Deciphering Multitarget Mechanisms of Cardiac Glycosides in Acute Myeloid Leukemia Using a Network Pharmacology and Molecular Docking

Abstract

Purpose

Acute myeloid leukemia (AML) remains the most aggressive form of leukemia, underscoring the urgent need for novel treatment strategies. Drug repurposing offers a promising approach to accelerate drug development process. Cardiac glycosides (CGs), traditionally used for heart conditions, have shown potential for AML therapy. As leukemic stem cells (LSCs) are key contributors to AML relapse and chemotherapy resistance, this study aims to elucidate the potential shared mechanisms of proscillaridin (PSN) and ouabain (OUA), two CGs with anti-LSC activity, focusing on their multitarget effects—an emerging strategy in cancer therapy.

Methods

An integrative in silico framework combining network pharmacology, bioinformatics, and molecular docking was employed to elucidate the molecular mechanisms, identify clinically relevant targets, and evaluate the binding potential of PSN and OUA, with predictions were subsequently supported by in vitro validation.

Results

Seventeen shared core targets of PSN–OUA in AML were identified by intersecting compound-predicted targets from SwissTargetPrediction with AML-relevant genes curated from clinical databases. Enrichment analyses using Gene Ontology, KEGG, and CancerGeneNet revealed strong associations of these core targets with key biological pathways involved in aggressive cancer phenotypes, including cell growth, apoptosis, and motility, which were subsequently validated to be affected by CGs in vitro. Protein–protein interaction mapping identified 10 hub targets—MTOR, PTGS2, ALB, ICAM1, SYK, PRKCA, MAPK14, MET, PRKCB, and MAP2K1—that are functionally interconnected and associated with clinical outcomes. Molecular docking and molecular dynamics simulations supported the high-affinity binding of PSN and OUA, particularly to MTOR and PTGS2. Consistently, both compounds suppressed MTOR- and PTGS2-associated downstream signaling in vitro.

Conclusion

Our findings elucidate multiple mechanisms by which CGs may exert anti-LSC activity, which could be crucial for the design of novel therapeutic strategies for AML. The proposed in silico framework is broadly applicable and may accelerate drug repurposing in AML and other cancers.

Introduction

Acute myeloid leukemia (AML) is considered the most clinically aggressive form of leukemia, accounting for nearly half of leukemia-related deaths, with a median age at diagnosis of approximately 69 years.1,2 Statistically, AML caused 0.145 million new cases and 0.130 million deaths worldwide in 2021, with incidence and mortality demonstrating a sustained upward trend over time, reflecting its high fatality.3,4 AML is driven by an uncontrolled proliferation of immature myeloid blasts, leading to their accumulation in peripheral blood and bone marrow. AML treatment strategies have evolved considerably—from traditional cytotoxic chemotherapy and hematopoietic stem cell transplantation to updated 7+3 regimens,5,6 as well as FDA-approved targeted therapies such as FLT3-ITD inhibitors and BCL2 inhibitors.7,8 However, long-term survival of AML patients remains poor, with a 5-year survival of approximately 29%.9 Additionally, outcomes are notably poorer in patients over the age of 60, with a 5-year survival of less than 15%.10 As the incidence of AML has gradually increased, while improvement in long-term survival remains limited,11,12 novel therapeutic agents for AML are urgently needed.

The process of developing a clinically viable drug is profoundly challenging, as reflected by the fact that only 3.4% of oncology candidate compounds ultimately proceed to clinical evaluation.13 To accelerate drug development, one of the most efficient approaches is to repurpose clinically approved non-oncology drugs for cancer therapy—a strategy widely referred to as drug repurposing (also known as drug repositioning).14,15 Cardiac glycosides (CGs), traditionally used as Na⁺/K⁺-ATPase inhibitors for heart diseases such as congestive heart failure and arrhythmias, show promising antitumor activity in several solid and hematologic malignancies, as supported by numerous experimental and clinical studies.16–18 Multiple hallmarks of cancer, namely sustained cell proliferation, apoptosis resistance, and activated cell motility, have been shown to be altered by CGs in multiple preclinical models.19 Epidemiological studies suggest that CG therapy in heart failure patients is associated with a lower risk of breast and prostate cancers,20,21 supporting its anticancer potential. Specifically in AML, CGs have been identified as novel drug candidates through computational analysis of gene signatures of leukemic stem cells (LSCs), which are key contributors to AML relapse and chemotherapy resistance.22 We previously reported that ouabain (OUA), identified as the most potent CGs in our study, exerts a strong cytotoxic effect against AML cells, with preferential activity toward LSCs and leukemic progenitor cells (LPCs), while having minimal effect in normal hematopoietic stem cells (HSCs).23 Likewise, proscillaridin (PSN), at a clinically relevant dose, was shown to have anticancer activity in high MYC expressing leukemic cells, including AML LSCs.24 Additionally, PSN has been reported as one of the most potent CGs across multiple screens against various cancers, including glioblastoma, colon and pancreatic cancer cells.25–27 As targeting LSCs has now become a key strategy to cure AML, the two potent and highly selective CGs, OUA and PSN were selected as drug candidates for further computational analysis to accelerate the comprehensive understanding of the molecular interactions and mechanisms of CGs in AML, which are not fully elucidated.

In the present study, we used in silico network pharmacology approaches as illustrated in Figure 1 to identify common molecular targets of OUA and PSN that are highly specific to AML and explore how CGs affect multiple pathways linked to AML aggressiveness. Hub targets with AML clinical relevance were also uncovered, and their potential interactions with OUA and PSN were further assessed via molecular docking. Our findings not only provide a basis for the clinical translation of CGs, particularly OUA and PSN, but also for the further development of other novel strategies targeting LSCs.

Figure 1.

Framework of this study illustrating an integrative strategy combining in silico network pharmacology, bioinformatics, and molecular docking to identify and validate AML-relevant targets of CGs, along with in vitro validation of CG effects in AML cells.

Material and Methods

Compound Preparation and Target Prediction

As mentioned, OUA and PSN were selected as drug candidates in the present study; their chemical information was obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov/) using the compound identifiers (CIDs) 439501 and 5284613, respectively.28 The structures of PSN and OUA were visually represented using DrugBank for detailed graphical illustration (Figure 2A and B). To predict potential molecular targets, the SMILES notation of compounds were uploaded into the SwissTargetPrediction database (http://www.swisstargetprediction.ch/).29 The crucial molecular targets shared by PSN and OUA (PSN–OUA intersected targets) were visualized using a Venn diagram by Venny 2.1.0 (https://bioinfogp.cnb.csic.es/tools/venny/).

Figure 2.

Identification of the core targets of PSN–OUA in AML. (A and B) Chemical structures of PSN and OUA, retrieved from DrugBank. (C) Venn diagram showing predicted molecular targets shared by PSN and OUA (PSN–OUA). (D) Venn diagram illustrating the intersection of PSN–OUA predicted targets with AML-associated genes, highlighting the shared AML-relevant targets. (E) Compound–target interaction network visualized in Cytoscape (3.10.3), depicting the relationships between the identified core molecular targets.

