The TKM formula granules (Batch Number: 2307003C) were purchased from China Resources Sanjiu Medical and Pharmaceutical Co., Ltd. According to the national standard (YBZ-PFKL-2021173), the corresponding TKM formula granules were weighed out and dissolved in sterile water, passed through a 0.22 μm filter, and configured into a 2000 mg/mL TKM original solution.

For qualitative chemical profiling, 0.10 g of TKM granules was extracted with 1.0 mL of 80% methanol in water (v/v), vortexed, centrifuged, and the supernatant was filtered for LC–MS/MS analysis.

Chromatographic separation was performed on a Thermo UPLC system using a reversed-phase column and a conventional water-acetonitrile gradient containing 0.1% formic acid at a flow rate of 0.3 mL/min; detailed gradient conditions are provided in the Supplementary Methods.

High-resolution MS detection was carried out on a Q Exactive Orbitrap mass spectrometer operated in positive/negative electrospray mode, and data acquisition and processing were performed with vendor software as described in the Supplementary Materials.

In addition, the content of schisandrin in the TKM granules was quantified by HPLC using UV detection and an external calibration curve constructed from a schisandrin reference standard. The chromatographic conditions were optimized based on the qualitative HPLC-Orbitrap-MS/MS analysis, and the measured schisandrin content in the TKM preparation is summarized in the Supplementary Materials.

For TKM, according to the results of HPLC-Orbitrap-MS/MS analysis, LC–MS features with an mzCloud Best Match ≥70 were retained as candidate constituents (Bi et al., 2020). To enrich for orally bioavailable, drug-like compounds, predicted gastrointestinal absorption and oral drug-likeness were evaluated using SwissADME; components with high GI absorption and satisfying at least two commonly used drug-likeness filters (Lipinski, Ghose, Veber, Egan, Muegge) were considered active ingredients for subsequent analyses (Daina et al., 2017). The corresponding targets were obtained by querying the SwissTargetPrediction database (Daina et al., 2017). In our research, we gathered known targets associated with breast cancer from the GeneCards database by utilizing the keyword “breast cancer” (Stelzer et al., 2016). The dataset GSE139038 comprised 41 patients alongside 17 controls sourced from the GPL27630 platform. We analyzed microarray datasets of differentially expressed mRNAs between the normal group and TNBC patients using the “limma” package in R. Genes were deemed pathogenic and related to TNBC if they had an adjusted p-value of less than 0.05 and an absolute log2 (fold change) greater than 2. Through the intersection of TKM and TNBC targets, we identified shared targets as potential therapeutic targets for TKM in the context of TNBC. The gene characteristics of these therapeutic targets were determined using the DisGeNet database (Li G. et al., 2022). Justification of thresholds: For MS/MS library matching, an mzCloud best-match score ≥70 was chosen to balance sensitivity and specificity and to reduce false positives from ambiguous spectra (Bi et al., 2020). Candidate TKM constituents were further required to have high predicted gastrointestinal absorption and to satisfy at least two standard oral drug-likeness rules (Lipinski, Ghose, Veber, Egan, Muegge) in SwissADME, in order to enrich for orally bioavailable, pharmaceutically tractable compounds. For the transcriptomic dataset (GSE139038), we applied an absolute log2 (fold change) > 2 together with an adjusted p-value <0.05 to focus on robust gene-expression changes. For KEGG pathway enrichment, Benjamini–Hochberg false discovery rate (FDR) correction was used, and pathways with FDR-adjusted p < 0.05 were considered significantly enriched.

The therapeutic targets' protein-protein interaction relationships were sourced from the STRING database (https://cn.string-db.org/), while Cytoscape software facilitated the integration, visualization, and analysis of biological networks. In the protein-protein interaction network, the core genes, which were identified as potential therapeutic targets, were prioritized using a comprehensive evaluation of various centrality measures. These measures included degree centrality, betweenness centrality, closeness centrality, Eigenvector centrality, and the local average connectivity-based method. The prioritization was performed using the CytoNCA plugin.

