Global burden of antimicrobial resistance

Antimicrobial resistance (AMR) is one of the most urgent public health challenges of the 21^st century, severely compromising the effectiveness of conventional antibiotics and complicating the management of infectious diseases [1,2]. In 2019, nearly 5 million deaths were associated with bacterial AMR, and projections suggest that this number may increase to 10 million annually by 2050 if current trends persist [1]. The rapid emergence and global dissemination of multidrug-resistant (MDR) pathogens, including methicillin-resistant Staphylococcus aureus (MRSA) and carbapenem-resistant Enterobacteriaceae, have rendered many frontline and last-resort antibiotics increasingly ineffective [2].

Biofilm-associated infections and the need for new antimicrobials

Microbial biofilm formation exacerbates the AMR crisis. Biofilms are structured microbial communities embedded in a self-produced extracellular matrix that confers enhanced tolerance to antimicrobial agents and host immune responses [3,4]. Biofilm-associated infections are notoriously persistent and frequently lead to chronic infections, therapeutic failure, and relapse [5]. Although natural compounds have emerged as promising candidates for antibacterial and antibiofilm interventions, substantial gaps remain in understanding their molecular mechanisms, target specificity, and translational potential [6–9].

Rationale for investigating Thymol (TM)

TM, a naturally occurring monoterpenoid phenol derived from Thymus species (thyme plants), has been reported to exhibit antibacterial, antifungal, and antibiofilm activities (Fig 1) [10,11]. Experimental studies have suggested that TM can disrupt microbial membranes, interfere with quorum-sensing pathways, reduce virulence factor expression, and inhibit biofilm formation [12,13]. Despite this growing body of evidence, several critical knowledge gaps persist. Specifically, TM has not been systematically evaluated across multiple essential microbial protein targets using long-timescale MD simulations; its activity has rarely been contextualized using mechanistic comparator compounds, and structurally related analogs with potentially improved drug-like properties remain underexplored. Addressing these gaps is essential for establishing a mechanistic and structural basis for TM’s antimicrobial activity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Chemical structures of the tested compounds: (A) TM, (B) SA, and (C) SLS.

https://doi.org/10.1371/journal.pone.0345977.g001

Rationale for using mechanistic comparator compounds

To strengthen the mechanistic interpretation, sodium azide (SA) and sodium lauryl sulfate (SLS) were included in this study as mechanistic reference compounds rather than therapeutic candidates. SA is a well-characterized inhibitor of cytochrome oxidase that rapidly suppresses microbial respiration (Fig 1) [14,15], whereas SLS is an anionic surfactant that disrupts cellular membranes and induces lysis (Fig 1) [16,17]. These distinct, single-mechanism modes of action provide clear interpretive benchmarks against which TM’s behavior can be compared. A conventional antibiotic was intentionally excluded, as the objective of this study was to distinguish mechanistic similarities and differences rather than to perform potency ranking against clinically optimized drugs.

Structural analogs in computational exploration

Although SA and SLS are unsuitable for therapeutic use owing to toxicity and irritation concerns, their structural analogs may offer more drug-like scaffolds. Accordingly, phenyl azide (PA), an aromatic analog of SA, and lauryl sulfate (LS), a neutral analog of SLS, were incorporated into molecular docking and MD simulations. This approach enabled the exploration of whether simplified or modified structures exhibit enhanced binding stability, favorable physicochemical properties, or improved target interactions suitable for future optimization.

Study hypotheses and objectives

Based on this rationale, we hypothesized that TM would demonstrate superior antimicrobial and antibiofilm activities relative to mechanistic comparators and form stable, biologically meaningful interactions with key microbial proteins. We further hypothesized that PA and LS may display improved interaction profiles, highlighting their potential as scaffolds for next-generation antimicrobial development. To test these hypotheses, this study employed an integrated workflow combining antimicrobial susceptibility testing, biofilm inhibition assays, in silico ADMET profiling, molecular docking, 300 ns MD simulations, principal component analysis (PCA), free-energy landscape (FEL) mapping, and Molecular Mechanics Generalized Born Surface Area (MM/GBSA) calculations to achieve comprehensive phenotypic and mechanistic characterization.

