Cell culture media and supplements, including RPMI 1640, DMEM, and fetal bovine serum (FBS), were purchased from Life Technologies (Thermo Fisher Scientific, Waltham, MA, United States). The details of the reagents and antibodies are presented in Supplementary Table S1.

Hormone-sensitive prostate cancer cells (LNCaP cells, RRID: CVCL_0395) were procured from Servicebio (Wuhan, China), castration-resistant prostate cancer cells (C4-2B cells, RRID: CVCL_4784) from BeNA Culture Collection (Henan, China), and human prostate stromal myofibroblast cell line (WPMY-1 cells, RRID: CVCL_3814) from Gain Biological (Shanghai, China). Each cell line was cultured in RPMI 1640 medium supplemented with 10% FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin (Sigma–Aldrich, St. Louis, MO, United States). The cells were incubated at 37°C in an atmosphere containing 5% CO₂. All cell lines used in this study were authenticated via short tandem repeat (STR) profiling within the past 3 years. Furthermore, all experiments were conducted using Mycoplasma-free cells to ensure the accuracy and reliability of the data.

CAF formation was induced using TGF-β1, as previously described (Zhang et al., 2020). Briefly, WPMY-1 cells were seeded in a six-well plate at a density of 5 × 10⁴ cells/well. After 48 h of incubation, the medium was discarded and the cells were incubated in complete medium containing TGF-β1 at varying concentrations (0, 10, and 20 ng/mL). Cell morphology was observed after 48 h of TGF-β1 treatment using an optical microscope (BX51; OLYMPUS, Tokyo, Japan). Additionally, the identity of the induced CAFs was confirmed by assessing the expression of fibroblast activation protein (FAP) and α-smooth muscle actin (α-SMA) using immunofluorescence and Western blot assays.

CAFs were treated with the Wnt inhibitor DKK-1 (20 ng/mL) or an SDF-1 neutralizing antibody (1000 nmol/L) for 24 h. WPMY-1 cells, along with untreated and treated CAFs, were seeded separately in six-well plates at a density of 1.5 × 10⁵ cells/mL and cultured for 72 h. After incubation, the medium was collected and centrifuged at 300 g for 5 min at room temperature to remove cellular debris. The resulting supernatants were collected as conditioned media (CM) for further analysis: WPMY-1-CM, CAF-CM, CAF^anti−Wnt-CM, and CAF^anti−SDF-1-CM.

The aforementioned CAF-CM and WPMY-1-CM were collected, and Wnt3a and SDF-1 levels were analyzed using ELISA. Detailed experimental procedures can be found in the Supplementary Methods.

C4-2B cells were washed 1–2 times with PBS containing 5% FBS (staining solution). To the staining solution (50 μL), 1–2 μL of anti-human Fc blocking antibody (1:100, eBioscience, San Diego, CA, United States) was added, and the cells were incubated at 4°C for 30 min. Next, 2 μL of APC anti-human CD133 antibody (1:100; Biolegend, San Diego, CA, United States) and 2 μL of PE anti-human CD44 antibody (1:100; Biolegend) or isotype control antibodies (1:100; Biolegend) were added to the staining solution (50 μL), and the samples were incubated at 4°C for 30 min. After incubation, the cells were washed twice with PBS. The cell concentration was adjusted to 1 × 10⁶/mL, and the CD133⁺CD44⁺ C4-2 cells were sorted using a flow cytometer (FACS Aria III; BD Biosciences, Franklin Lakes, NJ, United States). The sorted CD133⁺CD44⁺ cells were seeded in a serum-free tumor sphere culture medium supplemented with EGF and bFGF, and cultured in suspension on low-attachment plates. The cells were treated with WPMY-1-CM or CAF-CM for 24 h, and treatments were repeated every 3 days. After 14 days, sphere formation was observed under a microscope (BX51; OLYMPUS). To assess stemness and self-renewal capacity, the sphere formation rate was calculated based on the number of cell aggregates with a diameter greater than 50 μm and a compact, rounded morphology.