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were conducted to elucidate the biological relevance of therapeutic targets. For the biological process within the GO enrichment analysis, the ClueGo plugin was utilized, while the R package clusterProfiler was employed for KEGG pathway enrichment analysis (adjusted p-value <0.05). Ultimately, a network encompassing compounds, targets, pathways, and diseases was created with the help of Cytoscape software.

Molecular configurations of the core targets identified were ascertained through the Protein Data Bank (http://www.rcsb.org/) database, stored in Protein Data Bank format, and subsequently uploaded to AutoDockTools software (Velankar et al., 2021). In addition, the structure of the key candidate constituent for docking was sourced from the PubChem database and imported into AutoDockTools in the MOL2 format. PyMol, a robust visualization tool for proteins and molecules, was employed to modify the proteins by eliminating the original ligands and water molecules, as well as correcting the protein structures. A genetic algorithm based on Lamarckian principles was utilized for the calculation of binding energies.

The cell lines for TNBC, specifically MDA-MB-231, SUM-159, and 4T1, were obtained from the Cell Bank of the Chinese Academy of Sciences located in Shanghai, China. These cell lines were maintained in complete DMEM medium (Gibco, NY, United States), which was supplemented with 10% fetal bovine serum (Gibco, NY, United States) and 1% penicillin-streptomycin. They were kept in a cell culture incubator (Model 3111, Thermo, CA, United States) under a humidified environment with 5% CO2 at a temperature of 37 °C.

Cell viability was assessed utilizing the Cell Counting Kit-8 (Dojindo, Japan), in accordance with the protocol provided by the manufacturer. The cells were plated in 96-well plates at a concentration of 3000 cells per well. After an overnight incubation period, various concentrations of TKM were applied for either 24 or 48 h, with a blank control group included in the setup. The supernatant was subsequently discarded, and 100 µL of DMEM (without fetal bovine serum) combined with 10 μL of Cell Counting Kit-8 solution (Dojindo, Japan) was introduced into each well. After incubating for 30 min at 37 °C in the dark, absorbance readings were taken at 450 nm with a plate reader. The IC50 values of TKM were determined for 24 and 48 h across all 3 cell lines.

Cells undergoing logarithmic growth were plated in a 6-well dish at a density of 1500 cells per well and incubated overnight for the purpose of the colony formation assay. Following this, they were exposed to various concentrations of TKM for a duration of 2 weeks, with the media being changed every 3 days. After incubation, the cells were washed twice with PBS, fixed with paraformaldehyde for 30 min, and stained with 0.1% crystal violet solution for 30 min. The plates were dried overnight.

MDA-MB-231 cells were treated with TKM (0, 7.5, 15, or 30 mg/mL) for 24 h. The Annexin V APC Apoptosis Detection Kit (Biogems, United States) and Cell Cycle Assay Kit (Elabscience, China) were used to detect the effect of TKM on MDA-MB-231 cell apoptosis and the cell cycle, following the manufacturer’s instructions.

The Ethics Committee for Animal Experiments at Jining First People’s Hospital authorized the experimental protocol (NO. JNRM-2024-DW-075). Antitumor efficacy of TKM was investigated using 20 female BALB/c mice that had 4T1 tumors. Following a 5-day acclimatization period, each mouse received a subcutaneous injection of 1.0 × 10⁶ cells in PBS into the right armpit. The subsequent day, the mice with tumors were divided randomly into four groups: one control group (TC) receiving normal saline (0.2 mL/day), a doxorubicin (DOX) group (administered 2 mg/kg via tail vein injection daily starting from day four), and two groups treated with TKM at doses of 0.59 and 4.68 g/kg/day, beginning on the second day. The weight and volume of each tumor-injected mouse were recorded every other day from day four onwards. TKM treatment continued for a total of 14 days. Twelve hours after the last drug administration, the mice were euthanized, and the tumors were excised for further examination. Following the measurements, the tumor tissues were utilized for untargeted metabolomics analysis.