Test organisms

A total of 19 pathogenic microorganisms were used in this study, comprising 17 bacterial and two fungal strains. The bacterial panel included Staphylococcus aureus (S. aureus) ATCC 29213, S. aureus (clinical isolate; S. aureus-CI), methicillin-resistant S. aureus (MRSA-1 and MRSA-2), Staphylococcus saprophyticus (S. saprophyticus) ATCC 43867, Staphylococcus epidermidis (S. epidermidis) ATCC 12228, Streptococcus pyogenes (S. pyogenes; Group A) ATCC 19615, Streptococcus pneumoniae (S. pneumoniae) ATCC 49619, Enterococcus faecalis (E. faecalis) ATCC 29212, Bacillus cereus (B. cereus) ATCC 10876, Escherichia coli (E. coli) ATCC 25922, Klebsiella pneumoniae (K. pneumoniae) ATCC 27736, Pseudomonas aeruginosa (P. aeruginosa) ATCC 9027, Salmonella typhimurium (S. typhimurium) ATCC 13311, Shigella flexneri (S. flexneri) ATCC 12022, Proteus vulgaris (P. vulgaris) ATCC 6380, and Proteus mirabilis (P. mirabilis) ATCC 29906. The fungal strains used were Candida albicans (C. albicans) ATCC 10231 and Aspergillus niger (A. niger) ATCC 6275.

All ATCC reference strains were procured from Microbiologics® (St. Cloud, MN, USA). The non-ATCC microbial strains were obtained from the microbiology laboratory at a multi-speciality hospital, from Unaizah in Saudi Arabia. They were provided as archived, de-identified culture stocks that were previously collected for routine diagnostic purposes. No patient information or identifiers were shared, and no new samples were collected for this study. Therefore, no human subjects were involved, and ethical approval was not required under the institution’s guidelines.

Chemicals and reagents

The test compounds TM (CAS No. 89-83-8), SA (CAS No. 26628-22-8), and SLS (CAS No. 151-21-3) were procured from a certified supplier (Loba Chemie Pvt. Ltd., Mumbai, India). The specified purities of these compounds were as follows: SA, 99.5% w/w; TM, 99.0–101.0% w/w; and SLS, 91.0–92.0% w/w. This specification allows for slight analytical variances, particularly with TM, where the purity can exceed 100% owing to calibration deviations or methodological variations in testing. Unless otherwise specified, the remaining chemicals were procured from registered vendors (Sigma-Aldrich, USA; Oxoid Ltd., UK; Loba Chemie Pvt. Ltd., India).

Antimicrobial profiles of TM, SA, and SLS

Statistical analysis

All antimicrobial and antibiofilm experiments were performed using three independent biological replicates (n = 3). Statistical analysis was applied only to the preliminary antimicrobial activity (zone-of-inhibition data), as these measurements provided continuous numerical values. One-way analysis of variance (ANOVA) was performed using a customized Python script (pandas, matplotlib), followed by Tukey’s Honest Significant Difference (HSD) post hoc test using the Statsmodels library. Data distributions were visualized using box and violin plots. Statistical significance was set at p < 0.05 [18].

Molecular docking simulation

Molecular docking simulations were conducted to elucidate the binding orientations and relative affinities of the selected ligands toward the target proteins. Docking analyses were performed for thymol (TM; PubChem CID: 6989), lauryl sulfate (LS; an analog of SLS; PubChem CID: 8778), and phenyl azide (PA; an analog of SA; PubChem CID: 69319) using the Molecular Operating Environment (MOE) software (version 2022.02), in accordance with a previously established and validated protocol [18].

