Spatial and single-cell multi-omics reveal pro-angiogenic THY1⁺ fibroblast subtypes predicting prognosis in prostate cancer

Highlights

• •A novel THY1⁺ CAF subset drives angiogenesis via CXCL6/CXCR2 signaling.
• •THY1⁺ CAFs exhibit THY1-dependent VEGFA secretion enhancing endothelial tube formation.
• •An 8-gene CAF-derived signature robustly predicts biochemical recurrence across cohorts.
• •High THY1⁺ CAF infiltration correlates with Gleason score and poor survival.

Abstract

Background

Cancer-associated fibroblasts (CAFs) play a key role in prostate cancer (PCa) progression, though their heterogeneity and specific protumorigenic subsets remain poorly characterized. This study aimed to identify and validate a distinct THY1⁺ CAF subset associated with aggressive PCa.

Methods

Multiomics data from public (TCGA-PRAD, GEO) and prospective (FUSCC, n = 84) cohorts were analyzed. An 8-gene CAF-derived prognostic signature was constructed using LASSO Cox regression. THY1⁺ CAF clusters were identified via scRNA-seq. Primary CAFs were isolated from patient tissues, and THY1⁺/THY1⁻ subpopulations were purified via MACS/FACS. Angiogenic function and secretory profiles were assessed through tube formation assays, ELISA, and antibody arrays. THY1 knockdown and CXCR2 inhibition were used for mechanistic studies. Clinical relevance was evaluated via qPCR and multiplex immunohistochemistry on tissue microarrays.

Results

High CAF abundance correlated with aggressive clinicopathological features and poor prognosis in PCa. The 8-gene signature effectively predicted biochemical recurrence (BCR). scRNA-seq revealed THY1⁺ CAFs as a proangiogenic subpopulation. THY1⁺ CAFs enhanced angiogenesis via increased secretion of CXCL6 and VEGFA. CXCL6 promoted endothelial tube formation through CXCR2 activation, while THY1 knockdown downregulated VEGFA and impaired angiogenesis. High THY1⁺ CAF infiltration was associated with significantly worse recurrence-free survival.

Conclusion

THY1⁺ CAFs represent a proangiogenic subset that drives PCa progression via the CXCL6/CXCR2 axis and THY1-mediated VEGFA expression. These findings highlight stromal THY1 and the CXCL6/CXCR2 pathway as potential therapeutic targets.

Graphical abstract

Introduction

Prostate cancer (PCa) represents the most common noncutaneous malignancy in men within the United States and remains a formidable clinical challenge, with an estimated 299,010 new diagnoses and 35,250 deaths anticipated in 2024 [1]. Between 1990 and 2021, the age-standardized mortality rate (ASMR) of prostate cancer in China increased by 7.9 % (–18.4 to 43.6). Provinces with higher per-capita GDP exhibited the greatest disease burden. Prostate cancer has surpassed bladder cancer to become the predominant urological cancer subtype, with the most pronounced rise in age-standardized incidence rate (ASIR) occurring among men aged 55 years and older [2]. PCa is characterized by striking clinical heterogeneity, encompassing a spectrum from indolent, organ-confined lesions to highly aggressive, metastatic disease [3,4]. This variability is driven not only by cancer cell–intrinsic determinants but also by the profoundly influential tumor microenvironment (TME) [5]. The TME comprises immune cells, endothelial cells, and cancer-associated fibroblasts (CAFs) embedded within a remodeled extracellular matrix (ECM) [6,7]. Among these stromal elements, CAFs have emerged as critical orchestrators that facilitate angiogenesis, metastasis, and therapeutic resistance through sustained paracrine interactions and ECM reorganization [8,9].

CAFs exhibit marked heterogeneity in both cellular origin and functional roles, frequently exerting paradoxical effects on tumorigenesis [10,11]. This diversity constitutes a significant obstacle to the development of stromal-targeted therapies, as indiscriminate CAF depletion has produced inconsistent results, at times even accelerating tumor progression [12,13]. Consequently, increasing emphasis has been placed on the precise identification and functional characterization of specific protumorigenic CAF subpopulations [14]. scRNA-seq has been pivotal in disentangling this complexity, uncovering distinct CAF subsets—including myofibroblastic, inflammatory, and antigen-presenting phenotypes—across multiple malignancies such as breast [15,16], pancreatic [17], and colorectal cancer [18], each distinguished by unique molecular markers and specialized functional properties.

In PCa, however, the heterogeneity of CAFs remains comparatively underexplored. Although elevated CAF abundance is consistently associated with aggressive disease and unfavorable prognosis, the specific subpopulations underpinning this correlation have yet to be clearly defined [19]. Preliminary scRNA-seq analyses in PCa have indicated the presence of heterogeneous CAF clusters [20,21]; nevertheless, their functional significance, clinical implications, and definitive biomarkers remain to be conclusively validated [22]. Notably, few investigations have systematically delineated CAF subsets that drive tumor angiogenesis or promote invasion and metastasis. Elucidating this functional diversity is critical not only for refining prognostic stratification but also for advancing the development of targeted therapies capable of selectively suppressing protumorigenic CAFs while preserving stromal components with potentially protective roles.

