IntroductionProstate cancer is the most prevalent malignant tumor among men in Europe and the United States, ranking as the second leading cause of male cancer-related deaths [1]. This disease is highly heterogeneous, with its progression linked to multiple gene deletions or mutations, including FOXA1, ZNF292, CHD1, PTEN, and TP53 [2, 3]. Prostate growth and development are dependent on androgens, and androgen deprivation therapy (ADT) remains the cornerstone of current prostate cancer treatment [4]. However, due to the complex heterogeneity of prostate cancer, the therapeutic efficacy of ADT can vary significantly, thereby affecting patient prognosis. Furthermore, nearly all patients eventually develop resistance to castration therapy after a certain treatment period [5]. The mechanisms underlying castration resistance are extremely complex and ultimately result in the failure of treatments targeting the androgen receptor (AR) signaling pathway. Therefore, there is an urgent need to identify novel therapeutic targets for prostate cancer.Metabolic reprogramming is a distinct characteristic of prostate cancer. Normal prostate tissue exhibits unique metabolic features, notably the secretion of large amounts of citric acid, a semen component that is abundantly synthesized in prostate cells [6, 7]. Citric acid serves as a crucial hub linking cell metabolism, particularly by mediating lipid metabolism through the amphibolic pathway. Consequently, the substantial secretion of citric acid from normal prostate tissues can be utilized to synthesize fatty acids and cholesterol during cancer development [8]. On one hand, the metabolic shift in prostate cancer enables cancer cells to adapt to their energy requirements, while simultaneously, the utilization of cholesterol for the synthesis of endogenous steroid hormones promotes the proliferation of prostate cancer cells [9]. Consequently, exploring the metabolic vulnerabilities in prostate cancer may provide a direction for novel treatment strategies.Ferroptosis is a unique form of iron-dependent cell death characterized by the accumulation of excessive lipid peroxidation in the cell membrane [10]. Resistance to ferroptosis not only facilitates tumor development but also contributes to tumor resistance to treatment [11]. Multiple mechanisms are involved in ferroptosis resistance in prostate cancer. Liang et al. first reported that AR induce ferroptosis resistance in prostate cancer by regulating the activity of MBOAT2 [12]. Yi et al. demonstrated that SREBPs activation mediated by the PI3K-AKT-mTOR pathway can promote ferroptosis resistance in prostate cancer [13]. However, the relationship between SREBPs-mediated metabolic reprogramming and ferroptosis resistance in prostate cancer remains unclear.We confirmed the association between SREBF1-mediated metabolic reprogramming in prostate cancer and ferroptosis in human samples, using a combination of single-cell sequencing and Bulk-RNA analysis. Additionally, we validated SREBF1 as a potential therapeutic vulnerability and an effective target for prostate cancer by employing SREBF1 inhibitors.ResultsOverview of single-cell sequencing characteristics of normal prostate tissue and different types of prostate cancersOur study included normal prostate tissue cells, prostate cancer cells from radical prostatectomy (RP) samples representing primary cancer, and prostate cancer cells from castration-resistant prostate cancer (CRPC) samples. Following quality control (QC), a total of 51,092 cells were included in the study. Subsequently, all cells were divided into 33 clusters by setting the resolution to 0.6. These clusters were annotated based on characteristic gene expression differences, identifying T cells, B cells, Macrophages, Endothelial cells, Fibroblasts, Mast cells, Monocytes, and Epithelial cells (Fig. 1A). It is noteworthy that regardless of the sample origin, epithelial cells constituted the predominant cell type, accounting for 32.54%, 37.56%, and 59.37% of normal, primary cancer, and CRPC samples, respectively. Interestingly, in