In silico Identification of Key Genes and Repurposable Drugs for Triple-Negative Breast Cancer

Abstract

Triple-negative breast cancer (TNBC) is a highly aggressive and treatment-resistant subtype, lacking HER2, progesterone, and estrogen receptors. Systemic chemotherapy remains the primary treatment due to the absence of molecular targets, often leading to poor prognosis and high recurrence. This study used an integrated in silico transcriptomic and network-based bioinformatics approach to identify TNBC key genes (TKGs) and explore potential therapeutic candidates. Four TNBC microarray datasets (GSE36295, GSE38959, GSE45827, and GSE65194) were analyzed with the LIMMA algorithm in GEO2R, revealing 315 TNBC-shared differentially expressed genes (TSDEGs). Eight strongly connected TKGs were identified via topological analysis and protein-protein interaction (PPI) network construction across the STRING and IMEx databases from TSDEGs: CDK1, TOP2A, CCNB1, CDK2, FN1, UBC, PRKDC, and PARP1. Enrichment analysis of biological processes, molecular functions, cellular components, and KEGG pathways, combined with regulatory network analysis involving transcription factors and microRNAs, identified pathogenic roles of these genes. CDK1 and CDK2 were identified as the top pharmacological targets for molecular docking studies. Finally, TKG-guided top-ranked two drug molecules (Alsterpaullone and BLU-222) emerged as promising repurposed therapeutic candidates for TNBC. The absorption, distribution, metabolism, excretion, and toxicity (ADMET) and drug-likeness analyses showed these molecules have favorable pharmacokinetic properties. Molecular dynamics simulations over 100 nanoseconds demonstrated stable binding interactions and favorable behavior for the complexes CDK1-Alsterpaullone and CDK2-BLU-222, as indicated by root mean square deviation, fluctuation, and molecular interactions generalized Born surface area. Overall, these findings may aid in diagnosing and treating TNBC, providing valuable resources for future therapeutic strategies.

Introduction

Triple-negative breast cancer (TNBC) represents a severe oncological challenge, characterized by the absence of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) expression. ¹ Approximately 15-20% of global breast cancer cases are classified as this highly aggressive subtype, featuring early recurrence and a low overall survival rate.^1-7 TNBC is more common in younger women and those who carry the BRCA1 mutation. ⁸ The clinical signs of TNBC resemble typical breast cancer symptoms, including lump development, alterations in breast size or shape, recurrent rashes, pain, and discomfort. Its presentation as a moderate-to-high-grade malignancy with high proliferative activity further underscores the aggressive nature of TNBC. ⁹ The five-year survival rate for metastatic TNBC is below 12%, indicating a poor prognosis once the illness progresses beyond its initial stages. In early-stage TNBC, prompt treatment can enhance five-year survival rates to about 50-60%; however, this is still inadequate compared to those of hormone receptor-positive breast tumors. Globally, TNBC constitutes a significant percentage of breast cancer fatalities and a pronounced mortality burden, particularly in Asia.^10-15

TNBC is often classified as a high-grade tumor, correlating with elevated odds of relapse and mortality. In contrast to other breast cancer subtypes that include targets such as ER or HER2, there are currently no approved targeted therapies for TNBC. Consequently, systemic chemotherapy remains the primary form of treatment. The responses to chemotherapy are often transient, rendering this disease subtype a therapeutic challenge that necessitates the development of novel and more effective targeted medicines or combination treatments. ¹⁶ Despite an initial apparent response to chemotherapy, the overall results of conventional chemotherapy are still poor because of the high recurrence rates, increased mortality, and rapid emergence of resistance.

Current research efforts in TNBC are focused on unraveling the molecular heterogeneity of the disease to identify actionable targets and predictive biomarkers. Genomic and transcriptomic profiling studies have classified TNBC into distinct molecular subtypes, each with unique therapeutic vulnerabilities. ¹⁷ Moreover, there is growing interest in drug repurposing as a strategy to expedite the availability of effective TNBC treatments. ¹⁸ Despite these evident advances, the clinical translation of targeted therapies remains limited, and the heterogeneity of TNBC at the clinical, histological, and molecular levels continues to pose significant challenges for prognosis and treatment selection. ¹⁹ Therefore, there is a pressing need to identify robust biomarkers to predict outcomes and guide personalized therapy, as well as to discover novel therapeutic targets. A strong basis for logical drug repurposing initiatives is provided by the thorough identification of TNBC key genes (TKGs) and implicated dysfunctional pathways through in silico analysis. To identify TKGs, molecular pathways linked to TNBC tumorigenesis, and repurposable drug candidates, this study employed a thorough in silico analysis, as bioinformatics approaches are promising for reducing experimental load in the wet lab.

