An anti-androgen resistance-related gene signature acts as a prognostic marker and increases enzalutamide efficacy via PLK1 inhibition in prostate cancer

Abstract

Background

Anti-androgen resistance remains a major clinical challenge in the treatment of prostate cancer (PCa), leading to disease progression and treatment failure. Despite extensive research on resistance mechanisms, a reliable prognostic model for predicting patient outcomes and guiding therapeutic strategies is still lacking. This study aimed to develop a novel gene signature related to anti-androgen resistance and evaluate its prognostic and therapeutic implications.

Methods

Anti-androgen resistance-related differentially expressed genes (ARRDEGs) were identified through transcriptomic analysis of enzalutamide- and dual enzalutamide abiraterone-resistant PCa cell lines from the GEO database. Functional enrichment analysis was performed to determine the biological roles of these genes. A prognostic gene signature was developed using univariate Cox regression, LASSO, and multivariate Cox regression models. The model was validated in independent PCa cohorts from The Cancer Genome Atlas (TCGA). Additionally, we assessed the correlation between the signature, immune infiltration, immune checkpoint expression, and drug sensitivity. The efficacy of PLK1 inhibition combined with enzalutamide was further explored using in vitro and in vivo experiments.

Results

We identified 304 ARRDEGs, from which three key genes (LMNB1, SSPO, and PLK1) were selected to construct a prognostic signature. This gene signature effectively stratified PCa patients into high- and low-risk groups, with the high-risk group exhibiting shorter recurrence-free survival and distinct immune characteristics. High-risk patients demonstrated elevated immune checkpoint expression (B7H3, CTLA-4, B7-1, and TIGIT), increased M2 macrophage infiltration, and enhanced sensitivity to chemotherapy and targeted therapy. Mechanistically, PLK1 inhibition potentiated the antitumor effect of enzalutamide by downregulating SLC7A11 and inducing ferroptosis, providing a potential therapeutic strategy to overcome anti-androgen resistance.

Conclusion

We established a novel ARRDEGs-based prognostic signature that predicts PCa progression and response to chemotherapy and targeted therapy. The integration of this signature with immune profiling and drug sensitivity analysis provides a valuable tool for precision oncology in PCa. Our findings highlight the potential of PLK1 inhibition as a therapeutic strategy to enhance enzalutamide efficacy and overcome resistance.

Introduction

Prostate cancer (PCa) is the second most common malignancy among men and the sixth most common cause of cancer-related mortality globally [1]. Androgen deprivation therapy (ADT) remains the standard treatment for patients with locally progressed or metastatic PCa [2, 3]. Although second-generation anti-androgens, including enzalutamide (Enz) and abiraterone, induce an initial response, most tumors ultimately progress to metastatic castration-resistant prostate cancer (mCRPC) [4]. Recurrence arises through either reactivation or bypassing of androgen receptor (AR) signaling, and these mechanisms are predominantly driven by differential gene expression [5,6,7]. Therefore, the identification of anti-androgen resistance-related differentially expressed genes (DEGs) is critical. Furthermore, evaluation of the impact of these DEGs on prognosis and drug sensitivity is vital for identifying potential targets to mitigate resistance and optimize treatment outcomes. Despite the large body of research on resistance-related genes in PCa progression, the role of anti-androgen resistance-related DEGs (ARRDEGs) in predicting prognosis and clinicopathological features remains unclear.

The tumor immune microenvironment (TIME) is pivotal in influencing tumor growth, as immune cells within this microenvironment contribute to immune escape. Tumor cells exploit immune checkpoints to facilitate unchecked growth [8]. As a result, immune checkpoint inhibitors (ICIs), such as monoclonal antibodies targeting CTLA-4 and PD-1, have been developed to restore immune cell function and increase antitumor activity [9, 10]. Recent studies have indicated that the TIME significantly influences anti-androgen resistance [11,12,13]. Thus, elucidating the relationship between anti-androgen resistance and the immune response in PCa is pivotal for optimizing therapeutic strategies.

Advancements in genomics and bioinformatics have increased the precision of methods for analyzing large-scale clinical data, surpassing that of traditional in vitro methods or animal experiments. Current research has employed high-throughput genomics and proteomics to delineate the molecular characteristics of anti-androgen resistance [7]. Consequently, integration of multiple ARRDEGs via genomic and bioinformatic analyses and evaluation of their prognostic and immunotherapeutic implications in PCa are both feasible and necessary.

