Data sources and processing

The TNBC RNA-seq data was acquired from The Cancer Genome Atlas (TCGA) data portal, and the transcription profiles were downloaded as fragments per kilobase (FPKM) for further analysis. The “limma” R package was used to identify differentially expressed genes (DEGs) in the TNBC cohort by comparing paracancerous tissue and tumor tissue. A screening threshold of |log2FC| > 1 and adjusted P < 0.05 was applied.

Weighted gene co-expression network analysis (WGCNA)

The TCGA-TNBC expression profile was utilized as the input dataset for WGCNA. The goodSamplesGenes method within the “WGCNA” R package was utilized initially to remove outlier genes and samples. First, genes with small fluctuations in the dataset were removed based on the median absolute deviation (MAD), and the remaining gene expression files were entered into WGCNA; second, the pick-Soft-Threshold function was used to calculate the adjacency using the soft threshold power b obtained from the co-expression similarity. Third, the adjacency relationships were converted into a topological overlap matrix (TOM) and the respective dissimilarity (1-TOM) was evaluated. Fourth, hierarchical clustering combined with a dynamic tree cut function was used to identify modules. In order to classify genes with similar expression profiles into gene modules, average linkage hierarchical clustering was used based on the TOM-based dissimilarity measure, and the minimum size of the gene tree diagram (genomes) was 50. Then, for the modules most closely associated with TNBC, the module membership as well as the gene significance were calculated. Finally, the characteristic gene network was visualized.

Identification of potential targets for TNBC

Potential targets related to TNBC were screened from the GeneCards database by searching the keyword “triple-negative breast cancer”. The genes obtained from DEGs, WGCNA, and those screened from the GeneCards database were used as input sets to create a venn diagram.

Identification of potential anti-TNBC targets for nomilin

The molecular structure of nomilin was shown in Supplementary Fig. 1A-B. Nomilin’s potential targets were screened from the Swiss Target Prediction database, the medicinal chemistry database ChEMBL, and the STITCH database. In general, we first obtained the SMILES of nomilin from the PubChem website and then put it into the Swiss Target Prediction database. In addition, potential targets related to nomilin were screened from the medicinal chemistry database ChEMBL, and the STITCH database by searching the keyword “nomilin”. After summarizing the Uniprot IDs, removing duplicates, and drawing a venn diagram, 301 potential targets of nomilin were identified. A venn diagram was also used to analyze the overlap between nomilin’s potential targets and TNBC targets, resulting in the identification of 221 potential anti-TNBC targets for nomilin. These 221 potential anti-TNBC targets of nomilin were then imported into Cytoscape software to construct a “drug-target-disease” network.

Constructing a protein-protein interaction (PPI) network for the shared targets of nomilin and TNBC

To obtain PPI information, the interactive genetic database search tool String was utilized. The PPI network was constructed by inputting 221 potential targets of nomilin against TNBC into the String database, selecting the Homo sapiens species, setting the minimum interaction score to high confidence (0.700), and removing disconnected nodes in the network.

Core target identification of nomilin for TNBC

In order to ascertain the principal targets of nomilin against TNBC, the PPI network was integrated into Cytoscape, and the CytoNCA plug-in was employed to conduct an analysis of betweenness centrality (BC), closeness centrality (CC), degree (DC), eigenvector centrality (EC), and local average connectivity (LAC) parameters for the relevant targets (Shi et al. 2021). By selecting the top 20 genes with the most elevated scores in every analysis and utilizing a venn diagram, 8 core targets were identified due to the presence of overlapping genes.

Computer-aided molecular docking

Protein structures of the eight main targets were obtained from the PDB database (Bcl-2 ID:1K3K; Caspase-3 ID:1CP3; CCND1 ID:6P8E; EGFR ID:2XKN; HSP90AA1 ID:5NJX; KRAS ID:6N65; PARP1 ID:7KK2; TNF ID:1CA4). Proteins were purified by eliminating duplicate conformations and water molecules, and the protein pockets’ binding sites were forecasted. Original ligands were then deleted. The results were input into AutoDock, and Autodock Tools were used to dehydrate and hydrogenate the proteins and drug molecules in the computer. Afterwards, molecular docking was performed, the optimal docking conformation was chosen based on low-energy and reasonable conformation principles, set the docking site to Ligand Atoms, and set the amino acids within 45° from the ligand molecule as the docking pocket, and the obtained results were saved along with the recorded affinity values. Stable binding between the ligand and the receptor molecule is indicated by a negative binding energy, with greater absolute values representing increased stability. Subsequently, the outcomes were imported into PyMOL for additional visualization.