For untargeted metabolomics, differential metabolites were screened using a combination of orthogonal partial least squares–discriminant analysis (OPLS-DA) and univariate testing. Features with a variable importance in projection (VIP) > 1 and nominal p < 0.05 (two-sided t-test) were considered significantly altered, without applying an explicit fold-change cut-off. Benjamini–Hochberg FDR correction was applied at the level of KEGG pathway enrichment to control for multiple testing across pathways.

To validate amino-acid metabolism at the metabolite level, targeted LC–MS/MS analysis of amino acids and related metabolites in 4T1 TNBC cells treated with TKM or vehicle was performed by a commercial service provider (Shanghai Applied Protein Technology Co., Ltd., Shanghai, China). Briefly, 4T1 cell pellets were extracted with ice-cold methanol/acetonitrile (1:1, v/v) containing isotope-labeled internal standards, followed by low-temperature ultrasonication, protein precipitation at −20 °C, centrifugation, vacuum drying, and reconstitution in acetonitrile/water (1:1, v/v). The supernatants were analyzed on an Agilent 1290 Infinity UHPLC system coupled to an AB Sciex 6500+ QTRAP mass spectrometer equipped with a UPLC BEH Amide column (2.1 × 100 mm, 1.7 μm). Amino acids and related metabolites were separated using a hydrophilic-interaction liquid chromatography gradient with acetonitrile containing 0.1% formic acid and aqueous 10 mmol/L ammonium acetate containing 0.1% formic acid as mobile phases, and detected in positive/negative electrospray multiple-reaction monitoring mode. A pooled quality-control sample was injected at regular intervals to monitor system stability. Chromatographic peaks were integrated using MultiQuant software and metabolites were identified by comparison with authentic standards. Metabolites with a coefficient of variation <15% in quality-control samples were retained for statistical analysis, and differences between TKM- and vehicle-treated groups were assessed using Student’s t-test with p < 0.05 considered statistically significant.

Cell lysates were obtained through the use of RIPA assay lysis buffer. To determine protein concentration, a BCA assay kit was employed. The protein samples were subjected to separation via SDS-PAGE at a voltage of 70 V for 2 h, followed by transfer to PVDF membranes. These membranes were then incubated with 5% nonfat milk for 2 h at room temperature to block. After blocking, the membranes were incubated overnight at 4 °C with the following primary antibodies: PI3K, p-PI3K, AKT, p-AKT, β-catenin, and AKT1, each diluted at a ratio of 1:1000. After washing with TBST buffer, the membranes were treated with HRP-conjugated goat anti-rabbit antibody (1:20,000) for 2 h at room temperature. The protein bands were detected using the ECL kit on the ChemiDoc™ machine (BIO-RAD, United States). β-actin and vinculin acted as control proteins, and the expression levels of the other proteins were evaluated in relation to them.

An oligonucleotide synthesized chemically, which encodes a short hairpin RNA aimed at AKT1 (5′-GCT​ACT​TCC​TCC​TCA​AGA​A-3′), was introduced into a pPLKO.1-PURO vector (AZENTA, China). Similarly, a control plasmid was developed. Following the manufacturer’s instructions, the purified expression plasmid containing short hairpin RNA specific for AKT1 and the control plasmid were introduced into MDA-MB-231 cells using Lipofectamine™ 3000 (Invitrogen, United States). Two days post-transfection, drug selection commenced with 2 μg/mL puromycin (Sigma-Aldrich), leading to the harvesting and expansion of colonies into cell lines.

Surface plasmon resonance assays were performed to evaluate the binding of schisandrin to recombinant AKT1 protein. In this study, the recombinant AKT1 protein was obtained from Med Chem Express. It was then prepared in a sodium acetate buffer at pH 4.0, achieving a concentration of 20 μg/mL. AKT1 protein was immobilized onto a CM5 chip using amino coupling. Two-fold serial dilutions of schisandrin (100, 50, 25, 12.5, 6.25, and 3.125 μM) were prepared in PBS-P solution containing 5% DMSO to determine the K_D of AKT1 towards schisandrin. The K_D value was calculated using Biacore T100 Kinetics Summary Software (version 1.0.1).

