A Molecular-Dynamics-Guided Screening Pipeline Identifies Topical Glucocorticoid Receptor Modulators with Selective Epidermal Activation Authors: Feng-Chuan Tai, Heng-Hui Lien, Shih-Chang Lin, Yu-Li Hung, Yu-Ping Cheng, Jiun-Wen Guo Published: 2026 ABSTRACT AbstractIntroductionTopical corticosteroids remain the mainstay of psoriasis management but are limited by skin atrophy and barrier impairment. Developing safer, steroid-sparing therapy represents an unmet need in topical pharmacology. The glucocorticoid receptor (GR) plays a central role in epidermal homeostasis, yet selective activation by natural phytochemicals has not been systematically explored.MethodsA molecular modeling framework integrating docking, molecular dynamics simulations, and pharmacophore conservation was established to evaluate GR engagement by four phytochemicals – sakuranetin, salvianolic acid B, lithospermic acid, and magnolol. Computational findings were compared with immunohistochemical analysis of total GR and phosphorylated Ser211-GR in psoriatic epidermal specimens derived from residual archived tissues.ResultsSalvianolic acid B and lithospermic acid displayed high docking affinity and stable retention of canonical GR contacts, whereas sakuranetin and magnolol exhibited moderate but consistent binding. Immunohistochemistry confirmed preserved GR abundance for all compounds, with lithospermic acid producing the greatest increase in Ser211 phosphorylation, consistent with differential receptor-associated signaling patterns. Computational and histological findings showed preliminary directional concordance between computational interaction profiles and tissue-level receptor-associated readouts.ConclusionThis study demonstrates that integrating molecular modeling with receptor-associated histological analysis may assist in prioritizing natural compounds with potential GR-engagement properties. The proposed framework provides an exploratory and reproducible strategy for receptor-engagement prioritization in topical pharmacology and may support future investigation of bioactive phytochemicals in dermatologic research. Introduction Psoriasis is a common chronic inflammatory skin disease characterized by epidermal hyperplasia, impaired barrier function, and dysregulated immune responses, notably involving the IL-23/IL-17 axis [1, 2]. Beyond cutaneous manifestations, psoriasis is associated with systemic comorbidities including psoriatic arthritis, metabolic syndrome, and psychological disorders [1, 3]. Despite major therapeutic advances with biologic agents targeting TNF-α, IL-17, and IL-23, topical corticosteroids remain the mainstay of treatment for mild-to-moderate disease because of their rapid anti-inflammatory efficacy and clinical accessibility [4, 5]. However, chronic or extensive use can cause skin atrophy, barrier disruption, and systemic absorption, especially at sensitive anatomical sites [4]. These limitations underscore the need for safer steroid-sparing therapy for long-term management [6, 7]. Natural phytochemicals, particularly polyphenols such as flavonoids and phenolic acids, possess antioxidative, anti-inflammatory, and immunomodulatory properties and have demonstrated anti-psoriatic potential in preclinical models [7, 8]. Nevertheless, their molecular targets and receptor-level mechanisms remain largely undefined. Among the key regulators of skin homeostasis, the glucocorticoid receptor (GR) – a ligand-activated nuclear receptor governing inflammation, keratinocyte differentiation, and barrier integrity – emerges as a critical pharmacologic target [9, 10]. Selective GR modulators have been developed to retain anti-inflammatory efficacy while reducing steroid-related adverse effects, with vamorolone receiving regulatory approval in 2024 [11]. Notably, recent evidence suggests that certain phytochemicals – such as baicalein and agathisflavone – may interact with GR-associated signaling pathways and exhibit selective anti-inflammatory activity [12, 13], supporting the rationale for further investigation of phytochemical-GR interactions in topical pharmacology. Based on this framework, we focused on four natural compounds – sakuranetin, salvianolic acid B, lithospermic acid, and magnolol – that have shown therapeutic efficacy in imiquimod-induced psoriasis-like murine models [14–17]. These studies reported improvements in epidermal architecture, suppression of IL-23/Th17-mediated inflammation, and restoration of skin-barrier-related gene expression. While their in vivo benefits have been demonstrated, the receptor-mediated molecular mechanisms underlying these effects remain unclear. Given their partial structural similarity to steroidal scaffolds, we hypothesized that the anti-psoriatic activity of these phytochemicals may involve GR-associated receptor engagement. To test this hypothesis, we developed an integrative molecular modeling workflow combining molecular docking, molecular dynamics (MD) simulations, and protein-ligand interaction profiling into a composite GR-score that captures both structural and dynamical features of receptor engagement. The approach was benchmarked against epidermal immunohistochemistry of total GR and Ser211-phosphorylated GR, a marker of transcriptionally active receptor, in archived psoriatic skin specimens. The goal of this study was to establish a computational-experimental framework for prioritizing natural compounds with potential GR-engagement properties by linking in silico interaction profiling with receptor-associated histological readouts in psoriatic skin. Methods Software and Computational Environment All computational procedures in this study were conducted on a Linux-based workstation equipped with NVIDIA RTX-series GPUs for hardware acceleration. Molecular docking simulations were performed using AutoDock Vina (version 1.2.3), operated through the PyRx (version 0.8) graphical interface. Ligand structure conversion and geometry optimization were carried out using Open Babel (version 3.1.1), while protein structure preprocessing, including hydrogen addition, solvation, and ion neutralization, was conducted using PDBFixer (version 1.8.1). MD simulations were executed using OpenMM (version 8.0.0), with parameterization of protein-ligand systems based on the AMBER14SB force field for proteins and the General AMBER Force Field (GAFF) for ligands, generated via Antechamber and AmberTools23. Post-simulation trajectory analysis, including root mean square deviation (RMSD), root mean square fluctuation (RMSF), and hydrogen-bond occupancy calculations, was conducted using MDTraj (version 1.9.7) and PyTraj within the AmberTools suite. Protein-ligand interaction profiles were further examined using the Protein-Ligand Interaction Profiler (PLIP; an open-source web-based tool). All structural visualizations and interaction maps were rendered using PyMOL (version 2.5) and Discovery Studio Visualizer (version 21.1). Python 3.10 was used for script execution and automation. Protein Preparation The crystal structure of the human GR-LBD complexed with dexamethasone (DEX; PDB ID: 1M2Z) was obtained from the RCSB Protein Data Bank. Structural preprocessing was conducted using PyRx 0.8, incorporating Open Babel and AutoDock Tools. All water molecules, heteroatoms, and native ligands were removed. Polar hydrogens were added, and Kollman charges were assigned. The prepared GR structure was saved in PDBQT format for docking simulations. Ligand Preparation Seven compounds were selected for analysis: sakuranetin (CID: 73571), salvianolic acid B (CID: 6451084), lithospermic acid (CID: 6441498), magnolol (CID: 72300), and metformin (CID: 4091) as candidate or control molecules. Among them, desoximetasone (CID: 5311067), a clinically used topical corticosteroid, was employed as a positive control, while mifepristone (RU486, CID: 55245) served as a benchmark GR antagonist. In contrast, metformin – a clinically relevant non-steroidal small molecule lacking a steroid-like ring scaffold and without established GR-binding characteristics – was included as an exploratory negative comparator for receptor-engagement analysis. However, given its relatively small size and high polarity, metformin may exhibit inherently limited pocket retention in docking- and MD-based analyses compared with steroid-like ligands. Three-dimensional structures of these compounds were obtained from the PubChem database. For salvianolic acid B, which lacked a resolved 3D conformation, its 2D structure was converted to 3D using Open Babel. All ligands were subjected to geometry optimization using the Universal Force Field in PyRx and then saved in PDBQT format for molecular docking