Identification of potential natural BCL-2 inhibitors for leukemia through an integrated virtual screening approach

Abstract

Leukemia, a hematologic malignancy, frequently involves the overexpression of BCL-2 proteins, contributing to cell survival and apoptosis resistance. Although current BCL-2 inhibitors, including Venetoclax, have shown efficacy, they are limited by adverse side effects and restricted bioavailability. This study employs an integrated computational approach to identify natural BCL-2 inhibitors by screening 407,270 natural compounds. Using ligand- and structure-based virtual screening, we identified two promising compounds, CNP0237679 and CNP0420384, demonstrating strong binding affinities, favorable electronic properties, and stability in dynamic aqueous environments. Both compounds exhibited promising toxicity and pharmacokinetic profiles, indicating the possibility of low toxicity and high bioavailability. They are potential candidates for developing selective BCL-2 inhibitors because of these characteristics. The findings represent a positive step in developing new, naturally derived BCL-2 inhibitors and support additional in vitro and in vivo studies to validate their therapeutic efficacy and safety characteristics.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40360-025-01005-y.

Introduction

Leukemia, a malignancy affecting the bone marrow and blood, is primarily categorized into four types: Acute Lymphocytic Leukemia (ALL), Chronic Lymphocytic Leukemia (CLL), Acute Myeloid Leukemia (AML), and Chronic Myeloid Leukemia (CML). According to the World Health Organization (WHO), there were approximately 475,519 new leukemia cases and 311,594 related deaths globally in 2020 [1]. The overexpression of the B-cell lymphoma-2 (BCL-2) protein family is implicated in over 70% of lymphoid malignancies, highlighting its significant role in cancer pathogenesis [2].

A significant challenge in cancer therapy is reducing the risks of secondary malignancies and the cytotoxic and genotoxic impact of chemotherapy agents on normal cells. This necessitates the development of therapies that can specifically trigger cell death in cancerous cells without affecting healthy cells. Apoptosis, an evolutionarily conserved and essential cellular process, is crucial for eliminating damaged, aged, necrotic, or otherwise harmful cells, thereby supporting tissue development, physiological functions, and homeostasis [3, 4]. The BCL-2 protein family, comprising both pro-apoptotic (e.g., BAX, BAK) and anti-apoptotic proteins (e.g., BCL-2, MCL-1, BCL-XL), plays a pivotal function in controlling cell death and is a primary target in oncotherapy. The BCL-2 protein, initially identified in follicular lymphoma through characteristic 14;18 chromosomal translocations, inhibits apoptosis, thus promoting cancer cell survival and oncogenesis [5]. Anti-apoptotic BCL-2 proteins share structural similarities essential for cell survival, while pro-apoptotic members like BAX, BAK, and BH3-only proteins (e.g., BIM, PUMA, BAD) promote apoptosis by counteracting anti-apoptotic proteins. The levels of mitochondrial proteins, particularly BCL-2 and BAX, influence apoptosis initiation. A higher Bax/BCL-2 ratio correlates with mitochondrial cytochrome c release, activating caspases-9 and − 3, subsequently initiating programmed cell death. Additionally, an increase in the BAX/BCL-2 ratio is associated with poly (ADP-ribose) polymerase (PARP) fragmentation [5].

Despite the critical function of BCL-2 proteins in controlling apoptosis, present therapeutic approaches face limitations. Venetoclax, a selective BCL-2 inhibitor approved by the FDA in 2016 for treating CLL and AML, exemplifies these challenges. Although effective, Venetoclax can cause gastrointestinal side effects, neutropenia, poor in vivo stability, low oral bioavailability, and high cost, highlighting the need for new BCL-2 inhibitors that enhance efficacy, safety, and bioavailability.

The drug development process is costly and time-intensive, often requiring over $2.6 billion and 12 to 14 years to introduce a new pharmaceutical to the market [6]. Given the high cancer mortality rates and the slow pace of novel drug discovery, repurposing existing drugs with established safety profiles presents an efficient and appealing alternative, as this approach can significantly reduce both development time and costs. Advances in computational methods have further transformed drug discovery, particularly in virtual screening. Traditional molecular docking, which predicts preferred orientations of molecules to form stable complexes, often fails to capture the dynamic nature of proteins. Enhanced methods, such as quantum mechanical calculations, thermodynamic stability and dynamics assessments, and time-resolved trajectory analyses, provide a more precise representation of protein-ligand interactions. Numerous studies have utilized virtual screening to identify small molecules for oncotherapy targeting the BCL-2-BH4 domain, impairing its activity and initiating cell death in tumor cells [7, 8].

Natural products have proven valuable as lead compounds in drug discovery due to their chemical variability and biological activity. Although natural products have been investigated for in silico drug discovery, previous studies to identify potential BCL-2 inhibitors were generally limited to individual databases like NP Atlas and Zinc Natural Products [9, 10]. In contrast, the Collection of Open Natural Products (COCONUT) database compiles data from 53 databases and literature sources, including elucidated and anticipated natural compounds sourced from multiple open databases [11]. Screening this more comprehensive database provides a broader selection of compounds for discovery.