Protein–ligand interaction analysis

Protein–ligand interactions were analyzed using a customized Python script inspired by PLIP-style interaction detection, implemented with Biopython, NumPy, and pandas [25–28]. Each docked PDB structure was parsed to separate protein, ligand, and crystallographic water atoms, while common inorganic ions and small buffer components were excluded using a predefined ignore list. Ligands were identified either by user-defined residue names or, when not specified, by automatically selecting the largest non-water hetero-residue in the structure. Hydrogen bonds were detected using a distance-based criterion between polar heavy atoms (N/O) within a cutoff of 3.5 Å, with donor and acceptor roles assigned based on residue- and atom-specific rules. Hydrophobic interactions were defined as contacts between ligand carbon or sulfur atoms and carbon or sulfur atoms of hydrophobic protein residues (Ala, Val, Leu, Ile, Met, Phe, Trp, Pro, Cys, and Tyr) within 4.0 Å, excluding Gly by default. Salt bridges were assigned using distance-based charged-atom criteria, with interactions detected between positively charged Arg or Lys side-chain atoms and ligand oxygen atoms, or between negatively charged Asp or Glu side-chain atoms and ligand nitrogen atoms, within a 4.0 Å cutoff. Water-mediated hydrogen bonds were identified when both ligand–water and protein–water polar contacts were ≤ 3.5 Å. All detected interactions were compiled into detailed and summary tables and cross-validated against the corresponding two-dimensional (2D) and 3D interaction maps to ensure consistency with the figures presented in the manuscript.

MD simulation

Post-simulation analysis and binding affinity.

Post-simulation analyses were performed to evaluate complex stability and binding affinity. RMSD was used to assess global structural stability, whereas ligand RMSD monitored retention within the active site. Root-mean-square fluctuation (RMSF) was used to identify flexible or dynamic regions of the protein. Rg was calculated to examine protein compactness throughout the simulation, and hydrogen bond analysis quantified the frequency and persistence of protein–ligand hydrogen bonds, reflecting their contribution to complex stabilization. The solvent-accessible surface area (SASA) was computed for both the protein and ligand to characterize solvation behavior and exposure of the binding interfaces [18]. Binding free energies were estimated using the MM/GBSA method implemented in the gmx_MMPBSA tool. A 100 ns trajectory window (200–300 ns) comprising 200 evenly spaced, backbone-aligned frames was used after the removal of translational and rotational motions. The calculated energy components included van der Waals energy (E_vdW), electrostatic energy (E_ele), polar solvation energy (G_polar), and non-polar solvation energy (G_nonpolar). The total binding free energy (ΔG_binding) was computed as follows:

Determination of dissociation constants (K_d) and half maximal inhibitory concentrations (IC₅₀)

K_d and IC₅₀ values were estimated from the MM/GBSA-derived ΔG values [18,29]. K_d values were derived using the Gibbs free energy equation:

where R = 1.987 × 10 ⁻ ³ kcal/mol·K (universal gas constant) and T = 310 K (physiological temperature). In principle, IC₅₀ can be related to the inhibition constant (K_i) using the following equation:

IC₅₀ values were estimated under the assumption of competitive inhibition, with the approximation K_i ≈ K_d under equilibrium binding conditions, as previously described by Sahakyan et al. [29]. However, owing to the absence of experimental substrate concentration ([S]) and Michaelis–Menten constant (K_m) values, IC₅₀ was inferred directly from the calculated K_d values and therefore represents preliminary estimates of inhibitory potential rather than exact kinetic parameters.

All thermodynamic and binding calculations (ΔG, K_d, and IC₅₀) were performed using in-house Python scripts to ensure accuracy and reproducibility in modeling molecular interactions and estimating inhibitory potential from MM/GBSA analyses [18,29]. Although this approach provides a preliminary estimate of the inhibitory potential, experimental validation is necessary to confirm its biological accuracy.

Principal component analysis (PCA) and free energy landscape (FEL) construction

PCA was conducted to characterize the dominant collective motions of each protein–ligand complex over 300 ns of MD simulations. Trajectories were first corrected for periodic boundary conditions and recentered on the protein using gmx trjconv with the pbc mol and center options, followed by superimposition of all frames onto the Cα atoms of the protein to remove global translational and rotational motions. A covariance matrix of the Cα positional fluctuations was generated and diagonalized using the gmx covar to obtain eigenvalues and eigenvectors, with the first few eigenvectors representing the essential motions of the system. Projections of the trajectories along the first two principal components (PC1 and PC2) were computed using the gmx anaeig, producing projection-versus-time plots and PC1–PC2 conformational scatter maps. FELs were constructed by mapping the conformational population distribution along PC1 and PC2, where the free energy of each state was calculated using the Boltzmann relation:

where P(PC1, PC2) is the probability of sampling a given conformation, k_B is the Boltzmann constant, and T is the simulation temperature (310.15 K). The visualization of the FEL enabled the identification of the low-energy basins and metastable conformational states of the protein in the presence of each ligand [30,31].

