Preparation of receptor proteins

The three-dimensional crystal structure of the human Epidermal Growth Factor Receptor (EGFR) tyrosine kinase domain was retrieved from the Protein Data Bank (PDB; https://www.rcsb.org/) using PDB ID 1M17 on 15 November 2025. In addition, two experimentally determined EGFR structures carrying clinically relevant mutation combinations were retrieved from the PDB for comparative molecular docking analysis: PDB ID 4I22, representing the L858R/T790M EGFR mutant structure, and PDB ID 6LUD, representing the L858R/T790M/C797S EGFR mutant structure. These structures were used to evaluate the predicted docking characteristics of the investigated phytochemicals across wild-type and mutation-containing EGFR backgrounds. All three protein was prepared for molecular docking and simulation analyses by removing all heteroatoms, including any co-crystallized ligands and water molecules, using Discovery Studio (Studio, 2020). Hydrogen atoms were then added to the protein structure to ensure proper valency and geometry. Energy minimization of the receptor was performed using Swiss-PdbViewer (version 4.1.0) to remove any steric clashes and optimize the overall stability of the protein structure [31]. The processed protein was finally saved in PDB format and used for subsequent molecular docking and molecular dynamics simulation studies. Structural quality parameters and essential properties have been summarized in S1Table and S1 Fig in S1 File.

Preparation of ligands

From the Indian Medicinal Plants, Phytochemistry, and Therapeutics (IMMPAT) (https://cb.imsc.res.in/imppat/) database [32], a total of 16 bioactive phytochemicals of Adenium obesum were collected for this study on 15 November 2025. In addition, four well-established EGFR tyrosine kinase inhibitors Osimertinib, Afatinib, Erlotinib, and Lazertinib were selected as reference (control) drugs due to their proven therapeutic efficacy in the treatment of non-small-cell lung cancer (NSCLC) [33–35]. The chemical structures of the selected Adenium obesum phytochemicals and reference drugs were retrieved from the PubChem database [36] in SDF format. Using PyRx (PyRx 0.8), the three-dimensional structures of all ligands were optimized and converted into PDBQT format to make them suitable for molecular docking studies. ChemDraw Pro-8 was used to generate and present the two-dimensional structures of the phytochemicals and standard drugs, which are illustrated in S2 Table in S1 File. These prepared ligand structures were subsequently used for molecular docking and molecular dynamics simulation analyses to evaluate their binding affinity and stability against the EGFR receptor.

Prediction of Biological Activity (PASS) analysis

The PASS analysis was performed using the widely recognized PASS Online server (http://way2drug.com/PassOnline/predict.php) access on 17 November 2025. The probability of activity (Pa) and probability of inactivity (Pi) values generated by the platform were used to assess the potential biological activities of the studied compounds [37].

Molecular docking

Molecular docking is a fundamental computational approach in drug discovery used to predict the interaction of small molecules with target proteins and to characterize their putative binding modes [38]. In the present study, molecular docking was performed using the AutoDock Vina wizard integrated into PyRx (version 0.8, accessed November 17, 2025) [39]. Docking was performed using a grid encompassing the ATP-binding region and surrounding catalytic residues to allow comprehensive exploration of ligand-binding conformations [40]. Docking was performed with an exhaustiveness value of 8 and a maximum of nine output modes per ligand. The final pose selection was based not only on the lowest predicted binding energy but also on the consistency of the binding orientation, favorable interaction geometry with key catalytic and regulatory residues, and absence of steric clashes, ensuring biologically plausible binding conformations. The docking grid box was defined using grid box parameters detailed in Table 1, ensuring coverage of the entire receptor structure. AutoDock Vina’s default scoring function was employed to estimate the binding affinities between the ligands and the EGFR tyrosine kinase domain. For each ligand, nine different binding poses were generated. Among them, the top-ranked pose exhibiting the lowest binding energy and a root-mean-square deviation (RMSD) value of 0.000 Å for both lower and upper bounds was selected for further analysis. While AutoDock Vina provides efficient and reliable estimations of ligand–receptor binding affinity, it does not fully account for protein flexibility or solvent effects. The protein–ligand interaction patterns and binding conformations were visualized and analyzed using Discovery Studio 2024 (version 24.1.0.23298, accessed November 18, 2025).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Grid box parameters used for PyRx docking analysis in this study.

https://doi.org/10.1371/journal.pone.0357096.t001

To investigate whether clinically relevant EGFR mutations were associated with changes in the predicted docking characteristics of the investigated phytochemicals, comparative molecular docking was additionally performed against two experimentally determined mutation-containing EGFR structures. PDB ID 4I22 [41], representing an EGFR structure carrying the L858R and T790M mutations, and PDB ID 6LUD [42], representing an EGFR structure carrying the L858R, T790M, and C797S mutations, were included for comparative analysis. All 16 Adenium obesum phytochemicals investigated in the present study were docked against the wild-type EGFR structure (1M17) and the two mutation-containing structures (4I22 and 6LUD) using the same docking protocol and parameter settings described above. The resulting docking scores were compared across the three structural backgrounds to examine compound-dependent differences in predicted docking behavior.

Docking protocol validation

To assess the reliability and reproducibility of the molecular docking methodology, a re-docking validation procedure was performed using the co-crystallized native ligand present in the EGFR crystal structure (PDB ID: 1M17). The crystallographic ligand was extracted from the receptor complex and subsequently prepared using the same protocol applied to the screened phytochemicals. Briefly, hydrogen atoms were added, torsional bonds were assigned, and the ligand geometry was energy minimized prior to docking analysis. The prepared native ligand was then re-docked into the ATP-binding pocket of EGFR using AutoDock Vina implemented in PyRx under the same grid coordinates (Table 1) and docking parameters employed during virtual screening. The resulting docked conformation was superimposed onto the experimentally resolved crystallographic pose using PyMOL, and the root mean square deviation (RMSD) between the two conformations was calculated to evaluate docking accuracy. An RMSD value below 2.0 Å was considered indicative of successful reproduction of the experimental binding orientation and validation of the docking protocol [43].

Prediction of drug-likeness and absorption, distribution, metabolism, and excretion (ADMET) properties

The assessment of ADME profiles is a crucial step in pharmacological research and drug development, as it provides valuable insights into a compound’s pharmacokinetic behavior, safety, and therapeutic potential [44]. In the present study, all selected ligand compounds were subjected to comprehensive drug-likeness and ADMET analysis. The Canonical SMILES of each compound were retrieved from the PubChem database and used as input in SwissADME (http://www.swissadme.ch, accessed November 18, 2025) to predict key physicochemical and pharmacokinetic parameters. These included molecular weight, hydrogen bond donors and acceptors, lipophilicity (iLOGP), water solubility (Log S, ESOL), gastrointestinal absorption, bioavailability score, and compliance with Lipinski’s Rule of Five. In addition, medicinal chemistry parameters such as synthetic accessibility were assessed to evaluate the feasibility of compound development. To further strengthen the pharmacokinetic and toxicity assessment, additional in silico tools ADMET SAR, pkCSM, and ProTox-III were employed. These tools were used to predict critical ADMET features such as human intestinal absorption, blood–brain barrier (BBB) permeability, cytochrome P450 enzyme interactions (including CYP2D6 and CYP3A4), AMES mutagenicity, skin sensitization, and general toxicity risks. The predicted ADMET profiles of the selected Adenium obesum compounds were then compared with those of the standard reference drugs to determine their relative suitability as potential drug candidates.

Molecular Dynamics simulations

Molecular dynamics (MD) simulations were conducted to evaluate the structural stability and dynamic behavior of the docked protein–ligand complexes, serving as a critical post-docking validation step.

