In silico identification of bilobetin and ginsenoside as dual CK2 and ULK2 inhibitors targeting triple-negative breast cancer

Abstract

Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer characterized by the absence of hormone receptors and HER2 amplification, which limits the effectiveness of standard targeted therapies. In this study, we employed a comprehensive computational approach to identify natural compounds with therapeutic potential against TNBC by targeting Casein Kinase 2 (CK2) (PDB ID: 3H30) and ULK2 (Unc-51-like autophagy-activating kinase 2) (PDB ID: 6YID), two proteins critically implicated in tumor progression and autophagy regulation. Ten bioactive natural compounds were initially screened through molecular docking, among which bilobetin and ginsenosides demonstrated the most favorable binding affinities toward CK2 (–10.1 kcal/mol) and ULK2 (–9.4 kcal/mol), respectively, outperforming the reference drug capecitabine (–7.6 kcal/mol). Molecular dynamics (MD) simulations revealed that bilobetin–CK2 and ginsenosides–ULK2 complexes maintained stable conformations, supported by consistent RMSD and RMSF values, compact structural organization (Rg), and favorable residue correlations, indicating robust protein–ligand interactions. Free energy calculations using MM/GBSA further validated these findings, with bilobetin (–61.72 kcal/mol) and ginsenosides (–47.91 kcal/mol) showing substantially stronger binding free energies compared with capecitabine. Importantly, ADMET profiling indicated favorable pharmacokinetic properties, including high intestinal absorption, acceptable solubility, and low predicted toxicity, reinforcing their potential for drug development. Collectively, these results highlight bilobetin and ginsenosides as promising natural inhibitors of CK2 and ULK2, respectively, and provide a strong foundation for their advancement as novel therapeutic candidates against TNBC. Future in vitro and in vivo validation will be critical to confirm their efficacy and translational potential.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-33464-y.

Introduction

One of the leading causes of death globally is cancer, a disease marked by uncontrolled cell growth and spread. It may start practically anywhere in the human body and spread to any part of the body, consisting of trillions of cells. About 10 million fatalities were due to cancer in 2020, accounting for one in every six deaths^1–3. Different cancers have varying prevalence rates in men and women. While both genders are more susceptible to stomach, liver, prostate, and lung cancers, women are particularly prone to breast cancer, which accounts for one in eight cancer cases worldwide. Around 2.3 million documented incidences with 685,000 fatalities were attributed to breast cancer in 2020. Forecasts for 2040 show that this figure will rise by almost three million new cases and one million deaths annually⁴. Various factors are responsible for that, such as increased exposure to environmental pollutants, lifestyle factors like poor diet, lack of physical activity, and alcohol consumption, as well as advancements in early detection leading to more diagnoses^5,6.

TNBC is one of the most prevalent forms of breast cancer, which itself represents a collection of diseases rather than a single illness. It is characterized by the lack of progesterone, estrogen, and the human epidermal growth factor receptor 2 (HER2), and it accounts for over 15% of all cases of breast cancer^7,8. CK2 overexpression has been associated with increased tumor growth, chemoresistance, and poor prognosis in breast cancer. Moreover, ULK2, a key regulator of autophagy, plays an essential role in cell survival under stress, and its dysregulation has been linked to therapy resistance and cancer progression, particularly in TNBC. TNBC is often associated with overexpression of casein kinase 2 (CK2), which promotes chemoresistance and tumor-initiating cell survival. It has been demonstrated that CK2 inhibitors, such as multi-kinase drugs like 108,600, decrease clonogenic survival, slow the formation of TNBC stem cell populations, and make resistant cells more sensitive to conventional chemotherapy treatments⁹. In the meantime, ULK2, which is frequently taken into account in conjunction with ULK1, is becoming a more viable target. TNBC cells’ autophagic flow is disrupted by dual ULK1/2 inhibition (for instance, by SBP-7455), which lowers viability¹⁰. Furthermore, ULK1/2 play roles that are independent of autophagy. According to recent research, ULK1/2 control the migration of breast cancer cells by mechano-transduction by phosphorylating paxillin (PXN), which implies that ULK2 may influence both survival and the likelihood of metastasis in TNBC¹¹.

TNBC is more prevalent in women under 40 years of age than hormone receptor-positive breast cancer, and research indicates that women in this age range are twice as likely to develop TNBC compared with women over 50 years. Apart from that, the prevalence of TNBC is relatively higher among black women compared to white women, but the etiology of TNBC remains elusive, researchers believe that mutations in the BRCA1 (Breast Cancer 1) gene, a tumor suppressor gene responsible for DNA repair and maintaining genomic stability, along with factors such as smoking and body mass index, may contribute to its development^12,13.

Certain FDA-approved drugs, such as pembrolizumab (Keytruda), have shown promise for treating TNBC. Pembrolizumab, an immune checkpoint inhibitor targeting PD-1, was approved in combination with chemotherapy for advanced TNBC patients whose tumors exhibit a PD-L1 combined positive score (CPS) of at least 10. Despite its approval, its efficacy is limited to a subset of patients. The KEYNOTE-355 clinical trial demonstrated that only patients with a PD-L1 CPS of 10 or higher experienced a significant improvement in overall survival, whereas patients with lower PD-L1 expression derived minimal benefits. Furthermore, pembrolizumab therapy can lead to immune-related adverse effects, which may restrict its use in certain patients¹⁴,¹⁵.

Other FDA-approved therapies include PARP inhibitors such as Olaparib, which are indicated for patients with BRCA1/2 mutations. However, these drugs are effective only in a subset of TNBC patients with specific genetic alterations¹⁴. Chemotherapy, either alone or in combination with immunotherapy, remains a therapeutic option for advanced TNBC. However, its efficacy is limited, and it frequently causes significant adverse effects owing to the loss of healthy cells¹⁶. The chemotherapeutic agents such as anthracyclines, taxanes, capecitabine, and gemcitabine are often available as single or combined to treat TNBC, but their efficacy is limited¹⁷. As a consequence, chemotherapeutic drugs offer modest benefits but substantial toxicity; therefore, there is an urgent need for more efficient, tailored therapies capable of targeting TNBC with minimal harm to normal cells.

Naturally occurring phytoconstituents have been extensively investigated for their therapeutic potential across multiple cancer types¹⁸. Representative compounds, including lycopene, fucoxanthin, cucurbitacin, hesperidin, ginsenosides, lupenone, cypripedin, sesquiterpenes, carotenoids, and bilobetin (Supplementary Figure S1), have demonstrated anticancer activities and may serve as candidates for TNBC therapy with reduced toxicity and broader accessibility. The anticancer properties of ginsenosides, a varied class of triterpenoid saponins derived from Panax species, have been extensively researched. These effects include the induction of apoptosis and the prevention of tumour growth¹⁹. Although ginsenosides have several subtypes (such as Rg1, Rg2, and Rg3), we used a representative ginsenoside entry (PubChem CID: 3,086,007) as a scaffold for computational screening in this study. More research is planned to examine the activities of certain ginsenoside subtypes. Among the molecular targets of interest, CK2 (Casein kinase II) and ULK2 (Unc-51-like autophagy-activating kinase 2) are critical regulators of cell survival, proliferation, and autophagy, and their dysregulation has been strongly implicated in TNBC progression and therapy resistance^20,21. Given their pivotal roles in TNBC signaling, these proteins were selected for in silico screening of natural compounds. In breast cancer, CK2 overexpression has been linked to poor prognosis, chemoresistance, and accelerated tumour development. Additionally, ULK2, a crucial autophagy regulator, is vital for cell survival in stressful situations, and its dysregulation has been connected to treatment resistance and the advancement of cancer, especially TNBC.

