Heterogeneous Responses to High-Dose Testosterone in Castration-Resistant Prostate Cancer Tumors with Mixed Rb-Proficient and Rb-Deficient Cells

Abstract

Androgen deprivation therapy remains a cornerstone in managing prostate cancer. However, its recurrence often leads to the more aggressive castration-resistant prostate cancer (CRPC). Although second-line androgen receptor signaling inhibition treatments such as enzalutamide and abiraterone are available, their effectiveness against CRPC is only transient. High-dose testosterone (Hi-T) has recently emerged as a promising treatment for CRPC, primarily through the suppression of E2F and MYC signaling. However, the roles of Rb family proteins in influencing this therapeutic response remain debated. In this study, we utilized a CRPC patient-derived xenograft model that includes both Rb pathway–proficient and -deficient cell populations based on the positive or negative expression of RB family genes. Single-cell RNA sequencing analysis revealed that Rb-proficient cells displayed a robust response to Hi-T, whereas Rb-deficient cells exhibited significant resistance. Notably, our analysis indicated increased enrichment of the hypoxia signature in the Rb-deficient cell population. Further studies in RB1-silenced CRPC cell lines showed that treatment with a hypoxia-inducible factor-1α inhibitor can restore the sensitivity of Rb-deficient cells to high-dose dihydrotestosterone treatment. In conclusion, our research provides new molecular insights into CRPC tumor cell responses to Hi-T and proposes a new strategy to resensitize Rb-deficient CRPC cells to Hi-T treatment.

Introduction

Androgens and androgen receptors (AR) play pivotal roles in prostate cancer initiation and progression (1). The primary therapeutic strategy for metastatic prostate cancer involves suppressing systemic testosterone levels or using AR antagonists, known as androgen deprivation therapies (ADT). Despite its effectiveness, tumors often recur and progress to castration-resistant prostate cancer (CRPC), typically with increased AR expression and partial restoration of AR signaling (2). Although further addressed with aggressive AR signaling inhibitor (ARSi) treatments, such as enzalutamide and abiraterone (3, 4), tumors inevitably relapse.

An alternative approach leveraging elevated AR gene expression in metastatic CRPC (mCRPC) is the use of high-dose testosterone (Hi-T), either alone or in combination with ADT or ARSi treatment (5). Such approaches, termed bipolar androgen treatment (BAT), involve periodic treatment of CRPC patients with rapid cycling of Hi-T and ADT/ARSi agents and have shown promising outcomes in recent phase II clinical trials (6–9). Whereas Hi-T may function to resensitize tumor cells to ADT/ARSi treatment, it also exhibits strong antitumor effects through multiple mechanisms (10–18). One critical mechanism involves the transcriptional repression activity of AR (19–22). We previously reported that Hi-T–stimulated AR can suppress cell proliferation by recruiting the Rb protein to chromatin, strengthening the E2F–Rb transcription repressor complex and preventing G1/S cell-cycle progression, highlighting a crucial function of Rb in this process (20, 22). Furthermore, studies have indicated that Rb-like proteins may play an important compensatory role even when RB1 is completely silenced (20, 22, 23). Additionally, AR-mediated repression of MYC activity has been suggested as a contributing mechanism to tumor responses to Hi-T (17, 23).

Because RB1 deletions or mutations are common in CRPC, which often exhibits significant heterogeneity, we aimed to explore how tumors with mixed genetic compositions respond to Hi-T. In our study, we conducted single-cell RNA sequencing (scRNA-seq) on the LuCaP 96CR patient-derived xenograft (PDX) model, which contains both RB1⁺ and RB1⁻ cells and generally responds to Hi-T (22–25). Our results showed that many RB1⁻ cells express high levels of RBL2, which likely compensates for the loss of RB1. Therefore, we redefined Rb deficiency based on the absence of RB family gene (RBF) expression rather than solely on RB1 loss. We found that Rb-proficient cells (RBF⁺) demonstrated an almost complete response to Hi-T, whereas Rb-d eficient cells (RBF⁻) showed significant resistance. Additionally, our analysis revealed a strong enrichment of hypoxia signatures in Rb-deficient cells, and we found that targeting hypoxia-inducible factor-1α (HIF-1α) could resensitize these cells to Hi-T. Overall, our research provides new molecular insights into the heterogeneous responses of CRPC to Hi-T and proposes a novel approach for overcoming therapy resistance.