Given the therapeutic potential of natural compounds and the efficiency of computational methods in identifying new drug candidates, this study employs an integrated computational approach to screen 407,270 natural compounds for potential BCL-2 inhibitors. This approach aims to address the limitations of earlier screening methods and enhance the chances of identifying novel, effective, and natural BCL-2 inhibitors, thereby advancing new therapeutic options for leukemia and other BCL-2-associated cancers.

Methodology

This study aims to identify and screen potential natural BCL-2 inhibitors through integrated virtual screening approaches, with the overall workflow illustrated in Fig. 1.

Fig. 1.

A flowchart illustrating the workflow employed in this study

Ligand retrieval

The COCONUT database (v2022), containing 407,270 compounds, was screened to identify potential BCL-2 inhibitors [11]. Using the QikProp module within the Schrödinger Release 2024-1, compounds were filtered based on Lipinski’s Rule of Five, resulting in a refined subset of 276,409 compounds. Subsequent compound pre-processing was conducted using the LigPrep module with the OPLS4 force field. Epik was employed to predict protonation states at a physiological pH of 7.0 ± 2.0, ensuring that chirality was preserved, tautomers were generated, and up to 16 low-energy conformations were produced for each ligand [12].

Pharmacophore modelling and screening

To create the pharmacophore model, the top 100 compounds targeting BCL-2 with the lowest IC₅₀ values were selected from the BindingDB database [13]. These compounds underwent computational refinements, including desalination and tautomer generation, to ensure high-quality 3D structures. Using the Develop Pharmacophore Model module, a hypothesis match threshold of 75% was set, with an optimal configuration of five hypothesis features. Compounds were ranked based on the “Phase Hypo Score,” fifty conformers per compound were generated and minimized [14]. Parameters like enrichment factor (EF) and receiver operating characteristic (ROC) curves, BEDROC, and RIE were applied to evaluate model performance [15, 16]. Decoys were generated with the DUDE tool and converted into 3D structures using Open Babel v2.4.1 [17, 18].

Molecular docking

The BCL-2 protein’s X-ray crystallographic structure, with a resolution of 1.62 Å, was used and is available from the RCSB Protein Data Bank (PDB ID: 6O0K). Structural assessment of the protein was conducted using the PDBsum web server (https://www.ebi.ac.uk/thornton-srv/databases/pdbsum), where the Ramachandran plot provided by PDBsum confirmed the validity of the BCL-2 structure [19]. Protein preparation was conducted using Maestro, with the PROPKA module refining hydrogen bonding, followed by structural minimization until heavy atoms achieved a convergence at an RMSD of 0.3Å using the OPLS4 force field, excluding water molecules beyond 5Å from the ligands [20]. The receptor grid focused on the hydrophobic region and active site where Venetoclax interacts, with coordinates defined as X=-15.23, Y = 2.24, and Z=-10.27, accommodating a maximum ligand size of 18 Å. Ligands for docking were restricted to fewer than 100 rotatable bonds and fewer than 500 atoms, with the Van der Waals radius scaling factor adjusted to 0.80 and a partial charge cutoff of 0.15. Nitrogen inversion and ring conformations were explored, ensuring the flexibility of the ligand was preserved, and torsional sampling for functional groups was enhanced. The docking framework optimized intramolecular hydrogen bonding and the planar structure of conjugated pi systems. Extra precision (XP) docking was carried out using the Glide, Schrödinger, LLC, New York, NY, 2024, with the docking workflow involving multiple filtering stages, starting with high-throughput virtual screening (HTVS) of millions of compounds, followed by standard precision (SP) mode for tens to hundreds of thousands of ligands, and concluding with XP mode to eliminate false positives, resulting in a reduction from 59,644 compounds in HTVS to 5,959 in SP and approximately 602 in XP mode.

Binding free energy of the complexes

Energy parameters from Molecular Mechanics Generalized Born Surface Area (MM-GBSA) simulations were calculated using the Prime, Schrödinger, LLC, New York, NY, 2024, to estimate stabilization energies resulting from ligand interactions with the BCL-2 receptor. The analysis employed the VGSB solvation model and OPLS4 force field [21]. Ligands ranked in the top 10% from XP docking were subjected to additional MM-GBSA calculations.

Molecular stability and reactivity insights

Density Functional Theory (DFT) analysis used Gaussian software to examine selected molecules’ electronic properties and chemical reactivity. Geometric optimization was conducted using the B3LYP functional and the 6-31G(d, p) basis set, followed by calculations of key quantum descriptors encompassing the energies of the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO). The energy difference between HOMO and LUMO (ΔE), calculated as ΔE = E_LUMO - E_HOMO, served as a critical parameter for evaluating molecular reactivity and stability; a smaller ΔE value suggests higher reactivity, whereas a larger gap indicates increased stability. The dipole moment and total energy were also assessed to determine molecular polarity and stability.