PCSCs were seeded in six-well plates at a density of 5 × 10⁴ cells per well. CAFs were cultured separately on the 0.4-μm polyester membrane of a 12-mm Transwell insert (Corning, New York, NY, United States). After 24 h, the Transwell inserts were transferred into the wells containing PCSCs to establish an indirect co-culture system. PCSCs and CAFs were co-cultured at ratios of 1:3, 1:6, and 1:9. Cell proliferation was evaluated using the CCK-8 assay to determine the optimal ratio for subsequent experiments.

TOPFlash and FOPFlash luciferase activities were measured using a dual-luciferase reporter assay to assess the activity of the Wnt/β-catenin signaling pathway after the treatment of PCSCs with CAF-CM and CAF^anti−SDF-1-CM. Detailed experimental procedures are provided in the Supplementary Methods.

Male NSG mice (4 weeks old, average weight 20 ± 3 g) were obtained from Vital River (Beijing, China). To establish a subcutaneous tumor model mimicking the tumor microenvironment, prostate cancer stem cells (PCSCs, 1 × 10⁶) were injected alone or in combination with cancer-associated fibroblasts (CAFs, 6 × 10⁶) at a 1:6 ratio into the right flank of each mouse in 100 μL of PBS/Matrigel (1:1). This co-injection model was designed to reflect the stromal support provided by CAFs in vivo. The mice were randomly divided into four groups (n = 4 per group). Group A received PCSCs alone. Groups B–D received PCSC + CAFs. Once tumor volumes reached approximately 50 mm³, the mice were treated as follows: (a) PCSC alone + saline; (b) PCSC + CAFs + saline; (c) PCSC + CAFs + XAV939 (10 mg/kg, i.p.; MedChemExpress, Monmouth Junction, NJ, United States); and (d) PCSC + CAFs + AMD3100 (5 mg/kg, i.p.; MedChemExpress). XAV939 and AMD3100 were administered daily for 21 consecutive days. The drug concentrations were selected based on previously published studies and validated through preliminary experiments for determining dose range to ensure tolerability and efficacy. After 21 days of treatment, the mice were anesthetized with pentobarbital sodium (80 mg/kg, i.p.) and euthanized. Tumors were excised and measured; their volumes were calculated as V = 0.5 × H² × L, where H is the shorter diameter and L is the longer diameter. All animal procedures were approved by the Ethics Committee of Guang’anmen Hospital (Approval No.: IACUC-GAMH-2024-058-SQ) and conducted in accordance with institutional and national guidelines.

Immunofluorescence was used to determine Wnt, β-catenin, SDF-1, CXCR4, and p-GSK-3β expression in excised tumor tissues, and immunohistochemistry was employed to assess the expression of FAP and α-SMA in tumor tissues. Detailed experimental steps can be found in the Supplementary Methods.

Publicly available single-cell RNA sequencing data (GSE193337) (Song et al., 2015) and transcriptome sequencing data (GSE32269) (Zhao et al., 2014) were obtained from the Gene Expression Omnibus database. The GSE193337 dataset contains data of human tumor and benign prostate adenocarcinoma samples, including both single-cell and bulk RNA sequencing data. The GSE32269 dataset contains gene expression data of primary localized and castration-resistant bone metastatic PCa samples. For the GSE193337 dataset, analysis was performed using the Seurat package (Lu et al., 2015). The analysis steps included quality control (removal of low-quality cells and genes), dimensionality reduction (principal component analysis and t-SNE), cell clustering, and differential gene expression analysis. For the GSE32269 dataset, differential expression analysis was conducted using the edgeR software package (version 4.0.2). First, the raw data were normalized, followed by quantitative analysis to identify genes with significant expression differences between groups. mRNA expression data for correlation analysis were obtained from the UCSC XENA database. The data were converted into transcripts per million format and standardized using log₂ transformation. To preprocess raw RNA-seq count data (GSE32269), low-quality reads were removed, gene expression was calculated, and a count matrix was generated. These preprocessing steps were performed using STAR software for genome alignment and featureCounts for counting reads mapped to genes. The final count matrix was used as input for differential expression analysis.