Shared PSN–OUA and AML-Associated Targets

To comprehensively identify AML-related therapeutic targets, an in silico high-throughput screening was conducted using the DisGeNET (http://www.disgenet.org)30 and GeneCards (https://www.genecards.org/)31 databases with the keyword “acute myeloid leukemia”, incorporating all available sub-datasets. The two gene sets were merged, redundancies removed, and a refined list of candidate genes was generated. This list was then intersected with the PSN–OUA intersected targets to identify the shared targets most relevant to AML. The intersection was visualized using a Venn diagram by Venny 2.1.0. Subsequently, a compound–target network was constructed using Cytoscape 3.10.3 (https://cytoscape.org/)32 to map key interactions between CGs (PSN and OUA) and AML-related molecular targets.

Functional Annotation and Enrichment Analyses

To investigate the key pathways associated with PSN–OUA targets in AML, Gene Ontology (GO)33 and Kyoto Encyclopedia of Genes and Genomes (KEGG)34 enrichment analyses were performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID) tools (https://davidbioinformatics.nih.gov/).35 We ranked the terms by –log₁₀ P-value on the y-axis and plotted fold enrichment on the x-axis, providing an intuitive and informative overview of the most significant functional categories. Bubble charts and Sankey diagrams were created using SRplot (http://www.bioinformatics.com.cn/SRplot).36

Additionally, the same gene set was analyzed using the CancerGeneNet database (https://signor.uniroma2.it/CancerGeneNet/) to examine associations with canonical cancer hallmark-related pathways.37 Gene names were submitted via the “Connect Proteins” module, with the network complexity set to level 3 (all) to allow comprehensive mapping. For visualization, a minimum interaction score of 0.7 was applied under the “relaxed” setting, facilitating interpretation of functional relationships and the potential contributions of these genes to cancer aggressiveness.

Construction of Network and Identification of Hub Targets

To identify hub molecules driving network connectivity, a three-tiered clustering strategy was employed. First, a primary protein–protein interaction (PPI) network was constructed using the STRING 12.0 database (https://string-db.org/)38 via the “Multiple Proteins” analysis tool. The curated list of PSN–OUA and AML overlapping genes was submitted with the organism set to Homo sapiens. Interactions with a confidence score greater than 0.4 were retained to ensure medium or higher reliability. In the resulting network, nodes represent proteins and edges denote protein–protein associations. Second, the core molecular interaction landscape was refined using Cytoscape software32 to further explore connectivity patterns derived from the STRING output. The network data, exported as string_interactions_short.tsv, were imported in TSV format as a molecular interaction network. Finally, the topological significance of each node was evaluated using the cytoHubba. Degree centrality was used to rank network nodes, and the top 10 hub genes—those with the highest number of direct connections—were selected for subsequent analyses.

Molecular Docking

Molecular docking was performed to investigate the binding specificity between PSN and OUA and the top 10 hub targets from PPI. The 3D structures of PSN and OUA were downloaded in SDF format from the PubChem database. Protein structures were retrieved from the RCSB Protein Data Bank (PDB), including MTOR (PDB ID: 4JSV), PTGS2 (PDB ID: 5F1A), ALB (PDB ID: 1AO6), ICAM1 (PDB ID: 1IC1), SYK (PDB ID: 4FL2), PRKCA (PDB ID: 2GZV), MAPK14 (PDB ID: 6SFO), MET (PDB ID: 3DKG), PRKCB (PDB ID: 2I0E), and MAP2K1 (PDB ID: 3W8Q). Molecular docking simulations were carried out using the CB-Dock2 web server (https://cadd.labshare.cn/cb-dock2/),39 which performs automated cavity detection followed by blind docking based on AutoDock Vina. For each target–compound pair, five predicted binding cavities were generated, and the docking pose with the lowest Vina score within the best-fit pocket (CurPocket) was selected and presented in this study. The binding affinity was expressed in kcal/mol. Docking complexes were visualized in 3D with receptor surfaces hidden to highlight the ligand–target interface. Hydrogen bonds were displayed as solid blue dashed lines, while weak hydrogen bonds were shown as lighter blue dashed lines. The top two predicted targets were verified using molecular dynamics (MD) simulation. To further support the molecular docking results, biological activities of PSN and OUA on the predicted targets were also tested.

MD Simulation

MD simulations were performed to evaluate the stability of protein–ligand complexes between PSN and the predicted cancer targets MTOR and PTGS2 using GROMACS 2021.7 with the CHARMM36 force field for protein parameterization. For MTOR, the ligand-interacting functional domain corresponding to the predicted binding site was used for MD simulations. Ligand parameters were generated using the CGenFF server based on optimized geometries obtained from ORCA calculations. Each protein–ligand complex was solvated in a cubic simulation box filled with TIP3P water molecules, and appropriate counterions were added to neutralize the system. Energy minimization was performed using the steepest descent algorithm until the maximum force fell below 1000 kJ/mol·nm, followed by system equilibration in two stages: a 100-ps NVT ensemble at 300 K and a 100-ps NPT ensemble at 1 bar using the Parrinello–Rahman barostat. Production MD simulations were subsequently conducted for 100 ns with a 2-fs integration time step. Post-simulation trajectory analyses, including backbone and ligand root mean square deviation (RMSD) calculations, were performed using standard GROMACS analysis tools, and RMSD plots were generated using GraphPad Prism.

Evaluation of Clinical Associations and Disease Relevance

To explore the clinical relevance of the top 10 hub genes, we analyzed their mRNA expression patterns and prognostic significance in AML patients. Expression data were obtained through GEPIA (http://gepia.cancer-pku.cn/), a web-based platform that integrates RNA-seq data from The Cancer Genome Atlas (TCGA) and the Genotype-Tissue Expression (GTEx) databases.40 Tumor samples (N = 173) were sourced from the TCGA-LAML dataset, while normal tissue data (N = 70) were derived from GTEx. Differential expression was assessed using a |Log₂FC| threshold of 1, with statistical significance defined as P < 0.01. To assess prognostic value, overall survival analysis was conducted using the Kaplan–Meier Plotter (https://www.kmplot.com/analysis/),41 applying the JetSet best probe set for each gene: MTOR (Probe ID: 202288_at), PTGS2 (Probe ID: 204748_at), ALB (Probe ID: 211298_s_at), ICAM1 (Probe ID: 202638_s_at), SYK (Probe ID: 226068_at), PRKCA (Probe ID: 213093_at), MAPK14 (Probe ID: 202530_at), MET (Probe ID: 203510_at), PRKCB (Probe ID: 207957_s_at), and MAP2K1 (Probe ID: 202670_at). Gene expression was categorized into high and low groups according to the upper and lower quartiles, with a total of 1,608 patients included. P-value was calculated using the Log rank test, and hazard ratio (HR) was obtained as illustrated in the corresponding figure.