Methods

Data Acquisition and Determination of Differentially Expressed Genes

To explore repurposable candidate drugs for TNBC, it is required to identify genes that are differentially expressed in this condition. Transcriptomics profile analysis using bioinformatics tools is a popular approach for identifying disease-causing differentially expressed genes (DEGs).^20-24

To identify TNBC-shared DEGs (TSDEGs) compared to healthy control cells, the publicly available transcriptomic data repository at NCBI, the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/), was used. GEO provides access to various genomic datasets. For consistency, we selected case/control microarray datasets suitable for analysis with GEO2R (an integrated platform within GEO that performs differential gene expression analysis). GEO2R employs the R programming language and the linear models for microarray (LIMMA) statistical package, applying empirical Bayes statistics to identify genes with differentially expression between distinct patient groups. Inclusion criteria for dataset selection were: human-derived samples, data type defined as expression profiling by microarray, the presence of TNBC patient samples and healthy control samples. The analysis yielded p-values and log fold change (log FC) values, which facilitated the evaluation of statistical significance and the extent of expression differences, respectively. Significant TSDEGs were identified for each dataset by computing adjusted p-values and log2 fold-change (log2FC) values. Threshold values of adjusted p < 0.05 and log2FC > 1 were used to identify upregulated TSDEGs, while adjusted p < 0.05 and log2FC < −1 were considered to identify downregulated TSDEGs. For each dataset, quality control plots, such as volcano, mean-difference, and mean-variance trend plots, were used to assess data quality (Supplementary Figures 1, 2, and 3). InteractiVenn (https://www.interactivenn.net/) was used to visualize the TSDEGs across datasets, including upregulated and downregulated genes. ²⁵

After careful review, four case-control datasets were selected with the following accession IDs: GSE36295,^26,27 GSE38959, ²⁸ GSE45827, ²⁹ and GSE65194.^10-12 The details of the datasets are provided in Table 1.

Table 1.

Overview of Transcriptomics Datasets Analyzed

Accession IDs for datasets	Disease name	Platform	Case	Control
GSE36295	Triple Negative Breast Cancer	GPL6244 [HuGene-1_0-st] Affymetrix Human Gene 1.0 ST Array [transcript (gene) version]	11	5
GSE38959	Triple Negative Breast Cancer	GPL4133 Agilent-014850 Whole Human Genome Microarray 4x44K G4112F (Feature Number version)	30	13
GSE45827	Breast Cancer Basal Type (TNBC)	GPL570 [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array	41	11
GSE65194	Triple Negative Breast Cancer	GPL570 [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array	55	11

Identification of Key Genes

We initially constructed protein-protein interaction (PPI) networks for the TSDEG set across two independent repositories, STRING ³⁰ and IMEx, ³¹ employing the NetworkAnalyst platform (https://www.networkanalyst.ca/). ³² Network visualization was subsequently performed using Cytoscape software. ³³ Within these mapped interactomes, candidate TKGs were screened via the CytoHubba plugin. ³⁴ Node prioritization across both database networks was evaluated using six distinct graph topology algorithms: degree, closeness, stress, betweenness, maximum neighborhood component (MNC), and maximal clique centrality (MCC).

Disclosing Common Pathogenic Pathways in TNBC

To elucidate the pathogenetic contributions of TKGs in TNBC, functional profiling was performed across biological processes (BPs), cellular components (CCs), molecular functions (MFs), signaling pathways, and regulatory networks. Functional enrichment of the prioritized TKG set, covering both Gene Ontology (GO) categories and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, was executed using two complementary platforms: DAVID (https://david.ncifcrf.gov/tools.jsp) ³⁵ and Enrichr (https://maayanlab.cloud/Enrichr/). ³⁶ Statistical evaluation across both tools was determined via Fisher’s exact test, using p < 0.05 as the significance threshold.

Analysis of Regulatory Factors in TNBC

An analysis of the regulatory network was conducted using transcription factors (TFs) and microRNAs (miRNAs) to identify the primary regulators of TKGs. To identify TFs associated with TKGs, the TF-TKG network was analyzed using the JASPAR ³⁷ database. Through the analysis of the associations between miRNAs and TKGs utilizing the TarBase ³⁸ databases, it was possible to identify the significant miRNAs that modulate TKGs at the post-transcriptional stage. NetworkAnalyst (https://www.networkanalyst.ca/) ³² was employed for modeling these interactions. The post-transcriptional regulators of TKGs were chosen from the highest-ranked miRNAs. Cytoscape ³³ was used to visualize the networks of their interactions.

Identification of Candidate Drug Molecules

To identify possible therapeutic compounds targeting TKGs, prospective repurposable drugs were sourced from three principal repositories: the DrugBank (https://go.drugbank.com/) database, ³⁹ the DGIdb (https://dgidb.org/) database, ⁴⁰ and other relevant published literature. A gene-guided drug repurposing technique was utilized to identify candidate medicines with potential efficacy against TNBC. This involved molecular docking analysis targeting the top TKGs, followed by thorough evaluations of drug-likeness, pharmacokinetics, toxicity, and molecular dynamics (MD) simulations to assess the therapeutic potential and safety of the chosen compounds.

Molecular Docking Analysis

We performed in silico molecular docking to assess and evaluate potential therapeutic agents for TNBC. Docking was employed to predict the binding affinities of candidate drug molecules with target proteins, specifically receptors and transcription factors associated with TKGs. Three-dimensional structures of receptor proteins were collected from the Protein Data Bank (PDB)(https://www.rcsb.org), ⁴¹ while chemical structures of drug candidates were sourced from PubChem (https://pubchem.ncbi.nlm.nih.gov/). ⁴² Receptors were prepared using AutoDock Tools ⁴³ coupled with Swiss-PdbViewer, ⁴⁴ and ligand structures were energy-minimized and optimized using Avogadro. ⁴⁵ Molecular docking was carried out using AutoDock Vina via PyRx software,^46,47 and binding affinities were calculated for each receptor-ligand complex. The docked complexes were visualized in PyMOL, and their interactions were analyzed in Discovery Studio Visualizer. ⁴⁸ To identify the most promising drug candidates for further validation against TNBC, the resulting interactions were ranked by binding affinities.

