Clinical characteristics, sample distribution, and identification of Pseudomonas aeruginosa
A total of 613 clinical specimens were collected from patients with suspected bacterial infections, including sputum, urine, wound swabs, and burn samples, representing a broad range of infection sites. Following primary culturing and initial phenotypic screening, 44 isolates (7.2%) were identified as P. aeruginosa (Table 1). The presumptive identification of P. aeruginosa was based on characteristic colony morphology, pigmentation patterns, and Gram-staining features consistent with Gram-negative, non-fermentative bacilli. Species-level confirmation was subsequently achieved using the VITEK2 ND automated identification system, which showed complete concordance with the phenotypic findings. These clinically confirmed isolates were selected for subsequent antimicrobial susceptibility testing, molecular characterization of gyrase-associated genes, and downstream in silico analyses.

Table 1. Distribution of Clinical Specimens and Isolation Rates of P. aeruginosa

Antimicrobial susceptibility profiles of clinical Pseudomonas aeruginosa isolates
Antimicrobial susceptibility testing of the 44 clinical P. aeruginosa isolates revealed a high level of resistance across multiple antibiotic classes. Notably, elevated resistance was observed against β-lactam/β-lactamase inhibitor combinations, cephalosporins, carbapenems, aminoglycosides, fluoroquinolones, and folate pathway inhibitors. A detailed summary of resistance rates for each tested antimicrobial agent is presented in Table 2.

Table 2. Resistance Rates of Clinical P. aeruginosa Isolates (n = 44)

Overall, the resistance profiles demonstrated non-susceptibility to agents from six distinct antimicrobial categories, fulfilling the established criteria for MDR. Furthermore, the broad resistance distribution indicates that a substantial proportion of the isolates exhibited patterns consistent with an extensively drug-resistant phenotype.

Molecular detection and sequence confirmation of target genes
Molecular screening using PCR revealed the presence of MexB and GyrB genes in all 44 clinical P. aeruginosa isolates, confirming their universal distribution among the examined strains. The consistent detection of these target genes across all isolates underscores their essential role in bacterial viability and validates their selection for further molecular and in silico analyses. Representative PCR amplification profiles of the GyrB and MexB genes are shown in Figure 1.

For sequence-level characterization, a representative subset of 10 PCR-positive isolates was selected for sequence analysis, including six isolates for the GyrB gene and four isolates for MexB. DNA sequencing was performed by Macrogen (South Korea). The obtained sequences were analyzed using the BLAST algorithm available through NCBI to confirm gene identity and sequence similarity before performing multiple sequence alignment, mutation assessment, and protein structure modeling. The nucleotide sequence of the GyrB gene was deposited in the DDBJ under accession number LC919849, and the corresponding protein sequence was assigned GenBank accession number BHR60112.1.

Genetic variability, sequence analysis, and phylogenetic analysis of the GyrB gene
Sequence analysis of the GyrB gene revealed a high degree of similarity to P. aeruginosa reference sequences available in the NCBI database. BLAST analysis demonstrated sequence identities ranging from 95%-99%, indicating that the GyrB gene is highly conserved among the examined clinical isolates.

Multiple sequence alignment further confirmed the conserved nature of the analyzed coding region, with no significant insertions or deletions detected. Although minor nucleotide variations were observed, these changes did not affect the conserved functional motifs of DNA GyrB. The translated protein sequence (GenBank accession No. BHR60112.1), derived from the deposited GyrB nucleotide sequence (DDBJ accession No. LC919849), retained the conserved ATP-binding and catalytic domains characteristic of GyrB, supporting its suitability as a stable molecular target for inhibitor binding.

Phylogenetic analysis of the GyrB gene demonstrated that the representative clinical isolates clustered closely with reference P. aeruginosa strains retrieved from the NCBI database, indicating a high level of genetic conservation and confirming the molecular identity of the studied isolates (Figure 2). These findings further support the use of the GyrB gene as a reliable molecular marker and a suitable target for subsequent computational analyses.

The phylogenetic tree based on GyrB gene nucleotide sequences shows the clustering of representative clinical isolates (highlighted) alongside reference P. aeruginosa strains retrieved from the NCBI database. The phylogenetic tree was constructed using the Fast Minimum Evolution method. The scale bar (0.0002) represents the evolutionary distance, expressed as the number of nucleotide substitutions per site.

Protein preparation, structural modeling, and validation of target proteins
The three-dimensional structures of the target proteins, GyrB and MexB, were prepared for molecular docking analysis. The GyrB nucleotide sequence obtained from P. aeruginosa isolates was translated into its corresponding amino acid sequence based on the predicted ORF. The translated protein sequence (GenBank accession No. BHR60112.1), derived from the deposited GyrB nucleotide sequence (DDBJ accession No. LC919849), was subsequently used for structural modeling. Three-dimensional structural models of GyrB and MexB were generated using the SWISS-MODEL server, based on the best available AlphaFold reference models. The GyrB model was generated using the AlphaFold reference model Q917C2.1.A, which has 100% sequence identity, a GMQE score of 0.88, and a predicted monomeric oligomeric state. Structural validation using MolProbity demonstrated excellent model quality, with a MolProbity score of 0.58, a clash score of 0.00, and 97.64% of residues located in favored regions of the Ramachandran plot.