All simulations were performed using the Desmond MD engine (Schrödinger, LLC) on 25 November 2025, which offers high computational efficiency and scalability for biomolecular systems. The production simulations were carried out for 100 ns under an NPT ensemble, maintaining constant pressure and temperature to mimic physiological conditions. The protein–ligand complex was prepared using the Protein Preparation Wizard in Maestro (Schrödinger Suite). Missing side chains and loops were rebuilt, bond orders were corrected, and hydrogen atoms were added. The protonation states of ionizable residues and ligands were assigned at physiological pH (7.2) using Epik and PROPKA, ensuring accurate charge distribution throughout the system. The OPLS4 force field was employed to parameterize both protein and ligand, providing a consistent and reliable description of molecular interactions. The system was solved in an explicit TIP3P water model using a trucked octahedron simulation box, maintaining a 10 Å buffer distance around the complex to prevent boundary artifacts. To replicate physiological ionic strength, 0.15 M NaCl was added, and periodic boundary conditions were applied in all three dimensions. Prior to the production phase, the system underwent energy minimization followed by a multi-step equilibration process, according to Desmond’s default relaxation protocol, ensuring system stability and relaxation of steric clashes.

Following the completion of the simulation, multiple structural and dynamic parameters were analyzed using the Simulation Interaction Diagram (SID) and Maestro Analysis tools. These included roots mean square deviation (RMSD) and root mean square fluctuation (RMSF) to assess structural stability and residue flexibility, radius of gyration (Rg) to determine compactness, and total and potential energy profiles to monitor system convergence. Additionally, protein–ligand interaction profiles such as hydrogen bonds, hydrophobic interactions, π–π stacking, salt bridges, and water-mediated contacts were systematically evaluated to provide a comprehensive understanding of the conformational stability and interaction persistence of the complexes throughout the simulation trajectory.

DFT analysis

The Jaguar module of Schrodinger Suite with the B3LYP (Becke, 3-parameter, Lee-Yang-Parr) functional and the 6-31G* basis set was used to conduct DFT calculations [45,46]. Single-point energy calculations were performed. Analysis of frontier molecular orbitals provided insights into the energies of the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO), allowing for the derivation of the energy gap and global reactivity descriptors. Molecular Electrostatic Potential (MEP) maps were generated to visualize the distribution of surface charge.

1. (1) 𝐸𝑔𝑎𝑝 = (𝐸𝐿𝑈𝑀𝑂 − 𝐸𝐻𝑂𝑀𝑂)
2. (2) 𝐼 = −𝐸𝐻𝑂𝑀𝑂
3. (3) 𝐴 = −𝐸𝐿𝑈𝑀𝑂
4. (4) 𝜒 = (𝐼+𝐴) 2
5. (5) ω = 𝜇 2 (2𝜂)
6. (6) 𝜇 = − (𝐼+𝐴) 2
7. (7) 𝜂 = (𝐼−𝐴) 2
8. (8) 𝑆 = 1 (2𝜂)

Evolutionary conservation and structural hotspot mapping of the EGFR kinase domain

To assess the biological relevance of the predicted binding site, a protein-level conservation analysis was performed on the human epidermal growth factor receptor (EGFR). The EGFR kinase-domain structure used in the docking workflow, PDB ID 1M17, was selected as the structural template for residue mapping. Residues implicated in ligand binding were first identified from the docking and molecular dynamics analyses of the present study and then evaluated for evolutionary conservation using the ConSurf framework, which estimates residue-wise conservation from homologous sequence alignments and assigns conservation grades on a scale of 1–9, where higher values indicate stronger conservation [47]. The final conservation scores were mapped onto the EGFR kinase domain and integrated with the previously identified binding-pocket residues. In addition, structurally corresponding hotspot positions relevant to clinically important EGFR mutation sites associated with NSCLC, including L858R, T790M, and C797S, were also examined in the wild-type structural context. Structural visualization and figure preparation were performed in UCSF ChimeraX [48].

Clinical expression and pharmacogenomic profiling of EGFR in NSCLC

To provide clinical support for the structural findings, EGFR was evaluated in lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) using GEPIA3 [49]. Tumor-versus-normal expression and pathological stage analyses were performed to assess the clinical expression profile of EGFR in NSCLC. The Cell Line Screen module of GEPIA3 was additionally used to examine the association between EGFR expression and drug-sensitivity profiles across lung cancer cell lines.

Mutational landscape analysis of EGFR in lung adenocarcinoma

To further contextualize the structural and pharmacogenomic findings, the mutation distribution of EGFR was examined using cBioPortal, a cancer genomics platform for exploring multidimensional molecular alteration data across large cancer cohorts [50]. A lung adenocarcinoma cohort was queried, and the positional distribution of EGFR alterations across the full-length receptor was visualized using the mutation-domain plot. This analysis was used to summarize the location of recurrent EGFR alterations relative to the receptor architecture, with particular attention to hotspot events in the kinase-domain region.

PASS-Based Bioactivity Prediction

The PASS online platform was applied to predict the biological activities of the sixteen selected compounds, generating probability of activity (Pa) and probability of inactivity (Pi) values for five major pharmacological properties, including antineoplastic, apoptosis agonist, antiproliferative, antibacterial, and antifungal activities (Table 2). The PASS prediction spectrum revealed a remarkably high likelihood of anticancer and cytotoxic-related activities among the analyzed compounds, as evidenced by consistently elevated Pa values and minimal corresponding Pi values. In general, predicted activities with Pa > 0.7 were considered high-confidence, Pa values between 0.5 and 0.7 were interpreted as moderate confidence, and Pa < 0.5 indicated low-confidence predictions, consistent with standard PASS interpretation guidelines. However, PASS predictions are based on structure–activity relationships derived from known bioactive compounds and do not account for target-specific binding, dose dependency, or pharmacokinetic behavior; therefore, the results were treated as qualitative indicators and were further validated using molecular docking, ADME profiling, toxicity assessment, and molecular dynamics simulations.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. PASS-based prediction of antineoplastic, apoptosis agonist, antiproliferative, antibacterial, and antifungal activities (Pa/Pi) of selected compounds.

https://doi.org/10.1371/journal.pone.0357096.t002

For antineoplastic activity, the probability of activity (Pa) values ranged from 0.791 to 0.928, while the probability of inactivity (Pi) remained extremely low (0.005–0.013), indicating strong and highly confident predictions. Among the screened compounds, Hongheloside A demonstrated the highest antineoplastic potential (Pa = 0.928), followed by Hongheloside C (Pa = 0.925), Mono-O-acetylacoschimperoside P (Pa = 0.917), Neritaloside (Pa = 0.911), and Obebioside B (Pa = 0.910). All sixteen compounds displayed Pa values greater than 0.79, satisfying the high-confidence threshold for biological activity prediction. Additionally, the apoptosis against activity revealed Pa values in the range of 0.690 to 0.885, suggesting that the majority of compounds possess a moderate to high capacity to induce programmed cell death mechanisms. Isokaempferide exhibited the highest predicted apoptosis agonist activity (Pa = 0.885), followed by SCHEMBL5933906 (Pa = 0.831) and Somalin (Pa = 0.809). The remaining compounds maintained robust apoptosis-related Pa values, all exceeding 0.690, reinforcing their cytotoxic potential at the molecular level.

Even more pronounced results were observed for antiproliferative activity, which exhibited Pa values ranging from 0.606 to 0.988. Obetrioside B recorded the highest antiproliferative probability (Pa = 0.988; Pi = 0.000), followed by Echujin (Pa = 0.983), Hongheloside C (Pa = 0.982), and Obebioside B (Pa = 0.983). These results point to a strong predicted ability of the screened compounds to inhibit uncontrolled cellular multiplication, a critical feature in anticancer activity screening. In contrast, the antibacterial and antifungal predictions exhibited relatively moderate probabilities. Antibacterial Pa values ranged from 0.200 to 0.680, with the highest activity observed in Somalin (Pa = 0.680), followed by Hongheloside C (Pa = 0.674) and Obetrioside B (Pa = 0.669). Similarly, antifungal Pa values ranged between 0.277 and 0.810, where Obetrioside B showed the maximum predicted antifungal activity (Pa = 0.810), followed by Hongheloside C (Pa = 0.801) and Echujin (Pa = 0.800).

