Computational Evaluation of Fig-Derived Bioactive Compounds as COX-2 Inhibitors: Molecular Dynamics Reveals Desolvation-Driven Ranking Inversion Between Flavonoids and Furanocoumarins

Computational Evaluation of Fig-Derived Bioactive Compounds as COX-2 Inhibitors: Molecular Dynamics Reveals Desolvation-Driven Ranking Inversion Between Flavonoids and Furanocoumarins | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Computational Evaluation of Fig-Derived Bioactive Compounds as COX-2 Inhibitors: Molecular Dynamics Reveals Desolvation-Driven Ranking Inversion Between Flavonoids and Furanocoumarins Faycal Ferhat This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9155203/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Background. Figs ( Ficus carica ) have long been used in traditional medicine for anti-inflammatory effects, and they contain polyphenolic compounds with structural diversity across flavonoid and furanocoumarin scaffolds. Whether these compounds directly inhibit COX-2 — a key target in inflammation — has not been systematically tested with molecular dynamics and binding free-energy calculations. Methods. We applied a three-tier computational screening workflow to six fig-derived compounds (flavonoids: luteolin, quercetin; furanocoumarins: bergapten, psoralen; secoiridoid derivative: elenolic acid; phenol: hydroxytyrosol): (1) AutoDock Vina docking with smina/Vinardo scoring consensus; (2) 5 ns all-atom MD simulation (GROMACS 2026, CHARMM36 + GAFF2, 4 independent replicates per compound); (3) MM-GBSA binding free-energy calculation (igb = 2, saltcon = 0.150 M, gmx_MMPBSA v1.6.4) benchmarked against celecoxib. Structural reliability of the COX-2 template was verified by AlphaFold2 cross-validation (global RMSD 0.38 Å, pocket RMSD 0.25 Å vs AF-P35354-F1 raw prediction). Oleocanthal from the companion study served as an upper-bound reference. Two glycoside compounds (oleuropein, rutin) underwent docking only due to force-field conversion limitations. Results. All six tested fig-derived compounds bound less favourably than celecoxib in this computational model (ΔΔG range: +18.6 to + 28.5 kcal mol⁻¹, Welch t-test p ≤ 8.5×10⁻⁷ for all comparisons), suggesting that none of the compounds screened here approaches celecoxib-level COX-2 active-site affinity under these conditions. Within the test panel, furanocoumarins ranked above flavonoids: bergapten (ΔΔG = + 18.56 vs celecoxib; absolute ΔG = − 20.32 ± 0.97 kcal mol⁻¹) and psoralen (ΔΔG = + 19.93; −18.95 ± 1.16) outranked luteolin (− 16.54 ± 0.88) and quercetin (− 14.36 ± 1.53) despite weaker Vina docking scores — a desolvation-driven ranking inversion. Elenolic acid (− 14.88 ± 1.02) and hydroxytyrosol (− 10.34 ± 0.56) were the weakest binders. Although bergapten and psoralen show the strongest fig-derived binding in this model, both are known CYP3A4 inhibitors and phototoxic agents, raising concerns that would limit their direct therapeutic use. Conclusions. This computational screening phase indicates that none of the six tested fig-derived compounds approaches celecoxib potency at the COX-2 active site in this model, suggesting that experimental focus for fig bioactives may be better directed toward alternative anti-inflammatory mechanisms. The desolvation-driven ranking inversion between furanocoumarins and flavonoids is a methodologically important finding: Vina docking scores would have predicted flavonoids as the stronger binders, while physics-based MM-GBSA correctly captures the desolvation penalty that reverses this ranking — demonstrating why single-tier docking is insufficient for polar natural products. All raw data are deposited on Zenodo for full reproducibility. Biological sciences/Computational biology and bioinformatics Biological sciences/Drug discovery COX-2 figs molecular dynamics MM-GBSA binding free energy flavonoids furanocoumarins natural products 1. Introduction The fig ( Ficus carica ) is among the oldest cultivated plants, with archaeological evidence of human consumption predating cereal agriculture [12]. Fig extracts have demonstrated anti-inflammatory activity in multiple in vitro and in vivo models, and fig-derived compounds span several pharmacologically relevant chemical classes — flavonoids, furanocoumarins, secoiridoid derivatives, and simple phenols [11,12]. However, the molecular basis for these effects at specific drug targets has not been systematically evaluated using physics-based methods. Cyclooxygenase-2 (COX-2) catalyses the first committed step in prostaglandin synthesis, converting arachidonic acid to prostaglandin endoperoxide H₂ (PGH₂), the immediate precursor of pro-inflammatory prostanoids. Unlike COX-1, which is constitutively expressed, COX-2 induction by inflammatory stimuli makes it a validated pharmaceutical target [1,2,3]. NSAIDs block COX enzymes broadly, but selective COX-2 inhibitors carry cardiovascular risk (rofecoxib, valdecoxib withdrawn from market) and non-selective NSAIDs cause gastrointestinal injury [ 4 , 5 ]. This safety gap has driven sustained interest in plant-derived COX-2 inhibitors with potentially better tolerability profiles [ 6 , 7 ]. Figs contain a chemically diverse set of bioactive compounds: flavonoids (luteolin, quercetin), furanocoumarins (psoralen, bergapten), a secoiridoid derivative (elenolic acid), and a simple phenol (hydroxytyrosol) [11,12]. Luteolin has been docked repeatedly against COX-2, with published Vina scores clustering around − 9.5 to − 10.2 kcal mol⁻¹ [13,15,16]. Quercetin, abundant in fig pulp and leaves, shares the flavonoid scaffold with luteolin but differs by hydroxyl substitution pattern. Psoralen and bergapten are furanocoumarins with documented inhibitory activity against CYP450 enzymes and known phototoxicity [13,14]. However, published studies of fig compounds against COX-2 rely entirely on docking; to our knowledge, no prior work has applied replicated molecular dynamics simulation with explicit MM-GBSA binding free-energy calculation to a systematic fig-compound panel. The COX-2 cyclooxygenase channel is a hydrophobic pocket flanked by aromatic residues (Tyr385, Trp387 in canonical numbering) and selective for inhibitors with appropriate size and hydrophobic complementarity. The channel floor is anchored by Val523, while selectivity versus COX-1 derives from an Ile side pocket at the Val523 position [19,20]. Binding energetics split among van der Waals contacts, hydrogen bonding, and desolvation costs; distinguishing these contributions requires dynamic simulation with explicit solvation treatment rather than rigid-body docking alone. MM-GBSA methods combine molecular mechanics force fields with implicit Born solvation to estimate binding free energies while accounting for protein flexibility, ligand desolvation, and conformational sampling [21,22]. Multi-replicate MD simulations with distinct initial velocity seeds produce reproducible free-energy estimates with quantifiable standard error, allowing statistical comparisons [23]. This study constitutes a computational screening phase in the evaluation of fig bioactives as COX-2 inhibitors. We apply a hierarchical three-tier screening workflow: (1) Vina docking as a rapid pose filter; (2) 5 ns all-atom MD (4 replicates each) for dynamic validation; (3) MM-GBSA binding free-energy calculation for quantitative ranking. This workflow is specifically designed to identify which compounds, if any, warrant experimental follow-up (in vitro COX-2 inhibition assays, cell-based prostaglandin suppression) versus which can be deprioritised at the computational stage. Results are benchmarked against celecoxib and oleocanthal. Drug-likeness and predicted ADMET profiles contextualise affinity within therapeutic feasibility. Two glycoside compounds (oleuropein, rutin) underwent docking only due to force-field limitations. 2. Methods 2.1 Protein Preparation The crystal structure of human COX-2 (PDB: 5KIR, resolution 2.4 Å) was obtained from the RCSB Protein Data Bank. The structure was processed using OpenMM: water molecules and co-crystallized ligands were removed, missing hydrogen atoms were added at physiological pH 7.4, and residues were protonated using the ff14SB force field protonation states. The processed structure (5KIR_fixed.pdb) was used for all subsequent calculations. OpenMM renumbers residues sequentially from 1, so the numbering in 5KIR_fixed.pdb does not match canonical COX-2 literature numbering. The mapping was established by sequential Cα alignment between 5KIR_fixed.pdb and the original RCSB 5KIR structure (5KIR_original.pdb; 551/551 chain A residues matched by residue name). All residue numbers used in this manuscript refer to the OpenMM-renumbered structure; canonical equivalents for all active-site residues cited in Results are listed in Table S1. AlphaFold2 structural validation and simulation software cross-check. The COX-2 crystal structure (5KIR) was validated against the AlphaFold2 monomer prediction (UniProt P35354, AF-P35354-F1-model_v6.pdb, downloaded directly from the EMBL-EBI AlphaFold Database). AlphaFold2 generates structures from sequence alone — no force field, no experimental target data — making it a fully independent reference. Global Cα RMSD across 551 matched residues was 0.38 Å; pocket Cα RMSD across the 10 active-site residues was 0.25 Å (locally aligned; 0.36 Å post-global alignment). All pLDDT values were read directly from the B-factor column of the raw AF2 PDB file and verified using a deposited script (Zenodo, `alphafold/validate_af2_vs_5kir.py`); all 10 pocket residues carry pLDDT ≥ 94.6 (mean 97.0). This validates the 5KIR template (crystal vs AF2 pocket RMSD 0.251 Å) and simultaneously cross-validates the simulation software. The raw MD average structure (last 2 ns, 400 frames, `avg_structure_last2ns.pdb`) was compared directly against the raw AF2 PDB by extracting Cα coordinates for the 10 pocket residues from both files and computing post-Kabsch distances. Result: MD trajectory vs AF2 pocket RMSD = 0.307 Å , per-residue range 0.204–0.432 Å. CHARMM36+GAFF2 dynamics in explicit solvent and AF2 machine learning, operating on entirely different physical principles, converge on the same binding pocket geometry to sub-Ångström precision. Raw coordinate table: Zenodo `alphafold/MD_vs_AF2_raw_comparison.txt`. 2.2 Ligand Preparation Compounds were prepared from canonical SMILES strings obtained from PubChem. Three-dimensional coordinates were generated using RDKit (v2023.09) with the ETKDGv3 algorithm followed by MMFF94 geometry optimization. Partial charges and GAFF2 atom types were assigned using Antechamber (AmberTools25) with the AM1-BCC charge model (flag: -c bcc -at gaff2). GROMACS-compatible topologies were generated using ACPYPE (v2022.7). For Quercetin, SDF input was used for Antechamber to preserve bond order information and ensure correct assignment of aromatic carbon atom types (ca rather than c3). The compound panel comprised six phytochemicals found in Ficus carica (fig): Luteolin, Psoralen, Bergapten, Elenolic acid, Hydroxytyrosol, and Quercetin. Quercetin was included on the basis of its documented occurrence in fig pulp and leaves and its structural relatedness to Luteolin (flavone/flavonol scaffold), enabling direct SAR comparison within the flavonoid series. Two additional glycoside compounds (Oleuropein and Rutin) were prepared using the same protocol. Celecoxib served as a reference inhibitor standard. 2.3 Molecular Docking Molecular docking was performed using AutoDock Vina 1.2.6. The receptor PDBQT file was prepared from 5KIR_fixed.pdb using OpenBabel with Gasteiger partial charges. Ligand PDBQT files were prepared using OpenBabel. The docking search box was centered at coordinates (24.2, 1.6, 35.4) Å—derived from the center of mass of oleocanthal in the energy-minimized (EM) complex structure—with dimensions of 25 × 25 × 25 Å, encompassing the full COX-2 cyclooxygenase channel. Docking was performed with exhaustiveness = 32 and 9 binding modes generated. The best-scoring pose was selected for MD simulation. Pose validation was performed by computing heavy-atom contacts (≤ 4.5 Å) against 14 experimentally identified COX-2 binding-site residues; a pose was considered bound when ≥ 2 of 6 designated core residues (PHE486, SER321, VAL491, ALA495, TRP355, TYR353) were contacted. Protein–ligand interaction fingerprinting was performed using MDAnalysis with a 4.5 Å distance cutoff; ProLIF was evaluated but is not compatible with the ARM64 build of RDKit due to a fatal assert in the bond-order inference path. 2.4 System Setup and Energy Minimization Protein–ligand complexes were assembled as follows. The ACPYPE-generated ligand GRO file was translated to the docked pose center of mass. The ligand was merged with the protein GRO file (from pdb2gmx with CHARMM36-jul2022 force field, TIP3P water model). The combined system was placed in a dodecahedral periodic box with a minimum solute-to-box-face distance of 1.2 nm using gmx editconf. The box was solvated with TIP3P water using gmx solvate, and charge-neutralized to 0.15 M NaCl using gmx genion. All GAFF2 LIG atom type parameters were inlined in topol.top to prevent duplicate atomtype conflicts, with the ligand moleculetype block included as a separate ITP file. Energy minimization was performed using the steepest descent algorithm (emtol = 100 kJ mol⁻¹ nm⁻¹, nsteps = 50,000) with CHARMM36-jul2022 for the protein and GAFF2 for the ligand (GROMACS 2026, GPU-accelerated with CUDA on a single A100 GPU). 2.5 Molecular Dynamics Simulations MD simulations were conducted using GROMACS 2026. For each compound, the following protocol was applied: NVT equilibration: 100 ps, dt = 1 fs, V-rescale thermostat (τ = 0.1 ps, T = 300 K), protein backbone and ligand position-restrained (k = 1000 kJ mol⁻¹ nm⁻²), H-bond constraints (LINCS), PME electrostatics with rcoulomb = rvdw = 1.2 nm. NPT equilibration: 100 ps, dt = 2 fs, Parrinello–Rahman barostat (τ_p = 2.0 ps, P = 1 bar), same temperature coupling, position restraints maintained. Production MD: 5 ns × 4 independent replicates per compound, dt = 1 fs (chosen to ensure accurate force evaluation for flexible ligand torsions, at the cost of higher computation relative to 2 fs), all-bonds constrained, NPT ensemble at 300 K, 1 bar (Parrinello–Rahman). Three replicates used gen_vel=yes with seeds 1234, 42, and 9999 to assign independent initial velocities at the start of production; the main run used gen_vel=no, continuing velocities from the NPT endpoint. All four replicates started from the same NPT-equilibrated coordinates, ensuring identical initial structures while sampling distinct regions of phase space. Coordinates were saved every 5 ps (nstxout-compressed = 5000, dt = 1 fs; 1,001 frames per 5 ns replicate) for all compounds. Trajectory PBC correction was applied using gmx trjconv (-pbc mol, centering on Protein+Ligand group) prior to MM-GBSA analysis. 2.6 Binding Free Energy Calculations Binding free energies were calculated using the MM-GBSA method implemented in gmx_MMPBSA (v1.6.4) with sander (AmberTools25). For each replicate trajectory, frames were extracted at uniform intervals using gmx_MMPBSA parameters startframe=1, endframe=5000, interval=50. Because interval refers to frame indices (not time), this yields 21 frames per replicate for compounds with 1,001-frame trajectories (every 50th frame = every 250 ps). The MM-GBSA input parameters were: igb = 2, saltcon = 0.150 M, forcefields = CHARMM36,GAFF2. The final ΔG_bind for each compound was reported as the mean ± standard deviation across the four replicates, with SEM = SD/√4. Statistical comparisons against the celecoxib reference were performed using the Welch two-sample t-test (two-tailed). Per-residue binding free energy decomposition (idecomp=2, 1–4 interactions included) was performed using print_res="within 8" for all test compounds across all four replicates. Decomposition coverage (decomposed residue sum / system-level ΔG) is reported alongside per-residue data to quantify the fraction of binding energy accounted for by residues within the cutoff shell. 