Among the proposed markers, the Thy-1 cell surface antigen (THY1 or CD90) has attracted growing attention. THY1, a glycosylphosphatidylinositol (GPI)-anchored protein, is expressed across diverse cell types, including fibroblasts [23,24], stem cells [25,26], and certain populations of cancer cells [27], as well as other cell types [28]. Its functional role is highly context-dependent, acting as either an oncogene or a tumor suppressor in different malignancies. Although it remains uncertain whether THY1 is expressed by PCa epithelial cells or what role it may play therein, mounting evidence indicates that CAFs in PCa can express THY1. For example, Kwon et al. [29] identified two fibroblast subsets in the murine prostate stroma using scRNA-seq: Sca1⁺THY1⁺ fibroblasts, which may support epithelial proliferation, and Sca1⁺THY1⁻ myofibroblast-like cells. Pascal et al. [30] observed that THY1-high stromal cells encircled tumor cells in human PCa. True et al. [23] reported a greater frequency of THY1⁺ stromal cells in PCa tissue than in normal tissue and proposed THY1 as a CAF marker, identifying a urinary THY1 variant that decreased postprostatectomy. Additionally, Hesterberg et al. [31] linked ASPN⁺THY1⁺ stromal infiltration with elevated Gleason scores, suggesting a protumorigenic role. On the basis of these observations, we hypothesize that THY1 delineates a protumorigenic CAF subset that drives PCa progression through specific mechanistic pathways.

This study seeks to provide a comprehensive dissection of CAF abundance and heterogeneity in PCa, with particular emphasis on the identification, validation, and functional characterization of the THY1⁺ CAF subpopulation. We first established the prognostic relevance of overall CAF infiltration across multiple patient cohorts. Building upon this, we developed and validated a robust CAF-derived gene signature predictive of biochemical recurrence. Leveraging scRNA-seq datasets, we identified THY1 as a defining marker of a distinct CAF subset. Through multiplex immunohistochemistry in a prospective clinical cohort, we confirmed the clinical significance of THY1⁺ CAFs, demonstrating their enrichment in aggressive pathological features and their association with adverse patient outcomes. Functional analyses using primary CAFs isolated from patient tumors further revealed that THY1⁺ CAFs possess markedly enhanced proangiogenic activity compared with their THY1⁻ counterparts. Mechanistically, we identified and validated a distinctive secretory program enriched in proangiogenic mediators, most notably CXCL6, and delineated the CXCL6/CXCR2 axis as a pivotal pathway driving this phenotype. Moreover, we demonstrated that THY1 itself directly regulates the proangiogenic capacity of CAFs, primarily through modulation of VEGFA secretion.

Collectively, our findings deliver a comprehensive multiomics and functional framework of CAF heterogeneity in PCa, positioning THY1⁺ CAFs as central drivers of tumor angiogenesis and disease progression. This work not only enhances mechanistic understanding of the PCa TME but also underscores novel stromal therapeutic targets, including the CXCL6/CXCR2 axis, for future clinical intervention.

Method

Datasets and tissue specimens

This investigation integrated data from multiple publicly available cohorts together with a prospectively collected clinical cohort to evaluate CAFs in PCa. A comprehensive overview of all datasets is presented in Supplemental Table S1.

RNA-sequencing data and corresponding clinical annotations for the TCGA-PRAD cohort were obtained from the Genomic Data Commons (GDC) portal. This cohort comprised 471 primary tumor samples, with complete Gleason score information for all patients, pathological T stage data for 464 patients, N stage data for 403 patients, and BCR information for 361 patients. This dataset served as the foundation for the initial assessment of CAF abundance, differential gene expression, and survival analyses.

For independent validation, the following Gene Expression Omnibus (GEO) datasets were incorporated: GSE116918 [32] (n = 248), MSKCC [33] (n = 112), and GSE54460 [34] (n = 105). Additional datasets, GSE138503 [35] and GSE147493 [36], were employed for signature derivation. Single-cell RNA-seq dataset GSE181294 [37] were derived from GEO platform, from which fourteen prostate tumor samples with documented Gleason scores were retained for downstream analyses. Tumor malignancy was further summarized using the 2014 International Society of Urological Pathology (ISUP) Grade Group (GG) classification, which stratifies prostate cancer into five grade groups based on Gleason scoring patterns. Specifically, Grade Group 2 (GG2) corresponds to Gleason score 3 + 4, characterized by a predominant pattern 3, whereas Grade Group 3 (GG3) corresponds to Gleason score 4 + 3, characterized by a predominant pattern 4. Although both categories share a total Gleason score of 7, they are considered biologically and clinically distinct, with GG3 generally associated with more aggressive disease behavior. This classification was used to ensure a more accurate representation of tumor malignancy in subsequent analyses. We previously established two prostate adenocarcinoma (PRAD) single-cell RNA-seq datasets: the PRAD single-cell RNA-seq dataset1 [21] and dataset2 [20], also employed for analysis. External validation cohort (GSE176031): An additional independent scRNA-seq dataset comprising 21,743 cells was obtained from GEO and utilized for the external validation of the eight-gene signature expression patterns [38]. Expression data were normalized according to platform-specific protocols; for microarray data, the robust multiarray average (RMA) algorithm was applied, with multiple probes representing the same gene averaged.

Between 2015 and 2020, a prospective cohort of 84 patients with prostate cancer was recruited at the Fudan University Shanghai Cancer Center (FUSCC). A tissue microarray for multiplex immunofluorescence analysis was constructed using FFPE specimens, with the detailed clinical information in Supplemental Table S2. In addition, fresh tumor tissues from two supplementary patients were collected for primary CAF isolation and subsequent functional assays. The study was approved by the FUSCC Institutional Review Board, and written informed consent was obtained from all participants.

