The cell lines MCF-7 and human umbilical vein endothelial cells (HUVECs) were sourced from the Shanghai Cell Bank, Chinese Academy of Sciences (Shanghai, China). API (98%) was purchased from Shanghai Macklin Biochemical Co., LTD. (Shanghai, China). Salinomycin (SAL) (98%) was purchased from MedChemExpress (Monmouth Junction, NJ, United States). A 0.05% trypsin solution, containing ethylene diamine tetraacetic acid (EDTA) dissolved in PBS, was procured from Wuhanpssel Life Technology Co., LTD. (Wuhan, China). Dulbecco’s modified eagle medium (DMEM), fetal bovine serum (FBS), penicillin-streptomycin solution, and goat serum were purchased from Gibco Life Technologies (Grand Island, NE, United States). Endothelial cell complete medium was sourced from VivaCell Biotechnology GmbH (Denzlingen, Germany). Epidermal growth factor (EGF) and basic fibroblast growth factor (bFGF) were purchased from PeproTech (Hamburg, Germany). Heparin was sourced from Sigma-Aldrich Chemical Company (St. Louis, MO, United States). The B27 supplement was acquired from Gibco (Grand Island, NY, United States). A cell counting kit-8 (CCK-8) was purchased from Apexbio (Houston, TX, United States). An Annexin V-FITC/PI apoptosis kit was procured from Beyotime Biotechnology (Shanghai, China). The ALDEFLUORTM kit for aldehyde dehydrogenase 1 family member (ALDH1) activity assays was purchased from STEMCELL Technologies China Co., LTD. (Shanghai, China). The PE-conjugated mouse anti-human cluster of CD24 molecule (CD24) antibody (Clone ML5) and the FITC-conjugated mouse anti-humancluster of CD44 molecule (CD44) antibody (Clone G44-26) were purchased from BD Biosciences (San Diego, CA, United States).

Using “Apigenin” as the keyword, we searched for gene targets associated with API across multiple databases, including traditional chinese medicine systems pharmacology database and analysis platform (TCMSP) (https://tcmsp-e.com) (Ru et al., 2014), the encyclopedia of traditional chinese medicine (ETCM) (http://www.tcmip.cn/ETCM/index.Php) (Xu et al., 2019), Pharm Mapper (http://www.lilab-ecust.cn/pharmmapper) (Wang et al., 2017), SwissTarget Prediction (https://swisstargetprediction.ch) (Daina et al., 2019) and GeneCards (https://www.genecards.org) (Stelzer et al., 2016). In addition, BCSC-related targets were collected from the GeneCards database (https://www.genecards.org) (Stelzer et al., 2016), Cancer SEA (hrbmu.edu.cn) (Yuan et al., 2019) and the NCBI Count database (https://www.ncbi.nlm.nih.gov/gene) (Sharma et al., 2018), specify the species-specific characteristics of “Homo sapiens” in the search parameters. To identify possible common targets of API and BCSCs, the two targets were crossed.

The predicted potential common targets were imported into the STRING database (https://www.string-db.org/) (Szklarczyk et al., 2023) for protein-protein interactions (PPIs) analysis. To validate the veracity and precision of our data analysis, we established a stringent criterion by setting the minimum interaction score to a high confidence level of 0.9 and the species was restricted to “Homo sapiens”. Subsequently, the results were integrated into Cytoscape 3.9 for visualization of the interaction network between common protein targets, with nodes having a connectivity score of 0 being omitted. The cytoNCA plugin facilitated the calculation of degree values across various pathways, while the cytoHubba plugin employed a topological network algorithm to assign values to each gene. Our examination focused on the top 10 targets with the highest maximum clique centrality (MCC), aiming to discern specific pathways and pivotal targets that may be significantly influenced by API in the context of BCSCs. This approach was essential for uncovering the potential impact of API on BCSC biology and identifying key molecular intervention points.