2.7 Reference Compounds Celecoxib (ΔG = −38.875 ± 1.199 kcal mol⁻¹) and Oleocanthal (ΔG = −61.765 ± 1.477 kcal mol⁻¹, from companion paper [2]) were simulated using the identical protocol (N = 4 replicates each, igb = 2, saltcon = 0.150 M) and serve as in-house benchmarks. All reference values were obtained under the same computational conditions as the test compounds. 2.8 AI Writing Assistance Disclosure Large language model (LLM) assistance was used during the preparation of this manuscript for writing support, editing, and structural guidance. The LLM used was a locally deployed model running within the NVIDIA Ecosystem. All scientific content, computational results, raw data, statistical analyses, and conclusions are the sole work of the author and were independently verified against raw output files. LLM assistance was not used for data generation, analysis, or interpretation of results. 3. Results 3.1 Molecular Docking and Pose Validation All eight compounds (six test + two glycosides) were successfully docked into the COX-2 cyclooxygenase channel. Binding pose validation against 14 active-site residues confirmed that seven of eight compounds occupied the primary active site (≥ 2 core residue contacts; Table 1). Rutin, despite registering all six core residue contacts, exhibited steric clashes (< 1.5 Å) at SER321 and ALA495, consistent with its large disaccharide moiety being geometrically incompatible with the deep hydrophobic channel; its Vina score of −2.54 kcal mol⁻¹ was also substantially weaker than the remaining compounds. Luteolin achieved the strongest Vina affinity (−9.68 kcal mol⁻¹) and formed contacts with all six core residues, including eight hydrogen bonds (SER498, TYR353, TYR323, HIS57, PHE486, SER321, ILE485). The flavone scaffold showed aromatic stacking with TRP355 and PHE486. Bergapten and Psoralen (furanocoumarins, −7.97 kcal mol⁻¹ each) were well-accommodated in the channel, with Bergapten forming four hydrogen bonds to SER321, SER498, MET490, and VAL317. Elenolic acid (−6.35 kcal mol⁻¹) contacted all six core residues with five hydrogen bonds. Oleuropein (−6.31 kcal mol⁻¹) and Hydroxytyrosol (−6.23 kcal mol⁻¹) occupied the active site with ≥ 5 core residue contacts. Quercetin (−9.60 kcal mol⁻¹) contacted all six core residues and formed eight hydrogen bonds (PHE486, SER321, TYR353, TYR323, HIS57, SER498), comparable to Luteolin in binding complementarity despite lacking the 5-OH group. Table 1. Docking scores and active-site contact summary for all eight compounds. Compound Vina (kcal/mol) In site Contacts H-bonds Core residues hit Luteolin −9.68 YES 13 8 PHE486, SER321, VAL491, ALA495, TRP355, TYR353 Quercetin −9.60 YES 14 8 PHE486, SER321, VAL491, ALA495, TRP355, TYR353 Psoralen −7.97 YES 9 2 PHE486, SER321, VAL491, ALA495 Bergapten −7.97 YES 11 4 PHE486, SER321, VAL491, ALA495, TRP355, TYR353 Elenolic acid −6.35 YES 12 5 PHE486, SER321, VAL491, ALA495, TRP355, TYR353 Oleuropein −6.31 YES 9 0 PHE486, SER321, VAL491, ALA495, TRP355 Hydroxytyrosol −6.23 YES 12 5 PHE486, SER321, VAL491, ALA495, TRP355 Rutin −2.54 YES 14 11 PHE486, SER321, VAL491, ALA495, TRP355, TYR353 Rutin contacts all core residues but exhibits steric clashes (< 1.5 Å) at SER321/ALA495; poor Vina score consistent with size mismatch. Scoring-function consensus validation. To assess binding-mode robustness, all seven test compounds were independently re-docked using smina with Vinardo scoring (exhaustiveness=32, same receptor and box). Rank-1 poses were compared to the original AutoDock Vina poses by heavy-atom RMSD (Table 1b). Table 1b. Vina– smina (Vinardo) pose consensus. Compound Vina (kcal/mol) Smina (kcal/mol) RMSD (Å) Verdict Luteolin −9.68 −10.4 0.27 AGREE Quercetin −9.60 −10.7 0.38 AGREE Oleuropein −6.31 −4.3 1.03 AGREE Bergapten −7.97 −7.6 5.94 Partial overlap Psoralen −7.97 −7.4 16.6 Different site Elenolic acid −6.35 −5.5 13.8 Different site Hydroxytyrosol −6.23 −6.6 14.8 Different site Rutin† −2.54 N/A N/A Excluded †Rutin excluded from smina rescoring: the systematic steric clash at SER321/ALA495 precludes any productive binding pose in the deep hydrophobic channel; only the Vina reference score is reported. Three compounds (luteolin, quercetin, oleuropein) achieved strong Vina–smina consensus (RMSD < 2 Å), confirming scoring-function-independent binding modes. Bergapten showed partial overlap (5.9 Å RMSD), suggesting pose ambiguity at the channel entrance. Three smaller, weaker compounds (psoralen, elenolic acid, hydroxytyrosol) diverged to an alternative surface site under Vinardo scoring (RMSD 13–17 Å). Per-residue analysis showed that all three Vina poses retain 6–7 channel residues in their top-8 contacts, indicating the disagreement reflects scoring-function sensitivity for marginally-bound small molecules rather than incorrect docking. Vina scores differed from Vinardo by less than 1 kcal mol⁻¹ for these three compounds, affirming marginal binding status at both sites. MM-GBSA calculations use the Vina poses, providing explicit dynamic and solvation corrections for affinity ranking. 3.2 MD Simulation Stability Protein backbone RMSD stabilized rapidly for all compounds. Across all trajectories (main + replicates), mean backbone RMSD ranged from 0.9 to 1.3 Å (maxima 1.2–1.7 Å), showing stable protein structure throughout (Table 3). Ligand RMSD analysis used two reference frames. The main run measured RMSD relative to the docked pose (quantifying drift from initial binding). Replicates measured RMSD relative to their own first frame, because the equilibrated ligand position after 5 ns main production differs 0.4–1.4 nm from the original dock (an equilibration feature, not compound-specific). All simulations maintained protein–ligand contact (mean minimum distance 0.20 nm). In the main run, hydroxytyrosol (mean 2.5 Å) and rutin (mean 1.9 Å) showed the lowest ligand RMSD. However, rutin's stability reflects steric confinement of its large disaccharide (MW 610 Da) rather than productive binding; this aligns with its poor Vina score (−2.54 kcal mol⁻¹), steric clashes at SER321/ALA495, and strongly positive MM-GBSA ΔG (Section 3.3). Psoralen showed the largest displacement (mean 5.3 Å), consistent with its flat, compact scaffold providing weak steric fit in the deep hydrophobic channel. Replicates showed uniformly lower ligand RMSD (mean 0.8–2.2 Å), as expected from starting at the equilibrated pose. Table 3. MD stability summary (main production + replicates, 5 ns each). Compound Run BB mean (Å) BB max (Å) Lig mean (Å) Lig max (Å) Lig ref Hydroxytyrosol main 1.3 1.7 2.5 4.2 TPR s1234 1.1 1.5 1.4 3.4 self s42 1.2 1.6 1.1 3.5 self Psoralen main 1.2 1.5 5.3 6.5 TPR s1234 1.1 1.5 1.9 4.0 self s42 0.9 1.2 2.2 4.0 self Bergapten main 1.2 1.4 3.4 5.8 TPR s1234 1.1 1.5 1.1 4.0 self s42 1.2 1.6 1.6 3.5 self Elenolic acid main 1.3 1.7 4.6 5.1 TPR s1234 1.0 1.4 0.9 2.1 self s42 1.0 1.3 1.0 2.7 self Luteolin main 1.2 1.7 3.6 5.6 TPR s1234 1.0 1.2 1.2 2.3 self s42 1.1 1.6 1.3 2.9 self Rutin main 1.3 1.6 1.9 2.8 TPR s1234 1.0 1.4 1.0 1.9 self s42 1.1 1.5 0.9 1.6 self s9999 1.1 1.4 0.8 1.5 self BB RMSD: backbone, least-squares fit to backbone. Lig RMSD: fit to backbone, computed on ligand heavy atoms. "TPR" = docked-pose reference; "self" = frame 0 of PBC-corrected replicate trajectory (self-referenced, see Methods 2.5). s9999 RMSD data for non-Rutin test compounds not shown (all confirmed stable; BB RMSD < 1.6 Å). Oleuropein omitted (ligand-unstable). 3.3 Binding Free Energies (MM-GBSA) MM-GBSA binding free energies were calculated for all test compounds using gmx_MMPBSA v1.6.4 (igb=2, saltcon=0.150 M, CHARMM36+GAFF2 force fields, 300 K). Each compound was simulated across 4 independent 5 ns production replicates, with 21 frames sampled per replicate (interval=50 over 1001 trajectory frames). Results are summarised in Tables 2b (test compounds, N=4) alongside reference celecoxib data for benchmarking. Two compounds were excluded from the binding free energy analysis. Oleuropein and Rutin — both glycosides containing ≥ 11 GAFF2 c6 (sp3 ring carbon) atom types — yielded strongly positive ΔG values (> +300 and > +550 kcal mol⁻¹, respectively) across all four replicates despite confirmed ligand binding (minimum protein–ligand distance 1.5–2.0 Å throughout all trajectories). The anomalous energies are attributable to van der Waals cross-term incompatibility in the GROMACS→AMBER topology conversion employed by gmx_MMPBSA when mixing CHARMM36 (protein) and GAFF2 (ligand) force fields for large glycoside ligands; all covalent bond distances in the extracted ligand coordinates were verified intact (max 2.03 Å for Oleuropein, 2.12 Å for Rutin). Both compounds remain in the docking analysis (Section 3.1) and are discussed further in Section 4. Table 2b. MM-GBSA binding free energies — test compounds (N=4). Compound Rep_main Rep_s1234 Rep_s42 Rep_s9999 Mean ΔG ±SD SEM ΔVDW ΔEEL ΔEGB ΔESURF ΔΔG vs Cel Welch t p Bergapten‡ −19.48 −21.09 −19.48 −21.23 −20.32 0.97 0.49 −29.51 −6.27 +19.40 −3.94 +18.55 +24.05 <0.001 Psoralen‡ −17.99 −19.87 −17.89 −20.03 −18.95 1.16 0.58 −26.12 −4.93 +15.77 −3.67 +19.93 +23.86 <0.001 Luteolin −16.12 −17.51 −15.54 −16.97 −16.54 0.88 0.44 −34.31 −2.13 +25.13 −5.21 +22.34 +30.11 <0.001 Quercetin −15.05 −15.32 −14.99 −12.07 −14.36 1.53 0.77 −33.70 +2.48 +21.81 −4.96 +24.52 +25.21 <0.001 Elenolic acid −13.52 −14.81 −15.23 −15.94 −14.88 1.02 0.51 −25.38 −9.20 +23.30 −3.61 +24.00 +30.54 <0.001 Hydroxytyrosol −10.16 −9.63 −10.87 −10.71 −10.34 0.56 0.28 −22.33 −9.62 +25.29 −3.69 +28.53 +43.07 <0.001 All values in kcal mol⁻¹. ‡Furanocoumarin — CYP450 inhibitor and phototoxic; see Discussion. Welch two-sample t-test (two-tailed) vs celecoxib. Reference celecoxib: ΔG = −38.88 ± 1.20 kcal mol⁻¹, N=4. Energy components (averaged across all replicates): ΔVDW = van der Waals, ΔEEL = Coulombic, ΔEGB = Generalized Born solvation, ΔESURF = non-polar solvation. The MM-GBSA ranking was: Bergapten > Psoralen > Luteolin > Elenolic acid ≈ Quercetin > Hydroxytyrosol. All test compounds bound significantly less favourably than celecoxib (Welch t-test, all p ≤ 8.5×10⁻⁷). Affinity gaps ranged from +18.6 to +28.5 kcal mol⁻¹ versus celecoxib. All six fig compounds were weaker than oleocanthal (−61.77 ± 1.48 kcal mol⁻¹, companion paper [2]), with deficits of 41.5–51.4 kcal mol⁻¹. Quercetin (−14.36 ± 1.53) and luteolin (−16.54 ± 0.88) differed by 2.2 kcal mol⁻¹, consistent with their single-group difference (3-OH on C-ring). The quercetin s9999 replicate (−12.07 kcal mol⁻¹) was substantially weaker than the other three (−15.05 to −15.32), widening the SD from 0.19 (N=2) to 1.53 (N=4). This outlier reflects atypical electrostatic behavior (ΔEEL = +14.57 vs −0.6 to −2.6 for other replicates), suggesting that quercetin's additional 3-OH can adopt a conformation where desolvation cost exceeds hydrogen-bonding benefit. The Vina→MM-GBSA ranking inverted between furanocoumarins and flavonoids: bergapten and psoralen (Vina −7.97 kcal mol⁻¹ each) outranked luteolin (−9.68) and quercetin (−9.60) in MM-GBSA. The flavonoid hydroxyl groups incur large ΔEGB penalties (+23–25 kcal mol⁻¹), while the hydrophobic furanocoumarins pay less (+16–20 kcal mol⁻¹). This desolvation advantage offsets the furanocoumarins' weaker gas-phase interactions. 3.4 Per-Residue Energy Decomposition Per-residue binding contributions were computed as the difference between Complex and Receptor energies for each protein residue (idecomp=2, pairwise decomposition with 1-4 interactions folded into VDW and EEL terms, dec_verbose=3, full residue totals). Table 4 reports the top contributing residues for each compound, ranked by ΔTotal (Complex − Receptor, backbone + sidechain combined). All six test compounds engaged the same core channel residues: ALA495, LEU320, VAL491, VAL317, SER321, MET490, SER498, GLY494, and TYR353. Each compound buried 7–10 of its top-10 contributing residues within the active site, confirming consistent channel occupancy across all scaffolds. Bergapten and psoralen (furanocoumarins) engaged 8 pocket residues in their top-10, with ALA495 and LEU320 as primary anchors (bergapten: −1.78, −1.57; psoralen: −1.59, −1.51). Bergapten's methoxy group strengthened the SER498 contact (−0.87 vs −0.26), accounting for the 1.4 kcal mol⁻¹ MM-GBSA difference between them. Luteolin and quercetin (flavonoids) engaged 9–10 pocket residues, led by ALA495 and LEU320/VAL317 (luteolin: −1.45, −1.31; quercetin: −0.53, −0.43, −0.37 across three residues). Quercetin showed smaller per-residue magnitudes and a more diffuse footprint; large ΔEGB desolvation penalties (Section 3.3) result in weaker binding despite similar residue contacts. Elenolic acid (max VAL491 −1.37) and hydroxytyrosol (max ALA495 −1.52) showed the weakest per-residue profiles, consistent with their position at the weak-binding end of the ranking. Table 4. Top-5 per-residue binding contributions (ΔTotal, kcal mol ⁻ ¹) for test compounds. Compound Res-1 (ΔTotal) Res-2 (ΔTotal) Res-3 (ΔTotal) Res-4 (ΔTotal) Res-5 (ΔTotal) Pocket in top-10 Bergapten ALA495 (−1.78) VAL317 (−1.57) LEU320 (−1.57) VAL491 (−1.26) SER498 (−0.87) 8/10 Psoralen ALA495 (−1.59) LEU320 (−1.51) SER321 (−1.20) VAL491 (−0.94) VAL317 (−0.91) 8/10 Quercetin VAL317 (−0.53) LEU320 (−0.43) VAL491 (−0.37) TYR316 (−0.37) ALA495 (−0.30) 9/10 Luteolin ALA495 (−1.45) LEU320 (−1.31) VAL317 (−1.29) SER321 (−0.92) PHE349 (−0.79) 10/10 Elenolic acid VAL491 (−1.37) GLY494 (−1.09) LEU320 (−1.01) ALA495 (−0.86) VAL317 (−0.76) 10/10 Hydroxytyrosol ALA495 (−1.52) VAL317 (−0.88) SER321 (−0.77) LEU320 (−0.77) GLY494 (−0.75) 7/10 Values are full residue ΔTotal (backbone + sidechain, idecomp=2, dec_verbose=3). All compounds: N=4 replicates, per-residue values are mean across all four replicates. Quercetin values extracted from FINAL_DECOMP_MMPBSA.dat Delta section (N=4 confirmed: main, s1234, s42, s9999); replicate range VAL317 −0.44 to −0.65, CV < 20%. 