simulations. Molecular Docking Molecular docking simulations were conducted using AutoDock Vina integrated within PyRx. The docking search space was automatically generated within the PyRx environment using the prepared GR-LBD structure from PDB ID: 1M2Z as the receptor reference, without manual adjustment of grid-box dimensions or center coordinates. Default AutoDock Vina parameters implemented by PyRx were used unless otherwise specified, with the exhaustiveness parameter set to 8. For each ligand, the top-ranked pose based on binding affinity (kcal/mol) was selected for downstream analyses. Visualization and preliminary interaction analysis were performed using PyMOL and Discovery Studio Visualizer. Two-Dimensional Residue Interaction Mapping Two-dimensional residue interaction diagrams were generated to compare GR-binding residues across ligands. Key steroid-binding residues were annotated according to previous GR crystallographic studies [9, 10]. Metformin, used as the negative control, was not included in the two-dimensional interaction-map analysis because it could not be reliably selected or parsed as a ligand within the Discovery Studio Visualizer environment. MD Simulation To evaluate the conformational stability and receptor-engagement behavior of GR-ligand complexes, MD simulations were performed using a unified ligand-alignment workflow. Three-dimensional ligand structures were retrieved from the PubChem database or generated using RDKit (v2022.09) followed by MMFF94 energy minimization. Ligands were aligned to their respective AutoDock Vina docking poses using rigid or torsion-adjusted superposition based on maximum common substructure (MCS) matching and Kabsch alignment. The resulting poses were subsequently oriented relative to GR-bound DEX (residue A-301) using three or four reference atoms (C5, C8, O1, and O3). For ligands lacking oxygen atoms, such as metformin, chemically analogous atoms were selected to enable spatial superposition. GR-ligand docking complexes were preprocessed using PDBFixer (v1.8.1), including removal of nonstandard residues and solvent molecules, correction of missing atoms, and hydrogen addition at physiological pH (7.0). The resulting complexes were loaded into OpenMM’s Modeler class for system preparation. The AMBER14SB force field was applied to proteins, while ligand parameters were generated using the Open Force Field Toolkit (v0.11.5) with GAFFTemplateGenerator. Systems were solvated in cubic TIP3P water boxes with 10 Å padding and neutralized using 0.15 m NaCl. Production MD simulations were conducted using OpenMM (v8.0.0) with a Langevin integrator at 300 K, a friction coefficient of 1 ps−1, and a 2 fs timestep under periodic boundary conditions with CUDA GPU acceleration. Each system underwent a 100-ns production simulation (50,000,000 steps), with trajectory frames saved every 5,000 steps (10 ps/frame) as .dcd and .pdb files. Simulation metadata, including runtime, frame intervals, and total simulation steps, were retained for downstream analyses. Representative scripts used for ligand-pose alignment, MD system preparation, trajectory simulation, and downstream structural analyses are provided in supplementary materials. Trajectory and Interaction Analysis Post-simulation analyses were conducted using MDTraj (v1.9.7) and Python scripts customized for high-throughput evaluation of ligand-GR binding dynamics. Trajectories were loaded from .dcd files using the corresponding solvated .pdb as topology reference. Analyses included the following. (1) RMSD of the GR backbone and ligand heavy atoms to evaluate structural stability over time. RMSD values were computed using the initial frame as reference and plotted as time-series graphs. (2) RMSF of Cα atoms was calculated for each residue to assess flexibility, with special emphasis on residues 500–650 encompassing the GR ligand-binding domain. Output CSV files included both full-range and zoom-in RMSF profiles. (3) Radius of gyration (Rg) was used to monitor global protein compactness across time. (4) Protein-ligand center distance was measured by computing the Euclidean distance between the geometric centers of the protein and ligand at each frame, providing an index of ligand retention in the binding pocket. Protein-Ligand Interaction Profiling The protein-ligand interaction networks were systematically evaluated using the PLIP, a web-based open-source platform [18, 19]. For each GR-ligand complex, PLIP analyses were conducted at two key stages: (1) the initial docking-aligned conformation prior to MD simulation and (2) the final frame extracted after 100 ns of MD refinement. PLIP identified pharmacologically relevant non-covalent interactions, including hydrogen bonds, hydrophobic contacts, π-π stacking, and salt bridges. This comparative profiling strategy enabled assessment of interaction persistence and structural stability throughout the simulation trajectory. Additionally, the co-crystallized structure of DEX bound to GR (PDB ID: 1M2Z) was analyzed via PLIP in its native conformation (without MD processing) to serve as a canonical reference. The resulting interaction fingerprint of DEX provided a structural benchmark for evaluating the degree of mimicry achieved by each test compound. All interaction profiles were subsequently tabulated to identify conserved or critical contact residues across the compound panel. Composite GR-Score for Receptor-Engagement Profiling To integrate structural, docking-derived, and MD-associated features related to GR-binding and receptor-engagement patterns, we defined a transparent linear composite score termed the GR-score:GR‐score=Structure‐score+Dock‐score+MD‐score Structure-Score quantifies steroid-likeness from ring architecture and core unsaturation:Structure−score=0.20FR+0.1TR+0.2CO−0.05DBcore Here, Fused Ring (FR) count is the number of rings in the largest edge-fused system (rings share a bond; spiro junctions are excluded). A canonical steroid nucleus corresponds to FR = 4 (6–6–6–5). Total Ring (TR) count is the total number of rings in the molecule, including non-fused/peripheral rings. Carbonyl (CO) count is the number of in-core C=O groups within the fused steroidal nucleus, counted as pharmacophoric features that stabilize GR binding. DB_core (in-core double bond count) is the tally of in-ring C=C within the fused core. Exocyclic C=C and carbonyls are excluded since they do not aromatize the ring. Dock-score incorporates AutoDock Vina affinity and pose pharmacophore agreement:Dock‐score=0.06‐E_vina+0.08N_dock E_vina is the AutoDock Vina binding affinity (kcal/mol) of the lowest-energy docking pose; we use −E_vina so that more favorable (more negative) energies contribute positively. N_dock is the residue-wise exact match count between the docking pose’s receptor-ligand interaction profile and a predefined DEX canonical-contact set (e.g., ASN564, GLN570, ARG611, GLN642, PHE623, TYR735, THR739); a residue contributes one count if at least one standard geometric interaction (hydrogen bond, ionic/salt bridge, hydrophobe, or π/CH-π contact) is present in the docking pose. MD-score measures post-relaxation retention of the docked pharmacophore:MD−score=−2.0RMSD_mean−0.3Disp_100ns+0.20N_post RMSD_mean is the mean heavy-atom ligand RMSD (nm) over 0–100 ns relative to the minimized docking pose, emphasizing pose stability. Disp_100ns is the ligand center-of-mass displacement (nm) at the 100-ns endpoint relative to the starting pose, reflecting endpoint drift. N_post is the residue-wise exact match count to the same DEX canonical-contact set evaluated over the terminal MD window (e.g., the last 10 ns averaged), quantifying pharmacophore conservation after dynamics. Coefficient weights were calibrated with canonical steroids (cortisol, desoximetasone, RU486) as positive anchors and polyphenolic scaffolds as non-steroidal references. The chosen weights ensure that steroidal ligands generally exceed unity under hand-countable rules; while docking and MD features incrementally refine receptor-engagement and binding-stability characteristics. Under this construction, GR-score >1 denotes a steroid-like chemotype with stronger composite receptor-engagement and binding-stability features, whereas GR-score <1 indicates weaker overlap with canonical GR-associated interaction patterns. Regression and Correlation Analysis Associations between GR-score and the experimental pSer211/total GR ratio were evaluated by linear regression. Goodness-of-fit (R2), slope, intercept, and 95% confidence intervals were calculated using GraphPad Prism (v8.0.2). Non-parametric rank correlation was assessed by Spearman’s ρ with two-sided significance (p < 0.05). Tissue Source and Ethical Statement Archived paraffin-embedded dorsal skin tissues were obtained from prior imiquimod-induced psoriasis studies conducted under Cathay General Hospital IACUC approvals (CGH-IACUC-107-028, 108-006, 110-025, 112-024). No new animal experiments were performed. The archived tissues were derived from treatment groups involving sakuranetin, salvianolic acid B, lithospermic acid, magnolol, or desoximetasone under previously published and comparable imiquimod-induced psoriasiform dermatitis protocols. For each phytochemical, archived tissues were selected from treatment conditions previously reported to demonstrate anti-psoriatic activity within the respective experimental framework. All studies employed comparable 7-day topical treatment frameworks with dorsal application to similar skin regions (∼4 × 6 cm2) followed by endpoint tissue collection on Day 7. While vehicle systems differed according to compound formulation requirements (e.g., microemulsion or ethanol-based formulations), residual dorsal skin samples were selected based on consistent psoriasiform histopathology confirmed by hematoxylin-eosin staining. All paraffin sections were re-analyzed specifically for GR and phospho-GR (Ser211) immunohistochemistry under unified staining, imaging, and digital quantification workflows to minimize inter-experimental variability. The reuse of residual tissues for receptor pathway verification complied with institutional ethical standards and ARRIVE guidelines. Immunohistochemical Staining of GR and Phospho-GR (Ser211) Formalin-fixed, paraffin-embedded dorsal skin sections (4 µm thickness) were subjected to immunohistochemical staining to evaluate total GR expression and ligand-induced GR activation. After deparaffinization and rehydration, heat-induced antigen retrieval was performed in citrate buffer (pH 6.0) using a microwave system. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide for 10 min. Primary antibodies included anti-GR (polyclone, SAB4501309, Sigma-Aldrich, MA, USA, dilution: 1:100) to detect total GR, and anti-phospho-GR (Ser211) (polyclone, PA5-99335, Invitrogen, MA, USA, dilution: 1:100) to detect Ser211-phosphorylated GR, a receptor-associated activation marker. Slides were incubated with primary antibodies at 4°C overnight, followed by secondary antibody incubation using the ultraView Universal DAB Detection Kit (Ventana Medical Systems Inc.). Visualization was performed using DAB as chromogen, and slides were counterstained with hematoxylin. All stained sections were scanned at 200× magnification using a high-resolution slide scanner (3DHISTECH Pannoramic 250, Budapest, Hungary) for digital analysis. Immuno-Intensity Quantification and H-Score Analysis Digital images of GR and phospho-GR-stained sections were analyzed using CellQuant software (3DHISTECH Ltd.), which enables automated quantification of immuno-intensity across designated regions of interest. The H-score was calculated as the product of staining intensity (0 = no staining, 1 = weak, 2 = moderate, 3 = strong) and the percentage of positively stained cells at each intensity level. The final H-score ranged from 0 to 300, as per established quantification protocols. At least five non-overlapping high-power fields were selected per sample for quantification. The H-scores of total GR and phospho-GR (Ser211) were compared across all treatment groups, including desoximetasone, and test compound-treated mice (e.g., sakuranetin, lithospermic acid, magnolol, and salvianolic acid B). Statistical Analysis All data are presented as mean ± standard deviation. GR and GR-Ser211 immunohistochemical staining was semi-quantitatively assessed using the H-score method. Statistical differences in H-scores among treatment groups were analyzed using one-way analysis of variance (ANOVA), followed by Scheffé’s post hoc test for multiple comparisons. A p value <0.05 was considered statistically significant. All statistical analyses were conducted using SPSS version 20.0 (IBM Corp., Armonk, NY, USA). Bar graphs were generated using GraphPad Prism version 8.0.2 (GraphPad Software, San Diego, CA, USA). Results Ligand and Target Selection for Docking Simulation Based on prior in vivo studies demonstrating anti-psoriatic efficacy, four plant-derived small molecules were selected for molecular docking analysis. These included sakuranetin (CID: 73571), salvianolic acid B (CID: 6451084), lithospermic acid (CID: 6441498), and magnolol (CID: 72300) – each previously reported to exert therapeutic effects in imiquimod-induced murine models of psoriasis. In addition, desoximetasone (CID: 5311067), a clinically used topical corticosteroid, was included as a positive reference control, while mifepristone (RU486, CID: 55245) was employed as a benchmark GR antagonist, serving to contrast binding patterns and identify potential inverse or noncanonical interactions. To establish a negative reference for GR binding, metformin (CID: 4091), a small molecule with no reported affinity for steroid receptors, was included. This allowed comparative evaluation of ligand-receptor dynamics, enabling distinction between pharmacologically active ligands and structurally unrelated compounds with minimal GR engagement. The docking target was the LBD of the human GR (PDB ID: 1M2Z), co-crystallized with DEX, which provides a high-resolution structure of the steroid-binding pocket. All ligands were geometry-optimized and converted to PDBQT format using PyRx with Universal Force Field minimization, ensuring compatibility with AutoDock Vina-based docking simulations. A summary of the selected compounds, including their sources, PubChem CIDs, structure types, and designated roles in this study, is provided in Table 1. Table 1.Compounds and target structure used for GR docking simulationCompound nameSourcePubChem CID/PDB IDStructure typeUsed asGR ligand-binding domainHuman receptor protein1M2ZX-ray crystalDocking targetDesoximetasoneCorticosteroid53110673DPositive controlMifepristone (RU486)Synthetic steroid552453DGR antagonist controlSalvianolic acid BPolyphenol64510842D to 3DCandidate ligandLithospermic acidPhenolic acid64414983DCandidate ligandSakuranetinFlavonoid735713DCandidate ligandMagnololBiphenol723003DCandidate ligandMetforminBiguanide40913DNegative control Compounds and target structure used for GR docking simulation Binding Affinities and Structural Interaction Profiles of GR-Ligand Complexes To evaluate the potential of phytochemicals as GR modulators, molecular docking simulations were performed using AutoDock Vina. Desoximetasone, a potent synthetic glucocorticoid, exhibited a predicted binding affinity of −7.6 kcal/mol. RU486 (mifepristone), a benchmark GR antagonist, demonstrated the highest binding affinity at −8.3 kcal/mol. Among the plant-derived compounds, salvianolic acid B matched RU486 in docking score (−8.3 kcal/mol), followed by lithospermic acid (−7.9 kcal/mol), sakuranetin (−7.1 kcal/mol), and magnolol (−6.9 kcal/mol). In contrast, metformin – used as a negative control – displayed a substantially weaker affinity (−5.0 kcal/mol), consistent with its lack of known GR-binding activity. Three-dimensional surface renderings of GR-ligand complexes (Fig. 1) illustrate distinct binding orientations and insertion depths for each compound. Desoximetasone (Fig. 1a) occupied a deep position within the canonical ligand-binding cavity formed by helices H3, H5, and H7, consistent with its strong predicted affinity. RU486 (Fig. 1b), while also penetrating the binding pocket, exhibited a lateral shift toward the H12 region, which may reflect its antagonist profile. Salvianolic acid B (Fig. 1c) showed a well-anchored pose with broad surface contact, matching its high docking score. Lithospermic acid (Fig. 1d) adopted a similarly deep conformation, reinforcing its favorable GR engagement. Sakuranetin (Fig. 1e) and magnolol (Fig. 1f) resided closer to the pocket entrance, with less hydrophobic embedding, potentially explaining their moderate affinities. Metformin (Fig. 1g), the negative control, failed to stably occupy the pocket, remaining largely solvent exposed. Predicted GR-ligand binding poses reveal differential pocket engagement. Surface renderings of GR-ligand complexes (PDB ID: 1M2Z) show the docking positions of desoximetasone (a), RU486 (b), salvianolic acid B (c), lithospermic acid (d), sakuranetin (e), magnolol (f), and metformin (g). Ligands differ in both orientation and insertion depth within the canonical ligand-binding cavity, reflecting variations in predicted receptor affinity and binding mode. Two-dimensional interaction diagrams (Fig. 2) illustrated distinct