The DAVID database was utilized for conducting both a gene ontology (GO) functional enrichment analysis and a Kyoto encyclopedia of genes and genomes (KEGG) signaling pathway enrichment analysis for the common targets of API and BCSCs (Sherman et al., 2022). The GO functional enrichment analysis encompassed three primary ontologies: molecular function (MF), biological process (BP), and cellular component (CC). Filter the results with p-values less than 0.01. Select the top 10 GO terms and 20 KEGG pathways, and then use SRplot (http://www.bioinformatics.com/) (Tang et al., 2023) for graphical representation. SRplot is a tool for data visualization and graphical depiction.

We integrated potential API targets, BCSC-associated targets, and enriched pathways into Cytoscape 3.9 to construct the “component-target-pathway” network. This network uses distinct graphical nodes and color-coding to visually represent its various components, targets, and pathways, whereas the interaction between any two nodes is denoted by connected edges. This approach facilitated a comprehensive visualization of the interactions and relationships within the network, providing a deeper understanding of the complex biological processes at play.

Molecular docking between key targets and API was conducted utilizing AutoDock software (https://vina.scripps.edu/). The binding affinity value, expressed in kcal/mol, corresponds to the thermodynamic binding affinity of API for the target protein. An increased absolute value of the binding free energy indicates greater stability of the ligand-receptor complex. Having identified the precise location of the original ligand of the receptor protein, we proceeded to remove it. Following this, we defined the dimensions of the grid box to 30 in the x, y, and z dimensions, with a modes parameter of 20, an exhaustiveness parameter of 15, and a number of genetic algorithm runs (GA) set to 10. We then performed molecular docking of the original ligand and the API with the receptor protein. By comparing the binding energies of the API and the original ligand, we assessed the binding stability of the API with the key target, which provided valuable insights into the potential efficacy of the API as a therapeutic agent.

MCF-7 cells were cultured in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin solution. HUVECs were maintained in endothelial cell complete medium. All cultures were incubated at 37°C in a humidified atmosphere containing 5% CO₂. BCSCs were derived from MCF-7 cells using a suspension culture method, as previously reported (Cao et al., 2018). MCF-7 cells were cultured in 6-well ultra-low adherent plates (Corning, NY, United States) at a density of 2 × 10⁵ cells per well. Each well contained 2 mL of DMEM supplemented with 2% B27, 1% penicillin-streptomycin solution, 0.4% bovine serum albumin (BSA), 20 ng/mL EGF, bFGF, and 10 ng/mL heparin, and was maintained at 37°C with 5% CO₂ in air. Following an uninterrupted culture period of at least 3 weeks, stem cells were successfully cultured and assessed for stmness indices.

The CCK-8 assay was utilized to assess the cytotoxic effects of API and SAL on enriched BCSCs and MCF-7 microspheres. Enriched BCSCs and MCF-7 microspheres were dissociated into single-cell suspensions and then seeded into 96-well plates at a density of 5 × 10³ cells per well. Various concentrations of API were added to each well, with the BCSC experimental group exposed to concentrations ranging from 5 to 80 μM, and the MCF-7 group to concentrations ranging from 5 to 100 µM. Concurrently, a blank control group was included and all plates were incubated for 48 h. Similarly, SAL was added to each well at concentrations ranging from 0 to 10 µM and incubated under identical conditions. Following the incubation period, 10 µL of CCK-8 reagent was added to each well, and the cells were further incubated at 37°C for 3 h. The optical density was measured at 450 nm using a Synergy H1 Hybrid multimode microplate reader. The half-maximal inhibitory concentration (IC₅₀) values of API and SAL on BCSCs and MCF-7 cells were subsequently calculated using GraphPad Prism 9.5 software.

To ascertain whether API exhibits greater selectivity for BCSCs relative to normal cells, HUVECs were prepared for assessment of their response to API. The HUVECs were allocated into groups: a control group with no treatment, an API treatment group with concentrations ranging from 5 to 100 μM, and a SAL treatment group with concentrations ranging from 0.25 to 20 µM. Following the same experimental protocol, the IC₅₀ values for API and SAL on HUVECs were calculated. This analysis aimed to evaluate the selectivity of API for BCSCs over normal cells.