Reagents

PSN and OUA were purchased from Sigma-Aldrich (St. Louis, MO, USA). Stock solutions were prepared under sterile conditions, aliquoted, and stored at −20 °C.

Cell Culture

Human AML-derived cell lines, including HL-60 (JCRB0085) and KG-1 (JCRB9051), were obtained from the Japanese Collection of Research Bioresources (JCRB) Cell Bank (Osaka, Japan). HL-60 cells were cultured in RPMI-1640 medium, while KG-1 cells were maintained in Iscove’s Modified Dulbecco’s Medium (IMDM; Thermo Fisher Scientific, Waltham, MA, USA); both media were supplemented with 10% fetal bovine serum (FBS), 100 U/mL penicillin, and 100 µg/mL streptomycin. Cells were maintained in a humidified incubator at 37 °C with 5% CO₂. Mycoplasma contamination was routinely checked using the MycoAlert™ PLUS Mycoplasma Detection Kit (Lonza, Cologne, Germany), and any contaminated cultures were discarded.

Apoptosis Assay

Apoptosis was evaluated using the Annexin V/7-AAD assay (BioLegend, San Diego, CA, USA) in accordance with the manufacturer’s instructions. Briefly, cells were washed with cold PBS and stained with PE-conjugated Annexin V and 7-AAD in Binding Buffer at room temperature while protected from light. Samples were immediately analyzed using a FACSCanto flow cytometer (BD Biosciences, San Jose, CA, USA). Live cells were identified as Annexin V^–/7-AAD^–, whereas dead cells comprised early apoptotic (Annexin V⁺/7-AAD^–), late apoptotic (Annexin V⁺/7-AAD⁺), and necrotic (Annexin V^–/7-AAD⁺) populations.

Cell Proliferation Assay

Cell proliferation was assessed based on the relative growth rate at Days 0, 3, and 5 under the indicated treatments. HL-60 and KG-1 cells were seeded in 96-well plates at a density of 10,000 cells/well and 20,000 cells/well, respectively. At each time point, 10 μL of Cell Counting Kit-8 (CCK-8) reagent (Dojindo Molecular Technologies, Kumamoto, Japan) was added to each well and incubated for 4 h in a standard cell culture incubator. Absorbance was measured at 450 nm using a VICTOR Nivo S multimode microplate reader (Revvity, Waltham, MA, USA). Background absorbance was subtracted, and the signal was normalized relative to Day 0.

Cell Motility Assay

Cell migration and invasion were evaluated using a 24-well Transwell system with 8.0 μm pore-size inserts (Corning Costar, Kennebunk, ME, USA). For the invasion assay, the upper surface of the membrane was precoated with Matrigel (BD Bioscience) to mimic the extracellular matrix barrier. AML cells were pretreated with the indicated conditions for 24 h, after which 30,000 treated cells suspended in serum-free medium were added to the upper chamber, while the lower chamber was filled with complete medium to serve as a chemoattractant. After 48 h of incubation, migrated or invaded cells were transferred to a 96-well plate, stained with Hoechst 33342 (Thermo Fisher Scientific), and imaged using an inverted fluorescence microscope (Eclipse Ti-U, Nikon, Tokyo, Japan). Quantitative analysis was performed using ImageJ software, with results expressed as a fold change relative to non-treated (NTX) control group.

Western Blotting

KG-1 cells were treated with the indicated concentrations of PSN or OUA and then lysed in lysis buffer (Cell Signaling Technology, Danvers, MA, USA) supplemented with protease inhibitor cocktail (Roche Molecular Biochemicals, Indianapolis, IN, USA) and phenylmethylsulfonyl fluoride (PMSF), followed by incubation at 4 °C for 1 h. Protein concentrations were determined using a BCA assay kit (Thermo Fisher Scientific). Proteins (30–50 µg) were separated by SDS–PAGE and transferred to PVDF membranes. Membranes were blocked with 5% non-fat dry milk in TBST for 2 h at room temperature and incubated with primary antibodies overnight at 4 °C: phospho-mTOR (Ser2448) (#5536), mTOR (#2983), Raptor (#2280), Rictor (#2114), G-β-L (#3274), COX-2 (#12282), and c-Myc (#5605) (Cell Signaling Technology). After washing, membranes were incubated with HRP-conjugated secondary antibodies (EMD Millipore, Burlington, MA, USA) for 2 h at room temperature. Bands were visualized using an ImageQuant™ digital imaging system (GE Healthcare, Pittsburgh, PA, USA). β-actin served as a loading control. Band intensity was quantified using ImageJ software.

RNA Isolation and Quantitative Real-Time PCR (RT-qPCR)

Total RNA was extracted from KG-1 cells following the indicated treatments using TRIzol reagent (Molecular Research Center, Cincinnati, OH, USA) and quantified using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific). Complementary DNA (cDNA) was synthesized using SuperScript III reverse transcriptase and oligo(dT) primers (Thermo Fisher Scientific) according to the manufacturer’s instructions. Quantitative real-time PCR (RT-qPCR) was performed on a CFX384 Touch Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA) using SYBR^TM Select Master Mix (Thermo Fisher Scientific). The primers used for qPCR were as follows: human PTGES2, forward 5′-CCTCTATGAGGCTGCTGACAAG-3′ and reverse 5′-ATCACACGCAGCACGCCATACA-3′; and GAPDH, forward 5′-AGCCACATCGCTCAGACAC-3′ and reverse 5′-GCCCAATACGACCAAATCC-3′. Relative mRNA expression levels were calculated using the 2^−ΔΔCt method, with GAPDH serving as the housekeeping gene.

Statistical Analysis

Data are presented as mean ± standard deviation (SD). Statistical analyses were performed using GraphPad Prism from five independent experiments. One-way ANOVA followed by Dunnett’s multiple comparisons test, and two-way ANOVA followed by Tukey’s multiple comparisons test were used to determine statistical significance. A P-value of less than 0.05 was considered statistically significant, with statistical power >80% unless otherwise specified. Half maximal inhibitory concentration (IC₅₀) values were calculated using a nonlinear regression model: log(inhibitor) versus normalized response with variable slope.

Results

Identification of Core Molecular Targets of PSN and OUA in AML

CGs have gained renewed interest for their anti-tumor activity against AML, specifically targeting LSCs. However, their precise molecular targets, crucial for further drug design, remain poorly defined. PSN and OUA, two well-characterized CGs with anti-LSC activity, were used as drug candidates to predict common anti-LSC targets. Using SwissTargetPrediction, 100 targets were identified separately for PSN and OUA based on their chemical structures, shown in Figure 2A and B. Then, the two sets of data for PSN and OUA were intersected to identify the shared molecules. The resulting 24 overlapping candidates, denoted as PSN–OUA targets, are shown in the Venn diagram (Figure 2C) and listed in Table 1. To increase target specificity for AML, the PSN–OUA targets were intersected with a total of 10,077 AML-related targets, retrieved from DisGeNET and GeneCards using the keyword “acute myeloid leukemia”, and visualized in the Venn diagram (Figure 2D and Table 1). The resulting 17 overlapping targets—MTOR, MAPK14, PRF1, DHODH, MET, PRKCA, ADORA2A, PTGS2, SYK, ICAM1, MAP2K1, BCL2A1, PDE5A, PRKCB, PRKCE, PRKCH, and ALB—may represent the core targets of PSN–OUA in AML. The compound–target interaction network was constructed using Cytoscape (Figure 2E). The network contains 18 nodes and 17 edges, with 17 core targets displayed as green ovals and the PSN–OUA as an orange rounded rectangle. Taken together, a set of predicted core targets of CGs in AML was identified.