CRPC-derived samples, the proportion of T cells significantly decreased, accounting for only 10.39% of all cells, compared to 29.55% and 37.22% in normal and primary cancer samples, respectively (Fig. 1B, C). Differential gene expression analysis identified several highly expressed genes in epithelial cells, including KLK3, PRAC1, and TSPAN1. KLK3, which encodes the prostate-specific antigen (PSA), is specifically expressed in prostate tissues, and is widely used in prostate cancer screening. Prostate Cancer Susceptibility Candidate Protein 1(PRAC1) is highly expressed in the prostate [14]. TSPAN1 is upregulated in various cancers and is regulated by androgens, promoting the proliferation and migration of prostate cancer [15] (Fig. 1D). Furthermore, we identified characteristic differences between different clusters through Gene Set Variation Analysis (GSVA) of the average expression levels. Epithelial cells exhibited higher activity in pathways, such as the PI3K-AKT pathway, P53 pathway, MYC, androgen response, glycolysis, fatty acid metabolism, and bile acid metabolism. T cells showed high activity in the interferon, IL-2, and IL-6 signaling pathways, whereas fibroblasts exhibited uniquely high activity in the epithelial-mesenchymal transition pathway (Fig. 1E).Fig. 1: Overview of single-cell RNA sequencing characteristics of prostate cancer.A tSNE plot showing cell clusters after dimensionality reduction and cell type annotation following quality control. B tSNE plot displaying facet diagrams from Normal, Primary cancer, and CRPC samples. C Distribution of different cell types in Normal, primary cancer, and CRPC samples. D Heatmap of the top 10 differentially expressed genes for each cell type. E Heatmap of Gene Set Variation Analysis (GSVA) based on the average gene expression for each cell type. Gene set: Hallmark from MSigDB.Full size imageCharacteristics of different types of epithelial cellsGiven the lipid-related metabolic signatures observed in epithelial cells, we further analyzed the epithelial cells (Fig. 2B). Citrate, as a hub linking glycolysis and lipid metabolism, including fatty acid and cholesterol synthesis, plays an important role in prostate cancer (Fig. 2A). Through gene differential analysis, we identified significant gene expression differences in epithelial cells from normal, primary cancer, and CRPC samples. In the primary cancer samples, the expression of PCA3, ERG, NPY, and AMACR was significantly upregulated, making them the characteristic genes with the largest expression differences. PCA3, a long non-coding RNA (lncRNA), is highly expressed specifically in prostate cancer and has been used for urine detection of prostate cancer [16]. The ERG gene serves as a prostate cancer marker, and the ERG-TMPRSS gene fusion is one of the most common gene rearrangements in prostate cancer [17]. AMACR holds significant value in the pathological diagnosis of prostate cancer, serving as a characteristic marker, and its expression is closely related to the fatty acid metabolism in prostate cancer [18]. ATP-related genes were significantly upregulated in the CRPC cells (Fig. 2C). Subsequently, we observed significant differences in the characteristics of the epithelial cells derived from these three sources using GSVA. In normal samples, some inflammation-related pathways showed higher activity, whereas epithelial cells derived from primary cancer samples mainly exhibited higher activity in androgen response, fatty acid metabolism, cholesterol metabolism, and other pathways. Epithelial cells derived from the CRPC samples were primarily concentrated in the E2F, MYC, and DNA repair pathways. Epithelial cells derived from primary cancer and CRPC samples showed higher activity in the glycolysis and MTORC1 pathways (Fig. 2D). Overexpression of ACLY, a key gene in the flow of citric acid to lipid metabolism, was observed in epithelial cells derived from tumor samples. The average expression levels in primary cancer samples were 0.6657, while in CRPC samples, it was 0.3280, and in normal samples, it was only 0.1954. FASN, a