receptor-ligand contact patterns across the docked compounds. Desoximetasone (Fig. 2a) formed conventional hydrogen bonds with GLU542 and TYR545, supplemented by two water-mediated hydrogen bonds involving HOH80 and HOH98. A single unfavorable interaction with VAL543 was also observed. RU486 (Fig. 2b) exhibited prominent π-alkyl and alkyl interactions with VAL543, ARG614, and LEU620, consistent with its hydrophobic steroidal scaffold. It additionally formed a carbon hydrogen bond with VAL543, and one unfavorable bump was noted with HOH80. No conventional hydrogen bonding was detected. Salvianolic acid B (Fig. 2c) displayed a dense hydrogen-bonding network, forming conventional hydrogen bonds with GLU537, GLU540, ARG611, ARG614, ASN619, and TYR660. It also engaged in a π-cation interaction with ARG614 and a π-π T-shaped contact with VAL543. Two unfavorable contacts were identified, including a donor-donor clash with TRP610 and a steric bump near HOH80, implying spatial constraints at the binding interface. Lithospermic acid (Fig. 2d) established five conventional hydrogen bonds with PRO541, TYR545, TRP610, ARG611, and ARG614, along with π-donor hydrogen bonds to SER616 and TYR660. Additional hydrophobic interactions included a π-sigma contact with VAL543 and a π-alkyl interaction with PRO625. A single unfavorable contact with HOH80 suggested mild steric interference near solvent exposure. Sakuranetin (Fig. 2e) formed conventional hydrogen bonds with GLU540, PRO541, and GLN615, and a carbon hydrogen bond with ARG611. It also interacted with VAL543 and PRO625 through π-alkyl contacts, and with LEU620 and ILE628 via alkyl interactions, highlighting a moderate degree of hydrophobic engagement. Magnolol (Fig. 2f) predominantly relied on π-alkyl and alkyl interactions, contacting VAL543, ALA607, ARG611, ARG614, LEU620, PRO625, ILE628, and TYR663. A single π-donor hydrogen bond was detected with TRP610, supporting anchorage at the hydrophobic core. One unfavorable bump with HOH56 was also present, indicating potential solvent-facing steric hindrance. In contrast, metformin (negative control) is not shown as it could not be selected or analyzed as a ligand in the Discovery Studio Visualizer software environment. Receptor-ligand interaction profiles derived from AutoDock Vina docking poses. Two-dimensional interaction diagrams of GR-ligand complexes were generated using Discovery Studio Visualizer based on AutoDock Vina-predicted conformations. Panels display desoximetasone (a), RU486 (b), salvianolic acid B (c), lithospermic acid (d), sakuranetin (e), and magnolol (f). Annotated interactions include conventional hydrogen bonds (green), carbon hydrogen bonds (light green-gray), and water-mediated hydrogen bonds (blue). In addition, a variety of π-interactions and hydrophobic contacts are observed and visualized in various shades of pink. These include π-alkyl and alkyl interactions (light pink), π-π T-shaped stacking (deeper pink), π-sigma and π-donor hydrogen bonds (pale pink-green), and distinct π-cation interactions (orange). Unfavorable interactions are depicted in solid red, representing steric clashes or donor-donor repulsion between ligand and receptor atoms. Metformin (negative control) is not shown as it could not be selected or parsed as a ligand within the Discovery Studio Visualizer environment. Structural Stability and Ligand Retention during GR-Ligand Complex Simulations To evaluate the conformational dynamics of GR-ligand complexes, 100 ns MD simulations were analyzed using four key parameters: ligand RMSD, ligand displacement, protein Rg, and residue-level RMSF. All Ligands Maintained a Stable Pocket Retention Based on RMSD Ligand RMSD analysis revealed that all compounds remained below 0.25 nm during the 100 ns trajectories, indicating stable pocket retention (Fig. 3a). Sakuranetin exhibited the lowest and most stable RMSD (∼0.08–0.12 nm), followed closely by desoximetasone (∼0.10–0.12 nm), reflecting highly consistent binding poses. RU486 also demonstrated stable retention with slightly higher RMSD values (∼0.12–0.16 nm). Lithospermic acid showed moderate stability (∼0.15–0.18 nm), maintaining confined fluctuations. In contrast, salvianolic acid B, magnolol, and metformin exhibited higher RMSD values (∼0.18–0.22 nm), suggesting greater internal flexibility within the pocket, although none exceeded the conventional stability threshold (≤0.3 nm). Structural stability and ligand retention in GR-ligand complexes during 100 ns MD simulation. a Ligand RMSD (aligned to initial pose). b Displacement from initial ligand position (center of mass). c Rg of GR-ligand complex. d Residue-level RMSF of GR ligand-binding domain (residues 523–777, Chain A). Each colored line corresponds to a specific compound as indicated in the legend. Data illustrate dynamic stability, ligand mobility, and protein flexibility across all simulated systems. RMSD, root mean square deviation; Rg, radius of gyration; RMSF, root mean square fluctuation. Metformin Showed the Largest and Most Fluctuating Displacement within the Binding Pocket To complement RMSD assessment, we analyzed the framewise displacement of each ligand relative to its initial center-of-mass position during the 100 ns simulations (Fig. 3b). Among all ligands, RU486 showed the most stable retention, with displacement values remaining below ∼0.2 nm throughout the trajectory, reflecting its strong anchoring within the GR pocket. Lithospermic acid, salvianolic acid B, and magnolol exhibited similarly stable confinement, with displacements consistently restricted to ∼0.2–0.3 nm. Sakuranetin displayed slightly larger fluctuations, approaching ∼0.4 nm at later stages but still confined within the binding cavity. In contrast, desoximetasone remained stable initially but began to deviate markedly after ∼60 ns, reaching displacements up to ∼0.75 nm, suggestive of late-stage positional rearrangement. As expected for a negative control, metformin exhibited the highest instability, with displacement progressively increasing beyond 1.0 nm after ∼70 ns, indicating poor retention and potential escape from the canonical binding site. All GR-Ligand Complexes Preserved Global Structural Compactness To evaluate the global conformational compactness of the GR-ligand complexes, we calculated the Rg of the receptor backbone atoms (residues A:523–777) throughout the 100 ns MD simulation (Fig. 3c). Rg reflects the mass-weighted root mean square distance of atoms from their common center of mass, with lower values indicating tighter packing and greater overall structural compactness. All GR complexes exhibited stable Rg profiles within a narrow range of ∼5.05–5.07 nm over the course of the simulation. Among them, RU486-bound GR displayed the most compact and least fluctuating Rg values (∼5.050–5.055 nm), while the desoximetasone complex showed slightly higher Rg levels, fluctuating around 5.065–5.070 nm. Other phytochemical-bound systems – salvianolic acid B, lithospermic acid, sakuranetin, and magnolol – remained in a stable intermediate range (5.055–5.065 nm), without pronounced deviation. Metformin, despite its weak binding affinity, did not induce notable expansion or collapse of the receptor structure, maintaining an Rg trajectory comparable to that of sakuranetin and magnolol. These results suggest that none of the ligands caused significant destabilization or loosening of the GR ligand-binding domain in terms of global compactness during the 100 ns simulation window. Ligand Association Did Not Disrupt Local Flexibility of GR-Binding Domain To evaluate local fluctuations within the GR ligand-binding domain during ligand association, we calculated the RMSF values of each residue from chain A523 to 777 across the 100 ns MD simulation (Fig. 3d). RMSF reflects the positional deviation of each Cα atom relative to its average trajectory, with lower values indicating more rigid and stable regions. Across all ligand-bound GR systems, the majority of residues exhibited low RMSF values (∼0.1–0.25 nm), suggesting a relatively stable backbone architecture. Slightly elevated fluctuations were observed at the N-terminus (residues 523–530) and C-terminus (near residue 777), consistent with terminal flexibility commonly seen in MD simulations. Among the tested ligands, RU486 and desoximetasone were associated with slightly lower RMSF values in central residues (∼550–750), while salvianolic acid B and sakuranetin showed marginally higher fluctuations in select loop regions. Metformin, used as a negative control, produced a noticeable increase in RMSF at the N-terminal region, but its influence on the core binding pocket remained limited. Overall, the RMSF