Further quantum descriptors derived from HOMO-LUMO energies provided insight into molecular behavior and interactions with biological targets. These included global hardness (η), calculated as η = (E_LUMO - E_HOMO)/2, global softness (σ = 1/η), electronegativity (χ = -(E_LUMO + E_HOMO)/2), chemical potential (µ = (E_HOMO + E_LUMO)/2), and electrophilicity index (ω = µ²/2η). Based on the HOMO-LUMO gap, Frontier Molecular Orbital (FMO) analysis highlighted molecular reactivity, identifying compounds with lower ΔE values as more reactive and thus exhibiting greater biological potential [22].

Predicted activity and chemical insights

The Quantitative Structure-Activity Relationship (QSAR) model was developed using the AutoQSAR module in Schrödinger to predict the IC₅₀ values of selected compounds. This model construction was grounded in a dataset of the top 100 BCL-2 inhibitors obtained from the BindingDB database, prioritized based on their IC₅₀ values. After preparing the ligands, 367 compounds were generated for the QSAR analysis. These compounds were divided into two subsets: 75% of the dataset (275 compounds) was allocated for training, while the remaining 25% (92 compounds) was reserved for testing. To optimize model accuracy, parameters included a maximum allowable correlation coefficient of 0.80 among independent variables, incorporating binary fingerprints and numerical descriptors. The resulting model demonstrated the best predictive performance and was subsequently employed to estimate IC₅₀ values for 61 compounds identified through MMGBSA screening [13].

Pharmacokinetics and toxicity assessment

The pharmacokinetic profiles of the compounds were evaluated using several tools: Schrödinger’s QikProp, Schrödinger, LLC, New York, NY, 2024, SwissADME, ProTox-3.0, ADMETlab 3.0, and pkCSM to assess the drug-like characteristics and determining the toxicity profiles of small molecules in drug development [23–25].

Molecular dynamics simulations

The top two selected complexes were modeled using CHARMM-GUI, employing the CHARMM36 force field for parameterization. Solvation was carried out with the rigid 3-site TIP3P water model, and the system was balanced by incorporating Na⁺ and Cl⁻ ions. The molecular dynamics (MD) protocol was initiated with a 50,000-step energy minimization for system relaxation, followed by equilibration under both constant volume (NVT) and constant pressure (NPT) conditions, regulated by the Nose-Hoover thermostat and Parrinello–Rahman barostat [26].

The production phase was subsequently carried out at 310.15 K and 1 atm for 100 ns to assess system stability in an aqueous medium. Trajectory analysis was performed using GROMACS (gmx) tools, incorporating Principal Component Analysis (PCA) and Free Energy Landscape (FEL) assessments. These evaluations utilized the covariance matrix, offering insights into the protein’s structural integrity, folding behavior, and functional characteristics [27, 28].

Results

BCL-2 protein and structural assessment

BCL-2 serves as a crucial regulator of apoptosis, characterized by four conserved BCL-2 homology (BH) domains: BH1, BH2, BH3, and BH4 (Fig. 2). The anti-apoptotic activity of BCL-2 is primarily mediated by these BH domains, enabling interactions with both pro- and anti-apoptotic members of the BCL-2 family. The N-terminal BH4 domain is essential for BCL-2’s anti-apoptotic function, as it prevents the activity of pro-apoptotic proteins. The BH1 and BH2 domains create a hydrophobic groove, enhancing structural stability and enabling binding with pro-apoptotic factors. The BH3 domain is central to apoptosis regulation, serving as the active site where BCL-2 binds and sequesters pro-apoptotic proteins like BAX and BAK. This interaction prevents mitochondrial outer membrane permeabilization, blocking downstream apoptotic signaling pathways.

Fig. 2.

Structural depiction of the BCL-2 protein with the active site

Utilizing the PDBsum web server for analysis, 9 α-helices, 3 β-hairpins, and 3 β-turns were identified, along with 17 interactions between helices. The Ramachandran plot assessed BCL-2, revealing that 94.4% of the amino acid residues fell within favored regions, and 5.6% were classified into other permissible regions. In contrast, no residues were in generous or disallowed regions, yielding an overall G-Factor of 0.23 (Supporting Information Figure S1).

The pharmacophore model and validation

The generated Pharmacophore models were systematically evaluated based on their PhaseHypo Score (Supporting Information Table S1). The model designated AHHRR_1 obtained the highest PhaseHypo Score of 1.424, incorporating one hydrogen bond acceptor (A), two aromatic rings (R), and two hydrophobic (H) features. The spatial relationships and angles among these pharmacophore components are illustrated in Fig. 3A and B, respectively. This model has a survival score of 7.069, with a site score of 0.808, a vector score of 0.919, a volume score of 0.652, a selectivity score of 2.134, and a BEDROC score of 1.000.

