Patient cohorts and study design

We utilized the FUSCC dataset, consisting of 465 cases, 360 of which had transcriptomic data, 279 samples with whole-exome sequencing (WES) results, and 401 samples with somatic copy-number alteration (SCNA). In this cohort, we analyzed and compared the clinical and pathological characteristics between LAR subtype and non-LAR subtype patients. Our findings revealed significant differences between the two subtypes in terms of age at onset, menopausal status, histological grade, Ki67 proliferation index and lymph node status (Table 1). Notably, LAR TNBCs exhibited higher histological grade, lower Ki67 proliferation index and a higher frequency of lymph node metastasis.

Table 1 Clinicopathological characteristic of LAR vs non-LAR patients in the FUSCC cohort

Based on different clinical and molecular characteristics, a specific targeted therapeutic strategy was proposed for the LAR subtype without ERBB2 mutation in the FUTURE clinical trial (anti-AR plus anti−CDK4/6). However, the trial results showed that only one of eight assessable patients presented with stable disease (SD), while the remaining seven patients displayed progressive disease (PD) [7]. This finding suggests that combining anti-AR therapy with other targeted therapies might be more appropriate. Therefore, in this study, we conducted a comprehensive analysis of multiomics data within the FUSCC TNBC cohorts. Then, we performed corresponding in vitro and in vivo experiments to explore other potential therapeutic strategies, as well as the underlying mechanisms (Fig. 1).

Fig. 1

Workflow of the analytical process conducted in this study. The analytical process performed in this study followed a structured framework: This study utilized multi-omics cohort from FUSCC to identify potential therapeutic targets for the LAR subtype. The efficacy of treatment strategies and potential molecular mechanisms were investigated through in vivo in vitro ex vivo drug sensitivity experiments, as well as RNA sequencing. The aim of this study was to propose potential clinical treatment strategies specifically for the LAR subtype. LAR luminal androgen receptor, IM immunomodulatory, BLIS basal-like and immune-suppressed, MES mesenchymal-like, FUSCC Fudan University Shanghai Cancer Center

Patients with LAR TNBC are subject to epigenetic regulation

Initially, we performed differential gene analysis using epigenetic-related genes (ERGs) obtained from the EpiFactors database [30]. We found significant differences between LAR and non-LAR subtype patients (Fig. 2A). Gene Ontology (GO) enrichment analysis also highlighted chromosome organization and histone modification as prominent biological processes associated with the LAR subtype (Fig. 2B). Additionally, KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway analysis demonstrated a greater connection between the LAR subtype and epigenetic-related pathways such as transcriptional misregulation, cell cycle, and human immunodeficiency (Fig. 2C). Then, we developed a weighted gene coexpression network analysis (WGCNA) to identify a coexpression model for the ERGs in the LAR subtype (Fig. 2D,E). Next, we conducted GO and KEGG analyses on the hub genes. Likewise, this analysis revealed enrichment of epigenetic-related pathways, such as cell cycle, chromatin organization and chromatin remodeling processes, within the LAR subtype (Fig. 2F). A PPI (Protein–Protein Interaction) network was constructed to explore the coexpression proteins. Through functional enrichment analysis, we found that this network was mainly associated with histone kinase activity and the cell cycle, which was consistent with our previous results (Fig. 2G).

Fig. 2

Patients with LAR TNBC are subject to epigenetic regulation. A Heatmap showing the top 30 differentially expressed epigenetic-related genes (ERGs) filtered by P  <  0.05 and log2FC > 1. B–C Enrichment of ERGs in signaling pathways as determined by GO (B) and KEGG analysis (C). D–E Construction of ERG WGCNA network. The demonstration of module feature vector clustering (D). Correlation analysis between each module and TNBC subtype feature (E). Color labels are exclusively employed to differentiate between various gene modules and hold no intrinsic significance. F Functional annotation of hub genes in the LAR subtype processed by GO enrichment analysis. G PPI network showing coexpressed proteins of WCGNA hub genes. H ConsensusCluster determined by ERGs in the FUSCC TNBC cohort. I The Sankey diagram displaying relationships among TNBC subtypes and gene clusters. J Chord plot demonstrating the top 10 enriched signaling pathways. K Copy number alterations (left) and transcriptional expression levels (right) of representative HDAC genes. L Univariate Cox regression analysis of ERGs in the LAR subtype of TNBC. Abbreviations: LAR, luminal androgen receptor; TNBC: Triple-Negative Breast Cancer; WGCNA: Weighted Gene Co-expression Network Analysis; ERGs: Epigenetic-Related Genes; HDAC: Histone Deacetylases. Statistical analysis was performed by the Student’s t test (E, L, K) or Mann–Whitney test (B, C, J). ****Indicates P < 0.0001; ***indicates P < 0.001; **indicates P < 0.01; *indicates P < 0.05; ns indicates no significance