In this study, ARRDEGs were identified via analysis of transcriptomic data from PCa cell lines in the GEO database, followed by clustering and functional enrichment analyses. We further incorporated these ARRDEGs with PCa prognosis-associated genes from The Cancer Genome Atlas (TCGA) database to develop a robust prognostic model and a corresponding nomogram. Additionally, the associations among ARRDEG expression, immune infiltration levels, and immune checkpoint levels in PCa were evaluated. Our results revealed key genes involved in anti-androgen resistance and were used to construct a gene signature that may serve as a prognostic biomarker and guide the personalized treatment of PCa patients.

Materials and methods

Identification of ARRDEGs

The raw gene expression data for enzalutamide (Enz)-resistant cells (GSE104935) and dual enzalutamide/abiraterone-resistant cells (GSE109708) were retrieved from the GEO database. We conducted a comparative analysis to identify DEGs between anti-androgen therapy-resistant and sensitive PCa cells via the"limma"package (version 3.5.1) in R. The criteria for DEG selection were established as an absolute fold change (FC) ≥ 1.5 and a p value < 0.05.

Functional enrichment analysis

We performed Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment studies to investigate the probable functions of the ARRDEGs via public databases such as OmicsBean (www.omicsbean.cn), DAVID (https://david.ncifcrf.gov), and Metascape [14]. Pathways with significant enrichment were identified via a p value threshold of < 0.05, and the top 20 pathways were chosen for subsequent research.

Identification of prognosis-related genes in PCa patients

The mRNA expression patterns and associated clinical data of 498 PCa patients were acquired from The Cancer Genome Atlas (TCGA, https://cancergenome.nih.gov/). We employed the"limma"package in R to identify DEGs between PCa tissues and normal tissues, establishing the selection criteria of a fold change (FC) ≥ 1.5 and a p value < 0.05. Thereafter, univariate Cox regression analysis was conducted with the"survival"package in R to identify prognosis-associated DEGs (p < 0.05).

Construction and validation of the prognostic signature

We identified genes linked to recurrence-free survival (RFS) in PCa patients via univariate Cox regression analysis and designated genes with p < 0.05 as significant prognostic factors. A prognostic model for anti-androgen resistance was developed by utilizing a multivariate Cox regression model to ascertain coefficients for each gene in the risk score model. The prognostic index (PI) was computed using the following formula: PI = (β1 * expression level of LMNB1) + (β2 * expression level of SSPO) + (β3 * expression level of PLK1).

The median PI served as a threshold to classify patients into high-risk and low-risk groups, demonstrating notable prognostic disparities between the two groups. Kaplan‒Meier survival curves and time‒dependent receiver operating characteristic (ROC) curves were used to evaluate the predictive efficacy of the prognostic signature for RFS.

Construction and evaluation of the nomogram

To evaluate the independence of the three-gene signature in predicting RFS relative to other clinical variables (including TNM stage, T stage, M stage, N stage, age, Gleason score, and PSA level), univariate and multivariate Cox regression analyses were conducted; p < 0.05 was considered to indicate statistical significance. The hazard ratio (HR) and 95% confidence interval (CI) for each variable were computed. A nomogram was developed utilizing the risk score and clinical characteristics to precisely predict the survival probability of PCa patients. Calibration curves were constructed to assess the predictive accuracy of the nomogram; the curves were close to the 45° line indicating superior predictive value. The"rms"package in R was utilized to develop the nomogram and generate calibration curves. Furthermore, a time-dependent ROC curve analysis was performed to assess the sensitivity and specificity of the nomogram in predicting RFS.

Prediction of the chemotherapy response

The"pRRophetic"package in R was utilized to assess the treatment response of high-risk and low-risk PCa patients. This analysis was based on the half-maximal inhibitory concentration (IC50) values for each patient sourced from the Genomics of Drug Sensitivity in Cancer (GDSC) database (https://www.cancerrxgene.org/).

Evaluation of immune cell infiltration

The"CIBERSORT"method in R was employed to assess the infiltration levels of 22 immune cell types in PCa patients and to assess their correlations with the risk score. The TIMER database (http://timer.cistrome.org/) was utilized to examine the relationships between the expression levels of anti-androgen resistance-associated genes and the infiltration levels of six immune cell subtypes: B cells, CD8 + T cells, CD4 + T cells, macrophages, neutrophils, and dendritic cells.