Molecular dynamics simulation

The simulation was conducted utilizing the Schrodinger 2019 software module “Desmond”. The predefined simple point-charge (SPC) water model was employed, along with the OPLS2005 force field, to simulate water molecules. To neutralize the system charge, chloride ions/sodium ions were randomly added in appropriate amounts to balance the charge within the solvation system. Energy minimization is conducted utilizing the standard protocol incorporated within the Desmond module. Following this, a normal pressure and temperature (NPT) simulation was carried out for a duration of 100 ns, with data being recorded at intervals of 50 ps. The progress of the simulation was tracked and saved accordingly.

Proteome microarray assay

The HuProt™ 20 K Human Proteome Microarrays from CDI Laboratories, Inc, were employed for chip hybridization, washing, and detection following standard chip detection procedures. In a nutshell, the chip was immersed in a 5% BSA blocking solution and incubated with biotin (10 µM) and biotin-nomilin (10 µM) for 1 h. After thorough washing, the chip was placed in a 0.1% Cy5-Streptavidin Solution (Cy5-SA) and allowed to react for 20 min in the dark. Following further washing and centrifugation, the microarray was scanned at 635 nm using the GenePix 4000B scanner. The median (M) and standard deviation (SD) of the raw signal intensities (I) from all sites were calculated to determine the Z-Score of the corrected data for each site. Z-Score = (I-M) / SD.

Enrichment analysis

The 8 core targets were analyzed using R software to identify the biological processes and pathways through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis via the R “ClusterProfiler” package, P < 0.05 were regarded as significant. The above results were visualized by R“ggplot2” and “enrichplot” packages eventually. Histograms were used to visualize the initial 10 entries of GO analysis, encompassing cellular component, molecular function (MF), and biological process (BP), whereas bubble charts were employed to represent the first 30 entries of KEGG analysis results. Afterwards, the R programming language was employed to produce circular diagrams illustrating the main signaling pathways discovered in the GO and KEGG examinations.

Reagents and cell culture

Nomilin (purity 99.38%), the compound with the CAS number 1063-77-0 was bought from MedChemExpress (MCE, NJ, USA). Curcumin (CAS number 458-37-7) with a purity of 98.16%, obtained from MCE, was utilized as a positive control. The MDA-MB-231 and BT-549 cell lines, which are human TNBC cell lines (TNBCCs), were acquired from the Shanghai Institute of Cell Biology, Chinese Academy of Sciences. TNBCCs were cultured in DMEM medium and RPMI-1640 medium, respectively. In order to augment cellular culture conditions, a mixture of 10% FBS and 1% penicillin-streptomycin was incorporated into all media. The cells were subsequently incubated in a humidified environment with 5% CO2.

Cell counting Kit-8 (CCK-8) assay

TNBCCs were placed in 96-well plates with a cell density of 0.8 × 104 cells per well. After exposing the cells to different amounts of nomilin (0, 10, 20, 50, 100, 150, 200, 250, 300, 400, 500 µM) for a duration of 24 h, 10 µL of cck-8 reagent (Targetmol, Shanghai, China) was introduced into each well and left to incubate for 4 hours. The absorbance of each well was then measured at 450 nm.

Colony formation

TNBCCs were placed in 6-well plates with a concentration of 1000 cells per well. After the formation of visible cell clumps, they were incubated with 0, 150, and 300 µM nomilin for 24 h. After counting approximately 100 cells in the control group’s single-cell colony, the cells were fixed using 4% paraformaldehyde, stained with crystal violet for a duration of 15 min, air-dried, and subsequently photographed and enumerated.

Western blot (WB)

WB analysis was conducted as described previously (Wu et al. 2023). In summary, cells were disrupted using RIPA solution, and the overall protein concentration was assessed utilizing the BCA technique. The denatured total protein was then subjected to polyacrylamide gel electrophoresis. To remove protein bands of comparable molecular weights, we employed a stripping buffer (Beyotime Biotechnology, Shanghai, China) for membrane stripping. Supplementary Table 1 contains a list of the main antibodies utilized. Quantitative analysis of the images was carried out using Image J software.