Across all biological activity categories, the probability of inactivity (Pi) remained significantly lower than the corresponding Pa values, reinforcing the reliability of the predicted bioactivities. Notably, antineoplastic and antiproliferative activities consistently demonstrated the highest Pa scores among the screened properties, highlighting these compounds as strong candidates for further cancer-focused computational evaluation. Consequently, based on their superior anticancer-associated predictive scores, these compounds were prioritized for downstream molecular docking, ADMET analysis, and molecular dynamics simulations with selected therapeutic target proteins.

Molecular docking analysis

Molecular docking was performed to evaluate the predicted binding poses and relative docking scores of Adenium obesum-derived phytochemicals within the epidermal growth factor receptor (EGFR) tyrosine kinase domain (PDB ID: 1M17). Docking scores, expressed in kcal/mol, were used as computational ranking metrics within the applied AutoDock Vina protocol, with more negative scores indicating more favorable predicted ligand–receptor interactions according to the docking scoring function. Importantly, these scores represent model-dependent computational estimates and should not be interpreted as experimentally measured binding free energies, equilibrium binding affinities, or inhibitory potencies. The docking results are summarized in Table 3 and demonstrate a heterogeneous distribution of predicted binding scores across the investigated phytochemical library.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Docking scores (kcal/mol) of Adenium obesum phytochemicals and standard EGFR inhibitors against EGFR (PDB ID: 1M17).

https://doi.org/10.1371/journal.pone.0357096.t003

For descriptive purposes, the compounds were grouped according to their relative docking-score ranges to facilitate comparative interpretation within the present dataset. Compounds with the most favorable predicted docking scores (approximately ≤ −9.0 kcal/mol) included SCHEMBL5933906 (−10.0 kcal/mol), Echujin (−9.9 kcal/mol), Cardenolide (−9.9 kcal/mol), and Obetrioside A (−9.4 kcal/mol). Under the applied docking protocol, these compounds received more favorable scores than the reference EGFR-targeting compounds Osimertinib (−7.9 kcal/mol) and Erlotinib (−7.0 kcal/mol). However, these differences should be interpreted strictly as differences in computational docking scores within the same protocol and should not be taken as evidence that these phytochemicals possess greater experimental binding affinity or inhibitory potency than the approved EGFR-targeting drugs. The results instead identify these compounds as candidates for further structural and computational evaluation based on their predicted compatibility with the EGFR binding site.

Compounds with intermediate docking scores, approximately ranging from −8.0 to −9.0 kcal/mol, included Obetrioside B (−8.5 kcal/mol), Obebioside B (−8.6 kcal/mol), and Hongheloside C (−8.8 kcal/mol). Their predicted docking scores were within a similar numerical range to those obtained for the reference compounds Afatinib (−8.1 kcal/mol) and Lazertinib (−8.2 kcal/mol). These numerical similarities indicate comparable ranking under the applied docking protocol; however, they do not establish equivalent binding affinity, inhibitory activity, or pharmacological efficacy. Accordingly, these compounds were considered candidates for further evaluation based on complementary structural, pharmacokinetic, and dynamic analyses rather than on docking scores alone.

Compounds with relatively less favorable docking scores within the investigated set included Obebioside A (−7.8 kcal/mol), Obebioside C (−7.8 kcal/mol), and Neritaloside (−7.9 kcal/mol). Although these compounds received less favorable scores than the highest-ranked phytochemicals, their docking results alone are insufficient to exclude them from further consideration. Their potential relevance should instead be assessed in conjunction with binding-mode plausibility, protein–ligand interaction patterns, physicochemical characteristics, ADMET profiles, and molecular dynamics behavior.

Overall, the docking analysis identified several A. obesum phytochemicals with favorable predicted docking scores within the EGFR kinase domain under the applied computational protocol. SCHEMBL5933906, Echujin, Cardenolide, and Obetrioside A were among the top-ranked compounds based on docking score, whereas other phytochemicals exhibited intermediate or relatively less favorable scores. Although several phytochemicals received numerically more favorable docking scores than the reference EGFR-targeting compounds, these findings should not be interpreted as evidence of superior inhibitory potency or therapeutic efficacy. Instead, the docking results served as an initial computational prioritization step for subsequent analyses integrating protein–ligand interaction patterns, drug-likeness, ADMET and toxicity predictions, molecular dynamics simulations, and, ultimately, experimental validation. This integrated approach was used to distinguish compounds with favorable predicted docking characteristics from those with a more balanced overall computational profile for further investigation.

Validation of molecular docking

The reliability of the molecular docking protocol was validated through re-docking of the co-crystallized native ligand into the ATP-binding cavity of EGFR (PDB ID: 1M17). The predicted docking pose was superimposed onto the experimentally determined crystallographic conformation to evaluate the accuracy and reproducibility of the docking methodology. The detailed validation parameters and docking outcomes are summarized in S3 Table in S1 File. The superimposed structures demonstrated strong spatial agreement between the native and re-docked conformations, particularly within the rigid aromatic scaffold occupying the ATP-binding pocket (S2 Fig in S1 File). The calculated RMSD between the crystallographic and predicted poses was 1.64 Å, indicating excellent reproduction of the experimentally observed binding orientation and confirming the accuracy of the selected docking parameters. These findings validated the reliability and robustness of the docking protocol and supported its suitability for subsequent virtual screening and molecular interaction analyses of the investigated phytochemicals against EGFR.

Analysis of protein-ligand interactions

Molecular docking of twenty selected compounds against the EGFR kinase domain (PDB ID: 1M17) revealed a broad range of binding affinities, with docking energies spanning from −7.0 to −10.0 kcal·mol^-1. The highest binding affinity was observed for SCHEMBL5933906 (−10.0 kcal·mol^-1), followed by Echujin (−9.9 kcal·mol^-1), Cardenolide (−9.9 kcal·mol^-1), and Obetrioside A (−9.4 kcal·mol^-1). Several ligands, including Echujin, Cardenolide, and Obetrioside A, outperformed clinically approved inhibitors such as Osimertinib (−7.9 kcal·mol^-1), Afatinib (−8.1 kcal·mol^-1), and Erlotinib (−7.0 kcal·mol^-1), suggesting their potential as high-affinity EGFR binders.

Detailed inspection of hydrogen bonding patterns highlighted that the most promising compounds formed multiple short, directional hydrogen bonds with key catalytic residues, which are crucial for stabilizing the ligand within the ATP-binding site. The analysis of non-bond interactions is detailed in S4 Table in S1 File, while Fig 3 presents the interaction profile of the selected cardenolide compound; the remaining selected compounds are shown in S3 Fig in S1 File Echujin formed seven hydrogen bonds, including strong conventional interactions with LYS851 (1.89 Å) and MET769 (2.03 Å), as well as moderate bonds with GLU734, GLU738, and additional LYS851 contacts (2.23–2.88 Å). Obetrioside A established eight hydrogen bonds, notably engaging LYS721 (2.04 Å), MET769 (2.07 Å), and PHE699 (2.41–2.78 Å), demonstrating a dense network of short-distance interactions with residues critical for EGFR catalysis. Obetrioside B formed six hydrogen bonds, including a very strong bond with ASP831 (1.97 Å), along with interactions with MET769 and CYS773 (2.27–2.32 Å), highlighting the importance of these conserved residues in stabilizing ligand binding. Conversely, SCHEMBL5933906, despite its highest docking score, lacked strong conventional hydrogen bonds, with reported interactions limited to two hydrophobic contacts at >5.1 Å (LYS721 and MET742). This discrepancy indicates that its predicted binding energy may be primarily driven by non-directional hydrophobic interactions or scoring algorithm bias, underscoring the necessity for pose validation and subsequent MM-GBSA or molecular dynamics evaluation.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Cardenolide docking poses with EGFR (1M17): 3D ligand orientations in the binding pocket and 2D interaction maps showing H-bonds and hydrophobic.