Materials and Methods

Cell culture

The C4-2 cell line (RRID: CVCL_4782) was purchased from the ATCC and maintained in RPMI medium with 2% FBS plus 8% charcoal-stripped serum. C4-2 RB-KO-1 and RB-KO-2 cell lines were generated previously by us using the CRISPR/Cas9 approach (22). All cell lines were frequently examined for Mycoplasma contamination using MycoAlert Mycoplasma Detection Kit (Lonza) and authenticated using short tandem repeat profiling.

Cell proliferation assay

Cells were treated with a HIF-1α inhibitor, PX-478 (MedChemExpress, HY-10231; ref. 26) at concentrations of 0 to 50 µmol/L, either alone or in the combination with dihydrotestosterone (DHT, 10 nmol/L). Six days after treatment, the cells were collected, trypsinized, and counted using Countess Automated Cell Counter (Thermo Fisher Scientific) following the manufacture’s protocol.

Immunoblotting

For immunoblotting, protein extracts were prepared by boiling cells in 2% SDS. Proteins were detected with primary antibodies, including anti-Rb (BD Biosciences, 554144, RRID: AB_395264), anti-AR (EMD Millipore, 06-680, RRID: AB_310214), anti–HIF-1α (Cell Signaling, 36169, RRID: AB_2799095), anti-GAPDH (Abcam, ab8245, RRID: AB_2107448), and anti-H3 (Abcam, ab1791, RRID: AB_302613).

Mouse xenografts

All animal experiments were approved by the University of Massachusetts Boston Institutional Animal Care and Use Committee and were conducted following institutional and national (United States) guidelines. 96CR PDX tumors were passaged by subcutaneous implantation in the flanks of castrated male SCID mice (4–6 weeks old, Taconic, RRID: IMSR_TAC:ICRSC). The housing conditions were ambient temperatures of 65 to 75°F with 40% to 60% humidity and a 12-hour light/12-hour dark cycle.

scRNA-seq analysis

A single-cell solution of tissue samples was obtained following a published protocol (27). Briefly, tissue fragments were incubated in 3 mL of Accumax (Innovative Cell Technologies, AM105) for 10 to 20 minutes at room temperature on a rocking shaker. Cell suspensions were then filtered using a 70-μm cell strainer and spun at 580 g for 5 minutes at 4°C. Red blood cells were lysed with ACK Lysing Buffer (Thermo Fisher Scientific, A1049201) on ice for 1 minute, followed by quenching with PBS and then filtered again using a 40-μm cell strainer. Cell suspensions were washed three times and resuspended in PBS plus 0.04% BSA. The final cell viability of the suspensions was determined using Countess Automated Cell Counter (Thermo Fisher Scientific).

For scRNA-seq data processing and quality control, sequencing reads were processed into FASTQ format. Custom references were generated using CellRanger (10x Genomics, v6.0.1, RRID: SCR_023221) for the human reference genome (GRCh37). Single-cell reads were quantified using CellRanger with default parameters. The resulting gene-by-cell matrices were imported into Seurat (v.4.3.0, RRID: SCR_007322) for downstream analysis. Low-quality cells were filtered with min.cells = 3 and min.features = 100 genes and with greater than 75,000 read counts. Filtered cells were normalized using NormalizeData, and 3,000 highly variable features were selected by FindVariableFeatures. Principal components and Uniform Manifold Approximation and Projections (UMAP) were implemented using RunPCA and FindNeighbors with 20 principal components. Clusters were calculated using the FindNeighbours and FindClusters functions with the resolution parameter set to 0.4 and visualized using the Seurat implementation of the dimensional reduction algorithm UMAP. To identify the effects of cell-cycle heterogeneity for each cell population, the Seurat R function CellCycleScoring was used to score the cell-cycle phases.