Single-cell RNA sequencing analysis

For the incorporated single-cell datasets, preprocessing was performed using the R package Seurat [39]. DoubletFinder was applied to identify and remove potential doublets or multiplets [40]. Low-quality cells were filtered according to the following predefined quality control (QC) criteria: cells with fewer than 200 or more than 6000 detected genes (nFeature_RNA), total UMI counts (nCount_RNA) lower than 1000 or higher than 50,000, or a mitochondrial transcript proportion (percent.mt) of 20 % or greater were excluded. Applying these QC filters sequentially removed 9.8 % of cells based on nFeature_RNA, an additional 5.1 % based on nCount_RNA, and a further 4.3 % based on percent.mt, resulting in a total of 56,924 high-quality cells retained for downstream analyses. Subsequently, CellCycleScoring() was employed to assess cell cycle status, and Harmony was used for batch effect correction, followed by the selection of the top 2000 highly variable genes. Principal Component Analysis (PCA) was performed based on these 2000 highly variable genes, and UMAP embeddings were constructed based on the distances of the top 15 principal components. The annotation of clustering results was referenced from published literature [41]. The markers of immunocytes provided in Table S3. To quantitatively evaluate the enrichment of specific cellular features across different groups, we applied the Ro/e (Observed-to-Expected Ratio) index [42]. The Ro/e index is used to determine whether a given feature exhibits enrichment beyond random expectation within a particular group. When Ro/e index > 1, it indicates that the feature is overrepresented in that group, meaning its occurrence is higher than expected by chance; when Ro/e index< 1, it indicates that the occurrence is lower than expected, suggesting that the feature may be diminished or suppressed in that group.

Spatial transcriptome RNA sequencing analysis

The 10x Visium spatial transcriptomics data were downloaded from GEO, including GSE230282 (castration resistant prostate cancer) [43] and GSE278936 (primary prostate cancer) [44]. Quality control and preprocessing involved calculating the proportion of mitochondrial genes, removing mitochondrial and ribosomal-associated genes at the expression matrix level, filtering out low-abundance genes with fewer than 10 spot expressions, and normalizing the data using SCTransform (assay = Spatial). Subsequently, dimensionality reduction and clustering were performed, with spatial clustering visualized using SpatialDimPlot. Using an annotated scRNA-seq object as a reference, anchors were established with FindTransferAnchors, and cell type probabilities were transferred to the spatial object as a predictions assay through TransferData. Moran’s I was utilized on this assay to identify spatially variable cell type scores, which were visualized separately using SpatialPlot.

Estimation of CAF infiltration via computational deconvolution

The EPIC algorithm, implemented in the immunedeconv R package (v2.1.0), was employed to estimate CAF infiltration levels from bulk RNA-seq data. The EPIC method [45] was applied to deconvolute transcriptomic profiles and infer the relative proportions of CAFs among other cellular constituents within the TME. The derived CAF abundance values were subsequently incorporated into correlation and survival analyses alongside clinical variables. Data visualization was performed using GraphPad Prism, version 8.4.3.

Creation of a CAFs associated signature for prognostic evaluation

An individualized prognostic signature based on CAF-associated genes was developed to predict BCR in patients with PCa. Utilizing three independent cohorts (TCGA-PRAD, GSE138503, and GSE147493), we filtered the genes consistently exhibiting a positive correlation with CAF abundance (R > 0.6, FDR < 0.05). Univariate Cox regression (p < 0.05) were further applied to identify these genes that link to biochemical recurrence. To enhance model robustness and mitigate overfitting, the gene set was further refined through LASSO Cox regression with 10-fold cross-validation, implemented using the glmnet R package [46]. The model integrated gene expression levels with corresponding coefficients to generate the risk score, defined as:

$$\text{riskScore}=\sum _{i=1}^{\mathrm{n}}{\mathrm{X}}_{\mathrm{i}}\times {\mathrm{B}}_{\mathrm{i}}$$

Where ${\mathrm{X}}_{\mathrm{i}}$ represents the coefficient assigned to each gene, and ${\mathrm{B}}_{\mathrm{i}}$ denotes the corresponding gene expression value.

Patients were stratified into high- and low-risk groups based on the median risk score. The prognostic relevance and clinical utility of this signature were further validated across three independent external cohorts, MSKCC, GSE54460, and the prospective FUSCC cohort, thereby underscoring its robustness and broad applicability.

THY1⁺ vs. THY1⁻ CAF differential gene expression and enrichment analysis

We performed single-cell RNA-seq analysis on datasets 1 and 2 to identify genes differentially expressed between THY1-high and THY1-low cancer-associated fibroblast (CAF) subpopulations. Dataset 1 analysis: CAFs were subclustered (resolution = 0.2) into 6 populations via Seurat and manual annotation based on marker genes: cluster 1 (CXCL12⁻NOTCH3⁺), cluster 2 (CXCL12⁺NOTCH3⁺), cluster 3 (CXCL12⁻/NOTCH3⁻), and cluster 4 (FAP⁺/NOTCH3⁺). Clusters 1 and 4 (high THY1) were defined as THY1⁺ CAFs; Clusters 2 and 3 were defined as THY1⁻ CAFs. Differential expression analysis (edgeR, |log₂FC| > 0.322, FDR < 0.05) identified 322 upregulated genes in THY1⁺ CAFs. Dataset 2 analysis: Clustering and annotation followed the original methods. Clusters 1–4 (high THY1) were defined as THY1⁺ CAFs; cluster 5 was defined as THY1⁻. Using the same criteria, 435 genes were upregulated in THY1⁺ CAFs.

Isolation and culture of human primary prostate cancer CAFs

Primary CAFs were isolated from human PCa tissues obtained from patients undergoing radical prostatectomy. Fresh tumor specimens were collected under sterile conditions within one hour post-resection, immediately transferred into tissue preservation solution, and maintained on ice during transport. All samples were processed within four hours of collection to preserve optimal cell viability.

Tissue Processing and Digestion: The collected tissues were thoroughly rinsed with PBS to remove residual blood. Representative portions were fixed and subjected to hematoxylin and eosin (HE) staining for histopathological evaluation of tumor content and Gleason scoring. The remaining tissues were minced into ∼1 mm³ fragments using sterile surgical scissors in Solution A (PBS supplemented with 1 % penicillin/streptomycin and 1 μg/mL amphotericin B). Tissue fragments were then digested in Solution B (DMEM/F12K medium supplemented with 2 % BSA, 1 mg/mL collagenase I, and 0.1 mg/mL hyaluronidase) at 37 °C for 18 h under gentle agitation.