profiles suggest that none of the ligands induced major destabilization of the GR-binding domain, although minor differences in local flexibility may reflect the extent of ligand-receptor engagement and conformational stabilization. Collectively, these dynamic metrics highlight distinct retention profiles among the GR-ligand complexes. Sakuranetin exhibited the lowest RMSD values, indicating a highly consistent binding pose, although its displacement reached ∼0.4 nm, suggesting limited translational drift within the pocket. RU486 showed the most stable displacement (<0.2 nm) alongside low RMSD, confirming its strong anchoring as a reference binder. Lithospermic acid also demonstrated robust confinement with low-to-moderate RMSD and displacement, supporting stable retention. In contrast, desoximetasone maintained low RMSD yet displayed marked late-stage displacement (>0.6 nm after ∼60 ns), consistent with positional rearrangement despite pocket retention. Salvianolic acid B and magnolol exhibited moderately elevated RMSD (∼0.18–0.22 nm) but remained confined in displacement (<0.35 nm), indicating pocket stability with greater internal flexibility. As expected, metformin showed the highest displacement (>1.0 nm) and instability, reflecting weak retention as a negative control. Overall, these findings suggest that RU486, sakuranetin, and lithospermic acid represent the most stably anchored ligands, whereas desoximetasone undergoes late rearrangements and phytochemicals such as salvianolic acid B and magnolol maintain moderate but confined binding dynamics. Post-MD PLIP Interaction Analysis Reveals Conserved and Dynamic Binding Features To evaluate the stability and persistence of receptor-ligand interactions beyond initial docking poses, we conducted PLIP analyses before and after the 100 ns MD simulations. Table 2 summarizes the dynamic evolution of key contacts, including hydrophobic interactions, hydrogen bonds, and π-stacking, for each ligand. Table 2.Dynamic changes in protein-ligand interactions before and after 100 ns MD simulationLigandTimepointResultshydrophobic interactionhydrogen bondπ-cation interactionπ-stackingDEX (co-crystallized ligand in PDB: 1M2Z)–PHE623ASN564––GLN642GLN570ARG611TYR735THR739DesoximetasonePre-MDMET604GLN570AGR611–ALA607TRP610LEU608PHE623ARG611Post-MDLEU753GLN642–​CYS736PHE740RU486Pre-MDLEU563GLN570––MET604ALA605LEU608ARG611PHE623MET646LEU732Post-MDLEU563GLN642––GLN570LEU608PHE623TYR735Salvianolic acid BPre-MDASP641LEU636–PHE737GLN642ASP641GLN642LEU732ASN731TYR735LEU733PHE737GLN738TYR735THR739Post-MDMET560LEU563––LEU563ASN564ALA605GLN570LEU608GLN642LEU732VAL729LEU733LEU732Lithospermic acidPre-MDGLN570GLY567––MET601GLN570LEU603TYR598MET604TRP600ALA605ALA605ALA607PHE606LEU608LEU732ARG611Post-MDLEU563ASN564––LEU566GLN570TRP600LEU603SER602MET604LEU603PHE606PHE623PHE623SakuranetinPre-MDLEU566ARG611––GLN570LEU608PHE623Post-MDMET604GLN570––LEU608MET604PHE623GLN642MagnololPre-MDTRP600GLN570–PHE623MET604LEU608LEU621PHE623MET646Post-MDLEU563LEU563––ASN564LEU608GLN642TYR735THR739MetforminPre-MDNo interactionPost-MD–LEU544–​ALA607Protein-ligand interactions for each GR-ligand complex were analyzed using the PLIP at two timepoints: the initial docking pose (pre-MD) and after 100 ns of molecular dynamics simulation (post-MD). Interaction types include hydrophobic contacts, conventional hydrogen bonds, π-cation interactions, and π-stacking. Bolded entries represent interactions preserved throughout the simulation, while newly emerged or lost contacts reflect ligand repositioning, receptor flexibility, or altered engagement within the binding pocket during the MD trajectory. This comparative profiling highlights both stable and dynamic features of ligand binding across compounds. Dynamic changes in protein-ligand interactions before and after 100 ns MD simulation Protein-ligand interactions for each GR-ligand complex were analyzed using the PLIP at two timepoints: the initial docking pose (pre-MD) and after 100 ns of molecular dynamics simulation (post-MD). Interaction types include hydrophobic contacts, conventional hydrogen bonds, π-cation interactions, and π-stacking. Bolded entries represent interactions preserved throughout the simulation, while newly emerged or lost contacts reflect ligand repositioning, receptor flexibility, or altered engagement within the binding pocket during the MD trajectory. This comparative profiling highlights both stable and dynamic features of ligand binding across compounds. Desoximetasone initially engaged multiple hydrophobic contacts (MET604, ALA607, LEU608, PHE623) and hydrogen bonds with GLN570, TRP610, and ARG611, as well as π-cation interaction with ARG611. After MD, its interaction network shifted to new contacts with LEU753, GLN642, CYS736, and PHE740. Importantly, the canonical ARG611 was lost, but retention of GLN642 alongside new hydrophobic anchoring suggests a reoriented yet stable binding conformation. RU486 demonstrated the most extensive and consistent interaction profile. Pre-MD, it engaged GLN570 and ARG611 in addition to a wide hydrophobic network (LEU563, MET604, ALA605, LEU608, PHE623, MET646, LEU732). Post-MD, RU486 retained GLN570 and PHE623 while incorporating GLN642 and TYR735, both canonical DEX residues. This shift indicates enhanced mimicry of the co-crystallized agonist and underscores RU486’s structural rigidity and adaptability within the GR pocket. Salvianolic acid B exhibited a broad interaction network prior to MD, including hydrophobic and hydrogen-bond contacts with ASP641, GLN642, LEU732, LEU733, TYR735, LEU636, ASN731, PHE737, GLN738, and THR739, as well as π-stacking with PHE737. This extensive interaction map suggested multiple anchoring modes within the GR-binding pocket. After 100 ns MD, Sal B’s interaction profile underwent a major reorganization. New hydrophobic contacts were established with MET560, LEU563, ALA605, LEU608, LEU732, and LEU733, while hydrogen bonds stabilized with ASN564, GLN570, and GLN642. Notably, three of these residues (ASN564, GLN570, and GLN642) overlap directly with the canonical DEX binding map, which also includes TYR735 and THR739. Compared to its pre-MD state, where Sal B transiently contacted TYR735 and THR739, the post-MD profile demonstrates consolidation into a more focused hydrogen-bonding network. Prior to MD, lithospermic acid exhibited multiple contacts, including hydrophobic and hydrogen-bond interactions with GLN570, MET601, LEU603, MET604, ALA605, PHE606, LEU732, GLY567, TYR598, ALA607, LEU608, and ARG611. This broad interaction profile incorporated two canonical DEX residues (GLN570, ARG611). After 100 ns MD, its binding network reorganized to include LEU563, LEU566, LEU603, MET604, TRP600, SER602, LEU603, PHE606, and PHE623, with hydrogen bonds involving ASN564 and GLN570. Three of these residues (ASN564, GLN570, PHE623) overlap directly with the canonical DEX binding map, indicating a stabilized anchoring mode enriched in polar and aromatic contacts following simulation. Before MD, sakuranetin formed interactions with LEU566, GLN570, LEU608, PHE623, and ARG611, engaging three canonical DEX residues (GLN570, ARG611, and PHE623). After 100 ns MD, its interaction network was reorganized to include MET604, LEU608, PHE623, GLN570, and GLN642. This profile retained two canonical residues (GLN570, GLN642, PHE623) but lost ARG611, indicating a shift from its pre-MD binding configuration toward an alternative anchoring arrangement. Before MD, magnolol established interactions with TRP600, MET604, LEU608, LEU621, PHE623, and MET646, including hydrogen bonding with GLN570 and π-stacking with PHE623. After 100 ns MD, its interaction network shifted to LEU563, ASN564, LEU608, GLN642, TYR735, and THR739, with the loss of PHE623 π-stacking. Three of these residues (ASN564, GLN642, TYR735) overlap with canonical DEX contacts. Metformin, consistent with its weak binding affinity, formed only transient contacts. Post-MD, it established hydrogen bonds with LEU544 and ALA607 but lacked overlap with key DEX residues, supporting its instability and validating its role as a negative control. Benchmarking against the co-crystallized DEX structure – which engages ASN564, GLN570, ARG611, GLN642, PHE623, TYR735, and THR739 – revealed distinct patterns of canonical residue conservation among the test ligands. RU486 preserved four residues (GLN570, PHE623, GLN642, TYR735), representing the most extensive overlap. Salvianolic acid B retained three canonical contacts (ASN564, GLN570, GLN642), while lithospermic acid also maintained three (ASN564, GLN570, PHE623). Sakuranetin preserved three residues as well (GLN570, GLN642, PHE623), whereas magnolol