3.5 Comparison with Reference Inhibitor and Positive Control Test compounds were compared against celecoxib (−38.88 ± 1.20, N=4, reference standard) and oleocanthal (−61.77 ± 1.48, companion paper [2], positive control). No tested fig-derived compound approached celecoxib affinity in this computational model. Luteolin, the strongest fig binder at −16.54 ± 0.88, was 22.3 kcal mol⁻¹ weaker (Welch t = 30.10, p = 2.6×10⁻⁷). Quercetin (−14.36 ± 1.53) was 2.2 kcal mol⁻¹ below luteolin, with one weaker replicate (s9999, −12.07) driving higher variance. Elenolic acid (−14.88 ± 1.02) and hydroxytyrosol (−10.34 ± 0.56) ranked lowest. Large ΔEGB desolvation penalties for flavonoid hydroxyl groups (Section 3.3) limit their MM-GBSA performance. Bergapten (−20.32 ± 0.97) and psoralen (−18.95 ± 1.16) ranked first and second overall, exceeding the flavonoids despite weaker docking scores. However, both are potent CYP450 inhibitors and phototoxic agents, raising significant concerns for therapeutic development (elaborated in Section 4.5). In summary, no tested fig-derived compound matched celecoxib under these computational conditions. The furanocoumarins showed the strongest fig-derived binding but carry documented CYP450 and phototoxicity liabilities. The flavonoids showed moderate-to-weak binding, consistent with anti-inflammatory activity operating through mechanisms beyond direct COX-2 active-site inhibition. All six fig compounds were substantially weaker than oleocanthal, which remains the strongest predicted COX-2 binder across both compound sets. 3.6 Drug-Likeness and ADMET Profiles Physicochemical properties and ADMET profiles were computed using RDKit and standard Lipinski filters (Table 5). Six of eight compounds passed Lipinski's Rule of Five with zero violations, including bergapten and psoralen (strongest binders) and all flavonoids. Oleuropein failed with three violations (HBD > 5, HBA > 10, TPSA > 140 Ų), and rutin failed with four (MW > 500, HBD > 5, HBA > 10, TPSA > 140 Ų), typical of glycoside natural products. Bergapten (MW 220) and psoralen (MW 190) achieved excellent drug-likeness scores (QED 0.68 and 0.63, highest in the panel). Despite favorable physicochemistry, their CYP3A4 inhibition and phototoxicity preclude therapeutic use without scaffold redesign (Section 4.5). Hydroxytyrosol (MW 154, LogP 0.6) and elenolic acid (MW 294, LogP 1.7) have favorable physicochemistry but weak COX-2 binding. Luteolin (MW 286, LogP 2.28) and quercetin (MW 302, LogP 1.99) satisfy Lipinski criteria and show reasonable bioavailability predictions, but moderate-to-weak COX-2 binding (−16.54 and −14.36 kcal mol⁻¹). Table 5. Drug-likeness and predicted ADMET properties (test compounds). Compound MW LogP HBD HBA TPSA (Ų) Rot Lipinski Bioavail. BBB QED Bergapten 220.18 1.53 0 5 57.9 1 PASS HIGH YES 0.682 Psoralen 190.15 1.52 0 4 48.7 0 PASS HIGH YES 0.632 Luteolin 286.24 2.28 4 6 111.1 1 PASS HIGH NO 0.511 Quercetin 302.24 1.99 5 7 131.4 1 PASS HIGH NO 0.434 Elenolic acid 294.30 1.67 0 6 78.9 6 PASS HIGH YES 0.421 Hydroxytyrosol 154.16 0.63 3 3 60.7 2 PASS HIGH YES 0.547 Oleuropein 468.46 −0.75 6 11 183.2 7 FAIL LOW NO 0.128 Rutin 610.52 −1.69 10 16 269.4 6 FAIL LOW NO 0.140 QED = quantitative estimate of drug-likeness (Bickerton et al. 2012). Excluded from MM-GBSA analysis (Section 3.3). 4. Discussion 4.1 Scoring-Function Sensitivity and Pose Validation Across Scaffold Classes The smina/Vinardo consensus (Table 1b) stratified compounds by polarity. The three most polar (luteolin, quercetin, oleuropein) agreed across scoring functions (RMSD < 2 Å), while three hydrophobic compounds (psoralen, elenolic acid, hydroxytyrosol) diverged to an alternative site (RMSD 13–17 Å). Bergapten showed intermediate disagreement (5.9 Å). This divergence does not invalidate the Vina poses. Per-residue contact analysis confirmed that all three "disagreement" compounds retain 6–7 of their top-8 contacts within the COX-2 channel residue set (LEU320, GLY494, ALA495, SER321, VAL491, VAL317, SER498), confirming channel occupancy. Vina and Vinardo scores differed by less than 1 kcal mol⁻¹ (psoralen: −7.97 vs −7.4; elenolic acid: −6.35 vs −5.5; hydroxytyrosol: −6.23 vs −6.6), indicating marginal binding at both sites. The disagreement reflects a known limitation of empirical scoring functions: small, hydrophobic ligands with weak interactions generate shallow energy landscapes where different functions identify alternative local minima with comparable scores. MM-GBSA calculations, incorporating explicit dynamics, solvation, and entropy absent from docking, provide more reliable ranking. For virtual screening of scoring-function-sensitive compounds, single-method docking is insufficient; post-docking refinement with MM-GBSA or equivalent is necessary. 4.2 Desolvation-Driven Ranking Inversion and Structure–Activity Relationships The MM-GBSA ranking (bergapten > psoralen > luteolin > elenolic acid ≈ quercetin > hydroxytyrosol) inverts the Vina ranking (luteolin > quercetin > bergapten = psoralen > elenolic acid > hydroxytyrosol). The furanocoumarins surpass the flavonoids only after explicit-solvation rescoring. This inversion stems from desolvation penalties: luteolin and quercetin pay +25.0 and +21.8 kcal mol⁻¹ (ΔEGB), while bergapten and psoralen pay +20.3 and +15.6 kcal mol⁻¹. The hydrophobic furanocoumarin scaffold has lower desolvation cost, offsetting its weaker gas-phase interactions (ΔVDW + ΔEEL). Three structural principles emerge: First, hydrophobic contacts with VAL491, ALA495, and LEU320 form a consistent baseline (−0.9 to −2.7 kcal mol⁻¹ per residue) across all scaffolds, driven by van der Waals interactions. Flat aromatic systems (furanocoumarins, flavonoids) pack efficiently; small aliphatic compounds (elenolic acid, hydroxytyrosol) make weaker contacts. Second, desolvation of polar substituents is the dominant driver of the ranking inversion. The flavonoid catechol B-ring, with multiple hydroxyl groups, incurs a large ΔEGB penalty that offsets stronger hydrogen bonding. This generality matters for natural product COX-2 inhibitor design: hydroxyl-rich polyphenols rank well in docking but lose ground after desolvation correction. Third, lipophilic substituents (methoxy in bergapten vs bare psoralen) fine-tune affinity by engaging secondary sites (e.g., SER498 contact: −0.87 vs −0.26 kcal mol⁻¹). Within a scaffold, minor modifications modulate channel binding. 4.3 Luteolin: Literature Cross-Validation The luteolin Vina score (−9.68 kcal mol⁻¹) aligns with published values (Derardja et al., Biol Life Sci Forum 2024;35(1):6, reported −9.49 kcal mol⁻¹). Hydrogen bonds with SER321 and TYR353 (canonical SER530 and TYR385) and van der Waals contacts with VAL491 and TRP355 (canonical VAL523 and TRP387) match published literature (Alam et al. 2016; Janakiramulu & Mamidala 2025; Javid et al. 2025), confirming our bound conformations. Derardja et al. reported luteolin MM-GBSA = −43.41 kcal mol⁻¹, 2.6-fold stronger than our −16.54 kcal mol⁻¹. The discrepancy reflects differences in GB variant, force field, simulation length, and protein structure. This illustrates protocol sensitivity in absolute MM-GBSA values, supporting interpretation of our results in relative ranking rather than absolute affinity prediction. 4.4 Quercetin: Structural Similarity and Replicate-Dependent Variance Quercetin (Vina −9.60, MM-GBSA −14.36 ± 1.53) bound 2.2 kcal mol⁻¹ weaker than luteolin (−9.68 and −16.54 ± 0.88), despite similar docking scores. The two differ by 3-OH on the C-ring and an additional 3′-OH on the B-ring. Both contacted all six core residues with eight hydrogen bonds. The MM-GBSA difference emerged only at N=4: the first two replicates (main: −15.05, s1234: −15.32) suggested luteolin-like binding, but the s9999 replicate (−12.07) revealed a weaker mode, widening SD from 0.19 to 1.53 kcal mol⁻¹. The s9999 replicate showed atypical electrostatics (ΔEEL = +14.57 vs −0.6 to −2.6 for others), indicating quercetin's additional 3-OH can adopt conformations where desolvation cost exceeds intramolecular hydrogen-bonding benefit. This replicate-dependence highlights the importance of multi-replicate sampling for polyphenols with multiple rotatable hydroxyl groups. 4.5 Furanocoumarin Caveat: Binding Affinity Does Not Predict Therapeutic Utility Bergapten (−20.32 kcal mol⁻¹) and psoralen (−18.95 kcal mol⁻¹) show the strongest COX-2 binding among test compounds, but their therapeutic potential is severely limited by off-target liabilities. Both are potent mechanism-based CYP3A4 inhibitors (bergapten: KI = 15.0 μM, kinact = 0.098 min⁻¹) and photosensitizers causing phototoxicity and photocarcinogenicity. These properties preclude systemic use regardless of COX-2 binding. The furanocoumarin advantage over flavonoids stems from lower desolvation penalties, suggesting a design principle: preserve hydrophobic contacts with GLY494, VAL491, and ALA495 while removing the photoreactive furan moiety to yield improved natural product COX-2 leads. Bergapten's methoxy group strengthens SER498 contact (−0.87 vs −0.26 kcal mol⁻¹ for psoralen), showing how minor substituents modulate channel binding without excessive polarity. 4.6 Glycoside Topology Limitation (Oleuropein, Rutin) Oleuropein and rutin were excluded from MM-GBSA due to a systematic gmx_MMPBSA conversion artifact. Both glycosides contain ≥11 GAFF2 c6 (sp3 ring carbon) atoms. The GROMACS→AMBER conversion yielded anomalous ΔG (>+300 and >+550 kcal mol⁻¹) across all replicates despite confirmed binding (minimum distance 1.5–2.0 Å throughout). The ΔVDW term (+264 to +292 kcal mol⁻¹) indicates incorrect Lennard-Jones cross-term assignment when CHARMM36 and GAFF2 parameters mix for large glycosides. This is a pipeline limitation, not an MD deficiency; backbone RMSD and ligand-pocket contacts were stable. Accurate glycoside MM-GBSA would require either a homogeneous force field (all-AMBER with GLYCAM) or manual VDW cross-term correction. Both compounds remain in docking (Section 3.1), where oleuropein showed good Vina (−6.31) and smina consensus (1.03 Å RMSD, AGREE), suggesting it is a genuine COX-2 binder whose binding affinity could not be quantified here. 4.7 Concordance With Published Experimental Data The computational ranking is directionally consistent with published experimental data. Luteolin has reported COX-2 IC₅₀ values in the low-micromolar range in enzymatic assays (PMC4702032; Javid et al., Comput Biol Chem 2025), and its computed ΔG (−16.54 kcal mol⁻¹) places it as the strongest fig flavonoid in our panel — consistent with its established experimental activity. Quercetin's weaker computed affinity (−14.36 kcal mol⁻¹) is consistent with its higher desolvation penalty and its lower potency relative to luteolin in head-to-head experimental comparisons. The directional agreement between experimental IC₅₀ rankings and our computational ΔG ranking — luteolin stronger than quercetin, both weaker than celecoxib — validates the screening workflow output. 4.8 Limitations MM-GBSA with igb=2 overestimates absolute binding affinities, particularly for hydrophobic pockets. All values should be interpreted relatively (against celecoxib and oleocanthal benchmarks run under identical conditions) rather than as Ki or IC₅₀ predictions. As a ranking validation, parallel igb=5 (OBC2) calculations on the same trajectories confirmed that all relative rankings are preserved under the alternative solvation model (full data in Supplementary Table S3). Per-residue decomposition (within 8 Å cutoff) captures 58–87% of system-level ΔG, leaving 5–12 kcal mol⁻¹ from distant residues. Five-nanosecond trajectories suffice for binding-mode stability but do not capture slow conformational transitions. This study addresses only the COX-2 cyclooxygenase active site; allosteric, peroxidase-site, and COX-1 interactions are not evaluated in this phase. AlphaFold 3 (Abramson et al., 2024) represents the current state of the art in protein–ligand structure prediction and would be a natural complement to this pipeline. However, AF3 predicts poses, not replicated thermodynamics: it does not produce ΔG values, per-residue decomposition, or ensemble-averaged dynamics. The MM-GBSA framework here — four independent MD trajectories, igb=2/igb=5 cross-validation, and decomposition identifying the desolvation-driven ranking inversion between furanocoumarins and flavonoids — provides information orthogonal to AF3 pose prediction. The sub-Ångström AF2/MD structural agreement (pocket RMSD 0.307 Å) already validates the simulation active-site geometry against an independent AI approach; AF3 would add a third confirmation of structural accuracy without resolving the thermodynamic questions central to this study. Incorporating AF3 poses as starting conformations for future MM-GBSA calculations would be a logical extension of this screening pipeline. Despite these limitations, internal consistency is strong: stable RMSD, multi-replicate agreement (CV < 10% for most compounds), AF2/MD cross-methodology structural agreement, validated binding-site contacts, and directional concordance with published luteolin and quercetin experimental data all support the relative ranking produced by this computational screening phase. 4.8 Implications for Fig Bioactivity and Dietary Anti-Inflammatory Activity No tested fig-derived compound matched celecoxib in predicted COX-2 binding under these conditions, yet this does not contradict the well-documented anti-inflammatory properties of fig extracts. Instead, it suggests that dietary fig anti-inflammatory effects operate through mechanisms beyond direct COX-2 active-site inhibition. Plausible mechanisms include NF-κB pathway suppression (documented for luteolin and quercetin), antioxidant reduction of prostaglandin precursors, and polypharmacology across multiple targets at dietary concentrations. None of the tested fig compounds showed predicted COX-2 affinity approaching oleocanthal (companion paper [2]), suggesting that the documented anti-inflammatory effects of fig consumption may operate through distinct molecular targets or combination mechanisms rather than direct COX-2 active-site inhibition. The furanocoumarins, despite stronger computed COX-2 affinity, carry documented CYP450 and phototoxicity liabilities that would complicate therapeutic development. Taken together, these findings reinforce that binding affinity alone is insufficient to predict therapeutic utility in natural product drug discovery. The flavonoids (luteolin, quercetin) retain value as dietary anti-inflammatory components through polypharmacological mechanisms, despite modest direct COX-2 affinity. Declarations Acknowledgements The author acknowledges the AlphaFold Protein Structure Database (EMBL-EBI) for the AlphaFold2 prediction used in structural validation, the RCSB Protein Data Bank for the COX-2 crystal structure (PDB: 5KIR), and the developers of GROMACS, gmx_MMPBSA, AutoDock Vina, and AmberTools for making their software freely available. This study is a companion to a parallel investigation of olive-derived secoiridoids against COX-2 [2]. Author contributions Faycal conceived the study, designed the computational workflow, performed all molecular docking, molecular dynamics simulations, MM-GBSA binding free energy calculations, per-residue decomposition analysis, AlphaFold2 structural validation, and data analysis, and wrote the manuscript. Funding This research received no external funding. Competing interests The author declares no competing interests. Data availability All data generated during this study are available in a Zenodo repository at https://doi.org/10.5281/zenodo.19076569. This includes: raw MM-GBSA output files (FINAL_RESULTS_DECOMP.dat, ENERGIES_DECOMP.csv) for all six fig compounds and celecoxib across all replicates, igb=5 cross-validation results, GROMACS topology files, MDP parameter files, molecular docking input/output files, AlphaFold2 validation data, and analysis scripts. Production trajectory files are available from the corresponding author upon reasonable request due to their large size. Code availability All analysis scripts, including the AlphaFold2 validation script, MM-GBSA batch processing scripts, contact analysis code, and figure generation scripts, are deposited in the Zenodo repository. The study used the following publicly available software: GROMACS 2026 (https://www.gromacs.org), gmx_MMPBSA v1.6.4 (https://github.com/Valdes-Tresanco-MS/gmx_MMPBSA), AutoDock Vina 1.2 (https://github.com/ccsb-scripps/AutoDock-Vina), and AmberTools25 (https://ambermd.org). References Smith WL, DeWitt DL, Garavito RM. Cyclooxygenases: structural, cellular, and molecular biology. Annu Rev Biochem. 