Following digestion, the suspension was centrifuged at 600 rpm for 3 min to pellet undigested tissue fragments. The supernatant was collected and centrifuged again at 800 rpm for 10 min to obtain a fibroblast-enriched pellet. This pellet was resuspended in Solution C (DMEM/F12K supplemented with 10 % FBS, 1 % penicillin/streptomycin, and 50 ng/mL human insulin) and seeded into 6-well plates. Cultures were maintained at 37 °C with 5 % CO₂ for 48–72 h to facilitate fibroblast adhesion and proliferation. The medium was refreshed every 48 h to remove nonadherent cells and debris.

Early adherent fibroblasts exhibited the characteristic spindle-shaped morphology and were successfully expanded for downstream experiments. All primary CAFs utilized in this study were maintained between passages 3 and 6 to ensure phenotypic stability. A summary of the principal reagents and materials is provided in Supplemental Table S4.

Purification of THY1⁺ and THY1⁻ CAFs

Primary CAFs were isolated and subsequently separated into THY1⁺ and THY1⁻ subpopulations through a combined approach utilizing magnetic-activated cell sorting (MACS) and fluorescence-activated cell sorting (FACS).

MACS enrichment: Initial enrichment was performed using human THY1 MicroBeads (Miltenyi Biotec, Cat. No. 130–096–253) with LS columns. Briefly, 5 × 10⁷ cells were incubated with MicroBeads in MACS buffer for 20 min at 4 °C and then applied to a pre-wetted LS column positioned within a magnetic field. THY1⁻ cells were collected from the flow-through, whereas THY1⁺ cells were retained on the column and subsequently eluted upon removal from the magnet.

FACS purification: To achieve higher purity, enriched fractions were subjected to FACS. Following Fc receptor blocking, cells were incubated with an anti-THY1 antibody (BioLegend, #328,124) and sorted using a flow cytometer. The resulting THY1⁺ and THY1⁻ CAFs were cultured for 72 h, and their purity, routinely exceeding 99 %, was confirmed by flow cytometry.

Construction of THY1-knockdown CAFs

Stable THY1-knockdown CAFs were generated through lentiviral shRNA transduction. Lentiviral particles were produced by cotransfecting 293FT cells with either THY1-targeting shRNA constructs (shTHY1#1 and shTHY1#2) or a non-targeting control shRNA (shNC), along with the packaging plasmids pMD2.G and psPAX2, using Neofect™ reagent. Viral supernatants were harvested 72 h post-transfection, concentrated, and employed to transduce primary THY1⁺ CAFs at a multiplicity of infection (MOI) of 10 in the presence of 8 μg/mL polybrene.

Following 7 days of selection with 2 μg/mL puromycin, stable polyclonal populations were established, yielding shNC, shTHY1#1, and shTHY1#2 CAFs. Knockdown efficiency was validated through quantitative PCR and Western blotting. The sequences for shNC, shTHY1#1, and shTHY1#2 are listed in Supplemental Table S5.

Quantitative real-time PCR (qRT‒PCR)

Total RNA was extracted using TRIzol reagent (Invitrogen), and complementary DNA (cDNA) was synthesized with the PrimeScript RT reagent kit (Takara). Quantitative PCR was conducted using SYBR Green Master Mix (Takara), with GAPDH serving as the internal control. Relative gene expression levels were calculated using the 2^–ΔΔCt method. All reactions were performed in triplicate. Primers were designed via Primer-BLAST and synthesized by Tsingke Biotechnology (Beijing, China). The primer sequences utilized in this study are provided in Supplemental Table S5.

Western blot

Western blotting was performed via standard procedures: 1: Extracting and quantifying proteins; 2: Separating them by size via electrophoresis; 3: Transferring them to a membrane; 4: Using specific antibodies and chemiluminescent detection to identify and analyze a target protein. The antibodies utilized in this study are provided in Supplemental Table S4.

Enzyme-Linked immunosorbent assay (ELISA)

Secretions of VEGFA and CXCL6 in CAFs were quantified via commercial ELISA kits following the manufacturer’s instructions. CAFs were plated at a density of 5 × 10⁵ cells per well in 6-well plates. The supernatant was harvested after 24 h. The next steps involved centrifugation at 1500 × g for 15 min and storage of the sample at −20 °C until analysis. All standards and samples were run in quadruplicate.

The assay procedure was as follows: 100 μL of standards, blanks, or samples were added per well and the reaction was subjected to incubation at 25 °C for 2 h. Following four washes with 1 × wash buffer, 50 μL of 1 × detection antibody was introduced, and the mixture was incubated for one hour protected from light. After four washes with 1 × wash buffer, 50 μL of 1 × detection antibody was added and incubated in the dark for 1 hour. Following four additional washes, 100 μL of 1 × HRP-streptavidin was added and incubated for 30 min. The plate underwent four additional washes before adding 100 μL of TMB substrate, followed by a 15-minute incubation in the dark. The reaction was halted using 100 μL of stop solution, and the absorbance was promptly recorded at 450 nm.

HUVEC tube formation assay

For the assessment of proangiogenic functions across heterogeneous CAF subsets, tube formation assays were performed with HUVECs. Conditioned media were collected from THY1⁺ and THY1⁻ CAFs after 36 h of culture, subjected to 0.45 μm membrane filtration, and either used immediately or stored at −20 °C.