To further verify these findings, we employed ConsensusCluster to stratify the FUSCC TNBC dataset with epigenetic-related genes, resulting in the identification of three clusters: cluster 1 (n = 128), cluster 2 (n = 109), and cluster 3 (n = 123). Notably, a majority of LAR-subtype patients were categorized under cluster 2 (Fig. 2H, I). This cluster was enriched with the cell cycle, transcriptional misregulation in cancer, and the necroptosis pathways, which is consistent with the enriched pathways in the LAR subtype (Fig. 2J).

Next, our focus shifted toward exploring key epigenetic-related regulators in the LAR subtype, which might be potential drivers and therapeutic targets. As previously described, HDACs play a vital role in epigenetic regulation [8]. Interestingly, the LAR subtype of TNBC exhibited a lower frequency of loss/deletion or a higher frequency of gain/amplification in several HDACs such as HDAC1, HDAC3, HDAC8, and HDAC10 (Fig. 2K). Consistently, LAR TNBCs had elevated transcriptional levels of most of the aforementioned HDACs. Moreover, using univariate Cox regression analysis, histone-related genes, including HDACs, predominantly predicted a higher risk of recurrence in the LAR subtype (Fig. 2L). This suggests that HDACs could serve as promising targets for LAR TNBCs.

Taken together, epigenetic regulation has a greater significance in the LAR subtype of TNBC, particularly in the context of histone acetylation and deacetylation. Therefore, LAR TNBCs may have a higher potential for benefiting from therapeutic interventions targeting epigenetic regulators, such as HDACs.

Synergistic effect of chidamide combined with enzalutamide in vitro

In this study, we chose chidamide, a subtype-selective HDAC inhibitor of HDAC 1/2/3/10 [31]. As a candidate drug compatible with the anti-AR agent enzalutamide for treating LAR TNBC. It is worth noting that chemotherapy is the main strategy for TNBC, and LAR TNBCs are characterized by a high frequency of PI3K mutations [6]. Therefore, we selected the PI3Ki alpelisib and paclitaxel as controls.

We conducted growth inhibition assays to investigate the drug sensitivity of MDA-MB-453 and CAL-148 cell lines, both categorized as the LAR subtype. Our findings indicate that MDA-MB-453 cells were significantly inhibited by chidamide and showed overall sensitivity to all tested drugs (Table 2). The IC50 values were 1.09 μM, 18.67 μM, 4.12 nM, and 0.77 μM for chidamide, enzalutamide, paclitaxel, and alpelisib, respectively (Fig. 3A, Table 2). CAL-148 cells exhibited increased sensitivity to chidamide, paclitaxel, and alpelisib but demonstrated resistance to enzalutamide, with IC50 values of 2.40 μM, 3.9 nM, 1.09 μM, and 149.3 μM for each drug, respectively (Fig. 3B, Table 2). Interestingly, we demonstrated that both MDA-MB-453 and CAL-148 cells exhibited high sensitivity to chidamide through the mono-drug IC50 assay and confirmed that the non-LAR cell lines exhibited relatively lower sensitivity to chidamide than did the LAR cell lines (Additional file 1: Fig. S1A, B).

