3D-QSAR Modeling of the Anti-Inflammatory Activity of 28 Rare Biflavones

Abstract

Inflammation underlies a wide range of chronic diseases. Current treatments, such as nonsteroidal anti-inflammatory drugs, are effective, but they often result in adverse side effects. This has driven the search for alternative anti-inflammatory agents with improved safety profiles. Biflavones have emerged as promising candidates, yet existing studies primarily focus on naturally occurring biflavones that are limited to those with C3′–C8″ linkages. In this study, we explore the anti-inflammatory potential of 28 biflavones featuring the underexplored C2′–C6″ linkage. Using LPS-induced RAW 264.7 macrophage assays, biflavone 10 (synthesized from the flavones acacetin and 5,3′,4′-trihydroxyflavone) inhibited nitric oxide production (20.94 ± 7.56%). A 3D-QSAR model was developed based on selected nitric oxide (NO) inhibition data (R ² = 0.960, R ²CV = 0.831, and Pearson’s r = 0.950), revealing that hydrogen bond acceptor interactions are key determinants of bioactivity. The presence of HBAs at C-5 and C-7 of ring A1 enhanced activity, while their presence at C-4′ of ring B2 reduced it. Molecular docking studies revealed a strong binding affinity between biflavone 10 and inflammation-related protein COX-2 (−6.881 kcal/mol). This prediction was validated by a COX-2 enzyme inhibitor assay; the biflavone 10 inhibits COX-2 activity in a dose-dependent manner. Molecular dynamics simulations confirmed the stability of the biflavone 10-COX-2 complex through stable root mean square deviation (RMSD) profiles, consistent hydrogen bonding, and minimal structural fluctuation. Together, this study provides a comprehensive assessment of the anti-inflammatory potential of structurally novel biflavones, offering mechanistic insight into their activity and guiding the rational design of biflavone-based therapeutics.

1. Introduction

Inflammation is an innate defense mechanism that protects the body. Acute inflammation is often triggered by infection or injury, whereas chronic inflammation can occur in the absence of perturbation and is associated with a myriad of diseases, such as bronchitis, gastritis, and cardiovascular diseases. Commonly employed therapeutic agents include nonsteroidal anti-inflammatory drugs (NSAIDs) and corticosteroids. NSAIDs exert their effects by inhibiting cyclooxygenase (COX) enzymes, thereby reducing the synthesis of prostaglandins (PGs), which amplify cytokine signaling. , Corticosteroids, on the other hand, modulate the transcription of pro-inflammatory genes through activation of glucocorticoid receptor pathways. Despite their clinical efficacy, these treatments are often associated with significant side effects, such as increased cardiovascular risk, gastrointestinal irritation, and multiple organ pathologies.

This necessitates the search for new anti-inflammatory agents with improved efficacy and safety profiles. Natural products such as biflavones have emerged as promising candidates. Li et al. reported that ginkgetin, isolated from Ginkgo biloba leaves, exhibits strong anti-inflammatory activity. This biflavone downregulates pro-inflammatory cytokines and inhibits the toll-like receptor 4 (TLR4)/nuclear factor κB nuclear translocation (NF-κB) pathway in the rat model. In addition, ginkgetin suppresses proatherogenic and inflammatory processes in a transient receptor potential vanilloid 4 (TRPV4)-dependent manner. Amentoflavone, another well-studied biflavone found in plants such as Selaginella tamariscina, also demonstrates notable anti-inflammatory effects. At a concentration of 10 μM, amentoflavone has been shown to reduce the phosphorylation of p38 mitogen-activated protein kinase (MAPK) in RAW 264.7 macrophages. It also inhibits inducible nitric oxide (iNOS) and COX-2 expression in rats with ulcerative colitis.

These findings support the potential of biflavones as scaffolds for the development of anti-inflammatory agents. However, the studies on the anti-inflammatory effects of biflavones are mainly restricted to commonly occurring biflavones with a C3′–C8″ linkage (Figure ) between the flavone units. , Moreover, the relationship between biflavone structures and their anti-inflammatory activity remains poorly understood, primarily due to the challenges in obtaining a diverse collection of biflavones. Most studies have focused either on individual biflavones isolated from specific plant species or a small subset of synthetic biflavones, with no evaluation of the structure–activity relationship. − The reported activities are summarized in Table S1.