Fifty microliters of thawed Matrigel were dispensed into each well of a precooled 96-well plate using prechilled pipette tips. The plate was subjected to a 30-minute incubation at 37 °C to facilitate polymerization. HUVECs were resuspended in serum-free medium and seeded onto the gel at 30,000 cells per well (10 μL). Then, 80 μL of CM from THY1⁺ or THY1⁻ CAFs was introduced into each well. After incubation was performed at 37 °C for 6 h, tube formation was visualized and imaged under a microscope. Quantification was carried out via ImageJ software.

Protein microarray analysis of cell culture supernatants via antibody arrays

The cell culture supernatants were analyzed via a Raybiotech AAH-BLG-1 protein microarray. After centrifugation, the samples underwent dialysis and were labeled with biotin as per the manufacturer's guidelines. Following array blocking, the labeled samples were diluted and hybridized overnight at 4 °C. Subsequent to washing, the arrays were incubated with Cy3-streptavidin, air-dried, and scanned. Signal quantification was carried out via GenePix Pro 6.0. Technical support for this experiment was provided by Shanghai Wayen Biotechnologies Co., Ltd.

Multiplex fluorescence immunohistochemistry (mFIHC) of the tissue microarray

Multiplex fluorescence immunohistochemistry (mFIHC) was conducted on a prostate cancer tissue microarray (TMA) comprising 225 cores from 84 patient samples via a six-color tyramide signal amplification (TSA) kit (TSA-RM-24,259, Panovue, Cat. No. 10,003,100,100). Antibodies used included anti-THY1 (diluted 1:500), anti-α-SMA (1:1000), and anti-COL1A1 (1:500). The fluorophores assigned were as follows: THY1 is visualized in yellow, α-SMA in red, and COL1A1 in green.

Following deparaffinization and rehydration, microwave-heat induced epitope retrieval (HIER) was conducted in citrate buffer. Peroxidase quenching was achieved using a 3 % H₂O₂ methanol solution. The sections were initially blocked using 5 % goat serum for 40 min at room temperature and were subsequently incubated with primary antibodies for 2 h at 37 °C. Subsequent to washing, incubation with HRP-conjugated secondary antibodies was performed on the membranes for 30 min at room temperature. Tyramide signal amplification (TSA) was conducted using fluorophore-conjugated tyramides in accordance with the manufacturer's guidelines. Nuclei were counterstained with DAPI. Between successive rounds of staining, microwave treatment was used to remove the antibodies prior to subsequent labeling.

Finally, the slides were coverslipped with antifade mounting medium and sealed with nail polish. Images for multiplex fluorescence analysis were captured with a microscope and quantitatively evaluated with HALO® image analysis software. The proportions of ACTA2⁺ and THY1⁺ CAFs within each tumor core were quantified.

Statistical analysis

Statistical analyses were performed in R 4.4.3 or GraphPad. For group comparisons, the Wilcoxon rank-sum test (continuous) or Chi-Square test/Fisher’s exact test (categorical) was used. Survival analysis included Kaplan–Meier and log-rank tests. For comparisons between subgroups defined by continuous variables, the student’s t-test was applied. Prognostic factors were assessed with univariate/multivariate Cox regression. Time‒dependent ROC curves were generated with the timeROC package. Functional enrichment analysis was conducted via Metascape [46] (https://metascape.org), adjusted p < 0.05 was used. The differential gene expression (edgeR) thresholds were |log₂FC| > 0.6 and FDR < 0.05. All tests were two-sided; p < 0.05 was considered significant.

Results

Single-cell and spatial transcriptomics reveal the cellular heterogeneity and fibroblast functional remodeling associated with the Gleason grading of PCa

Distribution of cells remains consistent across different Gleason grades (GG1+2 vs GG3+4 + 5) (Fig. 1A). Similarly, the cell cycle states (G1, S, G2M) are evenly distributed among the subpopulations (Fig. 1B), without significant shifts, further validating the robustness of the clustering structure and suggesting that the cell cycle is not a primary driving factor. Based on classical marker genes, we defined ten major cell lineages including epithelial cells, endothelial cells, fibroblasts, T/NK/B cells, myeloid immune cells, and mast cells (Fig. 1C-D). Ro/e index revealed a significant enrichment trend for cells such as Monocyte/Macrophage and Fibroblast in GS3+4 + 5 (Fig. 1E), indicating enhanced immune infiltration and matrix remodeling in high-grade tissues, while B cells and Endothelial cells are more prominent in low-grade tissues, reflecting the grade-related heterogeneity of the tumor microenvironment. Upon extracting fibroblast cells and comparing GS3+4 + 5 with GS1+2, we identified a set of differentially expressed genes (Fig. 1F), with higher expression of genes such as CCL4, CYP1B1, RPS2, THY1, ITM2B, and PAM in the GS3+4 + 5 group associated with ECM remodeling. GO/KEGG enrichment analysis further indicates that the functional characteristics of fibroblasts in the low Gleason group are biased towards pathways related to immune regulation, antigen processing and presentation, cell adhesion, and stress response (Fig. 1G). This suggests that fibroblasts may play a more active role in immune evasion and ECM remodeling in high-grade tumors. Spatial transcriptomics analysis has revealed that fibroblasts aggregate at the tumor margins and stroma, encapsulating tumor cells, and this phenomenon is more pronounced in castration-resistant prostate cancer (CRPC), thereby inhibiting immune infiltration and creating a localized cold environment (Fig. 1H).

Fig. 1.

Single-cell and spatial transcriptome multi-omics analysis. (A) UMAP plots of different Gleason grades (GG1+2 vs GG3+4 + 5). (B) UMAP plots of various cell populations under different cell cycle states (G1, S, G2M). (C) DotPlot showing the expression profiles of classical marker genes used for annotation. (D) Expression of representative marker genes in UMAP space. (E) Ro/e index. (F) Analysis of expression differences in fibroblasts with high and low Gleason scores, with the top 20 genes displayed. (G). GO and KEGG enrichment analyses further demonstrate. (H) Spatial transcriptomics results showing the spatial distribution of cell types in two Visium slices (GSM8557993 and GSE230282).