Fig. 3.

(A) The distances between the pharmacophore features, (B) The angles between the pharmacophore features of the most effective pharmacophore model, AHHRR_1 and (C) The ROC curve demonstrates the validation of the pharmacophore model

The AHHRR_1 model was tested against 100 inactive compounds generated using the DUDE Decoys tool, effectively reducing the decoy set to just 8 structures, achieving an efficiency rate of 92%. The evaluation of the AHHRR_1 hypothesis showed an enrichment factor (EF) of 11.42% in the top 1% of the decoy dataset, indicating that the pharmacophore model is 11.42 times more efficient at identifying true positives. The receiver operating characteristic (ROC) curve area, relative information efficiency (RIE), and area under the average concentration curve (AUAC) were calculated as 1.00, 9.42, and 0.96, respectively (Fig. 3C), highlighting its statistical significance in identifying active compounds. The model’s relevance was further validated by the BEDROC calculation, which assesses the early enrichment of active compounds, with values of 0.999, 0.999, and 1.000 for tuning parameters of α = 8.0, α = 20.0, and α = 160.9, respectively.

Binding affinity of the complexes

The docking assessments for the top 500 compounds yielded scores between − 9.20 and − 5.688 kcal/mol due to the XP docking procedure. A subsequent analysis using MM-GBSA binding energy highlighted two leading compounds: CNP0237679, which presented an MM-GBSA ΔGbind of -75.06 kcal/mol and a docking score of -7.993 kcal/mol, and CNP0420384, which displayed an MM-GBSA ΔGbind of -71.15 kcal/mol with a docking score of -8.320 kcal/mol. These compounds were selected for further examination. In addition, the co-crystallized FDA-approved BCL-2 inhibitor Venetoclax was also docked for reference, yielding a docking score of − 6.793 kcal/mol and an MM-GBSA ΔG_bind of − 62.19 kcal/mol, both of which were lower than those obtained for the two screened compounds. These findings indicate that the identified candidates possess stronger predicted binding affinities than the reference inhibitor and were therefore selected for further examination.

The interactions at the molecular level between the BCL-2 protein and the top five ligands with the strongest binding affinities are thoroughly outlined in Table 1, along with the key interacting residues and their corresponding distances. The interactions involving the top two ligands, CNP0161565 and CNP0405001, are depicted in Fig. 4. The findings indicated a favorable binding conformation characterized by diverse interactions. Additionally, the interaction profiles of the remaining three ligands and a control drug Venetoclax within the BCL-2 binding site (Supporting Information Figure S2 and S3).

Table 1.

Best two compounds with COCONUT ID, molecular formula, XP Docking Score, MM-GBSA ΔGbind Energy, principal interacting residues, their interaction types, and interaction distances

COCONUT ID	Docking score (kcal/mol)	ΔGbind (kcal/mol)	Atoms and residues in H-Bond interactions and distance (Å)	Residues in hydrophobic interactions
Venetoclax	-6.793	-62.19	Venetoclax: H –:Asn142:OD1 (1.987), Venetoclax: H –:Tyr202:OH (1.967)
CNP0237679	-7.993	-75.06	Asn143:HD21 - CNP0237679:O1 (1.947), CNP0237679:H7 – Asp111 OD2 (1.973), Arg146:HE - CNP0237679:O6 (2.323), Asn143:HD21 - CNP0237679:O2 (2.630), and Arg146:HH12 - CNP0237679:O6 (2.976)	Gly145:HA1 – CNP0237679 (π- σ bond)
CNP0420384	-8.320	-71.15	CNP0420384:H25 – Glu136:O (1.973), Arg146:HE - CNP0420384:O6 (2.323), CNP0420384:H7 – Asp111 OD2 (1.973), CNP0420384:H7 – Asp111 OD2 (1.973), CNP0420384:H7 – Asp111 OD2 (1.973), CNP0420384:H7 – Asp111 OD2 (1.973), and CNP0420384:H7 – Asp111 OD2 (1.973)
CNP0385544	-8.254	-71.08	CNP0385544:H26 – Asp103:OD2 (1.970), CNP0385544:H41 – Asp103:OD2 (2.406), Arg107:HD2 - CNP0385544:O4 (2.494), CNP0385544:H39 – Asp103:OD2 (2.603), Phe104:HA - CNP0385544:O4 (2.494), and Arg146:HH11 - CNP0385544:O1 (2.826)
CNP0339788	-7.937	-68.81	Tyr108:HH - CNP0339788:O1 (1.771), Asn143:HD22 - CNP0339788:O5 (1.887), Arg146:HH11- CNP0339788:O4 (2.624), Ghy145:HA2 - CNP0339788:F2 (2.727), CNP0339788:H10 – Asp111:OD2 (2.869), and CNP0339788:H1 – Asp103:OD2 (2.916)
CNP0155847	-7.813	-67.69	CNP0155847:H26 – Asp103:OD2 (1.936), and CNP0155847:H5 – Asp103:OD2 (1.980)

Fig. 4.