Differential gene expression for each cluster was calculated using the function Seurat FindAllMarkers and between the identity classes using the FindMarkers function. To identify the Hi-T–repressed gene signature in Rb-proficient and Rb-deficient cells, Hi-T–regulated genes in Rb-proficient and Rb-deficient cells were filtered by an average −log(fold change) of greater than 2 and adjusted P value < 0.05. Gene set enrichment analysis (GSEA) was performed with the fgsea R function (v.1.18.0, RRID: SCR_020938) using the MSigDB hallmark (v.7.5.1, RRID: SCR_022870) as the gene set and log-ranked gene expression as inputs. Kaplan–Meier survival analysis was performed on the Stand Up To Cancer (SU2C) mCRPC cohort (28). The z-score of mRNA expression of SU2C polyA⁺ (n = 266) was used as the input. The data were filtered to include patients who received ARSi treatment (abiraterone, enzalutamide, or apalutamide) prior to taxane therapy or combination therapy with another agent in the clinical trials. The expression profile of tissue samples was generated before the start of therapy or within 90 days of starting the first-line therapy. The remaining tumor samples (n = 81) were divided into two groups based on their signature score (high: the top 25% signature score; low: the bottom 75% signature score). To determine the enrichment of the target gene set over the sample population, we applied nonparametric and unsupervised GSEA using the GSVA (version 1.40.1, RRID: SCR_021058) R package. The enrichment score of each patient was used as the input for survival analysis. Survival curves were calculated by survfit R function (version 3.2-0), and the statistical significance between two groups was assessed by the log-rank test. The results were visualized using the survminer R package (version 0.4.9, RRID: SCR_021094).

Statistical analysis

Data presented in bar graphs represent the mean ± SD of at least three biological replicates. Statistical analysis was performed using an unpaired two-tailed Student t test comparing treatment versus vehicle or as otherwise indicated. A P value of less than 0.05 was considered statistically significant (*, <0.05; **, <0.01; ***, <0.001; and ****, <0.0001). Immunoblotting results are representative of at least three experiments. Boxplots of signature scores and gene expression were compared using the Wilcoxon test for comparison between two groups of samples. These tests were parametric and based on the assumption of normal distribution and equal variance across all experimental groups. All statistical analyses and visualizations were performed with R (version 3.6.0) unless otherwise specified.

Data availability

The scRNA-seq data have been deposited in the Gene Expression Omnibus with the accession number GSE275027.

Results

Identification of Rb-proficient and Rb-deficient cell populations in the 96CR model

mCRPC is highly heterogeneous, and subclones carrying distinct genomic backgrounds, such as loss of RB1, can differentially contribute to tumor progression and treatment response (28). To gain deeper molecular insight into how this heterogeneity influences CRPC responses to Hi-T therapy, we focused on a unique CRPC PDX model, LuCaP 96CR. Originating from the hormone-dependent LuCaP 96 PDX, which initially presented a heterozygous RB1 deletion, the 96CR model developed a homozygous RB1 deletion in tumor cells that progressed upon ADT treatment (24). Interestingly, our recent findings still identified Rb-positive tumor populations within the 96CR model (22), indicating the coexistence of both Rb-deficient and Rb-proficient cells. Additionally, the presence of AR amplification in these tumors (22) further positions the 96CR model as an ideal candidate for evaluating Hi-T therapy responses.