https://doi.org/10.1371/journal.pone.0357096.g003

Hydrophobic interactions and π-based contacts further contributed to ligand stabilization, with residues LEU694, PHE699, VAL702, ALA719, LEU820, CYS773, and MET742 frequently involved. Cardenolide, for example, combined a strong hydrogen bond with ASP831 (2.16 Å) with multiple alkyl contacts to LEU694, VAL702, and ALA719 (3.98–5.22 Å), reflecting an optimal balance of directional and non-directional stabilization. Other high-ranking compounds, such as Hongheloside A, Obebioside A, and Strospeside, displayed extensive hydrophobic interactions, often complemented by moderate hydrogen bonds (2.7–3.5 Å) with key residues, contributing to their moderate-to-high docking energies (−7.8 to −8.5 kcal·mol^-1). Analysis of residue-level engagement revealed several conserved “hotspots” critical for ligand recognition: MET769, ASP831, and LYS721 were consistently involved in strong hydrogen bonds, while PHE699, LEU694, VAL702, ALA719, GLU734, GLU738, CYS773, and LYS851 contributed to hydrophobic packing. High-affinity compounds preferentially interacted with multiple hotspots simultaneously, a pattern reminiscent of clinically validated EGFR inhibitors.

Integrating docking energies with detailed interaction patterns, Echujin, Obetrioside A, Obetrioside B, and Cardenolide emerged as the most structurally and energetically plausible ligands. These compounds combined multiple strong hydrogen bonds (<2.2 Å) with key hotspot residues and extensive hydrophobic contacts, indicating favorable enthalpic stabilization and optimal binding geometry. Compounds whose high docking scores relied primarily on long-range hydrophobic interactions, such as SCHEMBL5933906, require additional validation via molecular dynamics simulations to confirm binding stability and reliability. Overall, the protein–ligand interaction analysis demonstrates that several phytochemicals from Adenium obesum possess stronger predicted binding affinities than reference EGFR inhibitors. The integration of docking scores, hydrogen bond quality, and hydrophobic contacts provides a robust framework for prioritizing candidates for further molecular dynamics and free energy simulations, supporting their potential as novel EGFR-targeting ligands.

Molecular docking poses and its interaction analysis

The binding modes of the selected ligands within the EGFR active site were analyzed using BIOVIA Discovery Studio and PyMOL, providing detailed 2D and 3D interaction visualizations of the docked complexes. The analysis revealed a combination of conventional hydrogen bonds, non-conventional hydrogen bonds, and hydrophobic interactions (including π-alkyl, π-sigma, and alkyl interactions) that collectively stabilize the ligand within the ATP-binding pocket. Key hydrogen bond donors and acceptors were identified, highlighting their critical role in orienting the ligands and enhancing binding affinity. The 3D representations illustrated the spatial orientation of each ligand relative to the binding cavity, while the 2D diagrams provided a clear depiction of specific residue interactions, including distances and bond types. As shown in Fig 3 and S3 Fig in S1 File, these visualizations collectively demonstrate the molecular interactions that underpin ligand binding, providing a structural rationale for the observed docking energies and supporting the selection of high-affinity compounds for further molecular dynamics simulations.

Comparative docking across wild-type and mutant EGFR structures

To further examine the predicted binding behavior of the 16 Adenium obesum phytochemicals in clinically relevant EGFR mutation-containing structural backgrounds, all compounds were additionally docked against EGFR structures carrying the L858R/T790M mutation combination (PDB ID: 4I22) and the L858R/T790M/C797S mutation combination (PDB ID: 6LUD). The wild-type EGFR structure (PDB ID: 1M17), used in the initial docking analysis, served as the reference system (Table 4). This comparative analysis was performed to assess whether the predicted docking characteristics of the investigated phytochemicals varied across wild-type and mutation-containing EGFR structural backgrounds.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Comparative docking scores of 16 Adenium obesum phytochemicals against wild-type and clinically relevant mutant EGFR structural backgrounds.

https://doi.org/10.1371/journal.pone.0357096.t004

The docking results demonstrated compound-dependent variation across the three evaluated EGFR structures. SCHEMBL5933906 exhibited consistently favorable predicted docking scores across all three structural backgrounds, with values of −10.0 kcal/mol for wild-type EGFR, −10.0 kcal/mol for the L858R/T790M structure, and −10.8 kcal/mol for the L858R/T790M/C797S structure. Several other phytochemicals also exhibited more negative docking scores in at least one mutation-containing structural background compared with the wild-type structure. For example, Obebioside C showed docking scores of −7.8 kcal/mol against wild-type EGFR, −9.0 kcal/mol against the L858R/T790M structure, and −8.8 kcal/mol against the L858R/T790M/C797S structure. Similarly, Obebioside A, Strospeside, Obeside C, Obetrioside B, Obebioside B, and Hongheloside C exhibited more negative predicted docking scores in at least one of the evaluated mutation-containing structures relative to wild-type EGFR.

Conversely, several compounds exhibited less negative docking scores in one or both mutation-containing structural backgrounds compared with wild-type EGFR. For instance, Cardenolide exhibited docking scores of −9.9 kcal/mol against wild-type EGFR, −8.6 kcal/mol against the L858R/T790M structure, and −8.5 kcal/mol against the L858R/T790M/C797S structure. Despite this relative reduction in docking score compared with the wild-type structure, Cardenolide retained favorable predicted docking characteristics in both mutation-containing structural backgrounds. Echujin similarly showed docking scores of −9.9, −9.2, and −9.6 kcal/mol against wild-type, L858R/T790M, and L858R/T790M/C797S EGFR, respectively. Other compounds, including Hongheloside A, Somalin, Neritaloside, Isokaempferide, and Mono-O-acetylacoschimperoside P, also displayed compound-specific variations in their predicted docking scores across the evaluated structural systems.

Overall, the comparative analysis demonstrated that the predicted docking characteristics of the investigated phytochemicals varied according to the EGFR structural background, with no uniform pattern of either increased or decreased docking scores across all compounds. Some phytochemicals exhibited comparable or more negative predicted docking scores in mutation-containing structures, whereas others showed relative changes in docking scores compared with wild-type EGFR. These observations indicate that mutation-associated structural differences may influence the predicted ligand–EGFR interaction landscape in a compound-dependent manner. Importantly, the docking scores were interpreted only as computational indicators of predicted ligand–protein interaction propensity and were not used as direct measures of experimental binding affinity or inhibitory potency. Furthermore, the present analysis does not establish whether any of the investigated phytochemicals can inhibit mutant EGFR enzymatic activity or overcome clinically established EGFR tyrosine kinase inhibitor resistance. Because the evaluated mutant structures contain combinations of EGFR mutations rather than isolated single mutations, the observed differences cannot be attributed to the independent effects of L858R, T790M, or C797S individually. Therefore, these findings should be considered hypothesis-generating and provide a comparative computational assessment of the predicted interaction behavior of the investigated phytochemicals across wild-type and clinically relevant mutation-containing EGFR structural backgrounds.