An essential part of chemotherapy treatments for TNBC is capecitabine, an oral prodrug of 5-fluorouracil. Its use has been shown to considerably enhance survival outcomes for individuals with TNBC in a number of randomised trials and meta-analyses. Patients receiving capecitabine in combination with standard chemotherapy showed improvements in disease-free survival (DFS) and overall survival (OS) of 23% and 27%, respectively, according to a recent meta-analysis of 11 RCTs involving 5,175 patients²². Because of its therapeutic importance in TNBC, capecitabine is a suitable reference drug for evaluating the binding affinities of natural compounds in this study, even if it is not a direct inhibitor of CK2 or ULK2. In order to identify and assess bioactive natural compounds with potential inhibitory activity against CK2 and ULK2, we employed molecular docking, ADMET profiling, and molecular dynamics simulations in this study, utilising capecitabine as a comparator.

Materials and methods

Screening and selection of phytocompounds

Fifty phytocompounds were shortlisted through a structured literature survey of natural products reported to exhibit anticancer activity, particularly in breast cancer and TNBC contexts. Compounds were prioritized based on structural diversity (e.g., terpenoids, flavonoids, carotenoids, saponins) and prior mechanistic evidence such as pro-apoptotic, anti-proliferative, and anti-metastatic effects. Representative classes included ginsenosides from Panax spp., cucurbitacins, sesquiterpenes (e.g., β-caryophyllene, humulene), carotenoids (e.g., lycopene, fucoxanthin, cypripedin), citrus flavonoids (e.g., hesperidin), and biflavonoids from Ginkgo biloba (e.g., bilobetin)^23,24. All of these classes are supported by recent reviews highlighting their anticancer potential and, in several cases, direct activity against breast cancer^25–27.

Ginsenosides’ PubChem entry (CID: 3,086,007) was used as an example scaffold for this family of compounds. Future research will investigate the unique biological actions of individual ginsenosides, such as Rg1, Rg2, and Rg3. To ensure the shortlisted candidates had a plausible spectrum of activity, we further profiled each molecule using the PASS/Way2Drug platform, extracting Pa/PI values for anticancer- and kinase-related endpoints as an initial filter. PASS has been extensively validated for predicting biologically relevant activities from molecular structures, and the Way2Drug interface enables efficient high-throughput triage of candidate compounds²⁸.

Pass prediction

The PASS prediction was conducted using the online tool (http://www.way2drug.com/passonline/), which evaluates organic molecules’ biochemical and physiological characteristics based on their structural formula. The tool provides predictions for over 4,000 categories of biochemical and physiological activities, with an accuracy exceeding 95%. In this study, 50 natural compounds were initially screened for their biological activities, and 10 were selected based on their potential antineoplastic properties for further analysis (Supplementary Table S1). Pa (probability of being active) and Pi (probability of being passive) figures indicate the forecast; Pa is greater than Pi, and Pa + Pi always equals 1. Pa does not equal Pi, and the values of Pa and Pi are often at most one²⁹. Based on these findings, the human protein kinase CK2 (PDB ID:3H30) and the crystal structure of ULK2 in complex (PDB ID:6YID) were chosen as target receptors.

A number of software tools were employed in the computational workflow. For ligand synthesis and virtual screening, PyRx version 0.8 (https://pyrx.sourceforge.io)³⁰ was used, while AutoDock Vina version 1.1.2³¹ was applied for molecular docking. Open Babel version 2.4.1 (http://openbabel.org) was utilized for structure conversions and optimization. Swiss-PDB Viewer version 4.1.0 (https://spdbv.vital-it.ch)³² was employed for protein energy minimisation, and PyMOL version 1.3 (https://pymol.org) was used for protein visualisation and water molecule removal. GROMACS version 2021.4 was utilised to perform molecular dynamics simulations, and ADMETlab 2.0 (https://admetmesh.scbdd.com/)³³ was used to estimate ADMET characteristics.

Preparing ligands and molecular optimization

This study selected ten phytocompounds based on their reported anticancer potential (Supplementary Figure S1). From the initial 50, ten phytocompounds were advanced using a predefined triage funnel: (i) PASS filters favoring anticancer-relevant terms (e.g., “antineoplastic,” “kinase inhibitor”) with Pa ≥ 0.5 and Pa–Pi ≥ 0.3; (ii) drug-likeness heuristics with ≤ 1 Lipinski RO5 violation and acceptable lipophilicity; and (iii) oral-exposure proxies (Veber criteria: rotatable bonds ≤ 10 and TPSA ≤ 140 Ų) to avoid extreme chemotypes prior to docking/MD^23,28,34,35.

Ligand 3D structures were retrieved from open repositories (e.g., PubChem) and energy-minimized to physiologically relevant, low-strain conformations before docking standardizing protonation/tautomer states and relieving steric clashes to improve pose reliability in subsequent simulations³⁴.

For ADME/T considerations during triage, we cross-checked computed physicochemical envelopes against well-used in silico guidelines (RO5/Veber) and documented, where available, literature flags on tolerability for each class (e.g., ginsenosides, carotenoids, sesquiterpenes), reserving prospective toxicity modeling for later stages^36,37.

All chemical structures were drawn using ChemDraw Professional and saved in Protein Data Bank (PDB) format for subsequent computational investigations, including molecular docking, molecular dynamics, and ADMET analyses. Structural optimization was performed using vibrational frequencies derived from the B3LYP functional with the DND basis set (diffuse basis set) and semi-core pseudopotential³⁸.

Molecular docking analysis

Target protein selection and preparation

The three-dimensional structures of human CK2 alpha kinase (PDB ID: 3H30) and ULK2 (PDB ID: 6YID) were retrieved from the RCSB Protein Data Bank^39,40 (Supplementary Figure S2). All water molecules, co-crystallized ligands, and heteroatoms were removed using PyMOL v1.3⁴¹. Energy minimization of the protein structures was performed using Swiss-PDB Viewer v4.1.0⁴² and the resulting files were saved in PDB format. The proteins were then prepared for docking using AutoDock Vina implemented in the PyRx v0.8 virtual screening tool³¹.

Molecular docking

For CK2 (PDB ID: 3H30) with bilobetin, the docking grid was centered at X = 23.1863, Y = 9.6133, Z =  − 0.1349 with dimensions of 58.5789 × 69.5290 × 52.6330 Å. For the CK2–Capecitabine reference complex, the same grid parameters were applied. For ULK2 (PDB ID: 6YID) with ginsenoside, the docking grid was centered at X = 0.7417, Y =  − 1.3544, Z = 25.7607 with dimensions of 52.2261 × 63.9886 × 59.6734 Å, and the same grid parameters were used for the ULK2–Capecitabine reference complex. In all cases, the exhaustiveness was set to 8 to ensure adequate sampling of ligand conformations.

Notably, the dimensions of the docking grid were purposefully selected to be bigger than the standard binding pocket size. This tactic was used to guarantee a thorough investigation of possible binding modes and to prevent the ligands—especially the bulky and flexible natural substances like ginsenosides—from being artificially constrained. The final validation stage, in which the native co-crystallized ligand was re-docked with an RMSD < 2.0 Å, verifies that the methodology successfully replicated the right binding pose, even if larger grids may expand the computational search area. Despite the large grid size, this, together with the significant convergence between our docking postures, MD simulation stability, and MM/GBSA results, gives confidence in the found binding modes.

Docking accuracy was validated by re-docking the native co-crystallized ligands into their respective binding pockets. The RMSD values between the docked and experimental poses were below 2.0 Å, indicating the robustness of the docking protocol, in line with established benchmarks⁴². The final docking poses and active site visualizations were prepared using PyMOL v1.3 and Discovery Studio Visualizer v21.1.0 (BIOVIA, Dassault Systèmes). Hydrogen bond and hydrophobic interaction analyses were performed following established approaches⁴³.