Tumor tissues were sectioned and promptly cryopreserved in liquid nitrogen. Briefly, tissue samples were pulverized into powder using a tissue grinder, followed by RNA extraction using an adequate quantity of TRIzol reagent. cDNA was generated from the extracted RNA using reverse transcriptase. Thereafter, RT-qPCR was performed using ABI 7500 RT-PCR equipment (Thermo Fisher Scientific) with specific reagents and primers. The PCR sequences are listed in Supplementary Table S2.

Proteins were extracted on ice using RIPA lysis buffer containing protease inhibitors, and protein concentration was measured using a BCA protein assay kit (Thermo Fisher Scientific). Protein samples (20–50 μg) were separated using SDS-PAGE and transferred onto PVDF membranes. The membranes were blocked with 5% non-fat milk for 1 h and incubated overnight with primary antibody (1:1000) at 4°C, and then with secondary antibody (1:2000) for 1 h at 20°C–25°C. Immunoreactive bands were detected using a ChemiDoc XRS + chemiluminescence detection system (Bio-Rad, Hercules, CA, United States) and analyzed using ImageJ software. Protein expression was normalized to the expression of the internal control GAPDH.

Statistical analysis was performed using GraphPad Prism 9 (GraphPad Software, San Diego, CA, United States). All data are presented as mean ± standard deviation from at least three independent experiments. For comparisons among multiple groups, one-way analysis of variance (ANOVA) was conducted, followed by appropriate post hoc tests, including Tukey’s, Sidak’s, and Dunnett’s multiple comparison tests. For comparisons between groups, independent sample t-tests were used. Results with p-value <0.05 were considered statistically significant.

To investigate the role of CAFs in PCSC stemness regulation during CRPC progression, we established mouse models bearing hormone-sensitive prostate cancer (HSPC) (LNCaP) or CRPC cells (C4-2B). In vivo, C4-2B tumors grew significantly faster and larger (p = 0.0105) than LNCaP tumors; C4-2B and LNCaP tumor volumes reached 1092.78 ± 154.71 and 720.00 ± 84.49 mm³, respectively, by 27 days post-inoculation (Figure 1A).

Further immunohistochemical analysis showed that the CAF markers FAP and α-SMA exhibited significantly higher expression in C4-2B tumor tissues than in LNCaP tumors (FAP: p = 0.0047, α-SMA: p = 0.0262) (Figure 1B). Similarly, immunofluorescence staining confirmed that the expression of FAP (p = 0.0024) and α-SMA (p = 0.0092) was significantly higher in C4-2B than in LNCaP tumors (Figure 1B). Additionally, Western blotting showed that FAP (6.88 ± 0.45 vs. 4.80 ± 0.53; p = 0.0014) and α-SMA (8.61 ± 0.33 vs. 5.52 ± 0.93; p < 0.0001) protein levels were significantly higher in C4-2B than in LNCaP tumors (Figure 1C). Moreover, the expression of the CSC markers CD133 (4.76 ± 0.49 vs. 3.46 ± 0.47; p = 0.0475) and CD44 (5.20 ± 0.44 vs. 3.64 ± 0.28; p = 0.0144) was significantly higher in C4-2B than in LNCaP tumors (Figure 1C).

Notably, α-SMA expression and the relative fluorescence intensities of CD133 (p = 0.0359) and CD44 (p = 0.0156) were significantly higher in C4-2B than in LNCaP tumors. Additionally, dual staining revealed spatial proximity between the CAF marker α-SMA and tumor stem cell markers CD133 and CD44 (Figure 1D).

TGF-β1 is known for its essential role in the induction of CAFs (Pan et al., 2024). In this study, WPMY-1 cells were treated with 10 ng/mL TGF-β1, a concentration selected based on optimization experiments using different TGF-β1 concentrations (Supplementary Figure S1). Immunofluorescence staining revealed a marked increase in FAP and α-SMA expression following TGF-β1 treatment compared with that in the untreated group (Figure 2A). Western blotting confirmed the increased expression of FAP (p < 0.0001) and α-SMA (p < 0.0001) in CAFs, highlighting that the induced cells displayed typical CAF characteristics (Figure 2B). Overall, these results indicate that TGF-β1 successfully induced the transformation of WPMY-1 cells into highly activated CAFs.