ScRNA-seq analysis

We obtained single-cell RNA sequencing (scRNA-seq) data from Enz-resistant PCa samples (GSE213667) from the GEO database and processed the data via the R package"Seurat."To eliminate batch effects between samples, we utilized the R package"harmony"and normalized the data via the ScaleData function. Principal component analysis (PCA) was then performed on the scaled data. We subsequently applied the RunUMAP function for dimensionality reduction. Differential gene expression within each cluster was assessed via the FindAllMarkers function. Finally, the cell types were annotated via the"SingleR"package.

Cell culture

The human PCa cell lines 22RV1 and C4-2 were acquired from the Center for Excellence in Molecular Cell Science, CAS (Shanghai, China). The cells were cultivated in RPMI-1640 media (GIBCO, #C11875500BT) supplemented with 10% fetal bovine serum (FBS) (ExCell Bio, FSP500) and 1% penicillin‒streptomycin (GIBCO, #15140-122). Polymerase chain reaction (PCR) analysis was conducted to verify the absence of mycoplasma contamination in the cells.

Stable PLK1 knockdown cell line

The pLVX cloning vector (Youbio) was used to establish short hairpin RNA (shRNA) targeting PLK1 (5′-GGCAAGATTGTGCCTAAGTCTCTGCTGCT-3′) or a negative control (5′-GGAATCTCATTCGATGCATAC-3′). The cells were transduced with lentiviral particles for 3 days and then selected with puromycin (10 mg/mL) for 14 days to establish stable cell lines.

Cell viability assay

Cell viability was evaluated via the Cell Counting Kit-8 (CCK-8) assay (APExBIO, K1018) in accordance with the manufacturer's guidelines. In summary, 5 × 10³ PCa cells (wild-type, shNC, or shPLK1) were treated with DMSO (control) or different doses of Enz or GSK461364 for the designated time periods. Prior to the experiment (1–2 h in advance), 10 μL of CCK-8 reagent and 90 μL of RPMI 1640 solution was added to each well. The optical density (OD) at 450 nm was assessed to evaluate cell viability. The effects of drug combinations (synergy, antagonism, or additivity) were assessed according to the combination index (CI) calculated via CalcuSyn 2.0 software; specifically, CI < 1 indicates synergy, CI = 1 denotes additivity, and CI > 1 signifies antagonism.

Cell death assay

PCa cells were treated with the specified medicines for 24 h, subsequently washed, and stained with 2 μg/mL DAPI (Servicebio, G1012) and 10 μg/mL PI (Servicebio, G1021). Following a 10-min incubation in the dark, fluorescence images were acquired with an inverted fluorescence microscope (Olympus IX71).

Colony formation assay

Approximately 1000 PCa cells were inoculated onto six-well plates and cultivated for 14 days. The colonies were subsequently fixed with 4% paraformaldehyde and stained with crystal violet for 20 min. Imaging and colony enumeration were conducted with Quantity One 1D analysis software (Bio-Rad, Hercules, CA, USA).

Xenograft mouse model

All animal experiments were approved by the Institutional Animal Care and Use Committee of Sun Yat-Sen University (Approval no. SYSU-IACUC-2021-000041). Male BALB/c nude mice were purchased from GemPharmatech LLC. For the animal experiments, 3 × 10⁶ 22Rv1 PCa cells were subcutaneously injected into the flank of each mouse. When the average tumor volume reached 50–60 mm³, Enz (10 mg/kg), GSK461364 (20 mg/kg), or their combination was administered via intraperitoneal injection every 3 days. The treatment lasted for a total of 21 days, with tumor growth data recorded every 3 days. On day 22, the mice were euthanized, and the tumors were collected. The tumor volume was calculated via the following formula: a × b² × 0.5, where"a"is the diameter perpendicular to"b,"and"b"is the shortest diameter.

