Repurposing acetyldigitoxin as a potential EZH2 inhibitor for non-small cell lung cancer: a computational and experimental approach.

Introduction
Non-small cell lung cancer (NSCLC) remains one of the most prevalent and deadly malignancies worldwide, accounting for approximately 85% of all lung cancer cases [1, 2]. Despite advances in targeted therapies and immunotherapies, the prognosis for NSCLC patients varies significantly depending on the stage at diagnosis. According to recent data, the 5 year survival rate is 65% for localized disease, 37% for regional spread, and only 9% for distant metastasis, with an overall 5-year survival rate of 28% across all stages [3]. These stark statistics highlight the urgent need for novel therapeutic strategies that can effectively target the molecular drivers of NSCLC progression.
In recent years, innovative approaches such as therapeutic protein-based vaccines have emerged as promising immunotherapeutic avenues. These vaccines are designed to elicit targeted immune responses against tumor-associated antigens, offering a potential strategy for personalized cancer therapy. However, their clinical application in NSCLC remains challenging due to issues related to antigen selection, immune evasion, and vaccine delivery [4]. Alongside immunotherapies, computational drug discovery has transformed oncology by enabling the rapid identification and optimization of novel inhibitors. Advanced in silico strategies, particularly those employing multi-sampling algorithms and multi-simulation validation, have proven critical for uncovering multitargeted inhibitors with robust binding affinity and stability. For instance, recent studies have successfully identified compounds like 5-nitroindazole and Imidazolidinyl urea as promising multitargeted agents in lung cancer through integrated computational pipelines, demonstrating the power of these approaches in accelerating early-stage drug discovery [5, 6]. Among the key molecular drivers of NSCLC, the enhancer of zeste homolog 2 (EZH2), a catalytic subunit of the polycomb repressive complex 2 (PRC2), has emerged as a critical epigenetic regulator implicated in tumorigenesis, metastasis, and drug resistance [7–9]. EZH2 mediates gene silencing through the trimethylation of histone H3 at lysine 27 (H3K27me3), leading to the repression of tumor suppressor genes and the activation of oncogenic pathways [10, 11]. Overexpression of EZH2 has been observed in various cancers, including NSCLC, and is associated with poor clinical outcomes, making it an attractive therapeutic target [12–14]. Recent efforts to develop EZH2 inhibitors, such as tazemetostat and EPZ-6438, have demonstrated promising results in preclinical and clinical studies, particularly in hematologic malignancies and solid tumors [15–17]. However, their efficacy in solid tumors, including NSCLC, has been more variable, often hampered by primary or acquired resistance. Reported resistance mechanisms include compensatory activation of other signaling pathways and the acquisition of secondary mutations in EZH2 that reduce drug binding [18]. These limitations underscore the need to identify novel EZH2 inhibitors with distinct chemical scaffolds and improved therapeutic profiles.
Given that the development of new drugs is a time-consuming and costly process, drug repurposing has gained considerable interest as a strategy to identify existing FDA-approved compounds with potential anticancer activity. This approach is further strengthened by computational methodologies, including molecular docking, dynamics simulations, and binding free energy calculations, which allow for the efficient screening and validation of drug-target interactions. By using existing drugs with known safety profiles, we can accelerate the translation of potential therapies into clinical trials.
In this study, we aim to repurpose FDA-approved drugs as EZH2 inhibitors for the treatment of NSCLC. Adopting an integrated computational and experimental framework inspired by recent successful in silico campaigns [5, 6], we combine virtual screening, molecular dynamics simulation, and in vitro validation to identify and characterize promising compounds. Our goal is to uncover new potential EZH2-targeting agents that may offer improved efficacy and selectivity against NSCLC, thereby contributing to the expanding arsenal of targeted therapies for this challenging malignancy.

Material and methods

Docking-based virtual screening
The crystal structure of EZH2 was obtained from the Protein Data Bank (PDB) under the accession code 5WG6. FDA-approved drug structures were retrieved from the DrugBank database. The protein structure was preprocessed using AutoDock Tools by adding hydrogen atoms, assigning charges, and removing water molecules and extraneous ligands present in the crystal structure. Ligand structures were similarly prepared by adding hydrogen atoms and assigning appropriate charges using the same software. Docking studies were performed using AutoDock Vina [19]. All docking simulations were performed using Chain A of the EZH2 crystal structure (PDB: 5WG6). To validate our docking protocol, the co-crystallized inhibitor GSK126 was extracted and re-docked. The grid box (40 × 40 × 40 Å) was centered on the coordinates of the native GSK126 ligand (X = −81.89, Y = 2.43, Z = −55.64) to ensure comprehensive sampling of the canonical binding pocket. The re-docked pose of GSK126 demonstrated excellent agreement with the experimental conformation, with a root-mean-square deviation (RMSD) of 0.92 Å, confirming the accuracy of our docking parameters. We used the BIOVIA Discovery Studio Visualizer software to visualize the binding modes and evaluate their pose, score, and affinity [20]. This approach aligns with the emphasis on methodological transparency in computational drug discovery.

Molecular dynamics simulation
The molecular dynamics (MD) simulation was performed using GROMACS 2020 [21]. Employing the CHARMM36 force field for the protein and SwissParam for the ligand topology. The system was solvated in a rectangular box using the TIP3P water model, with a minimum distance of 10 Å from the protein to the box edges. Sodium and chloride ions were added to neutralize the system and to achieve physiological ionic strength. The simulation protocol included energy minimization to remove bad contacts, followed by a heating phase from 0 to 300 K in the NVT ensemble, and equilibration in the NPT ensemble. The production run was conducted for 100 ns under NPT conditions, with a time step of 2 fs and the leapfrog integration algorithm. Non-bonded interactions were managed using the Particle Mesh Ewald (PME) method for electrostatics, with a cutoff of 10 Å for van der Waals interactions. Temperature and pressure were controlled using the Nose–Hoover thermostat and the Parrinello-Rahman barostat, respectively. Analysis of the MD trajectories included calculation of root-mean-square deviation (RMSD), root-mean-square fluctuation (RMSF), and hydrogen bond interactions to assess the stability and dynamics of the system. Principal Component Analysis (PCA) was performed using GROMACS (gmx covar and gmx anaeig) to capture conformational transitions in the EZH2-ligand complexes. The covariance matrix was computed from backbone atoms (Cα, N, C) over the last 50 ns (5000 frames, every 10 ps), with the trajectory fitted to the backbone to remove rigid-body motions. Eigenvalues and eigenvectors were obtained, and the trajectory was projected onto the top three PCs. 2D projections, 3D motion trajectories, and extreme structures were generated for each complex to visualize sampling and transitions.