ADMET analysis

The absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of TM, SLS, and SA were predicted using ADMETlab 3.0 [18,32] with SMILES strings as the input. Key pharmacokinetic and toxicity parameters, including lipophilicity (LogP), aqueous solubility (LogS), intestinal permeability (Caco-2 cell model), cytochrome P450 (CYP450) enzyme inhibition, and predicted toxicity endpoints, were evaluated. Drug-likeness considerations are discussed separately as supportive physicochemical context and are not treated as a substitute for comprehensive ADMET evaluation. This analysis provides an integrated overview of the predicted pharmacokinetic behavior and safety profiles of the investigated compounds.

Preliminary antimicrobial activity

The preliminary antimicrobial screening revealed distinct activity profiles for TM, SA, and SLS against Gram-positive bacteria, Gram-negative bacteria, and fungi (Fig 2 and S5 Fig). TM exhibited the strongest and broadest activity, particularly against C. albicans (24.8 ± 0.2 mm), A. niger (62.0 ± 0.1 mm), S. epidermidis (26.3 ± 0.3 mm), and S. typhimurium (25.5 ± 0.2 mm). SA also demonstrated notable effects, especially against C. albicans (33.5 ± 0.0 mm), A. niger (46.0 ± 0.0 mm), S. aureus (22.7 ± 0.2 mm), and S. typhimurium (29.9 ± 0.1 mm). SLS produced moderate inhibition, with the largest zones recorded against A. niger (38.0 ± 0.2 mm), S. epidermidis (17.8 ± 0.2 mm), and P. aeruginosa (9.8 ± 0.2 mm), whereas the negative control showed no activity (6.0 ± 0.0 mm). TM’s potent antifungal effects, particularly against A. niger, are consistent with reports of its ability to disrupt fungal membrane integrity and interfere with ergosterol biosynthesis [33]. Its activity against S. epidermidis is consistent with that of Xu et al. [22], who demonstrated TM’s efficacy against biofilm-forming staphylococci. TM’s broad-spectrum activity involves multiple mechanisms, including disruption of cytoplasmic membrane integrity [35], altered membrane permeability [36], quorum-sensing inhibition [37], and interaction with membrane proteins and enzymes essential for cellular function [38]. SA’s activity against C. albicans and S. typhimurium is consistent with earlier findings by Lichstein and Soule [39], with its mechanism attributed primarily to cytochrome oxidase inhibition, impairing bacterial respiration [40]; its potency against Gram-negative bacteria also supports observations by Fager and Alben [41]. SLS exhibited moderate activity, likely because of its capacity to solubilize membrane lipids and denature proteins [42]. However, its comparatively lower performance aligns with reports that higher concentrations are required for effective antibacterial action [33]. Its minimal efficacy against P. aeruginosa reflects the organism’s intrinsic resistance mechanisms, including low outer membrane permeability [43], efflux pump expression [44], and the presence of antibiotic-inactivating enzymes [45]. Collectively, these findings highlight the markedly superior performance of the natural compound TM compared to the synthetic agents SA and SLS, reinforcing the growing interest in essential-oil-derived antimicrobial scaffolds and suggesting that TM’s multimodal mechanisms may reduce the likelihood of resistance development [46].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Preliminary antimicrobial activity of TM, SA, SLS, and NC against the tested pathogens.

Representative zone-of-inhibition images are shown for (A) S. aureus ATCC 29213, (B) S. aureus-CI, (C) MRSA-1, (D) MRSA-2, (E) S. saprophyticus ATCC 43867, (F) S. epidermidis ATCC 12228, (G) S. pyogenes-A ATCC 19615, (H) S. pneumoniae ATCC 49619, (I) E. faecalis ATCC 29212, (J) B. cereus ATCC 10876, (K) E. coli ATCC 25922, (L) K. pneumoniae ATCC 27736, (M) P. aeruginosa ATCC 9027, (N) S. typhimurium ATCC 13311, (O) S. flexneri ATCC 12022, (P) P. vulgaris ATCC 6380, (Q) P. mirabilis ATCC 29906, (R) C. albicans ATCC 10231, and (S) A. niger ATCC 6275. All assays were performed using three independent biological replicates (n = 3).