Wound healing assay

TNBCCs were placed in a six-well dish until they developed a complete layer of cells. To create a linear gap on each plate, a 1000 µl pipette tip was used to scratch once. After rinsing the isolated cells with PBS, a serum-free solution was introduced, which included nomilin at concentrations of 0, 150, or 300 µM. Following a 2-day period, the culture area was examined under an inverted microscope to capture images.

Transwell assay

Dispense 200 µl of serum-free medium containing 8 × 104 cells into the upper chamber, and add 500 µl of medium with 20% FBS to the lower chamber. After cells have adhered, replace the medium in the upper chamber with serum-free medium containing 0, 150, or 300 µM nomilin. Subsequently, fix and stain the cells with crystal violet. Utilize a microscope to tally the migrating cells in various regions of the slide.

Flow Cytometry

Following a 24-hour incubation period of TNBCCs with nomilin, the cells were subjected to trypsin digestion and subsequent centrifugation. Following three washes with PBS, the cells were resuspended in a buffer solution. Next, the reconstituted cells were cultured with 5 µl FITC for a duration of 15–20 min. Following this, 5 µl 7-AAD was added for 5 min in order to conduct apoptosis analysis. The BD FACSVerse™ system was utilized to examine the apoptosis of TNBCCs through flow cytometry.

Immunofluorescence (IF)

Following a 24-hour treatment of TNBC cells with nomilin, the cells were then fixed and permeabilized using 0.1% Triton X-100. After that, the cells were obstructed using goat serum for a duration of 30 min and then subjected to the Cleaved Caspase3 primary antibody, which was left to incubate overnight at 4 °C. Afterward, the cells underwent three rounds of washing with PBS and were subsequently exposed to secondary antibodies at 37 °C for 1 h. Afterward, the cells were incubated with DAPI for 7 min, and finally mounted with an anti-fade fluorescence mounting medium. Fluorescence microscopy was utilized to capture IF images of the cells.

Xenograft tumor experiments

A group of sixteen BALB/c mice, all female and without fur, weighing between 18 and 22 g and aged 6 to 8 weeks, were housed in a controlled environment with regulated temperature, humidity, and lighting. They were provided with standard nourishment and access to water. Four groups were formed by randomly dividing mice. Subsequently, a total of 5 × 106 MDA-MB-231 cells were subcutaneously implanted into the right thigh root of mice. The mice in the nomilin group were administered nomilin at a dose of 15/30 mg/kg·2d (ip). The control group mice received injections of DMSO as a solvent whereas mice in positive control group were treated with curcumin (25 mg/kg·2d). Carbon dioxide asphyxiation was used to euthanize all mice, and tissues from the tumor, heart, liver, lung, and kidney were gathered for further experiments. The Institutional Animal Care and Utilization Committee of Wenzhou Medical University granted approval for this study involving animals on December 7, 2022 (WYYY-AEC-2022-062). Moreover, animal tests were conducted in compliance with the regulations set by all animal welfare governing bodies(Kilkenny et al. 2010).

Hematoxylin-eosin (HE) staining and immunohistochemistry (IHC) staining

The tumor and organ samples were preserved in formalin, encased in paraffin, and subsequently sliced into sections that were 4 μm in thickness. HE was used to stain both the organs and tumor tissues. IHC analysis was conducted following established protocols (Dai et al. 2018). The tumor tissue sections underwent deparaffinization using xylene, followed by rehydration and antigen retrieval, prior to incubation with primary antibodies. Subsequently, the sections were treated with enzyme-labeled secondary antibodies and diaminobenzidine, and finally analyzed.

Statistical analysis

In the field of network pharmacology, all statistical analyses were conducted using R software. Experimental data were analyzed using GraphPad Prism for statistical purposes. The data results were subjected to one-way analysis of variance, with statistical significance defined as p < 0.05. To ensure reliability, each independent experiment was conducted a minimum of three times. Comparative analyses between two groups were executed using Student’s t-test, whereas one-way ANOV A was employed for comparisons involving multiple groups. p-value < 0.05 was considered to indicate statistical significance.