1.

Chemical structures of ginkgetin and amentoflavone.

To address this gap, this study expands the chemical space of anti-inflammatory biflavones by introducing a library of C2′–C6″ linked biflavones, which diverges from the conventionally studied structures. In our previous work, we synthesized a library of 28 rare biflavones (Figure ) using a food-grade oxidative coupling method. In this present study, we leverage this structurally diverse library to explore its anti-inflammatory potential.

2.

Structures of 28 rare biflavones synthesized.

A nitric oxide (NO) inhibition assay was employed to evaluate anti-inflammatory activity. Inflammatory responses in macrophages are often characterized by the overproduction of NO via iNOS upon stimulation with lipopolysaccharide (LPS). , The expression of iNOS is regulated by several signaling pathways, including NF-kB and Jak-STAT. Some cyclooxygenase inhibitors, such as aspirin and sodium salicylate, have also been reported to inhibit both the PG synthesis and iNOS expression. , Since NO production is modulated by multiple inflammation-related pathways, its inhibition serves as a quantifiable marker of anti-inflammatory activity. Additionally, the assay can be performed spectrophotometrically, making it suitable for high-throughput screening. This makes it ideal for evaluating the anti-inflammatory effects of the rare biflavones and structurally related analogues. As such, a systematic evaluation of NO inhibition was performed to develop a quantitative structure–activity relationship (QSAR) model. This model identifies key structural features that influence anti-inflammatory activity, providing comprehensive insights into the enhanced activity observed in certain biflavones and elucidating their structure–activity relationships in the context of inflammation modulation.

To further explore potential mechanisms of action, molecular docking studies were conducted for the most bioactive biflavones (biflavone 10) between the biflavone and inflammation-relevant proteins such as COX-2, hematopoietic cell kinase (HCK), iNOS, inhibitory κB-α/nuclear factor-κB (I-κB-α/NF-κB), tumor necrosis factor-alpha (TNF-α), and monocyte chemoattractant protein-1­(MCP-1). The protein–drug with the lowest binding energy was used for molecular dynamics (MD) simulations to assess the stability and behavior of the flavonoid–protein complex. The COX-2 inhibition ability was experimentally verified using a COX-2 inhibition kit.

2. Materials and Methods

2.1. Materials

The 28 biflavones used in this study were synthesized from our laboratory (Zheng et al., 2025). Biograde DMSO, LPS derived from Escherichia coli serotype 055:B5, high-glucose Dulbecco’s Modified Eagle Medium (DMEM, supplemented with l-glutamine and phenol red, without sodium pyruvate and HEPES), penicillin–streptomycin antibiotic solution (1000 IU/mL penicillin and 1000 μg/mL streptomycin), fetal bovine serum (FBS), phosphate-buffered saline, BCA Protein Assay Kit, and Cell Counting Kit-8 (CCK-8) were all procured from Dojindo Molecular Technologies, Inc. (Kumamoto, Japan). Griess reagent and GoTaq qPCR Master Mix were sourced from Promega Pte Ltd. (Madison, Wisconsin, USA). Tissue culture-treated 96-well microplates were obtained from Corning Inc. (New York, NY, USA).

2.2. Cell Culture

The murine macrophage cell line RAW 264.7 was used to evaluate the anti-inflammatory effect. The RAW 264.7 cell line (TIB-71) was propagated in high-glucose DMEM fortified with 10% FBS and 1% penicillin–streptomycin antibiotic solution. The cells were maintained under standard culture conditions at 37 °C in a 5% CO₂ atmosphere, with medium replenishment and subculture performed at 2–3 day intervals. For treatment assays, cells were seeded at a density of 30,000 cells per well and incubated for 24 h to allow sufficient cell attachment prior to compound treatment. Additionally, biflavones 1–28 were dissolved in DMSO to yield concentrated stock solutions, which were then subsequently diluted in DMEM to the required experimental concentrations before application.