Table 2 IC50 of a single drug in LAR cell linesFig. 3

Synergistic effect of chidamide combined with enzalutamide in vitro. A, B Cell viability of MDA-MB-453 and CAL-148 cell lines treated with chidamide or enzalutamide. The IC50 values are shown in the left corner of each figure. C Heatmap showing the cell viability and CI value of chidamide and enzalutamide in MDA-MB-453 and CAL-148 cell lines at different drug concentrations

Then, we investigated the potential synergistic effects of anti-AR therapy in combination with chidamide treatment using drug combination assays and the Chou-Talalay method [32]. A combination index (CI) > 1 indicated an antagonistic effect between the two drugs, while a CI < 1 suggested a synergistic effect. In both MDA-MB-453 and CAL-148 cells, combining low concentrations of chidamide with enzalutamide demonstrated significant synergy, resulting in a 20−50% decrease in growth rate (CI = 0.2, 0.21, respectively) (Fig. 3C). However, the synergistic effects of chidamide in combination with paclitaxel or alpelisib were strongly correlated with drug concentration and cell lines. In MDA-MB-453 cells, synergistic effects were observed only when combining chidamide with high concentrations of paclitaxel (3.7 nM, CI = 0.6) or high concentrations of alpelisib (5.56 μM, CI = 0.6) (Additional file 1: Fig. S1C, D). In CAL-148 cells, chidamide combined with paclitaxel slightly decreased the growth rate, indicating antagonism between the two drugs. Similarly, the combination of chidamide and alpelisib exhibited antagonism, except for the high concentration of alpelisib in CAL-148 cells (Additional file 1: Fig. S1E, F).

Overall, we discovered that chidamide had a more pronounced inhibitory effect on the proliferation of LAR TNBC cell lines than other monotherapies. Furthermore, the combination of chidamide and enzalutamide exhibited a significant synergistic effect.

Synergistic effect of chidamide combined with enzalutamide in vivo and ex vivo

Next, we explored the efficacy and toxicity of chidamide and enzalutamide, both alone and in combination through in vivo experiments. The combination of chidamide and enzalutamide resulted in greater tumor regression in TS/A (mouse LAR tumor cell line) allograft models than treatment with chidamide or enzalutamide alone (Fig. 4A–C). Specifically, we observed a 59% decrease in tumor volumes after treatment with chidamide, a 25% decrease after enzalutamide treatment, and an 84% decrease after the combination treatment. Additionally, we found that tumor weights decreased by 18%, 6%, and 72% following treatment with chidamide, enzalutamide, and the combination, respectively. Importantly, there were no significant differences in body weight loss between the combination group and the groups treated with single agents, indicating the acceptable toxicity of the combination treatment (Fig. 4D). Furthermore, the synergetic effect of chidamide and enzalutamide was confirmed using LAR-subtype patient-derived organoids (PDOs) (Fig. 4E).

Fig. 4

Synergistic effect of chidamide combined with enzalutamide in vivo and ex vivo. A-D Representative tumor images A, tumor growth curves B, endpoint tumor weight C bar plots and endpoint mouse weight bar plots (D) for different treatment groups. E Representative brightfield images of patient-derived organoids (PDOs) on day 5 after drug treatment. PDO models of LAR and non-LAR subtypes were treated with DMSO (control), 1 μM chidamide, 1 μM enzalutamide or the combination therapy for 5 days. Scale bar, 200 μm.Viability was calculated by each group compared with group blank (PDOs without any additional treatment). Abbreviations: LAR, luminal androgen receptor. The data are presented as the mean ± SEM and were compared using Student’s t test (B-D) or Kruskal-Wallis test (E) ****indicates P < 0.0001; ***indicates P < 0.001; **indicates P < 0.01; *indicates P < 0.05; ns indicates no significance

In summary, our in vivo experiments demonstrated that the combination of chidamide and enzalutamide led to superior tumor regression. This synergetic effect was further validated ex vivo using LAR-subtype PDOs.

Potential mechanisms of the synergistic effect of chidamide combined with enzalutamide