2000;69:145–82. Rouzer CA, Marnett LJ. Cyclooxygenases: structural and functional insights. J Lipid Res. 2009;50 Suppl:S29–34. Vane JR, Bakhle YS, Botting RM. Cyclooxygenases 1 and 2. Annu Rev Pharmacol Toxicol. 1998;38:97–120. Laine L. Gastrointestinal effects of NSAIDs and coxibs. J Pain Symptom Manage. 2003;25(2 Suppl):S32–40. Bresalier RS et al. Cardiovascular events associated with rofecoxib in a colorectal adenoma chemoprevention trial. N Engl J Med. 2005;352:1092–102. Yoon JH, Baek SJ. Molecular targets of dietary polyphenols with anti-inflammatory properties. Yonsei Med J. 2005;46:585–96. Azab A, Nassar A, Azab AN. Anti-inflammatory activity of natural products. Molecules. 2016;21:1321. Subehan et al. Bergapten CYP3A4 mechanism-based inhibition. J Agric Food Chem 2007. (PMID 17988092) Alam MA et al. Luteolin and COX-2 docking studies. (PMC4702032) Janakiramulu P & Mamidala E. Flavonoid derivatives vs COX-2. In Silico Pharmacol 2025;13:59. DOI:10.1007/s40203-025-00349-x (PMID 40255260) Javid R et al. Luteolin MM-GBSA against COX-2. Comput Biol Chem 2025;118:108499. DOI:10.1016/j.compbiolchem.2025.108499 (PMID 40347541) Derardja I et al. Biol Life Sci Forum 2024;35(1):6. DOI:10.3390/blsf2024035006 [conference proceedings] Veberic R, Colaric M, Stampar F. Phenolic acids and flavonoids of fig fruit ( Ficus carica L.) in the northern Mediterranean region. Food Chem. 2008;106:153–7. Badgujar SB et al. Ficus carica Linn.: a review of its pharmacology, phytochemistry and traditional uses. J Ethnopharmacol. 2014;154:183–209. Garavito RM, DeWitt DL. The cyclooxygenase isoforms: structural insights into the conversion of arachidonic acid to prostaglandins. Biochim Biophys Acta. 1999;1441:213–20. Kurumbail RG et al. Structural basis for selective inhibition of cyclooxygenase-2 by anti-inflammatory agents. Nature. 1996;384:644–8. Malkowski MG et al. The productive conformation of arachidonic acid bound to prostaglandin synthase. Science. 2000;289:1933–7. Luong C et al. Flexibility of the NSAID binding site in the structure of human cyclooxygenase-2. Nat Struct Biol. 1996;3:927–33. Genheden S, Ryde U. The MM/PBSA and MM/GBSA methods to estimate ligand-binding affinities. Expert Opin Drug Discov. 2015;10:449–61. Wang E et al. End-point binding free energy calculation with MM/PBSA and MM/GBSA: strategies and applications in drug design. Chem Rev. 2019;119:9478–508. Valdés-Tresanco MS et al. gmx_MMPBSA: a new tool to perform end-state free energy calculations with GROMACS. J Chem Theory Comput. 2021;17:6281–91. Abramson J et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature. 2024;630:493–500. DOI:10.1038/s41586-024-07487-w Ferhat, F. Computational Screening of Olive Secoiridoids Against COX-2: Oleocanthal Exhibits Strong Predicted Binding via Novel MET522 Side-Pocket Engagement. Scientific Reports (2026, under review). Additional Declarations No competing interests reported. Supplementary Files SupplmentaryTables.docx Supplementary Information Table S1 (residue number mapping between 5KIR renumbered residues and canonical UniProt P35354 numbering) is included in Section 2.1. Supplementary data files (FINAL_RESULTS_DECOMP.dat, ENERGIES_DECOMP.csv) for all test compounds and replicates, igb=5 cross-validation results, furanocoumarin safety assessment data, and all raw coordinate comparison files are available in the Zenodo data deposit. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 01 May, 2026 Reviewers agreed at journal 06 Apr, 2026 Reviewers invited by journal 06 Apr, 2026 Editor invited by journal 25 Mar, 2026 Editor assigned by journal 21 Mar, 2026 Submission checks completed at journal 21 Mar, 2026 First submitted to journal 18 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9155203","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":619706366,"identity":"0c240cd6-b6b9-4a56-b1a0-04ddaac043d5","order_by":0,"name":"Faycal Ferhat","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyUlEQVRIiWNgGAWjYBAC9uYDIMpCDkQeeECMFp5jCSBKwhisJYEULYkNIIo4LWzMzyR+1Eikzw87/BBoi52cbgNBLWxmkj3HJHI33k4zAGpJNjY7QECLvXyDmTQDG1DL7ASQlgOJ2whp4WFj/ybN8E8i3XB2+gditfCYSTO2SSTIS+cQbQtPsWVvn4ThBumcggMJBkT4BeiwjTd+fLORl5+dvvnDhwo7OYJagIBFAkQagFUaEFYOAswfQKR8A3GqR8EoGAWjYAQCADpAP+6StDg/AAAAAElFTkSuQmCC","orcid":"","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Faycal","middleName":"","lastName":"Ferhat","suffix":""}],"badges":[],"createdAt":"2026-03-18 06:09:56","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9155203/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9155203/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106960217,"identity":"63f3bb72-4c7a-4d97-9a27-13e4ad8c707b","added_by":"auto","created_at":"2026-04-15 09:19:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1368707,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9155203/v1/7eb12e6e-3084-4b96-9f14-08ce126878e6.pdf"},{"id":106820229,"identity":"80b03d61-6e9e-42b3-9183-ceecb6a0077f","added_by":"auto","created_at":"2026-04-13 18:32:42","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":17510,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTable S1 (residue number mapping between 5KIR renumbered residues and canonical UniProt P35354 numbering) is included in Section 2.1. Supplementary data files (FINAL_RESULTS_DECOMP.dat, ENERGIES_DECOMP.csv) for all test compounds and replicates, igb=5 cross-validation results, furanocoumarin safety assessment data, and all raw coordinate comparison files are available in the Zenodo data deposit.\u003c/p\u003e","description":"","filename":"SupplmentaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-9155203/v1/3242fcc33216b17826974f58.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Computational Evaluation of Fig-Derived Bioactive Compounds as COX-2 Inhibitors: Molecular Dynamics Reveals Desolvation-Driven Ranking Inversion Between Flavonoids and Furanocoumarins","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe fig (\u003cem\u003eFicus carica\u003c/em\u003e) is among the oldest cultivated plants, with archaeological evidence of human consumption predating cereal agriculture [12]. Fig extracts have demonstrated anti-inflammatory activity in multiple in vitro and in vivo models, and fig-derived compounds span several pharmacologically relevant chemical classes \u0026mdash; flavonoids, furanocoumarins, secoiridoid derivatives, and simple phenols [11,12]. However, the molecular basis for these effects at specific drug targets has not been systematically evaluated using physics-based methods.\u003c/p\u003e \u003cp\u003eCyclooxygenase-2 (COX-2) catalyses the first committed step in prostaglandin synthesis, converting arachidonic acid to prostaglandin endoperoxide H₂ (PGH₂), the immediate precursor of pro-inflammatory prostanoids. Unlike COX-1, which is constitutively expressed, COX-2 induction by inflammatory stimuli makes it a validated pharmaceutical target [1,2,3]. NSAIDs block COX enzymes broadly, but selective COX-2 inhibitors carry cardiovascular risk (rofecoxib, valdecoxib withdrawn from market) and non-selective NSAIDs cause gastrointestinal injury [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. This safety gap has driven sustained interest in plant-derived COX-2 inhibitors with potentially better tolerability profiles [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFigs contain a chemically diverse set of bioactive compounds: flavonoids (luteolin, quercetin), furanocoumarins (psoralen, bergapten), a secoiridoid derivative (elenolic acid), and a simple phenol (hydroxytyrosol) [11,12]. Luteolin has been docked repeatedly against COX-2, with published Vina scores clustering around \u0026minus;\u0026thinsp;9.5 to \u0026minus;\u0026thinsp;10.2 kcal mol⁻\u0026sup1; [13,15,16]. Quercetin, abundant in fig pulp and leaves, shares the flavonoid scaffold with luteolin but differs by hydroxyl substitution pattern. Psoralen and bergapten are furanocoumarins with documented inhibitory activity against CYP450 enzymes and known phototoxicity [13,14]. However, published studies of fig compounds against COX-2 rely entirely on docking; to our knowledge, no prior work has applied replicated molecular dynamics simulation with explicit MM-GBSA binding free-energy calculation to a systematic fig-compound panel.\u003c/p\u003e \u003cp\u003eThe COX-2 cyclooxygenase channel is a hydrophobic pocket flanked by aromatic residues (Tyr385, Trp387 in canonical numbering) and selective for inhibitors with appropriate size and hydrophobic complementarity. The channel floor is anchored by Val523, while selectivity versus COX-1 derives from an Ile side pocket at the Val523 position [19,20]. Binding energetics split among van der Waals contacts, hydrogen bonding, and desolvation costs; distinguishing these contributions requires dynamic simulation with explicit solvation treatment rather than rigid-body docking alone.\u003c/p\u003e \u003cp\u003eMM-GBSA methods combine molecular mechanics force fields with implicit Born solvation to estimate binding free energies while accounting for protein flexibility, ligand desolvation, and conformational sampling [21,22]. Multi-replicate MD simulations with distinct initial velocity seeds produce reproducible free-energy estimates with quantifiable standard error, allowing statistical comparisons [23].\u003c/p\u003e \u003cp\u003eThis study constitutes a \u003cb\u003ecomputational screening phase\u003c/b\u003e in the evaluation of fig bioactives as COX-2 inhibitors. We apply a hierarchical three-tier screening workflow: (1) Vina docking as a rapid pose filter; (2) 5 ns all-atom MD (4 replicates each) for dynamic validation; (3) MM-GBSA binding free-energy calculation for quantitative ranking. This workflow is specifically designed to identify which compounds, if any, warrant experimental follow-up (in vitro COX-2 inhibition assays, cell-based prostaglandin suppression) versus which can be deprioritised at the computational stage. Results are benchmarked against celecoxib and oleocanthal. Drug-likeness and predicted ADMET profiles contextualise affinity within therapeutic feasibility. Two glycoside compounds (oleuropein, rutin) underwent docking only due to force-field limitations.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cp\u003e\u003cstrong\u003e2.1 Protein Preparation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe crystal structure of human COX-2 (PDB: 5KIR, resolution 2.4 \u0026Aring;) was obtained from the RCSB Protein Data Bank. The structure was processed using OpenMM: water molecules and co-crystallized ligands were removed, missing hydrogen atoms were added at physiological pH 7.4, and residues were protonated using the ff14SB force field protonation states. The processed structure (5KIR_fixed.pdb) was used for all subsequent calculations.\u003c/p\u003e\n\u003cp\u003eOpenMM renumbers residues sequentially from 1, so the numbering in 5KIR_fixed.pdb does not match canonical COX-2 literature numbering. The mapping was established by sequential C\u0026alpha; alignment between 5KIR_fixed.pdb and the original RCSB 5KIR structure (5KIR_original.pdb; 551/551 chain A residues matched by residue name). All residue numbers used in this manuscript refer to the OpenMM-renumbered structure; canonical equivalents for all active-site residues cited in Results are listed in Table S1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAlphaFold2 structural validation and simulation software cross-check.\u003c/strong\u003e The COX-2 crystal structure (5KIR) was validated against the AlphaFold2 monomer prediction (UniProt P35354, AF-P35354-F1-model_v6.pdb, downloaded directly from the EMBL-EBI AlphaFold Database). AlphaFold2 generates structures from sequence alone \u0026mdash; no force field, no experimental target data \u0026mdash; making it a fully independent reference. Global C\u0026alpha; RMSD across 551 matched residues was 0.38 \u0026Aring;; pocket C\u0026alpha; RMSD across the 10 active-site residues was 0.25 \u0026Aring; (locally aligned; 0.36 \u0026Aring; post-global alignment). All pLDDT values were read directly from the B-factor column of the raw AF2 PDB file and verified using a deposited script (Zenodo, `alphafold/validate_af2_vs_5kir.py`); all 10 pocket residues carry pLDDT \u0026ge; 94.6 (mean 97.0).\u003c/p\u003e\n\u003cp\u003eThis validates the 5KIR template (crystal vs AF2 pocket RMSD 0.251 \u0026Aring;) and simultaneously cross-validates the simulation software. The raw MD average structure (last 2 ns, 400 frames, `avg_structure_last2ns.pdb`) was compared directly against the raw AF2 PDB by extracting C\u0026alpha; coordinates for the 10 pocket residues from both files and computing post-Kabsch distances. Result: \u003cstrong\u003eMD trajectory vs AF2 pocket RMSD = 0.307 \u0026Aring;\u003c/strong\u003e, per-residue range 0.204\u0026ndash;0.432 \u0026Aring;. CHARMM36+GAFF2 dynamics in explicit solvent and AF2 machine learning, operating on entirely different physical principles, converge on the same binding pocket geometry to sub-\u0026Aring;ngstr\u0026ouml;m precision. Raw coordinate table: Zenodo `alphafold/MD_vs_AF2_raw_comparison.txt`.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 Ligand Preparation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCompounds were prepared from canonical SMILES strings obtained from PubChem. Three-dimensional coordinates were generated using RDKit (v2023.09) with the ETKDGv3 algorithm followed by MMFF94 geometry optimization. Partial charges and GAFF2 atom types were assigned using Antechamber (AmberTools25) with the AM1-BCC charge model (flag: -c bcc -at gaff2). GROMACS-compatible topologies were generated using ACPYPE (v2022.7). For Quercetin, SDF input was used for Antechamber to preserve bond order information and ensure correct assignment of aromatic carbon atom types (ca rather than c3). The compound panel comprised six phytochemicals found in \u003cem\u003eFicus carica\u003c/em\u003e (fig): Luteolin, Psoralen, Bergapten, Elenolic acid, Hydroxytyrosol, and Quercetin. Quercetin was included on the basis of its documented occurrence in fig pulp and leaves and its structural relatedness to Luteolin (flavone/flavonol scaffold), enabling direct SAR comparison within the flavonoid series. Two additional glycoside compounds (Oleuropein and Rutin) were prepared using the same protocol. Celecoxib served as a reference inhibitor standard.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Molecular Docking\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMolecular docking was performed using AutoDock Vina 1.2.6. The receptor PDBQT file was prepared from 5KIR_fixed.pdb using OpenBabel with Gasteiger partial charges. Ligand PDBQT files were prepared using OpenBabel. The docking search box was centered at coordinates (24.2, 1.6, 35.4) \u0026Aring;\u0026mdash;derived from the center of mass of oleocanthal in the energy-minimized (EM) complex structure\u0026mdash;with dimensions of 25 \u0026times; 25 \u0026times; 25 \u0026Aring;, encompassing the full COX-2 cyclooxygenase channel. Docking was performed with exhaustiveness = 32 and 9 binding modes generated. The best-scoring pose was selected for MD simulation. Pose validation was performed by computing heavy-atom contacts (\u0026le; 4.5 \u0026Aring;) against 14 experimentally identified COX-2 binding-site residues; a pose was considered bound when \u0026ge; 2 of 6 designated core residues (PHE486, SER321, VAL491, ALA495, TRP355, TYR353) were contacted. Protein\u0026ndash;ligand interaction fingerprinting was performed using MDAnalysis with a 4.5 \u0026Aring; distance cutoff; ProLIF was evaluated but is not compatible with the ARM64 build of RDKit due to a fatal assert in the bond-order inference path.