Elevated CAF abundance correlate with aggressive stages and unfavorable clinical outcomes

We first evaluated the associations between CAF infiltration and diverse clinicopathological characteristics, building upon existing evidence implicating CAFs in the malignant progression of solid tumors [[47], [48], [49], [50], [51]]. Utilizing EPIC deconvolution in the TCGA-PRAD cohort (n = 471), we observed significantly elevated CAF abundance in tumors with GS > 7 compared with those with GS ≤ 7, in T3–T4 versus T1–T2 stages, and in node-positive (N1) versus node-negative (N0) disease (all p < 0.05, Fig. 2A). Stratification by the median EPIC score further demonstrated a marked enrichment of high-CAF cases among patients exhibiting advanced features, including GS > 7, T3–T4 stage, and N1 status (all p < 0.05, Fig. 2B).

Fig. 2.

Cancer-associated fibroblast (CAF) abundance and its clinical significance in prostate cancer. (A) CAF levels across the Gleason score (GS), T stage, and N stage in the TCGA-PRAD cohort (n = 471), as estimated by EPIC deconvolution. Statistical significance was determined by Student's t-test. (B) Proportion of samples classified into high-CAF (red) and low-CAF (blue) groups, stratified by the median EPIC score, across clinical parameters in the TCGA-PRAD cohort. p values were calculated via the chi-square test. (C) CAF abundance compared by GS, T stage, and M stage in the GSE116918 cohort (n = 248). Student's t-test was used for statistical comparisons. (D) Distribution of high- and low-CAF groups across clinical features in the GSE116918 cohort, as analyzed by the chi-square test. (E) Plot displaying differentially expressed genes (DEGs) between the high- and low-CAF groups in the TCGA-PRAD cohort. (F) Heatmap of representative DEGs with clinical annotations, including CAF status, biochemical recurrence (BCR), and GS. (G) Functional enrichment analysis of upregulated genes in high-CAF samples performed via Metascape. (H, J) Kaplan‒Meier curves of recurrence-free survival for patients with high vs. low CAF abundance in the TCGA (H) and GSE116918 (J) cohorts. Hazard ratios (HRs) and p values were calculated via the log-rank test. (I, K) Time‒dependent receiver operating characteristic (ROC) curves evaluating the predictive performance of CAF abundance for BCR in the TCGA (I) and GSE116918 (K) cohorts.

These results were corroborated in an independent dataset, the GSE116918 cohort (n = 248), which included information on metastatic status. In line with the TCGA findings, CAF levels were significantly elevated in samples with GS > 7, T3–T4 stage, and metastatic (M1) disease compared with their lower-risk counterparts (all p < 0.05, Fig. 2C). Moreover, the high-CAF group was consistently enriched in advanced disease categories (all p < 0.05, Fig. 2D).

To elucidate the molecular features associated with tumors exhibiting high CAF content, we performed differential gene expression analysis within the TCGA cohort. This analysis identified 539 genes significantly upregulated in high-CAF–infiltrated samples (Fig. 2E–F). Functional enrichment analysis using Metascape demonstrated that these genes were predominantly enriched in biological processes related to blood vessel development and cell migration (Fig. 2G, Table S6), suggesting a potential role for CAFs in promoting angiogenesis and invasive tumor behavior.

Elevated CAF abundance was also a strong predictor of poor recurrence-free survival in both the TCGA (HR = 2.74, p < 0.001; Fig. 2H) and GSE116918 (HR = 2.83, p < 0.001; Fig. 2J) cohorts. Time-dependent ROC curves further confirmed the robust predictive performance of CAF levels for BCR at 1, 3, and 5 years (Fig. 2I, K).

Taken together, these findings establish CAF abundance as a clinically relevant biomarker associated with aggressive pathological features and adverse prognosis in patients with prostate cancer.

Construction and validation of a CAF-associated prognostic signature

Given the prognostic relevance of CAF abundance, we aimed to develop a robust gene signature reflective of CAF activity. Genes positively correlated with CAF abundance (R > 0.6, FDR < 0.05) were identified across three independent cohorts: TCGA-PRAD (116 genes), GSE138503 (1140 genes), and GSE147493 (225 genes). A total of 54 genes overlapped among all three datasets (Fig. 3A). Functional enrichment analysis revealed a strong association of these genes with key biological processes, including blood vessel development and extracellular matrix organization (Fig. 3B). Univariate Cox regression analysis identified 19 genes significantly associated with recurrence-free survival (all p < 0.05, Fig. 3C). An 8-gene prognostic signature was refined using LASSO regression (Fig. 3D, E) and integrated into a risk calculate formula.

Fig. 3.

Construction and validation of a CAF-related prognostic signature in prostate cancer. (A)Upset and venn diagram illustrating the overlap of genes positively correlated with CAF abundance across the TCGA-PRAD, GSE138503, and GSE147493 cohorts. (B) Functional enrichment analysis by Metascape highlighting key biological processes. (C) Forest plot showing the results of univariate Cox regression analysis of the 54 genes associated with recurrence-free survival (p < 0.05). (D) 10-fold cross-validation curve for tuning parameter (λ) selection in the LASSO regression. (E) LASSO coefficient profiles of the 8 genes selected for the prognostic signature. (F) Kaplan‒Meier curves comparing recurrence-free survival between the high- and low-risk groups stratified by the median risk score in the TCGA cohort. (G) Time‒dependent ROC curves evaluating the predictive accuracy of the risk score for BCR. (H) Univariate and multivariate Cox regression analyses evaluating the risk score and clinical parameters (age, Gleason score, T stage, N stage) as predictors of BCR. (I) Distribution of risk scores across different Gleason scores and T stages in the TCGA-PRAD cohort (n = 471). p values were derived from Student’s t-test. (J) Proportion of high-risk (red) and low-risk (blue) patients stratified by clinical parameters. p values were calculated via the chi-square test.