Table 1.

List of Predicted Molecular Targets of PSN and OUA as Identified by the SwissTargetPrediction.

No.	Intersection of PSN and OUA Targets (PSN–OUA)	Intersection of PSN–OUA and AML Genes (PSN–OUA in AML)
1.	MTOR	MTOR
2.	MAPK14	MAPK14
3.	PRF1	PRF1
4.	DHODH	DHODH
5.	HCAR2	MET
6.	MET	PRKCA
7.	PRKCA	ADORA2A
8.	ADORA2A	PTGS2
9.	SLC5A2	SYK
10.	SLC28A2	ICAM1
11.	PTGS2	MAP2K1
12.	TTL	BCL2A1
13.	SYK	PDE5A
14.	ICAM1	PRKCB
15.	MAP2K1	PRKCE
16.	PSEN2 PSENEN NCSTN APH1A PSEN1 APH1B	PRKCH
17.	SLC5A1	ALB
18.	BCL2A1
19.	ADK
20.	PDE5A
21.	PRKCB
22.	PRKCE
23.	PRKCH
24.	ALB

GO Enrichment Analysis Highlights the Biological Significance of PSN–OUA Targets in AML

To further explore the biological functions of PSN–OUA in AML, we submitted the identified 17 core targets for GO enrichment analysis using DAVID tools. The GO data were categorized into three main aspects: biological process (BP), cellular component (CC), and molecular function (MF). For BP, 45 out of 57 GO terms were statistically significant (P < 0.05). Among these, apoptotic process (GO:0006915) and signal transduction (GO:0007165) involved the largest number of target genes—7 in each group. The genes associated with apoptotic process were BCL2A1, ADORA2A, PRKCB, PRKCE, PRF1, PRKCA, and MAPK14, while those linked to signal transduction included MAP2K1, PRKCH, PRKCB, PRKCE, PDE5A, MAPK14, and MET. Concurrently, protein kinase C signaling (GO:0070528) exhibited the greatest fold enrichment. Figure 3A summarizes the top 20 BP GO terms ranked by P-value (see Supplementary Table S1 for the complete gene lists). For CC, target genes were predominantly associated with the cytosol (GO:0005829), endoplasmic reticulum (GO:0005783), plasma membrane (GO:0005886), cytoplasm (GO:0005737), immunological synapse (GO:0001772), and mitochondrion (GO:0005739), all statistically significant (P < 0.05). Notably, the cytosol contained the largest number of associated genes (N = 13), including MAP2K1, PRKCH, SYK, BCL2A1, PRKCB, PRKCE, PRF1, PRKCA, PTGS2, MAPK14, MTOR, DHODH, and PDE5A (Figure 3B). Figure 3C presents the top 16 significantly enriched MF (P < 0.05), with all 17 targets were associated with protein binding (GO:0005515), and 11 of MF (68.75%) were linked to kinase-related activity, suggesting a predominant association with kinase-related processes in AML. Given that aberrant activation of kinase has been linked to leukemogenesis,42 targeting kinase activity with CGs may offer therapeutic benefit in AML.

Figure 3.

GO enrichment analysis of the core targets of PSN–OUA in AML. (A) Top 20 enriched BP, ranked by statistical significance (P < 0.05). (B) CC significantly enriched at P < 0.05. (C) MF significantly enriched at P < 0.05. GO enrichment analysis was performed via the DAVID bioinformatics platform, and results were visualized as a bubble plot using SRplot (see Supplementary Table S1 for the complete gene lists). The diameter indicates the number of genes overlapping with the specified GO term and the color indicates the enrichment P-value.

Connections of PSN–OUA Targets in AML with Cancer-Related Pathways and Phenotypes

KEGG is a well-established database that links genes to cellular pathways involved in both normal physiological processes and various diseases.34 Using the identified 17 core targets of PSN–OUA in AML, a total of 81 KEGG pathways were initially identified. Upon filtering with a P-value < 0.05, 73 pathways remained significantly enriched. To better visualize the signaling pathways associated with the identified core targets, we plotted the top 20 enriched pathways as a Sankey diagram and dot plot, sorted by P-values in ascending order, in Figure 4A (see also Supplementary Table S2). Top-ranked pathways include those directly involved in leukemogenesis, including VEGF signaling pathway43,44 and EGFR tyrosine kinase inhibitor resistance pathway,45 which exhibited notably high fold enrichment values of 41.83 and 31.38, respectively (Supplementary Figure S1A and B). Additionally, NF-κB signaling pathway, recurrently implicated in AML pathogenesis,46 was also enriched. Therefore, PSN and OUA are likely involved in AML-relevant oncogenic signaling pathways.

Figure 4.

Pathway enrichment analyses of the core targets of PSN–OUA in AML. (A) Top 20 KEGG pathways identified using the DAVID bioinformatics platform, and visualized with SRplot as Sankey and dot plots, ranked by P-value to highlight key molecular mechanisms (see also Supplementary Table S2). The diameter indicates the number of genes overlapping with the specified KEGG pathway and the color indicates the enrichment P-value. (B) Enrichment analysis of pathways commonly contributing to hallmarks of cancers using CancerGeneNet. The analysis, visualized with a level 3 relaxed layout and a score threshold of 0.7, highlights the pivotal pathways involved in AML progression. This figure was redrawn from the original output of the analytical platform, with layout adjustments made to improve readability.

To explore how PSN–OUA core targets in AML align with the established hallmarks of cancer, we performed the CancerGeneNet network analysis. As shown in Figure 4B, several hub proteins were connected to phenotypes that drive aggressive tumor cell behaviors. Most prominently, PRKCA, MAPK14, and the components of the mTORC1 complex, appear to play a central role in signaling cascades that promote cell proliferation. PRKCA was also associated with apoptotic resistance, while ICAM1 and MET were implicated in the context of cell motility through chemotaxis and metastasis, respectively. PTGS2 was linked to inflammation—a key driver of tumor progression.47 Interestingly, MAPK14 may also be connected to M2 macrophage polarization, a phenotype often correlated with aggressive leukemia development.48 All hallmarks of cancer were predicted to be activated.