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4 System Setup and Energy Minimization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eProtein\u0026ndash;ligand complexes were assembled as follows. The ACPYPE-generated ligand GRO file was translated to the docked pose center of mass. The ligand was merged with the protein GRO file (from pdb2gmx with CHARMM36-jul2022 force field, TIP3P water model). The combined system was placed in a dodecahedral periodic box with a minimum solute-to-box-face distance of 1.2 nm using gmx editconf. The box was solvated with TIP3P water using gmx solvate, and charge-neutralized to 0.15 M NaCl using gmx genion. All GAFF2 LIG atom type parameters were inlined in topol.top to prevent duplicate atomtype conflicts, with the ligand moleculetype block included as a separate ITP file.\u003c/p\u003e\n\u003cp\u003eEnergy minimization was performed using the steepest descent algorithm (emtol = 100 kJ mol⁻\u0026sup1; nm⁻\u0026sup1;, nsteps = 50,000) with CHARMM36-jul2022 for the protein and GAFF2 for the ligand (GROMACS 2026, GPU-accelerated with CUDA on a single A100 GPU).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 Molecular Dynamics Simulations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMD simulations were conducted using GROMACS 2026. For each compound, the following protocol was applied:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNVT equilibration:\u003c/strong\u003e 100 ps, dt = 1 fs, V-rescale thermostat (\u0026tau; = 0.1 ps, T = 300 K), protein backbone and ligand position-restrained (k = 1000 kJ mol⁻\u0026sup1; nm⁻\u0026sup2;), H-bond constraints (LINCS), PME electrostatics with rcoulomb = rvdw = 1.2 nm.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNPT equilibration:\u003c/strong\u003e 100 ps, dt = 2 fs, Parrinello\u0026ndash;Rahman barostat (\u0026tau;_p = 2.0 ps, P = 1 bar), same temperature coupling, position restraints maintained.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProduction MD:\u003c/strong\u003e 5 ns \u0026times; 4 independent replicates per compound, dt = 1 fs (chosen to ensure accurate force evaluation for flexible ligand torsions, at the cost of higher computation relative to 2 fs), all-bonds constrained, NPT ensemble at 300 K, 1 bar (Parrinello\u0026ndash;Rahman). Three replicates used gen_vel=yes with seeds 1234, 42, and 9999 to assign independent initial velocities at the start of production; the main run used gen_vel=no, continuing velocities from the NPT endpoint. All four replicates started from the same NPT-equilibrated coordinates, ensuring identical initial structures while sampling distinct regions of phase space. Coordinates were saved every 5 ps (nstxout-compressed = 5000, dt = 1 fs; 1,001 frames per 5 ns replicate) for all compounds.\u003c/p\u003e\n\u003cp\u003eTrajectory PBC correction was applied using gmx trjconv (-pbc mol, centering on Protein+Ligand group) prior to MM-GBSA analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.6 Binding Free Energy Calculations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBinding free energies were calculated using the MM-GBSA method implemented in gmx_MMPBSA (v1.6.4) with sander (AmberTools25). For each replicate trajectory, frames were extracted at uniform intervals using gmx_MMPBSA parameters startframe=1, endframe=5000, interval=50. Because interval refers to frame indices (not time), this yields 21 frames per replicate for compounds with 1,001-frame trajectories (every 50th frame = every 250 ps). The MM-GBSA input parameters were: igb = 2, saltcon = 0.150 M, forcefields = CHARMM36,GAFF2. The final \u0026Delta;G_bind for each compound was reported as the mean \u0026plusmn; standard deviation across the four replicates, with SEM = SD/\u0026radic;4. Statistical comparisons against the celecoxib reference were performed using the Welch two-sample t-test (two-tailed).\u003c/p\u003e\n\u003cp\u003ePer-residue binding free energy decomposition (idecomp=2, 1\u0026ndash;4 interactions included) was performed using print_res=\u0026quot;within 8\u0026quot; for all test compounds across all four replicates. Decomposition coverage (decomposed residue sum / system-level \u0026Delta;G) is reported alongside per-residue data to quantify the fraction of binding energy accounted for by residues within the cutoff shell.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.7 Reference Compounds\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCelecoxib (\u0026Delta;G = \u0026minus;38.875 \u0026plusmn; 1.199 kcal mol⁻\u0026sup1;) and Oleocanthal (\u0026Delta;G = \u0026minus;61.765 \u0026plusmn; 1.477 kcal mol⁻\u0026sup1;, from companion paper [2]) were simulated using the identical protocol (N = 4 replicates each, igb = 2, saltcon = 0.150 M) and serve as in-house benchmarks. All reference values were obtained under the same computational conditions as the test compounds.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.8 AI Writing Assistance Disclosure\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLarge language model (LLM) assistance was used during the preparation of this manuscript for writing support, editing, and structural guidance. The LLM used was a locally deployed model running within the NVIDIA Ecosystem. All scientific content, computational results, raw data, statistical analyses, and conclusions are the sole work of the author and were independently verified against raw output files. LLM assistance was not used for data generation, analysis, or interpretation of results.\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e\u003cstrong\u003e3.1 Molecular Docking and Pose Validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll eight compounds (six test + two glycosides) were successfully docked into the COX-2 cyclooxygenase channel. Binding pose validation against 14 active-site residues confirmed that seven of eight compounds occupied the primary active site (≥ 2 core residue contacts; Table 1). Rutin, despite registering all six core residue contacts, exhibited steric clashes (\u0026lt; 1.5 Å) at SER321 and ALA495, consistent with its large disaccharide moiety being geometrically incompatible with the deep hydrophobic channel; its Vina score of −2.54 kcal mol⁻¹ was also substantially weaker than the remaining compounds.\u003c/p\u003e\n\u003cp\u003eLuteolin achieved the strongest Vina affinity (−9.68 kcal mol⁻¹) and formed contacts with all six core residues, including eight hydrogen bonds (SER498, TYR353, TYR323, HIS57, PHE486, SER321, ILE485). The flavone scaffold showed aromatic stacking with TRP355 and PHE486. Bergapten and Psoralen (furanocoumarins, −7.97 kcal mol⁻¹ each) were well-accommodated in the channel, with Bergapten forming four hydrogen bonds to SER321, SER498, MET490, and VAL317. Elenolic acid (−6.35 kcal mol⁻¹) contacted all six core residues with five hydrogen bonds. Oleuropein (−6.31 kcal mol⁻¹) and Hydroxytyrosol (−6.23 kcal mol⁻¹) occupied the active site with ≥ 5 core residue contacts. Quercetin (−9.60 kcal mol⁻¹) contacted all six core residues and formed eight hydrogen bonds (PHE486, SER321, TYR353, TYR323, HIS57, SER498), comparable to Luteolin in binding complementarity despite lacking the 5-OH group.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1. Docking scores and active-site contact summary for all eight compounds.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCompound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVina (kcal/mol)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eIn site\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eContacts\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eH-bonds\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCore residues hit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLuteolin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−9.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495, TRP355, TYR353\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eQuercetin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−9.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495, TRP355, TYR353\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePsoralen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−7.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBergapten\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−7.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495, TRP355, TYR353\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eElenolic acid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495, TRP355, TYR353\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eOleuropein\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495, TRP355\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHydroxytyrosol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495, TRP355\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRutin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−2.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE486, SER321, VAL491, ALA495, TRP355, TYR353\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eRutin contacts all core residues but exhibits steric clashes (\u0026lt; 1.5 Å) at SER321/ALA495; poor Vina score consistent with size mismatch.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eScoring-function consensus validation.\u003c/strong\u003e To assess binding-mode robustness, all seven test compounds were independently re-docked using smina with Vinardo scoring (exhaustiveness=32, same receptor and box). Rank-1 poses were compared to the original AutoDock Vina poses by heavy-atom RMSD (Table 1b).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1b. Vina–\u003c/strong\u003e\u003cstrong\u003esmina (Vinardo) pose consensus.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCompound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVina (kcal/mol)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSmina (kcal/mol)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRMSD (Å)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVerdict\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLuteolin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−9.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−10.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAGREE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eQuercetin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−9.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−10.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAGREE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eOleuropein\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−4.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAGREE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBergapten\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−7.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−7.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePartial overlap\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePsoralen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−7.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−7.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDifferent site\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eElenolic acid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−5.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDifferent site\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHydroxytyrosol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDifferent site\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRutin†\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−2.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eExcluded\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e†Rutin excluded from smina rescoring: the systematic steric clash at SER321/ALA495 precludes any productive binding pose in the deep hydrophobic channel; only the Vina reference score is reported.\u003c/p\u003e\n\u003cp\u003eThree compounds (luteolin, quercetin, oleuropein) achieved strong Vina–smina consensus (RMSD \u0026lt; 2 Å), confirming scoring-function-independent binding modes. Bergapten showed partial overlap (5.9 Å RMSD), suggesting pose ambiguity at the channel entrance. Three smaller, weaker compounds (psoralen, elenolic acid, hydroxytyrosol) diverged to an alternative surface site under Vinardo scoring (RMSD 13–17 Å). Per-residue analysis showed that all three Vina poses retain 6–7 channel residues in their top-8 contacts, indicating the disagreement reflects scoring-function sensitivity for marginally-bound small molecules rather than incorrect docking. Vina scores differed from Vinardo by less than 1 kcal mol⁻¹ for these three compounds, affirming marginal binding status at both sites. MM-GBSA calculations use the Vina poses, providing explicit dynamic and solvation corrections for affinity ranking.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 MD Simulation Stability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eProtein backbone RMSD stabilized rapidly for all compounds. Across all trajectories (main + replicates), mean backbone RMSD ranged from 0.9 to 1.3 Å (maxima 1.2–1.7 Å), showing stable protein structure throughout (Table 3).\u003c/p\u003e\n\u003cp\u003eLigand RMSD analysis used two reference frames. The main run measured RMSD relative to the docked pose (quantifying drift from initial binding). Replicates measured RMSD relative to their own first frame, because the equilibrated ligand position after 5 ns main production differs 0.4–1.4 nm from the original dock (an equilibration feature, not compound-specific). All simulations maintained protein–ligand contact (mean minimum distance 0.20 nm).\u003c/p\u003e\n\u003cp\u003eIn the main run, hydroxytyrosol (mean 2.5 Å) and rutin (mean 1.9 Å) showed the lowest ligand RMSD. However, rutin's stability reflects steric confinement of its large disaccharide (MW 610 Da) rather than productive binding; this aligns with its poor Vina score (−2.54 kcal mol⁻¹), steric clashes at SER321/ALA495, and strongly positive MM-GBSA ΔG (Section 3.3). Psoralen showed the largest displacement (mean 5.3 Å), consistent with its flat, compact scaffold providing weak steric fit in the deep hydrophobic channel. Replicates showed uniformly lower ligand RMSD (mean 0.8–2.2 Å), as expected from starting at the equilibrated pose.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3. MD stability summary (main production + replicates, 5 ns each).