The 96CR xenograft tumors were subcutaneously established in castrated male SCID mice, which were then exposed to a Hi-T regimen for approximately 1 week. Following this treatment, tumor cells were subjected to scRNA-seq. We combined expression data from both untreated (vehicle) and testosterone (Hi-T)-treated cells for analysis, identifying 14 distinct clusters, which were visualized using UMAP (Fig. 1A and B). An examination of RB1 expression revealed two major cell subpopulations. Subpopulation 1 included clusters 1, 2, 4, 7, 11, and 12 and exclusively harbored RB1-positive (RB1⁺) cells [Fig. 1C (left)], although the overall proportion of these cells was relatively low (fewer than 10%). In contrast, subpopulation 2 consisted of clusters 0, 3, 5, 6, 8, 9, 10, and 13, which contained only RB1-negative (RB1⁻) cells. Previous studies by our group and others have indicated that Rb-like proteins p107 (encoded by RBL1) and p130 (encoded by RBL2) can functionally compensate for Rb loss by binding to similar chromatin sites (22, 23). Indeed, although RBL1-positive cells remained at low abundance in subpopulation 1 [Fig. 1C (middle)], RBL2 was highly expressed in most cells within this group [Fig. 1C (right)], indicating its primary role in compensating for RB1 loss. Conversely, neither RBL1 nor RBL2 was expressed in subpopulation 2. Based on the scores of this RBF gene signature (RB1/RBL1/RBL2), we redefined subpopulation 1 as Rb-proficient/RBF⁺ and subpopulation 2 as Rb-deficient/RBF⁻ (Fig. 1D). Interestingly, when we examined a mCRPC patient cohort from Stand Up To Cancer (SU2C) (28), RBL2 expression was significantly lower in tumors lacking RB1, suggesting that CRPC may further adapt to an RBF⁻ state to achieve complete Rb deficiency (Fig. 1E). In contrast, AR was strongly expressed in both Rb-proficient and Rb-deficient subgroups and repressed by Hi-T (Fig. 1F). Notably, Rb-deficient clusters exhibited substantial transcriptomic overlap between the vehicle-treated and Hi-T–treated groups, whereas Rb-proficient clusters were more distinctly separated under these two treatment conditions.

Figure 1.

Identification of Rb-proficient and Rb-deficient subpopulations in the 96CR PDX model. A and B, 96CR tumors were subcutaneously established in castrated male SCID mice and treated daily with testosterone (i.p. injection) at 40 mg/kg for 1 week. The tumor samples were then subjected to scRNA-seq analysis. Projection of scRNA-seq data onto UMAP coordinates and clusters (A) or UMAP for Hi-T–treated vs. vehicle-treated cells (B). C, UMAP plot for RB1, RBL1, and RBL2 expression. D, UMAP plot for RBF signature scores. E, Gene expression of RBL2 in human mCRPC samples (SU2C cohort, Abida 2019) grouped by different RB1 statuses. F, UMAP plot for AR expression. Sub, subpopulation; Veh, vehicle.

We further analyzed cell lineages using various markers: luminal (KRT8, KRT18, and KRT19), basal (KRT14 and TP63), mesenchymal (VIM and S100A4), and neuroendocrine (SYP). As shown in Fig. 2A, luminal epithelial markers were present in all clusters except for clusters 10 and 13. Notably, basal markers were absent across all clusters (Fig. 2B). Neuroendocrine markers sporadically appeared in some cells within the RBF⁺ population, as shown in Fig. 2C. This observation is intriguing, considering previous research indicating that loss of RB1 may facilitate the neuroendocrine transformation of CRPC cells (29, 30). Yet, in the 96CR model, cells in the RBF⁻ population did not exhibit neuroendocrine marker expression. Lastly, mesenchymal markers were specifically found in clusters 10 and 13, belonging to the Rb-deficient group, as shown in Fig. 2D.

Figure 2.

The 96CR tumor contains luminal epithelial and mesenchymal cells. A–D, UMAP plot for gene expression of luminal markers (A), basal markers (B), a neuroendocrine marker (C), and mesenchymal markers (D).

Rb-deficient cells exhibit partial resistance to Hi-T treatment

Our previous research demonstrated that the antitumor effect of Hi-T primarily induces growth arrest by inhibiting the G1/S phase transition (20). Hi-T–activated AR facilitates the formation of the AR–Rb/p130–E2F complex, which suppresses the transcription of G1/S genes, thereby blocking entry into the S phase (20, 22). To understand how Hi-T affects cell-cycle progression in the 96CR model, we used cell-cycle markers for the analysis. As shown in Fig. 3A–C, vehicle-treated cells had a considerable proportion of cells in the S and G2/M phases, indicative of active proliferation. In contrast, Hi-T–treated cells predominantly accumulated in the G1 phase, indicating general growth arrest in tumor cells.