The stem cell markers CD44 and CD133 can be used to sort cell subtypes among C4-2B cells (Li and Mu, 2023; Wang et al., 2023). Flow cytometry analysis showed that CD44⁺CD133⁺ prostate cancer stem-like cells accounted for 0.86% of the LNCaP population and 5.30% of the unsorted C4-2B population. After sorting, the proportion of CD44⁺CD133⁺ cells in C4-2B increased to 97.56%, indicating successful enrichment of the stem-like subpopulation (Figure 2C). In the Transwell co-culture system, the CCK8 assay indicated a marked increase in the proliferation rate of PCSCs cultured with CAFs (1:6 ratio) (Figure 2D). To further investigate the effect of CAFs on PCSC stemness, we examined the spheroid formation rate in the PCSC and PCSC–CAF (1:6) groups. Compared with that in the PCSC-only group (76.13 ± 4.33 spheroids/well), spheroid formation increased significantly (p = 0.0125) to an average of 108.43 ± 9.66 spheroids/well in the PCSC–CAF co-culture group, underscoring the essential role of CAFs in sustaining PCSC stemness (Figure 2E).

To explore the cell-type-specific expression landscape of key signaling molecules in PCa, we performed dimensionality reduction using the t-SNE algorithm based on single-cell transcriptomic data (Figure 3A). This analysis enabled visualization of diverse stromal and immune subpopulations within the PCa microenvironment. Among these, CAFs formed a distinct cluster, prompting further investigation of their molecular characteristics.

A dot plot analysis of key genes related to Wnt signaling and stemness regulation revealed that CXCR4 and CTNNB1 (the gene encoding β-catenin) were widely expressed across multiple cell types, including fibroblasts, with particularly high levels in CRPC (Figure 3B). The high variability in expression suggests their dynamic involvement in tumor progression. Furthermore, analysis of TCGA data revealed a statistically significant but weak positive correlation between CXCR4 and CTNNB1 expression (R = 0.3, p < 0.0001), as well as modest correlations between CXCR4 and other Wnt-related genes, such as FZD1 and GSK-3β (Figure 3C), suggesting that these molecules may be co-regulated within a shared signaling axis.

Additionally, immunofluorescence staining was performed to investigate whether Wnt3a, SDF-1, and α-SMA are co-localized to C4-2B and LNCaP tumors. In the immunofluorescence co-localization experiment, the expression of Wnt3a and SDF-1 was significantly higher in C4-2B tumors than in LNCaP tumors. Moreover, Wnt3a/α-SMA (p = 0.0125) and SDF-1/α-SMA (p = 0.0073) co-localization ratios were significantly higher in C4-2B (4.31% ± 2.85% and 20.02% ± 3.30%, respectively) than in LNCaP tumors (Figure 3D). Furthermore, RT-qPCR indicated that the Wnt3a (p < 0.0001) and SDF-1 (p < 0.0001) mRNA levels were significantly higher in CAFs than in WPMY-1 cells (Figure 3E). Collectively, these results suggest that CAF activation enhanced the tumorigenicity of CRPCs by secreting or locally overexpressing Wnt3a and SDF-1, potentially activating downstream pathways.

Microscopic images showed that CAF-CM treatment significantly enhanced the sphere-forming ability of PCSCs compared to WPMY-1-CM (Figure 4A). Western blotting showed the marked upregulation of CD133 and CD44 expression in PCSCs treated with CAF-CM (Figure 4C), suggesting that CAFs enhanced stem cell-like properties in PCa via a paracrine pathway.

The ELISA showed that the Wnt3a and SDF-1 levels were markedly higher in CAF-CM than in WPMY-1-CM (Figure 4B). Upon inhibition of Wnt3a and SDF-1 expression in CAFs-CM, the number of spheres formed by PCSCs was significantly decreased (Figure 4A). Additionally, inhibition of Wnt or SDF-1 expression in CAF-CM significantly downregulated CD133 and CD44 expression in PCSCs (Figure 4C), indicating that CAF-derived Wnt3a and SDF-1 were essential for promoting stem cell-like characteristics in PCSCs.