Statistical methods

Two-tailed unpaired Student's t test and one-way analysis of variance (ANOVA) were used to assess statistical significance. A p value < 0.05 was considered to indicate statistical significance. Statistical analyses were conducted via GraphPad Prism 9.0 software. Quantitative values are presented as the means ± standard deviations. Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Results

Identification and functional annotation of anti-androgen resistance-related differentially expressed genes (ARRDEGs)

We assessed the mRNA expression profiles of samples from the Enz-resistant (GSE104935) and dual Enzalutamide/abiraterone-resistant (GSE109708) GEO datasets via the"limma"R package to identify DEGs between anti-androgen therapy-resistant and sensitive PCa cells. By establishing thresholds of p < 0.05 and |fold change|≥ 1.5, we identified 2847 and 1277 DEGs between resistant and sensitive samples in the Enz-related dataset and dual Enzalutamide/abiraterone-related dataset, respectively (Fig. 1A, B). A total of 304 common ARRDEGs were extracted for further study (Fig. 1B).

Fig. 1

Identification and functional enrichment analysis of ARRDEGs. A Volcano plot of differentially expressed genes (DEGs). B Venn diagram of 304 common ARRDEGs. C–E Bar chart and protein‒protein interaction network for the GO and KEGG analyses of the DEGs

We then performed GO and KEGG pathway enrichment analyses to assess the potential functions of the ARRDEGs. The findings indicated substantial enrichment of the DEGs in biological processes associated with tumor progression, such as"negative regulation of cellular component organization,""Rho GTPase signaling,"and"regulation of kinase activity."(Fig. 1C–E). These pathways are closely involved in cancer cell invasion and metastasis, indicating that ARRDEGs may be pivotal in the malignant evolution of PCa.

Clustering of ARRDEGs revealed two PCa subtypes with distinct baseline characteristics and survival outcomes

To further explore the clinical significance of ARRDEGs in PCa, we performed unsupervised consensus clustering via the"ConsensusClusterPlus"package. The optimal cluster number was determined to be k = 2, according to the cumulative distribution function (CDF) plot and delta area (Fig. 2A–C). This approach was used to classify 498 PCa samples from the TCGA database into two distinct subtypes. Differential expression analysis was performed via the"limma"package, revealing 941 DEGs between the subtypes (fold change = 1.5, p = 0.05) (Fig. 2D, E). The Kaplan‒Meier analysis indicated that patients with subtype 1 had markedly longer RFS than did those with subtype 2 (Fig. 2F), suggesting that these subtypes may have distinct prognostic outcomes. GO function and KEGG pathway enrichment analyses of these DEGs revealed significant enrichment of pathways such as"cell-substrate adhesion,""integrin-mediated signaling pathway,""ameboidal-type cell migration,"and"ECM-receptor interaction"(Fig. 2G, H). These pathways are implicated in cancer cell invasion and metastasis, suggesting that ARRDEGs may promote PCa progression by regulating cellular adhesion, migration, and the tumor microenvironment.

Fig. 2

Clustering of ARRDEGs revealed two PCa Subtypes with distinct baseline characteristics and survival outcomes. A–C Consensus matrix for k = 2, 3, and 4. D Heatmap illustrating the expression profiles of 941 DEGs between the two subtypes. E Volcano plot of the DEGs between the subtypes. F Recurrence-free survival of the two subtypes. G, H KEGG pathway and GO term enrichment analysis for DEGs between the two subtypes

Establishment and validation of a prognostic model using ARRDEGs

We conducted a univariate Cox regression analysis utilizing the TCGA database to identify ARRDEGs associated with RFS in PCa patients. Among the 304 ARRDEGs, 32 genes were significantly associated with PCa prognosis (Fig. 3A). We then applied the LASSO algorithm to further refine these genes, ultimately selecting three core ARRDEGs: LMNB1, SSPO, and PLK1. On the basis of their expression levels, we constructed a risk score model with the following formula:

Fig. 3

Construction and validation of the ARRDEG-based prognostic model. A Univariate Cox regression analysis revealed 32 genes associated with RFS in PCa patients. B LASSO coefficient profiles of the 32 RFS-associated genes. C Selection of the optimal lambda in the LASSO model, resulting in three key ARRDEGs (LMNB1, SSPO, and PLK1). D Multivariate Cox regression analysis confirming the independent prognostic value of the risk score. E Forest plot showing the hazard ratios of the risk score and other clinical factors. F Calibration plot demonstrating the agreement between the predicted and observed RFS values. G, H Kaplan–Meier survival curves and ROC curves of high- and low-risk patients in the training cohort. I, J Kaplan‒Meier survival curves and ROC curves of high- and low-risk patients in the testing cohort

Risk score = (0.02181362 * LMNB1 expression level) + (0.17958708 * SSPO expression level) + (0.25686309 * PLK1 expression level) (Fig. 3B, C).