engaged a different triad (ASN564, GLN642, TYR735). By contrast, desoximetasone retained only a single canonical residue (GLN642), indicating a marked reduction in pharmacophoric overlap. Metformin, consistent with its unstable binding, exhibited no meaningful conservation of DEX-associated interactions. Collectively, these results indicate that RU486 exhibits the strongest post-MD mimicry of the DEX interaction profile, preserving four canonical contacts. Salvianolic acid B, lithospermic acid, sakuranetin, and magnolol each retained three canonical residues, representing a consistent and moderate degree of overlap with the DEX binding pattern. By contrast, desoximetasone preserved only a single canonical contact, and metformin failed to retain any, underscoring their minimal to negligible conservation within the GR-binding pocket. MD-Informed GR-Score for Receptor-Engagement Prioritization To integrate structural and dynamic determinants of GR engagement, a composite GR-score was derived from three quantitative components – Structure-score, Dock-score, and MD-score – each reflecting distinct mechanistic contributions (Table 3). Among reference ligands, the antagonist RU486 achieved the highest GR-score (1.46), consistent with its strong receptor affinity and stable retention, whereas desoximetasone, the clinical glucocorticoid control, scored 1.01, representing canonical steroidal engagement. Importantly, the high GR-score observed for RU486 despite its established antagonist activity suggests that the current framework primarily captures receptor-binding, pharmacophore-overlap, and ligand-retention characteristics rather than definitive agonistic transcriptional activity. Table 3.Composite GR-score integrating structure-, docking-, and dynamics-based parameters for receptor-engagement prioritization​Structure-scoreDock-scoreMD-scoreGR-scoreDesoximetasone0.80.616−0.401.01RU4860.90.4980.061.46Salvianolic acid B0.620.818−0.101.34Lithospermic acid0.520.8740.211.60Sakuranetin0.620.666−0.111.17Magnolol0.20.414−0.090.53Metformin00.3−0.71−0.41 Composite GR-score integrating structure-, docking-, and dynamics-based parameters for receptor-engagement prioritization Among the phytochemicals tested, lithospermic acid exhibited the greatest composite score (1.60), surpassing both desoximetasone and RU486, indicating favorable structural mimicry and post-dynamics stability. Salvianolic acid B followed closely (1.34), reflecting its extensive hydrogen-bonding network and three-point canonical residue overlap with the DEX binding motif. Sakuranetin reached a moderate score (1.17), consistent with its compact but partially flexible pose, while magnolol (0.53) and metformin (−0.41) ranked lowest, confirming weak steroid-like characteristics and poor dynamic retention. Supplementary module-ablation analyses demonstrated that docking-associated features contributed most strongly to the composite GR-score framework (online suppl. Table S1). IHC Validation of the MD-Informed GR-Score IHC staining was performed to assess total GR expression and its phosphorylation at Ser211, the latter serving as a receptor-associated marker linked to transcription-associated GR signaling. As shown in Figure 4a total GR staining was predominantly localized in the nuclei of keratinocytes across all groups, with relatively uniform distribution throughout the basal and suprabasal layers. No apparent reduction in staining intensity or alteration in epidermal morphology was observed among the treatment groups. In contrast, Ser211 staining (Fig. 4b) displayed distinct differences in nuclear intensity across treatment groups. Desoximetasone, the positive control, induced a moderate nuclear staining pattern in both basal and suprabasal keratinocytes. Lithospermic acid treatment resulted in the strongest Ser211 immunoreactivity, with intense nuclear staining extending across the full epidermal thickness. Salvianolic acid B and sakuranetin produced staining comparable to desoximetasone, though slightly less intense, whereas magnolol treatment yielded weaker nuclear staining, particularly in the suprabasal layers. Immunohistochemical assessment of GR expression and pSer211-associated receptor signaling, and correlation with the MD-informed GR-score. a Immunohistochemical staining of total GR in epidermal tissues treated with desoximetasone (positive control) or the indicated natural compounds. Nuclear GR localization was observed throughout the basal and suprabasal keratinocyte layers across all groups. b Immunohistochemical staining of phosphorylated GR at Ser211, serving as a proximal indicator associated with GR activation. Distinct differences in nuclear staining intensity were evident among treatment groups, with lithospermic acid producing the strongest signal. c Semi-quantitative analysis of total GR immunoreactivity. No significant differences were observed among the four natural compounds relative to desoximetasone. d Semi-quantitative analysis of GR-Ser211 immunoreactivity. Lithospermic acid significantly increased Ser211 staining intensity compared with desoximetasone and all other treatments. e Relative pSer211/GR ratio (GR-Ser211/GR), indexing the proportion of receptor in the phosphorylated state. Lithospermic acid exhibited the highest pSer211/GR ratio, whereas salvianolic acid B and sakuranetin were comparable to desoximetasone, and magnolol trended lower without significance. Scale bar = 20 μm. Results are expressed as mean ± SD (n = 6). Statistical significance: **p < 0.01. Subsequent semi-quantitative analysis confirmed these observations. As shown in Figure 4c, mean GR intensity values were comparable among the four natural compounds, with no significant differences relative to desoximetasone, indicating that all compounds-maintained GR expression at levels similar to the positive control. In contrast, phosphorylation at Ser211 showed distinct differential signaling patterns (Fig. 4d). Lithospermic acid significantly increased Ser211 intensity compared with desoximetasone and all other treatments (p < 0.01), demonstrating increased pSer211-associated receptor signaling intensity. By comparison, salvianolic acid B and sakuranetin yielded values close to desoximetasone without significant differences, while magnolol exhibited a slightly lower trend that did not reach statistical significance. A receptor-associated pSer211/GR ratio was calculated by dividing the GR-Ser211 signal by total GR (GR-Ser211/GR), providing a relative index of Ser211-phosphorylated receptor abundance. Using this ratio (Fig. 4e), lithospermic acid exhibited the highest pSer211/GR ratio – approximately two-fold higher than desoximetasone and significantly greater than all other treatments (p < 0.01). Salvianolic acid B and sakuranetin were comparable to desoximetasone (slightly lower on average, not significant), whereas magnolol trended lower – at approximately 75% of the desoximetasone level – without statistical significance. Collectively, the four natural compounds produced overall IHC profiles broadly comparable to desoximetasone. Lithospermic acid consistently showed the strongest pSer211-associated receptor signal among the tested compounds – significantly higher GR-Ser211 intensity and an approximately two-fold increase in the GR-Ser211/GR ratio relative to desoximetasone (p < 0.01). Salvianolic acid B and sakuranetin were largely comparable to desoximetasone across readouts (minor decreases on average, not significant). Magnolol trended lower – about 75% of the desoximetasone level in the GR-Ser211/GR ratio – yet without statistical significance. Exploratory Concordance between GR-Score and pSer211-Associated Readouts The exploratory concordance of the MD-informed GR-score with the GR-Ser211/GR ratio, a receptor-associated immunohistochemical readout, was examined by linear regression analysis. A positive association was observed (slope = 1.79, intercept = 0.20, R2 = 0.61, p = 0.12; Fig. 5a), although the 95% confidence interval (−0.82 to 4.40) indicated that the relationship did not reach statistical significance. Residual diagnostics demonstrated errors centered near zero without obvious heteroscedasticity or systematic curvature (Fig. 5b). A rank-based Spearman correlation (ρ = 0.60, p = 0.35) showed directionally consistent results, although statistical significance was likewise not achieved. Exploratory concordance analysis between GR-score and pSer211-associated readouts. a Linear regression of the GR-Ser211/GR ratio against the MD-informed GR-score (slope = 1.79, intercept = 0.20, R2 = 0.61, p = 0.12). Dashed lines represent the 95% confidence interval. b Residual plot of the regression model, showing