Figure 3.

Rb-deficient cells exhibit partial resistance to Hi-T treatment. A and B, UMAP plots for cell-cycle distribution of Hi-T–treated and vehicle-treated cells by cell-cycle phases (A) or treatments (B). C and D, Stacked bar plots for the percentage of cells at each cell-cycle phase in vehicle vs. Hi-T subgroups (C) or in vehicle vs. Hi-T subgroups with different RBF statuses (D). E and F, UMAP plots for the cell-cycle phase distribution in RBF⁺ (E) or RBF⁻ subgroups (F). Veh, vehicle.

We then examined the differential responses between the Rb-proficient and Rb-deficient populations. We categorized cell populations into four groups: Veh (RBF⁻), Hi-T (RBF⁻), Veh (RBF⁺), and Hi-T (RBF⁺). As shown in Fig. 3D and E, RBF⁺ cells showed an almost complete response to Hi-T, with very few cells progressing to the S and G2/M phases. However, RBF⁻ cells exhibited partial resistance to Hi-T, with a significant number of cells continuing to proliferate through the S and G2/M phases (Fig. 3D and F). These findings suggest that although Rb-deficient tumor cells may still show some degree of reduced proliferation in response to Hi-T, they clearly exhibit significant resistance to Hi-T therapy compared with Rb-proficient tumor cells.

Hi-T treatment represses E2F and MYC signaling

We then evaluated the transcriptional impact of Hi-T in this 96CR model by conducting GSEA on RBF⁺ and RBF⁻ cell groups. As demonstrated in Fig. 4A, the results indicated enrichment of Hi-T upregulated genes in the classic androgen response and cholesterol synthesis pathways, the latter being a key AR target (31), in both cell types. Notably, in RBF⁺ cells, Hi-T downregulated genes primarily linked to E2F, MYC, and cell-cycle pathways, aligning with previous studies (17, 22, 23, 25). In contrast, this downregulation was less prominent in RBF⁻ cells, suggesting a diminished repression capability of AR in these cells. The mechanism behind Hi-T’s suppression of the E2F and MYC pathways did not seem to be primarily due to the inhibition of E2F1 and MYC genes themselves (Fig. 4B). Interestingly, Hi-T specifically suppressed the oxidative phosphorylation pathway in RBF⁻ cells, indicating a unique Hi-T activity in repressing metabolic functions in this cell type. Importantly, this effect may favor tumor hypoxia, in which the oxidative phosphorylation is shifted to anaerobic glycolysis. Overall, these findings reinforce the notion that Hi-T primarily exerts its growth-inhibitory effects by hindering cell-cycle progression, likely through the suppression of the E2F and MYC signaling pathways.

Figure 4.

Hi-T treatment suppresses E2F and MYC signaling. A, GSEA bubble plot for differentially expressed genes in Hi-T–treated vs. vehicle-treated cells (RBF⁺ or RBF⁻ subpopulation). B, Violin plots for E2F1 and MYC expression in the four subgroups of cells. NES, normalized enrichment score; ns, not significant; Veh, vehicle.

Hi-T–repressed genes are associated with prostate cancer aggressiveness

Next, we sought to develop Hi-T target signatures with potential clinical significance. We established a 19-gene signature representing the transcriptional repression effect of Hi-T in Rb-proficient cells (Fig. 5A) and a distinct 21-gene signature for the Rb-deficient group (Fig. 5B). To assess the clinical impact of these Hi-T repressed markers in patients with mCRPC resistant to ARSi treatment, we conducted Kaplan–Meier survival analysis using the SU2C mCRPC dataset (28). This involved examining enrichment scores categorized into the top 25% and bottom 75%. As shown in Fig. 5C and D, the RBF⁺ Hi-T signature (19-gene) correlated significantly with poorer overall survival after ARSi treatment, whereas the RBF⁻ Hi-T signature (21-gene) showed no significant association with clinical outcomes. For additional validation, we analyzed publicly available bulk RNA-seq data from the 35CR, 77CR, and 96CR PDX models (GSE188174; ref. 25) and our previous RNA-seq data for 35CR (GSE179688; ref. 22), all of which are recognized as Rb-proficient models. The results revealed that only the 19-gene RBF⁺ signature was consistently and significantly repressed by Hi-T across all these models (Fig. 5E and F). Collectively, these analyses established a clinically relevant signature that can effectively measure the transcriptional response of Hi-T in Rb-proficient CRPC.