2.3. Cytotoxicity Test

Cytotoxicity was investigated using CCK-8. After treating the cells in the 96-well plates with the 28 biflavonoids for 24 h, the culture medium was discarded. CCK-8 working solution (CCK/DMEM, 1:10) was dispensed into each well at a volume of 100 μL, followed by the incubation of the cells for 40 min in the dark at 37 °C. After which, the absorbance was measured at 450 nm using a Tecan Infinite F200 microplate reader. The cell viability was expressed as a percentage relative to the untreated control group (DMEM only). Images of the control and biflavone 10-treated cells were captured using a cell imaging multimode reader (BioTek, Cytation 5).

2.4. Nitric Oxide Production Assay

NO production was evaluated using Griess reagent, following a previously reported protocol. For the mass screening, the cells were treated with various dimers at a concentration of 20 μM for 24 h. A 30 μL aliquot of cell culture supernatant from each well was subjected to reaction with Griess reagent. The absorbance measurements were obtained at a wavelength of 540 nm using a Tecan Infinite F200 microplate reader. The measured NO concentrations were normalized and expressed as a percentage of the LPS-treated DMSO reference group, which comprises DMEM with 100 ng/mL LPS in the absence of dimeric compounds.

2.5. Data Set Preparation, Alignment, and 3D-QSAR Modeling by Partial Least Squares Analysis

In general, the 24 synthesized dimers were used to evaluate the anti-inflammatory effects of biflavonoids, and the NO concentration (% of DMSO) was utilized to build the QSAR model. The two-dimensional structures of the ligands were transformed into three-dimensional configurations through the LigPrep tool available in Schrodinger Suite under the OPLS4 force field. Further geometric optimization of the 3D structures was carried out using the minimization tool in Maestro under the same force field with a gradient convergence threshold established at 0.01 kcal/mol/Å. After the minimization process, Tanimoto coefficient-based structural clustering was conducted to categorize the compounds. Ligands that belonged to the same clusters were aligned through ligand alignment, and the resulting aligned conformers were utilized for subsequent field-based 3D-QSAR modeling.

The model was constructed through the PHASE module, where PLS analysis was employed. NO inhibition (%) values were the dependent variables, while Gaussian intensities acted as the descriptors (independent variables). The 24 data set was randomly divided into training and test sets following an 80:20 distribution. The maximum PLS factor was set as 4, and the finalized field-based models were characterized by their steric, electrostatic, hydrophobic, hydrogen bond donor, and hydrogen bond acceptor (HBA) field contributions.

2.6. Molecular Docking

Computational molecular docking was performed through the Glide docking module available within the Schrödinger software environment (version 2023-1, Schrödinger, LLC, New York, NY, USA). Biflavone 10 was prepared with LigPrep, and the X-ray crystal structures of inflammatory targets, including cyclooxygenase-2 (COX-2) (Protein Data Bank (PDB) ID: 5IKT), hematopoietic cell kinase (HCK) (PDB ID: 2HCK), inhibitory κB-α/nuclear factor-κB (I-κB-α/NF-κB) complex (PDB ID: 1NFI), tumor necrosis factor-alpha (TNF-α) (PDB ID: 2AZ5), inducible nitric oxide synthase (iNOS) (PDB ID: 4NOS), and monocyte chemoattractant protein-1 (MCP-1) (PDB ID: 1ODK) were retrieved from the PDB. Structural preparation of each protein involved the removal of water molecules, the addition of hydrogen atoms, and the subsequent assignment of all-atom charges and atom types. Molecular docking was then conducted using Schrödinger. Biflavone 10 was docked into the active site of the target protein using the standard precision algorithm (SP).