Binding free-energy and residue decomposition analysis
To calculate the binding free energies of the complexes formed between the selected hits and EZH2, we used the MMPBSA method implemented in the gmx_MMPBSA software package [22]. We input the complex structures and the trajectory data from the last 50 ns of the molecular dynamics simulations, sampling 5000 frames (saved every 10 ps). Convergence of the binding free energy was verified by calculating cumulative averages over the sampled frames, ensuring stable energy values. The MMPBSA calculations were performed using default parameters. The results of the MMPBSA calculations were used to evaluate the strength of the interactions between the selected hits and EZH2. We used the MMPBSA to calculate the binding free energy of each complex, as well as the decomposition of the binding free energy into its components, such as the van der Waals and electrostatic interactions, as well as the solvation free energy.

Inhibitor
Acetyldigitoxin (CAS No: 1111-39-3) was obtained from MCE (Monmouth Junction, NJ, USA) and prepared in sterile dimethyl sulfoxide (DMSO; Merck Millipore, Billerica, MA, USA). The stock solution was kept at − 20 °C until needed. The concentration of DMSO used as the vehicle control was maintained at or below 0.1%.

Cell culture
The NSCLC cell lines A549 and H1299, along with the normal lung cell line Beas-2B, were obtained from the China Cell Bank at the Shanghai Institute of Life Sciences, Chinese Academy of Sciences. The cancer cells were grown in RPMI-1640 medium (Gibco, USA) supplemented with 10% Fetal Bovine Serum (FBS, Sijiqing) and 1% penicillin (100 units/ml)-streptomycin (0.1 g/l) (Gibco, USA), and incubated at 37 °C in a 5% CO2 environment. The Beas-2B cell line, used as a control, was cultured in DMEM medium (Gibco, USA) under the same conditions. All cells were maintained in a humidified incubator at 37 °C with 5% CO2.

RNA-extraction, cDNA synthesis, and real-time PCR
To select appropriate cell lines, we assessed the expression of EZH2 at the mRNA level in two cell lines compared to Beas-2B using RT-PCR. Total RNA was extracted from the cells using TRIzol (Invitrogen, Life Technologies) and treated with DNase I to eliminate DNA contamination. The RNA concentrations were measured using a nanodrop instrument, and 1 μg of RNA was used for cDNA synthesis. RT-PCR was conducted using the StepOnePlus™ RT-PCR system (Thermo Fisher Scientific, MA, USA) and the SYBR Green detection method (Invitrogen/ThermoFisher Scientific, MA, USA). The primer sequences for EZH2 were as follows: forward (F), 5'-AAT CAG AGT ACA TGC GAC TGA GA-3'; reverse (R), 5'-GCT GTA TCC TTC GCT GTT TCC-3'. GAPDH was used as an internal control with the following primer sequences: 5'-GAGTCAACGGATTTGGTCGT-3' (F) and 5'-AATGAAGGGGTCATTGATGG- 3' (R). Each reaction was performed in triplicate, and gene expression was quantified using the -2ΔΔCt method, with GAPDH serving as the reference gene for normalization.

Cell viability assay
Cell viability was assessed using the MTT [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide] assay [23]. A549 and Beas-2B cells were plated in triplicate in a 96-well plate at a density of 5,000 cells per well and treated with acetyldigitoxin (ADT) at concentrations ranging from 0 to 120 nM for 48 h. To assess the specificity of ADT's effects toward EZH2 inhibition, parallel MTT assays were performed using GSK126 at concentrations ranging from 0 to 10 μM for 48 h. Following incubation, 20 μL/well of MTT solution (5 g/L) was added, and the cells were further incubated for 3 h at 37 °C. The medium was then removed, and 100 μL of DMSO was added to each well to dissolve the formazan crystals. Absorbance was measured at 570 nm and 630 nm using an ELISA plate reader (BioRad, USA), and IC₅₀ values were determined using GraphPad Prism software (version 8.0).
The cytotoxic selectivity for cancer cells over normal cells was expressed as the selectivity index (SI) [24], using the following formula based on IC₅₀ values:

