Flowers from A. amica, N. oleander, and C. morifolium were acquired from Vellore district, Tamil Nadu, India. They flowers were authenticated by Dr. D. Stephen, Assistant Professor in the Department of Botany, American College, Madurai, Tamil Nadu, India. The flower petals were carefully removed and washed with distilled water to remove any impurities. In brief, 60 gm of fresh flower petals were correctly ground with 1× PBS at 1:2 ratio using a mixer grinder. The flower mixture was filtered through Whatman filter paper (pore size: 11 μm), then subjected to a series of centrifugations: 3,000 × g for 10 min, 6,000 × g for 20 min, and 10,000 × g for 30 min to eliminate larger cellular debris. The supernatant was carefully collected in a fresh tube, filtered through a 0.22-µm syringe filter, and subjected to ultracentrifugation for 2 h at 1,20,000 × g in 4 °C. The pellet obtained after ultracentrifugation was further washed with 1× PBS and again ultracentrifuged at 1,20,000 × g for 2 h. The supernatant was removed; the pellet was dissolved in ice-cold 1× PBS and stored at −80 °C for further experiments.

Following the isolation of the three different flower-derived ELNs of N. oleander (NELNs), C. morifolium (CELNs), and A. amica (AELNs), their size, distribution, and concentration were characterized by nanoparticle tracking analysis (NTA) using a NanoSight NS300 system. In addition, to further assess the colloidal stability of the isolated ELNs, zeta potential analysis was performed using a Particle Metrix ZetaView analyzer. Furthermore, FE-SEM (fluorescent emission scanning electron microscopy) was used to morphologically characterize floral ELNs. To visualize the ELNs through FE-SEM, floral ELNs were diluted ten-fold, and 100 μL of the sample was aliquoted on a cover slip. The ELNs were fixed using 2.5% glutaraldehyde, washed thrice with 1× PBS, and further dehydrated using 30%, 50%, 70%, 80%, 90%, and 100% ethanol. The samples were finally subjected to FE-SEM (Model: FEI Quanta 250 FEG, Thermo Fisher Scientific) for morphological characterization.

To characterize the bioactive constituents present in flower-derived ELNs, GC-MS analysis was employed for comprehensive metabolite profiling. For phytochemical screening, the floral ELN pellet was dissolved in 5 mL of absolute methanol until homogeneity in the solution was achieved (Iriawati et al., 2024), and the metabolome was further processed for GC-MS analysis. This analysis for the flower-derived ELNs was performed in TUV-SUD, Ranipet, India using an Agilent system with a DB-5 MS capillary column (30 m × 250 µm × 0.1 µm). Initially, the oven temperature was maintained at 50 °C and later increased to 150 °C until reaching the final temperature of 250 °C. As a carrier, helium gas was inserted at a pressure of 7.5 psi, and 1 µL of sample aliquot was injected into the column. Total mass spectra were obtained at 70 eV with sections ranging from 45 to 450 Da, with a scan interval of 0.5 s. The MassHunter software platform was used to analyze the mass spectra and chromatograms, and the relative percentage quantity of each component was determined by comparing its average peak area to the total areas. Metabolites retrieved at different retention times were represented as individual mass spectra and identified using the NIST (National Institute of Standards and Technology) database (Chikowe et al., 2025). From the hit list obtained, a score threshold of minimum 50% was considered on the basis of confidence level while selecting the metabolites (Chikowe et al., 2025).

The protocol for DPPH (2,2-diphenyl-1-picrylhydrazyl) assay was adopted following Iriawati et al. (2024). In brief, 0.2 mM of DPPH reagent was prepared with 100 mL of absolute methanol and incubated in the dark for 2 h. The ELN metabolites were purified using methanol, and 900 µL of 0.2 mM DPPH was added to 100 µL of ELN metabolites and further incubated in the dark for 30 min at room temperature. Absorbance was measured at 517 nm using a microplate reader (BioTrek, United States). Absolute methanol was used as blank, and only DPPH reagent was used as a control. The percentage of radical scavenging activity was measured using the equation “% radical scavenging = {(AC–AS)/AC} × 100,” where AC denotes absorbance of control and AS denotes absorbance of sample. Ascorbic acid was used as a positive control.

An MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) assay was performed to evaluate the effect of flower-derived ELNs on the cell viability of RAW264.7 macrophages, which were maintained using DMEM (Dulbecco’s modified Eagle medium) supplemented with 10% FBS (fetal bovine serum) and 5% antibiotic–antimycotic solution. In brief, 5 × 10³ cells were seeded in a 96-well plate and incubated in a humidified incubator at 37 °C, 5% CO₂ for 24 h. The cells were then treated with the different flower-derived ELNs—NELNs, CELNs, and AELNs—at varying concentrations ranging from 1 to 10 μg/mL overnight. MTT (5 mg/mL) was added to the cells post treatment and incubated for 4 h, followed by the addition of DMSO (100 μL) to assess the cells’ viability (Zhai et al., 2025).

Following the instructions provided by the nuclear extraction kit’s manufacturer (Cayman Chemicals, India), nuclear extracts were prepared from control and treated RAW264.7 macrophages. To further prevent protein degradation, the cells were quickly homogenized in phosphatase-inhibitor-containing ice-cold PBS. The homogenates were centrifuged at 300 × g for 5 min. Thence, the pellets were again suspended in the hypotonic buffer and detergent. The cytoplasmic fractions were then created by centrifuging at 14,000 × g for 30 s. The nuclear proteins were dissolved in lysis solution containing proteasome inhibitors after the cell nuclei had been lysed by the nuclear extraction buffer. Following the manufacturer’s instructions, the NF-κB (p65) transcription factor assay ELISA kit (Cayman) was used to assess the NF-κB DNA binding activity in the nuclear extract after being prepared.

In brief, LPS-induced RAW cells were treated with flower-derived ELNs and indomethacin as a positive control for 24 h. Post treatment, mRNA was isolated from the cells using the TRIzol method, and the mRNA was quantified using a NanoDrop (Eppendorf). This mRNA was further converted to cDNA using a PrimeScript™ RT reagent kit, and the cDNA was further amplified using SYBR Premix Ex TaqTMII (Tli RNase Plus) against specific forward and reverse primers (listed in Table 1) following the kit protocol. We used GAPDH as the internal standard, and the results were expressed in terms of fold change compared to control.

For in vitro validation, we used GraphPad Prism 8.0 for statistical analysis. The data were represented as mean ± SD, n = 3 (three technical replicates), and three independent biological replicates. Moreover, one-way ANOVA with Dunnett’s or Tukey’s multiple comparisons as required was utilized to analyze the group differences, with P-value <0.05 being statistically significant.

The metabolites identified from flower-derived ELNs were evaluated for their drug-likeliness property and toxicity profile using ProTox II. Drug-likeliness property can be demonstrated using the Lipinski rule of five, where the molecular weight of the ligand should be ≤500 kDa, H-bond donors ≤5, H-bond acceptor ≤10, and C log P (octanol-water partition coefficient) ≤5. The canonical smiles for each of the metabolites were inserted into ProTox II and the compounds were screened for favoring the Lipinski rule of five.

To understand the interaction of the flower-derived ELN metabolites with NF-κB p65–p50 heterodimer, the metabolites identified from GC-MS analysis were further subjected to molecular docking against NF-κB p65–p50 heterodimer (PDB id: 1VKX) to assess their binding affinities with 1VKX.