In the TCGA cohort, patients were stratified into high- and low-risk groups based on the median risk score. High-risk patients exhibited significantly poorer recurrence-free survival (HR = 6.31, p < 0.001; Fig. 3F). The risk score demonstrated excellent predictive accuracy for recurrence at 1, 3, and 5 years (AUC = 0.820, 0.810, and 0.800, respectively; Fig. 3G). Crucially, both univariate and multivariate analyses confirmed that the risk score was an independent predictor of BCR after adjusting for Gleason score and T stage (Fig. 3H). Furthermore, risk scores were significantly elevated in patients with higher Gleason scores (GS > 7) and advanced T stages (T3–T4) (all p < 0.01, Fig. 3I). Notably, the risk score was markedly higher in the Gleason grade group 4 (GS 4 + 3) compared with the group defined by GS 3 + 4 (p < 0.01, Fig. 3I). Consistently, high-risk patients were enriched in these more aggressive clinicopathological subsets, with a significantly greater proportion of high-risk cases in GS 4 + 3 relative to GS 3 + 4 (all p < 0.05, Fig. 3J).

Collectively, this CAF-derived gene signature represents a powerful and independent prognostic biomarker for prostate cancer.

Validation of the CAF-Derived prognostic signature across independent cohorts

The eight-gene CAF-derived prognostic signature was further validated in three independent cohorts. In the GSE54460 cohort (n = 100), patients in the high-risk group (n = 50) exhibited significantly poorer recurrence-free survival (HR = 3.430, p < 0.001; Fig. 4A). The signature demonstrated robust predictive accuracy for recurrence at 1, 3, and 5 years, with AUC values of 0.665, 0.705, and 0.728, respectively (Fig. 4B). Moreover, risk scores were significantly higher in tumors with GS > 7 compared with those with GS ≤ 7 (p = 0.008) and in tumors with GS 4 + 3 compared with GS 3 + 4 (p = 0.008, Fig. 4C).

Fig. 4.

External validation of the CAF-related prognostic signature in three independent cohorts. (A–C) Validation in the GSE54460 cohort: (A) Kaplan–Meier analysis of recurrence-free survival between risk groups; (B) time-dependent ROC curves predicting BCR at 1, 3, and 5 years; (C) risk score distribution in GS ≤ 7 vs. GS > 7 and in GS 3 + 4 vs. 4 + 3. (D-F) Validation in the MSKCC cohort: (D) Kaplan–Meier survival analysis; (E) ROC curves for BCR prediction at 1, 3, and 5 years; (F) risk score by GS ≤ 7 vs. > 7 and by T1+T2 vs. T3+T4. (G-I) Validation in the FUSCC cohort: (G) Kaplan–Meier analysis; (H) ROC analysis for 1-, 3-, and 5-year BCR prediction; I) risk score comparison between GS ≤ 7 and > 7 and between T1+T2 vs. T3+T4.

Consistent results were obtained in the MSKCC cohort (n = 140). High-risk patients (n = 70) demonstrated a substantially elevated risk of BCR (HR = 2.140, p = 0.032; Fig. 4D). ROC analysis yielded AUC values of 0.687, 0.684, and 0.655 for predicting recurrence at 1, 3, and 5 years, respectively (Fig. 4E). Once again, risk scores were significantly higher in patients with GS > 7 and those at T3–T4 stage (Fig. 4F).

Finally, validation in the prospective FUSCC cohort (n = 84) confirmed the clinical utility of the signature. High-risk patients (n = 42) displayed significantly reduced recurrence-free survival (HR = 2.790, p = 0.002; Fig. 4G). The signature achieved AUC values of 0.930, 0.801, and 0.817 for predicting recurrence at 1, 3, and 5 years, respectively (Fig. 4H). Furthermore, risk scores were markedly elevated in tumors with GS > 7 and in those with T3–T4 stage compared with T1–T2 stage (Fig. 4I).

Immunohistochemical analyses validated the prognostic value of THY1⁺ CAFs

Although the 8-gene signature was derived from CAF-correlated transcripts, we sought to determine whether these genes were preferentially expressed by CAFs. Analysis of a prostate cancer single-cell RNA-seq dataset (n = 21,743 cells) revealed that among the eight signature genes, COL1A2, COL1A1, COL3A1, SULF1, COL5A1, and THY1 exhibited the highest mean expression in CAFs compared with eight other major cell types (Fig. 5A, B). Notably, COL1A1 expression was minimal in non-CAF populations, whereas THY1, a surface marker, was also highly enriched in CAFs. In contrast, RAB3IL1 expression in CAFs was lower than that observed in Mono/Macro cells, and PLXND1 was expressed at comparatively reduced levels in CAFs relative to other cell populations. On the basis of these findings, we selected COL1A1 and THY1 for further investigation.

Fig. 5.