The multivariate Cox regression analysis subsequently confirmed that the risk score, in conjunction with T stage, N stage, age, PSA, and Gleason score, was significantly correlated with RFS in PCa patients and confirmed that the risk score was an independent prognostic indicator for PCa (Fig. 3D, E). To evaluate the model's accuracy, calibration plots were generated and revealed excellent agreement between the predicted and observed RFS values, confirming the model's robust predictive performance (Fig. 3F). The TCGA dataset was randomly divided into training and testing sets. Patients were categorized into high-risk and low-risk groups on the basis of the median risk score as a threshold. The Kaplan‒Meier survival analysis indicated that patients in the low-risk cohort experienced markedly prolonged RFS than those in the high-risk cohort (Fig. 3G). The receiver operating characteristic (ROC) curve analysis yielded AUC values of 0.94, 0.78, and 0.71 for predicting 1-, 3-, and 5-year survival, respectively, highlighting the model’s robust predictive power in the training set (Fig. 3H). Comparable findings were observed in the test set, further validating the model’s prognostic value (Fig. 3I, J).

These findings reinforce that the risk score model, which incorporates LMNB1, SSPO, and PLK1 expression, serves as a reliable and specific prognostic tool for PCa patients. Notably, PLK1 is essential for cell cycle regulation and tumor proliferation, which may explain its inclusion as a key factor in this model.

Immune landscape and tumor mutational burden (TMB) in high-risk and low-risk PCa patients

We subsequently assessed the correlation between the risk score and the expression levels of common immune checkpoints. As shown in Fig. 4A, the heatmap displays the distribution of immune checkpoint expression across different risk groups. Notably, the expression levels of B7H3, CTLA-4, B7-1, and TIGIT were markedly increased in the high-risk group, indicating a more immunosuppressive tumor microenvironment (Fig. 4B–E). KEGG pathway analysis further supported these findings (Fig. 4F).

Fig. 4

Immune landscape and tumor mutational burden (TMB) analysis in high- and low-risk PCa patients. A Heatmap showing immune checkpoint expression differences between the high- and low-risk groups. B–E Boxplots comparing the expression levels of immune checkpoints such as B7H3, CTLA-4, B7-1, and TIGIT between the two groups. F KEGG pathway enrichment analysis of differentially expressed genes related to immune response pathways in high- and low-risk patients. G Analysis of the differential response to immune checkpoint inhibitors (ICIs) between the high- and low-risk groups. H The CIBERSORT algorithm was used to estimate the infiltration levels of 22 immune cell types in the high- and low-risk groups. I, J Tumor mutational burden (TMB) analysis identifying the top TMB-related genes in high- and low-risk patients

We evaluated the efficacy of ARRDEGs in predicting the immune checkpoint inhibitor response by utilizing the IMvigor210 dataset, a well-characterized bladder cancer cohort treated with anti–PD-L1 therapy and has been widely adopted as a benchmark to explore immunotherapy-related signatures across multiple cancers. The results demonstrated that a markedly greater percentage of patients in the low-risk cohort than in the high-risk cohort responded to immunotherapy, underscoring the value of ARRDEGs as prospective biomarkers for ICI sensitivity across various malignancies (Fig. 4G).

Additionally, CIBERSORT analysis of 22 tumor-infiltrating immune cell types revealed that the high-risk group had increased infiltration of naïve B cells, regulatory T cells, and M2 macrophages, whereas the low-risk group displayed a greater proportion of CD4 memory T cells and resting mast cells (Fig. 4H). These differences in immune cell composition may underlie the distinct immunotherapy responses observed between the risk groups.

Finally, we evaluated the TMB and identified the top 20 TMB-related genes in PCa patients. The high-risk cohort presented an increased incidence of TP53 mutations, whereas mutations in TTN and SPOP were more common in the low-risk cohort (Fig. 4I, J). These findings highlight the close associations among ARRDEGs, TMB, and the tumor immune microenvironment, suggesting that these factors collectively influence PCa progression and therapeutic outcomes.