PSN and OUA Induce Apoptosis and Inhibit Cell Growth and Motility in AML Cells

Having demonstrated through enrichment analyses that PSN–OUA may disrupt homeostasis in AML by tiling the balance of cell death, particularly apoptosis, and survival, we first examined the anti-leukemic effects of PSN and OUA for in vitro validation. AML HL-60 (FAB M2) and KG-1 (FAB M0) cells were treated with various concentrations of PSN and OUA (0–300 nM) for 48 h, and their IC₅₀ values were calculated from the dose-response curve. PSN demonstrated IC₅₀ values of 42.79 nM in HL-60 cells and 26.14 nM in KG-1 cells, whereas OUA displayed IC₅₀ values of 48.30 nM and 39.16 nM, respectively (Figure 5A). Figure 5B shows that apoptosis is a major mode of cell death induced by PSN and OUA, thus strengthening the roles of PSN–OUA targets in apoptosis regulation. Further, Hoechst 33342 staining revealed hallmark apoptotic features—nuclear shrinkage, chromatin condensation, and apoptotic body formation—in both AML cell lines upon PSN or OUA treatment (Supplementary Figure S2). Quantitative analysis confirmed a dose-dependent increase in the percentage of apoptotic nuclei, with significant apoptosis observed at 20 nM for PSN and 30 nM for OUA.

Figure 5.

PSN and OUA significantly induce apoptosis and suppress AML cell growth. AML HL-60 and KG-1 cells were treated with the indicated concentrations of PSN and OUA, ranging from 0–300 nM, for 48 h. (A) IC₅₀ values of PSN and OUA in AML cells were determined using a variable-slope log-dose response model. (B) PSN and OUA primarily induce apoptosis, with minimal necrosis, as determined by Annexin V/7-AAD assay using flow cytometry. Live cells are defined as Annexin V⁻/7-AAD⁻, whereas dead cells comprised early apoptotic (Annexin V⁺/7-AAD⁻), late apoptotic (Annexin V⁺/7-AAD⁺), and necrotic (Annexin V⁻/7-AAD⁺) cells. Data are presented as mean ± SD (N = 4). (C) PSN and OUA inhibit AML cell proliferation, as determined by the CCK-8 assay. Proliferation rates on Days 3 and 5 were normalized relative to Day 0. Data are presented as mean ± SD (N = 5). *P < 0.05 for PSN or OUA at 20 nM versus NTX; ^#P < 0.05 for PSN or OUA at 30 nM versus NTX, two-way ANOVA followed by Tukey’s multiple comparisons test. Statistical power >80%, except for PSN and OUA at 20 nM versus NTX in KG-1 cells on Day 5 (power = 63.30% and 55.70%, respectively).

Next, we evaluated the effects of PSN and OUA on AML cell growth over 5 days. HL-60 and KG-1 cells were treated with lower concentrations of PSN and OUA (0–30 nM), and cell viability was determined by CCK-8 assay. Figure 5C shows that cell viability was markedly suppressed by PSN and OUA in HL-60 and KG-1 cells when compared to the NTX control. As (i) OUA at 20 nM was subtoxic and (ii) the cells in most conditions retained the ability to expand, we postulated that the reduced cell growth by PSN and OUA was attributable, at least in part, to their antiproliferative effect. To further substantiate this notion, we performed additional cell-cycle analysis using propidium iodide (PI) staining. The results showed that both PSN and OUA (30 nM) induced a marked arrest at the G2/M phase after 48 h treatment (Supplementary Figure S3), indicative of disruption of cell-cycle progression. Notably, PSN, and to a much lesser extent OUA, at the tested concentration (30 nM) also showed a marked increase in the sub-G1 population in both HL-60 and KG-1 cells, indicative of enhanced apoptotic cell loss and consistent with the apoptosis assays.

PSN–OUA core targets in AML were also predicted by CancerGeneNet network analysis to be implicated in metastasis and motility-related process, ie, chemotaxis. To further assess the effects of PSN and OUA on cancer-related phenotypes, cell motility—including cell migration and invasion, a key step underlying metastasis and extramedullary involvement in AML—was evaluated using Transwell migration and invasion assays in PSN- or OUA-treated cells at 48 h. HL-60 and KG-1 cells were pretreated with low-dose PSN or OUA at 10 nM, after which viable cells were counted and seeded onto the Transwell inserts with or without Matrigel. Figure 6 shows that PSN and OUA markedly impaired the ability of HL-60 and KG-1 cells to migrate and invade through an extracellular matrix-like barrier, supporting that PSN and OUA exert multiple effects on AML phenotypes, in agreement with the in silico predictions.

Figure 6.

PSN and OUA inhibit AML cell migration and invasion. HL-60 and KG-1 cells were pretreated with PSN or OUA (10 nM) for 24 h prior to Transwell migration and Matrigel invasion assays. After 48 h, migrated and invaded HL-60 (upper) and KG-1 (lower) cells were stained with Hoechst 33342, imaged, and quantified relative to NTX. Scale bar = 100 μm. Data are presented as mean ± SD (N = 5). *P < 0.05, **P < 0.01, ***P < 0.001 versus NTX by one-way ANOVA with Dunnett’s multiple comparisons test.

PPI Network Analysis Reveals Hub Targets Among PSN–OUA Core Targets in AML

We applied the three-tiered clustering strategy to identify hub targets with the highest connectivity, which are often key molecules in compound-target interactions. First, the identified 17 core targets of PSN–OUA in AML were submitted to the STRING tool for PPI network construction. These core targets were represented as network nodes, of which 14 nodes exhibited direct interaction edges (40 edges), while DHODH, ADORA2A, and PDE5A remained unconnected to other nodes (Figure 7A). Important network parameters included an average node degree of 4.71 and an average local clustering coefficient of 0.629. Additionally, the significant PPI enrichment (P-value = 3.36 × 10^–11) implies that the observed protein interactions are not random, highlighting a potential functional relationship among these proteins. To better visualize the main cluster of 14 hub proteins, the STRING-exported network was imported into Cytoscape (Figure 7B), and the cytoHubba plugin was used to identify the top 10 hub targets (Figure 7C and Table 2). Node color ranges from red to yellow to reflect degree scores from high to low, respectively. The hub targets, listed in order of degree centrality, were MTOR, PTGS2, ALB, ICAM1, SYK, PRKCA, MAPK14, MET, PRKCB, and MAP2K1, and were used for further investigation into their clinical relevance and molecular docking with PSN and OUA.

Figure 7.

Identification of the hub targets of PSN–OUA in AML using PPI analysis. (A) PPI network constructed from the intersection of PSN–OUA-predicted AML-relevant targets, as retrieved from the STRING database. Colored nodes represent input proteins and their direct interactors, with node fill indicating the presence of known or predicted 3D structures. Edges are color-coded by evidence type: database (blue), experimental (pink), co-occurrence (light blue), text mining (yellow-green), co-expression (black), and homology (purple). (B) Refined PPI network visualized in Cytoscape, demonstrating only the connected proteins. (C) Top 10 hub proteins ranked by degree centrality using the cytoHubba plugin network analyzer. Node colors represent degree values, ranging from red (highest) to yellow (lowest).

Table 2.