Flow cytometry analysis of apoptosis and cell cycle
Apoptosis was evaluated using Annexin V-FITC/propidium iodide (PI) staining with the Annexin V-FITC/PI Apoptosis Detection Kit (BD Biosciences, San Jose, CA, USA). A549 and Beas-2B cells were treated with the IC₅₀ concentration of ADT (32.4 nM) for 24 h. Cells were then collected, washed with cold phosphate-buffered saline (PBS), and resuspended in 1X binding buffer. Each sample was stained with 5 µL of Annexin V-FITC and 5 µL of PI for 15 min in the dark at room temperature. For compensation and gating controls, unstained cells, cells stained with Annexin V-FITC only, and cells stained with PI only were prepared in parallel. Flow cytometry was performed immediately after staining using a BD FACSCalibur flow cytometer (BD Biosciences). For each sample, a minimum of 10,000 events were acquired. Data were analyzed using FlowJo software (Version 10.8.1). The gating strategy was as follows: cells were first selected on a forward scatter (FSC-A) vs. side scatter (SSC-A) plot to exclude debris. Doublets were excluded using FSC-H vs. FSC-A. The viable cell population was then analyzed on a FITC-A (Annexin V) vs. PE-A (PI) dot plot. Quadrants were set based on unstained and single-stained controls. Cells in the lower right quadrant (Annexin V + /PI-) were considered early apoptotic, and cells in the upper right quadrant (Annexin V + /PI +) were considered late apoptotic or necrotic. The total apoptotic cell percentage was calculated as the sum of early and late apoptotic populations.
For cell cycle analysis, A549 cells were harvested 48 h after treatment with varying concentrations of ADT (0, 20, 40, 80 nM). Cells were washed with PBS, fixed in cold 70% ethanol, and stored at −20 °C for at least 24 h. Before analysis, fixed cells were washed with PBS, treated with 100 µg/mL RNase A for 30 min at 37 °C, and stained with 50 µg/mL PI for 15 min in the dark at room temperature. Cell cycle distribution was analyzed using the same BD FACSCalibur flow cytometer, with a minimum of 15,000 events acquired per sample. Data were analyzed with FlowJo. The gating strategy involved selecting singlet events using FSC-H vs. FSC-A, followed by analysis of the PI-A signal (FL2-A channel) on a linear scale to generate a histogram of DNA content. The percentages of cells in the G0/G1, S, and G2/M phases were determined by the modeling software. Flow cytometry data were collected using a flow cytometer (BD Biosciences, San Jose, CA, USA) and analyzed using FlowJo software.

Gene expression modulated by EZH2 inhibitor
After exposure of A549 cells with ADT, their total RNA was extracted, converted to cDNA, and RT-PCR performed as described. Then, the expression of EZH2, Bax, Caspase-3, Bcl-2 and Cyclin D1 were evaluated. The relative expression level of genes was calculated using the 2−ΔΔCt method with GAPDH as an internal control and comparing the treated samples with the control (untreated) samples. The sequence primer pairs are listed in Table 1.

Western blot analysis
Cells were treated with ADT at concentrations of 0, 25, and 50 nM for 48 h in complete medium. Following treatment, cells were washed twice with ice-cold PBS and lysed in RIPA buffer (Sigma-Aldrich). Protein concentrations were determined using the BCA Protein Assay Kit (Thermo Fisher Scientific). Equal amounts of protein were then separated by 10% sodium SDS-PAGE and transferred to PVDF membranes (Millipore). Subsequently, membranes were incubated overnight at 4 °C with the following primary antibodies (Cell Signaling Technology) diluted in blocking buffer: anti-EZH2 (1:1000, #5246), anti-Bax (1:1000, #5023), anti-Bcl-2 (1:1000, #15,071), anti-Caspase-3 (1:1000, #9662), anti-Cyclin D1 (1:1000, #2978), anti-H3K27me3 (1:1000, #9733) and anti-Histone H3 (total H3; 1:1000, #4499) for normalization, and anti-β-actin (1:2000, #4967) as a loading control. After washing three times with TBST, membranes were incubated with HRP-conjugated secondary antibodies for 1 h at room temperature. Protein bands were then visualized using an ECL kit (Thermo Fisher Scientific), analyzed using ImageJ software, and normalized to β-actin levels.

Statistical analysis
All experiments were performed in triplicate, and data are presented as mean ± standard deviation (SD). Statistical significance was determined using one-way ANOVA followed by Tukey’s post-hoc test (p < 0.05) using GraphPad Prism software (version 8.0).

Results

Docking-based virtual screening
The docking scores and rankings for the top 40 compounds identified from the virtual screen of 1,450 FDA-approved drugs are provided in Supplementary S1. From these compounds, ADT and Cyproheptadine emerged as the top two scoring ligands with binding energies of -10.90 kcal/mol and −10.74 kcal/mol, respectively. These two top hits were selected for further validation alongside the known EZH2 inhibitor GSK126 (−10.60 kcal/mol), which was utilized as a positive control. Among these, ADT demonstrated the highest binding affinity, surpassing the GSK126. The interaction profiles of the GSK126 and the top two hits were analyzed in detail (Fig. 1) (Supplementary S2) (Table 2). GSK126 formed hydrogen bonds with TRP624 and hydrophobic interactions with residues such as TYR111, PHE665, and ILE109. Notable interactions included Pi-Pi stacking with TYR111 and Pi-Alkyl interactions with PHE665 and TYR661 (Fig. 1A). Cyproheptadine formed hydrogen bonds with TYR111 and ILE109, along with Pi-Pi stacking interactions involving TYR111 and TYR661 (Fig. 1B). ADT exhibited a diverse interaction profile, forming hydrogen bonds with SER664, GLY623, and CYS663. Hydrophobic interactions were also observed with residues like TYR111, TYR661, and CYS663, which contributed to its strong binding to EZH2 (Fig. 1C). Among the tested ligands, ADT demonstrated the most robust binding affinity and interaction diversity, suggesting its potential as a strong EZH2 inhibitor.

Table 2 provides an overview of the molecular docking results and interaction analysis of the ADT, Cyproheptadine and the GSK126. Each compound exhibits specific interactions with key residues of the EZH2 protein, contributing to their binding affinity.

Molecular dynamics simulations and stability analysis
To evaluate the stability and dynamics of the selected compounds (ADT and Cyproheptadine) in complex with EZH2, alongside GSK126, we performed 100 ns molecular dynamics simulations using the GROMACS 2023 software package. GROMACS is a versatile tool for simulating biomolecular systems, enabling detailed analysis of protein–ligand interactions in a solvated environment. Key metrics, including RMSD, RMSF, radius of gyration (Rg), solvent accessible surface area (SASA), and hydrogen bond occupancy, were calculated to assess conformational stability, flexibility, compactness, solvent exposure, and intermolecular interactions (Fig. 2A-E). Figure 2A illustrates the RMSD profiles of the protein backbone in the EZH2-ligand complexes. All simulated systems reached equilibrium and demonstrated stable binding profiles throughout the production phase. Notably, ADT displayed the lowest and most consistent RMSD (averaging ~ 0.12 nm), suggesting superior conformational stability compared to both GSK126 (~ 0.15–0.2 nm) and Cyproheptadine.