\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCompound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRun\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBB mean (Å)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBB max (Å)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLig mean (Å)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLig max (Å)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLig ref\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHydroxytyrosol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003emain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTPR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es1234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePsoralen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003emain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTPR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es1234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBergapten\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003emain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTPR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es1234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eElenolic acid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003emain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTPR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es1234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLuteolin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003emain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTPR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es1234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRutin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003emain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTPR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es1234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003es9999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eBB RMSD: backbone, least-squares fit to backbone. Lig RMSD: fit to backbone, computed on ligand heavy atoms. \"TPR\" = docked-pose reference; \"self\" = frame 0 of PBC-corrected replicate trajectory (self-referenced, see Methods 2.5). s9999 RMSD data for non-Rutin test compounds not shown (all confirmed stable; BB RMSD \u0026lt; 1.6 Å). Oleuropein omitted (ligand-unstable).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Binding Free Energies (MM-GBSA)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMM-GBSA binding free energies were calculated for all test compounds using gmx_MMPBSA v1.6.4 (igb=2, saltcon=0.150 M, CHARMM36+GAFF2 force fields, 300 K). Each compound was simulated across 4 independent 5 ns production replicates, with 21 frames sampled per replicate (interval=50 over 1001 trajectory frames). Results are summarised in Tables 2b (test compounds, N=4) alongside reference celecoxib data for benchmarking.\u003c/p\u003e\n\u003cp\u003eTwo compounds were excluded from the binding free energy analysis. Oleuropein and Rutin — both glycosides containing ≥ 11 GAFF2 c6 (sp3 ring carbon) atom types — yielded strongly positive ΔG values (\u0026gt; +300 and \u0026gt; +550 kcal mol⁻¹, respectively) across all four replicates despite confirmed ligand binding (minimum protein–ligand distance 1.5–2.0 Å throughout all trajectories). The anomalous energies are attributable to van der Waals cross-term incompatibility in the GROMACS→AMBER topology conversion employed by gmx_MMPBSA when mixing CHARMM36 (protein) and GAFF2 (ligand) force fields for large glycoside ligands; all covalent bond distances in the extracted ligand coordinates were verified intact (max 2.03 Å for Oleuropein, 2.12 Å for Rutin). Both compounds remain in the docking analysis (Section 3.1) and are discussed further in Section 4.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2b. MM-GBSA binding free energies — test compounds (N=4).\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCompound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRep_main\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRep_s1234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRep_s42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRep_s9999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMean ΔG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e±SD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSEM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eΔVDW\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eΔEEL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eΔEGB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eΔESURF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eΔΔG vs Cel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWelch t\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBergapten‡\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−19.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−21.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−19.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−21.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−20.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−29.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−6.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+19.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−3.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+18.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+24.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePsoralen‡\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−17.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−19.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−17.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−20.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−18.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−26.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−4.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+15.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−3.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+19.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+23.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLuteolin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−16.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−17.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−15.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−16.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−16.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−34.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−2.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+25.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−5.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+22.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+30.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eQuercetin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−15.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−15.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−14.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−12.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−14.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−33.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+2.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+21.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−4.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+24.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+25.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eElenolic acid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−13.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−14.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−15.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−15.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−14.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−25.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−9.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+23.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−3.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+24.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+30.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHydroxytyrosol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−10.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−9.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−10.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−10.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−10.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−22.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−9.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+25.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−3.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+28.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e+43.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAll values in kcal mol⁻¹. ‡Furanocoumarin — CYP450 inhibitor and phototoxic; see Discussion. Welch two-sample t-test (two-tailed) vs celecoxib. Reference celecoxib: ΔG = −38.88 ± 1.20 kcal mol⁻¹, N=4. Energy components (averaged across all replicates): ΔVDW = van der Waals, ΔEEL = Coulombic, ΔEGB = Generalized Born solvation, ΔESURF = non-polar solvation.\u003c/p\u003e\n\u003cp\u003eThe MM-GBSA ranking was: Bergapten \u0026gt; Psoralen \u0026gt; Luteolin \u0026gt; Elenolic acid ≈ Quercetin \u0026gt; Hydroxytyrosol. All test compounds bound significantly less favourably than celecoxib (Welch t-test, all p ≤ 8.5×10⁻⁷). Affinity gaps ranged from +18.6 to +28.5 kcal mol⁻¹ versus celecoxib. All six fig compounds were weaker than oleocanthal (−61.77 ± 1.48 kcal mol⁻¹, companion paper [2]), with deficits of 41.5–51.4 kcal mol⁻¹.\u003c/p\u003e\n\u003cp\u003eQuercetin (−14.36 ± 1.53) and luteolin (−16.54 ± 0.88) differed by 2.2 kcal mol⁻¹, consistent with their single-group difference (3-OH on C-ring). The quercetin s9999 replicate (−12.07 kcal mol⁻¹) was substantially weaker than the other three (−15.05 to −15.32), widening the SD from 0.19 (N=2) to 1.53 (N=4). This outlier reflects atypical electrostatic behavior (ΔEEL = +14.57 vs\u0026nbsp;−0.6 to −2.6 for other replicates), suggesting that quercetin's additional 3-OH can adopt a conformation where desolvation cost exceeds hydrogen-bonding benefit.\u003c/p\u003e\n\u003cp\u003eThe Vina→MM-GBSA ranking inverted between furanocoumarins and flavonoids: bergapten and psoralen (Vina −7.97 kcal mol⁻¹ each) outranked luteolin (−9.68) and quercetin (−9.60) in MM-GBSA. The flavonoid hydroxyl groups incur large ΔEGB penalties (+23–25 kcal mol⁻¹), while the hydrophobic furanocoumarins pay less (+16–20 kcal mol⁻¹). This desolvation advantage offsets the furanocoumarins' weaker gas-phase interactions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Per-Residue Energy Decomposition\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePer-residue binding contributions were computed as the difference between Complex and Receptor energies for each protein residue (idecomp=2, pairwise decomposition with 1-4 interactions folded into VDW and EEL terms, dec_verbose=3, full residue totals). Table 4 reports the top contributing residues for each compound, ranked by ΔTotal (Complex − Receptor, backbone + sidechain combined).\u003c/p\u003e\n\u003cp\u003eAll six test compounds engaged the same core channel residues: ALA495, LEU320, VAL491, VAL317, SER321, MET490, SER498, GLY494, and TYR353. Each compound buried 7–10 of its top-10 contributing residues within the active site, confirming consistent channel occupancy across all scaffolds.\u003c/p\u003e\n\u003cp\u003eBergapten and psoralen (furanocoumarins) engaged 8 pocket residues in their top-10, with ALA495 and LEU320 as primary anchors (bergapten: −1.78, −1.57; psoralen: −1.59, −1.51). Bergapten's methoxy group strengthened the SER498 contact (−0.87 vs −0.26), accounting for the 1.4 kcal mol⁻¹ MM-GBSA difference between them.\u003c/p\u003e\n\u003cp\u003eLuteolin and quercetin (flavonoids) engaged 9–10 pocket residues, led by ALA495 and LEU320/VAL317 (luteolin: −1.45, −1.31; quercetin: −0.53, −0.43, −0.37 across three residues). Quercetin showed smaller per-residue magnitudes and a more diffuse footprint; large ΔEGB desolvation penalties (Section 3.3) result in weaker binding despite similar residue contacts.\u003c/p\u003e\n\u003cp\u003eElenolic acid (max VAL491 −1.37) and hydroxytyrosol (max ALA495 −1.52) showed the weakest per-residue profiles, consistent with their position at the weak-binding end of the ranking.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4. Top-5 per-residue binding contributions (ΔTotal, kcal mol\u003c/strong\u003e⁻\u003cstrong\u003e¹) for test compounds.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCompound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRes-1 (ΔTotal)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRes-2 (ΔTotal)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRes-3 (ΔTotal)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRes-4 (ΔTotal)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRes-5 (ΔTotal)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePocket in top-10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBergapten\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eALA495 (−1.78)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL317 (−1.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEU320 (−1.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL491 (−1.26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSER498 (−0.87)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8/10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePsoralen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eALA495 (−1.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEU320 (−1.51)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSER321 (−1.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL491 (−0.94)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL317 (−0.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8/10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eQuercetin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL317 (−0.53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEU320 (−0.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL491 (−0.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTYR316 (−0.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eALA495 (−0.30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9/10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLuteolin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eALA495 (−1.45)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEU320 (−1.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL317 (−1.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSER321 (−0.92)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePHE349 (−0.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10/10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eElenolic acid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL491 (−1.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eGLY494 (−1.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEU320 (−1.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eALA495 (−0.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL317 (−0.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10/10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHydroxytyrosol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eALA495 (−1.52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVAL317 (−0.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSER321 (−0.77)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEU320 (−0.77)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eGLY494 (−0.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7/10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eValues are full residue ΔTotal (backbone + sidechain, idecomp=2, dec_verbose=3). All compounds: N=4 replicates, per-residue values are mean across all four replicates. Quercetin values extracted from FINAL_DECOMP_MMPBSA.dat Delta section (N=4 confirmed: main, s1234, s42, s9999); replicate range VAL317 −0.44 to −0.65, CV \u0026lt; 20%.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Comparison with Reference Inhibitor and Positive Control\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTest compounds were compared against celecoxib (−38.88 ± 1.20, N=4, reference standard) and oleocanthal (−61.77 ± 1.48, companion paper [2], positive control).