A. Three-dimensional representation of the protein-ligand complexes. Corresponding two-dimensional interactions highlighting types of contacts. B. (BCL-2-CNP0237679) and C. (BCL-2-CNP0420384)

The binding energies and interacting residues of Venetoclax and the co-crystallized ligand were employed as reference standards. Comparative analysis revealed that both screened ligands exhibited significantly stronger binding affinities than Venetoclax as well as the native co-crystallized ligand of the 6O6K structure. Furthermore, redocking of the co-crystallized ligand produced a RMSD of 0.8204 Å between the docked and experimental poses, thereby confirming the reliability and accuracy of the docking protocol in reproducing the native binding conformation.

Geometry optimization and quantum chemical calculations

The geometry optimization of the leading five compounds was executed using Density Functional Theory (DFT), revealing stable molecular configurations without imaginary frequencies, indicating their presence in the most favorable energy states. Following optimization, an analysis of Frontier Molecular Orbitals (FMO) assessed the electronic characteristics, focusing on the energies of the HOMO and the LUMO. CNP0237679 exhibited a HOMO value of -0.19883 kcal/mol and a LUMO value of -0.05578 kcal/mol, resulting in a band gap (ΔE) of 0.14305 kcal/mol. CNP0420384 showed a HOMO of -0.22082 kcal/mol and a LUMO of -0.04473 kcal/mol, yielding a band gap of 0.17609 kcal/mol (Fig. 5). In contrast, the other compounds, CNP0385544, CNP0339788, and CNP0155847, displayed higher band gaps of 0.19991 kcal/mol, 0.195 kcal/mol, and 0.21518 kcal/mol, respectively (Figure S4).

Fig. 5.

Molecular orbital (MO) energy levels and electron densities for the top two molecules - CNP0237679 and CNP0420384

The quantum chemical descriptors derived from the FMO analysis revealed that the global hardness (η) for CNP0237679 was 0.071525 kcal/mol, while CNP0420384 exhibited a hardness of 0.088045 kcal/mol. The global softness (σ) values were calculated at 13.980 kcal/mol⁻¹ for CNP0237679 and 11.358 kcal/mol⁻¹ for CNP0420384. The electronegativity (χ) was slightly elevated for CNP0420384 at 0.132775 kcal/mol compared to CNP0237679’s 0.127305 kcal/mol. The electrophilicity index (ω) was recorded as 0.1133 kcal/mol for CNP0237679 and 0.1001 kcal/mol for CNP0420384.

QSAR model and predicted activity

The ten leading QSAR models were selected, as detailed in Table 2. The most accurate model identified was Kernel Partial Least Squares (KPLS)_Dendritic_1, which demonstrated an R² value of 0.8997, standard deviation (SD) of 0.0022, root mean square error (RMSE) of 0.0021, and a Q² value of 0.8977. This model was employed to estimate the IC₅₀ values of the compounds that were screened (Fig. 6). The IC₅₀ values for the 61 compounds evaluated in this investigation ranged between 0.0287 nM and 0.0447 nM, with the average, median, and mode being 0.0389 nM, 0.0396 nM, and 0.0447 nM, respectively. Notably, the two compounds with the highest rankings, CNP0237679 and CNP0420384, showed predicted IC₅₀ values of 0.0363 nM and 0.0382 nM, which fall below the central tendency IC₅₀ values.

Table 2.

Performance metrics of 10 QSAR models

Model	Score	SD	R 2	RMSE	Q2	Q2 MW (Null Hypothesis)
kpls_dendritic_1	0.8997	0.0022	0.8997	0.0021	0.8977	-0.0108
kpls_dendritic_2	0.8942	0.0022	0.8941	0.0022	0.8925	0.0268
kpls_dendritic_30	0.8932	0.0022	0.8917	0.0020	0.9077	0.0255
kpls_linear_30	0.8866	0.0022	0.8972	0.0022	0.8907	0.0255
kpls_dendritic_40	0.8773	0.0022	0.8923	0.0023	0.8860	0.0148
kpls_linear_40	0.8560	0.0022	0.8998	0.0024	0.8781	0.0148
kpls_linear_2	0.8543	0.0022	0.8924	0.0024	0.8690	0.0268
kpls_dendritic_34	0.8103	0.0022	0.8940	0.0026	0.8460	0.0272
kpls_radial_13	0.8089	0.0022	0.8914	0.0026	0.8417	0.0348
kpls_radial_35	0.8028	0.0023	0.8823	0.0027	0.8422	0.0026

Fig. 6.