Figure 5.

Hi-T–repressed genes in Rb-proficient cells are associated with prostate cancer aggressiveness. A and B, Heatmaps of scaled expression of a 19-gene Hi-T–repressed signature established from RBF⁺ cells (A) or a 21-gene signature from RBF⁻ cells (C). C and D, Kaplan–Meier analyses for the overall survival from the initiation of first-line ARSi treatment in tumors with higher scores (yellow, top 25%) of the 19-gene signature (C) or 21-gene signature (D) vs. with lower scores (blue, bottom 75%). E and F, Box plots for Hi-T–repressed gene signature scores in LuCaP CRPC PDX models treated with vehicle vs. Hi-T from two studies: GSE188174 (E) and GSE179688 (F). Veh, vehicle.

Targeting hypoxia signaling restores Hi-T sensitivity in Rb-deficient cells

We next sought to identify vulnerabilities in Rb-deficient cells that could be exploited to restore their sensitivity to Hi-T treatment. GSEA analysis comparing RBF⁺ and RBF⁻ cells under vehicle treatment revealed significant activation of TNF signaling via NF-κB and hypoxia signaling pathways in Rb-deficient cells (Fig. 6A–C). This finding is consistent with previous research indicating that RB1 loss can enhance the transcriptional response to hypoxia (32). Notably, two hypoxia pathway genes, BTG1 and PMEPA1, were expressed at higher levels in Rb-deficient cells than in Rb-proficient cells (Fig. 6D). To further investigate this phenomenon, we constructed a 10-gene hypoxia signature comprising well-established hypoxia pathway genes (HIF1A, VEGFA, EPO, LDHA, SLC2A1, CA9, PGK1, EGLN1, BNIP3, and LOX; refs. 33–35). Analysis of the SU2C patient data revealed a significant elevation of this signature in RB1-loss CRPC (Fig. 6E), suggesting that these aggressive tumors may display a more robust hypoxia response than their Rb-proficient counterparts.

Figure 6.

Rb-deficient cells exhibit enhanced hypoxia signaling. A, GSEA bubble plot for differentially expressed genes in vehicle-treated RBF⁻ vs. vehicle-treated RBF⁺ subpopulations. B and C, Heatmap of a panel of TNFα–NFκB pathway genes (B) and hypoxia pathway genes (C). D, UMAP plot for BTG1 and PMEPA1 expression. E, Hypoxia signaling score (10 common genes) in human mCRPC samples (SU2C cohort) grouped by different RB1 statuses. Veh, vehicle.