key gene in mediating fatty acid metabolism, also exhibited high expression levels in primary cancer samples and CRPC samples, with values of 0.6295 and 0.6141, respectively, compared to 0.1640 in normal samples. SCD, which mediates the formation of monounsaturated fatty acids (MUFA) and is related to ferroptosis resistance, showed expression levels of 0.7122, 0.2897, and 0.1811 in primary cancer, CRPC, and normal samples, respectively (Fig. 2E).Fig. 2: Characteristic differences in epithelial cells of prostate tissue and different types of prostate cancer.A Schematic diagram of intracellular energy metabolism, with highlighted products and enzymes. B tSNE plot of epithelial cells, colored by sample types: Normal, Primary cancer, and CRPC. C Differential gene analysis in epithelial cells from Normal, Primary cancer, and CRPC samples, showing genes with the largest average log fold change (LogFC). D Heatmap displaying the average gene set variation analysis (GSVA) scores in epithelial cells from Normal, Primary cancer, and CRPC samples. The gene set used is Hallmark from MSigDB. E Violin plot showing the expression levels of ACLY, FASN, and SCD genes in epithelial cells from Normal, Primary cancer, and CRPC samples.Full size imageTranscription factor regulatory network of prostate epithelial cellsWe analyzed the transcription factor activity in epithelial cells using SCENIC, resulting in the identification of 219 transcription factors. Among these, 30 transcription factors exhibited a higher activity based on an RSS value greater than 0.2 and a Z value greater than 1.4. These included SREBF1, SREBF2, and FOXA1 (Table S1). The average area under the curve (AUC) of the regulons in each group was calculated. The activities of SREBF1 and FOXA1 were higher in the primary cancer samples, while the activities of FOXC1 and TP73 were higher in normal samples, and the activities of transcription factors such as FOXA3 were higher in the CRPC samples (Fig. 3A). The RSS values of SREBF1 were higher in primary cancer and CRPC samples, with values of 0.32 and 0.40, respectively, while the value was only 0.21 in the Normal sample (Fig. 3B). We visualized the target genes of SREBF1, including HMGCS1, DHCR7, SC5D, SCD1, ACLY, FASN, and LDLR. (Fig. 3C and Table S2). Furthermore, the transcriptional activity of SREBF1 was higher in primary cancer samples (Fig. 3D, E).Fig. 3: Analysis of transcription factors in prostate cancer epithelial cells.A Heatmap showing the average Area Under the Curve (AUC) values of 30 transcription factors. B RSS plot of 30 transcription factors. Color represents the Z value, and the size of the circle represents the RSS value. C Visualization of SREBF1 and its target genes. D Violin plot showing the AUC values of SREBF1 in epithelial cells from Normal, Primary cancer, and CRPC samples. E tSNE plot displaying the expression of SREBF1 in epithelial cells from Normal, Primary cancer, and CRPC samples. Color represents the AUC value.Full size imageCharacteristic differences based on SREBF1 transcriptional activity groupingWe divided all cells into SREBF1-positive and SREBF1-negative groups based on the binary regulon AUC matrix in SCENIC, and analyzed the differences between the two groups (Fig. 4A). Notably, “positive” and “negative” mentioned above do not represent the presence or absence of SREBF1 activity, but only the classification of the binary results of regulon AUC, representing the level of SREBF1 activity in cells. We conducted GSVA on the two groups and found significant differences in the pathways related to cholesterol and fatty acid metabolism. Among them, the “Cholesterol Metabolism with Bloch and Kandutsch-Russell Pathways” showed the most obvious enrichment difference (t Value: 60.78), while in fatty acid metabolism, “Omega9 Fatty Acid Synthesis” ranked second (t Value: 55.68). Both the Bloch and Kandutsch-Russell pathways are involved in cholesterol synthesis. Omega 9 Fatty Acid (MUFA) Synthesis is associated with anti-ferroptosis and is mainly regulated by SCD. Other pathways, such as the “Mevalonate