2.7. Molecular Dynamics Simulations

Molecular dynamics simulations were carried out using GROMACS 2025.0. The protein–ligand complex was first parametrized with the CHARMM27 force field and TIP3P water model. The system was solvated in a triclinic box with a 1.0 nm distance between the solute and the box edges. System neutralization was achieved by adding counterions to reach a physiologically relevant 0.1 M concentration. The energy minimization was achieved by applying the steepest descent algorithm with a maximum of 50,000 steps and a force tolerance criterion of 1000 kJ/mol/nm. After which, the minimized system was then subjected to a sequential two-stage equilibration process: the first stage involved a 100 ps NVT equilibration at 300 K and the temperature was controlled via the V-rescale thermostat that employs a coupling constant of τ = 0.1 ps. The second stage comprises a 100 ps NPT equilibration at a constant pressure of 1 bar that was regulated by the Parrinello–Rahman barostat. Position restraints of 1000 kJ/mol/nm² were consistently applied to the protein backbone atoms and non-hydrogen ligand atoms throughout both equilibration stages to ensure preservation of the structural conformation. The production MD was then conducted for 50 ns. Trajectories were saved every 10 ps, and extensive trajectory analysis was performed using various GROMACS utilities. The system was centered on the protein, and periodic boundary conditions were removed using gmx trjconv. Structural stability was assessed via root mean square deviation (RMSD) (gmx rms) for protein backbone and ligand independently. Residue flexibility was determined using root mean square fluctuation (RMSF) (gmx rmsf). Protein–ligand interactions were characterized through hydrogen bonding patterns (gmx hbond) between the protein and ligand groups. The overall compactness of the system was monitored using the radius of gyration (gmx gyrate), while various energy components were analyzed using gmx energy to assess thermodynamic stability. Visualization of the MD trajectories was performed using ChimeraX, and graphical analysis of quantitative data was conducted using Grace software.

2.8. COX-2 Enzyme Inhibition Assay

The inhibitory activity of biflavone 10 against COX-2 was evaluated using a colorimetric COX-2 (human) inhibitor screening assay kit (Cayman Chemical, Cat no. 701080). The assay was performed according to the manufacturer’s protocol. Briefly, biflavone samples were dissolved in DMSO and diluted to the working concentrations (23 μM, 115 μM, 230 μM, 1150 μM, 2300 μM, and 11,500 μM) using DMSO. Reactions were initiated by adding arachidonic acid substrate to a mixture containing the COX-2 enzyme and the test compounds. The final concentrations in the wells were 1 μM, 5 μM, 10 μM, 50 μM, 100 μM, and 500 μM. The reaction was stopped after a 2 min incubation at 37 °C, and the resultant PG yield was quantitatively assessed via enzyme-linked immunosorbent assay. Optical absorbance was subsequently recorded at a wavelength of 405 nm employing a BioTek Epoch 2 Microplate Spectrophotometer interfaced with BioTek Gen5 data acquisition software (version 3.02). The data were analyzed using the Excel template provided by the manufacturer, and the inhibition curve was plotted using GraphPad Prism 9.

2.9. Statistical Analysis

Data comparisons between the control and each dimer were performed using an unpaired t-test with GraphPad Prism (version 9.5.1) for the cell viability test and NO inhibition. The values shown are mean ± SD, data comparisons across the groups for NO inhibition were performed using one-way analysis of variance (ANOVA) with posthoc Tukey’s test.

3. Results and Discussion

3.1. Nitric Oxide Inhibition and Cytotoxicity

As a preliminary study, the cytotoxicity of the 28 biflavones was determined at a concentration of 20 μM (Figure ). No significant morphological changes were observed in the biflavone-treated cells (Figure S1). The cell viabilities of all the biflavones are more than 75%, indicating that these biflavones have very low cytotoxicity at this concentration. Next, the NO inhibition of the biflavones was evaluated and reported as normalized NO concentration (Figure ). Biflavone 10 showed the highest anti-inflammatory activity, with the lowest normalized NO concentration at 20.94 ± 7.56%. In addition, the biflavones exhibited a variation in NO concentration, suggesting differences in their anti-inflammatory potency. This variability provides valuable data for establishing QSAR, which can help identify key structural features influencing NO inhibition.

3.

Cell viability of RAW 267.4 cells treated with 28 biflavones at a concentration of 20 μM. The values shown are mean ± SD, n = 6. *p < 0.05, **p < 0.01, and ***p < 0.001.