Drug-likeness and pharmacokinetics analysis

The Lipinski rule is a crucial criterion in computer-aided drug design. When chemicals interact with the lipid bilayers in the body, they adhere to the “rule of five” to determine the degree of permeability or absorption. These qualities are assessed using Lipinski’s “rule of five”; a chemical is said to fit the drug-like characteristics if it fulfills each of the five requirements listed below^44,45.

• The chemicals have a molecular weight of less than 500 g/mol.

• The computed partition coefficient between octanol and water is less than 5 (logP5).

• The number of donors for hydrogen bonds is somewhat below five.

• The maximum number of hydrogen bond acceptors is ten, especially N and O atoms.

Finding and creating novel therapeutic compounds requires careful evaluation of ADMET data prediction. Assessing the properties and in silico pharmacokinetic parameters of pharmaceutical substances under investigation is another excellent use for ADMET data. Since these are naturally occurring compounds, ADMET characteristics were evaluated to predict potential toxicity and adverse effects, enabling a rapid pre-screening process before comprehensive in vitro analysis. As a result, ADMET data have been predicted using the online program pkCSM, and these data are computed and quantified using the molecules’ structural and chemical characteristics⁴⁶.

Molecular dynamic simulations

Molecular dynamics simulations are an essential tool for assessing protein stability for the reason that they reveal in great detail the changes in structure and fluctuations that occur when the ligand–protein complex is stabilized⁴⁷. Receptor–ligand complexes derived from molecular docking approaches were used in the study to confirm stability and binding processes in a dynamic system. This was accomplished by running a 100 ns (ns) MD simulation of the two best-docked compounds: Ligand-A (bilobetin) toward the crystal structure of the human CK2 alpha kinase catalytic subunit (PDB ID: 3H30) and Ligand-B (ginsenoside) toward the crystal structure of ULK2 in complex (PDB ID: 6YID). The CHARMM-GUI internet-based visual tool was used to build the protein and ligand force fields.

All of the CHARMM-GUI, OpenMM, AMBER, GROMACS, and CHARMM/OpenMM simulations used the CHARMM36 Additive Force Field⁴⁸. Using the CHARMM36 force field and GROMACS 2020 software, the simulations ran for 100 ns in a periodic water box. The papers titled "CHARMM36 all-atom additive protein force field: validation based on comparison to NMR data" and "GROMACS: High-performance molecular simulations through multilevel parallelism from laptops to supercomputers" provided the necessary background information⁴⁹. The complexes were positioned inside a rectangular box, with a buffer space of 10 units between each cardinal direction. The steepest descent method was employed to reduce energy after introducing salt and chloride ions to guarantee charge neutrality and TIP3P molecules of water to dissolve the container⁵⁰. It took 5000 steps (10 picoseconds) for each system to attain equilibrium at 310 K. The NPT ensemble was used for 100 ns. A two-fs step of time was used to enforce the hydrogen restrictions using the Lincs technique.

The switching function was used to compute the Van-der Waals forces, which had a range of 12–14 units and a threshold of 14 units. A maximum grid spacing of 1.2 units was used to calculate long-range electrostatic interactions using particle mesh Ewald (PME). Instead of multi-time stepping, PME computations were done at each time step. The temperature was 310 K, and initial velocities were randomly determined from a Maxwellian distribution. The barostat was set to maintain a target pressure of 1 bar, regardless of any alterations in system size. The integrating time interval was maintained at 2 fs. The simulation yield was re-centered, and the trajectories were analyzed using QTGRACE and VMD software (University of Illinois at Urbana-Champaign, Urbana, IL, USA). The stability of the system was evaluated using a combination of analytical techniques, including principal component analysis (PCA), radius of gyration (Rg), root mean square deviation (RMSD), hydrogen bond analysis and root mean square fluctuation (RMSF). Furthermore, dynamic cross-correlation matrix (DCCM) studies were performed on the MD trajectories using the Bio3D (Version: 2.4–5) package in the R program^51,52.

Molecular mechanics generalized born surface area (MM-GBSA) calculations

In order to offer a more accurate, physics-based estimation of binding affinity than docking scores, calculations were conducted using molecular mechanics/generalized Born surface area (MM/GBSA). To be clear, these calculations were not performed on an ensemble of snapshots from the molecular dynamics (MD) trajectories, but rather on the energy-minimized structures of the protein–ligand complexes that were directly derived from the molecular docking investigations.

This technique, sometimes referred to as “static” or "single-trajectory" MM/GBSA, is a tried-and-true and computationally effective way to estimate free energy and optimise binding poses after docking^53–55.

By explicitly computing molecular mechanics energies in conjunction with implicit solvation and surface area factors, it provides a notable benefit over docking, even though it does not include the whole conformational sampling of the protein–ligand complex acquired by MD.

The Prime module of the Schrödinger Suite 2020–3 (Maestro, commercial edition) was used to perform the calculations. To maximise local contacts and alleviate steric conflicts, the OPLS-2005 force field was used to minimise each protein–ligand complex that was produced by docking before MM/GBSA analysis. The solvent-accessible surface area (SASA) was used to estimate non-polar contributions, and the generalised Born solvation model (VSGB 2.0) was used to account for polar solvation effects. The following formula was used to determine the binding free energy (ΔG_bind):

where G represents the total free energy of each component.

Ginsenosides with ULK2 (PDB ID: 6YID), bilobetin with CK2 (PDB ID: 3H30), and the reference medication capecitabine with both targets were assessed for relative binding free energies. Compared to docking alone, this MM/GBSA step’s main function in our approach was to give an intermediate validation utilising a more thorough energetic model. A convergent, multi-method validation of the natural compounds’ superior binding is provided by the strong correlation between their favourable MM/GBSA scores and their subsequent stability in the 100 ns MD simulations (as shown by stable RMSD, Rg, and interaction profiles), which lessens the inherent drawbacks of the single-pose approximation^54,55.

Rationale for the integrated multi-stage computational workflow

This study’s computational pipeline makes use of force fields and specialised tools at various stages, which is a conventional, tiered process in contemporary structure-based drug development^56,57. Each step was chosen to take use of its unique benefits, resulting in a complementing process where the high accuracy of later validation procedures balances the computational efficiency of initial screening.

Because of its established speed, precision, and reliable empirical scoring function for evaluating ligand affinities, Molecular Docking with AutoDock Vina was utilised for the first high-throughput screening and pose prediction⁵⁸.

The stability and dynamic interactions of the top-ranked docked complexes in a solvated, physiological environment were assessed using Molecular Dynamics (MD) with CHARMM36/TIP3P. The CHARMM36 force field offers a high-fidelity, dynamic model of protein–ligand interactions across time and has been well tested for biomolecular system simulation^59,60.

MM/GBSA with OPLS-2005/VSGB 2.0 was used to get more accurate, thermodynamic-based binding free energy estimates. A useful link between static docking scores and complete thermodynamic integration is offered by the OPLS-2005 force field and VSGB 2.0 solvation model, which are parameterised especially for this use in Schrödinger^54,55.

The purpose of this multi-tiered strategy is to get around the drawbacks of any one methodology. A solid, multifaceted validation of bilobetin and ginsenosides as prospective inhibitors is provided by the significant convergence of data, where the top compounds found by docking also showed stable binding in MD simulations and favourable free energies in MM/GBSA.

Result and discussion

Pass prediction spectra analysis

Data on the antiviral, antibacterial, antifungal, and antineoplastic efficacy of the compounds, as determined by SAR studies, were obtained using the web-based application PASS, which is used to predict the activity spectrum of molecules. With Pa scores of 0.813–0.973 for antineoplastic, 0.190–0.973 for antiviral, 0.273–0.803 for antifungal, and 0.197–0.650 for antibacterial, the acquisition pass spectrum values were examined. Here, the likelihood of antineoplastic activity or a higher Pa score is displayed. Based on these findings, the compounds were further analyzed for their potential as inhibitors against TNBC through molecular docking and ADMET evaluations to validate their suitability for targeting TNBC proteins (Supplementary Table S1).