The MexB model was generated using the AlphaFold reference model A0A5K1SFB4.1.A, which has 100% sequence identity, a GMQE score of 0.91, and a QMEAN score of 0.89 ± 0.05, and is predicted to be monomeric. The model exhibits 60% sequence similarity across the entire protein length (residues 1-721), indicating a robust and complete structural model suitable for docking analysis. Additionally, the model is predicted to contain a transmembrane segment, consistent with the membrane-associated nature of MexB.

Protein preparation involved adding hydrogen atoms and performing structural optimization through energy minimization to ensure model stability. Active binding sites were identified based on conserved functional domains, particularly the ATP-binding pocket of GyrB and the substrate-binding catalytic site of MexB. The modeled three-dimensional structures of GyrB and MexB are presented in Figure 3A and B.

Selection, structural classification, and ADMET profiling of candidate ligands
A panel of reference inhibitors and candidate ligands was selected for molecular docking analysis against GyrB and MexB. The selection strategy included established reference inhibitors as well as structurally related candidate ligands selected based on previously reported antibacterial activity and their potential to interact with conserved functional regions of the respective target proteins. The selected ligands were classified according to their target specificity and chemical class. The chemical structures, PubChem CID numbers, molecular formulas, MWs, and SMILES notations of the selected ligands are summarized in Table 3. Novobiocin and PAbN were used as reference inhibitors for GyrB and MexB, respectively, while a coumarin derivative and an indole derivative were evaluated as candidate ligands (Table 4).

Table 3. Chemical Characteristics of Reference Inhibitors and Candidate Ligands

Table 4. Reference Inhibitors and Candidate Ligands Selected for Molecular Docking and ADMET Analysis

Prior to docking, ligand structures were retrieved from the PubChem database, prepared using DSV, and subjected to geometry optimization. Drug-likeness and pharmacokinetic properties were subsequently evaluated through in silico ADMET analysis, which included assessments of physicochemical characteristics, Lipinski’s rule of five parameters, GI absorption, bioavailability, and toxicity prediction. This integrated approach enabled the prioritization of candidate ligands based on both their predicted binding affinity and favorable pharmacokinetic profiles.

Molecular docking validation using reference inhibitors
To validate the molecular docking protocol, established reference inhibitors were used for each target protein. Novobiocin and PAbN were selected as reference ligands for GyrB and MexB, respectively. Both reference inhibitors exhibited favorable binding affinities and stable interactions within the functional binding sites of their respective proteins. Novobiocin bound to the ATP-binding pocket of GyrB, whereas PAβN interacted with the substrate-binding cavity of MexB. These findings confirm the suitability of the docking protocol for the subsequent evaluation of candidate ligands.

Molecular docking results for selected ligands
Molecular docking analysis was performed to evaluate the binding affinity of selected reference inhibitors and candidate ligands against the GyrB and MexB proteins. Multiple binding cavities were automatically predicted for each target protein using the CurPocket platform. The cavity with the lowest predicted binding energy was selected as the optimal binding site for each ligand, and the corresponding cavity parameters are summarized in Table 5.

Table 5. Molecular Docking Results and Predicted Binding Cavity Parameters for Selected Ligands Targeting GyrB and MexB

For GyrB, the reference inhibitor novobiocin exhibited the strongest binding affinity, with a binding energy of -10.2 kcal/mol in the highest-ranked binding pocket (C1). The coumarin derivative also demonstrated favorable binding (-8.2 kcal/mol) within the same pocket, indicating its potential to interact with the ATP-binding site of GyrB. For MexB, the reference inhibitor PAβN showed a binding energy of -8.1 kcal/mol in the top-ranked binding pocket (C4). The indole derivative exhibited a slightly lower binding affinity (-7.7 kcal/mol) but maintained favorable interactions within the predicted binding cavity (C1), suggesting its potential as an efflux pump inhibitor.

Overall, the docking results indicate that GyrB exhibited stronger ligand-binding interactions than MexB. Although the candidate ligands displayed slightly lower binding affinities than their respective reference inhibitors, they showed comparable potential as promising antibacterial lead compounds.

Binding interaction analysis of GyrB and MexB complexes
The protein-ligand interaction analysis was conducted to further evaluate the binding modes and stability of the selected ligands within the active sites of the GyrB and MexB proteins.

For GyrB, the reference inhibitor novobiocin demonstrated a well-established interaction network within the ATP-binding pocket. It formed a conventional hydrogen bond with Asn267, a π-cation interaction with Arg482, and π-π stacking with Phe460. Additional hydrophobic interactions were observed with residues such as Ala387, along with polar interactions involving Arg384 and Arg391, supported by multiple van der Waals contacts and charged interactions. The coumarin-based candidate ligand exhibited a similar binding orientation within the same pocket, forming a hydrogen bond with Asn265,
π-π stacking with Phe460, and a π-cation interaction with Arg391. These interactions indicate that the candidate ligand adopts a binding orientation comparable to that of the reference inhibitor, although it has fewer predicted interactions and a lower predicted binding affinity.