4.

Normalized NO concentration of cells treated with 28 biflavones at a concentration of 20 μM. The values shown are mean ± SD, n = 6. *p < 0.05, **p < 0.01, and ***p < 0.001. (a–f) Bar represented by different letters means normalized [NO] values are significantly different (p < 0.05) by one-way ANOVA with posthoc Tukey’s test.

3.2. Field-Based 3D-QSAR Analysis

The data set was allocated into a training set of 20 compounds and a test set of 4 compounds (Table S4). A field-based 3D-QSAR model was constructed to correlate the normalized NO concentration (% of DMSO) of the aligned flavonoid dimers with interaction energy field contributions at individual grid points. Model development was carried out using PLS regression by incorporating four PLS factors. The statistical performance of the optimal PLS4 model was assessed through various internal and external validation parameters. The PLS4 model demonstrated strong goodness-of-fit with a non-cross-validated R² of 0.86, surpassing the widely accepted threshold of R ² > 0.6 for a statistically valid QSAR model (Table ). Leave-one-out (LOO) cross-validation further yielded an R ²CV of 0.60, confirming the internal robustness of the model. The high variance ratio (F) and correspondingly low error probability (p-value) collectively indicated the overall predictive capacity of the selected model. External validation was subsequently evaluated, yielding a Q ² value of 0.652 and a Pearson correlation coefficient of 0.885, thereby reflecting a strong positive linear association between experimentally observed and model-predicted NO concentrations. Figure illustrates the scatterplots of experimental versus predicted NO concentrations, where the data points from both the training and test sets align closely with the linear regression line, thereby confirming the predictive accuracy and statistical credibility of the PLS4 model.

1. Partial Least Squares (PLS) Statistics for Field-Based 3D-QSAR Model.

factors	SD	R 2	R2CV	R 2 scramble	stability	F	P	RMSE	Q 2	Pearson-r
1	10.50	0.40	–0.04	0.21	0.770	11.8	2.96 × 10–3	16.30	–0.173	0.638
2	7.95	0.67	0.36	0.43	0.889	17.5	7.52 × 10–5	12.38	0.323	0.647
3	6.23	0.81	0.49	0.51	0.876	22.8	5.05 × 10–6	10.28	0.534	0.769
4	5.40	0.86	0.60	0.54	0.902	24.4	2.05 × 10–6	8.88	0.652	0.885

5.

Regression plot of experimental vs predicted NO concentration (% of DMSO) for the (A) training and (B) test set of field-based 3D-QSAR.

Among the contribution values for steric (0.247), electrostatic (0.109), hydrophobic (0.153), HBA (0.283), and hydrogen bond donor interactions (0.206), the molecular field analysis indicated that HBA and steric interactions contributed more significantly than others. Figure A provides the numbering system for the flavonoid to facilitate interpretation of the contour maps, where rings A1, B1, and C1 correspond to the radical precursor, and rings A2, B2/B3, and C2 correspond to the nucleophile precursor. As demonstrated in Figure B, the incorporation of sterically bulky substituents, such as hydroxyl or methoxy groups, at the C-8 position of ring A2, the C-2′ and C-3′ positions of ring B2, or the C-4′ position of ring B3 was found to be associated with enhanced anti-inflammatory activity. On the other hand, an increased steric hindrance at C-4′ of ring B3 appeared to decrease the anti-inflammatory effects. As illustrated in the HBA contour map in Figure E, the magenta contour regions at the C-4′ position of ring B2 suggest that the incorporation of HBA moieties, including hydroxyl or methoxy groups at this site, would lead to a decrease in anti-inflammatory effects. In contrast, red regions indicated that the presence of these groups at C-5 and C-7 of ring A1 would enhance the anti-inflammatory effect.

6.

Field-based 3D-QSAR contour maps based on biflavone 22 (highest anti-inflammatory effect) and biflavone 1 (lowest anti-inflammatory effect) in the training sets, (A) structure of biflavone 22 and 4, (B) steric, (C) electrostatic, (D) hydrophobic, (E) H-bond acceptor, and (F) H-bond donor contours maps.