A 24-well ultra-low adhesion plate was filled with 3 mL of serum-free medium DMEM/F12⁺. BCSCs derived from MCF-7 cells were seeded into the wells at a density of 5 × 10⁴ cells/well and subjected to drug treatment. The positive control group received a single concentration equivalent to the IC₅₀ value, which is 2 µM of SAL. In contrast, the experimental groups were designed at different concentration gradients based on the IC₅₀ value, specifically at 1/4, 1/2, and 1 times the IC₅₀ value, corresponding to 7 μM, 14 μM, and 28 µM of the API, respectively. Additionally, a negative control (NC) group was included to ensure the accuracy and reliability of the experimental results. Each condition was assayed in triplicate, and the cells were incubated for 6 days without medium replacement. Microscopic images were captured on days 0, 3 and 6 to monitor the effects of the treatments on spheroid formation. Only spheres with a cell mass/spherical area greater than or equal to 5,000 µM² were quantified.

BCSCs treated with the aforementioned drugs were collected, and 6 × 10⁵ cells were resuspended in 200 µL of binding buffer. Annexin V-FITC and PI staining solutions (5 µL each) were added, and the cells were incubated for 15 min in the dark at room temperature. Subsequently, the stained cells were subsequently analyzed using a CytofLEX S flow cytometer, employing the following gating strategy: First, cells were gated using forward scatter (FSC) and side scatter (SSC) to separate the cellular population from debris. Then, single cells were selected by gating on FSC-A (area) vs. FSC-H (height) to exclude doublets and aggregates. Lastly, apoptotic cells were identified using an FITC (Annexin V-FITC) vs. PI plot, categorizing live cells as Annexin V−/PI−, early apoptotic as Annexin V+/PI−, late apoptotic or necrotic as Annexin V+/PI+, and dead cells as Annexin V−/PI+. The percentage of apoptotic cells was calculated for each group using CytExpert software after Annexin V-FITC/PI staining.

To determine the frequency of CD44⁺/CD24^−/low cells in MCF-7 cells and BCSCs, we conducted a flow cytometric analysis. Cells were harvested and stained with phycoerythrin (PE)-conjugated CD24 and fluorescein isothiocyanate (FITC)-conjugated CD44 antibodies following a standard protocol. We selected 2.5 × 10⁵ cells per sample, washed them twice with 1 mL staining buffer, and centrifuged at 250 × g for 3 min. After supernatant removal, the cell pellet was resuspended in 100 µL staining buffer containing 2.5 µL each of PE-conjugated CD24 and FITC-conjugated CD44 antibodies. After a 30-min incubation at 4°C, the stained cells were then analyzed using a CytofLEX S flow cytometer with the following gating strategy: Initially, cells were gated on FSC and SSC to distinguish the cellular population from debris. Next, single cells were selected by gating on FSC-A vs. FSC-H to exclude doublets and aggregates. Finally, CD44⁺/CD24^−/low cells were identified using an FITC (CD44-FITC) vs. PE (CD24-PE) plot.

To quantify the ALDH^high population in MCF-7 cells and BCSCs, we utilized the ALDEFLUOR™ kit for staining. Harvested cells (6 × 10⁵) were resuspended in 600 µL of ALDEFLUOR™ assay buffer, and 3 µL of activated ALDEFLUOR™ reagent was added. A control group was prepared by mixing half of the cell suspension with 6 µL of the ALDH inhibitor, diethylaminobenzaldehyde (DEAB). Following a 45-min incubation at 37°C, cells were centrifuged at 250 × g for 3 min, and the supernatant was aspirated. Cell pellets were resuspended in 300 µL of ALDEFLUOR™ assay buffer, and ALDH activity was measured using the CytofLEX S flow cytometer. The stained cells were analyzed with the following gating strategy: Initially, cells were gated on FSC and SSC to distinguish the cellular population from debris. Next, single cells were selected by gating on FSC-A vs. FSC-H to exclude doublets and aggregates. Finally, the ALDH^high population was identified using an FL-1 (green fluorescence) vs. SSC plot.