Pharmacokinetics, Toxicity, and Drug-likeness Properties Analysis

The pharmacokinetics, toxicity, and drug-likeness profiles of the selected compounds were evaluated using a combination of web-based tools. The absorption, distribution, metabolism, excretion, and toxicity (ADMET) features were predicted using ADMETlab 3.0 (https://admetlab3.scbdd.com/) ⁴⁹ and SwissADME (https://www.swissadme.ch/), ⁵⁰ while ProTox 3.0 (https://tox.charite.de/protox3/) ⁵¹ and pkCSM (https://biosig.lab.uq.edu.au/pkcsm/) ⁵² were used to assess toxicity-related parameters. Drug-likeness based on Lipinski’s Rule of Five ⁵³ and physicochemical properties, such as molecular weight, LogP, hydrogen bonding, and synthetic accessibility, were also evaluated using these tools.

Molecular Dynamics Simulations

To account for protein flexibility, conformational adaptations, and solvent dynamics that static molecular docking cannot predict, 100 ns MD simulations were conducted to evaluate the physical stability of the top protein-ligand complexes over time. GROMACS version 2024 ⁵⁴ was used to perform MD simulations of protein-ligand complexes for 100 nanoseconds (ns). Simulations were performed with the CHARMM36m force field. The TIP3P water model was utilized to create a water box with edges 1 nanometer (nm) from the protein surface. The necessary ions were employed to neutralize the systems. After energy minimization, isothermal-isochoric (NVT) equilibration, and isobaric (NPT) equilibration of the system, 100 ns MD simulations were conducted with periodic boundary conditions and a time integration step of 2 femtoseconds (fs). The trajectory data was analyzed with a snapshot interval of 100 picoseconds (ps). The root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), and solvent accessible surface area (SASA) analyses were conducted using the rmsd, rmsf, gyrate, and sasa modules incorporated within the GROMACS software after the simulation was completed. The graphs for each of these analyses were generated using the ggplot2 package in RStudio.

Figure 1 illustrates the entire workflow of this study.

Figure 1.

Workflow of the bioinformatics analysis in the study

Results

Differentially Expressed Genes Mediated Data Processing

The TSDEGs between TNBC and control samples were identified using the statistical LIMMA method across four datasets. A total of 315 TSDEGs were identified, comprising 206 upregulated and 109 downregulated genes (Figure 2 and Supplementary Table 1). These were derived from counts of individual DEGs across datasets GSE36295, GSE45827, GSE38959, and GSE65194, which reported 491, 4709, 2475, and 4730 upregulated genes as well as 484, 2397, 1125, and 2587 downregulated genes, respectively.

Figure 2.

Identification of the TSDEGs. Venn diagram showing the (A) upregulated and the (B) downregulated DEGs shared across four datasets. (C) TSDEGs, represents the combined set of commonly upregulated and downregulated genes identified across all datasets

Identification of Key Genes

To isolate core TKGs, candidate TSDEGs were mapped onto two separate PPI interactomes derived from the IMEx and STRING databases. We identified highly connected genes critical for maintaining network integrity and functional crosstalk by applying six distinct topological algorithms: degree, closeness, stress, betweenness, MNC, and MCC. Within the STRING-derived interactome, CDK1, TOP2A, and CCNB1 emerged as the three most prominent hub TSDEGs based on overall connection density.

Within the IMEx-supported PPI network, five of the highest-ranking TSDEGs were identified: CDK2, FN1, UBC, PRKDC, and PARP1. So, the eight TSDEGs: CDK1, TOP2A, CCNB1, CDK2, FN1, UBC, PRKDC, and PARP1 were termed TKGs because they were consistently identified across all six topological analyses in both databases. (Figure 3).

Figure 3.

PPI network of TKGs compiled from (A) STRING and (B) IMEx databases. The nodes represent top 8 ranked TKGs identified from TNBC and control samples across both databases

Exploring Common Pathogenic Pathways

Functional profiling across the eight prioritized TKGs using GO annotations and KEGG repositories revealed key biological mechanisms involved in TNBC pathogenesis. Primary enrichment analysis was conducted via the Enrichr server and subsequently cross-validated using DAVID to identify overlapping, statistically significant terms. This dual-platform evaluation yielded seventeen significant BP terms (the top five are presented in Table 2), three CC terms, five MF terms, and six KEGG pathways. Comprehensive summaries of these enriched terms appear in Table 2 and Supplementary Table 2.

Table 2.

Significantly Enriched GO Terms and KEGG Pathways With TKGs in Enrichr and DAVID

Annotation ID	BP term	Associated hKGs
GO:0006974	DNA damage response	CDK1, CDK2, PARP1, PRKDC, TOP2A
GO:0000086	G2/M transition of mitotic cell cycle	CCNB1, CDK1, CDK2
GO:0031571	Mitotic G1 DNA damage checkpoint signaling	CDK2, PRKDC
GO:0160049	Negative regulation of cGAS/STING signaling pathway	PARP1, PRKDC
GO:0018105	Peptidyl-serine phosphorylation	CDK1, CDK2, PRKDC

Analysis of Key Regulators for TNBC Key Genes

To determine the key regulatory elements of the TKGs, we analyzed both transcriptional and post-transcriptional regulators. Among the TFs, the top five candidates, GATA2, YY1, FOXC1, NFIC, and NFYA, were identified based on their high-ranking positions in the regulatory network (Figure 4A). Similarly, for post-transcriptional regulation: hsa-let-7a-5p, hsa-let-7b-5p, hsa-let-7c-5p, hsa-miR-15a-5p, and hsa-miR-20a-5p emerged as the top-ranking miRNAs interacting with TKGs (Figure 4B). These regulators were selected based on their topological importance in the interaction networks and are likely to play significant roles in modulating TKGs expression. Their details are given in Tables 3 and 4.