Based on this finding, we hypothesized that elevated hypoxia signaling in Rb-deficient cells may significantly contribute to their partial resistance to Hi-T. We selected the C4-2 CRPC cell line model for subsequent testing, as these AR-positive cells have been previously used as an Rb-proficient model and display a robust response to Hi-T, whereas silencing RB1 or other Rb-like genes significantly impairs this response (20, 22). Indeed, analysis of a previously generated RNA-seq dataset (GSE176404; ref. 36) revealed that Hi-T suppressed the Rb-proficient Hi-T signature in control C4-2 cells, whereas it suppressed the Rb-deficient Hi-T signature in RB1-silenced C4-2 cells (Fig. 7A and B). These results suggest that the mechanism of Hi-T action in C4-2 cells may mirror that observed in the 96CR model. Using two previously established RB1 knockout (KO) C4-2 CRPC cell clones (22, 37), we found that the expression of HIF-1α, a key hypoxia transcription factor, was stimulated by 10 nmol/L DHT (mimicking the physiologic androgen levels in uncastrated men), and this effect was much more noticeable in RB1 KO lines (Fig. 7C). To further examine the hypoxic response under normoxic conditions, we treated these cells with the clinically used hypoxia inducer roxadustat, which inhibits HIF-1α hydroxylation and prevents its degradation (38). Notably, we observed a more dramatic increase of HIF-1α protein in RB1 KO cells compared with control cells (Fig. 7C), suggesting that the HIF-1α–mediated hypoxic response is more active in Rb-deficient lines. We then proposed that targeting hypoxia signaling could potentially restore Hi-T sensitivity. To test this, we treated RB1 KO C4-2 cells with the HIF-1α inhibitor PX-478, which blocks HIF-1α transcription and promotes its degradation (26, 39, 40). Consistent with previous findings (22), the tumor-suppressive effect of 10 nmol/L DHT was substantially weakened by RB1 loss (Fig. 7D), although not completely abolished, presumably because of the compensatory function of other Rb family proteins. Strikingly, although PX-478 alone only modestly affected cell growth, combining PX-478 with DHT almost completely restored the antitumor activity of high-dose DHT under normoxia. This effect was also confirmed using an E2F/Rb target, MCM2, a DNA replication and S phase cell-cycle marker (Fig. 7E). Together, this pilot study highlights a potential combination therapy strategy for CRPC, in which hypoxia inhibitors can resensitize Rb-deficient tumors (or tumor cell subpopulations) to Hi-T treatment.

Figure 7.

HIF-1α inhibition resensitizes Rb-deficient CRPC cells to Hi-T treatment. A and B, Heatmap plots for Hi-T–repressed Rb-proficient signature genes in control C4-2 cells (A) or Rb-deficient signature genes in RB1-silenced C4-2 cells (B) using our previously generated RNA-seq dataset (GSE176404). C, Immunoblotting for Rb, HIF-1α, and AR expression in control C4-2 cells vs. C4-2-derived RB1-KO lines (two clones: RB1-KO-1 and RB1-KO-2) treated with 10 nmol/L DHT, 5 µmol/L roxadustat, or the combination for 24 hours. D, Proliferation assay for control C4-2 cells and RB1-KO-1 with the indicated treatments (10 nmol/L DHT, 5–50 µmol/L PX-478, or the combination for 6 days). E, Immunoblotting for MCM2 expression (DNA replication marker) in RB-KO-1 cells treated with 10 nmol/L DHT, 5 µmol/L PX-478, or the combination for 24 hours. Ctrl, control; Veh, vehicle. ***, P value < 0.001; ****, P value < 0.0001.

Discussion

Hi-T treatments have shown promise in treating a subgroup of patients with CRPC by hindering tumor growth (6–9). The primary mechanism behind this effect seems to involve blocking DNA synthesis and cell-cycle progression by inhibiting the E2F and MYC signaling pathways (11, 16, 19–23, 25). Recent research efforts are focused on identifying predictors of tumor response to these treatments (41, 42). Our previous studies indicate that this antitumoral effect is achieved through the enhanced interaction of the AR and Rb protein at chromatin, which strengthens the Rb–E2F repressive complex, and recent studies have also emphasized the crucial role of Rb-like proteins in modulating the tumor-suppressive effects of AR (20, 22, 23). Nevertheless, the heterogeneity of CRPC tumors has not been fully explored in this context. To address this, we conducted scRNA-seq on a CRPC PDX model, 96CR. Initially identified as an RB1 homozygous deletion model, this PDX line was later found to contain a mix of cell types, including both Rb-proficient and Rb-deficient cell populations (22, 24).