Analysis of drug-like properties

A comprehensive analysis of the physicochemical and pharmacokinetic characteristics of the 16 Adenium obesum phytochemicals was performed to evaluate their predicted drug-likeness, compliance with Lipinski’s Rule of Five, and bioavailability-related properties (Table 5). The investigated compounds exhibited substantial physicochemical diversity, with molecular weights ranging from 300.00 to 917.00 g/mol, hydrogen-bond acceptor counts ranging from 2 to 20, hydrogen-bond donor counts ranging from 0 to 9, and topological polar surface areas (TPSA) ranging from 26.30 to 299.00 Ų. This broad variation in molecular size, polarity, hydrogen-bonding capacity, and lipophilicity was reflected in differences in predicted Lipinski rule compliance, bioavailability scores, and gastrointestinal (GI) absorption across the investigated compounds.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Predicted drug-likeness and pharmacokinetic properties of Adenium obesum-derived phytochemicals evaluated as computationally prioritized EGFR-binding candidates.

https://doi.org/10.1371/journal.pone.0357096.t005

Six compounds, Hongheloside A, Strospeside, Obeside C, Somalin, Isokaempferide, and Cardenolide showed comparatively favorable drug-likeness profiles, with no more than one predicted Lipinski rule violation. Among these compounds, molecular weights ranged from 300.00 g/mol for Isokaempferide to 592.72 g/mol for Neritaloside, while TPSA values ranged from 26.30 to 140.98 Ų. Isokaempferide and Cardenolide exhibited the lowest molecular weights within this group, at 300.00 and 342.00 g/mol, respectively, and both contained a limited number of rotatable bonds. These compounds, together with Hongheloside A, Strospeside, Obeside C, and Somalin, showed a predicted bioavailability score of 0.55. High predicted GI absorption was observed for Hongheloside A, Strospeside, Obeside C, Somalin, Isokaempferide, and Cardenolide, whereas Neritaloside, despite exhibiting only one predicted Lipinski violation, showed low predicted GI absorption.

In contrast, several phytochemicals exhibited multiple predicted Lipinski rule violations. SCHEMBL5933906, Obebioside A, Obetrioside B, Obebioside B, Echujin, Hongheloside C, Obebioside C, Obetrioside A, and Mono-O-acetylacoschimperoside P showed two or three predicted violations. These compounds generally possessed higher molecular weights (634.75–917.00 g/mol), greater hydrogen-bonding capacities, and elevated TPSA values. In particular, Obetrioside B exhibited the highest molecular weight (917.00 g/mol) and TPSA (299.00 Ų) among the investigated compounds. Echujin also showed a relatively high molecular weight (843.00 g/mol), nine hydrogen-bond donors, 17 hydrogen-bond acceptors, and a TPSA of 252.75 Ų. These physicochemical characteristics may present challenges for passive membrane permeability and oral absorption.

The predicted bioavailability scores and GI absorption profiles were broadly consistent with the observed physicochemical characteristics. Most compounds with no more than one predicted Lipinski violation showed a bioavailability score of 0.55, whereas compounds with two or three predicted violations generally exhibited a lower predicted bioavailability score of 0.17. Similarly, the majority of compounds with multiple Lipinski violations showed low predicted GI absorption. However, Lipinski rule compliance did not uniformly correspond to high predicted absorption, as Neritaloside demonstrated one predicted violation but low GI absorption. These findings indicate that drug-likeness and absorption are influenced by multiple interacting physicochemical properties rather than by Lipinski rule compliance alone. The predicted consensus Log Po/w values ranged from 0.21 to 5.22, indicating considerable variation in lipophilicity among the investigated phytochemicals. Several compounds showed moderate predicted lipophilicity, whereas Cardenolide exhibited the highest predicted consensus Log Po/w value (5.22). This relatively high predicted lipophilicity may represent a potential developability consideration and should be evaluated alongside its other physicochemical and pharmacokinetic characteristics. Similarly, the predicted water solubility values varied considerably among the compounds, further demonstrating that favorable drug-likeness parameters in one dimension do not necessarily indicate an optimal overall pharmacokinetic profile.

Overall, the drug-likeness assessment highlights a clear distinction between the smaller, less polar compounds and the larger, highly polar glycosylated phytochemicals. Isokaempferide, Cardenolide, Somalin, and Hongheloside A exhibited comparatively favorable predicted drug-likeness and absorption-related profiles, with physicochemical characteristics that may be more compatible with further oral drug-development investigation. However, these findings should not be interpreted as evidence of confirmed oral bioavailability or therapeutic suitability. Compounds with multiple predicted Lipinski rule violations and low gastrointestinal absorption may still possess pharmacological potential, but their predicted physicochemical and pharmacokinetic characteristics could present additional challenges that may require structural optimization, formulation strategies, alternative delivery approaches, or experimental pharmacokinetic evaluation. Collectively, these findings emphasize the importance of considering drug-likeness and predicted pharmacokinetic characteristics alongside target-binding potential when prioritizing natural-product candidates for further investigation.

ADME analysis

The pharmacokinetic properties of the selected Adenium obesum–derived phytochemicals were evaluated (Table 6) using in silico absorption, distribution, metabolism, and excretion (ADME) profiling in order to assess their suitability as potential therapeutic candidates against non-small-cell lung cancer (NSCLC). Key parameters including water solubility (Log S), Caco-2 cell permeability, human intestinal absorption, volume of distribution at steady state (VDss), blood–brain barrier (BBB) permeability, cytochrome P450 (CYP) inhibition potential, total clearance, and renal OCT2 substrate affinity were analyzed and compared.

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Predicted ADME (Absorption, Distribution, Metabolism, and Excretion) properties of selected Adenium obesum phytochemicals.

https://doi.org/10.1371/journal.pone.0357096.t006

Most of the investigated phytochemicals exhibited moderate to high predicted intestinal absorption, indicating favorable oral bioavailability potential. Compounds such as Cardenolide (98.34%), Somalin (96.75%), Hongheloside A (95.62%), and Mono-O-acetylacoschimperoside P (90.62%) demonstrated particularly strong absorption profiles. Moderate absorption was also observed for Neritaloside (81.95%), Isokaempferide (84.05%), and Strospeside (72.26%), suggesting these compounds may be efficiently absorbed through the human gastrointestinal tract. In contrast, a few compounds such as Obetrioside B (19.17%) and Echujin (33.73%) displayed comparatively lower absorption rates, indicating relatively limited bioavailability. Additionally, this absorption trend was further supported by Caco-2 permeability predictions, where Somalin (1.457 × 10 − ⁶ cm/s) and Cardenolide (1.370 × 10 − ⁶ cm/s) showed the highest permeability, followed by Hongheloside A (0.726 × 10 − ⁶ cm/s) and Strospeside (0.711 × 10 − ⁶ cm/s). These findings indicate that the leading compounds have a strong capacity to cross intestinal epithelial barriers via passive diffusion. However, several compounds such as Echujin, Obetrioside B, and Hongheloside C showed lower permeability values, which may limit their absorption efficiency. All compounds exhibited low aqueous solubility (log S values between −2.92 and −5.58), which is characteristic of many plant-derived secondary metabolites. Although low solubility may present formulation challenges, it is often associated with improved membrane permeability and enhanced interaction with hydrophobic protein targets such as EGFR.

In terms of distribution, most phytochemicals displayed moderate predicted volumes of distribution (VDss), ranging from −0.867 to 0.765 log L/kg. Cardenolide (0.765) and Somalin (0.332) indicated broader systemic distribution, which may enhance access to tumor tissues. Meanwhile, compounds such as Isokaempferide (−0.954) and Obebioside C (−0.647) showed more restricted tissue distribution, suggesting confinement primarily within the plasma compartment. Moreover, Blood–brain barrier (BBB) permeability prediction revealed that nearly all investigated phytochemicals demonstrated negative BBB scores, indicating limited penetration into the central nervous system. Notably, Cardenolide exhibited a marginally positive BBB value (0.239), suggesting only a weak potential for CNS permeation. Although slight BBB permeability is generally undesirable for peripheral cancer targets, this low positive value remains within an acceptable range and is unlikely to result in significant neurological exposure. Moreover, such limited permeability could be contextually beneficial in cases where metastasis to the brain occurs in advanced NSCLC.