The QSAR model scatter plot of the best model kpls_dendritic_1

Pharmacokinetic and toxicity profiling

CNP0237679 and CNP0420384 passed safety assessments, were considered potential inhibitors, and were utilized for further assessments (Fig. 7). Supporting Information Tables S2, S3, and S4 offer comprehensive pharmacokinetic and toxicity assessments of the compounds.

Fig. 7.

The Bioavailability Radar Plot illustrates the favourable drug-likeness of the two compounds. The blue region indicates the ideal range for each property, while the yellow dots represent the properties of the compounds. A: CNP0237679 and B: CNP0420384

Trajectory assessment of the complexes

MD simulation analysis investigated the dynamic behavior of the complexes and evaluated the stability of protein–ligand interactions. The RMSD analysis (Fig. 8A) revealed that ligand binding stabilized the backbone conformation of BCL-2, with the complexes achieving convergence within the first 100 ns and maintaining consistent structures thereafter. RMSF profiles (Fig. 8B) showed localized fluctuations around the binding-site residues, while the radius of gyration (Fig. 8C) indicated that the complexes remained compact relative to the apo form. Solvent-accessible surface area analysis (Fig. 8D) reflected conformational shifts in the protein backbone upon ligand engagement. The hydrogen bond profile (Fig. 9A) demonstrated persistent interactions across the trajectory, and non-covalent contacts, including hydrophobic stabilization, remained consistent at 25 ns intervals and interaction profile of the complexes throughout the trajectory (Fig. 10). Average values for all simulation parameters are summarized in Table 3.

Fig. 8.

The BCL-2 backbone and ligand complex trajectory assessment during a 100 ns trajectory. (A) RMSD, (B) RMSF, (C) Rg, and (D) SASA

Fig. 9.

(A) The H-bond contacts of the leads with the BCL-2 backbone during 100 ns simulation, (B) 2D projections of the complexes, and (C) The Gibbs FELs of the complexes using two eigenvectors

Fig. 10.

Trajectory snapshots of the complexes at 0, 25, 50, 75, and 100 ns. (A) CNP0237679 and (B) CNP0420384. The interaction profile of the complexes involved in wide range of non-covalent interactions throughout the 100 ns trajectory. (C) CNP0237679 and (D) CNP0420384

Table 3.

The parameters’ average values for assessing the 100 Ns trajectory of the BCL-2 backbone

Complex	RMSD (nm)	RMSF (nm)	Rg (nm)	SASA (nm2)
Apo-BCL-2	0.268	0.085	2.015	173.31
CNP0237679	0.167	0.085	1.430	180.220
CNP0420384	0.155	0.071	1.429	180.158

The PCA of the simulation trajectories illustrated the collective motion of the protein backbone in contrast to the unbound complex. The PCA analysis outcomes for the complexes are shown in Fig. 9B. The findings revealed that the binding of ligands significantly expanded the collective motion. The FEL was computed to describe the conformational changes in the ligand-bound complexes over the 100 ns trajectory. Stable conformation is represented by deep dark blue wells transitioning to red. Figure 9C displays the FEL plotted between the PC1 and PC2 coordinates for the complexes, with Gibbs free energy ranging from 0 to 13.6 kJ/mol, 13.8 kJ/mol, 11 kJ/mol, and 13.4 kJ/mol, respectively. A key observation was the total alteration of the primary free energy well within the global minimum energy region upon ligand binding during the 100 ns trajectory. Additionally, the energy landscape of the complex revealed multiple distinct minima, each corresponding to 2–4 metastable structural states of the ligand-bound complexes relative to the APO, separated by relatively modest energy barriers.

Discussion

This study examines the molecular interactions of CNP0237679 and CNP0420384 with the BCL-2 protein, an essential modulator of apoptosis and an established therapeutic target in various cancers, particularly hematologic malignancies. The significance of BCL-2 inhibition is illustrated by the success of Venetoclax, an FDA-approved BCL-2 inhibitor known for its efficacy in treating CLL and AML through the selective induction of apoptosis in BCL-2-dependent cells.

Pharmacophore modeling was used to identify critical structural characteristics like spatial orientation, hydrogen bond donors and acceptor sites, and electronic properties necessary for BCL-2 inhibition [29]. The ligand-based AHHRR model incorporated these elements, enabling the identification of potential inhibitors across a diverse chemical space. Rigorous validation procedures confirmed the model’s predictive accuracy for known compounds and new chemical entities, an essential step preceding virtual screening [30].

The QSAR model KPLS_Dendritic_1 demonstrated strong predictive power, reflected in its high R² and Q² values for IC₅₀ estimations, though experimental validation remains crucial to confirm these computational outcomes [31]. This dendritic model effectively captures the detailed architecture of molecular structures through its branched framework, facilitating the representation of local and global features. Consequently, this improves the extraction of molecular descriptors and topological indices, enhancing insights into the structural determinants of biological activity and reinforcing the model’s utility for drug discovery [32].