https://doi.org/10.1371/journal.pone.0345977.g002

MIC, MBC, MBIC, and MBEC

Antimicrobial analyses (S6–S8 Figs) revealed distinct MIC, MBC, MBIC, and MBEC profiles for TM, SA, and SLS against the tested pathogens. TM consistently exhibited low MIC and MBC values (0.10–0.20 mg/mL) against all organisms, demonstrating broad-spectrum bactericidal activity and strong efficacy against MRSA and clinical isolates (S6 Fig). TM also showed potent antibiofilm activity, with MBIC and MBEC values ranging from 0.20–0.39 mg/mL and 0.39–0.78 mg/mL, respectively, supporting its potential for treating biofilm-associated infections and aligning with reports documenting its synergistic effects with antibiotics [12,33,34,47]. SA displayed MIC values comparable to TM (0.10–0.78 mg/mL) but required substantially higher concentrations for bactericidal and antibiofilm effects (MBC, MBIC, and MBEC: 6.25–100 mg/mL), indicating predominantly bacteriostatic activity [39]. SA demonstrated strong antifungal activity (S7 Fig), with low MIC–MBEC values against C. albicans and A. niger, consistent with its known inhibitory effect on cytochrome oxidase [40]. Simultaneously, its reduced activity against Gram-positive bacteria compared to Gram-negative species may reflect differences in membrane permeability [43]. SLS showed variable performance, with low MICs for most Gram-positive bacteria (0.10–0.20 mg/mL) but higher MICs for Gram-negative pathogens, especially P. aeruginosa (0.78–3.12 mg/mL) and P. vulgaris (3.12 mg/mL) (S8 Fig). SLS’s MBC, MBIC, and MBEC values were generally high (50–100 mg/mL), particularly against P. aeruginosa, highlighting the difficulty of eradicating biofilms in highly resistant strains [45,48]. SLS was notably effective against A. niger but showed limited activity against C. albicans, likely due to differences in fungal cell wall architecture [49]. Overall, TM exhibited consistently low MIC, MBC, MBIC, and MBEC values across all pathogens, whereas SA and SLS required substantially higher concentrations, particularly for biofilm eradication. These quantitative differences form the foundation for the comparative computational and ADMET analyses presented in the subsequent sections.

Statistical analysis

The analysis of 57 observations showed that TM had the highest mean antimicrobial activity (22.22 ± 10.62), followed by SA (18.76 ± 11.00), and SLS had the lowest (12.70 ± 7.09). The data were non-normally distributed with unequal variances; however, both ANOVA and Kruskal-Wallis tests confirmed significant differences among the compounds (F(2,168) = 14.00, p < 0.0001). Tukey’s HSD test revealed significant differences between the TM and SLS (p < 0.001) groups and the SA and SLS (p = 0.0031) groups, but not between the TM and SA (p = 0.1425) groups. The Kruskal-Wallis test also confirmed significant differences (H = 37.76, p = 6.32e − 09). Box and violin plots (S9–S10 Figs) illustrate these trends, proving that TM and SA were more potent than SLS under the tested conditions.

Molecular dockings analysis

Molecular docking analysis showed that TM, PA, and LS bind stably within the active sites of their respective microbial protein targets, forming energetically favorable complexes supported by hydrogen bonding, hydrophilic, and hydrophobic interactions, and, in selected cases, salt bridge formation (Table 1). Of the 23 docked poses generated for each compound, the nine most favorable complexes—three per ligand, representing Gram-positive bacteria, Gram-negative bacteria, and fungi—were selected for detailed interaction analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of protein–ligand interactions for the top-scoring docked complexes. Hydrogen-bonding residues, hydrophilic and hydrophobic contacts, salt bridges, and water-mediated hydrogen bonds were identified using a Python-based interaction analysis pipeline inspired by PLIP. π–π stacking interactions are reported where present based on visual inspection of interaction maps. Distances are reported in Å, and ranges indicate multiple contacts involving the same residue. Hydrophobic residues correspond to non-polar ligand–protein contacts (C···C/S interactions), whereas hydrophilic residues represent hydrogen bond donors/acceptors or charged side chains contributing to polar interactions. n denotes the number of individual hydrogen bonds or salt-bridge contacts contributed by a given residue.