Furthermore, the metabolic profiles of the selected Adenium obesum phytochemicals were largely favorable. The majority of compounds were predicted not to inhibit key cytochrome P450 (CYP) isoforms, indicating a low likelihood of adverse drug–drug interactions. Among the tested phytochemicals, Isokaempferide was predicted to inhibit CYP1A2, whereas all other compounds showed no inhibitory activity toward CYP1A2 or CYP2C9. These findings suggest that, with the exception of Isokaempferide, the phytochemicals are unlikely to interfere significantly with hepatic metabolism, supporting their favorable pharmacokinetic profiles for potential use in combination therapies targeting non-small-cell lung cancer.

The predicted excretion profiles of the selected Adenium obesum phytochemicals indicated a range of clearance rates, reflecting balanced systemic retention across the panel. Total clearance values ranged from 0.385 to 0.617 ml/min/kg, representing slow, moderate, and relatively faster elimination. Compounds such as Mono-O-acetylacoschimperoside P (0.385 ml/min/kg) and Hongheloside C (0.420 ml/min/kg) exhibited slower clearance, suggesting prolonged systemic exposure, whereas Obebioside A (0.617 ml/min/kg) and Echujin (0.605 ml/min/kg) showed relatively higher clearance, indicative of faster elimination. Several compounds, including Cardenolide (0.478 ml/min/kg), SCHEMBL5933906 (0.445 ml/min/kg), and Neritaloside (0.433 ml/min/kg), display moderate clearance, providing an optimal balance between sufficient plasma retention and effective elimination. Importantly, none of the phytochemicals were predicted to be substrates of renal OCT2 transporters, reducing the risk of transporter-mediated renal accumulation or nephrotoxicity. Overall, the excretion profiles suggest that most compounds possess excretion profiles compatible with safe and potential therapeutic use.

Overall, the integrated ADME profiling highlights Cardenolide, Somalin, Hongheloside A, Neritaloside, and Mono-O-acetylacoschimperoside P as the most pharmacokinetically promising candidates, displaying a favorable balance of high intestinal absorption, adequate permeability, moderate distribution, CYP inhibition, and safe excretion potential. These pharmacokinetic advantages, combined with their strong molecular docking affinities toward EGFR, support their selection for further molecular dynamics simulations and advanced pharmacological investigation in the context of NSCLC targeted therapy.

Toxicity analysis

The in-silico toxicity profiling of the selected Adenium obesum phytochemicals revealed a broadly favorable safety landscape, strengthening their suitability as potential EGFR-targeted therapeutic candidates for non-small-cell lung cancer (NSCLC), as summarized in Table 7. Notably, none of the screened compounds exhibited predicted hERG channel inhibition, indicating a minimal risk of cardiotoxicity, a critical parameter in anticancer drug development. The acute oral toxicity predictions, expressed as LD₅₀ values in rats, ranged from 2.387 to 3.706 mol/kg, suggesting an overall low to moderate toxicity profile across the compound library. Compounds such as SCHEMBL5933906 (3.706 mol/kg), Hongheloside C (3.156 mol/kg), Obebioside C (3.159 mol/kg), and Obebioside B (3.232 mol/kg) demonstrated comparatively higher LD₅₀ values, indicating lower acute toxicity and improved safety margins. Conversely, Somalin (2.387 mol/kg) and Isokaempferide (2.395 mol/kg) presented relatively lower LD₅₀ values, suggesting a moderately increased acute toxicity, albeit still within a tolerable range for further investigation.

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. In silico toxicity profiling of selected Adenium obesum phytochemicals based on key safety parameters.

https://doi.org/10.1371/journal.pone.0357096.t007

Oral rat chronic toxicity values largely remained within a moderate window (1.207–3.092), with Obetrioside B (3.092), Echujin (3.051), and Hongheloside C (2.965) exhibiting higher thresholds for chronic exposure tolerability. Importantly, the vast majority of the phytocompounds tested negative in the AMES mutagenicity assay, reflecting a reduced likelihood of genotoxic or carcinogenic effects. Only Strospeside and Obeside C displayed positive AMES predictions, warranting cautious interpretation and prioritizing them lower in the lead selection process. Furthermore, all compounds were predicted to be non-sensitizing to the skin and non-hepatotoxic, underscoring a favorable systemic safety profile with minimal risk of liver-associated adverse effects, a key concern in chemotherapeutic regimens.

The predicted maximum tolerated dose (MTD) in humans demonstrated compound-specific variation, with most values remaining within a negative log-scale range, suggestive of acceptable dose tolerability for therapeutic application. Interestingly, Isokaempferide exhibited a slightly positive MTD value (0.340), implying relatively higher dose tolerance in comparison to other screened constituents. In contrast, SCHEMBL5933906 (−2.078), Echujin (−2.135), and Hongheloside C (−2.046) indicated comparatively lower human dose thresholds, which may necessitate careful dose optimization. Among the evaluated candidates, Cardenolide emerged as a particularly balanced compound, displaying no cardiotoxic, hepatotoxic, mutagenic, or sensitizing effects, along with moderate acute and chronic toxicity values (LD₅₀ = 2.489 mol/kg; chronic toxicity = 2.145) and a reasonable predicted human tolerability (MTD = −0.376). This overall toxicity signature, combined with its favorable performance in the preceding in silico analyses, positions Cardenolide as one of the most promising and pharmacologically safe lead candidates from Adenium obesum for further experimental validation in NSCLC therapy. Collectively, the toxicity evaluation highlights that the majority of the investigated phytochemicals possess an acceptable safety and tolerability profile, thereby supporting their progression into advanced preclinical and experimental validation phases as potential EGFR target candidate for lung cancer treatment.

Molecular dynamics simulation

To validate the structural stability and dynamic behavior of the top-ranked Adenium obesum phytochemical Cardenolide –EGFR complex obtained from molecular docking, a 100 ns all-atom molecular dynamics (MD) simulation was performed under physiological conditions using the Desmond simulation package. MD simulations were employed to move beyond static docking poses and evaluate whether the ligand–protein complex remains dynamically stable, structurally coherent, and energetically favorable under explicit solvent conditions.

Protein–ligand RMSD analysis

Protein–ligand RMSD analysis (Fig 4) was performed to assess the structural stability and equilibration behavior of the EGFR–Cardenolide phytochemical complex during the MD simulation. Following backbone alignment to the reference structure, the EGFR kinase domain rapidly equilibrated and maintained RMSD fluctuations within ~1–3 Å throughout the production phase, consistent with a stable, globular kinase fold. The absence of progressive drift or large-amplitude deviations indicates that the system reached thermodynamic equilibrium and that ligand binding does not induce destabilizing global conformational rearrangements.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Protein–ligand RMSD analysis of the EGFR in complex with the Adenium obesum phytochemical Cardenolide during molecular dynamics simulation.

https://doi.org/10.1371/journal.pone.0357096.g004

Ligand stability, evaluated as RMSD relative to the protein backbone, closely mirrored the protein RMSD profile across the trajectory. The lack of abrupt RMSD excursions or sustained increases confirms persistent ligand retention within the ATP-binding cleft, with no evidence of dissociation or pocket migration. Such coherence between protein and ligand RMSD is indicative of a robust and dynamically stable binding mode. Together, these results demonstrate that the EGFR–Cardenolide complex remains structurally stable under explicit-solvent conditions, validating that the predicted binding pose is maintained beyond static docking and persists throughout long-timescale molecular dynamics simulation.