First, total RNA was extracted from cells that had been treated with various drug concentrations for 48 h. RNA purity and quality were assessed to ensure suitability for downstream applications. Subsequently, reverse transcription and amplification of quantitative polymerase chain reaction (Q-PCR) were conducted using the established reaction system. We investigated whether the API could downregulate the expression of genes associated with stemness at the RNA level, including ALDHA1A1, MYC proto-oncogene, bHLH transcription factor (c-MYC), nanog homeobox (NANOG), and epithelial cell adhesion molecule (EpCAM), and designed corresponding primers based on the top 10 targets selected by MCC, thereby laying the groundwork for further investigation into the mechanism of API against BCSCs. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was utilized as an endogenous control for data normalization. Data were analyzed using the 2^−ΔΔCT method, enabling precise quantification of gene expression levels. The primer sequences, designed specifically for this study, are detailed in Table 1. This approach ensures the accuracy and reliability of the gene expression analysis.

By integrating Q-PCR expression data with KEGG signaling pathway enrichment analysis, we identified key signaling pathways. Specifically, we focused on validating the expression and activity of key proteins phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K), AKT serine/threonine kinase (AKT), and tumor protein 53 (p53) at the protein level. For the assessment of cellular stemness, we used GAPDH as an internal control to evaluate in depth the expression of stem cell markers at the protein level, including ALDH1A1, c-MYC, NANOG, and EpCAM. Furthermore, to elucidate the mechanism of action of the API, we detected the expression levels of phosphorylated PI3K (p-PI3K), phosphorylated AKT (p-AKT), and p53 proteins.

Cells were exposed to varying concentrations of the API and the positive control, SAL (2 µM), for a duration of 48 h. Following this incubation period, the residual medium was carefully aspirated, and the cells were gently rinsed. Cells were then lysed using a radioimmunoprecipitation assay (RIPA) buffer supplemented with a protease inhibitor cocktail (phenylmethylsulfonyl fluoride, PMSF) at a ratio of 100:1 for 20 min at 4°C. The lysate was then centrifuged at 12,000 g and 4°C for 10 min, and the supernatant was collected. The protein concentration in the supernatant was then determined using the bicinchoninic acid (BCA) assay, which involved preparing a standard curve by diluting a 0.5 mg/mL protein standard to concentrations of 0.5, 0.25, 0.125, and 0.0625 mg/mL and adding 10 µL of each to a 96-well plate. Following this, the protein samples were diluted tenfold by mixing 1 µL of sample with 9 µL of water and added to the plate. Then, 100 µL of BCA reagent was added to each well, incubated at 37°C for 30 min, after which the absorbance was measured at 562 nm. The standard curve was plotted using the absorbance values, and the protein concentrations of the samples were interpolated from it.

For the Western blot analysis, proteins were resolved on a 10% sodium dodecyl sulfate-polyacrylamide (SDS-PAGE) gel. The amount of protein was 30 mg per well. The resolved proteins were transferred onto polyvinylidene fluoride (PVDF) membranes, followed by blocking with a 5% solution of skim milk in tris-buffered SALine with Tween (TBST) for 1 h at ambient temperature. The membranes were incubated with the following primary antibodies overnight at 4°C, diluted as specified: anti-ALDH1A1 (1:500), anti-c-MYC (1:1000), anti-NANOG (1:1000), anti-EpCAM (1:1000), anti-phospho-PI3K (1:1000), anti-PI3K (1:1000), anti-phospho-AKT (1:1000), anti-AKT (1:1000), anti-tumor protein p53 (TP53) (1:1000), and anti-GAPDH (1:3000). After thorough washing with TBST buffer, the membranes were incubated with a secondary horseradish peroxidase (HRP)-conjugated goat anti-rabbit IgG antibody (1:2000) for 1 h at room temperature. After additional washes with TBST, the membranes were visualized using an ultra-sensitive enhanced chemiluminescence (ECL) detection kit reagent and imaged using a MiniChemi™ 910 imaging system. The intensity of the bands was quantified using VisionWorks software (UVP, Upland, CA) to assess the relative protein expression levels.