The RMSF analysis highlighted residue-specific flexibility in EZH2 (Fig. 2B). In the binding site (residues 335–340), Cyproheptadine induced RMSF values of 0.05–0.4 nm, indicating dynamic fluctuations. GSK126 showed similar patterns (0.053–0.3 nm), while ADT led to moderate flexibility in key interacting residues identified from docking and MMPBSA studies, such as ILE109, TYR111, ALA622, GLY623, CYS663, SER664, TYR661, HIS689, and TYR726. Within this pocket, TYR111 and CYS663/SER664 exhibited RMSF of 0.5–0.8 nm, consistent with their roles in hydrogen bonding and hydrophobic contacts. ALA622 and GLY623 remained stable (0.3–0.4 nm), supporting alkyl and carbon hydrogen bond contributions. Higher fluctuations were noted for HIS689 and TYR726 (0.4–0.5 nm), aligning with pi-alkyl interactions. Distal regions (e.g., 448–460 at 1.11 nm and 717–723 at 0.59 nm) displayed elevated RMSF, suggesting ligand-independent global motions. The radius of gyration (Rg) profiles (Fig. 2C) provided insights into the compactness of the complexes. All systems stabilized around 2.3–2.6 nm after initial adjustments, with ADT showing the lowest average Rg (~ 2.4 nm), indicative of a more compact and rigid structure. Cyproheptadine and GSK126 had slightly higher Rg values (~ 2.5 nm), with minor fluctuations suggesting comparable but less pronounced compactness.
SASA analysis revealed solvent exposure dynamics, with values fluctuating between 240 and 275 nm2 (Fig. 2D). ADT maintained lower SASA (~ 250 nm2), implying greater burial of hydrophobic surfaces and enhanced stability. Cyproheptadine exhibited peaks up to 270 nm2, reflecting transient exposure, while GSK126 showed stable intermediate values (~ 260 nm2), consistent with its reference role.
Finally, hydrogen bond occupancy demonstrated interaction persistence over time (Fig. 2E). ADT formed the most stable bonds (averaging 3–5 per frame), particularly with residues like TYR111 and CYS663, supporting its low RMSD. Cyproheptadine showed variable occupancy (1–4 bonds), with declines correlating to RMSD increases, while GSK126 maintained steady bonds (2–4), affirming strong EZH2 affinity.
PCA was conducted to delineate the essential dynamics and conformational transitions across the EZH2 complexes with ADT, Cyproheptadine, and GSK126, building on the stability indicators from RMSD, RMSF, Rg, and SASA. For each complex, the covariance matrix was derived from backbone atoms (Cα, N, C) over the last 50 ns of the trajectory (5,000 frames at 10 ps intervals) and diagonalized to obtain eigenvalues and eigenvectors. The eigenvalues consistently showed a steep decay, underscoring the prevalence of a few dominant modes in the fluctuation landscape. Specifically, the first three principal components (PCs) captured 65% of the total variance for ADT (PC1: 35%, eigenvalue ~ 2.5 nm2; PC2: 20%, ~ 1.5 nm2; PC3: 10%, ~ 1.0 nm2), 62% for GSK126 (PC1: 32%, ~ 2.8 nm2; PC2: 18%, ~ 1.6 nm2; PC3: 12%, ~ 1.1 nm2), and 58% for Cyproheptadine (PC1: 30%, ~ 3.2 nm2; PC2: 16%, ~ 1.8 nm2; PC3: 12%, ~ 1.3 nm2). This gradient in variance concentration—from ADT's highly coordinated motions to Cyproheptadine's more dispersed ones—mirrors the ligands' binding affinities and structural stabilization, with ADT exhibiting the most efficient reduction to an essential subspace.
The 2D projections onto PC1 and PC2 (Fig. 3) illuminated ligand-specific conformational sampling. For ADT (Fig. 3A), the landscape was markedly compact (~ 6 nm along PC1 from −2.1 nm to + 3.1 nm; ~ 8 nm along PC2 from −1.0 nm to + 6.9 nm), with > 80% of frames in a tight elliptical cluster near (0.0 to 1.0 nm, 0.5 to 2.5 nm), denoting a dominant stable basin and infrequent low-amplitude transitions. In contrast, GSK126 (Fig. 3B), displayed a moderately dispersed pattern (~ 9 nm along PC1 from −20.9 nm to −11.3 nm; ~ 11 nm along PC2 from -8.9 nm to + 2.4 nm), featuring a dense central cluster (~ 70% of frames at −17 to −14 nm, −5 to −1 nm) with a diffuse tail toward negative extremes, suggestive of transient domain adjustments. Cyproheptadine (Fig. 3C), showed the broadest exploration (~ 37 nm along PC1 from −25.0 nm to + 12.0 nm; ~ 15 nm along PC2 from −9.8 nm to + 5.7 nm), with a loose primary basin (~ 60% at −18 to −12 nm, −6 to −2 nm) and prominent extensions along positive PC1, indicating frequent large-scale rearrangements. These projections collectively reveal ADT's superior constraint on EZH2 dynamics (aligned with its lowest RMSD ~ 0.12 nm and Rg ~ 2.4 nm), GSK126's balanced stability (~ 0.15–0.2 nm RMSD, Rg ~ 2.5 nm), and Cyproheptadine's enhanced flexibility (~ 0.18–0.25 nm RMSD, Rg ~ 2.6 nm), correlating with their binding free energies (ADT: −34.73 kcal/mol; GSK126: −24.72 kcal/mol; Cyproheptadine: ≈ −28.50 kcal/mol).

