32%), LYS 21 (35%), and LYS 292 (32%) were the principal residues that facilitated the stability of the amaroswerin-IMPDH complex through robust hydrogen bond interactions (Figure 7). The binding stability was maintained by an average of 12 contacts over 200 ns.

A 150 ns MD simulation was performed to evaluate the dynamic stability of the amaroswerin–IMPDH complex beyond molecular docking. The stability of the complex structure was evaluated by RMSD, root mean square fluctuation (RMSF), residue interaction, and protein-ligand contact profiling (Figure 8). RMSD of the protein backbone initially rose during the first 20–30 ns of the simulation to allow for structural relaxation from the docked state, and then tended to stabilize around 3.0–4.2 Å over the rest of the simulation (Figure 8a). Both RMSD and RMSF values of the ligand were relatively stable after initial equilibration, ranging from 4.5–5.2 Å and 11.0–11.1 Å, respectively, suggesting that amaroswerin was bound in the same orientation in the active site, with only minor conformational changes. No large deviations were observed during the production phase, indicating that the protein-ligand complex equilibrated and remained structurally stable throughout the simulation. The fluctuations in amino acid residues, as measured by RMSF, showed that most residues were within 1.5 Å of their mean positions, indicating the overall rigidity of the protein structure (Figure 8b). The elevation of fluctuations was mainly found in the terminal regions and a few loops at residues 90, 180–190, 275–285, and the C-terminal region, which are typical of flexible solvent-exposed loops. Importantly, the residues that form the ligand-binding pocket showed relatively small fluctuations, which would help to preserve the integrity of the active site when the ligands are bound.

Amaroswerin formed stable interactions with several active-site residues during the simulation (Figure 8c), as confirmed by the protein-ligand interaction fraction analysis. Of these, GLU 411 had the highest interaction occupancy, being in contact for almost the entire simulation. Other residues, ASP 334, MET 384, THR 303, LYS 292, and ASP 244, also contributed to ligand stabilization through hydrogen bonding, hydrophobic interactions, and water-mediated interactions. The interaction profile indicates that multiple residues contribute to stabilizing the ligand within the catalytic pocket. The two-dimensional interaction map also showed that amaroswerin has a large hydrogen-bonding network with important catalytic residues (Figure 8d). It was determined that the residues GLU 411, MET 384, and LYS 292 form direct hydrogen bonds, and additional water-supported hydrogen bonds were detected between ASP 334 and ASP 244. Hydrophobic contacts with residues are further strengthened ligand binding. These complementary interactions helped maintain the ligand in an optimal conformation throughout the simulation. The protein-ligand contact timeline showed that Glu411 had the highest number of contacts that persisted throughout most of the simulation (Figure 8e), along with Met384, Thr303, Lys292, and Asp244. Transient interactions with a few residues were observed sporadically, but the overall stability of these core interactions suggests that amaroswerin remained tightly bound to the IMPDH active site.

To ensure the stability of the native ligand in the active site of C. jejuni inosine-5’-monophosphate dehydrogenase (IMPDH), a 100 ns MD simulation was performed. The structural stability and interaction profile of the co-crystal complex were assessed using RMSD, RMSF, interaction fraction, and protein-ligand contact analysis (Figure 9). The initial 10–20 ns showed a large increase in the protein backbone RMSD, corresponding to structural equilibration, and the RMSD then remained relatively constant between 2.5 and 3.3 Å throughout the rest of the simulation (Figure 9a). The ligand RMSD fluctuated moderately during the equilibration phase and converged to a steady value of 2.0–3.0 Å, suggesting a stable orientation of the co-crystal ligand within the catalytic pocket throughout the simulation. There is no significant conformational drift, indicating that the protein-ligand complex has equilibrated and remained stable during the 100 ns simulation.

In the RMSF profile, most protein residues showed fluctuations less than 1.5 Å, suggesting limited flexibility in the protein structure (Figure 9b). The general trend of higher fluctuations was observed in the flexible loops around residues 85–90, 190–200, and 275–285, as well as in the C-terminal residues, which are not close to the catalytic core. Residues lining the ligand-binding pocket, on the other hand, showed relatively small variations, indicating that ligand binding helped preserve the structural stability of the active site. To investigate the dynamics of the interaction between the co-crystal ligand and the active site residues, data from the interaction fraction analysis were used to simulate the interaction (Figure 9c). High interaction occupancies are observed for PRO 297, SER 299, ASP 334, GLY 336, and ILE 337, with the latter showing the highest occupancy. The interactions were mainly hydrogen bonding, water-mediated bridging, and hydrophobic interactions, collectively supporting the ligand within the binding cavity.