Arm of Cholesterol Biosynthesis Pathway” (t value: 30.43) and “Mevalonate Pathway” (t Value: 27.68), also showed significant differences. Cholesterol can synthesize endogenous androgens to promote prostate cancer cell proliferation, and high activity of the Mevalonate (MVA) pathway is associated with resistance to ferroptosis (Fig. 4B, D). We compared differences in the expression of some SREBF1 targeted genes related to ferroptosis between the two groups. SCD, FASN, and ACLY were significantly upregulated in SREBF1-positive cells, while cholesterol-related LDLR and DHCR7 were also significantly up-regulated in SREBF1-positive cells. Additionally, we observed the up-regulation of AR and MBOAT2, which are involved in the anti-ferroptosis mechanism of MBOAT2, in SREBF1-positive cells (Fig. 4C).Fig. 4: Differences between the two groups SREBF1_POS and SREBF1_NEG.A Volcano plot showing the differential gene expression analysis between SREBF1_POS and SREBF1_NEG groups. B Gene Set Variation Analysis (GSVA) comparing the SREBF1_POS and SREBF1_NEG groups. The gene set used is CP:WIKIPATHWAYS from MSigDB. C Violin plot displaying the expression levels of selected genes between SREBF1_POS and SREBF1_NEG groups. D Pathway map of ferroptosis and lipid metabolism. E Schematic diagram of cholesterol synthesis, from acetyl-CoA to cholesterol. F Dot plot showing the expression levels of cholesterol synthesis-related genes between SREBF1_POS and SREBF1_NEG groups.Full size imageFurthermore, we focused on the gene expression differences in the cholesterol synthesis pathway, from acetyl-CoA to cholesterol (Fig. 4E). Almost all genes were highly expressed in the SREBF1-positive group, with a higher expression ratio. Only PMVK and FDPS showed higher average expression levels in the SREBF1-negative cells, but their expression percentages were lower. This indicated that the cholesterol synthesis pathway was highly activated in the SREBF1-positive group, with an average expression percentage of genes in 7-dehydrocholesterol (7-DHC) synthesis, including EBP (81.05%), SC5D (95.72%), and MSMO1 (73.30%), which was significantly higher than that of DHCR7 (54.94%) (Fig. 4F).Single-cell sequencing hdWGCNA identifies modules associated with SREBF1 and their characteristicsWe performed hdWGCNA analysis of epithelial cell subsets. First, the topological overlap matrix (TOM) was calculated by selecting an optimal soft power of 10 (Figs. 5A and S1A). The eigenvalues between each module were then calculated for correlation analyses between modules (Fig. 5B) and between modules and traits. Highly connected genes within each module were determined by calculating the eigengene-based connectivity, kME (Fig. 5E). We performed a correlation analysis between the modules and single-cell sequencing data features. Single-cell data features were the activity scores (AUCell scores) of the transcription factors identified by SCENIC in each cell (Fig. 5C). Among them, the module with the strongest correlation with SREBF1 activity was epithelial cells-M1, with correlation coefficients of 0.91, 0.84, and 0.39 in the Normal, Primary cancer, and CRPC groups, respectively (Figs. 5C and S1B, C, D). The hub genes of Epithelial cells-M1 are shown in Fig. 5D. KEGG pathway enrichment analysis of the hub genes in Epithelial cells-M1 showed significant enrichment in ferroptosis and fatty acid metabolism-related pathways (Fig. 5F).Fig. 5: The relationship between SREBF1 and ferroptosis revealed by hdWGCNA.A Dendrogram of hdWGCNA in prostate cancer epithelial cells. B Correlation diagram between modules identified by hdWGCNA. C Correlation analysis between the modules identified by hdWGCNA and the transcription factor activities identified by SCENIC. D Top 40 hub genes of Epithelial cells-M1 module. E Hubgenes of all modules identified by hdWGCNA, ranked by kME. F KEGG pathway enrichment analysis of hub genes in the Epithelial cells-M1 module.Full size imageConstruction of a biochemical recurrence risk score for prostate cancer based on SREBF1 target genes combined with bulk RNA seq