Residue-Level Flexibility and Binding Pocket Dynamics (RMSF Analysis)

Residue-wise RMSF analysis (Fig 5) reveals that EGFR maintains a dynamically stable conformation upon binding the Adenium obesum phytochemical Cardenolide, with the majority of residues fluctuating below ~1.5 Å throughout the 100 ns simulation. Elevated flexibility is primarily confined to the N- and C-terminal regions, consistent with their solvent-exposed and functional peripheral nature. In contrast, residues directly involved in ligand binding exhibit markedly suppressed fluctuations, indicating ligand-induced stabilization of the kinase active site. This localized rigidity suggests effective conformational restraint of the binding pocket, a hallmark of productive EGFR inhibition. Moderate flexibility observed in adjacent loop regions likely reflects adaptive motion rather than destabilization, enabling conformational accommodation without compromising binding integrity. Notably, the RMSF profile qualitatively aligns with experimental B-factor trends, supporting the physiological relevance of the simulated dynamics. Collectively, these results demonstrate that the Cardenolide stabilizes EGFR through selective suppression of residue-level mobility at functionally critical regions, reinforcing its potential as a computationally prioritized candidate for targeted therapy in non-small-cell lung cancer.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Residue-level flexibility of EGFR in complex with the Adenium obesum phytochemical Cardenolide.

https://doi.org/10.1371/journal.pone.0357096.g005

Protein–ligand contact analysis

Protein–ligand contact profiling across the 100-ns MD simulation revealed a highly stable and cooperative interaction network between the Adenium obesum phytochemical Cardenolide and the EGFR kinase domain, indicating a well-defined and persistent binding mode essential for high-affinity inhibition. Throughout the trajectory, the ligand consistently engaged multiple residues spanning the hinge region, catalytic loop, and activation loop, forming a dense interaction architecture that remained preserved over time. The fractional contact histogram (Fig 6A) demonstrated that hydrophobic residues such as ALA719, PHE699, LEU820, VAL702, and MET742 made the largest contributions to ligand stabilization, with ALA719 and LEU820 displaying the highest fractional occupancies (>0.30). These residues constitute the EGFR hydrophobic core, and their extensive involvement suggests that the ligand embeds deeply within the ATP-binding cleft, exploiting van der Waals complementarity to restrict pocket fluctuations. Such persistent hydrophobic anchoring is a hallmark of clinically validated EGFR inhibitors and is typically associated with reduced dissociation frequency and strong binding enthalpy. In addition to hydrophobic interactions, the ligand formed highly stable hydrogen bonds with CYS751, THR830, ASP831 and PHE832, all of which showed notable fractional contact contributions (0.10–0.15). The presence of CYS751, a key catalytic-loop residue within the hydrogen-bonding network, is particularly significant as this interaction often governs the inhibitory mechanism in ATP-competitive EGFR inhibitors. The combined contribution of THR830 and ASP831 further strengthens the polar anchoring effect, ensuring that the ligand maintains its orientation even during subtle conformational breathing of the active site. The contact landscape also revealed frequent water-mediated interactions, especially with GLU738, ARG752, GLN767, PHE832, and THR830. Water bridges in EGFR are functionally important stabilizers, and their consistent formation suggests that the ligand efficiently integrates into the solvent shell of the pocket, allowing it to remain stably coordinated as local hydration layers reorganize during the simulation.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Protein–Ligand Contact Landscape During 100-ns MD Simulation.

(A) Fractional interaction plot showing hydrogen bonds, hydrophobic contacts, ionic interactions, water bridges, and halogen bonds formed between the ligand and EGFR residues across the trajectory. (B) Contact timeline map illustrating the temporal persistence of interactions across the simulation.

https://doi.org/10.1371/journal.pone.0357096.g006

Temporal interaction mapping across the entire 100-ns trajectory (Fig 6B) confirmed that the ligand maintained simultaneous contacts with 6–8 residues at nearly all timepoints, indicating a low probability of detachment and strong pocket residency. Several residues including PHE699, ALA719, CYS751, ARG752, THR830, and ASP831 showed dense, continuous contact patterns, reaffirming their roles as core stabilizing anchors. While hydrophobic residues displayed short intermittent bursts of contact, characteristic of flexible van der Waals interactions, polar residues preserved more continuous interactions, consistent with a rigid hydrogen-bond–based stabilization network. This mixed dynamic behavior rigid polar anchoring complemented by adaptive hydrophobic contacts is highly desirable for kinase inhibitor design, as it often correlates with high specificity and effective binding free energy. Mechanistically, the observed interaction patterns indicate that the phytochemical engages the EGFR ATP pocket in a manner analogous to known therapeutic EGFR inhibitors such as erlotinib and gefitinib. The strong hinge-region interactions with ALA719 and MET742, together with catalytic loop engagement at CYS751 and activation-loop involvement at THR830 and PHE832, highlight a binding mode that restricts ATP access and potentially disrupts kinase activation. The persistence and diversity of contacts, including the contribution of long-lived water bridges, further suggest that the ligand remains both geometrically and energetically favored within the active site. The combination of stable hydrophobic packing, well-positioned hydrogen bonding, and adaptive solvent-mediated interactions supports a robust inhibitory mechanism and aligns strongly with the low ligand RMSD, and favorable binding energy values associated with potent inhibitors.

Overall, the protein–ligand contact analysis provides compelling evidence that the Adenium obesum phytochemical Cardenolide is not only capable of stably occupying the EGFR active site, but also of establishing a sophisticated and persistent interaction network typically associated with highly potential kinase inhibitors. The long-lived contacts distributed across structural hot-spot residues, the continuous contact multiplicity throughout the trajectory, and the coordinated behavior of hydrophobic, polar, and water-bridged interactions collectively suggest that the Cardenolide possesses a strong inhibitory potential suitable for further optimization as a targeted therapeutic candidate for NSCLC.

Additionally, secondary structure element (SSE) analysis (S4 Fig in S1 File) demonstrated that approximately 46% of the protein structure was maintained as stable helices and β-strands throughout the simulation, with ~30% α-helical and ~16% β-sheet content. The temporal persistence of these elements confirms that ligand binding does not induce secondary structure destabilization or unfolding. From a mechanistic perspective, preservation of the kinase domain’s secondary structure is critical, as EGFR inhibitors must selectively restrain enzymatic activity without triggering non-physiological conformational collapse. The observed SSE stability therefore reinforces the structural compatibility of the Cardenolide with the EGFR binding landscape.

Moreover, ligand RMSF (S5 Fig in S1 File and torsional analyses (S6 Fig in S1 File) revealed limited internal flexibility, with only a small number of rotatable bonds exhibiting controlled conformational transitions. This restrained flexibility is advantageous from a drug design perspective, as excessive torsional freedom often incurs entropic penalties upon binding. The torsion profile demonstrated that the ligand predominantly occupied low-energy conformational states while bound to EGFR, indicating minimal conformational strain. Such behavior suggests that the ligand’s bioactive conformation is energetically favorable and does not require significant distortion to maintain binding a hallmark of efficient and selective inhibitors. Furthermore, stable radius of gyration (rGyr) and solvent-accessible surface area (SASA) values indicate that the ligand maintained a compact, well-buried conformation within the binding pocket, reducing solvent exposure and enhancing hydrophobic complementarity. These physicochemical trends collectively point toward favorable enthalpy–entropy compensation during binding. Also, Time-dependent analysis of ligand physicochemical Cardenolide properties (S7 Fig in S1 File) further corroborated binding stability. The radius of gyration remained essentially constant, indicating that the ligand neither collapsed nor excessively extended during the simulation. Solvent-accessible surface area (SASA) and polar surface area (PSA) exhibited only minor fluctuations, reflecting consistent burial of hydrophobic moieties and stable exposure of polar functional groups. Moreover, the limited presence of intramolecular hydrogen bonds suggests that ligand–protein interactions are favored over internal ligand stabilization, reinforcing the notion that binding is driven by productive intermolecular contacts rather than intramolecular constraints.