Identification of PLK1 as a key factor in Enz resistance

To further investigate the expression of the key ARRDEGs (LMNB1, SSPO, and PLK1) in different cell types within Enz-resistant tissues, we utilized single-cell RNA sequencing data from PCa tissues resistant to Enz (GSE213667). After data normalization, we employed uniform manifold approximation and projection (UMAP) for dimensionality reduction and the"SingleR"function for cell annotation. This analysis revealed eight distinct cell types, annotated as T cells, B cells, macrophages, epithelial cells, fibroblasts, endothelial cells, neutrophils, and circulating basal cells (Supplementary Fig. 1A). Further differential expression analysis revealed the expression patterns of LMNB1, SSPO, and PLK1 across these cell types. The results revealed that PLK1 was highly expressed in epithelial cells, endothelial cells, T cells, and circulating basal cells, whereas LMNB1 was expressed primarily in neutrophils, and SSPO was expressed predominantly in circulating basal cells (Fig. 5A, Supplementary Fig. 1B). This differential expression distribution suggests that these components may play distinct roles in immune cell infiltration, although further experiments are needed to elucidate the underlying mechanisms involved.

Fig. 5

Identification of PLK1 as a key factor in enzalutamide (Enz) resistance. A Gene expression profiles (PLK1, LMNB1, and SSPO) analyzed by UMAP. B mRNA expression levels of PLK1, LMNB1, and SSPO in enzalutamide-sensitive (Enz-S) and enzalutamide-resistant (Enz-R) cell lines. C Cell viability following DMSO or Enz treatment in shNC and shPLK1 cells. D Cell death was assessed by propidium iodide (PI) staining and fluorescence microscopy after 24 h of treatment. E Quantification of fluorescence intensity from (D). F Colony formation assay in shNC and shPLK1 cells after 14 days of treatment, with colony counts quantified via ImageJ software. G Number of clones quantified from (F). Statistical significance was determined by two-tailed unpaired t-test (B, C) and one-way ANOVA followed by Tukey’s multiple comparison test (E, G). B, C, E, G Data are presented as mean ± SD. Representative data of triplicate experiments are shown. *p < 0.05, **p < 0.01, ***p < 0.001, ns, not significant

However, as this study was primarily derived from the mRNA expression profiles of anti-androgen-resistant cell lines, our primary focus was on the expression of these three genes within the cells themselves. Consequently, we next validated the differences in the mRNA expression of these genes in Enz-resistant cell lines. The results indicated a marked upregulation of PLK1, suggesting that PLK1 may play a more significant role in mediating Enz resistance (Fig. 5B). We then constructed a PLK1-knockdown PCa cell line using lentiviral vectors (Supplementary Fig. 1 C, D). Cytotoxicity assays and dead cell staining experiments demonstrated that PLK1 knockdown significantly enhanced the cytotoxic effects of Enz (Fig. 5C–E). Moreover, the inhibitory effect of Enz on cell proliferation was also markedly amplified (Fig. 5F, G). These findings collectively suggest that PLK1 is a critical mediator of Enz resistance.

Synergistic antitumor effects of GSK461364 and Enz

Given the critical role of PLK1 in Enz resistance, we aimed to identify a safe, reliable, and potent PLK1 inhibitor to explore its potential in PCa treatment and in increasing sensitivity to Enz. GSK461364, a specific PLK1 inhibitor with demonstrated safety and efficacy in various solid tumors, has been reported to synergize with different anticancer agents [15,16,17,18]. Considering the pivotal role of PLK1 in anti-androgen resistance and prognosis in PCa, we hypothesized that its inhibition might increase the efficacy of Enz in resistant PCa cells. Our experiments revealed that the combination of GSK461364 and Enz significantly reduced PCa cell viability. The combination index (CI) calculated via Calcusyn software indicated synergistic interactions (CI < 1) between the two drugs (Fig. 6A, B). Propidium iodide (PI) staining further confirmed this synergy, showing enhanced Enz-induced cell death with the addition of GSK461364 (Fig. 6C, Supplementary Fig. 2 A). In colony formation assays, GSK461364 not only inhibited PCa cell growth but also augmented the cytotoxic effect of Enz (Fig. 6D, Supplementary Fig. 2B). In organoid models derived from P10P53 double-knockout mice, the combination treatment led to significant shrinkage and disintegration, as evidenced by increased red fluorescence intensity (Fig. 6E, Supplementary Fig. 2 C).