Notably, a systematic triage procedure led to the selection of these ten compounds for additional research. To guarantee a focus on compounds with a higher potential for oral bioavailability, candidates were ranked from an initial library of fifty phytocompounds based on both their high PASS prediction scores (Pa ≥ 0.5 for antineoplastic activity) and their adherence to drug-likeness heuristics, such as Lipinski’s Rule of Five and Veber’s criteria. Bilobetin and ginsenosides were among the chosen choices, and this multi-step filtering made sure they were promising in terms of their expected biological activity and pharmacokinetic appropriateness.

Lipinski rule, pharmacokinetics, and drug likeness

The Lipinski rule establishes pharmacokinetic drug features, including ADME, based on distinct molecular characteristics. The Lipinski rule suggests the largest molecular weight is 717, and the minimum molecular weight is 246.3. The largest number of hydrogen bond acceptors is 15, and the smallest is 0. The greatest number of hydrogen bond donors is 8, and the smallest is 0, with the topological polar surface area (Ų) ranging from 0.00 to 234.29.

Out of the 10 compounds, Hesperidin violates the Lipinski rule due to a high hydrogen bond donor count (8) and hydrogen bond acceptor count (15), as well as a large topological polar surface area (234.29 Ų). The other 9 compounds (i.e., Lycopene, Fucoxanthin, Cucurbitacin, Ginsenosides, Lupenone, Cypripedin, Sesquiterpenes, Carotenoids, and Bilobetin) complied with the Lipinski rule. The violation values range from 1 to 3, and the bioavailability score is approximately 0.85, as presented in (Supplementary Table S2).

Molecular docking against TNBC proteins

Molecular docking is commonly used in drug discovery, particularly to determine the bindin affinity of phytocompounds with the target proteins^61,62. The binding affinities of the ten selected phytocompounds were evaluated against two TNBC-associated proteins, CK2 (PDB ID: 3H30) and ULK2 (PDB ID: 6YID), and were compared with the reference drug Capecitabine. The numerical docking results are summarized in Table 1 and the corresponding bar-graph visualization is shown in (Supplementary Figure S3).

Table 1.

Binding affinities (kcal/mol) of selected phytocompounds and the standard drug Capecitabine against CK2 (PDB ID: 3H30) and ULK2 (PDB ID: 6YID).

Compound name	Binding affinity (kcal/mol)
	CK2 (PDB ID: 3H30)	ULK2 (PDB ID: 6YID
Bilobetin	− 10.1	− 9.1
Ginsenosides	− 9.4	− 8.9
Hesperidin	− 9.2	− 8.9
Lupenone	− 9.0	− 9.1
Sesquiterpenes	− 8.8	− 7.6
Cypripedin	− 8.4	− 7.3
Lycopene	− 7.9	− 8.7
Cucurbitacin	− 7.7	− 8.7
Carotenoids	− 7.3	− 8.2
Fucoxanthin	− 7.2	− 3.7
Standard capecitabine	− 7.6	− 6.6

Overall, the phytocompounds displayed stronger binding affinities than Capecitabine (− 7.6 kcal/mol for CK2; − 6.6 kcal/mol for ULK2), underscoring their potential as alternative TNBC inhibitors. Among all screened ligands, bilobetin demonstrated the most favorable docking energies, with − 10.1 kcal/mol for CK2 and − 9.1 kcal/mol for ULK2, surpassing all other compounds and the standard drug. Ginsenosides also exhibited consistently strong affinities (− 9.4 kcal/mol for CK2; − 8.9 kcal/mol for ULK2), followed by hesperidin (− 9.2 kcal/mol and − 8.9 kcal/mol, respectively) and lupenone (− 9.0 kcal/mol and − 9.1 kcal/mol).

Other phytocompounds, including lycopene, cucurbitacin, cypripedin, sesquiterpenes, and carotenoids, demonstrated moderate affinities across both targets, with values ranging from − 7.2 to − 8.8 kcal/mol for CK2 and − 7.3 to − 8.7 kcal/mol for ULK2. By contrast, fucoxanthin exhibited the weakest binding to ULK2 (− 3.7 kcal/mol), although its affinity for CK2 remained moderate (− 7.2 kcal/mol).

Taken together, the docking profiles presented in Table 1 and illustrated in Figs. 1 and 2 suggest that bilobetin and ginsenosides form particularly stable complexes with CK2 and ULK2, respectively. These interactions highlight their potential as lead candidates for the development of TNBC-targeted therapeutics.

Fig. 1.

Molecular docking and interaction analysis of bilobetin and capecitabine with CK2 (PDB ID: 3H30). (a) surface representation of CK2 showing bilobetin bound within the ligand-binding pocket; (b) hydrogen-bond donor/acceptor mapping of bilobetin–CK2 interactions; (c) hydrophobicity surface representation of bilobetin in the CK2 binding site; (d) 2D interaction diagram highlighting hydrogen bonds, hydrophobic contacts, and other non-covalent interactions of bilobetin with CK2 residues. (e) Surface representation of CK2 with capecitabine bound; (f) hydrogen-bond donor/acceptor mapping of capecitabine–CK2 interactions; (g) hydrophobicity surface representation of capecitabine in the CK2 binding site; (h) 2D interaction diagram showing hydrogen bonding, hydrophobic, and additional non-covalent interactions of capecitabine with CK2 residues. Hydrophobic areas are represented in brown, polar residues in light blue, and hydrogen bonds by yellow dashed lines.

Fig. 2.

Molecular docking and interaction analysis of ginsenosides and capecitabine with ULK2 (PDB ID: 6YID). (a) Surface representation of ULK2 showing ginsenosides bound within the ligand-binding pocket; (b) hydrogen-bond donor/acceptor mapping of ginsenosides–ULK2 interactions; (c) hydrophobicity surface representation of ginsenosides in the ULK2 binding site; (d) 2D interaction diagram highlighting hydrogen bonds, hydrophobic contacts, and other non-covalent interactions of ginsenosides with ULK2 residues. (e) Surface representation of ULK2 with capecitabine bound; (f) hydrogen-bond donor/acceptor mapping of capecitabine–ULK2 interactions; (g) hydrophobicity surface representation of capecitabine in the ULK2 binding site; (h) 2D interaction diagram showing hydrogen bonding, hydrophobic, and additional non-covalent interactions of capecitabine with ULK2 residues. Hydrophobic areas are represented in brown, polar residues in light blue, and hydrogen bonds by yellow dashed lines.

Hydrogen bond and hydrophobic interaction analysis

To better understand the stability of the protein–ligand complexes, we analyzed hydrogen bonds and hydrophobic interactions formed between the ligands and target proteins (Figs. 1 and 2). For CK2 with bilobetin, the docking results revealed the formation of four hydrogen bonds involving key residues, including LYS68, LYS158, ASP175, and ARG47 (Table 2 and Fig. 1b, d). These residues are located within or near the ATP-binding catalytic loop, which is essential for CK2’s phosphorylation activity. In addition to hydrogen bonding, bilobetin formed several hydrophobic interactions with VAL53, ILE174, PHE113, LYS49, and ILE95, further stabilizing the complex (Table 2 and Fig. 1c). In contrast, capecitabine formed fewer hydrogen bonds (SER51 and ASN161) and had a weaker hydrophobic interaction network with PHE113, VAL66, ILE95, ILE174, and HIS160, indicating a less stable binding profile compared to bilobetin (Table 2 and Figs. 1e–h. This suggests that bilobetin has a more favorable and stable binding profile compared to capecitabine.

Table 2.

Amino acid residues involved in hydrogen bonding, hydrophobic contacts, and other key intermolecular interactions between phytocompounds (and the standard drug Capecitabine) with CK2 (PDB ID: 3H30) and ULK2 (PDB ID: 6YID).