Taking together, the molecular dynamics simulation results provide compelling dynamic evidence that the selected Adenium obesum phytochemical Cardenolide forms a structurally stable, energetically favorable, and dynamically resilient complex with EGFR. The combined analyses of RMSD, RMSF, secondary structure preservation, interaction persistence, and ligand conformational behavior collectively validate the compound’s inhibitory potential at an atomistic level. These findings strongly justify further free energy calculations and experimental validation, positioning the identified phytochemical Cardenolide as a lead potential candidate for EGFR-targeted therapy against non-small-cell lung cancer.

Principal Component Analysis (PCA)

Principal component analysis (PCA) was performed to investigate the dominant collective motions and conformational dynamics of the protein–ligand complex throughout the 100 ns molecular dynamics simulation. The projection of the trajectory onto the first two principal components (PC1 and PC2) revealed that PC1 accounted for 33.1% of the total variance, whereas PC2 contributed 19.2%, together explaining 52.3% of the overall conformational motion observed during the simulation.

The PCA projection demonstrated (Fig 7) broad conformational sampling across the simulation trajectory, indicating that the complex explored multiple structural states during the simulation period. The trajectory initially occupied a localized region in conformational space and gradually expanded toward distinct regions along both PC1 and PC2 axes, suggesting progressive structural adaptation of the protein–ligand complex over time. The distribution of frames showed a continuous transition pattern rather than abrupt or highly scattered movements, indicating stable conformational evolution during the simulation. Several populated clusters were observed within the PCA landscape, reflecting the presence of metastable conformational states. These clustered regions imply that the complex adopted multiple energetically favorable conformations while maintaining structural integrity throughout the simulation. The separation between the starting and ending frames further suggests a conformational shift from the initial structure toward an adapted equilibrium state, likely associated with ligand accommodation within the binding pocket. Overall, the PCA results indicate that the complex underwent controlled conformational fluctuations while preserving dynamic stability across the 100 ns simulation. The absence of extreme dispersion or isolated outliers suggests that the system remained structurally coherent, supporting the stability of the ligand-bound complex during molecular dynamics simulation.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Principal Component Analysis (PCA).

https://doi.org/10.1371/journal.pone.0357096.g007

Electronic Reactivity and DFT-based descriptors

Density functional theory (DFT) calculations were performed to investigate the electronic characteristics and reactive behavior of the selected lead ligand Cardenolide and to complement the molecular docking and molecular dynamics findings at the quantum-chemical level. DFT calculations were performed on the selected ligand to obtain the optimized geometry and electronic descriptors (Table 8). The geometric optimization generated a stable molecular conformation, which was subsequently used for the calculation of electronic properties. Fig 8 illustrates the optimized molecular structure, frontier molecular orbital (HOMO–LUMO) distributions, and molecular electrostatic potential (MEP) map of the ligand obtained through DFT analysis. The frontier molecular orbital (FMO) analysis was conducted to determine the HOMO and LUMO energy levels and evaluate the electronic characteristics of the compound. The HOMO and LUMO energies were observed at −7.123 eV and −1.221 eV, respectively, resulting in an energy gap (ΔE) of 5.902 eV. The calculated ionization potential (I) and electron affinity (A) were 7.123 eV and 1.221 eV, respectively. Additionally, the chemical potential (μ), hardness (η), electronegativity (χ), softness (σ), and electrophilicity index (ω) were determined as −4.172 eV, 2.951 eV, 4.172 eV, 0.169, and 2.948, respectively. Overall, the DFT analysis provided complementary electronic-level insight into the interaction behavior of Cardenolide and supported its prioritization as a potential EGFR-binding phytochemical. Importantly, these quantum chemical descriptors were interpreted as supportive physicochemical indicators rather than standalone evidence of inhibitory activity.

Download:

• PNG
  larger image
• TIFF
  original image

Table 8. Density functional theory (DFT)-calculated quantum chemical descriptors of the selected lead ligand Cardenolide.

https://doi.org/10.1371/journal.pone.0357096.t008

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Optimized molecular structure, frontier molecular orbital (HOMO–LUMO) distributions, and molecular electrostatic potential (MEP) map of Cardenolide obtained through density functional theory (DFT) analysis.

https://doi.org/10.1371/journal.pone.0357096.g008

Evolutionary conservation and hotspot distribution in the EGFR kinase domain

Conservation mapping showed that the predicted ligand-binding pocket is located within a functionally constrained region of the EGFR kinase domain (S8 Fig in S1 File). Several residues identified as major interaction sites in the docking and molecular dynamics analyses were moderately to highly conserved, including Phe699 (8), Val702 (9), Ala719 (9), Lys721 (9), Met769 (7), Leu820 (9), Thr830 (8), Asp831 (9), and Phe832 (9) that are present in Table 9. These findings indicate that the principal ligand-contact residues are concentrated within an evolutionarily preserved segment of the kinase pocket, supporting the structural plausibility of the predicted binding mode. In addition, hotspot positions corresponding to major clinically relevant EGFR mutation sites were found within the same kinase region, further reinforcing the biological and translational relevance of the identified binding pocket. Overall, this conservation-guided structural analysis provides an additional layer of support for the EGFR-Cardenolide interaction model identified in the present study.

Download:

• PNG
  larger image
• TIFF
  original image

Table 9. Conservation grades of key EGFR kinase-domain residues involved in ligand binding and hotspot mapping.

https://doi.org/10.1371/journal.pone.0357096.t009

Clinical expression profile and pharmacogenomics context of EGFR in NSCLC

EGFR showed differential expression patterns between tumor and normal tissues in both LUAD and LUSC (Fig 9A). However, pathological stage analysis did not reveal marked stage-dependent variation in either cohort (Fig 9B-C), indicating that EGFR expression was not strongly associated with tumor stage in the present analysis. Drug-response screening further showed that EGFR expression was associated with distinct pharmacogenomic patterns across lung cancer cell lines (Fig 10A). Under the selected response metric, negative associations were observed for agents including afatinib, 993-D2, EphB4_9721, cetuximab, TANK_1366, trametinib, pelitinib, and RAF-9304, whereas positive associations were observed for compounds including BMN540215, C-75, A-484954, CD532, tricostatin A, parthenolide, sorafenib, PHA-665752, imatinib, and GNF-2. These results indicate that EGFR-linked drug-response patterns remain detectable in lung-derived cell-line datasets.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 9. Clinical expression profile of EGFR in non-small-cell lung cancer.

(A) Differential expression of EGFR in lung adenocarcinoma (LUAD) tumor and lung squamous cell carcinoma (LUSC) tumor and normal tissues. (B) Pathological stage-wise expression pattern of EGFR in LUAD. (C) Pathological stage-wise expression pattern of EGFR in LUSC.

https://doi.org/10.1371/journal.pone.0357096.g009

Download:

• PNG
  larger image
• TIFF
  original image

Fig 10. Pharmacogenomic and mutational landscape of EGFR in lung cancer.

(A) Correlation between EGFR expression and drug sensitivity across lung cancer cell lines. (B) Positional distribution of EGFR mutations across the receptor structure in a lung adenocarcinoma cohort, showing hotspot alterations within the kinase region.

https://doi.org/10.1371/journal.pone.0357096.g010

Mutational landscape of EGFR in lung adenocarcinoma

Mutation profiling showed that EGFR alterations were distributed across several structural regions of the receptor, with a clear concentration of clinically relevant events in the kinase-domain segment (Fig 10A, B). The mutation-domain plot highlighted a prominent L858R hotspot within the tyrosine kinase region, together with additional lower-frequency alterations positioned nearby. This pattern is consistent with the recognized importance of kinase-domain mutations in EGFR-driven lung adenocarcinoma and reinforces the biological relevance of targeting this region in the present study. Overall, the cBioPortal-based mutation analysis provides a concise genomic context for the EGFR-focused structural findings and supports the therapeutic significance of the kinase-domain binding pocket identified in this work.