Figure 4.

Regulatory interaction networks illustrating cross-talk between target genes and upstream regulators. (A) Interaction network mapping miRNAs to TKGs. (B) Interaction network mapping TFs to TKGs. Central TKGs in both subfigures are depicted as enlarged octagonal nodes. The prioritized top-ranked miRNAs in (A) are rendered as five quadrilateral nodes, whereas five hexagonal nodes denote the top-ranked TFs in (B)

Table 3.

Details of Top Five TFs

Rank	Name	Degree	Betweeness
1	GATA2	5	140.6151
2	YY1	4	58.0791
3	FOXC1	3	28.40852
4	NFIC	3	33.57915
5	NFYA	3	35.52458

Table 4.

Details of Top Five miRNAs

Rank	Name	Degree	Betweeness
1	hsa-let-7a-5p	8	531.0367
2	hsa-let-7b-5p	8	531.0367
3	hsa-let-7c-5p	8	531.0367
4	hsa-miR-15a-5p	8	531.0367
5	hsa-miR-20a-5p	8	531.0367

Molecular Docking Analysis

To identify potential candidate drug molecules for TNBC, CDK1 and CDK2 were selected as target receptors based on their topological prominence in PPI networks. CDK1 ranked highest across six network topology parameters in the STRING database, while CDK2 ranked highest among six topological measures in the IMEx database. We obtained the 3D structures for both proteins from the PDB database: 6GU6 for CDK1 and 1JVP for CDK2. Molecular docking analysis was used to assess the binding affinity scores by comparing the selected proteins with the identified drugs. These scores represent the strength of interaction between a ligand and its target protein, with lower values indicating stronger and more stable binding. Such scores are critical in drug discovery, as they help predict a compound’s potential efficacy in modulating its target’s biological activity. To shortlist effective drug candidates targeting CDK1 and CDK2, an in-depth literature review and a targeted search of the DrugBank and DGIdb databases were conducted. A group of compounds with known or predicted activity against these proteins was selected and prepared for docking analysis (Supplementary Table 3). Subsequently, molecular docking was performed using AutoDock Vina to predict binding interactions between the selected drugs and their respective protein targets. Based on significant binding affinity, the top-performing drugs for each kinase were identified. BLU-222 showed the strongest binding affinity (-8.8 kcal/mol) for CDK2 and was chosen as the lead candidate (Figure 5). This aligns with recent findings, which described BLU-222 as a selective CDK2 inhibitor with promising anticancer potential in preclinical models.^55-57 Similarly, Alsterpaullone ranked highest (-7.5 kcal/mol) for CDK1 (Figure 6), consistent with studies reporting its ability to block CDK1 activity and trigger apoptosis in cancer cells.^58,59 These two drugs, BLU-222 for CDK2 and Alsterpaullone for CDK1, were selected as the top hits for further validation. The binding affinities of both proteins for their respective drugs are given in Tables 5 and 6.

Figure 5.

Molecular docking analysis illustrating the interaction between the CDK2 protein and the BLU-222 ligand. (A) Surface depiction of the protein-ligand interaction. Binding residues were depicted in blue surrounding the ligand (green). (B) Ligand-binding pocket depicted in an enhanced 3D model. (C) A two-dimensional image depicting the interactions between the receptor and substrate via a conventional hydrogen bond and other bonding types

Figure 6.

Molecular docking analysis demonstrating the CDK1 protein with Alsterpaullone ligand. (A) Surface representation of the protein-ligand interaction. Binding residues were represented by the blue color that surrounded the ligand (green). (B) Ligand-binding pocket depicted in an enhanced 3D illustration. (C) A 2D image illustrating the interactions between the receptor and substrate through a conventional hydrogen bond and other bonds

Table 5.

Binding Affinity of Selected Drugs With CDK1

Drugs	Binding affinity scores (kcal/mol)
Alsterpaullone	-7.5
Fostamatinib	-7.0
Dinacicilib	-6.8
Flavopiridol	-6.7
Roscovitin	-6.4

Table 6.

Binding Affinity of Selected Drugs With CDK2

Drugs	Binding affinity scores (kcal/mol)
BLU-222	-8.8
Alsterpaullone	-8.2
Dinacicilib	-8.1
Roscovitin	-7.8
Flavopiridol	-6.8

Assessment of Pharmacokinetics and Toxicity Analysis of Drugs

ADMET parameters serve as critical benchmarks for evaluating candidate molecules during drug development, with drug-likeness profiling assessing essential physicochemical characteristics. High gastrointestinal absorption represents a fundamental requirement for orally administered agents, typically defined by a human intestinal absorption (HIA) score exceeding 30%. ⁶⁰ Both candidate compounds investigated in this study demonstrated high oral bioavailability efficiency, with exceptional HIA values of 91% or greater.