Protein	Ligands	Amino acid: bond distance (Å)
		Hydrogen bond	Hydrophobic and other bonds
CK2	Bilobetin	LYS 68: 2.17807, LYS 158: 2.05385, ASP 175: 2.60684, ARG 47: 3.50814	VAL 53: 3.86229, ILE 174: 3.25563, PHE 113: 5.28469, LYS 49: 4.90214, ILE 95: 5.24546
	Capecitabine	SER 51: 2.46239, ASN 161: 2.39718	PHE 113: 3.91874, VAL66: 4.18746, ILE95: 4.48252, ILE 174: 4.91647, HIS 160: 3.34194
ULK2	Ginsenosides	–	TYR 226: 3.97241, TRP 198: 3.82854, PRO 134: 5.13495, PRO 212: 4.36003, LYS 133: 4.08358, ARG 254: 3.57994
	Capecitabine	SER 199: 31,618, TYR 226: 3.42228	TRP 198: 3.56586, PRO 212: 5.27572

For ULK2, the binding analysis with ginsenosides demonstrated hydrophobic interactions with key residues such as TYR226, TRP198, PRO134, PRO212, LYS133, and ARG254, which are involved in stabilizing the kinase-substrate binding during autophagy initiation (Table 2 and Figs. 2a–d). Ginsenosides also formed strong interactions within the hydrophobic pocket of ULK2, further supporting their potential as a strong candidate for TNBC treatment (Fig. 2c). In comparison, capecitabine exhibited weaker interactions, with only one hydrogen bond to SER199 and TYR226 and fewer hydrophobic contacts with TRP198 and PRO212, resulting in a less favorable binding affinity (Table 2 and Figs. 2e–h).

Hydrophobicity surface and 2D interaction diagrams

The hydrophobicity surface representations in Figs. 1c and 2c showed that both bilobetin and ginsenosides penetrate deeper into the hydrophobic regions of the CK2 and ULK2 binding pockets compared to capecitabine (Figs. 1g and 2g), which binds more superficially. This deeper hydrophobic penetration may contribute to the higher stability and binding affinity of the natural compounds. The hydrophobicity surfaces further demonstrate that bilobetin and ginsenosides interact with hydrophobic residues in both proteins, forming stable and energetically favorable complexes.

Additionally, the 2D interaction diagrams in Figs. 1d, h, 2d, and h depict hydrogen bonding, hydrophobic contacts, and van der Waals interactions for bilobetin and ginsenosides, with capecitabine showing fewer and weaker interactions. Hydrogen bonds are represented by yellow dashed lines, while hydrophobic interactions are shown in brown.

These results indicate that bilobetin and ginsenosides may exhibit higher binding stability than capecitabine, supporting their potential for further experimental validation as TNBC inhibitors.

Protein–ligand interaction and molecular docking pose

The binding interactions between the selected ligands and their respective target proteins were characterized in detail using molecular docking simulations (Table 2, Figs. 1 and 2). For the CK2–bilobetin complex (PDB ID: 3H30), the docking score was − 10.1 kcal/mol, which was notably stronger than that of the standard anticancer drug capecitabine (− 7.6 kcal/mol). Bilobetin formed four hydrogen bonds with LYS68, LYS158, ASP175, and ARG47, all located within or near the ATP-binding catalytic loop, which is essential for CK2 phosphorylation activity. Additional hydrophobic interactions with VAL53, ILE174, PHE113, LYS49, and ILE95 further stabilized the complex. In contrast, capecitabine formed two hydrogen bonds (SER51, ASN161) and a reduced hydrophobic network, suggesting comparatively weaker stabilization within the CK2 binding pocket.

For the ULK2–ginsenoside complex (PDB ID: 6YID), the docking score was − 8.9 kcal/mol, also outperforming capecitabine (− 6.6 kcal/mol). Ginsenoside engaged in extensive hydrophobic contacts with key residues such as TYR226, TRP198, PRO134, PRO212, LYS133, and ARG254, occupying a deep hydrophobic channel in the ULK2 active site. Capecitabine, while forming hydrogen bonds with SER199 and TYR226, exhibited fewer hydrophobic interactions and a lower overall binding affinity.

Although ginsenosides did not form hydrogen bonds with key residues, they demonstrated strong hydrophobic interactions with several essential residues (e.g., TYR226, TRP198, PRO134), suggesting that these interactions contribute significantly to the binding affinity and stability of the complex. The absence of hydrogen bonds does not necessarily diminish their potential as promising inhibitors, as hydrophobic interactions often play a major role in stabilizing larger, non-polar compounds.

Hydrophobicity surface mapping (Figs. 1c, g, 2c, and g) revealed that both bilobetin and ginsenosides penetrate deeper into hydrophobic regions compared to capecitabine, which tends to bind more superficially. This enhanced hydrophobic burial may contribute to the higher stability and docking scores of the natural compounds. The 2D interaction diagrams (Figs. 1d, h, 2d, and h) further illustrate these differences: bilobetin and ginsenosides displayed a denser network of hydrophobic, van der Waals, and hydrogen bond contacts than capecitabine, supporting their stronger binding affinity and more stable complexes.

Collectively, these results suggest that bilobetin (CK2) and ginsenosides (ULK2) exhibit superior interaction profiles compared to the standard capecitabine, potentially translating into improved inhibitory effects on their respective cancer-related targets. While molecular docking provided strong evidence of favorable binding affinities and interaction profiles, computational pharmacokinetic and toxicity analyses were conducted to further evaluate the drug-likeness of the selected ligands. Accordingly, the ADMET predictions (Supplementary Table S3) were examined to assess absorption, distribution, metabolism, excretion, and toxicity characteristics.

Drug-likeness and ADMET evaluation

The ADMET properties of the selected compounds were assessed to predict their drug-likeness and pharmacokinetic profiles (Supplementary Table S3). Overall, the tested compounds demonstrated favorable safety characteristics, with none exhibiting AMES mutagenicity or hepatotoxicity. Only cypripedin (CID: 174,864) was flagged for potential skin sensitization, whereas all others were predicted to be non-sensitizing, supporting their safety for further investigation.

With respect to absorption, several compounds displayed high predicted human intestinal absorption (> 85%), including lycopene (CID: 446,925), fucoxanthin (CID: 5,281,239), cucurbitacin (CID: 5,281,316), ginsenosides (CID: 3,086,007), lupenone (CID: 92,158), cypripedin (CID: 174,864), and sesquiterpenes (CID: 667,450). In contrast, hesperidin (CID: 10,621) showed limited absorption (31.5%), potentially restricting its bioavailability despite otherwise acceptable safety characteristics. Water solubility varied across compounds, with hesperidin and bilobetin exhibiting relatively higher solubility, while highly lipophilic carotenoids and lycopene showed lower solubility values, consistent with their hydrophobic nature.

Regarding distribution, most compounds exhibited moderate to low predicted blood–brain barrier (BBB) permeability, suggesting limited central nervous system penetration, which may reduce off-target neurological effects. Among metabolic predictions, none of the tested compounds were identified as CYP2D6 substrates, and only cypripedin and sesquiterpenes were predicted to inhibit CYP1A2, indicating minimal risk of drug–drug interactions. Excretion analysis revealed acceptable total clearance values across all ligands, with no compounds predicted as renal OCT2 substrates, further supporting their pharmacological safety.

Taken together, these findings indicate that most phytochemicals investigated combine favorable pharmacokinetic profiles with low predicted toxicity. In particular, bilobetin (CID: 5,315,459) and ginsenosides (CID: 3,086,007) stood out due to their high intestinal absorption, non-toxic predictions, and compatibility with their strong binding affinities observed in molecular docking. The integration of docking and ADMET data thus provides a rational basis for selecting bilobetin as a lead compound targeting CK2 and ginsenosides as a lead compound targeting ULK2, with capecitabine included as a standard reference for both proteins. These compounds were therefore advanced to molecular dynamics (MD) simulations and subsequent post-docking analyses to further evaluate the stability and dynamic behavior of the protein–ligand complexes.