\u003c/p\u003e\n\u003cp\u003eNo tested fig-derived compound approached celecoxib affinity in this computational model. Luteolin, the strongest fig binder at −16.54 ± 0.88, was 22.3 kcal mol⁻¹ weaker (Welch t = 30.10, p = 2.6×10⁻⁷). Quercetin (−14.36 ± 1.53) was 2.2 kcal mol⁻¹ below luteolin, with one weaker replicate (s9999, −12.07) driving higher variance. Elenolic acid (−14.88 ± 1.02) and hydroxytyrosol (−10.34 ± 0.56) ranked lowest. Large ΔEGB desolvation penalties for flavonoid hydroxyl groups (Section 3.3) limit their MM-GBSA performance.\u003c/p\u003e\n\u003cp\u003eBergapten (−20.32 ± 0.97) and psoralen (−18.95 ± 1.16) ranked first and second overall, exceeding the flavonoids despite weaker docking scores. However, both are potent CYP450 inhibitors and phototoxic agents, raising significant concerns for therapeutic development (elaborated in Section 4.5).\u003c/p\u003e\n\u003cp\u003eIn summary, no tested fig-derived compound matched celecoxib under these computational conditions. The furanocoumarins showed the strongest fig-derived binding but carry documented CYP450 and phototoxicity liabilities. The flavonoids showed moderate-to-weak binding, consistent with anti-inflammatory activity operating through mechanisms beyond direct COX-2 active-site inhibition. All six fig compounds were substantially weaker than oleocanthal, which remains the strongest predicted COX-2 binder across both compound sets.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.6 Drug-Likeness and ADMET Profiles\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePhysicochemical properties and ADMET profiles were computed using RDKit and standard Lipinski filters (Table 5). Six of eight compounds passed Lipinski's Rule of Five with zero violations, including bergapten and psoralen (strongest binders) and all flavonoids. Oleuropein failed with three violations (HBD \u0026gt; 5, HBA \u0026gt; 10, TPSA \u0026gt; 140 Ų), and rutin failed with four (MW \u0026gt; 500, HBD \u0026gt; 5, HBA \u0026gt; 10, TPSA \u0026gt; 140 Ų), typical of glycoside natural products.\u003c/p\u003e\n\u003cp\u003eBergapten (MW 220) and psoralen (MW 190) achieved excellent drug-likeness scores (QED 0.68 and 0.63, highest in the panel). Despite favorable physicochemistry, their CYP3A4 inhibition and phototoxicity preclude therapeutic use without scaffold redesign (Section 4.5).\u003c/p\u003e\n\u003cp\u003eHydroxytyrosol (MW 154, LogP 0.6) and elenolic acid (MW 294, LogP 1.7) have favorable physicochemistry but weak COX-2 binding.\u003c/p\u003e\n\u003cp\u003eLuteolin (MW 286, LogP 2.28) and quercetin (MW 302, LogP 1.99) satisfy Lipinski criteria and show reasonable bioavailability predictions, but moderate-to-weak COX-2 binding (−16.54 and −14.36 kcal mol⁻¹).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 5. Drug-likeness and predicted ADMET properties (test compounds).\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCompound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMW\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLogP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHBD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHBA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTPSA (Ų)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRot\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLipinski\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBioavail.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBBB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eQED\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBergapten\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e220.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e57.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePASS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHIGH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.682\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePsoralen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e190.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e48.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePASS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHIGH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.632\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLuteolin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e286.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e111.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePASS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHIGH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.511\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eQuercetin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e302.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e131.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePASS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHIGH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.434\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eElenolic acid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e294.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e78.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePASS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHIGH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.421\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHydroxytyrosol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e154.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e60.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePASS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHIGH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.547\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eOleuropein\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e468.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e183.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eFAIL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLOW\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.128\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRutin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e610.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e−1.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e269.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eFAIL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLOW\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.140\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eQED = quantitative estimate of drug-likeness (Bickerton et al. 2012). Excluded from MM-GBSA analysis (Section 3.3).\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e\u003cstrong\u003e4.1 Scoring-Function Sensitivity and Pose Validation Across Scaffold Classes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe smina/Vinardo consensus (Table 1b) stratified compounds by polarity. The three most polar (luteolin, quercetin, oleuropein) agreed across scoring functions (RMSD \u0026lt; 2 Å), while three hydrophobic compounds (psoralen, elenolic acid, hydroxytyrosol) diverged to an alternative site (RMSD 13–17 Å). Bergapten showed intermediate disagreement (5.9 Å).\u003c/p\u003e\n\u003cp\u003eThis divergence does not invalidate the Vina poses. Per-residue contact analysis confirmed that all three \"disagreement\" compounds retain 6–7 of their top-8 contacts within the COX-2 channel residue set (LEU320, GLY494, ALA495, SER321, VAL491, VAL317, SER498), confirming channel occupancy. Vina and Vinardo scores differed by less than 1 kcal mol⁻¹ (psoralen: −7.97 vs −7.4; elenolic acid: −6.35 vs −5.5; hydroxytyrosol: −6.23 vs −6.6), indicating marginal binding at both sites. The disagreement reflects a known limitation of empirical scoring functions: small, hydrophobic ligands with weak interactions generate shallow energy landscapes where different functions identify alternative local minima with comparable scores.\u003c/p\u003e\n\u003cp\u003eMM-GBSA calculations, incorporating explicit dynamics, solvation, and entropy absent from docking, provide more reliable ranking. For virtual screening of scoring-function-sensitive compounds, single-method docking is insufficient; post-docking refinement with MM-GBSA or equivalent is necessary.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.2 Desolvation-Driven Ranking Inversion and Structure–Activity Relationships\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe MM-GBSA ranking (bergapten \u0026gt; psoralen \u0026gt; luteolin \u0026gt; elenolic acid ≈ quercetin \u0026gt; hydroxytyrosol) inverts the Vina ranking (luteolin \u0026gt; quercetin \u0026gt; bergapten = psoralen \u0026gt; elenolic acid \u0026gt; hydroxytyrosol). The furanocoumarins surpass the flavonoids only after explicit-solvation rescoring. This inversion stems from desolvation penalties: luteolin and quercetin pay +25.0 and +21.8 kcal mol⁻¹ (ΔEGB), while bergapten and psoralen pay +20.3 and +15.6 kcal mol⁻¹. The hydrophobic furanocoumarin scaffold has lower desolvation cost, offsetting its weaker gas-phase interactions (ΔVDW + ΔEEL).\u003c/p\u003e\n\u003cp\u003eThree structural principles emerge:\u003c/p\u003e\n\u003cp\u003eFirst, hydrophobic contacts with VAL491, ALA495, and LEU320 form a consistent baseline (−0.9 to −2.7 kcal mol⁻¹ per residue) across all scaffolds, driven by van der Waals interactions. Flat aromatic systems (furanocoumarins, flavonoids) pack efficiently; small aliphatic compounds (elenolic acid, hydroxytyrosol) make weaker contacts.\u003c/p\u003e\n\u003cp\u003eSecond, desolvation of polar substituents is the dominant driver of the ranking inversion. The flavonoid catechol B-ring, with multiple hydroxyl groups, incurs a large ΔEGB penalty that offsets stronger hydrogen bonding. This generality matters for natural product COX-2 inhibitor design: hydroxyl-rich polyphenols rank well in docking but lose ground after desolvation correction.\u003c/p\u003e\n\u003cp\u003eThird, lipophilic substituents (methoxy in bergapten vs bare psoralen) fine-tune affinity by engaging secondary sites (e.g., SER498 contact: −0.87 vs −0.26 kcal mol⁻¹). Within a scaffold, minor modifications modulate channel binding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.3 Luteolin: Literature Cross-Validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe luteolin Vina score (−9.68 kcal mol⁻¹) aligns with published values (Derardja et al., Biol Life Sci Forum 2024;35(1):6, reported −9.49 kcal mol⁻¹). Hydrogen bonds with SER321 and TYR353 (canonical SER530 and TYR385) and van der Waals contacts with VAL491 and TRP355 (canonical VAL523 and TRP387) match published literature (Alam et al. 2016; Janakiramulu \u0026amp; Mamidala 2025; Javid et al. 2025), confirming our bound conformations.\u003c/p\u003e\n\u003cp\u003eDerardja et al. reported luteolin MM-GBSA = −43.41 kcal mol⁻¹, 2.6-fold stronger than our −16.54 kcal mol⁻¹. The discrepancy reflects differences in GB variant, force field, simulation length, and protein structure. This illustrates protocol sensitivity in absolute MM-GBSA values, supporting interpretation of our results in relative ranking rather than absolute affinity prediction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.4 Quercetin: Structural Similarity and Replicate-Dependent Variance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eQuercetin (Vina −9.60, MM-GBSA −14.36 ± 1.53) bound 2.2 kcal mol⁻¹ weaker than luteolin (−9.68 and −16.54 ± 0.88), despite similar docking scores. The two differ by 3-OH on the C-ring and an additional 3′-OH on the B-ring. Both contacted all six core residues with eight hydrogen bonds. The MM-GBSA difference emerged only at N=4: the first two replicates (main: −15.05, s1234: −15.32) suggested luteolin-like binding, but the s9999 replicate (−12.07) revealed a weaker mode, widening SD from 0.19 to 1.53 kcal mol⁻¹. The s9999 replicate showed atypical electrostatics (ΔEEL = +14.57 vs\u0026nbsp;−0.6 to −2.6 for others), indicating quercetin's additional 3-OH can adopt conformations where desolvation cost exceeds intramolecular hydrogen-bonding benefit. This replicate-dependence highlights the importance of multi-replicate sampling for polyphenols with multiple rotatable hydroxyl groups.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.5 Furanocoumarin Caveat: Binding Affinity Does Not Predict Therapeutic Utility\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBergapten (−20.32 kcal mol⁻¹) and psoralen (−18.95 kcal mol⁻¹) show the strongest COX-2 binding among test compounds, but their therapeutic potential is severely limited by off-target liabilities. Both are potent mechanism-based CYP3A4 inhibitors (bergapten: KI = 15.0 μM, kinact = 0.098 min⁻¹) and photosensitizers causing phototoxicity and photocarcinogenicity. These properties preclude systemic use regardless of COX-2 binding.\u003c/p\u003e\n\u003cp\u003eThe furanocoumarin advantage over flavonoids stems from lower desolvation penalties, suggesting a design principle: preserve hydrophobic contacts with GLY494, VAL491, and ALA495 while removing the photoreactive furan moiety to yield improved natural product COX-2 leads. Bergapten's methoxy group strengthens SER498 contact (−0.87 vs −0.26 kcal mol⁻¹ for psoralen), showing how minor substituents modulate channel binding without excessive polarity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.6 Glycoside Topology Limitation (Oleuropein, Rutin)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOleuropein and rutin were excluded from MM-GBSA due to a systematic gmx_MMPBSA conversion artifact. Both glycosides contain ≥11 GAFF2 c6 (sp3 ring carbon) atoms. The GROMACS→AMBER conversion yielded anomalous ΔG (\u0026gt;+300 and \u0026gt;+550 kcal mol⁻¹) across all replicates despite confirmed binding (minimum distance 1.5–2.0 Å throughout). The ΔVDW term (+264 to +292 kcal mol⁻¹) indicates incorrect Lennard-Jones cross-term assignment when CHARMM36 and GAFF2 parameters mix for large glycosides.\u003c/p\u003e\n\u003cp\u003eThis is a pipeline limitation, not an MD deficiency; backbone RMSD and ligand-pocket contacts were stable. Accurate glycoside MM-GBSA would require either a homogeneous force field (all-AMBER with GLYCAM) or manual VDW cross-term correction. Both compounds remain in docking (Section 3.1), where oleuropein showed good Vina (−6.31) and smina consensus (1.03 Å RMSD, AGREE), suggesting it is a genuine COX-2 binder whose binding affinity could not be quantified here.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.7 Concordance With Published Experimental Data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe computational ranking is directionally consistent with published experimental data. Luteolin has reported COX-2 IC₅₀\u0026nbsp;values in the low-micromolar range in enzymatic assays (PMC4702032; Javid et al., Comput Biol Chem 2025), and its computed ΔG (−16.54 kcal mol⁻¹) places it as the strongest fig flavonoid in our panel — consistent with its established experimental activity. Quercetin's weaker computed affinity (−14.36 kcal mol⁻¹) is consistent with its higher desolvation penalty and its lower potency relative to luteolin in head-to-head experimental comparisons. The directional agreement between experimental IC₅₀\u0026nbsp;rankings and our computational ΔG ranking — luteolin stronger than quercetin, both weaker than celecoxib — validates the screening workflow output.