analysisWe performed a correlation analysis between all genes in the TCGA prostate cancer samples and SREBF1, and the results are shown in Fig. 6A. Among these, SCD1 showed the strongest correlation with SREBF1, with a correlation coefficient of 0.77. SCD-mediated MUFA synthesis is associated with anti-ferroptosis. Additionally, some genes directly related to cholesterol synthesis also have a high correlation, such as IDI1, LSS, and DHCR24. SREBF2 and SREBF1, both parts of the SREBPs family related to cholesterol metabolism, are often discussed together. ACAT2, HMGCS1, and HMGCR, key genes related to the MVK pathway in cholesterol synthesis, were highly correlated with SREBF1 (Fig. 6A and Table S3). Additionally, we used GSVA to analyze the activity of related gene sets in prostate cancer samples. Prostate cancer samples were divided into two groups based on the expression level of SREBF1. The GSVA score results showed that in the SREBF1 high expression group, the omega-9 fatty acid and the cholesterol synthesis pathways had higher activity (Fig. 6B).Fig. 6: Bulk-RNA seq analysis based on SREBF1 and its target genes and risk scoring model associated with biochemical recurrence of prostate cancer.A Histogram of correlation analysis between all genes and SREBF1 expression in TCGA-PRAD. B Violin plot of gene sets GSVA scores. C Risk forest plot of genes associated with biochemical recurrence of prostate cancer obtained by univariate Cox regression analysis. D Coefficient path diagram of genes in Lasso regression analysis. E Cross-validation curve in Lasso regression analysis, nfolds = 10. F Distribution of risk scores and biochemical recurrence characteristics of samples in TCGA cohort. G Distribution of risk scores and biochemical recurrence characteristics of samples in GSE116918 cohort. H KM curve of biochemical recurrence based on risk score grouping in the TCGA cohort. I KM curve of biochemical recurrence based on risk score grouping in the GSE116918 cohort. J ROC curve of the risk score group-based prediction model for biochemical recurrence in the TCGA cohort. K Multivariate Cox regression analysis of risk scores and clinical data associated with biochemical recurrence. L Nomogram based on risk score and clinical data associated with biochemical recurrence. BCR biochemical recurrence, p_T pathological T stage, ROC Curve receiver operating characteristic curve, AUC area under curve, KM Curve Kaplan–Meier curve.Full size imageSubsequently, we used the SREBF1 target gene to construct a risk score for the biochemical recurrence (BCR) of prostate cancer. First, a total of 48 genes associated with the biochemical recurrence of prostate cancer were screened using univariate Cox regression (Fig. 6C). Sixteen target genes associated with biochemical recurrence of prostate cancer were further screened out by Lasso regression, including MAN1C1, CTBS, LAMC1, DEGS1, TRNT1, DHX30, FSTL1, EIF4G1, FAM50B, GRB10, SPTBN2, NADSYN1, SYTL2, NAGLU, SERPINB5, and PRDM15 (Fig. 6D, E). In the univariate Cox analysis of these 16 genes associated with biochemical recurrence of prostate cancer, only three genes had a risk-reducing effect (Fig. S2). Subsequently, a multivariate Cox regression analysis based on these 16 genes was used to construct a risk score (Fig. 6F). The KM survival curve showed that the high-risk and low-risk groups, based on the median risk score of the TCGA cohort (0.9230274), had significant differences in BCR survival analysis, with the high-risk group having a higher risk of biochemical recurrence (Fig. 6H). We also validated the model using an external cohort (GSE116918), which showed that the high-risk group had a higher risk of biochemical recurrence (Fig. 6G, I). The ROC curve demonstrated the model efficiency of the risk score grouping based on TCGA cohort associated with BCR risk, with AUC values of 0.7843, 0.8153, and 0.8643 at one year, three years, and five years, respectively (Fig. 6J). We included clinical indicators, such as age, Gleason grouping (≤7 points for Gleason low-risk group), and pathological T stage, to