3.3. Molecular Docking

Biflavone 10 exhibited the most potent anti-inflammatory activity among the 28 biflavonoids screened and was selected for molecular docking studies to investigate its potential interactions with key inflammatory mediators. These docking studies aimed to elucidate the possible mechanism underlying its anti-inflammatory effects by identifying its binding affinities and modes of interactions with various protein targets involved in inflammatory pathways. Figure shows the optimal docking conformations of biflavone 10; the docking score for each protein were as follows: COX-2 (−6.881 kcal/mol), HCK (−6.433 kcal/mol), iNOS (−5.174 kcal/mol), I-κB-α/NF-κB complex (−5.051 kcal/mol), TNF-α (−4.807 kcal/mol), and MCP-1 (−4.785 kcal/mol). , In general, the binding of biflavone 10 to key inflammatory mediators suggests a multitargeted approach to inflammation inhibition. Notably, the strongest binding was observed with COX-2, suggesting potential inhibition of PG synthesis, which may contribute to its anti-inflammatory and analgesic properties. Biflavone 10 also demonstrated strong binding to HCK, a nonreceptor tyrosine kinase that regulates the expression of proinflammatory cytokines, including TNF and IL-6. By binding to HCK, biflavone 10 could suppress cytokine production and potentially inhibit NLRP3 inflammasome activation. Additionally, the binding to iNOS could inhibit the production of NO, which acts as a regulatory and pro-inflammatory mediator. The binding to the NF-κB regulatory complex suggests that biflavone 10 may interfere with the dissociation of I-κB-α from NF-κB, thereby preventing NF-κB nuclear translocation and subsequent transcriptional activation of pro-inflammatory genes. Furthermore, the moderate binding affinity to TNF-α (docking score −4.807 kcal/mol) and MCP-1 (docking score −4.785 kcal/mol) indicates potential inhibitory effects on cytokine and chemokine signaling. As a key pro-inflammatory cytokine, TNF-α triggers inflammatory responses. By interacting with TNF-α, biflavone 10 may prevent its binding to cell surface receptors, thereby attenuating downstream inflammatory signaling cascades, including NF-κB and MAPK activation. , Similarly, the interaction with MCP-1 suggests a potential interference with monocyte recruitment to inflammation sites, thereby reducing immune cell infiltration and associated tissue damage.

7.

Molecular docking of biflavone 10 with inflammatory mediators. 3D binding conformations (left panels) and 2D interaction diagrams (right panels) of biflavone 10 with (A,B) COX-2, (C,D) MCP-1, (E,F) TNF-α, (G,H) HCK, (I,J) I-κB-α/NF-κB complex, and (K,L) iNOS.

Hydrogen bonding and π–π stacking were identified as the predominant interactions between biflavone 10 and these inflammatory targets, facilitating favorable binding. This comprehensive multitarget interaction profile suggests that biflavone 10 may effectively mitigate inflammatory responses through simultaneous inhibition of cytokine signaling, chemokine-mediated immune cell recruitment, NF-κB-driven gene transcription, and PG production.

3.4. Molecular dynamics

Since biflavone 10 exhibited the strongest calculated binding affinity for COX-2, the interactions between biflavone 10 and COX-2 were further investigated through MD simulations of the docked protein–ligand complex. The solvent and ions of the simulation system converged to a minimum energy level during energy minimization and subsequently attained NVT and NPT equilibria. Energy statistics confirmed a stable simulation system throughout the 50 ns trajectory, with consistent temperature (300.004 ± 0.183 K) and pressure (1.25 ± 0.24 bar), while the total energy (−1.204 × 10⁶ kJ/mol) showed appropriate conservation, indicating a well-equilibrated system.