Binding energy and per-residue decomposition analysis of EZH2-ligand complexes
The binding affinity and stability of the ligands GSK126, Cyproheptadine, and ADT with the EZH2 protein were assessed through MM-PBSA binding free energy calculations using gmx_MMPBSA, utilizing the last 50 ns of MD simulations. Table 3 presents the binding free energies, decomposed into van der Waals (∆EvdW), electrostatic (∆Eele), solvation free energy (∆GSOLV), and total binding energy (∆Gbinding) terms. ADT exhibited the most favorable ∆Gbinding of -34.73 ± 7.21 kcal/mol, surpassing Cyproheptadine (−20.78 ± 5.17 kcal/mol) and GSK126 (-24.17 ± 1.39 kcal/mol), indicating its superior inhibitory potential. The ∆EvdW was the dominant favorable contributor across all compounds, ranging from -38.97 ± 0.79 kcal/mol for GSK126 to -21.60 ± 0.10 kcal/mol for Cyproheptadine, underscoring the importance of non-polar interactions in stabilizing the complexes. Electrostatic contributions (∆Eele) were significant, with Cyproheptadine and ADT showing values of -19.42 ± 4.16 kcal/mol and -132.71 ± 5.36 kcal/mol, respectively, suggesting strong charge-based interactions with EZH2. Conversely, the solvation term (∆GSOLV) was unfavorable, ranging from 135.91 ± 3.56 kcal/mol for ADT to 22.95 ± 1.02 kcal/mol for GSK126, reflecting the energetic cost of desolvating the binding interface.
Complementing this, per-residue energy decomposition via MM/GBSA (Fig. 4) provided detailed insights into specific residue contributions. For GSK126, key residues TYR111 (-2.58 ± 2.52 kcal/mol) and CYS663 (-2.06 ± 1.32 kcal/mol) exhibited the most favorable total energy contributions, driven by van der Waals and non-polar solvation interactions within the hydrophobic pocket. Other notable contributors included PHE686 (−2.27 ± 2.47 kcal/mol) and TYR661 (−1.59 ± 1.78 kcal/mol), with ALA622 (-0.90 ± 0.99 kcal/mol) and PHE665 (-0.59 ± 0.78 kcal/mol) showing moderate effects, consistent with its stable reference profile. Cyproheptadine shared a similar pattern, with TYR111 (−1.51 ± 1.08 kcal/mol) and TYR661 (−1.08 ± 1.01 kcal/mol) as primary contributors, though less favorable, and minor roles from VAL657 (−0.38 ± 0.42 kcal/mol) and ILE109 (−0.22 ± 0.35 kcal/mol), aligning with its dynamic adjustments during MD simulations. ADT mirrored this binding profile, with TYR661 (−1.60 ± 1.17 kcal/mol), CYS663 (−1.72 ± 0.87 kcal/mol), and TYR111 (−1.52 ± 1.59 kcal/mol) as key contributors, supported by PHE665 (−0.97 ± 1.08 kcal/mol), THR678 (−0.96 ± 1.00 kcal/mol), and ILE109 (−1.14 ± 1.41 kcal/mol). The lower standard deviations for ADT (0.87–1.59 kcal/mol) compared to GSK126 (0.78–2.52 kcal/mol) indicate a more consistent binding pose, reinforcing its stability observed in RMSD profiles (Fig. 2).

These combined analyses highlight ligand-specific interaction strengths, with GSK126 serving as a reliable benchmark. ADT demonstrated superior performance, as evidenced by its strongest binding affinity (−34.73 kcal/mol), high stability (low RMSD/RMSF), consistent hydrogen bonding (average 2.5 bonds), and robust electrostatic interactions. Based on this data, ADT was selected for in vitro experiments and may potentially outperform the reference inhibitor GSK126.

Comparative analysis of EZH2 expression in beas-2B and NSCLC cell lines
To evaluate the expression levels of EZH2 in normal and non-small cell lung cancer (NSCLC) cells, we performed RT-PCR analysis using Beas-2B (normal bronchial epithelial cells) and NSCLC cell lines (A549 and H1299). The results revealed a significant upregulation of EZH2 in NSCLC cells compared to Beas-2B. Specifically, the relative expression of EZH2 in Beas-2B was normalized to 1, while H1299 and A549 cells showed 2.2-fold and fourfold higher expression, respectively (Fig. 5A). These findings suggest that EZH2 is overexpressed in NSCLC cells, with the highest expression observed in the A549 cell line. Therefore, we selected A549 cells for further experiments.

Cytotoxic effects of ADT on beas-2B and A549 cells
To test whether ADT affects cell viability, we performed MTT assays on Beas-2B and A549 cells. The cells were treated with ADT, whose structure is illustrated in Fig. 5B, at concentrations ranging from 0 to 120 nM for 48 h. As shown in Fig. 5C, ADT caused a dose-dependent reduction in cell viability in both cell lines. The IC₅₀ revealing that A549 cells were more sensitive to ADT treatment, with an IC₅₀ of 32.4 nM, compared to Beas-2B cells, which exhibited an IC₅₀ of 190 nM. The greater cytotoxicity toward A549 cancer cells suggest a potential selective anti-cancer activity of ADT against NSCLC cells.
The SI, calculated as IC₅₀ (Beas-2B) / IC₅₀ (A549), was approximately 5.9, indicating a favorable differential cytotoxicity toward the cancer cells. This value, which exceeds the common threshold of 2 often used to indicate selective cytotoxicity [25], aligns with the concept of targeting EZH2, a protein overexpressed in A549 cells but present at lower levels in normal Beas-2B cells. This suggests that the heightened sensitivity of cancer cells may be due to their specific dependency on oncogenic pathways like those driven by EZH2.
To address whether the observed cytotoxicity is EZH2-specific or partially due to general effects of cardiac glycosides (e.g., Na + /K + -ATPase inhibition), we compared ADT's effects to those of GSK126, a selective EZH2 inhibitor, in parallel MTT assays. GSK126 treatment resulted in dose-dependent cytotoxicity in A549 cells with an IC₅₀ of 1.92 μM, while Beas-2B cells showed minimal response (IC₅₀ > 10 μM), confirming selective targeting of EZH2-overexpressing cells (Fig. 5D). Although ADT exhibited nanomolar potency compared to micromolar for GSK126, both compounds displayed similar selectivity patterns (higher sensitivity in A549 vs. Beas-2B cells), supporting that ADT's effects are at least partially mediated through EZH2 inhibition. However, the greater potency of ADT may involve additional mechanisms, such as Na + /K + -ATPase inhibition, contributing to its overall cytotoxicity.