https://doi.org/10.1371/journal.pone.0345977.t001

TM displayed the strongest binding affinity toward β-carbonic anhydrase (PDB: 5CXK; ΔG = −11.2 kcal/mol), where binding was stabilized by a hydrogen bond with TRP39 and reinforced by a well-defined hydrophobic pocket formed by ALA49, CYS42, LEU52, PRO48, TRP39, and TYR181. TM interaction with FtsZ (PDB: 4DXD; ΔG = −11.0 kcal/mol) was anchored by a hydrogen bond with GLY205 and further supported by hydrophobic contacts involving LEU200, LEU209, VAL203, and VAL297. In fungal sterol 14-α-demethylase (PDB: 5TZ1; ΔG = −9.0 kcal/mol), TM formed hydrogen bonds with HIS377 and SER378 and was stabilized by extensive hydrophobic interactions with LEU376, MET508, PHE233, PHE380, PRO230, and TYR118.

PA exhibited comparatively weaker yet meaningful binding interactions. In β-carbonic anhydrase (PDB: 5CXK; ΔG = −9.2 kcal/mol), PA formed multiple hydrogen bonds with ALA49, ARG64, TRP39, and VAL47, accompanied by hydrophobic contacts involving CYS42, CYS96, and TYR181. The binding of PA to topoisomerase IV (PDB: 4URN; ΔG = −7.8 kcal/mol) involved a hydrogen bond with SER50 and a salt-bridge interaction with ASP76, with additional hydrophobic stabilization provided by MET81. In sterol 14-α-demethylase (PDB: 5TZ1; ΔG = −8.8 kcal/mol), PA binding was supported by hydrogen bonds with ARG381 and HIS468 and hydrophobic contacts involving CYS470, ILE379, LEU376, and PHE463, without detectable π–π stacking interactions in the selected complex.

LS consistently showed strong predicted affinity for all evaluated targets. In peptide deformylase (PDB: 1LMH; ΔG = −10.7 kcal/mol), LS formed multiple hydrogen bonds with GLN65, HIS154, HIS158, and LEU112, supported by hydrophobic interactions with CYS111, LEU105, LEU112, LEU41, and LEU61. The most favorable binding energy was observed for LS in LpxC (PDB: 2VES; ΔG = −19.9 kcal/mol), driven by a dense network of hydrogen bonds with HIS237, HIS78, and THR190, along with extensive hydrophobic contacts involving ALA206, ALA214, ILE197, LEU18, LEU200, MET62, and PHE191. LS also strongly bound to serine/threonine phosphatase Z1 (PDB: 5JPF; ΔG = −13.4 kcal/mol), forming hydrogen bonds with ARG386, HIS231, HIS290, and HIS413, complemented by hydrophobic interactions with TRP371 and TYR299 and a stabilizing salt-bridge interaction involving ARG386.

Overall, these results reveal clear ligand- and target-specific interaction patterns across bacterial and fungal proteins. The comparative docking trends are illustrated in Fig 3, and Figs 4–6 present detailed two- and three-dimensional interaction maps. All nine selected complexes were subsequently subjected to 300 ns MD simulations to evaluate their structural stability, interaction persistence, and conformational behavior under simulated physiological conditions, providing further insight into their potential as antimicrobial candidates.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Heatmap of molecular docking scores for TM, PA, and LS against all target proteins.

Lower docking scores indicate stronger predicted binding affinity. The heatmap summarizes the comparative binding strengths of each compound across the full protein panel.

https://doi.org/10.1371/journal.pone.0345977.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Molecular interaction analysis of TM with selected microbial target proteins.

Combined 2D and 3D binding interaction representations illustrating the binding mode, key stabilizing interactions, and active-site accommodation of TM within: (A) β-carbonic anhydrase (PDB: 5CXK), (B) cell division protein FtsZ (PDB: 4DXD), and (C) sterol 14-α-demethylase (PDB: 5TZ1). The 2D interaction maps highlight hydrogen bonding, hydrophobic contacts, and aromatic interactions, whereas the corresponding 3D views depict TM within the respective binding pockets, demonstrating stable, target-specific binding conformations.

https://doi.org/10.1371/journal.pone.0345977.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Molecular interaction analysis of PA with selected microbial target proteins.