The RMSD analysis of COX-2 backbone atoms provided insights into conformational changes of the protein–ligand complex over time. The data (Figure ) revealed progressive stabilization of the complex. The protein exhibited initial conformational adjustments (0.10–0.14 nm) during the first 15 ns, followed by a semistable plateau around 0.15–0.17 nm between 15 and 30 ns, before slightly increasing to 0.20–0.22 nm during the final 20 ns. This pattern reflects reasonable structural stability without abrupt changes that would suggest binding instability. Residual flexibility was determined using RMSF, identifying the regions of high mobility within the protein structure. The RMSF profile (Figure B) demonstrated region-specific flexibility within the COX-2 enzyme, with most residues displaying fluctuations between 0.05 and 0.15 nm. Notably, a significant peak of approximately 0.31 nm was observed around atom 10,000, corresponding to a flexible loop region that is distant from the active site. Importantly, residues comprising the catalytic site showed minimal fluctuations (below 0.1 nm), suggesting an effective stabilization of the binding pocket by the biflavone dimer. Furthermore, regions involved in direct ligand contact displayed reduced mobility compared with peripheral domains.

8.

MD simulation analyses of COX-2 in complex with biflavone 10. (A) Root mean square deviations (RMSD) of COX-2 protein, (B) root mean square fluctuation (RMSF) of protein residues, (C) number of hydrogen bonds formed between COX-2 and biflavone 10, and (D) radius of gyration (R _g) analysis of the protein–ligand complex.

The ligand–protein interactions were characterized through an intermolecular hydrogen bonding analysis, offering valuable information about the persistence of key binding interactions. The data (Figure C) revealed that the biflavone dimer maintained 1–2 hydrogen bonds with COX-2 throughout most of the simulation, with occasional formation of 3 bonds. This consistent hydrogen-bonding pattern contributed significantly to the stability of the complex. The initial formation of multiple hydrogen bonds at the beginning of the simulation indicated the rapid establishment of an optimal binding orientation, followed by dynamic yet stable interactions throughout the remainder of the trajectory. The overall compactness of the system was monitored using the radius of gyration (R _g), and various energy components were analyzed to assess thermodynamic stability. The R _g plot (Figure D) demonstrated that the COX-2-biflavone complex maintained consistent structural compactness throughout the simulation. The R _g values fluctuated within a narrow range (3.16–3.22 nm), indicating a preserved tertiary structure without significant expansion or compression upon biflavone binding. This structural integrity suggests that biflavone 10 fits appropriately within the binding pocket without disrupting the overall protein architecture.

On the basis of the MD trajectory snapshots spanning 0–50 ns (Figure ), the COX-2-flavone dimer complex demonstrates overall stability throughout the simulation. While the ligand underwent some conformational adjustments within the binding pocket during the first 20 ns, it maintained consistent interactions with key COX-2 residues across the entire trajectory. The absence of significant displacement or drift from the binding site and the persistent engagement with various protein residues, particularly histidine residues, suggest that the complex reached equilibrium and remained stable through the latter half of the simulation. This stability indicates favorable binding energetics and suggests that the flavone dimer forms a relatively stable complex with COX-2 under the simulated conditions.

9.

Protein–ligand conformations at every 10 ns of simulation for COX-2 with biflavone 10, showing the trajectory at 0, 10, 20, 30, 40, and 50 ns time points.

The comprehensive MD simulation analysis demonstrates that the biflavone 10 formed a stable complex with COX-2 throughout the 50 ns trajectory. The gradual stabilization of RMSD values, limited fluctuations in catalytically important residues, consistent hydrogen-bonding interactions, and well-maintained protein compactness collectively indicate favorable and persistent binding. Biflavone 10 effectively occupies the binding pocket while allowing the enzyme to maintain its natural conformational dynamics, suggesting that this compound could be a promising candidate for COX-2 inhibition. These findings provide strong computational evidence supporting the potential anti-inflammatory activity of the biflavone dimer through direct interaction with COX-2.