Top 10 Hub Targets of the PSN–OUA in AML and Their Topological Parameter from PPI Network Analysis

No.	Gene	Name	Degree
1.	MTOR	Mechanistic Target of Rapamycin Kinase	9
2.	PTGS2	Prostaglandin-Endoperoxide Synthase 2	8
3.	ALB	Albumin	8
4.	ICAM1	Intercellular Adhesion Molecule 1	6
5.	SYK	Spleen Associated Tyrosine Kinase	6
6.	PRKCA	Protein Kinase C Alpha	5
7.	MAPK14	Mitogen-Activated Protein Kinase 14	6
8.	MET	MET Proto-Oncogene, Receptor Tyrosine Kinase	5
9.	PRKCB	Protein Kinase C Beta	3
10.	MAP2K1	Mitogen-Activated Protein Kinase Kinase 1	4

Clinical Significance of PSN–OUA Hub Targets in AML

To assess the clinical significance of the identified 10 hub targets of PSN–OUA in AML, we next examined their gene expression profiles and prognostic significance in AML patients. Expression analysis via GEPIA revealed that PTGS2, ALB, SYK, and MAPK14 were significantly upregulated in AML samples from the TCGA-LAML dataset when compared to GTEx-derived normal tissues (P < 0.01) (Figure 8). In contrast, ICAM1, PRKCA, MET, and MAP2K1 were markedly downregulated, while the remaining genes showed no significant differences. Furthermore, Kaplan–Meier analysis revealed that high expression of PTGS2, PRKCA, and MAP2K1, or low expression of MTOR, ICAM1, and PRKCB, was significantly associated with shorter overall survival (OS) in AML patients (P < 0.05; corresponding P-values and hazard ratios (HR) shown in Figure 9), while the remaining hub genes showed no significant association with OS. Taken together, these findings highlight that the hub genes of PSN–OUA in AML identified through the network pharmacology framework are clinically relevant, further supporting the therapeutic potential of CGs.

Figure 8.

mRNA expression of the identified 10 hub targets in AML clinical samples. Expression levels were derived from the GEPIA database, which integrates data from TCGA and GTEx datasets. Data are presented as log₂(TPM + 1)-transformed expression values. *P < 0.01 versus normal tissue (LAML vs normal), determined by one-way ANOVA as implemented in GEPIA.

Figure 9.

Kaplan–Meier overall survival analysis of the identified 10 hub targets in AML patients according to their expression level. The plots generated using the Kaplan–Meier plotter illustrate the clinical relevance of these hub targets in predicting overall survival, showing the significant differences between high- and low- expression groups. The Log rank test was used to assess statistical significance, as indicated.

Abbreviations: HR, hazard ratio.

Molecular Docking and MD Simulations of PSN–OUA Hub Targets in AML

To test whether PSN and OUA could directly interact with the identified 10 PSN–OUA hub targets in AML, molecular docking using CB-Dock2 was performed to simulate the binding poses between PSN and OUA with the targets and estimate their binding affinities by means of Vina score. The resulting binding interactions between protein structures and compound conformations are illustrated in Figure 10A and B (see also Supplementary Table S3 for a complete list of corresponding contact residues). Docking analysis revealed that most drug–target interactions exhibited favorable binding energies (Vina scores < –7.0 kcal/mol), indicating strong binding activity at the predicted binding sites. Notable exceptions included ICAM1–PSN (–6.9 kcal/mol), ICAM1–OUA (–6.7 kcal/mol), and PRKCA–OUA (–6.4 kcal/mol). Interestingly, the top two proteins with the strongest binding affinities were MTOR and PTGS2, both showing Vina scores well below –10 kcal/mol for PSN and, to a lesser extent, OUA. The Vina scores for PSN–MTOR and PSN–PTGS2 were –12.8 kcal/mol and –11.2 kcal/mol, respectively, while the Vina scores for OUA with the same targets were –10.7 and –10.6 kcal/mol (Table 3). Collectively, these molecular docking results strongly support the potential interactions of PSN and OUA with the identified hub targets in AML.

Figure 10.

Molecular docking reveals the interactions between PSN, OUA, and their key hub proteins in AML. (A and B) Predicted interactions between PSN (A) and OUA (B) with the top-ranked hub proteins. Only the docking complexes with the highest predicted binding scores for each compound are shown. Docking was performed using CB-Dock2 in Auto Blind Docking mode. Ligands (PSN and OUA) are displayed in 3D conformations, whereas the interacting protein residues are represented in 2D (see also Supplementary Table S3 for detailed contact residues).

Table 3.

Molecular Docking Scores of PSN and OUA Against the Top 10 Hub Target Proteins, Analyzed Using the CB-Dock2 Blind Docking Approach

Target	Cur Pocket ID	Vina Score (kcal/mol)	Cavity Volume (Å3)	Center			Docking Size
				x	y	z	x	y	z
PSN
MTOR	C1	–12.8	4623	63	–36	–24	34	35	27
PTGS2	C2	–11.2	6009	30	34	218	27	35	27
ALB	C5	–9.8	1023	28	38	15	27	27	27
ICAM1	C1	–6.9	175	61	21	40	27	27	27
SYK	C2	–7.9	1696	–29	–13	2	35	27	27
PRKCA	C3	–7.8	150	31	7	11	27	27	27
MAPK14	C2	–9.2	898	–22	17	–12	27	27	27
MET	C2	–8.6	299	20	16	78	27	27	27
PRKCB	C4	–9.9	910	9	50	28	27	27	27
MAP2K1	C4	–8.4	241	22	9	4	27	27	27
OUA
MTOR	C2	–10.7	4244	–6	3	–79	31	35	25
PTGS2	C2	–10.6	6009	30	34	218	25	35	25
ALB	C2	–8.8	7417	25	33	25	35	32	35
ICAM1	C4	–6.7	111	56	39	21	25	25	25
SYK	C3	–7.4	912	–10	–3	–1	25	25	25
PRKCA	C1	–6.4	258	27	24	5	25	25	25
MAPK14	C1	–7.5	1837	–2	6	–16	25	25	31
MET	C2	–7.7	299	20	16	78	25	25	25
PRKCB	C1	–8.2	7059	25	44	26	35	35	35
MAP2K1	C4	–7.5	241	22	9	4	25	25	25

To further examine protein structural flexibility and protein–ligand stability of the docked complexes, MD simulations were performed for PSN in complex with the two top-ranked targets shared by PSN and OUA, MTOR and PTGS2, selected based on favorable docking energies (Vina scores). MD simulations were conducted for 100 ns, and the stability of the protein–compound complexes was assessed by analyzing the root mean square deviation (RMSD) of both the protein backbone and the bound compound. As shown in Figure 11A, the average backbone RMSD values were 5.95 ± 1.12 Å for PSN–MTOR and 2.42 ± 0.47 Å for PSN–PTGS2. Although the MTOR backbone exhibited relatively higher RMSD values, the trajectory reached a stable plateau without abrupt fluctuations or progressive drift, indicating preserved overall structural integrity. Notably, ligand RMSD analysis showed stable positional behavior of PSN throughout the simulation, with average RMSD values of 4.39 ± 0.50 Å for PSN–MTOR and 2.68 ± 0.31 Å for PSN–PTGS2 (Figure 11B), indicating overall stability of the predicted binding poses. In particular, the lower ligand RMSD observed for PSN–PTGS2 suggests a comparatively stable retention of PSN within the binding pocket. Collectively, these MD simulation results provide dynamic support for the docking predictions used in this study and motivated subsequent in vitro validation.