MD simulation analysis

The best interactions between bilobetin and ginsenosides in their respective protein-binding complexes were examined using MD simulations to evaluate their efficacy and stability over time. During the simulation, we analyzed RMSD (Root Mean Square Deviation) for the protein backbone, complex structure, and ligand structure to assess the stability of the protein–ligand complexes. Additionally, a variety of features, including the dynamic cross-correlation matrix (DCCM), principal component analysis (PCA), hydrogen bond analysis, radius of gyration (Rg), and root-mean-square fluctuations (RMSF), were also evaluated by analyzing the trajectories from the molecular dynamics (MD) simulation.

Root mean square deviation (RMSD) analysis

RMSD is one of the most fundamental parameters for evaluating the stability of protein–ligand complexes, as it reflects the atomic positional deviations during the simulation. Figure 3a illustrates the RMSD profiles of CK2 complexed with bilobetin, ULK2 with ginsenosides, and their respective standard complexes with capecitabine. In the bilobetin–CK2 system, the protein RMSD demonstrated an initial equilibration phase during the first 10 ns, after which it stabilized consistently within the range of 1.5–2.1 Å. This steady profile indicates that CK2 maintained a well-folded structure in the presence of bilobetin, a hallmark of stable inhibitor binding. The ligand RMSD displayed minor fluctuations between 0.6 and 1.0 Å for most of the simulation, with transient excursions up to 1.5 Å observed after 60 ns, suggesting slight conformational adjustments without leaving the binding cavity. These results support the strong affinity and conformational stability predicted in molecular docking, which is consistent with reports of flavonoid inhibitors demonstrating persistent RMSD stability within CK2⁶³.

Fig. 3.

RMSD plots of protein backbone (blue) and ligands (brown) during 100 ns molecular dynamics simulations. (a) Bilobetin with human CK2 α kinase (PDB ID: 3H30), (b) ginsenosides with ULK2 (PDB ID: 6YID), (c) capecitabine with human CK2 α kinase (PDB ID: 3H30), and (d) capecitabine with ULK2 (PDB ID: 6YID). RMSD fluctuations illustrate the structural stability of the protein–ligand complexes over the simulation period.

In contrast, the ginsenosides–ULK2 complex showed higher backbone RMSD values, fluctuating between 4.0 and 4.8 Å, particularly beyond 70 ns (Fig. 3b). The ligand RMSD followed a similar pattern, reaching values as high as 10.5 Å, which highlights the structural flexibility induced by the bulky glycosyl moieties of ginsenosides. Nevertheless, the ligand remained bound within the ULK2 pocket throughout the simulation, indicative of adaptive stabilization rather than disengagement. This dynamic yet retained binding is consistent with previous reports where saponins and triterpenes induced loop mobility in autophagy-related kinases while preserving binding integrity⁶⁴.

By comparison, capecitabine demonstrated less favorable stability in both CK2 (Fig. 3c) and ULK2 (Fig. 3d) complexes. In CK2, the ligand RMSD gradually increased up to 5.0 Å, while in ULK2 it fluctuated between 2.4 and 4.8 Å, suggesting weaker anchoring and less persistent engagement with the active sites. Taken together, RMSD analysis emphasizes bilobetin’s superior stability with CK2 and ginsenosides’ retained adaptability within ULK2, outperforming the standard drug in maintaining stable complex formation.

Root mean square fluctuation (RMSF)

RMSF provides residue-level insights into protein flexibility, thereby distinguishing stable structural regions from those undergoing conformational changes. Figure 4 depicts the RMSF profiles for CK2 and ULK2 complexes. In the bilobetin–CK2 complex, most residues exhibited low RMSF values (≈ 1.0–2.5 Å), reflecting overall structural rigidity (Fig. 4a). Only loop regions around Lys100 and Glu200 displayed higher fluctuations, approaching 4.5 Å, which is characteristic of surface-exposed loops involved in conformational breathing. The stabilization of catalytic residues in the binding site suggests that bilobetin supports structural compactness while allowing minimal loop mobility, a feature previously reported for natural polyphenols binding to CK2⁶⁵.

Fig. 4.

RMSF profiles of protein residues during 100 ns molecular dynamics simulations. (a) CK2 α kinase in complex with bilobetin (PDB ID: 3H30), (b) ULK2 in complex with ginsenosides (PDB ID: 6YID), (c) CK2 α kinase in complex with capecitabine (PDB ID: 3H30), and (d) ULK2 in complex with capecitabine (PDB ID: 6YID). RMSF plots highlight the flexibility of individual amino acid residues, where lower fluctuations indicate stable regions and higher peaks correspond to flexible loop or terminal regions.

The ginsenosides–ULK2 complex exhibited greater fluctuations, with RMSF peaks up to 6.4 Å localized in residues near indices 100, 200, and at the C-terminal loops (Fig. 4b). This enhanced flexibility can be attributed to the intrinsic mobility of ULK2 loop regions, which are critical for autophagy regulation⁶⁶. Importantly, the key binding residues interacting with ginsenosides remained comparatively stable, suggesting that the ligand did not destabilize essential catalytic elements.

On the other hand, capecitabine complexes revealed increased fluctuations, particularly in CK2 residues spanning 50–100 and ULK2 residues 100–150 (Fig. 4c and d. Such fluctuations suggest lower binding efficiency, as the ligand failed to restrain local dynamics compared to bilobetin and ginsenosides. These findings reinforce the RMSD outcomes, where natural ligands demonstrated enhanced stabilization over the clinical standard.

Hydrogen bonding and interaction analysis

A crucial factor in determining the stability of proteins and ligands in MD simulations is hydrogen bonding. Residues Ser3, Arg14, Asp74, and Glu115 were shown to form persistent hydrogen bonds with the bilobetin–CK2 system, which together kept the ligand orientated favourably during the simulation (Fig. 5a and Simulation Interaction Diagram, Fig. 6a). These bonds’ stability is consistent with the usual interaction profile of potent CK2 inhibitors⁶⁷.

Fig. 5.

Hydrogen bond occupancy profiles of protein–ligand complexes during 100 ns molecular dynamics simulations. (a) Bilobetin with CK2 α kinase (PDB ID: 3H30), (b) ginsenosides with ULK2 (PDB ID: 6YID), (c) capecitabine with CK2 α kinase (PDB ID: 3H30), and (d) capecitabine with ULK2 (PDB ID: 6YID). The plots illustrate the number and persistence of hydrogen bonds formed between ligands and protein residues throughout the trajectory, where higher and more stable hydrogen bond counts indicate stronger and more consistent interactions.

Fig. 6.

Hydrogen bond, hydrophobic, ionic, and water-bridge interactions during 100 ns molecular dynamics simulations are depicted in the Simulation Interaction Diagram (SID). For example, ginsenosides with ULK2 (PDB ID: 6YID), bilobetin with CK2 α kinase (PDB ID: 3H30), capecitabine with CK2 α kinase, and capecitabine with ULK2. Bilobetin and ginsenosides exhibit more stable binding than capecitabine, as seen by SID stacked graphs that show the relative fractions of various interaction types across the trajectories.

Similarly, ginsenosides consistently engaged ULK2 residues Arg130, Leu132, and Ser199, supported by additional interactions mediated through their hydrophilic sugar moieties (Fig. 5b and Simulation Interaction Diagram, Fig. 6b).