errors centered around zero without evidence of heteroscedasticity. Data points correspond to individual compounds (desoximetasone, salvianolic acid B, lithospermic acid, sakuranetin, magnolol). To further examine the relative contribution of individual scoring modules, ablation analyses were performed (Table 3). Docking-associated features contributed most strongly to the composite scoring framework, whereas structure-only or MD-only modules demonstrated comparatively limited discriminatory capacity. Dual-module combinations retaining the docking component (S+D or D+M) showed improved concordance with receptor-associated readouts compared with S+M, although none exceeded the full S+D+M framework. Collectively, these findings suggest that the GR-score integrates complementary structural, docking-derived, and MD-associated information relevant to receptor-engagement profiling. Nevertheless, given the limited compound number and the absence of statistically significant correlations, the current framework should be interpreted as an exploratory proof-of-concept prioritization approach rather than a validated predictor of functional GR activation or agonism. Discussion The present study integrates docking, MD, and interaction profiling to systematically evaluate the GR engagement of four phytochemicals – sakuranetin, salvianolic acid B, lithospermic acid, and magnolol – previously shown to alleviate psoriasiform inflammation in vivo. Docking simulations identified salvianolic acid B and lithospermic acid as the strongest binders (−8.3 and −7.9 kcal/mol), while sakuranetin and magnolol displayed moderate affinities (−7.1 and −6.9 kcal/mol), and metformin confirmed its role as a negative control with weak binding (−5.0 kcal/mol). MD analyses further demonstrated stable pocket retention across all phytochemicals, with sakuranetin showing the lowest RMSD, lithospermic acid maintaining robust confinement, and salvianolic acid B and magnolol exhibiting modest flexibility. By contrast, desoximetasone, despite established clinical glucocorticoid efficacy, underwent late-stage positional drift and exhibited comparatively less canonical interaction retention, suggesting that clinically relevant glucocorticoid activity may not be fully captured by canonical-contact overlap or docking-stability metrics alone. Post-MD interaction profiling reinforced these observations, showing that salvianolic acid B, lithospermic acid, sakuranetin, and magnolol each retained three canonical DEX contacts, whereas desoximetasone preserved only one, highlighting a limitation of the current computational framework in fully capturing clinically relevant glucocorticoid activity. Experimental observations partially aligned with the in silico analyses. Total GR expression remained comparable among treatments, whereas pSer211 staining demonstrated differential receptor-associated signaling patterns across compounds. Lithospermic acid showed the highest pSer211/GR ratio, while salvianolic acid B and sakuranetin displayed profiles broadly comparable to desoximetasone. Regression analysis identified a positive but non-significant association (R2 = 0.61) between the MD-informed GR-score and the pSer211/GR ratio, suggesting preliminary concordance between computational interaction profiles and tissue-level receptor-associated readouts. Collectively, these findings support the utility of multi-parameter computational prioritization approaches for identifying candidate phytochemicals relevant to future GR-focused dermatologic investigations. Mechanistic Insights from Computational Modeling Across 100-ns trajectories, all phytochemical-GR complexes maintained pocket residency, although ligand-specific dynamic behaviors suggested differences in receptor-engagement patterns rather than definitive activation states. Ligand RMSD ranked sakuranetin as the most pose-rigid (∼0.08–0.12 nm), RU486 and desoximetasone as low-variance, and salvianolic acid B/magnolol as modestly flexible (≈0.18–0.22 nm). Center-of-mass displacement separated anchored binders (RU486 <0.2 nm; lithospermic acid confined at ∼0.2–0.3 nm) from rearrangers (desoximetasone drifting after ∼60 ns up to ∼0.75 nm) and the non-retainer metformin (>1.0 nm late escape). Protein-level metrics indicated no global destabilization (R_g ∼5.05–5.07 nm throughout) and only modest, loop-localized RMSF differences, suggesting the GR-LBD scaffold tolerates these chemotypes without major structural disruption. Post-MD interaction fingerprints rationalize the dynamics: RU486 gains/retains multiple canonical DEX contacts (GLN570, PHE623, GLN642, TYR735), whereas each phytochemical consistently preserves three canonical residues – e.g., lithospermic acid (ASN564, GLN570, PHE623) and salvianolic acid B (ASN564, GLN570, GLN642) – supporting stable pocket engagement despite limited internal mobility. In contrast, desoximetasone exhibited late-stage rearrangement associated with loss of ARG611 and redistribution toward alternative hydrophobic interaction nodes, consistent with a re-anchored but non-DEX-like binding orientation. Collectively, these simulations support the interpretation that different ligands stabilize distinct GR-LBD interaction patterns and conformational behaviors, rather than uniformly reproducing classical DEX-like signaling states. These observations align with previous structural and MD studies demonstrating ligand-dependent conformational plasticity of the GR-LBD and differential regulation of AF-2/Helix 12 accessibility [20–25]. Importantly, these MD-based observations complement our prior in vivo findings with lithospermic acid [16], magnolol [15], salvianolic acid B [14], and sakuranetin [17], each of which improved psoriasiform skin inflammation and barrier-associated outcomes in imiquimod-induced models. The present docking and MD analyses, therefore, provide an exploratory receptor-level framework potentially relevant to the previously observed anti-inflammatory effects of these phytochemicals. Nevertheless, the current computational analyses do not establish functional GR agonism, downstream transcriptional activation, or steroid-sparing efficacy, and further biological validation will be necessary to clarify the functional significance of these receptor-engagement patterns. Functional Implications of GR Engagement GR engagement inferred from structural, docking, and dynamics analyses should be interpreted cautiously when relating receptor-associated interactions to biologically meaningful tissue responses. In the present study, functional interpretation was primarily based on proximal receptor-associated readouts rather than direct assessment of downstream glucocorticoid-responsive transcriptional activity. Functionally, GR activation can be considered along a proximal-to-distal axis: (i) phosphorylation of GR at serine-211 (Ser211), which is commonly used as a proximal indicator associated with transcription-related receptor signaling; and (ii) downstream transcriptional effectors, such as FK506 binding protein 5 (FKBP5) and glucocorticoid-induced leucine zipper (GILZ), which are frequently implicated in anti-inflammatory and homeostatic glucocorticoid responses relevant to psoriatic skin. Because FKBP5 and GILZ were not experimentally evaluated in the present study, the following discussion is intended to provide biological context rather than direct functional validation of these pathways. Ser211 Phosphorylation as a Marker of Active GR Phosphorylation of the GR at Ser211 is widely regarded as a proximal marker associated with transcription-related GR signaling. Mechanistic studies have demonstrated that Ser211 phosphorylation contributes to GR-mediated transcriptional regulation and is influenced by upstream signaling pathways involved in nuclear receptor activation [26–28]. In addition, Ser211-based assays have shown concordance with endogenous GR target-gene responses in selected cellular systems [29]. However, Ser211 phosphorylation is also influenced by inflammatory context, ligand concentration, and cellular state. Altered Ser211/Ser226 balance has been associated with glucocorticoid insensitivity in inflammatory airway disease [30], while IL-8-driven inflammatory signaling can perturb Ser211 phosphorylation and contribute to steroid resistance [31]. Dose-dependent effects on Ser211 signaling have likewise been observed in fibroblast and macrophage systems, where excessive glucocorticoid exposure promoted profibrotic pathways despite increased nuclear Ser211 [32, 33]. In addition, tissue-specific variability has been reported under certain physiological conditions [34]. Collectively, these findings support pSer211 as