Combined 2D and 3D binding interaction representations illustrating the binding mode, key stabilizing interactions, and active-site accommodation of PA within: (A) β-carbonic anhydrase (PDB: 5CXK), (B) DNA topoisomerase IV, B subunit (PDB: 4URN), and (C) sterol 14-α-demethylase (PDB: 5TZ1). The 2D interaction maps highlight hydrogen bonding, hydrophobic contacts, and aromatic interactions, whereas the corresponding 3D views show PA positioned within the respective binding pockets, demonstrating stable, target-specific binding conformations.

https://doi.org/10.1371/journal.pone.0345977.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Molecular interaction analysis of LS with selected microbial target proteins.

Combined 2D and 3D binding interaction representations illustrating the binding mode, key stabilizing interactions, and active-site accommodation of LS within: (A) peptide deformylase (PDB: 1LMH), (B) lipid A deacetylase LpxC (PDB: 2VES), and (C) serine/threonine phosphatase Z1 (PDB: 5JPF). The 2D interaction maps highlight hydrogen bonding, hydrophobic contacts, and electrostatic interactions, whereas the corresponding 3D views depict LS within the respective catalytic pockets, demonstrating stable, target-specific binding conformations.

https://doi.org/10.1371/journal.pone.0345977.g006

MD simulation and binding affinity analysis

Three-hundred-nanosecond MD simulations were performed to evaluate the structural stability, flexibility, and interaction energetics of the nine selected protein–ligand complexes under physiologically relevant conditions. These analyses included RMSD, RMSF, Rg, SASA, hydrogen-bond profiles, and MM/GBSA binding free energies, providing a comprehensive view of ligand-induced stabilization and dynamic behavior across the complexes.

Comparative binding affinity and structure-activity relationships.

A comparative evaluation of the predicted binding affinities revealed clear differences among the studied protein–ligand systems. Based on the K_d and inferred IC₅₀ values, the LS-bound complex 2VES–LS (K_d = 1.7 × 10 ⁻ ¹¹ M) exhibited the strongest predicted affinity across all evaluated systems, followed by the TM-bound complex 4DXD–TM (K_d = 3.2 × 10 ⁻ ¹⁰ M). Additional high-affinity interactions were observed for the LS-bound 1LMH and TM- or PA-bound 5TZ1 complexes, which consistently demonstrated favorable ΔG(binding) values and low dissociation constants. In contrast, the 5CXK complexes, 5JPF–LS, and 4URN–PA exhibited comparatively weaker predicted affinities, in agreement with their less favorable MM/GBSA binding free energies.

Across ligand classes, LS displayed the strongest and most consistent binding behavior across different protein targets, reflecting its ability to establish stable noncovalent interaction networks. TM exhibited target-dependent high-affinity interactions, performing particularly well in selected systems, whereas PA showed more variable stabilization effects that were highly context-dependent. Importantly, all IC₅₀ values were derived for comparative ranking purposes only and should not be interpreted as absolute inhibitory potencies. Given the indirect estimation of IC₅₀ from K_d and the nonlinear relationship between ΔG, K_d, and biological inhibition, these results represent relative indicators of binding strength rather than quantitative measures of activity.

Fig 7 summarizes these trends by presenting a comparative heatmap of MM/GBSA-derived ΔG(binding) values alongside the predicted K_d estimates for all protein–ligand complexes. Although the computational results provide meaningful structure–activity insights and a relative ranking of ligand performance, experimental validation is required to confirm the predicted inhibitory potential of the identified lead systems.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Heatmap comparing MM/GBSA binding free energies (ΔG) and estimated dissociation constants (K_d) for selected protein–ligand docked complexes.

Lower (more negative) ΔG values and smaller K_d values indicated stronger predicted binding affinities across the evaluated complexes.

https://doi.org/10.1371/journal.pone.0345977.g007

Essential dynamics and free energy landscape analysis using PCA

PCA, combined with FEL mapping, were used to investigate the dominant motions, conformational stability, and energy minima of all protein–ligand complexes during the 300 ns MD simulations. The projection of trajectories onto PC1 and PC2 captured large-scale collective motions, whereas the FEL plots revealed the energetic distribution of sampled conformations, with deeper, more compact minima indicating higher structural stability.