Figure 11.

Molecular dynamics (MD) simulations and in vitro validation of MTOR and PTGS2 modulation by PSN–OUA. (A and B) MD simulations of PSN in complex with the predicted targets MTOR (blue) and PTGS2 (red) over a 100-ns simulation period. RMSD of protein backbones (A) and RMSD of the bound PSN ligand (B). (C–E) In vitro validation of MTOR- and PTGS2-associated downstream signaling. KG-1 cells were treated with PSN (0–40 nM) or OUA (0–60 nM) for 48 h, and Western blot or RT-qPCR analysis was performed to experimentally validate the PSN and OUA effects. (C and D) Protein levels of multiple components of the mTOR pathway (C), including phospho-mTOR, total mTOR, Rictor, Raptor, G-β-L, and c-Myc, and COX-2 (encoded by PTGS2) (D) were evaluated by Western blotting. Densitometric quantification normalized to β-actin is shown. Data are presented as mean ± SD (N = 5). *P < 0.05, **P < 0.01, ***P < 0.001 versus NTX by one-way ANOVA with Dunnett’s multiple comparisons test. (E) RT-qPCR analysis of PTGES2 mRNA expression downstream of COX-2 was performed. GAPDH was used as the internal control. Data are presented as mean ± SD (N = 5). *P < 0.05, **P < 0.01, ***P < 0.001 versus NTX by one-way ANOVA with Dunnett’s multiple comparisons test.

In Vitro Validation of MTOR- and PTGS2-Associated Downstream Signaling

Having demonstrated that MTOR and PTGS2 (encoding cyclooxygenase-2, COX-2) were potential PSN–OUA targets, we next examined whether their downstream signaling pathways were functionally modulated in AML cells, as inhibitory compounds do not necessarily reduce target protein levels. Figure 11C shows that following PSN and OUA treatment for 48 h, multiple components of the mTOR pathway, including phospho-mTOR, Rictor, Raptor, and G-β-L, were markedly reduced, indicating suppression of mTOR signaling activity. In addition, c-Myc, a reported downstream effector of mTOR and a functional mediator of CG activity in AML cells,23,24,49 also showed a pronounced decrease. Notably, PSN exerted a stronger inhibitory effect than OUA, as reflected by a reduction in total mTOR protein levels and more robust suppression of downstream components, whereas OUA-mediated effects were more evident at higher concentrations. For COX-2, two immunoreactive bands were detected at approximately 74 and 66 kDa, consistent with previously reported COX-2 variants.50 The lower-molecular-weight ~66-kDa form, which was more prominent, exhibited a significant reduction upon PSN or OUA treatment, whereas the canonical ~74-kDa form showed weaker and less consistent changes (Figure 11D). Importantly, PTGES2 mRNA, encoding prostaglandin E synthase 2, which acts downstream of COX-2 and is involved in the conversion of prostaglandin H₂ (PGH₂) to prostaglandin E₂ (PGE₂), was significantly decreased (Figure 11E), further supporting functional suppression of the COX-2 axis. Taken together, these results indicate that PSN and OUA suppress MTOR- and PTGS2-associated signaling pathways, consistent with the computational predictions generated using the proposed in silico framework.

Discussion

CGs are natural steroid compounds found in certain plants and animals, traditionally used in the treatment of heart diseases. The idea of repurposing CGs for cancer therapeutics originally emerged from epidemiological evidence and has been further strengthened by recent advancements in technologies, such as in silico analysis and high-throughput drug screening. In AML, LSCs are at the apex of the leukemic hierarchy, where they self-renew and give rise to more differentiated AML cells, resembling the hierarchical organization of HSCs.51 AML relapse is largely attributed to minimal residual disease (MRD) harboring LSCs.52 Therefore, targeting LSCs is fundamental for eradication of AML. To gain a comprehensive understanding of how CGs may benefit AML treatment, we used PSN and OUA—two potent CGs with anti-LSC activity—as candidates to construct the network pharmacology framework presented herein.

Network pharmacology is an emerging approach in drug design that combines multiple disciplines, such as systems biology, bioinformatics, and network science, to predict drug-target interactions, disease-target relationships, drug-disease associations, or their combinations, with an emphasis on the multitarget effects of drugs for systematic optimization.53,54 To the best of our knowledge, the present study represents the first application of network pharmacology to identify putative targets of CGs in AML. Following our comprehensive framework, 10 highly connected hub targets of PSN–OUA, specifically relevant to AML clinical data, were eventually identified, including MTOR, PTGS2, ALB, ICAM1, SYK, PRKCA, MAPK14, MET, PRKCB, and MAP2K1. These predicted hub targets clearly differ from the classical cardiac target, Na⁺/K⁺-ATPase pumps,55 suggesting that CGs act through distinct molecular pathways in AML, unlike those in cardiac tissue. This notion highlights the context-dependent nature of compound–target interactions and underscores the utility of our approach in uncovering cancer pharmacological targets that extend beyond the conventional profiles of CGs. Molecular docking suggested the potential direct interactions between PSN and OUA with several hub targets, particularly MTOR and PTGS2, which exhibited strong predicated binding affinities (Vina score < –10 kcal/mol) for both compounds. Subsequently, MD simulations using PSN as a representative compound in complex with MTOR and PTGS2 further supported the stability of the predicted binding interactions. Importantly, we experimentally validated that both PSN and OUA suppressed MTOR- and PTGS2-associated downstream signaling in AML cells in vitro (Figure 11), providing additional support for the robustness of the docking protocol and substantiating the shared mechanisms underlying the potent effects of PSN and OUA in AML cells.

mTOR signaling, a key pathway identified herein, has previously been reported as a downstream target of OUA in glioma and as a downregulated pathway following PSN treatment in non-small cell lung cancer (NSCLC).56,57 In AML, the PI3K/Akt/mTOR pathway, which regulates leukemic cell survival and proliferation,58 has been extensively studied and has emerged as a potential therapeutic target. Constitutive hyperactivation of this pathway is detected in 50–80% of AML patients, and is associated with decreased OS.58,59 PI3K/Akt/mTOR and its upstream or downstream inhibitors may preferentially target LSCs.60 For example, SYK inhibitors attenuate the activation of both mTOR and MAPK pathways by reducing phosphorylation of MEK and ERK, leading to decreased viability, enhanced myeloid differentiation, and inhibition of LSCs in both in vitro and in vivo models.61,62 Consistent with this, we identified MTOR, SYK, and MAP2K1 (encoding MEK1) as hub targets that are functionally interconnected, thus representing actionable target nodes in AML. The combination of mTOR and MEK inhibitors has been proposed and tested in clinical trials alongside conventional chemotherapy, demonstrating favorable outcomes in the early-phase trials.63–65 However, further evaluation is warranted to determine their long-term clinical benefits in diverse AML subsets. By contrast, PTGS2 has limited reports regarding its roles in AML, except for its consistent overexpression in AML patients, which is associated with poor OS.66 PTGS2, through PGE₂, along with VEGF-C may play a key role in angiogenesis, which is crucial for AML progression.66 Given the limited available information, PTGS2 may serve as a potential prognostic biomarker in AML; however, further studies are required to investigate its functional role in LSCs.