For MexB, the reference inhibitor PAβN formed a robust interaction network within the substrate-binding cavity. It established multiple hydrogen bonds with Glu97, His184, and Gln40, along with a π-cation interaction with Arg46 and amide-π stacking with Glu97. The indole-based candidate ligand exhibited a predicted interaction profile that included hydrogen bonds with Ser352, Asn351, and Arg358, as well as a π-anion interaction with Glu393 and several hydrophobic contacts involving Ala356, Val502, and Met391. However, the predicted interaction network was less extensive than that observed for PAβN.

Overall, the interaction analysis suggested that both candidate ligands could occupy the predicted functional binding sites of their respective targets and establish favorable protein-ligand interactions (Table 6 and Figure 4). Although the candidate ligands displayed slightly lower predicted binding affinities compared to the corresponding reference inhibitors, their binding orientations shared several key interactions with the reference compounds. These docking results provide computational evidence supporting the potential for further investigation; however, they should not be interpreted as direct evidence of inhibitory activity and require experimental validation.

Table 6. Key Interaction Residues in Protein-Ligand Complexes

ADMET and toxicity analysis of selected candidate ligands
The pharmacokinetic and drug-likeness properties of the selected candidate ligands were evaluated using in silico ADMET prediction tools. Key parameters analyzed included MW, TPSA, consensus LogP, Lipinski’s rule of five compliance, GI absorption, bioavailability score, PAINS alerts, Brenk structural alerts, and predicted toxicity endpoints. The results indicated that both the coumarin and indole derivatives exhibited favorable predicted drug-like characteristics (Table 7). Each compound complied fully with Lipinski’s rule of five and was predicted to have high GI absorption. Additionally, both ligands displayed moderate TPSA values (TPSA<60 Ų), consensus LogP values of 2.55 and 2.90, respectively, and a predicted bioavailability score of 0.55, suggesting favorable physicochemical properties for oral drug candidates.

Table 7. Predicted In Silico ADMET and Toxicity Properties of Selected Candidate Coumarin and Indole Derivative Ligands

MTD Human: Maximum tolerated dose in humans (log mg/kg/day); LD₅₀ Rat: Oral acute toxicity in rats (mol/kg); LOAEL Rat: Oral rat chronic toxicity (log mg/kg bw/day)

Neither compound exhibited PAINS alerts; however, the coumarin derivative showed one Brenk structural alert, while the indole derivative showed none. Toxicity predictions indicated that the indole derivative was negative for AMES mutagenicity, hepatotoxicity, hERG I/II inhibition, and skin sensitization. In contrast, the coumarin derivative showed a positive AMES prediction despite its otherwise favorable physicochemical and pharmacokinetic properties, indicating a potential mutagenic liability that warrants careful consideration and further experimental validation.

Overall, the in silico ADMET analysis provides a preliminary computational assessment of the pharmacokinetic and toxicity profiles of the selected ligands. These computational predictions can help prioritize candidate compounds for further investigation; however, they should not be considered definitive evidence of safety or biological efficacy and must be validated through experimental studies.

Molecular dynamics simulation of the MexB protein
MD simulations were performed for 25 ns to evaluate the structural stability of the MexB protein. The RMSD profile showed an initial increase during the early stage of the simulation, followed by fluctuations within approximately 0.4-0.5 nm for the remainder of the simulation. This pattern indicates that the protein underwent conformational adjustments before reaching a relatively stable dynamic state (Figure 5A).

The RMSF analysis demonstrated generally low residue fluctuations across most of the protein, with pronounced peaks observed only in a limited number of residues corresponding to flexible loop regions (Figure 5B). These localized fluctuations are consistent with the intrinsic flexibility commonly observed in protein structures. The Rg remained relatively stable throughout the simulation, fluctuating between 3.70 and 3.80 nm, suggesting that the overall compactness of the protein was maintained during the simulation period (Figure 5C).

The number of intramolecular hydrogen bonds exhibited moderate fluctuations but remained within a relatively narrow range throughout the simulation, indicating the preservation of the internal hydrogen-bonding network and the structural integrity of the protein (Figure 5D). The SASA initially decreased, followed by minor fluctuations, and eventually stabilized toward the end of the simulation. This behavior suggests only limited changes in solvent exposure during the simulation period (Figure 5E).

Overall, the MD simulation indicated that the modeled MexB protein maintained its structural integrity throughout the 25 ns simulation period. These findings support the protein’s structural stability under the simulated conditions. However, these results should be interpreted with caution due to the limited simulation duration. Longer simulations, combined with protein-ligand complex MD analyses, are required to further evaluate ligand-induced stability.