The Caco-2 permeability values further confirm gastrointestinal absorption levels; both drugs showed adequate Caco-2 permeability, as per the pkCSM model, with compounds with log Papp > 0.90 classified as having high Caco-2 permeability; values below this indicate moderate-to-low permeability. ⁵² Blood-brain barrier (BBB) permeability assessment shows that compounds with LogBB ≥ 0.3 can effectively cross the barrier, whereas values ≤ -1 indicate poor penetration. ⁶¹ Both the proposed drugs had LogBB <0.3, suggesting limited to poor BBB access. For volume of distribution, 0.04-20 L/Kg indicates optimal plasma distribution, and both our chosen drugs fall under this threshold. ⁶² BLU-222 showed no inhibitory activity against key cytochrome P450 enzymes, suggesting a low potential for drug interactions mediated by CYP inhibition, ⁶³ while Alsterpaullone showed selective CYP interaction, suggesting potential for drug repurposing, as cytochrome P450 is crucial for drug metabolism and interacts with many drugs. ⁶⁴ The total clearance values slightly exceed the typical 1-5 L/h range, suggesting relatively rapid elimination that may require dosing adjustments to maintain efficacy. ⁶⁵ Toxicity analyses, including the Ames test, indicate that BLU-222 does not exhibit mutagenicity or skin sensitization. Alsterpaullone showed Ames mutagenicity but no skin sensitization. Nevertheless, its potent kinase inhibition and therapeutic promise justify further investigation, with potential structural modifications to mitigate risks. ⁵⁹ Physicochemical screening confirmed that both candidate molecules possessed favorable pharmacokinetic profiles, fully adhering to Lipinski’s Ro5 without any violations. Furthermore, both compounds satisfied the specific oral bioavailability criteria established by Egan’s and Veber’s filtering rules. Both BLU-222 and Alsterpaullone support their potential for drug repurposing. The details are in Table 7.

Table 7.

ADMET Analysis of Selected Candidate Drug Molecules Using Different Web Tools

Properties	Alsterpaullone	BLU-222
Absorption
HIA (%)	91.972	91.67
LogS (ESOL)	-3.60	-3.46
Caco2	0.904	1.025
Distribution
BBB	0.067	-1.788
VDss	1.001	0.849
Metabolism (CYP450)
CYP1A2 Inhibitor	Yes	No
CYP2C19 inhibitor	No	No
CYP2C9 inhibitor	No	No
CYP2D6 inhibitor	Yes	No
Toxicity
TC	7.52	8.86
AMES Toxicity	Yes	No
Skin sensitization	No	No
Carcinogenicity	Inactive	Inactive

MD Simulations

RMSD

To assess system stability, the RMSD calculation was performed. A change in the RMSD value indicates conformational changes in proteins. Figure 7 illustrates the RMSD profile of the protein complex CDK1-Alsterpaullone (green) and CDK2-BLU-222 (purple). An initial steep increase in RMSD over the first 10 ns was observed in both complexes, suggesting initial conformational changes. Subsequently, both complexes attained a stable RMSD plateau within the range of 0.25-0.3 nm within 20 ns. The CDK1-Alsterpaullone complex exhibited consistent stability, with only a minor, transient increase in RMSD observed after 75 ns.

Figure 7.

CDK1-Alsterpaullone and CDK2-BLU-222 protein-ligand complexes backbone RMSD during a 100 ns MD simulation. The stabilization patterns of each complex are reflected in the temporal evolution of structural deviations from the initial conformations, as illustrated in the plot

RMSF

RMSF is used to assess a protein’s regional flexibility. The flexibility of a specific amino acid position increases as the RMSF increases. The RMSF profiles of the complexes CDK1-Alsterpaullone (brick red) and CDK2-BLU-222 (blue) are shown in Figure 8. Both complexes exhibited an overall RMSF profile spanning 0.1-0.5 nm. In the CDK-Alsterpaullone complex, the residues exhibited multiple peaks, with the majority of them having RMSF values below 0.3 nm. In contrast, the CDK2-BLU-222 complex had residues with comparatively lower RMSF amplitudes, below 0.25 nm, except at the C-terminal end, where fluctuations could reach 0.5 nm.

Figure 8.

The RMSF profiles by residue for the CDK1-Alsterpaullone and CDK2-BLU-222 complexes. The graph illustrates the dynamic behavior and flexibility of protein residues along the sequence as they are influenced by ligand binding throughout the simulation

Rg

The Rg determines the degree of compactness. A protein’s folding is stable when the Rg is relatively constant. The protein’s unfolding is inferred from fluctuations in the Rg. Figure 9 illustrates the Rg profile of the complex CDK1-Alsterpaullone (green) and CDK2-BLU-222 (red). The Rg profile of both complexes remained consistent throughout the simulation.

Figure 9.

CDK1-Alsterpaullone and CDK2-BLU-222 complexes Rg as a function of time during the 100 ns MD simulation. The trajectories demonstrate the structural stability and overall compactness of each protein-ligand complex

SASA

SASA is used in MD simulations to predict the exposure of proteins’ hydrophobic cores to solvent. Higher SASA values suggest that a significant portion of the protein is exposed to water. In contrast, lower values indicate that a substantial portion of the protein is concealed within the hydrophobic core. Figure 10 displays the SASA profile of the complex CDK1-Alsterpaullone (yellow) and CDK2-BLU-222 (blue). The SASA value remained relatively consistent throughout the simulation.

Figure 10.

The SASA profiles for the CDK1-Alsterpaullone and CDK2-BLU-222 complexes were calculated during the MD simulation. By indicating the degree of solvent exposure for each complex, the profiles emphasize the burial of the hydrophobic core and the structural integrity of the complex

Hydrogen Bond

To assess intermolecular interactions between the protein and ligands during the simulation, hydrogen-bond analysis was implemented. The quantity of hydrogen bonds determines the strength and stability of protein-ligand binding interactions. Figure 11 displays the hydrogen bond profile of the complex CDK1-Alsterpaullone (orange) and CDK2-BLU-222 (blue). Throughout the simulation, both complexes exhibited dynamic hydrogen bonding patterns, with bond numbers ranging from 0 to 5. During the simulation, the CDK1-Alsterpaullone complex showed 2-5 hydrogen bonds. In the initial stages of the simulation, the CDK2-BLU-222 complex formed 6-7 hydrogen bonds, with occasional peaks exceeding 4, suggesting the presence of additional stabilizing bonds. However, the number of H-bonds decreased and primarily fluctuated between 0 and 2, indicating that there were fewer interactions in the later stages.