In this study, the data were derived from at least 3 distinct experimental trials, ensuring a robust dataset for analysis. Prior to statistical analysis, the data were subjected to rigorous examination to assess their distribution and variance. To standardize the data and reduce variability, logarithmic transformations were applied to all variables. Statistical analysis involved one-way analysis of variance (one-ANOVA) to evaluate differences among multiple groups, and post hoc pairwise comparisons were conducted using Tukey’s honestly significant difference (HSD) test to precisely identify which specific groups differed significantly from one another. Results are expressed as mean ± standard deviation (Mean ± SD), which not only clearly delineates the central tendency of the data points but also reflects their dispersion. A p-value less than 0.05 was considered statistically significant.

To elucidate the potential effects of API against BCSCs, a comprehensive network pharmacology study was conducted utilizing established and authoritative databases. Initial confirmation of API’s structure (Figure 1A) was followed by extraction of two datasets: 1033 BCSC-associated target genes and 429 API-associated target genes from databases such as Genecards. Through the construction of a venn diagram, an overlapping set of 63 genes was identified (Figure 1B). For PPIs, the STRING database and Cytoscape 3.9 software were employed. Employing the cytoHubba plugin, the topological significance of the 63 potential core targets was identified and ranked to reveal hub genes and subnetworks (Figure 1C). Subsequent GO and KEGG enrichment analysis revealed significant enrichment of these targets in 117 pathways, with the PI3K/AKT pathway being the obvious enriched (Figures 2A, B). A “compound-target-pathway” network was generated using Cytoscape 3.9, integrating the potential API targets, BCSC targets, and enriched pathways. Within this network, the size and color of the nodes represent their level of importance (Figure 2C). Notably, TP53, heat shock protein 90 alpha family class A member 1(HSP90AA1), heat shock protein 90 alpha family class B member 1 (HSP90AB1) hypoxia inducible factor 1 subunit alpha (HIF1A), estrogen receptor 1 (ESR1), AKT1, epidermal growth factor receptor (EGFR), BCL2 apoptosis regulator (BCL2), glycogen synthase kinase 3 beta (GSK3B), and telomerase reverse transcriptase (TERT) emerged as prominent, occupying the top 10 target positions (Figure 2D; Table 2). The central role of TP53 as the most critical nuclear target and its high enrichment strongly indicates that API may play a crucial role in combating BCSCs via the “PI3K/AKT/p53” signaling pathway.

Molecular docking analysis was conducted to explore the potential targets of API against BCSCs. Based on network pharmacology findings, API was docked with the top 10 proteins, ranked by MCC scores, and the binding energy was calculated. A lower binding energy signifies a more stable ligand-receptor configuration, as reflected in the docking score. The binding energy of API was compared with that of the original ligands for each core target to evaluate stable binding. A binding energy close to or lower than −5 kcal/mol (Nimbannavar and Mane, 2019; Li et al., 2023; Ray and Mukherjee, 2024), suggests stable binding between API and the target protein. The results indicated that 55% of the targets had a docking energy below −8 kcal/mol, 27% below −7 kcal/mol, 9% below −6 kcal/mol, and another 9% below −5 kcal/mol. All binding energies were below −5 kcal/mol, demonstrating potential favorable binding activity of the top 10 targets with API (Table 3). The cartoon representation enhances visualization of the ligand-protein binding residues and illustrates the molecule’s propensity to form hydrogen bonds (Figure 3).

Notably, a significant difference in binding energy was observed between the original ligand of TP53 and API compared to other targets. Specifically, the binding energy of API to TP53 exhibited a more pronounced decrease compared to the original ligand, indicating a theoretical indication of tighter binding to TP53. This observation implies that APT might have potential effect with TP53.

MCF-7 cells cultured under serum-free conditions in DMEM/F12⁺ medium were subjected to a 3-week suspension culture in ultra-low attachment plates. Over the course of this period, cells progressively aggregated to form tumorospheres. Tumorosphere formation was observed and documented using microscopy (Figures 4A, B). To evaluate stemness properties, the proportions of CD44⁺/CD24^−/low and ALDH1^high subsets were determined by flow cytometry. Results demonstrated that following suspension culture in serum-free medium DMEM/F12⁺, the percentage of CD44⁺/CD24^−/low significantly increased from 22.76% ± 2.52% to 93.38% ± 0.65%, which was significantly higher compared to MCF-7 cells (p < 0.0001, 95% CI: [67.80, 73.35]) (Figures 4C, D). Furthermore, the expression level of ALDH1 rose from 5.69% ± 0.20% to 12.00% ± 1.00%, which is also a significant increase (p < 0.01, 95% CI: [3.05, 6.70]) (Figures 4E, F). Collectively, these findings suggest that BCSCs derived from MCF-7 cells through suspension culture exhibit robust stem-like characteristics.