Unsupervised clustering analysis revealed 14 unique cell clusters, consisting predominantly of luminal epithelial cells and a small group of mesenchymal cells. When integrating the data, two distinct subgroups emerged, RBF⁺ and RBF⁻, based on RBF expression. Notably, most RBF⁺ cells exhibited high levels of RBL2 but not RB1, consistent with our previous findings that identified RBL2/p130 as a key modulator of the Rb–E2F pathway (22). This observation also suggests that, in clinical settings, the expression of Rb-like proteins should be carefully assessed when evaluating Rb deficiency in CRPC tumors. Our cell-cycle analysis revealed that Hi-T induced nearly complete G1 arrest in RBF⁺ cells but was less effective at blocking the G1/S transition in RBF⁻ cells, thereby allowing a subset of tumor cells to continue proliferating. These results align with prior evidence indicating that Rb and Rb-like proteins play crucial roles in the tumor-suppressive effects of Hi-T (22, 23). Transcriptomic analysis further revealed that in Rb-proficient tumor cells, Hi-T downregulated genes involved in E2F and MYC signaling, whereas this effect was attenuated in Rb-deficient tumors, corroborating our earlier findings in the RB1-silenced C4-2 model (22). Another key discovery of this study is the development of a Hi-T target signature derived from Rb-proficient cells. This signature, which accurately predicts the transcriptional impact of Hi-T in other responsive PDX and cell line models, correlates significantly with poorer patient outcomes, suggesting its potential as a biomarker for evaluating the efficacy of Hi-T treatment in clinical trials and patient studies.

Finally, our study revealed that Rb-deficient cells show increased activity in hypoxia pathways compared with Rb-proficient cells. This finding is significant because the hypoxia response pathway plays a crucial role in the progression of prostate cancer (43, 44). Additionally, the metastatic bone tumor environment is inherently hypoxic (45). Interestingly, Rb regulation of hypoxia signaling is well documented (32), and enhanced hypoxia pathway seems to be a hallmark of RB1-loss CRPC (see Fig. 6E). Given the established connection between hypoxia signaling and tumor progression (46), we propose that this mechanism may contribute to the observed partial resistance of Rb-deficient cells to Hi-T therapy. Previous studies have indicated that HIF-1α signaling can significantly alter the transcriptome and cistrome of AR (47, 48), hinting at how hypoxia signaling might interfere with AR transcriptional repression activity. Our data also showed that androgens can stimulate HIF-1α expression, consistent with previous findings (49), and this effect was greatly enhanced in Rb-deficient cells. This suggests that Hi-T treatment may somehow induce the activation of HIF-1α–mediated hypoxia response, allowing tumor cells to alter their metabolic profile and potentially survive Hi-T–induced cell-cycle arrest. Although the exact mechanisms are yet to be fully understood, our pilot experiments with RB1-silenced CRPC cell line models indicated that inhibiting HIF-1α may resensitize Rb-deficient cells to Hi-T. PX-478, an HIF-1α inhibitor tested in preclinical models (39, 40), showed potential for restoring the effectiveness of Hi-T in these cases.

There are, however, some limitations to this study. First, the scRNA-seq data were derived from a single CRPC PDX model, 96CR, which contains mixed cell populations; therefore, future studies should include multiple AR-positive tumor models with diverse genetic backgrounds. Second, we tested the combination treatment only in the C4-2 cell line, as it mirrors the Hi-T response observed in the 96CR model. To obtain more comprehensive preclinical data, subsequent investigations should incorporate additional cell lines and PDXs and be tested in mouse models. Despite these limitations, we believe that our findings lay a strong foundation for further preclinical testing and rapid initiation of clinical trials, potentially leading to novel treatment strategies for AR-positive, Rb-deficient CRPC.

Authors’ Disclosures

C. Cai reports grants from the Department of Defense (United States) and grants from the National Cancer Institute (United States) during the conduct of the study. No disclosures were reported by the other authors.

Authors’ Contributions

M. Labaf: Conceptualization, software, formal analysis, investigation, methodology, writin–original draft. W. Han: Conceptualization, data curation, investigation, methodology. S. Zhang: Investigation, methodology. M. Liu: Software, formal analysis. N.D. Patten: Data curation, investigation. M. Li: Investigation, methodology. S. Patalano: Data curation, methodology. J.A. Macoska: Software, formal analysis, supervision. S.P. Balk: Conceptualization, resources, supervision. D. Han: Conceptualization, supervision, methodology, project administration, writing–review and editing. K. Zarringhalam: Software, formal analysis, supervision. C. Cai: Conceptualization, resources, data curation, supervision, funding acquisition, validation, investigation, writing–original draft, project administration, writing–review and editing.