ADMET analysis

All pharmacokinetic and toxicity parameters reported in this study were derived exclusively from in silico predictions using ADMETlab 3.0 with SMILES strings as input, and no in vitro or in vivo toxicity experiments were performed. Accordingly, all toxicity values should be interpreted strictly as computational estimates rather than experimentally validated measurements. ADMET profiling revealed distinct pharmacokinetic and safety characteristics among TM, SA, and SLS. TM exhibited moderate lipophilicity (LogP = 3.16), acceptable predicted solubility, and favorable synthetic accessibility. However, predicted cytochrome P450 inhibition and high plasma protein binding indicate a potential risk of drug–drug interactions. Additional predictions suggested moderate cellular permeability, limited oral absorption, and localized irritation alerts (ocular, dermal, and respiratory), supporting the feasibility of topical or localized therapeutic applications (S29 Fig). SA showed unfavorable pharmacokinetic characteristics, including poor predicted absorption, multiple structural alerts, and high predicted toxicity across hepatic, renal, neurological, and genotoxic endpoints, consistent with its established use as a laboratory biocide rather than a therapeutic agent (S30 Fig). SLS displayed moderate physicochemical parameters but exhibited poor predicted bioavailability, very high plasma protein binding (97.96%), and inhibition of multiple transporters and CYP isoenzymes. Predicted toxicities included hepatic injury, irritation, and environmental hazards, aligning with its known applications in topical or industrial formulations rather than systemic therapy (S31 Fig). SA demonstrated potent antifungal activity but inconsistent antibacterial activity, likely due to its mechanism of action as a respiration inhibitor. In contrast, SLS showed membrane-disruptive activity, consistent with its established non-systemic applications [49].

A comparative assessment integrating experimental antimicrobial data and computational analyses indicated that TM consistently exhibited superior antimicrobial performance among the tested compounds. This observation is supported by its lower MIC values (0.10–0.20 mg/mL) and stable predicted binding interactions with key microbial targets, including FtsZ (K_d = 3.2 × 10 ⁻ ¹⁰ M) and sterol 14-α-demethylase (K_d = 6.1 × 10 ⁻ ⁶ M), in agreement with previous reports [10]. SA demonstrated notable antifungal activity but variable antibacterial efficacy, likely reflecting its mechanism of action as a respiration inhibitor rather than a target-specific antimicrobial agent. In contrast, SLS primarily exhibits membrane-disruptive effects and fails to meet key pharmacological and toxicity criteria, which is consistent with its established non-systemic and industrial applications [49].

Clinical translation and formulation prospects for TM

Given TM’s lipophilicity and moderate oral bioavailability [10], clinically relevant applications are more feasible through topical and localized delivery systems. Recent studies have demonstrated that nanoemulsions [50,51], liposomal carriers [52], polymeric nanoparticles [13], hydrogels [53], and bioadhesive films [54] enhance TM’s solubility, stability, and penetration into infected tissues while minimizing irritation. These platforms are beneficial for treating skin, wound, and device-associated biofilm infections. TM has also shown synergistic activity with frontline antibiotics, including ciprofloxacin, vancomycin, linezolid, and cefazolin [12,55], suggesting that combination therapies may reduce effective doses and help overcome resistance. Together, these formulation strategies and synergistic regimens provide feasible translational pathways to advance TM for therapeutic development.

Limitations of the study

Although this study integrated in vitro antimicrobial and antibiofilm assays with comprehensive in silico analyses, several limitations must be acknowledged. SA and SLS were included primarily as comparative controls because of their known toxicity, and their observed activity should not be interpreted as evidence of therapeutic potential. The exceptionally high docking-derived affinities predicted for LS analogs may partially reflect computational bias arising from rigid receptor assumptions, scoring function limitations, and restricted conformational sampling; thus, caution should be exercised in their interpretation and experimental validation. IC₅₀ estimations were inferred from ΔG values under the approximation that K_i ≈ K_d and are intended solely for comparative ranking, rather than as absolute measures of inhibitory potency. In addition, although PCA and FEL analyses were performed to characterize large-scale conformational motions, additional convergence diagnostics, such as MD trajectory clustering, were not conducted and should be incorporated in future studies to further enhance robustness. Despite these constraints, this study provides meaningful comparative insights, supports TM as a viable repurposable antimicrobial candidate, and identifies LS-based analogs as promising scaffolds for rational optimization.