To elucidate the effect of API on the proliferation of BCSCs more clearly, we tested various concentrations of API on these cells and conducted comparative experiments using HUVECs to evaluate the specificity of API’s cytotoxicity. After 48 h of treatment, we observed a significant dose-dependent inhibition of BCSCs’ proliferation. Specifically, as the concentration of API increased from 0 µM to 100 μM, the cell viability of BCSCs, MCF-7, and HUVEC all significantly decreased. Through CCK-8 assays, we calculated the IC₅₀ values of API in BCSCs, MCF-7, and HUVECs to be 27.30 ± 1.23 µM, 50.12 ± 4.11 µM, and 69.75 ± 3.54 µM, respectively. The IC₅₀ value of API for BCSCs was approximately 2.55 times lower than that for HUVECs and 1.84 times lower than that for MCF-7 cells, indicating a significant selective cytotoxicity of API for BCSCs. Furthermore, while stem cells are generally more drug-tolerant, in this study, BCSCs exhibited greater sensitivity to API compared to non-stem breast cancer cells. For instance, under the influence of 50 µM API, the average viability of MCF-7 cells was 56.78%, while the average viability of BCSCs was 19.68%, exhibiting approximately a 2-fold difference (Figure 5A). In contrast, the IC₅₀ values of SAL in BCSCs, MCF-7, and HUVECs were 1.66 ± 0.51 µM, 1.60 ± 0.12 µM, and 1.08 ± 0.23 µM, respectively, indicating no selective cytotoxicity for BCSCs (Figure 5B). Although API is less effective than SAL in inhibiting BCSCs, it shows superior selectivity in targeting BCSCs.

To substantiate the inhibitory effects of API on BCSCs, apoptosis was measured by flow cytometry following exposure to various API concentrations for 48 h. The results revealed that the NC group exhibited an apoptosis rate of 1.10% ± 0.13%, indicating a low level of apoptosis in BCSCs under natural conditions. In the experimental groups, treatment with API at a concentration of 7 µM led to an apoptosis rate of 8.20% ± 1.58% (p < 0.001, 95% CI: [5.14, 9.07]), significantly higher than NC group, demonstrating that API can effectively promote apoptosis in BCSCs even at low concentrations. As the concentration of API increased, so did the apoptosis rate, with rates of 15.41% ± 1.00% (p < 0.0001, 95% CI: [13.05, 15.56]) and 18.78% ± 2.26% (p < 0.0001, 95% CI: [14.87, 20.48]) observed at 14 μM and 28 µM concentrations, respectively, showing a clear concentration-dependent effect. Compared with the positive control drug SAL, with an apoptosis rate of 8.28% ± 0.20%, API demonstrated a stronger or at least equivalent ability to promote cell apoptosis at all tested concentrations, with the effect becoming more pronounced as the concentration increased (Figures 5C, D). These results indicate that API has a significant capacity to promote apoptosis in BCSCs, and this effect is enhanced as the concentration of API increases, showing a good concentration dependence.

A defining characteristic of CSCs is their capacity to form spheres. Consequently, we developed an inhibition assay to evaluate the effects of API on BCSCs. After treatment with varying concentrations of API and subsequent continuous cultivation for 6 days, we observed a significant increase in cell numbers in the NC group, primarily characterized by the aggregation of single cells into clusters and the formation of spheroids. However, with increasing concentrations of API treatment, there was a reduction in cell numbers, leading to a significant decrease in the number of observable spheroids. Additionally, the positive control drug SAL showed effectiveness in inhibiting the formation of cell spheroids without significantly affecting the total cell count. The statistical analysis demonstrated that API exerted a significant inhibitory effect on BCSCs’ formation even at low doses, with an efficacy similar to the positive control at a concentration of 28 µM (Figures 6A, B).