Figure 11.

Analysis of the protein-ligand hydrogen bond for the CDK1-Alsterpaullone and CDK2-BLU-222 complexes during the 100 ns MD simulation session. The number of hydrogen bonds formed at each time point indicates the stability and strength of intermolecular interactions and binding dynamics over time

Discussion

TNBC is a highly aggressive subtype of breast cancer, marked by significant intra-tumoral heterogeneity with a tendency to develop resistance to treatments. Additionally, there are currently no approved targeted treatments for this rapidly proliferating subtype. ⁶⁶ Transcriptomic analyses have proven to be a powerful tool in studying various diseases, including cancer. ⁶⁷ This work utilized advanced in silico methods to identify dysregulated key genes and pathways in TNBC. An integrated analysis of four distinct transcriptome datasets of TNBC (GSE36295, GSE38959, GSE45827, GSE65194) identified 315 TSDEGs that distinguished TNBC tissues from healthy controls. Analyzing the resulting protein-protein interaction networks from the STRING and IMEx databases helped refine this long list of genes into a shortlist of potential candidates. Six different algorithms for the topological analysis pinpointed eight TKGs: CDK1, TOP2A, CCNB1, CDK2, FN1, UBC, PRKDC, and PARP1.

Among our identified TKGs, CDK1 is known to be essential for mitosis, coordinating spindle formation and chromosome alignment by activating key proteins involved in kinetochore assembly. ⁶⁸ In breast cancer, including TNBC, overexpression of CDK1 is connected with tumor cell proliferation, migration, and survival. On the other hand, its inhibition initiates cell cycle arrest and apoptosis. In TNBC, CDK1 has been shown to support tumor growth by stabilizing critical proteins and driving therapy-resistant pathways. Targeting CDK1 has shown potential to suppress breast cancer progression, which further highlights its importance as a therapeutic target.^69-72 CDK2 also plays a significant role in breast cancer by regulating cell cycle progression through the cyclin E/CDK2 complex. ⁷³ This specific complex is often overexpressed in TNBC, contributing to tumor cell proliferation and aggressiveness. CDK2 phosphorylates key substrates such as EZH2, which promotes malignancy and tumor growth. CDK2 activity is also associated with radioresistance in TNBC via modulation of pathways such as the CDK2/TRIM32/STAT3 axis, which promotes tumor survival under radiotherapy. Inhibition of CDK2 has already shown promising anti-tumor effects, including sensitization to chemotherapy and immunotherapy.^74-76

Amplification of TOP2A gene is associated with enhanced sensitivity to anthracycline chemotherapy, making TOP2A a predictive biomarker of treatment response. ⁷⁷ CCNB1 is a key regulator of the G2/M transition. During the cell cycle, it couples with CDK1 to form the maturation-promoting factor (MPF). CCNB1 overexpression in breast cancer is associated with more aggressive tumor characteristics, including larger tumor size, higher grade, lympho-vascular invasion, and hormone receptor negativity. ⁷⁸ Multiple cell types widely express the fibronectin (FN) family and participate in various functions, such as cell adhesion and migration during host defense, blood coagulation, wound healing, embryogenesis, and cell proliferation. FN1 is also involved in NKp46 receptor-mediated interferon-γ production by natural killer cells, which contributes to the control of tumor architecture and metastasis. This protein promotes cell migration and invasion in breast cancer, participates in immune cell infiltration, and functions through pathways such as ECM-receptor interaction and focal adhesion, which are interconnected with metastasis. ⁷⁹ UBC plays a vital role in protein degradation pathways, which regulate the stability of oncogenes. Dysregulation of UBC can lead to breast cancer progression by promoting cell survival, invasion, and therapy resistance through effects on signaling pathways such as NF-κB and TGF-β. ⁸⁰ PRKDC plays a pivotal role in DNA repair and maintaining genome stability. Its overexpression is frequently observed in breast cancer, where it correlates with enhanced tumor cell growth as well as resistance to DNA-damaging chemotherapy. ⁸¹ In TNBC, which often exhibits defects in homologous recombination repair due to BRCA1/2 mutations, PARP1 inhibitors have shown efficacy by exploiting this vulnerability, thereby inducing synthetic lethality and tumor cell death. This therapeutic approach has also been validated clinically, positioning PARP1 inhibition as a key strategy in TNBC treatment. ⁸²

The regulatory analysis highlighted the involvement of the top five TFs (GATA2, YY1, FOXC1, NFIC, and NFYA) and five miRNAs (hsa-let-7a-5p, hsa-let-7b-5p, hsa-let-7c-5p, hsa-miR-15a-5p, and hsa-miR-20a-5p) in modulating the expression of TKGs of TNBC. The strong connection between both of these factors and the TKGs suggests that they profoundly influence the expression of the TKGs across various physiological and pathological conditions. In TNBC, these TFs contribute to the aggressive phenotype and response to therapy. GATA2 promotes metastatic traits and survival pathways in TNBC. ⁸³ YY1 also supports tumor progression through enhancing cellular proliferation and invasion and by repressing tumor suppressors. ⁸⁴ FOXC1 is a well-established regulator of TNBC aggressiveness, promoting oncogenic signaling and metastatic potential. ⁸⁵ Particularly specific isoforms of NFYA exhibit malignant behavior in TNBC by regulating metabolism and sustaining tumor growth. ⁸⁶