CNP0237679, a dihydroflavonoid, and CNP0420384 exhibited higher binding energies than Venetoclax, indicating their potential as potent novel BCL-2 inhibitors. While molecular docking has traditionally been employed to predict receptor-ligand interactions, the MM-GBSA method provides superior accuracy by incorporating binding free energy calculations, aligning predictions more closely with empirical data [33, 34]. This method’s accuracy stems from its ability to account for dynamic molecular interactions, such as solvation effects, entropic contributions, and the flexibility of both ligands and target proteins. In alignment with our study, these compounds exhibited strong binding affinities to the BH3 binding groove, as observed in docking and MM-GBSA analyses.

Flavonoids, such as apigenin, butein, and quercetin, are known to upregulate BAX expression in malignant cells while reducing BCL-2 and Bcl-XL synthesis. Key interacting residues, including ASN143, ARG146, TYR108, PHE104, ASP111, and GLU136, were identified in this study, supporting previous findings that these residues are critical for stabilizing inhibitor binding within the BCL-2 pocket [35–37]. CNP0237679 and CNP0420384 demonstrated interactions with hydrophobic residues within the BH3 binding groove, an interaction pattern consistent with effective BCL-2 inhibitors [38].

DFT analysis further supports binding affinity predictions from docking and MM-GBSA assessments, confirming CNP0237679 and CNP0420384 as leading candidates with high binding affinities. The DFT analysis revealed that these compounds exhibit favorable electronic properties, with CNP0237679 showing a smaller band gap, indicating higher reactivity and potential interaction strength, while CNP0420384 demonstrated moderate reactivity due to a slightly larger band gap. The increased softness and reactivity of CNP0237679, indicated by lower hardness and greater electron-donating/accepting potential, further support its efficacy as a promising biological candidate [39]. CNP0420384, with its marginally higher electronegativity, shows stronger electron-attracting capability. Both compounds exhibit moderate electrophilicity, with CNP0237679 having a slight advantage. The dipole moment of CNP0237679 (7.76 D) was higher than that of CNP0420384 (6.15 D), which may contribute to stronger hydrogen-bond interactions within the active site. The predicted acidic and basic pKa values suggest that both molecules remain partially ionized at physiological pH, potentially enhancing solubility and binding adaptability.

The pharmacokinetic and pharmacological profiles of CNP0237679 and CNP0420384 show promise, with favorable ADME characteristics. Both compounds exhibit high levels of human intestinal and oral absorption, suggesting good bioavailability. They are both substrates for P-glycoprotein, though their profiles differ as inhibitors, which could influence drug efflux and systemic exposure [40]. CNP0420384 has extensive tissue distribution, potentially expanding its therapeutic applications, while CNP0237679, with higher blood-brain barrier (BBB) permeability, could hold additional promise in CNS-targeted therapies. Both compounds bind extensively to plasma proteins, with distribution volumes indicating potential for broad tissue interactions. CNP0420384 interacts with a broader range of CYP enzymes as a substrate and inhibitor, suggesting a complex metabolic profile, while CNP0237679 shows more selective metabolic properties [41]. Both compounds display moderate clearance rates, suggesting balanced elimination and the potential for maintaining stable therapeutic plasma concentrations [42]. Additionally, CNP0237679 presents a more favorable toxicity profile, with low chronic and respiratory toxicity, compared to CNP0420384, while non-AMES toxic and low acute oral toxicity indicate a slightly higher chronic risk [43].

The binding of these compounds to the BCL-2 active site induces conformational changes within the protein. The extent of atomic fluctuations around the protein backbone’s mean positions, assessed through molecular dynamics simulation, suggests that ligand interactions are crucial for maintaining the stability of the protein structure [44]. Analysis of the Rg and SASA supports these findings by highlighting the protein’s compact conformation and the stability of the active site during the entire simulation [45]. The formation of stable hydrogen bonds throughout a simulation duration of 100 ns was critical to the observed stabilization of the protein-ligand complexes [46].

The analysis of hydrogen bonds, a key factor in complex stability, indicated that these interactions varied throughout the simulation [46]. The compounds primarily engaged with target proteins through hydrogen bonding and hydrophobic interactions, producing significant conformational changes influencing the protein’s structure [47]. While CNP0237679 displayed transient hydrogen bond interactions, CNP0420384 formed more stable and consistent bonds, particularly with residues strongly associated with BCL-2 inhibition, such as Asp111 and Arg146. This suggests that both complexes exhibit high stability [48]. PCA indicated increased collective motion in the ligand-bound BCL-2 structure compared to the APO, suggesting enhanced flexibility and larger conformational shifts upon ligand binding [49].