The ligand–protein interactions and binding stability of the selected docking poses were further investigated through classical all-atom molecular dynamics (MD) simulations (Vishwakarma and Mittal, 2024) with the help of GROMACS software. Initial coordinates for the simulations were obtained from the docking-derived conformations. The protonation states of protein residues were assigned at physiological pH 7.0, and hydrogen atoms were added accordingly. Initially, the topology files for proteins and ligands were prepared using the CHARMM27 all-atom force field and SwissParam web server. Each protein–ligand complex was solvated in a cubic box containing water molecules using the TIP3P model and neutralized with appropriate counter ions (Jorgensen et al., 1983; Jyoti et al., 2025). Energy was minimized to 10 kJ/mol. Following initial energy minimization, the system was subjected to a 500 ps NVT simulation with the protein atoms constrained, allowing relaxation of the solvent environment, followed by a 500 ps NPT equilibration to relax the protein atoms. Subsequently, 200 ns production MD simulations were performed under an NPT ensemble. The coordinates for every MD trajectory file were recorded for every 10 ps. Following this, parameters such as RMSD (root mean square deviation), RMSF (root mean square fluctuation), SASA (solvent-accessible surface area), Rg (radius of gyration), and H-bond were analyzed using XM-grace software.

Binding free energies were computed using MMPBSA (molecular mechanics Poisson–Boltzmann surface area) following MD simulation (Subramani and Venugopal, 2025). This software calculates the binding free energy between receptor and ligand. In this study, the final 10 ns were utilized to estimate the binding energies between the complexes, which were extracted from a 200 ns trajectory. The below approach was used for calculating the binding free energies, where ΔE_MM represents total molecular mechanics energy, ΔS denotes conformational entropy contribution (-TΔS), ΔG_PB represents polar contribution, and ΔG_SA refers to non-polar contribution. Δ G bind = G complex − G protein − G ligand = Δ H + Δ G solvation − T Δ S = Δ E MM + Δ G PB + Δ G SA − T Δ S .

ELNs were isolated using ultracentrifugation, then purified and resuspended in 1× PBS. These ELNs were also analyzed using NTA for size distribution and particle concentration. The NTA results confirmed that most of the isolated ELNs ranged 100–200 nm in diameter, and the particle concentration was 9 × 10⁹ particles/mL (NELNs), 8.5 × 10⁹ particles/mL (CELNs), and 9.7 × 10⁹ particles/mL (AELNs) (Figure 2). The average zeta potential of the isolated ELNs was found to range from −40 mV to −10 mV (Figure 2). The particle sizes of the ELNs were in accordance with previously evaluated morphological characterization of ELNs, and the negative zeta potential exhibited better stability for all the flower-derived ELNs; hence, we proceeded with further studies. For morphological assessment, isolated floral ELNs were visualized using FE-SEM (Figure 3). The ELNs exhibited vesicular morphology, with dimensions of 100–200 nm—consistent with our NTA findings.

Metabolome extraction was carried out from three different flower-derived ELNs using methanol following the protocol of Iriawati et al. (2024); the metabolite profiling was carried out using Agilent GC-MS system. A total of 33, 26, and 28 peaks were observed in ELNs derived from N. oleander, C. morifolium, and A. amica, respectively. Each peak designated the presence of metabolites relating to their molecular formula, retention time (RT), and molecular weight referred from NIST (Figure 4). However, based on a confidence level of at least 50% (Chikowe et al., 2025), the compounds were selected and documented with their peak area (%) and retention time in Tables 2–4. Most of the metabolites identified in the flower-derived ELNs have diverse protective roles such as antimicrobial, antioxidant, anti-inflammatory, anticancer and neuroprotective effects. For instance, the presence of compounds such as acetamide, 2-(adamantan-1-yl)-N-(1-adamantan-1-ylethyl), phenol, 2,5-bis(1,1-dimethylethyl)-, and n-hexadecanoic acid identified in N. oleander-derived ELNs are known to possess anticancer, antimicrobial, and antioxidant properties. On the other hand, the presence of 2-Caren-4-ol in C. morifolium-derived ELNs has antioxidant and antimicrobial properties, and hexadecanoic acid present in A. amica-derived ELNs is known to have a potent anti-inflammatory role.