compare their relationship with BCR risk with the risk grouping we constructed in multivariate Cox analysis. The T3 stage in pathological T stage was associated with increased BCR risk (HR = 3.65, P = 0.0186), and the high-risk group of the risk score was also associated with increased BCR risk (HR = 3.59, P = 0.0033) (Fig. 6K). Based on these features, a nomogram was constructed (Fig. 6L).The SREBF1 inhibitor Betulin significantly promotes ferroptosis in prostate cancerConsidering the regulation of metabolic reprogramming in prostate cancer by SREBF1 and its relationship with ferroptosis, we investigated the effect of the SREBF1 inhibitor Betulin on promoting ferroptosis in prostate cancer. RSL3, an inhibitor of the classical ferroptosis-resistant pathway, was used as a positive control to promote ferroptosis. Our experiments involved the androgen-sensitive prostate cancer cell line LNCaP and the castration-resistant prostate cancer cell line PC3.We verified the expression of SREBF1 target genes and ferroptosis-related genes in drug-treated cell lines using quantitative real-time polymerase chain reaction (qRT-PCR). In the LNCaP cell line, the expression of target genes such as SCD1, DHCR7, MSMO1, CYP51A1, and EBP decreased significantly, and GPX4 decreased to a certain extent (Fig. 7A). However, in the PC3 cell line, ferroptosis-related genes such as SCD5, GPX4, and SLC7A11 were not affected much in the Betulin group (Fig. 7B). Next, we detected intracellular reactive oxygen species (ROS) to determine the degree of ferroptosis in prostate cancer cells after Betulin treatment. Fluorescence microscopy revealed a significant increase in ROS levels in both LNCaP and PC3 cells following Betulin treatment (Fig. 7C, D). The results of intracellular ROS detection using flow cytometry were consistent. In LNCaP cells, the Mean Fluorescence Intensity (MFI) of ROS detection in the Betulin group was significantly increased (126), compared with 89 in the RSL3 group and 64.73 in the control group (Fig. 7E). Similarly, in the PC3 cell line, the average MFI of ROS detection in the Betulin and RSL3 group were 85.07 and 67, respectively, which were significantly higher than the 30.87 in the control group (Fig. 7F). Intracellular glutathione (GSH) levels were measured to observe intracellular oxidative stress. In LNCaP cells, GSH decreased significantly in the Betulin group and RSL3 group, with levels of 13.94 μg/106 cells and 11.82 μg/106 cells, respectively, compared to 19.33 μg/106 cells in the control group. Similarly, in PC3 cells, GSH also decreased in the Betulin group and RSL3 group, with levels of 16.16 μg/106 cells and 12.73 μg/106 cells, respectively, compared to 17.28 μg/106 cells in the control group (Fig. 7G). Finally, we examined whether castration and Betulin treatment had a synergistic effect on hormone-sensitive prostate cancer cell lines. We observed that the Q value was greater than 1 at Betulin concentrations ranging from 0.625 μg/mL to 40 μg/mL, indicating a synergistic effect between the two treatments. The Q value was highest around 10 μg/mL, indicating the most significant synergistic effect (Fig. 7H).Fig. 7: Validation of the SREBF1 inhibitor Betulin’s ability to promote ferroptosis in prostate cancer cells.qRT-PCR analysis of the expression of some SREBF1 target genes and ferroptosis-related genes in LNCaP (A) and PC3 (B) cell lines (n = 3). Fluorescence microscopy observation of ROS content in LNCaP (C) and PC3 (D) cells. Hoechst was used to stain the nuclei, and ROS was visualized using the DCFH-DA probe (n = 3). Cell flow cytometry analysis of ROS content in LNCaP (E) and PC3 (F) cells. Quantitative comparison by Mean Fluorescence Intensity (MFI) (n = 3). G Measurement of GSH content in LNCaP and PC3 cells (n = 6). H Cell viability assay in LNCaP cells after androgen deprivation culture and Betulin treatment. The left y-axis represents cell viability, and the right y-axis represents the Q value calculated for each concentration group (n = 3). CS charcoal Stripped. Data were presented as mean ± SD. (ns, P ≥ 0.05; *P