3.5. COX-2 Enzyme Inhibitor Assay

Since the computational docking results indicated that biflavone 10 exhibited strong binding affinity toward the COX-2 enzyme, experimental validation was performed using a COX-2 enzyme inhibitor assay. The COX-2 reaction was conducted, and the production of PG was quantified using an ELISA-based assay. Biflavone 10 was tested at five concentrations ranging from 1 to 500 μM (Figure ). This concentration range was selected based on the reported half-maximal inhibitory concentration (IC₅₀) value of amentoflavone, a structurally related biflavone. The IC₅₀ of amentoflavone is 0.123 mg/mL (equivalent to approximately 228 μM). At lower concentrations, biflavone 10 exhibited modest inhibition, with a percentage (%) inhibition of 22.02 ± 9.54% and 40.95 ± 1.70% at 1 μM and 10 μM, respectively. At the concentration of 100 μM, the % inhibition was 75.30 ± 7.99%. These results demonstrate that biflavone 10 inhibits COX-2 activity in a concentration-dependent manner. The observed COX-2 enzymatic inhibition observed in vitro, together with the suppression of NO production in RAW 264.7 macrophages, underscores the potential of biflavone 10 as an anti-inflammatory agent. However, as NO production is regulated by multiple signaling pathways beyond COX-2, , further studies are needed to delineate the molecular targets and pathways involved.

10.

Plot of percent (%) inhibition against different concentrations of biflavone 10. The values shown are mean ± SD, n = 2. (a–d) Bar represented by different letters means % inhibition are significantly different (p < 0.05) by one-way ANOVA with posthoc Tukey’s test.

4. Conclusions

This study investigated the anti-inflammatory effects of 28 biflavonoids, identifying biflavone 10 as the most potent compound through the inhibition of LPS-induced NO production in the RAW 264.7 cell line, with a normalized NO concentration of 20.94 ± 7.56%. These experimental results were used to construct a reliable 3D-QSAR model (R ² = 0.960, R²CV = 0.831, and Pearson’s r = 0.950), which highlighted the HBA interaction as the key determinant. Structural analysis revealed that the presence of hydrogen bond acceptors at C-4′ of ring B2 diminished anti-inflammatory effects, while their presence at C-5 and C-7 of ring A1 enhanced anti-inflammatory activity. To explore the possible molecular mechanism, biflavone 10 was subjected to docking and MD analyses. Molecular docking predicted the strongest binding affinity to COX-2 (−6.881 kcal/mol), followed by HCK (−6.433 kcal/mol), iNOS (−5.174 kcal/mol), I-κB-α/NF-κB complex (−5.051 kcal/mol), TNF-α (−4.807 kcal/mol), and MCP-1 (−4.785 kcal/mol), suggesting that biflavone 10 may exhibit anti-inflammatory effects through multiple targets. The experimental validation showed dose-dependent inhibition of COX-2 activity by biflavone 10, corroborating its strong predicted binding affinity.

Molecular dynamics simulations demonstrated that biflavone 10 forms a stable complex with COX-2, as evidenced by stable RMSD values, minimal fluctuations in key residues, consistent hydrogen bonding, and the preserved protein compactness. Taken together, this integrative approach provides a comprehensive understanding of the structure–activity relationships and possible molecular mechanisms underlying the anti-inflammatory activity of rare biflavones. These findings support the rational design of biflavone-based anti-inflammatory agents.

Glossary

Abbreviations

QSAR

quantitative structure–activity relationship

PLS

partial least squares

SD

standard deviation

HCK

hematopoietic cell kinase

COX-2

cyclooxygenase-2

I-κB-α/NF-κB

inhibitory-κB-α/nuclear factor-κB

TNF-α

tumor necrosis factor-alpha

iNOS

inducible nitric oxide synthase

RMSD

stable root mean square deviation

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c13338.

• Summary of reported NO inhibition activity of biflavones; original cell viability raw data from the CCK-8 assay; raw nitric oxide inhibition data; predicted normalized NO concentrations used for QSAR analysis; and images of control and biflavone 10-treated RAW 264.7 cells (PDF)

Hongling Zheng: data curation, conceptualization, and writingoriginal draft, review and editing. Yujia Cao: data curation. Fion Wei Lin Chin: data curation and conceptualization. Qiuyu Zhang: data curation. Ningping Zhan: data curation. Molan Zhang: data curation. Xin Yang: data curation. Joanne Yi Hui Toy: data curation and writingreview, and editing, Dejian Huang: conceptualization, supervision, and writingreview and editing.

The authors declare no competing financial interest.