A DPPH assay was performed to evaluate the antioxidant capacity of flower-derived ELNs at different concentrations, with ascorbic acid as a positive control. ELNs were observed to exhibit radical scavenging activity in a dose-dependent manner compared to the positive control (ascorbic acid) (Figure 5).

RAW264.7 macrophages were seeded and treated with NELNs, CELNs, and AELNs at different concentrations of 1–16 μg/mL for 24 h. The medium was then removed, and MTT (5 mg/mL) was used on the treated cells to check their viability. No toxicity was observed in NELN, CELN, and AELN doses ranging 1–8 μg/mL, 1–10 μg/mL, and 1–6 μg/mL, respectively. However, higher concentrations of all three types of flower-derived ELNs resulted in a marked reduction in cell viability (Figure 6). Based on the cell viability assay, the least cytotoxic concentrations of each ELN type were selected for subsequent experimental analyses.

To validate the anti-inflammatory potential of the flower-derived ELNs, LPS-induced RAW macrophages (1 μg/mL for 16 h) were treated with NELNs (2 μg/mL and 4 μg/mL), CELNs (4 μg/mL and 6 μg/mL), TELNs (2 μg/mL and 4 μg/mL), and indomethacin (positive control) (30 μg/mL) for 24 h. NF-κB binding activity was detected using ELISA, and it was significantly elevated in the LPS-treated condition, whereas flower-derived ELN-treated groups significantly reduced the condition compared to the control drug (Figure 7). Gene expression studies revealed an increase in IL-10 expression and decrease in TNF-α expression (expressed as fold change ±SD), confirming the anti-inflammatory role of NELNs, CELNs, and AELNs (Figures 8A,B) compared to positive control. TNF-α mRNA expression was significantly elevated during LPS simulation, increasing from 1 ± 0.05-fold in the untreated group to 1.41 ± 0.06-fold, while treatment with NELNs, CELNs, and AELNs significantly reduced the fold change increase to ∼1.25, ∼1.29, and ∼1.25, respectively. There was also a significant decrease in IL-10 mRNA expression, dropping from 1 ± 0.05 in untreated group to 0.20 ± 0.03 following LPS treatment. However, the same was restored upon treatment with floral ELNs, leading to an increase in fold change (NELNs: ∼0.6, CELNs: ∼0.6 and AELNs: ∼0.725) compared to LPS treatment. Notably, the higher dosage of flower-derived ELNs efficiently resulted in a significant increase of TNF-α expression and increase in IL-10 expression compared to the lower dosage for all the floral ELNs. Treatment with indomethacin as reference drug used at a dosage of 30 μg/mL showed significant reduction of NF-κB binding activity, TNF-α mRNA expression level (fold change: 1.29 ± 0.05), and a significant elevation of IL-10 mRNA expression (fold change: 0.45 ± 0.04) compared to the LPS-treated group in RAW macrophages.

The phytocompounds identified from the GC-MS analysis of ELNs derived from three different flowers were further evaluated for their drug-likeness properties. Ligand assessment was performed based on Lipinski’s rule of five, with the results summarized in Tables 5–7. Toxicity profiles were predicted using the ProTox-II server, highlighting the estimated LD₅₀ (median lethal dose), toxicity class, and organ-specific toxicities such as hepatotoxicity, neurotoxicity, and nephrotoxicity. As presented in Tables 5–7, most of the metabolites were consistent with Lipinski’s rule of five, indicating their favorable drug-likeness. Additionally, most compounds exhibited LD₅₀ values ranging from 500 to 10,000 mg/kg, corresponding to toxicity classes 3–6. Organ toxicity predictions revealed that most compounds were non-hepatotoxic, non-neurotoxic, and non-nephrotoxic in nature, suggesting a generally favorable safety profile. Based on the drug-likeness and toxicity assessments, the majority of the phytocompounds derived from flower-based ELNs demonstrated favorable pharmacological properties and a low predicted toxicity profile, highlighting that ELNs might serve as superior drug candidates with enhanced safety and efficacy profiles.