A more complex binding mechanism is shown by the dynamic interaction profile of ginsenosides with ULK2 than by the static docking study. Although the MD simulations showed that the complex is further stabilised by the subsequent development of persistent hydrogen bonds during the simulation trajectory, molecular docking indicated a binding mode that is primarily stabilised by hydrophobic interactions. This suggests an adaptive binding mechanism in which the ligand can gradually optimise its posture and establish important polar connections with the protein’s binding pocket thanks to the initial hydrophobic contact, which serves as a secure foundation. Complex natural product inhibitors such as saponins and glycosylated flavonoids are known to exhibit this dynamic and multifaceted binding, which involves both robust hydrophobic burial and specific hydrogen bonding. This type of binding frequently results in increased binding affinity and selectivity [cite a relevant review if possible, e.g., on natural product binding dynamics]. This balances the complex’s substantial binding free energy with its reported stability in MD simulations (despite a greater beginning RMSD).

Their capacity to stay bonded in spite of larger RMSD changes is explained by these multipoint interactions. Capecitabine complexes, on the other hand, showed less hydrogen bonds with both CK2 and ULK2 (Fig. 5c, d; Simulation Interaction Diagram, Fig. 6c, d), and these interactions were less sustained throughout the orbit. This analysis is presented as Simulation Interaction Diagrams (SID), which summarize the fractions of hydrogen bond, hydrophobic, ionic, and water-bridge interactions across the 100 ns trajectories (Fig. 6). In contrast to the weaker and less reliable profiles of capecitabine, higher proportions of sustained interactions in ginsenosides and bilobetin suggest more stable binding mechanisms.

Radius of gyration (Rg)

The Radius of Gyration (Rg) provides a measure of the overall compactness of a protein structure during simulation. Figure 7 shows the Rg trajectories for CK2 and ULK2 complexes. The bilobetin–CK2 complex maintained Rg values between 22.5 and 23.0 Å throughout the 100 ns simulation, with minimal fluctuations after 10 ns of equilibration (Fig. 7a). This stable compactness suggests that bilobetin binding promoted a tightly folded conformation, which is consistent with the inhibitory mechanism of natural polyphenols that often reduce structural plasticity in kinases⁶⁸.

Fig. 7.

Radius of gyration (Rg), molecular surface area (MolSA), solvent accessible surface area (SASA), and polar surface area (PSA) plots for protein–ligand complexes over 100 ns molecular dynamics simulations. (a) Bilobetin with CK2 α kinase (PDB ID: 3H30), (b) ginsenosides with ULK2 (PDB ID: 6YID), (c) capecitabine with CK2 α kinase (PDB ID: 3H30), and (d) capecitabine with ULK2 (PDB ID: 6YID). Rg reflects the compactness of the protein structure, while MolSA, SASA, and PSA indicate solvent exposure and conformational flexibility. Stable Rg and surface area profiles across the trajectories suggest that the overall folding and structural integrity of the protein–ligand complexes were preserved throughout the simulations.

In contrast, the ginsenosides–ULK2 complex displayed slightly elevated Rg values, averaging around 24.5–25.2 Å, with fluctuations correlating with loop flexibility observed in RMSF analysis (Fig. 7b). Although higher Rg indicates increased structural relaxation, the protein maintained global integrity, implying that ginsenosides induced adaptive but non-disruptive flexibility. Such ligand-induced expansion has been reported in ULK-family kinases, where bulky triterpenes or glycosylated inhibitors occupy extended binding pockets⁶⁹.

Capecitabine, however, failed to stabilize the compactness of either protein. Its CK2 complex showed increasing Rg values toward 24.0 Å (Fig. 7c), while ULK2 exhibited greater fluctuations between 25.0 and 26.5 Å (Fig. 7d). These findings indicate that capecitabine may induce destabilization of the protein scaffold compared to bilobetin and ginsenosides.

Solvent accessible surface area (SASA)

SASA analysis reflects the degree of solvent exposure of protein residues, thereby providing insights into conformational stability and protein–ligand interactions. Figure 7 (see SASA panel) depicts SASA variations across all complexes. For the bilobetin–CK2 complex, SASA values remained consistently within the range of 14,000–14,500 Ų, with minimal fluctuations throughout the trajectory (Fig. 7a-SASA). This constancy suggests that bilobetin maintained CK2 in a compact conformation with stable solvent accessibility. Such reduced SASA variability has been previously reported for CK2 bound to potent flavonoid inhibitors⁷⁰.

The ginsenosides–ULK2 complex displayed larger SASA fluctuations, oscillating between 15,500 and 16,800 Ų (Fig. 7b-SASA). These variations parallel the loop mobility observed in RMSF and Rg analyses, supporting the conclusion that ginsenosides allow conformational breathing while preserving global structure. Similar solvent exposure dynamics have been reported for ULK2 complexes with chemically synthesized new compound, XST-14⁷¹.

In contrast, capecitabine complexes demonstrated greater SASA variability (Fig. 7c, d-SASA), particularly in ULK2, where solvent exposure increased progressively over the trajectory. This observation indicates a less stable interaction profile and supports the conclusion that natural compounds (bilobetin and ginsenosides) stabilize their targets more effectively than the standard.

Principal component analysis (PCA)

PCA provides a global perspective on protein dynamics by extracting dominant motion patterns from the simulation trajectory. Figure 8 displays the PCA projections for the bilobetin–CK2, ginsenosides–ULK2, and capecitabine complexes. The bilobetin–CK2 system exhibited a compact clustering of conformations within the first two principal components (PC1 and PC2), suggesting a restricted conformational landscape (Fig. 8a). This indicates that bilobetin binding constrained CK2 into a stable conformational ensemble, consistent with the behavior of high-affinity CK2 inhibitors⁷². Conversely, the ginsenosides–ULK2 complex displayed broader dispersion across PC1 and PC2, indicative of enhanced conformational flexibility (Fig. 8b). While this suggests dynamic adaptation, the clustering remained confined within a stable subspace, implying that the fluctuations were not destabilizing but rather reflective of loop and domain breathing. Such broader PCA sampling has been reported for ULK2 in complex with triterpenoids, highlighting ligand-induced conformational plasticity^73,74.

Fig. 8.

Principal component analysis (PCA) of molecular dynamics trajectories for protein–ligand complexes over 100 ns. (a) Bilobetin with CK2 α kinase (PDB ID: 3H30), (b) ginsenosides with ULK2 (PDB ID: 6YID), (c) capecitabine with CK2 α kinase (PDB ID: 3H30), and (d) capecitabine with ULK2 (PDB ID: 6YID). The PCA plots represent the distribution of conformational sampling along the first two principal components (PC1 and PC2), reflecting the dominant motions of the protein–ligand complexes. More compact clustering indicates restricted conformational dynamics and higher stability, whereas wider dispersion reflects greater flexibility during the simulation.

Capecitabine complexes, however, exhibited scattered and less compact trajectories, especially in ULK2, where motions spanned a wide range of PC1 (Fig. 8c, d). This dispersal reflects poor stabilization and greater conformational heterogeneity, further corroborating its weaker inhibitory potential relative to natural ligands.

Dynamic cross-correlation matrix (DCCM)

DCCM analysis provides residue-level insights into correlated and anti-correlated motions within protein complexes. As shown in Fig. 9a, the bilobetin–CK2 complex demonstrated strong positive correlations across catalytic domain residues, suggesting coordinated motions that stabilize ligand binding. The presence of such correlated domains is consistent with effective allosteric regulation in CK2 when engaged by potent inhibitors⁷⁵.

Fig. 9.

Residue cross-correlation maps (DCCM) of protein–ligand complexes derived from molecular dynamics simulations. The maps depict correlated (blue) and anti-correlated (yellow/light) motions of residues during the simulation, with values ranging from –1.0 to + 1.0 as shown in the color bar. (a) CK2 α kinase in complex with bilobetin (PDB ID: 3H30), (b) ULK2 in complex with ginsenosides (PDB ID: 6YID), (c) CK2 α kinase in complex with capecitabine (PDB ID: 3H30), and (d) ULK2 in complex with capecitabine (PDB ID: 6YID). The diagonal indicates self-correlations (value = 1), while off-diagonal regions reveal inter-residue correlated or anti-correlated motions, reflecting the effect of ligand binding on global protein dynamics.