All data are expressed as the mean ± standard deviation, and each experiment was conducted independently on three occasions. Significant differences were assessed using the Student’s t-test (*p < 0.05).

To explore the impact of TKM on TNBC cells, MDA-MB-231, SUM-159, and 4T1 cell lines were subjected to various concentrations of TKM over different time periods, and cell viability was assessed using Cell Counting Kit-8. Figures 1A–C illustrate that TKM treatment led to a dose- and time-dependent inhibition of growth in all three TNBC cell lines. Additional results from the colony formation assay further supported the anti-TNBC efficacy of TKM in vitro (Figure 1D).

To verify the impact of TKM on tumor growth in vivo, BALB/c mice received injections of 4T1 cells, followed by daily oral administration of TKM at doses of 0.59 and 4.68 g/kg. As illustrated in Figure 1E, the tumor control group displayed the largest average tumor volume, measuring 1.070 cm³. In contrast, treatment with TKM resulted in a significant inhibition of tumor progression, with reductions of 66.01% and 76.84% observed at the dosages of 0.59 and 4.68 g/kg, respectively. Additionally, throughout the treatment duration, the mice did not show any notable fluctuations in body weight (Figure 1F), suggesting that TKM did not have any toxic effects. A photograph of the tumor tissue taken at the end of the treatment is displayed in Figure 1G.

The chemical compounds of TKM were identified using LC-MS analysis. Supplementary Figure S1 shows the LC-MS total ion chromatogram of TKM. Compared to the mzCloud database, we identified 810 putative compounds, 293 of which had an mzCloud best-match score higher than 70 (Supplementary Table S1). In total, 151 compounds showed high GI absorption and had more than two drug-like properties, including Lipinski, Ghose, Veber, Egan, and Muegge, and thus, 1164 targets of the 151 compounds were selected for further investigation (Supplementary Figure S2).

By mining the Gene Expression Omnibus (GEO) database, 950 genes satisfying the filtering conditions were identified. Additionally, 714 known breast cancer-related genes were identified in the GeneCards database. By merging and deduplicating these two datasets, 1450 breast cancer-related targets were obtained (Supplementary Figure S3). A total of 214 intersectional targets were identified between the disease and drug targets (Figure 2A). These intersectional targets are regarded as therapeutic targets in TKM-treated TNBC. By querying the DisGeNet database, these treatment targets were observed to involve numerous attributes, which reflects TKM’s multi-faceted and multi-dimensional regulation of breast cancer.

Through computational analysis of degree centrality, betweenness centrality, eigenvector centrality, closeness centrality, and local average connectivity (Figure 2B), 28 core therapeutic targets were identified. They are shown in Figures 2C,D. Using the ClueGo plugin for GOBP functional annotation (Supplementary Figure S4.), we observed that the primary biological processes were “regulation of telomerase activity, positive regulation of glial cell proliferation, regulation of reactive oxygen species biosynthetic process, alpha-beta T cell lineage commitment, and negative regulation of cellular response to oxidative stress” (Figure 3A). KEGG pathway enrichment analysis was conducted using the R language package Cluster Profiler. The results indicate that TKM may help treat TNBC by “proteoglycans in cancer, Ras signaling pathway, PI3K/AKT signaling pathway, Wnt/β-catenin signaling pathway, and ErbB signaling pathway” (Figure 3B). GO, BP, and KEGG pathway analyses revealed that “MYC, TP53, JUN, CCND1, STAT3, and AKT1” may facilitate the treatment (Figures 3C,D). Subsequently, using the Cytoscape software, we established a compound-target-pathway-disease network (Supplementary Figure S5), revealing that schisandrin binds to more treatment targets (Supplementary Table S2), suggesting that it may be the key bioactive constituent for breast cancer treatment. The molecular docking performances were satisfactory (Figure 4).