The different phytochemicals identified from flower-derived ELNs were further subjected to molecular docking against NF-κB p65–p50 heterodimer (PDB id: 1VKX) to assess the binding affinities of the different metabolites through AutoDock Vina software. The molecular docking data were assessed according to the binding affinities (in kcal/mol) represented in Tables 8–10. The top three compounds exhibiting the highest binding scores (Table 11) along with dexamethasone were visualized for molecular interactions using PyMOL and Discovery Studio Visualizer. Key hydrogen bonds, hydrophobic interactions, and π–π stacking interactions were noted to assess the quality of binding (Figure 9). The binding affinities varied between −2.6 and −8.0 kcal/mol, indicating moderate to strong predicted binding interactions. However, it is notable that among all the tested metabolites, 7H-3-[5-methyl-1-(4-methylphenyl)-1,2,3-triazol-4-yl]-1,2,3-triazol-4-yl]-6-(6-methoxynaphthalen-2-yl)-s-triazolo [3 (Ligand A) from N. oleander-derived ELNs (−10.0 kcal/mol), 2-methyl-5H-dibenz [b,f]azepine (Ligand B) from C. morifolium-derived ELNs (−6.4 kcal/mol), and methyl 2,2-diphenyl-3-methylbutanoate (Ligand C) from A. amica-derived ELNs (−8.7 kcal/mol) exhibited a higher binding score with 1VKX than the positive control dexamethasone (Ligand D) (−6.9 kcal/mol). The binding conformation of the top ligands from the three flower-derived ELNs were situated within the designated active site, forming a strong interaction with the key residues (Table 8).