Induction of apoptosis and cell cycle arrest by ADT in NSCLC cells
To evaluate the effects of ADT on apoptosis and cell cycle distribution, we performed flow cytometry analysis on Beas-2B and A549 cells following treatment with ADT. Cell cycle analysis in A549 cells treated with 0, 20, 40, and 80 nM ADT demonstrated a dose-dependent increase in the proportion of cells in the G0/G1 phase. The percentage of G0/G1 cells rose from 48.5% in untreated cells to 50.4%, 55.5%, and 66.3% at 20, 40, and 80 nM ADT, respectively (Fig. 6A). This indicates that ADT induces G0/G1 phase arrest in A549 cells in a concentration-dependent manner.

Furthermore, ADT treatment significantly induced apoptosis, with 26.5% of cells becoming apoptotic compared to 7.8% in the control. At the molecular level, this was associated with a 2.5-fold upregulation of the pro-apoptotic gene Bax and a concurrent 0.37-fold down-regulation of the anti-apoptotic gene Bcl-2 (Fig. 6B).
These findings collectively suggest that ADT exerts its anti-cancer effects by inducing apoptosis and cell cycle arrest in NSCLC cells, with minimal impact on normal bronchial epithelial cells.

Effect of ADT on mRNA expression of key genes in A549 cells
To further elucidate the molecular mechanisms underlying the anti-cancer effects of ADT, we assessed the mRNA expression levels of several genes, including EZH2, Bax, Caspase-3, Bcl-2, and CyclinD1, in A549 cells following treatment with ADT at its IC₅₀ concentration. Compared to the untreated control group (A549-NC), the relative mRNA expression levels were significantly altered. Specifically, EZH2 expression decreased to 0.52-fold, while pro-apoptotic genes Bax and Caspase-3 were upregulated to 2.5-fold and 2.08-fold, respectively. In contrast, the anti-apoptotic gene Bcl-2 and the cell cycle regulator CyclinD1 were downregulated to 0.37-fold and 0.56-fold, respectively (Fig. 6C).
These results indicate that ADT treatment modulates the expression of genes involved in apoptosis and cell cycle regulation, promoting pro-apoptotic pathways and suppressing anti-apoptotic and proliferative signals in A549 cells. This further supports the role of ADT in inducing apoptosis and cell cycle arrest in NSCLC cells.

Protein-level changes and EZH2 inhibition by ADT
To corroborate the gene expression data and assess EZH2's enzymatic activity, Western blot was performed on A549 cell lysates following ADT treatment. ADT dose-dependently reduced EZH2 protein levels, with a 0.7-fold decrease at 50 nM compared to controls (p < 0.05) (Fig. 6 D, E). Consistent with apoptosis induction, Bax protein increased to 2.1-fold, while Bcl-2 decreased to 0.5-fold (p < 0.001). Cleaved Caspase-3, indicative of active apoptosis, was elevated to 2.9-fold at 80 nM (p < 0.0001), with full-length Caspase-3 showing minimal change. Cyclin D1 protein levels dropped to 0.56-fold (p < 0.001), supporting G0/G1 arrest. Importantly, histone H3K27me3 levels, a direct marker of EZH2 methyltransferase activity, were significantly reduced to 0.36-fold (normalized to total H3; p < 0.001), confirming functional inhibition of EZH2 beyond mere downregulation. β-Actin served as the loading control for non-histone proteins, ensuring consistent normalization across samples. These protein-level validations reinforce ADT's mechanism as an EZH2-targeted therapeutic in NSCLC.