To further substantiate our findings, we conducted Q-PCR, Western blot, and flow cytometry analyses. The Q-PCR analysis revealed that API significantly downregulated the expression of ALDH1A1, NANOG, CD44, SRY-box transcription factor 2 (SOX2), and c-MYC. Statistical results indicated a decreasing trend in the expression levels of ALDH1A1 and c-MYC with increasing concentrations of API. Furthermore, treatment with 28 µM API led to a significant reduction in the expression levels of NANOG, CD44, and SOX2. This finding suggests that API was markedly more effective in downregulating the expression of NANOG and c-MYC compared to the SAL group (Figure 6C). Western blot analysis showed that MCF-7 cells induced to transform into BCSCs via suspension culture exhibited significantly increased expression levels of stem cell markers ALDH1A1, c-MYC, NANOG, and EpCAM. Subsequent treatment with varying concentrations of API and the positive control SAL resulted in a concentration-dependent decrease in the expression of these markers. Statistical analysis revealed that API significantly downregulated the expression of ALDH1A1 and NANOG at both tested concentrations, with greater efficacy than the SAL group. Additionally, API demonstrated a significant reduction in the protein expression of c-MYC and EpCAM, particularly in the inhibition of c-MYC expression (Figures 6D, E). Consistent with the Western blot data, flow cytometry analysis revealed that the inhibitory effect on BCSCs increased with higher concentrations of API after a 48-h treatment. In the NC group, the percentage of CD44⁺/CD24^−/low cells was 90.93% ± 0.29%. After treatment with 14 μM and 28 µM API, the percentages of CD44⁺/CD24^−/low cells were reduced to 81.66% ± 1.64% (p < 0.001, 95% CI: [6.34, 11.52]) and 78.76% ± 1.18% (p < 0.0001, 95% CI: [10.21, 14.11]), respectively, demonstrating significant differences. The percentage of ALDH1^high cells also decreased from 11.54% ± 1.55% in the NC group to 3.88% ± 0.41% (p < 0.01, 95% CI: [5.09, 10.00]) and 2.32% ± 0.28% (p < 0.001, 95% CI: [6.70, 11.74]) after treatment with 14 μM and 28 µM API, respectively, also showing significant differences (Figures 6F–I). Collectively, these findings indicate that API effectively reduced the formation of BCSCs.

The results of network pharmacology and molecular docking analysis indicate potential interactions between API with TP53 and the PI3K/AKT signaling pathways, and further investigations of BCSCs were performed using Q-PCR and Western blot. These findings substantiate the hypothesis that the PI3K/AKT/p53 signaling pathway is instrumental in mediating the effects of API on BCSCs. To delineate the molecular mechanisms underlying API’s inhibitory effects on BCSCs, we assessed the mRNA levels of the top 10 genes within our PPI network. Q-PCR data revealed that treatment with 14 µM or 28 µM API for 48 h resulted in a significant upregulation of TP53 and ESR1 expression in enriched BCSCs (Figure 7A). The p53 protein, a pivotal tumor suppressor, is integral to the regulation of the cell cycle, the induction of apoptosis, and the repair of damaged DNA, as evidenced by literature (Boutelle and Attardi, 2021; Rodriguez-Pastrana et al., 2023; Wang et al., 2023). In cancer therapy, the PI3K/AKT signaling pathway is often targeted to inhibit cancer cell proliferation (He et al., 2021). Moreover, the PI3K/AKT signaling pathway can indirectly regulate TP53 expression through MDM2 proto-oncogene (MDM2)-mediated pathways (Abraham and O’Neill, 2014). Therefore, based on previous pathway analysis, we chose to further investigate the TP53 and PI3K/AKT signaling pathways at the protein level. Western blot analysis revealed that with increasing concentrations of API, the relative expression levels of p-PI3K and p-AKT proteins gradually decreased, while the expression level of TP53 protein significantly increased with API concentration. These results indicate that API modulates the PI3K/AKT/TP53 signaling pathway by downregulating the phosphorylation levels of PI3K and AKT and upregulating TP53 expression, thereby exerting its inhibitory effect on BCSCs (Figures 7B, C).