Through network pharmacology analysis, we identified 28 key targets related to TNBC proliferation and survival and then focused our experimental validation on representative functional readouts rather than reiterating the full in silico workflow. Consistent with these predictions, TKM reduced the viability of MDA-MB-231, SUM-159, and 4T1 cells in a dose- and time-dependent manner, suppressed colony formation, induced apoptosis, and caused G2/M-phase cell-cycle arrest in MDA-MB-231 cells, indicating a robust antiproliferative effect on TNBC cells (Figure 5).

KEGG pathway analysis indicated enrichment of the PI3K/AKT and Wnt/β-catenin pathways, suggesting that these signaling axes may be involved in the cellular mechanisms by which TKM inhibits TNBC. To functionally validate these predictions, we performed Western blotting in MDA-MB-231 cells. TKM reduced the phosphorylation of PI3K and AKT and decreased β-catenin levels, and, importantly, it increased the pro-apoptotic protein BAX while decreasing the anti-apoptotic protein BCL2, thereby elevating the BAX/BCL2 ratio. In parallel, TKM treatment markedly reduced the expression of the oncogenic transcription factor c-MYC, a key downstream effector of Wnt/β-catenin signaling (Figures 6A,B). These findings indicate that TKM not only modulates PI3K/AKT and Wnt/β-catenin at the level of core kinases but also suppresses their downstream prosurvival and proliferative outputs in TNBC cells. Network pharmacology analysis identified schisandrin as a candidate active component, and molecular docking suggested that it can interact with AKT1 (Figure 4D). Surface plasmon resonance experiments provided additional support that schisandrin binds to recombinant AKT1 protein with a K_D of 1.525 × 10⁻⁴ M (Figure 6C). Moreover, schisandrin inhibited the proliferation of control MDA-MB-231 cells more strongly than that of AKT1-knockdown cells (Figures 6D,E), supporting AKT1 as one of the functional effectors engaged by schisandrin in this system.

To determine if TKM affects TNBC by influencing the metabolism of tumor tissues, an untargeted metabolomics strategy utilizing LC‒Q-TOF/MS was employed to examine the metabolic variations in tumor tissues treated with and without TKM. A supervised OPLS-DA model was created (Figures 7A,C) to pinpoint the elements responsible for group distinctions. The R² values for the positive and negative MS ion modes in the OPLS-DA model were recorded at 0.997 and 0.992, while the Q² values were 0.563 and 0.57, respectively. Differential metabolites were identified from all variables that fulfilled the criteria of VIP >1 in OPLS-DA and p < 0.05 based on the t-test (Supplementary Tables S4, 5). We then performed hierarchical clustering of metabolites potentially influenced by TKM across both positive and negative ion modes (Figures 7B,D). Furthermore, the pathways of the differential metabolites were analyzed as potential target pathways (Figure 7E). Within this context, representative metabolic pathways include the biosynthesis of amino acids; glycine, serine, and threonine metabolism; mineral absorption; and protein digestion and absorption. Alterations in central carbon metabolism in cancer, which aligned with the results of network pharmacology, were also detected. KEGG pathway enrichment of these differential metabolites was analyzed using Benjamini–Hochberg FDR correction, and pathways with FDR-adjusted p < 0.05 were regarded as significantly altered.

To validate the tumor-tissue metabolomics at the metabolite level in an independent system, we additionally performed targeted LC–MS/MS profiling of more than 80 amino acids and related metabolites in 4T1 TNBC cells treated with TKM versus vehicle. Quality-control samples showed a mean coefficient of variation of approximately 4.8%, with more than 95% of quantified metabolites exhibiting coefficient of variation (CV) < 15%, indicating good analytical robustness. Approximately 30 metabolites were significantly altered (p < 0.05) in TKM-treated cells, including decreases in glycine, threonine, phosphoserine, and asparagine and increases in glutamine and aspartic acid, in line with the amino-acid metabolism pathways highlighted by the untargeted tumor metabolomics. These targeted data further support amino-acid metabolic reprogramming as an important downstream effect of TKM in TNBC and are provided in detail in the Supplementary Table S6.