Beyond MTOR and PTGS2, several additional targets identified in this analysis have been linked to AML biology. A previous study has recently reported that inhibition of p38 MAPK, encoded by MAPK14, using doramapimod provides therapeutic benefits, particularly in monocytic AML characterized by CD14 expression.67 The combination of doramapimod with the BCL2 inhibitor venetoclax showed synergistic efficacy by expanding coverage across heterogeneous AML populations.67 Overexpression of classical PKC isoforms α and β, encoded by PRKCA and PRKCB, were observed in AML blasts.68 Following this, the PRKCB inhibitor enzastaurin has been shown to induce apoptosis in AML cell lines and primary AML patient-derived cells, suggesting its potential anti-leukemic activity, although this effect may not be primarily dependent on PRKCB inhibition.68 Interestingly, ICAM1, also known as CD54, is a membrane-bound glycoprotein of immunological significance in AML, promoting T and NK cell–mediated cytotoxicity upon upregulation.69,70 Inhibition of ICAM1 impairs the immune response, underscoring its prognostic value. In agreement with this, our analysis revealed that low ICAM1 expression correlates with inferior patient outcome in AML.

Emphasizing the multitarget effects of CGs, we next extended our analysis to a broader functional perspective by exploring the signaling networks and pathways implicated in the aggressive phenotypes of AML cells using functional annotation and enrichment analyses. To this end, the identified 17 core targets of PSN–OUA in AML, rather than the 10 hub targets, were utilized to capture the overarching mechanisms. Because apoptosis is a fundamental biological process that controls tumor growth and progression71 and serves a major mechanism underlying the majority of drug responses,72 including CGs,19,73 it is not surprising that “apoptotic process” emerged as the top GO BP term according to the P-value. Seven targets were mapped, including the three high-degree hubs identified in the PPI network, namely MAPK14, PRKCA, and PRKCB, highlighting their central role as a mechanistic focal point. “Protein kinase C signaling”, involving PRKCA and PRKCB, showed the highest fold enrichment among all GO BP terms, and its components are clinically established targets in AML.68,74 We herein validated the potent apoptotic effects of PSN and OUA in AML cells, demonstrating a dose-dependent response with IC₅₀ values in the nanomolar range. In parallel, KEGG pathway analysis interestingly pinpointed the “VEGF signaling pathway” and “EGFR tyrosine kinase inhibitor resistance”. Closer dissection of the two pathways revealed a connection to cellular processes involved in, but not limited to, proliferation and motility (Supplementary Figure S1A and B). Key signaling axes included PRC/MEK, c-MET/mTOR, c-MET/PKC/MEK, and p38 MAPK. We also mapped the identified core targets of PSN–OUA in AML to the KEGG AML pathway (KEGG ID: hsa05221) and found that both mTOR and MEK were linked to gene modules associated with anti-apoptotic signaling and proliferative control (Supplementary Figure S1C), suggesting that CGs may exert their pharmacological effects by controlling AML growth. These in silico predictions aligned with multiple complementary phenotypic analyses demonstrating that PSN and OUA suppress AML cell growth and induce cell death (Figure 5; Supplementary Figures S2 and S3), highlighting coordinated regulation of apoptosis and proliferation as central features of PSN–OUA activity. Additionally, Transwell migration and invasion assays confirmed the inhibitory effects of CGs on AML cell motility, consistent with the predicted involvement of metastasis-related processes highlighted by CancerGeneNet analysis. Beyond this, the network connected PRKCA, MAPK14, mTOR, MET, and PTGS2, which are hub targets, to signaling pathways involved in metastasis, inflammation, and differentiation. Overall, several of our identified core targets, along with their closely connected molecular components within the interaction network, overlapped with AML-specific cancer modules (Supplementary Table S4), underscoring their integrated relevance to AML therapeutic strategies.

Conclusion

In conclusion, our integrative in silico framework combining network pharmacology, bioinformatics, and molecular docking–MD simulation approaches, together with downstream in vitro validation, offers a powerful and reliable approach to explore the therapeutic potential of CGs in a drug-repurposing context. This framework emphasizes on the multitarget effects of CGs, an emerging strategy that is considered more efficient for tackling complex diseases like cancer. We identified key clinically relevant hub targets and pathways involved in AML development and progression that are commonly affected by both PSN and OUA, reported as potent anticancer CGs. Given that PSN and OUA exhibit potent anti-LSC activity, the identified targets and pathways also provide insights into the mechanisms underlying LSC regulation, which could be crucial for the design of novel therapeutic strategies for AML, including the development or optimization of CG analogs with improved pharmacokinetic properties. Further studies are warranted to verify direct biochemical target engagement and to clarify the functional roles of these targets in LSCs and AML blasts, as well as the pharmacokinetics, efficacy, and safety of CGs and their optimized analogs in vivo. The proposed framework is broadly applicable to other drugs, particularly those undergoing repurposing strategy, and provides a rational basis for prioritizing compounds with favorable molecular targeting profiles, with the potential to improve clinical translation, while reducing research and development time and costs.

Funding Statement

This research was funded by Siriraj Foundation for Stem Cell Research (D003276).

Abbreviations

AML, acute myeloid leukemia; BP, biological process; CC, cellular component; CCK-8, Cell Counting Kit-8; CGs, cardiac glycosides; CIDs, compound identifiers; COX-2, cyclooxygenase-2; DAVID, Database for Annotation, Visualization, and Integrated Discovery; GO, Gene Ontology; GTEx, Genotype-Tissue Expression; HR, hazard ratio; HSCs, hematopoietic stem cells; IC₅₀, half-maximal inhibitory concentration; JCRB, Japanese Collection of Research Bioresources; KEGG, Kyoto Encyclopedia of Genes and Genomes; LSCs, leukemic stem cells; MD, molecular dynamics; MF, molecular function; MRD, minimal residual disease; NSCLC, non-small cell lung cancer; NTX, non-treated; OS, overall survival; OUA, ouabain; PDB, Protein Data Bank; PGE₂, prostaglandin E₂; PPI, protein–protein interaction; PSN, proscillaridin; RMSD, root mean square deviation; TCGA, The Cancer Genome Atlas.

Data Sharing Statement

The data analyzed in the current study were obtained from publicly available datasets and are provided in the Supplementary Data or available from the corresponding author upon reasonable request.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas. S.L. supervised the project. The original draft of the article was written by P.S., and all authors revised or critically reviewed the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.