Fig. 6

Combination treatment with GSK461364 and Enz in PCa cells. A, B The viability of C4-2 and 22Rv1 PCa cells was assessed via the CCK-8 assay following 48 h of treatment with GSK461364, Enz, or their combination. C Cell death was evaluated via propidium iodide (PI) staining and fluorescence microscopy after 24 h of treatment. Scale bar, 100 μm. D Colony formation assays were performed on treated C4-2 and 22Rv1 cells after 14 days, with colony counts quantified via ImageJ software. E PI staining and bright-field imaging of PCa organoids were conducted to assess cell death following treatment with GSK461364, Enz, or their combination. Scale bar, 50 μm. In nude mice, the combination of Enz and GSK461364 more effectively inhibited tumor growth. At the end of the experiment, the mice were euthanized, and F Representative tumor images are shown. G Tumor growth curves were recorded every three days (mean ± SD, n = 5/group), and tumor tissues were weighed and summarized. H Representative H&E and IHC staining of the indicated proteins in tumor tissues from each group. Scale bar, 100 μm. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparison test (G). G Data are presented as mean ± SD. Representative data of triplicate experiments are shown. *p < 0.05, **p < 0.01, ***p < 0.001, ns, not significant

In vivo experiments using a subcutaneous xenograft mouse model revealed that the combination of GSK461364 and Enz resulted in significantly greater tumor inhibition than either drug alone (Fig. 6F, G). Hematoxylin and eosin (H&E) staining revealed larger areas of tumor necrosis in the combination treatment group, along with a significant reduction in the proportion of Ki67-positive proliferating cells, a marker of cell proliferation (Fig. 6H, Supplementary Fig. 2D). In summary, our data indicate that the combination of GSK461364 and Enz is a highly efficacious therapeutic strategy. These findings highlight the superior efficacy of this combination therapy.

PLK1 inhibition synergizes with enzalutamide treatment to lower SLC7A11 levels and enhance ferroptosis

Recent studies have shown that PLK1 can regulate the chemosensitivity of esophageal squamous cell carcinoma through the ferroptosis pathway [19, 20]. Additionally, Enz has been demonstrated to induce ferroptosis by suppressing glutathione (GSH) production and promoting lipid peroxidation [19]. We hypothesized that the synergistic sensitization of PLK1 inhibition and Enz treatment might be mediated by the promotion of ferroptosis. To explore this, we evaluated the changes in reactive oxygen species (ROS) and lipid peroxidation levels induced by the combination of PLK1 inhibition and Enz treatment. Compared with either treatment alone, the combined treatment significantly increased the ROS and lipid peroxidation levels (Fig. 7A, B, Supplementary Fig. 3A, B). Consistent with these findings, pharmacological inhibition of PLK1 with the combination of GSK461364 and Enz also increased these levels (Fig. 7C, D, Supplementary Fig. 3C, D). Further experiments demonstrated that PLK1 silencing and pharmacological inhibition increased the intracellular iron pool in Enz-treated PCa cells (Fig. 7E, F, Supplementary Fig. 3E, F). These results confirm that targeting PLK1 synergizes with Enz to induce ferroptosis, thereby enhancing tumor cell killing.

Fig. 7

GSK461364 and Enz induce ferroptosis via SLC7A11 downregulation in PCa cells. A, B Lipid peroxidation assays were performed to assess ferroptosis induction in shNC and shPLK1 cells following treatment. C, D Lipid peroxidation assays were conducted to evaluate ferroptosis induction in wild-type PCa cells after treatment. E Ferro-orange staining for intracellular Fe²⁺ was performed in shNC and shPLK1 cells treated for 48 h with or without Enz (10 μM). Scale bar, 100 μm. F Ferro-orange staining for intracellular Fe²⁺ in wild-type PCa cells treated for 48 h with or without Enz (10 μM). Scale bar, 100 μm. G Western blot analysis showing SLC7A11 expression levels following treatment with or without Enz in shNC and shPLK1 cells. H Western blot analysis demonstrating SLC7A11 expression levels after treatment with GSK461364, Enz, or their combination. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparison test (A–D). A–D Data are presented as mean ± SD. Representative data of triplicate experiments are shown. *p < 0.05, **p < 0.01, ***p < 0.001, ns, not significant

As a key source of cysteine for glutathione synthesis, the cystine/glutamate transporter SLC7A11 has been reported to play a crucial role in ferroptosis [20, 21]. Recent studies also suggest that SLC7A11 is closely involved in the mechanism of action of Enz. We speculated that SLC7A11 might be an important downstream target of PLK1 and Enz. To test this hypothesis, we assessed the protein expression levels of SLC7A11 and found that the combination of PLK1 targeting and Enz significantly reduced SLC7A11 levels compared with those in the control and single-treatment groups (Fig. 7G, H, Supplementary Fig. 3G, H). These data indicate that the mechanism by which targeting PLK1 synergizes with Enz to promote ferroptosis is likely mediated by a reduction in SLC7A11 expression.