Discussion
In this study, we employed a robust computational pipeline combining molecular docking and molecular dynamics simulations to identify the potential of repurposed FDA-approved drugs as EZH2 inhibitors, a key epigenetic regulator implicated in NSCLC progression. Our integrated computational and experimental strategy exemplifies the growing power of such approaches to rapidly identify novel therapeutic applications for existing drugs, accelerating the oncology drug discovery pipeline. This aligns with the expanding role of AI in lung cancer, which is proving transformative not only in diagnostics but also in the identification of novel therapeutic agents [26, 27].
Our study positions acetyldigitoxin (ADT) as a promising repurposed inhibitor of EZH2 for NSCLC treatment, outperforming the positive control GSK126 and the alternative candidate Cyproheptadine in key metrics of binding affinity, stability, and conformational dynamics. ADT is structurally related to its parent molecule, digitoxin, a well-known cardiac glycoside, and shares some of its pharmacological properties [28–30]. It is traditionally used in the treatment of heart failure and atrial arrhythmias due to its ability to inhibit the Na + /K + -ATPase pump, thereby increasing intracellular calcium levels and enhancing cardiac contractility [31, 32]. However, its potential as an anticancer agent has gained attention in recent years, with studies suggesting its role in inducing apoptosis, modulating cell cycle progression, and inhibiting cancer cell proliferation [29].
The integrated computational analyses collectively elucidate why ADT emerges as the superior candidate. Molecular docking highlighted ADT's superior affinity compared to GSK126 and Cyproheptadine, driven by a robust network of hydrogen bonds and hydrophobic interactions. These interactions anchor ADT firmly within the EZH2 binding pocket, as validated by MD simulations, where ADT exhibited the lowest RMSD and RMSF values, indicative of minimal structural deviations and enhanced rigidity in critical residues. In contrast, GSK126 showed stable binding with greater conformational flexibility, while Cyproheptadine displayed higher fluctuations, suggesting less optimal pocket occupancy.
Beyond achieving the highest docking score, ADT’s dominance is rooted in its capacity to form a stable and conformationally restricted complex with EZH2. This is evidenced by its persistent hydrogen-bond network, compact protein–ligand interface (low Rg/SASA), and most favorable MM-PBSA binding free energy. Crucially, Principal Component Analysis revealed that ADT binding effectively minimizes conformational entropy in EZH2, confining the protein to a narrow energetic basin. This contrasts with the broader dynamics induced by GSK126 and Cyproheptadine. Therefore, ADT’s predicted efficacy stems not only from strong initial affinity but from its ability to stabilize a specific, low-energy binding mode that likely prolongs target engagement, a key advantage for a therapeutic inhibitor.
Subsequent in vitro experiments align with these emerging insights demonstrate that ADT exhibits significant anti-cancer activity in NSCLC cells. Our RT-PCR analysis revealed that EZH2 is significantly overexpressed in NSCLC cell lines compared to normal bronchial epithelial cells. We specifically selected the A549 NSCLC cell line due to its significantly high basal EZH2 expression, making it an ideal model to test the efficacy of a putative EZH2 inhibitor. EZH2 expression was 2.2-fold higher in H1299 cells and fourfold higher in A549 cells, with the latter showing the highest expression. This finding aligns with previous studies that have implicated EZH2 overexpression in cancer progression and poor prognosis in various malignancies, including NSCLC [9, 33]. The elevated expression of EZH2 in A549 cells likely contributes to their aggressive phenotype, making them a suitable model for evaluating EZH2-targeted therapies.
The results demonstrate that ADT exhibits selective cytotoxicity toward A549 cells, a model for NSCLC. This selectivity is quantified by a selectivity index (SI) of ~ 6, which is well above the common threshold of 2 used to indicate promising selective cytotoxicity for anticancer agents, suggesting a potential therapeutic window for ADT. This aligns with its mechanism of targeting EZH2, a dependency overexpressed in the cancer cells, and suggests a potential for minimized off-target effects. This selectivity aligns with previous findings on cardiac glycosides, such as digitoxin [34], digoxin [29], and ouabain [35], which have been shown to inhibit the proliferation of lung cancer cells with 3- to tenfold greater cytotoxicity compared to nonmalignant lung cells [36, 37].
To address potential concerns regarding the specificity of ADT's effects, we compared its cytotoxicity to that of GSK126, a selective EZH2 inhibitor. Both compounds exhibited selective cytotoxicity toward A549 cells over Beas-2B cells, with similar patterns of higher sensitivity in EZH2-overexpressing cancer cells. However, ADT's nanomolar IC₅₀ compared to GSK126's micromolar IC₅₀ suggests that while EZH2 inhibition contributes to ADT's effects, its greater potency may involve synergistic mechanisms, such as Na + /K + -ATPase inhibition inherent to cardiac glycosides. This dual-action profile could enhance ADT's therapeutic efficacy but warrants further investigation to delineate the relative contributions of each pathway. Future studies incorporating EZH2 knockdown or overexpression could provide definitive evidence of specificity. The selective cytotoxicity of ADT may be attributed to its ability to inhibit EZH2, which is overexpressed in NSCLC cells but present at lower levels in normal cells. These results underscore the potential of cardiac glycosides, including ADT, as promising candidates for targeted cancer therapy, particularly in NSCLC, where selective anticancer activity is essential for improving therapeutic outcomes.
To elucidate the mechanisms underlying the antiproliferative effects of ADT, we conducted apoptosis analysis and cell cycle distribution assays. Our results revealed that ADT induced apoptosis in A549 cells and caused a dose-dependent increase in G0/G1 phase arrest, suggesting that its anticancer activity is mediated through disruption of cell cycle progression and promotion of apoptotic cell death in NSCLC. These findings are consistent with earlier studies highlighting the shared mechanism of action among cardiac glycosides in targeting cell cycle regulation. For instance, Lindholm et al. [30] demonstrated that digitoxin treatment induced G0/G1 phase arrest in pancreatic cancer cells (BxPC-3 and CFPAC-1), underscoring its ability to inhibit proliferation by disrupting the cell cycle. Similarly, evidence of G0/G1 arrest has been reported in non-small cell lung cancer and large cell lung cancer following treatment with cardiac glycosides. Chou and colleagues [38] further supported this mechanism, showing that digoxin and digitoxin significantly inhibited cell growth and induced G0/G1 arrest in the human ovarian cancer SKOV-3 cell line. Cells in the G0/G1 phase, characterized by a resting, non-proliferating state with basal oxidative metabolism, are often associated with high chemoresistance, particularly in migrating and invading cancer cells [39]. However, it is worth noting that some studies have reported cell cycle arrest in the G2 phase induced by cardiac glycosides. For example, Gan et al. [40] observed G2 arrest in HeLa cells treated with digitoxin, attributed to DNA double-strand breaks, while Wang et al. [41] found differential responses to digoxin in NSCLC-derived cell lines, with one line arrested in G1 and the other in G2. These variations suggest that the cell cycle effects of cardiac glycosides may be context-dependent, influenced by cell type and specific molecular mechanisms.
The observed G0/G1 phase arrest in our study is consistent with the role of EZH2 in regulating cell cycle progression. EZH2 is known to promote the expression of genes involved in cell cycle regulation, such as CyclinD1 [42], and its inhibition by ADT likely contributes to the observed cell cycle arrest. Additionally, the induction of apoptosis may be mediated through the modulation of pro-apoptotic and anti-apoptotic signaling pathways, as evidenced by the changes in gene expression observed in our RT-PCR analysis. Specifically, ADT treatment led to a significantly upregulation of pro-apoptotic genes Bax and Caspase-3, while notably downregulation of anti-apoptotic gene Bcl-2 and the cell cycle regulator CyclinD1. Cyclins and cyclin-dependent kinases (CDKs) function as major switches that regulate the cell cycle [43]. Specifically, Cyclin D1, which partners with Cdk4 and Cdk6, facilitates the cell cycle by promoting the transition from the G1 phase to the S phase [44]. The alterations in gene expression observed in our study are in line with the increased apoptosis and G0/G1 phase arrest, further supporting the anti-cancer effects of ADT. Prior studies demonstrate that digitoxin and its synthetic monosaccharide analog significantly reduce lung cancer cell viability by inhibiting Na + /K + -ATPase, disrupting intracellular ion homeostasis, and activating apoptotic pathways [45, 46].
The downregulation of EZH2 to 0.52-fold following ADT treatment suggests that ADT directly or indirectly inhibits EZH2 activity, leading to the restoration of tumor suppressor gene expression and the suppression of oncogenic pathways. Beyond mRNA expression, Western blot analysis confirmed that these transcriptional changes translated to significant alterations at the protein level. Critically, we observed a substantial decrease in global H3K27me3 levels, a definitive biochemical readout that confirms ADT’s functional inhibition of EZH2 enzymatic activity. This provides direct evidence of target engagement, demonstrating that ADT acts not merely by reducing EZH2 expression but by potently blocking its methyltransferase function within the PRC2 complex. The resulting pro-apoptotic protein signature (upregulated Bax and cleaved Caspase-3, downregulated Bcl-2) and the reduction of Cyclin D1 provide a coherent protein-level explanation for the induction of apoptosis and cell cycle arrest, respectively. This mechanism aligns with the known role of EZH2 in epigenetic silencing and highlights the potential of ADT as an epigenetic modulator in cancer therapy. These findings are consistent with the broader therapeutic potential of EZH2 inhibitors, such as tazemetostat and EPZ005687, which have demonstrated significant anticancer effects in various cancers by inducing cell cycle arrest, apoptosis, and modulating the tumor microenvironment [15]. For instance, tazemetostat has shown efficacy in relapsed/refractory follicular lymphoma [16], while EPZ005687 selectively targets EZH2-mutant lymphomas through H3K27 methylation inhibition [17]. Additionally, EZH2 inhibitors like EPZ6438 and GSK126 have been shown to repolarize tumor-associated macrophages, enhancing antitumor immunity in colorectal cancer [47]. However, the context-dependent effects of EZH2 inhibition, such as the recruitment of immunosuppressive monocytes in mesothelioma [48], underscore the need for combination therapies to maximize therapeutic outcomes. The ability of ADT to downregulate EZH2 further supports its potential as part of a broader strategy to target EZH2-driven oncogenic pathways and highlights its promise as an epigenetic modulator in cancer treatment. The potential underlying mechanism of ADT is displayed in Fig. 7.