On the other hand, the let-7 family of miRNAs, including hsa-let-7a-5p, hsa-let-7b-5p, and hsa-let-7c-5p, functions as tumor suppressors by regulating genes involved in cell proliferation, migration, and metabolism.^87-90 hsa-let-7b-5p suppresses TNBC tumor growth by targeting HK2 and AURKB. ⁹¹ hsa-miR-15a-5p inhibits breast cancer cell proliferation and promotes apoptosis by targeting anti-apoptotic genes such as BCL2, making it vital in controlling TNBC progression. ⁹² hsa-miR-20a-5p plays an important role in regulating cell cycle and angiogenesis-related genes, which influence tumor growth and metastatic potential in this breast cancer subtype. ⁹³ Collectively, these miRNAs modulate key pathways that govern TNBC aggressiveness and therapeutic response. ⁹⁴

As the number of targeted therapeutic agents for TNBC remains very limited, we considered TKG-targeted drugs for docking, sourced from DrugBank, DGIdb, and the existing literature. Molecular docking was performed with seventy-two compounds against the top two TKGs, CDK1 and CDK2, leading to the identification of two compounds (Alsterpaullone and BLU-222) with strong binding affinities. Interaction analysis revealed that Alsterpaullone bound CDK1 via conventional hydrogen bonds with Arg170, Asp128, Lys130, Gln132, and Thr14. Its fused core contributed a π-σ interaction with Thr14, π-anion forces with Asp128 and Glu163, and surrounding van der Waals contacts. Meanwhile, BLU-222 anchored to CDK2 through conventional hydrogen bonds with Asn132 and Asp145, a carbon–hydrogen bond with Asp145, a π-anion force with Asp145, and hydrophobic alkyl/π -alkyl contacts with Val18, Tyr15, Phe80, Ala144, and Ala149, supported by surrounding van der Waals interactions. These combined networks drove their favorable predicted binding energies despite the absence of π-π stacking. Further ADMET profiling of these candidates yielded moderately satisfactory results with identifiable areas for optimization. While Alsterpaullone exhibited higher cytotoxicity than ideal, it remains among the top-performing compounds in our docking analysis. Alsterpaullone potently inhibits the CDK family and induces cell cycle arrest and apoptosis in various cancer cell lines, underscoring its efficacy.^59,95 Specifically, Alsterpaullone’s pyrroloazepinoindole core forms crucial hydrogen bonds and hydrophobic contacts within the CDK1 hinge region, which can be further optimized by introducing halogen atoms (F or Cl) to strengthen ATP-pocket binding. Similarly, BLU-222 interacts with the CDK2 active site, where functionalizing its side chains toward the hydrophobic back-pocket could enhance kinase selectivity. Medicinal chemists can utilize these structural moiety modifications in wet-lab synthesis to optimize lead potency and selectivity against TNBC. Docking and simulation studies in structurally related CDK inhibitors showed that toxicity profiles can usually be improved through optimization, formulation, or simply the delivery method. BLU-222 is a novel, selective CDK2 inhibitor currently undergoing clinical evaluation, highlighting its therapeutic promise. The emerging potential of BLU-222 as a targeted agent aligns well with the urgent need for novel interventions in TNBC.^55,57

The 100 ns MD simulations allow a detailed comparison of the structural fluctuations and binding events within the CDK1 and CDK2 protein complexes with Alsterpaullone and BLU-222. The initial responses of both complexes were similar, with marked RMSD jumps within the first 10 ns, indicating rapid conformational adaptation to the simulation environment. A subsequent stabilization at similar RMSD plateaus (0.25-0.3 nm) occurred after 20 ns. The CDK1-Alsterpaullone did not fluctuate as much and maintained a relatively stable conformation, with only small transient fluctuations after 75 ns. On the other hand, both systems reached equilibration comfortably within the duration of the simulations.

Analysis of the RMSF uncovered varying patterns of local flexibility. CDK1-Alsterpaullone showed peaks and larger fluctuations (most residues below 0.3 nm, localized in higher values), compared to CDK2-BLU-222 with lower overall flexibility (below 0.25 nm) except for the C-terminal, where they reached 0.5 nm. This indicates that the two kinases show distinct regional dynamics.

The Rg profiles for both complexes showed a constant level of compactness throughout the simulations, indicating that protein folding was stable and that no unfolding events occurred. Similar levels of SASA were observed in the 100 ns trajectories of both systems, suggesting little change in protein solvent accessibility during the simulations and preserving the burial of the hydrophobic core. The pattern and dynamics of protein-ligand hydrogen bonds were quite different for the CDK1-Alsterpaullone complex. The complex had consistent hydrogen bonds (2-5 bonds) throughout the simulations, with transient peaks of more than 4 bonds, thereby supporting sufficient and stable protein-ligand interactions. In contrast, the CDK2-BLU-222 complex showed significant temporal variation (electrostatic complementarity), with 6-7 hydrogen bonds present at early stages, dropping to 0-2 hydrogen bonds during subsequent simulations, suggesting that binding interactions weakened over time. The CDK1-Alsterpaullone complex exhibited stronger binding stabilization and more stable intermolecular interactions than the slowly destabilizing interface of the CDK2-BLU-222 complex, even though both complexes showed rapid structural equilibration.