The two-dimensional representation of the interaction map also revealed numerous hydrogen bonds between the ligand and surrounding residues (Figure 9d). Stable hydrogen bonds were established with ASP 334, SER 299, SER 358, GLY 357, MET 384, and GLY 385 residues, and hydrophobic interactions with ILE 300, ILE 337, TYR 381, and PRO 297 also contributed to ligand binding. Several other hydrogen-bond interactions were also mediated by water molecules, underscoring the importance of solvent bridges in maintaining the protein-ligand complex. The protein-ligand contact timeline (Figure 9e) shows that most contacts were maintained for a large fraction of the 100 ns simulation. Contacts were maintained throughout the trajectory, and other residues were in intermittent contact, contributing to the overall stability of the complex. The same interaction pattern throughout the simulation suggests that the co-crystal ligand was well bound in the active site.

The ADMET profiles of the hit metabolites showed that afzelechin had the highest intestinal absorption (91.48%), followed by amarogentin (50.44%) and amaroswerin (49.294%). In terms of Caco-2 permeability, amaroswerin (–0.592) and eriocitrin (–1.101) had the lowest values, indicating low permeability potential. Interestingly, all the hit metabolites were found to act as substrates for P-glycoprotein, indicating possible transporter-mediated efflux. The probability distribution rates of these phytochemicals ranged from –1.406 to 1.529, with eriocitrin showing the highest distribution potential. In terms of BBB permeability, all phytochemicals showed a negative rate, indicating poor penetration across the BBB. Likewise, the central nervous system (CNS) permeability values for these chemicals are low, indicating a lower risk of neurotoxicity. The metabolic potential of these chemicals revealed that none were substrates or inhibitors of major cytochrome P450 (CYP450) enzymes, including CYP1A2, CYP2C19, CYP2C9, CYP2D6, and CYP3A4. The absence of significant CYP inhibition indicates that the phytochemicals are less likely to cause adverse drug-drug interactions and are metabolically compatible. In terms of excretion, amaroswerin and amarogentin have relatively high clearance rates. Furthermore, none of these metabolites were found to be substrates for renal OCT2 (organic cation transporter 2), indicating lower renal transport-associated toxicity. The phytochemicals were found to be non-hepatotoxic and non-sensitizing to the skin, whereas the oral chronic rat toxicity values ranged from 2.215 to 3.33, indicating an acceptable toxicity range. Additionally, the toxicity values for Tetrahymena pyriformis and the fathead minnow indicate that these substances are relatively safe for the environment and living organisms. Together, the ADMET profiles indicated that the molecules have favorable pharmacokinetic and acceptable safety profiles, supporting their potential as promising candidates for C. jejuni targeting of IMPDH (Table 7).

An online toxicity prediction platform was used to evaluate the potential organ toxicity and toxicity endpoints of the selected phytochemicals. Amaroswerin, eriocitrin, afzelechin, and amarogentin showed higher predicted probabilities of inactivity for hepatotoxicity and neurotoxicity, suggesting a lower computational likelihood of liver and neuronal toxicity. However, eriocitrin, afzelechin, and amarogentin exhibited relatively higher predicted probabilities for nephrotoxicity and respiratory toxicity (0.77, 0.75, and 0.80, respectively), indicating that these potential toxicities should be considered in future experimental investigations. Similarly, the higher predicted inactive probabilities for carcinogenicity (0.75, 0.91, 0.70, and 0.77, respectively), together with the low predicted probabilities for mutagenicity, cytotoxicity, blood–brain barrier penetration, ecotoxicity, and clinical toxicity, suggest a favorable preliminary computational safety profile (Table 8). Nevertheless, these predictions are generated using in silico models and should not be interpreted as definitive evidence of safety. Therefore, comprehensive in vitro and in vivo toxicological studies are required to validate these computational findings before considering further drug development.