In contrast, the ginsenosides–ULK2 complex exhibited both correlated and anti-correlated motions, particularly among residues in loop and hinge regions (Fig. 9b). This reflects the inherent flexibility of ULK2 when interacting with bulky ligands, where induced anti-correlated motions may play a functional role in stabilizing ligand engagement⁷⁶. Despite these fluctuations, the active site residues remained positively correlated, supporting effective inhibition.

For the CK2–capecitabine complex (Fig. 9c), the DCCM revealed markedly weaker positive correlations compared to the bilobetin-bound structure. Only limited sections surrounding the catalytic loop demonstrated moderate correlated movements, while substantial portions of the protein displayed diffuse or inconsistent correlation patterns. It appears that capecitabine is ineffective at stabilising CK2 dynamics based on this lack of coordinated residue migration. The scattered anti-correlated areas further indicate disrupted intramolecular communication, matching with its weaker binding affinity and larger structural fluctuations found in RMSD and RMSF investigations.

Capecitabine complexes displayed weaker correlations and more widespread anti-correlated motions, especially in ULK2 (Fig. 9d). Such disrupted communication between residues is indicative of unstable binding, consistent with the higher RMSD and SASA fluctuations observed earlier.

Molecular mechanics/ generalized Born surface area (MM/GBSA)

To complement the molecular docking and dynamic simulation results, the binding free energies of the selected ligands with their respective protein targets were calculated using the Prime MM/GBSA approach. Unlike docking scores, which primarily assess static binding poses, MM/GBSA integrates molecular mechanics energies and solvation effects, thus providing a more reliable estimation of protein–ligand affinity⁷⁷. The calculated ΔG_bind values are summarized in Table 3.

Table 3.

Predicted binding free energies (ΔG bind score, kcal/mol) calculated using the Prime MM-GBSA method for the ligand–receptor complexes.

Ligand	Protein	MM/GBSA ΔG bind score (kcal/mol)
Bilobetin	CK2 (PDB ID: 3H30)	− 61.72
Capecetabine (Standard)	CK2 (PDB ID: 3H30)	− 31.54
Ginsenosides	ULK2 (PDB ID :6YID)	− 47.91
Capecetabine (Standard)	ULK2 (PDB ID :6YID)	− 42.95

Bilobetin demonstrated the most favorable binding affinity toward CK2 (PDB ID: 3H30), with a ΔG bind score of –61.72 kcal/mol. This energy value is nearly twice as strong as that of the reference drug capecitabine (–31.54 kcal/mol), underscoring the superior stabilizing potential of bilobetin within the CK2 active site. This observation is consistent with the MD trajectory analyses, where bilobetin maintained stable RMSD values, limited SASA fluctuations, and strong positive correlations across active site residues. Together, these findings suggest that bilobetin not only binds tightly to CK2 but also promotes structural stabilization, reinforcing its candidacy as a natural CK2 inhibitor. Comparable stabilization effects of biflavonoids on CK2 have been reported in literature, supporting the mechanistic plausibility of bilobetin’s strong inhibitory action⁷⁸.

Ginsenosides also exhibited favorable binding toward ULK2 (PDB ID: 6YID), with a ΔG bind score of –47.91 kcal/mol, surpassing that of the standard capecitabine (–42.95 kcal/mol). Although the difference in energy values is more moderate than in the CK2 case, the higher affinity of ginsenosides is corroborated by their dynamic behaviour: the complex remained stable during MD simulations despite moderate conformational breathing, as evidenced by Rg and PCA analyses. This adaptive yet stable interaction profile reflects the bulky and glycosylated structure of ginsenosides, which has been similarly described in prior studies on saponin–kinase interactions⁷⁹.

Overall, the MM/GBSA results strongly reinforce the conclusions derived from docking and MD simulations. Both bilobetin and ginsenosides demonstrated superior binding energies compared to the standard drug, aligning with their stable dynamic performance and favorable ADMET profiles. These findings highlight their potential as promising natural inhibitors of CK2 and ULK2, respectively, and provide a rational framework for further in vitro and in vivo validation studies.

Bilobetin and ginsenosides are positioned as extremely promising lead compounds by the combined computational evidence, which includes favourable docking scores, stable MD profiles, and strong MM/GBSA binding energies. This analysis offers a strong basis for a number of important future actions. First, to establish their inhibitory effectiveness and anti-proliferative effects, experimental validation is required using cell-based investigations in TNBC models and in vitro kinase assays. Second, the comprehensive interaction maps produced here provide a guide for studying the structure–activity relationship (SAR); for example, the sugar moieties of ginsenosides could be optimised to improve the dynamic hydrogen bonds seen with ULK2, and the flavonoid core of bilobetin could be derivatised to improve its interactions with the catalytic residues of CK2. Lastly, since that these chemicals target two important pathways in TNBC, investigating their effectiveness both on their own and in conjunction with conventional chemotherapeutics is an intriguing translational approach.

Conclusion

This study investigated the therapeutic potential of natural compounds as inhibitors of TNBC, focusing on two key proteins implicated in its progression: CK2 (Casein Kinase II) and ULK2 (Unc-51-like autophagy-activating kinase 2). Through molecular docking, ADMET profiling, and molecular dynamics simulations, bilobetin and ginsenosides emerged as the most promising candidates, exhibiting stronger binding affinities and greater stability compared to the standard drug, capecitabine.

Bilobetin demonstrated robust and stable interactions with CK2, supported by both hydrogen bonding and hydrophobic contacts, indicating a superior inhibitory potential. Ginsenosides, despite lacking hydrogen bonds, established strong hydrophobic interactions within the ULK2 binding pocket, highlighting an alternative but equally effective binding mode. ADMET predictions further confirmed the bioavailability, solubility, and low toxicity of both compounds, reinforcing their suitability as drug-like candidates. MD simulations validated the dynamic stability of these complexes, confirming their capacity to maintain consistent binding throughout the simulation period.

Taken together, these findings suggest that bilobetin and ginsenosides hold substantial promise as natural inhibitors of CK2 and ULK2 in TNBC. While computational evidence strongly supports their therapeutic potential, experimental validation through in vitro and in vivo studies remains essential. Future research should also assess their combinatorial effects with existing chemotherapeutics and explore advanced delivery platforms, such as nanoparticle-based systems, to optimize therapeutic efficacy. Additionally, future optimisation using ginsenoside and bilobetin analogue synthesis and structure activity relationship (SAR) studies may improve their pharmacokinetic and efficacy against TNBC.

Author contributions

MAH, IJT, MSH, AP and SA: Writing – original draft, Visualization, Validation, Methodology, Simulation, Data curation. MSH, DRS, MTI and AAN: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Methodology, Formal analysis. MUA, ShS, SS, GSN, MRB, ZH and MJ: Writing – review & editing, Visualization, Validation, Supervision, Software, Methodology, Investigation, Formal analysis, Conceptualization. SRK, MT and MTS: Writing – review & editing, Validation, Methodology, Investigation, Formal analysis, Data curation. MSH and MTI: Conceptualization and Supervision.

Funding

Open access funding provided by Parul University.

Data availability

All data generated or analyzed during this study are included in the published article and its supplementary files. Protein structure information is publicly available in the Protein Data Bank (PDB) at https://www.rcsb.org: CK2 (PDB ID: 3H30) → [https://www.rcsb.org/structure/3H30) ULK2 (PDB ID: 6YID) → [https://www.rcsb.org/structure/6YID).

Declarations

Competing interests

The authors declare no competing interests.