Notably, CAF-CM treatment markedly upregulated β-catenin and CXCR4 expression in PCSCs. Compared to CAF-CM treatment, CAF^anti−Wnt-CM treatment significantly reduced β-catenin level in PCSCs, whereas CAF^anti−SDF-1-CM treatment significantly inhibited both CXCR4 and β-catenin expression (Figure 4C). Considering the strong association between the SDF-1/CXCR4 axis and β-catenin in PCa, this result suggests that CXCR4 enhanced stem cell-like properties in PCa by activating β-catenin.

To test this hypothesis, we used the TOPFlash luciferase assay to evaluate the effect of SDF-1/CXCR4 signaling on β-catenin downstream transcriptional activity. Importantly, TOPFlash and FOPFlash activities were significantly downregulated in PCSCs incubated in CAF-CM treated with an SDF-1 neutralizing antibody (Figure 4D), further supporting the role of SDF-1 in driving Wnt/β-catenin pathway activation in PCSCs. Collectively, these findings suggest that activated CAFs may be crucial in maintaining stemness and promoting CRPC progression by expressing Wnt3a and SDF-1, thereby activating the Wnt/β-catenin and SDF-1/CXCR4 signaling pathways.

Tumor growth was faster and tumor weight was significantly higher (p = 0.00413) in mice co-implanted with CAF–PCSC than in those implanted with PCSCs only. However, treatment with the Wnt inhibitor XAV939 (p = 0.0008) and the CXCR4 inhibitor AMD3100 (p = 0.0117) significantly reduced tumor weight in the CAF–PCSC co-implantation group (Figure 5A). Immunofluorescence staining showed that Ki-67 expression was significantly higher in the CAF–PCSC group than in the PCSC group, indicating that CAFs enhanced the in vivo growth of PCSCs. Similarly, XAV939 and AMD3100 treatments significantly suppressed Ki-67 expression in PCSCs in vivo (Figure 5A).

Western blotting showed that Wnt3a, β-catenin, SDF-1, and CXCR4 expression and the p-GSK-3β/GSK-3β ratio were significantly higher in the CAF–PCSC group than in the PCSC group (Figure 5B). RT-qPCR demonstrated that CAF treatment increased the expression of Wnt3a, β-catenin, TCF/LEF (the primary downstream effector of the Wnt pathway), SDF-1, and CXCR4 in PCSCs in vivo. Additionally, Wnt3a, β-catenin, p-GSK-3, SDF-1, and CXCR4 fluorescence intensities were higher in the CAF–PCSC group than in the PCSC group (Figure 6A).

Notably, XAV939 treatment suppressed Wnt3a, β-catenin, p-GSK-3, and TCF/LEF expression but did not affect SDF-1 or CXCR4 expression compared with that in the CAF–PCSC group. Additionally, AMD3100 treatment markedly decreased SDF-1, CXCR4, β-catenin, p-GSK-3, and TCF/LEF expression (Figures 5B, 6A,B).

Immunofluorescence staining showed a significant increase in the fluorescence intensities of β-catenin (p = 0.0160) and CXCR4 (p = 0.0345) in the CAF–PCSC group. Additionally, the co-localization ratio of β-catenin and CXCR4 was significantly higher (p = 0.0023) in the CAF–PCSC (39.71% ± 3.71%) group than in the PCSC group (21.32% ± 4.56%; Figure 5C). Moreover, XAV939 and AMD3100 treatments significantly reduced the relative expression and co-localization of β-catenin and CXCR4. XAV939 and AMD3100 treatments significantly decreased the co-localization ratios of β-catenin and CXCR4 to 11.04% ± 2.27% (p = 0.0001) and 6.96% ± 1.95% (p < 0.0001), respectively.

Based on the results of the in vitro and in vivo experiments, CXCR4 promoted prostate cancer stem-like properties by activating β-catenin and its downstream pathways. Conclusively, CAFs promoted PCSC proliferation and tumor growth by enhancing the Wnt/β-catenin and SDF-1/CXCR4 signaling pathways. Inhibitors targeting these pathways effectively suppressed the CAF-induced maintenance of cancer stemness.