To further validate and explore the dynamic behavior of the ligand–protein interactions predicted through molecular docking, molecular dynamics (MD) simulations were performed on the top-ranked ligand–NF-κB p65–p50 heterodimer complexes. The simulations were analyzed in terms of root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (ROG), solvent accessible surface area (SASA), and the number of hydrogen bonds (H-bonds) between molecules and the protein. To evaluate the conformational stability of the protein alone and the complexes over the course of MD simulations, RMSD was monitored over a period of 200 ns (Figure 10A). The unbound protein showed significant fluctuations, with RMSD values ranging approximately 2.5–4.5 nm with occasional spikes above 6 nm, indicating substantial flexibility and conformational rearrangements. In contrast, the ligand-bound complexes exhibited improved stability throughout the simulation. It is notable from Figure 10A that the red, blue, and green lines representing different ligand–protein complexes such as 7H-3-[5-methyl-1-(4-methylphenyl)-1,2,3-triazol-4-yl]-1,2,3-triazol-4-yl]-6-(6-methoxynaphthalen-2-yl)-s-triazolo [3 (ligand 1 in red line), 2-methyl-5H-dibenz [b,f]azepine (ligand 2 in green line), and methyl 2,2-diphenyl-3-methylbutanoate (ligand 3 in blue line) stabilized within the first 20–30 ns and maintained consistent RMSD values between 0.5 and 1.5 nm. Notably, 1VKX-ligand 1 displayed the highest RMSD among the ligated systems (∼1.4 nm), while 1VKX-ligand 3 maintained the lowest RMSD (∼0.8 nm), suggesting a more stable binding conformation. To understand whether the protein was flexible or rigid, RMSF analysis was carried out. Figure 10B depicts the fluctuation of the native protein 1VKX and the other three ligand–protein complexes. The apoprotein (black line, 1VKX) demonstrated higher fluctuations throughout the sequence, particularly in the loop and terminal regions; RMSF values peaked above 1.5 nm, with the highest flexibility in the amino acid residue regions 85–110, 250–290, and 310–340. Among all the protein–ligand complexes, 1VKX-ligand 3 (blue line) showed the lowest overall RMSF values, with fluctuations mostly below 0.3 nm, indicating a strong stabilizing effect on the protein backbone. On the other hand, 1VKX-ligand 1 and 1VKX-ligand 2 exhibited intermediate stabilization, where the former complex retained slightly higher flexibility, especially around residues 90–110 and 310–330. It is also notable that residues located within the predicted ligand binding pocket (residues 135–160 and 270–290) exhibited substantially reduced fluctuations in all ligand-bound systems compared to the apoprotein. The blue complex demonstrated the greatest dampening of motion in the binding site region, suggesting that it forms the most stable interactions and may represent the most potent binder among the tested ligands. Radius of gyration (ROG) was analyzed for the apoprotein and the three complexes to understand the overall compactness and folding behavior of the protein during MD simulations (Figure 10C). The ROG of the apoprotein (black line) fluctuated between 2.55 and 2.75 nm, with intermittent spikes approaching 2.85 nm, indicating a transient conformational expansion and a relatively flexible structure. In contrast, the ligand-bound complexes maintained a more compact and stable structure throughout the 200 ns simulation time period. It is notable that the ROG values for all three complexes remained consistent in the range of 2.30 to 2.45 nm with minimal fluctuations. Among all the complexes, 1VKX-ligand 3 exhibited the lowest ROG values (∼2.32–2.36 nm), suggesting maximum structural compactness and stability upon binding. Hence, the marked difference between the apoprotein and the ligand-bound states underscores the stabilizing effect of ligand association. These results suggest that ligand binding contributes to a more ordered and compact protein structure, likely due to enhanced intramolecular interactions and reduced structural flexibility. The degree of solvent exposure of the protein in the native and ligand-bound states was analyzed using the solvent-accessible surface area (SASA) (Figure 10D). The apoprotein (1VKX, in black) exhibited higher SASA values of 170–185 nm², indicating a more solvent-exposed and less compact structure. However, in comparing the SASA values, the ligand-bound complexes demonstrated reduced SASA values, 1VKX-ligand 3 (blue) maintained the lowest solvent accession (∼162–167 nm²), and the other two complexes (in green and red) showed moderate solvent accession (165–171 nm²) throughout the simulation. These findings support the observation that ligand binding reduces solvent accessibility by promoting protein compaction, in line with ROG and RMSF data. Notably, the blue complex again demonstrated the highest structural integrity and least solvent-exposed conformation. Subsequently, hydrogen bonding (H-bond) analysis was performed to monitor the stability and persistence of the binding interactions over the simulation period (Figure 10E). Consistently, 1VKX-ligand 3 and 1VKX-ligand 2 exhibited 5–7 and 2–4 instances of H-bonds, respectively, during most of the simulation time, while 1VKX-ligand exhibited 1–2 H-bonds, which were less persistent and stable over the simulation time. The H-bond analysis data for 1VKX-ligand 3 accords with the RMSD, RMSF, ROG, and SASA analyses. Hence, the trajectory analysis of the MD simulation revealed enhanced stability, compactness, and reduced flexibility in the ligand-bound complexes, with ligand 3 (methyl 2,2-diphenyl-3-methylbutanoate) exhibiting the most stable binding interaction with 1VKX.

MMPBSA calculations were used to determine the binding free energies—the energy change between bound and unbound states of the protein and ligand (Figure 11) for Complexes 1, 2, and 3. Table 12 demonstrates the total binding free energy as −14.42 kcal/mol for Complex 1, –15.48 kcal/mol for Complex 2, and –16.25 for Complex 3. It is notable from Figure 11 and Table 12 that Complex 3 (methyl 2,2-diphenyl-3-methylbutanoate – 1VKX) exhibited the least total binding free energy, demonstrating the ligand’s strong interactions with 1VKX.