Overall, the results above indicate that the CDK1-Alsterpaullone complex is more stable than the CDK2-BLU-222 complex. These findings are highly valuable for understanding the stability and long-term heterogeneity of these kinase-ligand complexes. However, wet-lab studies should be performed to validate whether these computational findings are biologically relevant and have a therapeutic impact.

Conclusion

The results of this study demonstrate that TNBC is associated with molecular dysregulation of critical genes. Through transcriptome profile analysis, eight prevalent TKGs: CDK1, CDK2, TOP2A, FN1, UBC, PARP1, PRKDC, and CCNB1, were recognized as essential to the disease’s pathophysiology. The TKG-set enrichment study, which included biological processes, molecular functions, cellular components, KEGG pathways, and regulatory factors (such as transcription factors and miRNAs), identified common pathogenetic mechanisms. Ultimately, TKG-guided molecular docking of repurposable drug candidates (Alsterpaullon and BLU-222) exhibited their potential as therapeutic agents for TNBC. These molecules may provide effective treatment alternatives while minimizing the necessity for several medications, which might trigger toxicity issues. However, additional experimental validation is necessary to determine the therapeutic efficacy of these results.

Appendix.

List of Abbreviations

ADMET

Absorption, Distribution, Metabolism, Excretion, and Toxicity

ATP

Adenosine Triphosphate

AUC

Area Under Curve

AURKB

Aurora Kinase B

BBB

Blood Brain Barrier

BCL2

B Cell Lymphoma 2

BP

Biological Process

BRCA1

Breast Cancer Gene 1

BRCA2

Breast Cancer Gene 2

CC

Cellular Component

CCNB1

Cyclin B1

CDK1

Cyclin Dependent Kinase 1

CDK2

Cyclin Dependent Kinase 2

cGAS

Cyclic GMP AMP Synthase

CI

Confidence Interval

CYP

Cytochrome P450

DAVID

Database for Annotation, Visualization and Integrated Discovery

DEG

Differentially Expressed Gene

DFS

Disease Free Survival

DGIdb

Drug Gene Interaction Database

DMFS

Distant Metastasis Free Survival

DNA

Deoxyribonucleic Acid

ECM

Extracellular Matrix

ER

Estrogen Receptor

ESOL

Estimated Aqueous Solubility

EZH2

Enhancer of Zeste Homolog 2

FC

Fold Change

FDR

False Discovery Rate

FN1

Fibronectin 1

FOXC1

Forkhead Box C1

fs

Femtosecond

GATA2

GATA Binding Protein 2

GEO

Gene Expression Omnibus

GEPIA

Gene Expression Profiling Interactive Analysis

GO

Gene Ontology

HER2

Human Epidermal Growth Factor Receptor 2

HIA

Human Intestinal Absorption

HK2

Hexokinase 2

HR

Hazard Ratio

IMEx

International Molecular Exchange Consortium

KEGG

Kyoto Encyclopedia of Genes and Genomes

LIMMA

Linear Models for Microarray Data

LogBB

Logarithm of Brain to Blood Ratio

LogP:

Logarithm of Partition Coefficient

log2FC

Log2 Fold Change

MCC

Maximal Clique Centrality

MD

Molecular Dynamics

MF

Molecular Function

miRNA

MicroRNA

MM-GBSA

Molecular Mechanics Generalized Born Surface Area

MNC

Maximum Neighborhood Component

MPF

Maturation Promoting Factor

NCBI

National Center for Biotechnology Information

NFIC

Nuclear Factor I C

NFYA

Nuclear Factor Y Subunit Alpha

NF-κB

Nuclear Factor Kappa B

NKp46

Natural Cytotoxicity Triggering Receptor 1

nm

Nanometer

NPT

Isothermal Isobaric Ensemble

ns

Nanosecond

NVT

Isothermal Isochoric Ensemble

OS

Overall Survival

PARP1

Poly (ADP-Ribose) Polymerase 1

PDB

Protein Data Bank

PPI

Protein Protein Interaction

PR

Progesterone Receptor

PRKDC

Protein Kinase DNA Activated Catalytic Subunit

ps

Picosecond

RFS

Relapse Free Survival

Rg

Radius of Gyration

RMSD

Root Mean Square Deviation

RMSF

Root Mean Square Fluctuation

Ro5

Rule of Five

ROC

Receiver Operating Characteristic

RRA

Robust Rank Aggregation

SASA

Solvent Accessible Surface Area

STING

Stimulator of Interferon Genes

STRING

Search Tool for the Retrieval of Interacting Genes

TCGA

The Cancer Genome Atlas

TF

Transcription Factor

TGF-β

Transforming Growth Factor Beta

TIMER

Tumor Immune Estimation Resource

TKG

TNBC Key Gene

TNBC

Triple Negative Breast Cancer

TOP2A

DNA Topoisomerase II Alpha

TSDEG

TNBC Shared Differentially Expressed Gene

UBC

Ubiquitin C

VDss

Volume of Distribution at Steady State

YY1

Yin Yang 1.

ORCID iD

Tasnim Hosen Tanha https://orcid.org/0009-0009-9344-2652

Ethical Considerations

Not applicable. This study did not involve human participants, live animal experiments, or tissue samples. All data used were obtained from publicly available repositories.

Consent to Participate

Not applicable. This study did not involve human participants, patient clinical trials, or individual person’s data.

Data Availability Statement

All datasets generated and analyzed during the current study are included within this published article and its supplementary information file.*