The stability observed in the MD simulations can be attributed to the engagement of our screened compounds with the hydrophobic interactions of the BH3-binding groove of BCL-2, comprising Phe104, Phe112, Met115, Val133, Leu137, and Ala149. Among these, Phe104 in the P2 pocket acted as a dominant anchoring site through π–π stacking and van der Waals contacts, consistent with prior evidence that the F104L mutation reduces venetoclax affinity ~ 25-fold [50]. Phe112, located at the P2/P3 base, was also strongly engaged, providing significant binding energy contributions; its conformational remodeling upon ligand binding parallels structural studies with venetoclax, which highlight its role in selectivity [37]. Met115 contributed stabilizing van der Waals contacts within the hydrophobic core, consistent with its recurrent involvement in natural product docking studies [51]. Val133 interacted through van der Waals and π-alkyl contacts, shaping the P1 pocket to support orientation of the ligands, while Leu137 cooperated with Phe104 and Phe112 in stabilizing ligand binding, in agreement with BH3 helix and inhibitor binding studies [50]. Ala149, though less dominant, provided additional van der Waals contacts that completed the lipophilic environment and augmented overall stabilization [7]. Thus, the interactions of our compounds across this hydrophobic surface directly correlate with existing reports on BH3-mimetic inhibitors such as venetoclax, and explain the reduced flexibility, compactness, and persistence of non-covalent interactions observed throughout the trajectory.

FEL analysis showed multiple distinct energy minima representing metastable states, emphasizing the dynamic characteristics of these complexes [52]. CNP0237679 appeared to stabilize BCL-2 in fewer conformational states, while CNP0420384 allowed for greater exploration of conformational space, suggesting potential applications across various BCL-2-dependent cancers. The binding of these ligands suggests an “induced fit” mechanism, where protein conformation adapts to optimize binding [53]. SASA decreased upon ligand binding, suggesting compaction of the protein structure, while the stable Rg values confirmed no unfolding occurred. The MD simulation analysis contributes valuable insight into the dynamic stability and intermolecular interactions in protein complexes bound to ligands.

This study highlights the limitations and offers guidance for future research directions. The next essential step involves validating these computational results via laboratory experiments and animal model investigations. Examining the molecular processes underlying these interactions will provide a deeper understanding of how these compounds modulate BCL-2 and related proteins, potentially broadening their therapeutic applications. Beyond preclinical evaluations, including pharmacokinetic and pharmacodynamic analyses, further research should be pursued to advance these compounds toward clinical trial stages. Considering the potential for these findings to inform personalized medicine, understanding cancer-related genetic and molecular heterogeneity may significantly improve the development of targeted cancer therapies.

Conclusion

This study successfully identifies two natural compounds, CNP0237679 and CNP0420384, as potential BCL-2 inhibitors, offering promise in targeted leukemia treatment. Using an integrated computational strategy, these compounds exhibited high binding affinities, robust free binding energies, and structural stability within the BCL-2 binding pocket. The pharmacokinetic and toxicity analyses further demonstrated favorable profiles, indicating the compounds’ potential for high bioavailability and low toxicity, which are suitable for further drug development.

The findings underscore the utility of leveraging a comprehensive virtual screening approach to discover bioactive natural compounds with therapeutic potential in cancer treatment. The molecular dynamics simulations provide evidence of protein-ligand stability, reinforcing these compounds’ potential efficacy as BCL-2 inhibitors. Moreover, the developed pharmacophore model may facilitate the identification of additional natural inhibitors, expanding the library of compounds for anti-apoptotic targeting therapies.

To advance this investigation, upcoming research should prioritize in vitro and in vivo validation of CNP0237679 and CNP0420384, alongside in-depth pharmacodynamic analyses to confirm therapeutic efficacy. Given their unique binding profiles, these compounds may also have broader applicability in other BCL-2-dependent cancers, contributing to developing more selective and potent therapeutic agents. This research represents a meaningful step toward developing natural, targeted BCL-2 inhibitors, emphasizing the significance of computational approaches in expediting the initial phases of drug discovery.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

UD: Conceptualization; Methodology; Software; Data curation; Formal analysis; Investigation; Validation; Visualization; Writing-original draft; Writing-review and editing, prepared Figs. 1, 2, 3, 5, 6, and 7. AM: Software; Data curation; Formal analysis; Investigation; Validation; Visualization; Writing-original draft, prepared Figs. 4, 8, 9, and 10. MKS: Conceptualization; Methodology; Supervision; Review and Editing. JK: Conceptualization; Supervision; Review.

Funding information

Open Access funding by Manipal Academy of Higher Education, Manipal, Karnataka, India.

Data availability

This study was carried out using publicly available Natural Compounds data from the COCONUT Database at https://coconut.naturalproducts.net/download and the Crystal Structure of BCL2 from the Protein Data Bank with PDB ID: 6O0K.

Declarations

Ethics approval

Not Applicable.

Competing interests

The authors declare no competing interests.