This study introduces acetyldigitoxin as a potential EZH2 inhibitor, demonstrating its potential as a repurposed drug for NSCLC treatment. The integration of computational and experimental approaches, including molecular docking, molecular dynamics, and in vitro assays, provides compelling evidence for ADT's selective cytotoxicity toward NSCLC cells and its multi-modal anti-cancer effects.
However, potential clinical translation of ADT must be considered in the context of the well-documented cardiotoxicity associated with cardiac glycosides. The therapeutic use of these compounds is historically narrow, primarily limited by their potential to cause severe arrhythmias through potent inhibition of the Na + /K + -ATPase pump in cardiac muscle [49, 50]. Nevertheless, several strategic considerations could potentially mitigate this risk in an oncological context. First, the plasma concentrations required for the anticancer effects observed here may be lower than those associated with overt cardiotoxicity, suggesting a possible therapeutic window. Second, developing targeted delivery systems, such as nanoparticle encapsulation or tumor-homing conjugates, could enhance drug accumulation at the tumor site while sparing cardiac tissue. Finally, exploring lower, metronomic dosing regimens or using ADT in combination with other agents could achieve synergistic anticancer effects while keeping peak plasma concentrations below the cardiotoxic threshold. Therefore, the repurposing of ADT for NSCLC should be pursued with a clear focus on these pharmacological challenges, where detailed pharmacokinetic/pharmacodynamic (PK/PD) studies and rigorous cardiac monitoring in preclinical models will be paramount to defining its true therapeutic index.
The current study has several limitations. First, it lacks in vivo validation, exploration of off‑target effects, and investigation of ADT’s efficacy in additional NSCLC cell lines or patient‑derived models. Furthermore, while our data strongly suggest EZH2 targeting by ADT, direct biochemical proof of enzyme inhibition, such as in vitro methyltransferase assay or surface plasmon resonance, is lacking. The observed reduction in H3K27me3 levels could result from both direct inhibition and downregulation of EZH2 expression; therefore, future studies employing such direct binding and activity assays are required to unequivocally establish ADT as a bona fide EZH2 inhibitor. Additionally, our functional characterization focused on the A549 NSCLC cell line, a choice based on its high basal EZH2 overexpression, which provided a robust proof‑of‑concept model but limits the generalizability of our findings. Future work should include validation in other relevant models, such as H1299 and additional molecular subtypes of NSCLC, to strengthen the biological relevance of ADT. Meanwhile, to further validate the specificity of ADT for EZH2, studies employing genetic models, such as EZH2‑knockdown cancer cells or EZH2‑overexpressing normal cells, will be essential. While our demonstration of reduced H3K27me3 provides supporting biochemical evidence of on‑target inhibition, these experiments will help definitively attribute the phenotypic effects to EZH2 modulation. Furthermore, given that ADT is a prodrug of digitoxin, future studies should investigate whether the parent compound digitoxin also exhibits EZH2‑modulatory activity, which could provide additional insights into the structure‑activity relationship of cardiac glycosides as epigenetic modulators. Despite these limitations, this study lays a strong foundation for future research into ADT as a targeted therapeutic agent for EZH2‑driven cancers.

Conclusion
In conclusion, our results provide strong preliminary evidence for the therapeutic potential of ADT as a putative EZH2 inhibitor in NSCLC. The repurposing of ADT for cancer therapy offers a promising strategy to accelerate the development of effective treatments for NSCLC, particularly for patients with EZH2-driven tumors. Future research should focus on elucidating the detailed molecular mechanisms of ADT and evaluating its efficacy in preclinical and clinical settings.

Supplementary Information
Below is the link to the electronic supplementary material.