Discussion

Anti-androgen therapy remains the frontline treatment for PCa. However, nearly all patients who initially respond to this therapy eventually develop resistance. Previous studies have shown that resistance-related genes significantly impact tumorigenesis and cancer progression. While numerous gene signatures have been developed to predict the survival outcomes and treatment responses of PCa patients, the role of anti-androgen resistance-related gene signatures as prognostic and immune biomarkers has yet to be fully explored [22,23,24]. We aimed to fill this gap by developing a gene signature that not only predicts prognosis but also identifies high-risk PCa patients who may benefit from alternative therapies.

Through transcriptomic analysis of the GEO database, we identified 304 ARRDEGs. Functional enrichment analysis revealed that these ARRDEGs were highly enriched in pathways related to tumor progression, including"signaling by Rho GTPases"and"regulation of kinase activity,"which aligns with previous findings. Using Cox regression models, we identified three key genes (LMNB1, SSPO, and PLK1) to construct a prognostic signature. Stratification according to this signature indicated that patients in the high-risk category had shorter RFS, highlighting the model's robust ability to predict PCa prognosis.

Our analysis of immune checkpoint genes (e.g., B7H3, CTLA-4, B7-1, and TIGIT) revealed elevated expression in the high-risk group, indicating that a more immunosuppressive tumor microenvironment may contribute to poorer clinical outcomes [25]. This observation aligns with the established notion that increased immune checkpoint expression is a key mechanism of tumor immune evasion [26]. Given the success of ICIs in cancer therapy, our model's potential to predict ICI responsiveness offers valuable insights for patient stratification and personalized treatment. Furthermore, our analysis suggests that high-risk patients may be more sensitive to 12 types of chemotherapy and targeted agents, underscoring the potential relevance of this gene signature in guiding clinical treatment decision-making (Supplementary Fig. 4A–L). However, due to the lack of further experimental validation, these findings should be regarded as exploratory and warrant additional investigation in future studies.

Given the pivotal role of PLK1 in anti-androgen resistance, we further explored the combined therapeutic effects of targeting PLK1 (through gene silencing and pharmacological inhibition) in conjunction with Enz treatment. Our data confirmed that the combination treatment significantly increased antitumor efficacy. Mechanistically, we found that combination therapy might act by downregulating SLC7A11, promoting lipid peroxidation, and inducing ferroptosis.

In summary, we have established a novel prognostic gene signature based on ARRDEGs that serves as a reliable predictor of tumor immune status and therapeutic response in PCa patients. The correlation of this signature with the tumor immune microenvironment highlights its potential value in predicting the immunotherapy response. In personalized medicine, this ARRDEG-based model serves as a valuable tool for clinical decision-making in PCa management. Nevertheless, our study has several limitations. Multicenter, prospective, and large-scale studies are needed to validate these findings. Additionally, further experimental investigations are essential to elucidating the mechanisms by which these genes influence PCa prognosis and the immune response.

Conclusion

With this study, a novel prognostic signature based on three ARRDEGs (LMNB1, SSPO, and PLK1) was established, which is significantly associated with tumor immune microenvironment features and may have potential utility in predicting the efficacy of immunotherapy in PCa patients. In the context of personalized medicine, this ARRDEG-based signature provides a valuable tool for guiding clinical decision-making in PCa, particularly in the selection of patients who may benefit from chemotherapy, targeted therapy, or immunotherapy. Nevertheless, this study has several drawbacks. Large-scale, multicenter, prospective clinical trials are essential to confirm the predictive significance of this signature across various patient populations. Further experimental investigations are needed to elucidate the precise mechanisms by which LMNB1, SSPO, and PLK1 contribute to PCa progression and influence the tumor immune environment. Studies on the potential of combining this signature with immunotherapeutic strategies could reveal new avenues for personalized PCa treatment.