To elucidate the mechanism associated with the synergistic effect of chidamide and enzalutamide in LAR-subtype TNBC, we conducted RNA sequencing of tumors from the TS/A models. Initially, using consensus clustering, we identified three distinct clusters that exhibited specific associations with different treatment groups (Fig. 5A, B). Specifically, the chidamide group was assigned to cluster 1, while the combination therapy group exclusively fell within cluster 3. This correspondence indicates the presence of distinct mechanisms governing tumor inhibition between the chidamide monotherapy and combination therapy groups. Additionally, resembling the control group, certain samples from the enzalutamide group were affiliated with cluster 2, thereby partially explaining the limited efficacy of enzalutamide treatment. To further elucidate the potential tumor-inhibiting mechanisms of different treatments, we explored the typical biological features of various clusters by GSVA (Gene Set Variation Analysis) enrichment analysis. Cluster 1 was primarily enriched in the cell cycle, cellular senescence and spliceosome, suggesting that these biological processes potentially mediate tumor inhibition led by chidamide monotherapy (Fig. 5C). In comparison to the other two clusters, Cluster 3 exhibited significant enrichment in autophagy, apoptosis and diverse metabolic pathways such as fatty acid metabolism and glyoxylate and dicarboxylate metabolism (Fig. 5D). This suggests that combination therapy may predominantly kill tumor cells by regulating programmed cell death and metabolic pathways. Subsequently, we determined the differentially expressed genes (DEGs) in the combination therapy group compared to the chidamide group and enzalutamide group separately using a cutoff of p value  <  0.05 and log2FC ≥ 1.5. Through KEGG and GO pathway analyses, combination therapy predominantly affects metabolic pathways, as well as autophagy, which aligns consistently with the aforementioned cluster comparison results (Fig. 5E, F).

Fig. 5

Potential mechanisms of the synergistic effect of chidamide combined with enzalutamide. A ConsensusCluster determined by ERGs in TS/A tumor samples. B The Sankey diagram displaying relationships among different treatment groups and gene clusters. C, D GSVA analysis of enriched pathways in cluster 2 compared with cluster 1 (C) and cluster 3 (D). E, F KEGG (E) and GO (F) analysis of differentially expressed genes among different treatment groups. G GSVA analysis of AR and steroid synthesis, and the P53 pathway among different treatment groups. H GSEA analysis of differentially expressed genes among different treatment groups. I Bar plots showing the percentage of CD3e+ cells among CD45+ cells, CD4+ cells among CD45+ T cells, CD8+ cells among CD45+ T cells, PRF1+ cells among CD3e+ CD45+ CD8+ T cells, CD206+ cells among CD45+ CD11b+ F4/80+ cells, and CD86+ cells among CD45+ CD11b+ F4/80+ cells. Data were compared using Student’s t test (I) or Mann–Whitney test (C–F)

As previously reported, the LAR subtype is characterized by AR positivity, as well as dysregulated cell-cycle signaling [6]. Such findings provide a rationale for the use of AR antagonists and CDK 4/6 inhibitors as a therapeutic strategy for LAR TNBC. However, this approach did not achieve the expected results [7]. Thus, our investigation aimed to determine whether combination therapy enhances the inhibition of the cell cycle-related pathways and the AR pathway, thereby increasing therapeutic efficacy. Strikingly, the combination therapy group exhibited a significant upregulation of P53 signaling, as well as downregulation of RB and overall cell cycle pathways (Fig. 5G and Additional file 2: Fig. S2A). Likewise, the combination therapy group exhibited distinct patterns compared to the other two monotherapy groups when examining the AR pathway and its downstream signaling through GSVA analysis. Not surprisingly, enzalutamide markedly suppressed AR signaling. In contrast, chidamide monotherapy had a relatively weak impact on AR signaling, while the combination therapy exhibited downregulation of AR and steroid synthesis (Fig. 5G, Additional file 2: Fig. S2A). These results were also confirmed through GSEA (Gene Set Enrichment Analysis) using the above identified DEGs (Fig. 5H). Additionally, GSEA results demonstrated a significant upregulation in the apical junction, along with a downregulation in mTORC1 signaling (Fig. 5H).

Finally, we investigated whether the tumor microenvironment (TME) is involved in the distinct antitumor effects of these treatment strategies. Although the ESTIMATE analysis results demonstrated a lower immune score and stromal score in the combination group, there was no statistically significant difference compared with single drug group (Additional file 2: Fig. S2B). Using flow cytometry and CIBERSORT, we also observed little alteration regarding immune cell composition (Fig. 5I, Additional file 2: Fig. S2C). These findings indicate that the inhibition of tumor proliferation by combination therapy may not mainly rely on the immune response.

In conclusion, the combination of chidamide and enzalutamide primarily inhibits tumor proliferation by regulating metabolism, particularly fatty acid metabolism and autophagy. Moreover, the effectiveness of combination therapy, which exhibits a more pronounced effect than solely antagonizing AR, may be attributed to the suppression of the steroid synthesis and cell cycle-related pathways. Additionally, the antitumor tumor microenvironment appears to have limited impacts on the synergistic effect of the combination therapy.