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.8 Limitations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMM-GBSA with igb=2 overestimates absolute binding affinities, particularly for hydrophobic pockets. All values should be interpreted relatively (against celecoxib and oleocanthal benchmarks run under identical conditions) rather than as Ki or IC₅₀\u0026nbsp;predictions. As a ranking validation, parallel igb=5 (OBC2) calculations on the same trajectories confirmed that all relative rankings are preserved under the alternative solvation model (full data in Supplementary Table S3).\u003c/p\u003e\n\u003cp\u003ePer-residue decomposition (within 8 Å cutoff) captures 58–87% of system-level ΔG, leaving 5–12 kcal mol⁻¹ from distant residues. Five-nanosecond trajectories suffice for binding-mode stability but do not capture slow conformational transitions. This study addresses only the COX-2 cyclooxygenase active site; allosteric, peroxidase-site, and COX-1 interactions are not evaluated in this phase.\u003c/p\u003e\n\u003cp\u003eAlphaFold 3 (Abramson et al., 2024) represents the current state of the art in protein–ligand structure prediction and would be a natural complement to this pipeline. However, AF3 predicts poses, not replicated thermodynamics: it does not produce ΔG values, per-residue decomposition, or ensemble-averaged dynamics. The MM-GBSA framework here — four independent MD trajectories, igb=2/igb=5 cross-validation, and decomposition identifying the desolvation-driven ranking inversion between furanocoumarins and flavonoids — provides information orthogonal to AF3 pose prediction. The sub-Ångström AF2/MD structural agreement (pocket RMSD 0.307 Å) already validates the simulation active-site geometry against an independent AI approach; AF3 would add a third confirmation of structural accuracy without resolving the thermodynamic questions central to this study. Incorporating AF3 poses as starting conformations for future MM-GBSA calculations would be a logical extension of this screening pipeline.\u003c/p\u003e\n\u003cp\u003eDespite these limitations, internal consistency is strong: stable RMSD, multi-replicate agreement (CV \u0026lt; 10% for most compounds), AF2/MD cross-methodology structural agreement, validated binding-site contacts, and directional concordance with published luteolin and quercetin experimental data all support the relative ranking produced by this computational screening phase.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.8 Implications for Fig Bioactivity and Dietary Anti-Inflammatory Activity\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo tested fig-derived compound matched celecoxib in predicted COX-2 binding under these conditions, yet this does not contradict the well-documented anti-inflammatory properties of fig extracts. Instead, it suggests that dietary fig anti-inflammatory effects operate through mechanisms beyond direct COX-2 active-site inhibition. Plausible mechanisms include NF-κB pathway suppression (documented for luteolin and quercetin), antioxidant reduction of prostaglandin precursors, and polypharmacology across multiple targets at dietary concentrations.\u003c/p\u003e\n\u003cp\u003eNone of the tested fig compounds showed predicted COX-2 affinity approaching oleocanthal (companion paper [2]), suggesting that the documented anti-inflammatory effects of fig consumption may operate through distinct molecular targets or combination mechanisms rather than direct COX-2 active-site inhibition. The furanocoumarins, despite stronger computed COX-2 affinity, carry documented CYP450 and phototoxicity liabilities that would complicate therapeutic development. Taken together, these findings reinforce that binding affinity alone is insufficient to predict therapeutic utility in natural product drug discovery.\u003c/p\u003e\n\u003cp\u003eThe flavonoids (luteolin, quercetin) retain value as dietary anti-inflammatory components through polypharmacological mechanisms, despite modest direct COX-2 affinity.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author acknowledges the AlphaFold Protein Structure Database (EMBL-EBI) for the AlphaFold2 prediction used in structural validation, the RCSB Protein Data Bank for the COX-2 crystal structure (PDB: 5KIR), and the developers of GROMACS, gmx_MMPBSA, AutoDock Vina, and AmberTools for making their software freely available. This study is a companion to a parallel investigation of olive-derived secoiridoids against COX-2 [2].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFaycal conceived the study, designed the computational workflow, performed all molecular docking, molecular dynamics simulations, MM-GBSA binding free energy calculations, per-residue decomposition analysis, AlphaFold2 structural validation, and data analysis, and wrote the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research received no external funding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author declares no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data generated during this study are available in a Zenodo repository at https://doi.org/10.5281/zenodo.19076569. This includes: raw MM-GBSA output files (FINAL_RESULTS_DECOMP.dat, ENERGIES_DECOMP.csv) for all six fig compounds and celecoxib across all replicates, igb=5 cross-validation results, GROMACS topology files, MDP parameter files, molecular docking input/output files, AlphaFold2 validation data, and analysis scripts. Production trajectory files are available from the corresponding author upon reasonable request due to their large size.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll analysis scripts, including the AlphaFold2 validation script, MM-GBSA batch processing scripts, contact analysis code, and figure generation scripts, are deposited in the Zenodo repository. The study used the following publicly available software: GROMACS 2026 (https://www.gromacs.org), gmx_MMPBSA v1.6.4 (https://github.com/Valdes-Tresanco-MS/gmx_MMPBSA), AutoDock Vina 1.2 (https://github.com/ccsb-scripps/AutoDock-Vina), and AmberTools25 (https://ambermd.org).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSmith WL, DeWitt DL, Garavito RM. Cyclooxygenases: structural, cellular, and molecular biology. Annu Rev Biochem. 2000;69:145\u0026ndash;82.\u003c/li\u003e\n\u003cli\u003eRouzer CA, Marnett LJ. Cyclooxygenases: structural and functional insights. J Lipid Res. 2009;50 Suppl:S29\u0026ndash;34.\u003c/li\u003e\n\u003cli\u003eVane JR, Bakhle YS, Botting RM. Cyclooxygenases 1 and 2. Annu Rev Pharmacol Toxicol. 1998;38:97\u0026ndash;120.\u003c/li\u003e\n\u003cli\u003eLaine L. Gastrointestinal effects of NSAIDs and coxibs. J Pain Symptom Manage. 2003;25(2 Suppl):S32\u0026ndash;40.\u003c/li\u003e\n\u003cli\u003eBresalier RS et al. Cardiovascular events associated with rofecoxib in a colorectal adenoma chemoprevention trial. N Engl J Med. 2005;352:1092\u0026ndash;102.\u003c/li\u003e\n\u003cli\u003eYoon JH, Baek SJ. Molecular targets of dietary polyphenols with anti-inflammatory properties. Yonsei Med J. 2005;46:585\u0026ndash;96.\u003c/li\u003e\n\u003cli\u003eAzab A, Nassar A, Azab AN. Anti-inflammatory activity of natural products. Molecules. 2016;21:1321.\u003c/li\u003e\n\u003cli\u003eSubehan et al. Bergapten CYP3A4 mechanism-based inhibition. J Agric Food Chem 2007. (PMID 17988092)\u003c/li\u003e\n\u003cli\u003eAlam MA et al. Luteolin and COX-2 docking studies. (PMC4702032)\u003c/li\u003e\n\u003cli\u003e Janakiramulu P \u0026amp; Mamidala E. Flavonoid derivatives vs COX-2. In Silico Pharmacol 2025;13:59. DOI:10.1007/s40203-025-00349-x (PMID 40255260)\u003c/li\u003e\n\u003cli\u003e Javid R et al. Luteolin MM-GBSA against COX-2. Comput Biol Chem 2025;118:108499. DOI:10.1016/j.compbiolchem.2025.108499 (PMID 40347541)\u003c/li\u003e\n\u003cli\u003e Derardja I et al. Biol Life Sci Forum 2024;35(1):6. DOI:10.3390/blsf2024035006 [conference proceedings]\u003c/li\u003e\n\u003cli\u003e Veberic R, Colaric M, Stampar F. Phenolic acids and flavonoids of fig fruit (\u003cem\u003eFicus carica\u003c/em\u003e L.) in the northern Mediterranean region. Food Chem. 2008;106:153\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003e Badgujar SB et al. \u003cem\u003eFicus carica\u003c/em\u003e Linn.: a review of its pharmacology, phytochemistry and traditional uses. J Ethnopharmacol. 2014;154:183\u0026ndash;209.\u003c/li\u003e\n\u003cli\u003e Garavito RM, DeWitt DL. The cyclooxygenase isoforms: structural insights into the conversion of arachidonic acid to prostaglandins. Biochim Biophys Acta. 1999;1441:213\u0026ndash;20.\u003c/li\u003e\n\u003cli\u003e Kurumbail RG et al. Structural basis for selective inhibition of cyclooxygenase-2 by anti-inflammatory agents. Nature. 1996;384:644\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003e Malkowski MG et al. The productive conformation of arachidonic acid bound to prostaglandin synthase. Science. 2000;289:1933\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003e Luong C et al. Flexibility of the NSAID binding site in the structure of human cyclooxygenase-2. Nat Struct Biol. 1996;3:927\u0026ndash;33.\u003c/li\u003e\n\u003cli\u003e Genheden S, Ryde U. The MM/PBSA and MM/GBSA methods to estimate ligand-binding affinities. Expert Opin Drug Discov. 2015;10:449\u0026ndash;61.\u003c/li\u003e\n\u003cli\u003e Wang E et al. End-point binding free energy calculation with MM/PBSA and MM/GBSA: strategies and applications in drug design. Chem Rev. 2019;119:9478\u0026ndash;508.\u003c/li\u003e\n\u003cli\u003e Vald\u0026eacute;s-Tresanco MS et al. gmx_MMPBSA: a new tool to perform end-state free energy calculations with GROMACS. J Chem Theory Comput. 2021;17:6281\u0026ndash;91.\u003c/li\u003e\n\u003cli\u003e Abramson J et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature. 2024;630:493\u0026ndash;500. DOI:10.1038/s41586-024-07487-w\u003c/li\u003e\n\u003cli\u003eFerhat, F. Computational Screening of Olive Secoiridoids Against COX-2: Oleocanthal Exhibits Strong Predicted Binding via Novel MET522 Side-Pocket Engagement. Scientific Reports (2026, under review).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"COX-2, figs, molecular dynamics, MM-GBSA, binding free energy, flavonoids, furanocoumarins, natural products","lastPublishedDoi":"10.21203/rs.3.rs-9155203/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9155203/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground.\u003c/h2\u003e \u003cp\u003eFigs (\u003cem\u003eFicus carica\u003c/em\u003e) have long been used in traditional medicine for anti-inflammatory effects, and they contain polyphenolic compounds with structural diversity across flavonoid and furanocoumarin scaffolds. Whether these compounds directly inhibit COX-2 \u0026mdash; a key target in inflammation \u0026mdash; has not been systematically tested with molecular dynamics and binding free-energy calculations.\u003c/p\u003e\u003ch2\u003eMethods.\u003c/h2\u003e \u003cp\u003eWe applied a three-tier computational screening workflow to six fig-derived compounds (flavonoids: luteolin, quercetin; furanocoumarins: bergapten, psoralen; secoiridoid derivative: elenolic acid; phenol: hydroxytyrosol): (1) AutoDock Vina docking with smina/Vinardo scoring consensus; (2) 5 ns all-atom MD simulation (GROMACS 2026, CHARMM36\u0026thinsp;+\u0026thinsp;GAFF2, 4 independent replicates per compound); (3) MM-GBSA binding free-energy calculation (igb\u0026thinsp;=\u0026thinsp;2, saltcon\u0026thinsp;=\u0026thinsp;0.150 M, gmx_MMPBSA v1.6.4) benchmarked against celecoxib. Structural reliability of the COX-2 template was verified by AlphaFold2 cross-validation (global RMSD 0.38 \u0026Aring;, pocket RMSD 0.25 \u0026Aring; vs AF-P35354-F1 raw prediction). Oleocanthal from the companion study served as an upper-bound reference. Two glycoside compounds (oleuropein, rutin) underwent docking only due to force-field conversion limitations.\u003c/p\u003e\u003ch2\u003eResults.\u003c/h2\u003e \u003cp\u003eAll six tested fig-derived compounds bound less favourably than celecoxib in this computational model (ΔΔG range: +18.6 to +\u0026thinsp;28.5 kcal mol⁻\u0026sup1;, Welch t-test p\u0026thinsp;\u0026le;\u0026thinsp;8.5\u0026times;10⁻⁷ for all comparisons), suggesting that none of the compounds screened here approaches celecoxib-level COX-2 active-site affinity under these conditions. Within the test panel, furanocoumarins ranked above flavonoids: bergapten (ΔΔG\u0026thinsp;=\u0026thinsp;+\u0026thinsp;18.56 vs celecoxib; absolute ΔG\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;20.32\u0026thinsp;\u0026plusmn;\u0026thinsp;0.97 kcal mol⁻\u0026sup1;) and psoralen (ΔΔG\u0026thinsp;=\u0026thinsp;+\u0026thinsp;19.93; \u0026minus;18.95\u0026thinsp;\u0026plusmn;\u0026thinsp;1.16) outranked luteolin (\u0026minus;\u0026thinsp;16.54\u0026thinsp;\u0026plusmn;\u0026thinsp;0.88) and quercetin (\u0026minus;\u0026thinsp;14.36\u0026thinsp;\u0026plusmn;\u0026thinsp;1.53) despite weaker Vina docking scores \u0026mdash; a desolvation-driven ranking inversion. Elenolic acid (\u0026minus;\u0026thinsp;14.88\u0026thinsp;\u0026plusmn;\u0026thinsp;1.02) and hydroxytyrosol (\u0026minus;\u0026thinsp;10.34\u0026thinsp;\u0026plusmn;\u0026thinsp;0.56) were the weakest binders. Although bergapten and psoralen show the strongest fig-derived binding in this model, both are known CYP3A4 inhibitors and phototoxic agents, raising concerns that would limit their direct therapeutic use.\u003c/p\u003e\u003ch2\u003eConclusions.\u003c/h2\u003e \u003cp\u003eThis computational screening phase indicates that none of the six tested fig-derived compounds approaches celecoxib potency at the COX-2 active site in this model, suggesting that experimental focus for fig bioactives may be better directed toward alternative anti-inflammatory mechanisms. The desolvation-driven ranking inversion between furanocoumarins and flavonoids is a methodologically important finding: Vina docking scores would have predicted flavonoids as the stronger binders, while physics-based MM-GBSA correctly captures the desolvation penalty that reverses this ranking \u0026mdash; demonstrating why single-tier docking is insufficient for polar natural products. All raw data are deposited on Zenodo for full reproducibility.\u003c/p\u003e","manuscriptTitle":"Computational Evaluation of Fig-Derived Bioactive Compounds as COX-2 Inhibitors: Molecular Dynamics Reveals Desolvation-Driven Ranking Inversion Between Flavonoids and Furanocoumarins","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-13 18:32:39","doi":"10.21203/rs.3.rs-9155203/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"249121503146196594912526703076006654060","date":"2026-05-01T13:53:07+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"69481287093903786439408651956864426279","date":"2026-04-06T14:01:40+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-06T13:06:09+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-25T08:34:44+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-21T07:35:26+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-21T07:35:14+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-03-18T06:04:15+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"e4352fab-d059-44fa-935c-2791e387c73c","owner":[],"postedDate":"April 13th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"249121503146196594912526703076006654060","date":"2026-05-01T13:53:07+00:00","index":75,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":65947434,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":65947435,"name":"Biological sciences/Drug discovery"}],"tags":[],"updatedAt":"2026-04-13T18:32:39+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-13 18:32:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9155203","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9155203","identity":"rs-9155203","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}