a practical proximal receptor-associated marker suitable for tissue-level and screening-oriented analyses. Nevertheless, nuclear GR localization and Ser211 phosphorylation alone do not establish functional GR agonism or downstream transcriptional outcomes. Accordingly, interpretation of pSer211 staining should be integrated with complementary functional assessments, including downstream target genes, cofactor recruitment, and phenotypic endpoints. Functional Implications of GR Engagement: FKBP5 and GILZ Beyond structural dynamics, the biological significance of GR engagement is ultimately linked to downstream glucocorticoid-responsive gene programs, among which FKBP5 and GILZ are commonly used markers of GR-associated signaling. FKBP5 encodes a co-chaperone involved in GR feedback regulation and has been widely utilized as a pharmacodynamic biomarker of glucocorticoid exposure across tissues [35]. In keratinocytes and skin, FKBP5 has also been implicated in modulation of GR phosphorylation, nuclear translocation, and steroid-associated responses [36, 37]. GILZ (glucocorticoid-induced leucine zipper) has been associated with suppression of inflammatory pathways relevant to Th1/Th17-driven psoriasis pathogenesis [38], although context-dependent and paradoxical effects have also been reported in imiquimod-induced psoriasis-like models [39]. Pharmacological studies further suggest that ligand-specific modulation of GR conformation may influence FKBP5- and GILZ-associated transcriptional responses [40]. Nevertheless, FKBP5 and GILZ were not experimentally evaluated in the present study, and the current discussion is intended to provide biological context rather than direct functional validation of these downstream pathways. Accordingly, while the observed docking and MD profiles may support potential relevance to receptor-associated signaling, further studies will be required to determine whether these phytochemicals directly regulate FKBP5-, GILZ-, or other GR-dependent transcriptional programs in vivo. Translational and Integrative Perspectives: GR Modulation as a Path to Safer Therapies in Psoriasis Chronic dermatologic disorders such as psoriasis often require prolonged anti-inflammatory treatment, yet long-term corticosteroid use remains limited by adverse cutaneous and systemic effects. These challenges have stimulated interest in selective GR modulators, which aim to retain anti-inflammatory efficacy while reducing undesirable glucocorticoid-associated responses [11]. Several plant-derived compounds have demonstrated GR-associated signaling activity in experimental systems [41–45], supporting the broader feasibility of receptor-focused strategies in inflammatory skin disease. At the same time, cutaneous glucocorticoid biology is influenced by interactions between GR and the mineralocorticoid receptor (MR), both of which contribute to epidermal homeostasis, inflammation, and barrier regulation [46, 47]. In parallel, natural-product chemical space continues to provide potential candidates for steroid-sparing therapeutic development, including phytosterols, flavonoids, terpenoids, and alkaloids with reported anti-psoriatic or GR-associated activities [48–51]. Vitamin D analogs remain an established non-steroidal topical therapy for psoriasis and exemplify the broader feasibility of receptor-focused therapeutic strategies in chronic inflammatory skin disease [52]. In this context, phytochemical-derived compounds with GR-associated signaling activity may provide an additional framework for receptor-focused anti-inflammatory modulation [1, 53]. Evidence that β-sitosterol attenuates psoriasiform inflammation with favorable skin tolerance compared with corticosteroids further supports the therapeutic potential of plant-derived steroid-sparing approaches [48]. Within this broader context, the present multi-parameter framework integrates docking-derived interaction patterns, MD stability, and tissue-level receptor-associated readouts to prioritize candidate phytochemicals for further investigation. Importantly, the current workflow should be regarded as an exploratory receptor-engagement prioritization approach rather than direct evidence of functional GR agonism or clinical steroid-sparing efficacy. Nevertheless, these findings support the potential utility of computationally guided screening strategies for identifying candidate compounds relevant to future psoriasis-focused GR modulation studies. Limitations and Future Directions This study should be regarded as an exploratory proof-of-concept framework rather than a generalizable predictive platform. Although the analysis was limited to four phytochemicals with prior anti-psoriatic evidence, the principal limitation lies in the biological interpretation of the current receptor-associated readouts. Experimental validation was confined to immunohistochemical assessment of total GR and Ser211 phosphorylation, without direct evaluation of downstream transcriptional activity, cofactor recruitment, GR antagonism, transactivation/transrepression balance, dose-response or time-course relationships, or steroid-associated atrophy/barrier-related functional outcomes. Accordingly, the observed association between the GR-score and pSer211 ratio should be interpreted cautiously as a preliminary receptor-engagement-associated observation rather than definitive evidence of functional GR agonism. Future studies incorporating larger and chemically diverse compound libraries, downstream GR target genes (e.g., FKBP5 and GILZ), transcriptomic analyses, and functional cellular or animal validation will be important for clarifying the biological relevance and translational utility of this framework. Nevertheless, the present work demonstrates that integration of structural analysis, docking-derived interaction patterns, MD stability, and tissue-level receptor-associated readouts may provide a practical strategy for prioritizing candidate compounds before resource-intensive experimental evaluation. Conclusion This study presents an exploratory multi-parameter framework integrating structural analysis, docking-derived interaction patterns, MD stability, and tissue-level receptor-associated readouts to prioritize phytochemicals with potential relevance to GR-associated signaling. Among the evaluated compounds, lithospermic acid, salvianolic acid B, magnolol, and sakuranetin demonstrated stable GR-LBD interaction profiles during 100-ns MD simulations. The observed association between computational profiles and pSer211-related histological findings provides preliminary receptor-engagement-associated support for further investigation, but should not be interpreted as definitive evidence of functional GR agonism or steroid-sparing efficacy. Overall, the present work highlights the potential utility of computationally guided screening strategies for prioritizing candidate compounds for future psoriasis-focused GR modulation studies. Acknowledgment The authors thank their respective institutions and colleagues for general support during the study. Statement of Ethics The dorsal skin tissue blocks used in this study were derived from previously approved animal studies conducted at Cathay General Hospital, Taipei, Taiwan, with no new animal experiments performed in the present work. The study protocols were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Cathay General Hospital, Taiwan (Approval Nos.: CGH-IACUC-107-028, 15 May 2018; CGH-IACUC-108-006, 19 December 2018; IACUC-110-025, 5 July 2021; CGH-IACUC-112-024, 27 June 2023). Conflict of Interest Statement The authors have no conflicts of interest to declare. Funding Sources This work was supported by the National Science and Technology Council (NSTC), Taiwan (Grant No. NSTC-113-2314-B-281-002) and by the Cathay General Hospital, Taipei, Taiwan (Grant Nos. CGH-MR-A113009, CGH-MR-A114017, and CGH-MR-A114018). The funders had no role in the study design, data collection, data analysis, interpretation, manuscript preparation, or the decision to publish. Author Contributions Feng-Chuan Tai: resources, investigation, project administration, supervision, and writing – review and editing. Heng-Hui Lien: conceptualization, methodology, resources, supervision, and writing – review and editing. Shih-Chang Lin: investigation, funding acquisition, and writing – review and editing. Yu-Li Hung: methodology and funding acquisition. Cheng Yu-Ping: investigation and writing – review and editing. Guo Jiun-Wen: writing – original draft, review, and editing; visualization; validation; supervision; project administration; methodology; investigation; funding acquisition; formal analysis; and data curation.