Single-cell and immunohistochemical validation of CAF subpopulations. (A) t-SNE visualization of major cell types in the GSE176031 single-cell RNA-seq dataset (n = 21,743 cells). (B) Heatmap showing the mean expression levels of the 8 signature genes across 9 major cell types. (C) UMAP projection of 948 CAFs from our single-cell RNA-seq dataset2 (GSE141445), colored according to the expression levels of ACTA2, VIM, COL1A1, and THY1. (D–G) Multiplex immunofluorescence staining of a prostate cancer TMA (84 tumor samples; four markers: DAPI [blue], COL1A1 [green], THY1 [yellow], α-SMA [red]): (D) Representative image from a BCR-positive, GS 4 + 3 sample with 46.9 % THY1⁺ CAFs (scale bar: 300 μm); (E) Representative image from a BCR-negative, GS 4 + 3 sample with 7.7 % THY1⁺ CAFs (scale bar: 300 μm); (F) enlarged views from 5 BCR-positive cases (scale bar: 100 μm); (G) enlarged views from 5 BCR-negative cases (scale bar: 100 μm). (H) Violin plot comparing the percentage of THY1⁺ CAFs between the GS >7 and GS ≤7 groups. (I) Kaplan–Meier analysis of recurrence-free survival between the high- and low-THY1⁺ CAF groups (log-rank test). (J) ROC curves evaluating the predictive accuracy of the THY1⁺ CAF percentage for BCR at 1, 3, and 10 years. (K) Schematic diagram illustrating two CAF subtypes: THY1⁻αSMA⁺ (left) and THY1⁺αSMA⁺ (right).

Analysis of our single-cell RNA-seq dataset2 (948 CAFs) confirmed pronounced heterogeneity within the CAF compartment: while canonical markers ACTA2 and VIM were broadly expressed, COL1A1 and THY1 expression was confined to specific CAF subsets (Fig. 5C). To assess the clinical relevance of these subpopulations, we performed multiplex immunofluorescence (TSA) on a prostate cancer tissue microarray (n = 84 samples), staining for ACTA2 (red), THY1 (yellow), COL1A1 (green), and DAPI (blue). Representative images demonstrated that a BCR-positive, GS 4 + 3 tumor exhibited elevated THY1 expression and a high proportion of THY1⁺ CAFs (46.9 %; Fig. 5D), whereas a BCR-negative, GS 4 + 3 tumor displayed low THY1 expression and a correspondingly low proportion of THY1⁺ CAFs (7.7 %; Fig. 5E). Consistent trends were observed across five additional BCR-positive and five BCR-negative cases (Fig. 5F, G).

Quantitative analysis revealed that the proportion of THY1⁺ CAFs was significantly higher in GS > 7 tumors compared with GS ≤ 7 tumors (Fig. 5H). Patients were stratified into high (n = 42) and low (n = 42) THY1⁺ CAF groups based on the median proportion. Elevated THY1⁺ CAF abundance was associated with significantly worse recurrence-free survival (HR = 2.63, p = 0.0032; Fig. 5I) and demonstrated good predictive accuracy for BCR at 1, 3, and 10 years (AUC = 0.793, 0.713, and 0.705, respectively; Fig. 5J). Collectively, these findings underscore the functional heterogeneity of CAFs, with the THY1⁺ subpopulation conferring a more aggressive phenotype and adverse clinical outcomes (Fig. 5K).

THY1⁺ CAFs represent a distinct proangiogenic subpopulation

To explore the functional heterogeneity of CAFs, we hypothesized that THY1⁺ and THY1⁻ CAFs may exhibit divergent protumorigenic properties. We first analyzed two independent prostate cancer single-cell RNA-sequencing (scRNA-seq) datasets to delineate CAF subpopulations based on THY1 expression. In our RNA-seq dataset1 (3193 CAFs), six CAF clusters were identified. THY1 expression was significantly enriched in clusters 1–3 compared with clusters 4–6 (p < 0.0001; Fig. 6A). Likewise, in our single-cell RNA-seq dataset2 (951 CAFs), five CAF clusters were defined, with THY1 expression markedly elevated in clusters 1 and 4, but nearly absent in clusters 2 and 3 (p < 0.001; Fig. 6B). These consistent findings across independent datasets establish THY1 expression as a defining marker of distinct CAF subpopulations.

Fig. 6.

Transcriptomic and functional analysis reveals THY1⁺ CAFs as a proangiogenic subpopulation in prostate cancer. (A) Analysis of our RNA-seq dataset1 (3193 CAFs). Left: UMAP projection colored according to the six identified CAF clusters. Middle: UMAP visualization of THY1 expression. Right: Violin plot showing the distribution of THY1 expression levels across all six clusters. Statistical significance was determined by one-way ANOVA with Tukey's post hoc test (p < 0.0001 for clusters 1–3 vs. clusters 4–6). (B) Analysis of our single-cell RNA-seq dataset2 (951 CAFs). Left: t-SNE projection colored according to the five identified CAF clusters. Middle: t-SNE visualization of THY1 expression. Right: Violin plot of THY1 expression across clusters. Statistical significance was determined by one-way ANOVA with Tukey's post hoc test (*: p < 0.001 for clusters 1 and 4 vs. clusters 2 and 3). (C) Volcano plot of differentially expressed genes (DEGs) between THY1-high (clusters 1–3) and THY1-low (clusters 4–6) CAFs in our RNA-seq dataset1. (D) Volcano plot of DEGs between THY1-high (clusters 1–4) and THY1-low (cluster 5) CAFs in our single-cell RNA-seq dataset2. (E) Venn diagram identifying 121 common genes that were upregulated in THY1-high clusters in both datasets. (F) Functional enrichment analysis (GO biological process) of the 121 common upregulated genes was performed via Metascape. The bar chart shows the top significantly enriched terms ranked by -log10 (p value). (G) Validation of primary CAF isolation. Top: Representative immunofluorescence images of sorted THY1⁺ and THY1⁻ CAFs. Middle: Flow cytometry analysis confirming the purity (> 99 %) of the sorted populations. Bottom: Western blot analysis confirming THY1 protein expression in the sorted populations (scale bar: 200 μm). (H) Representative images of the HUVEC tube formation assay after treatment with serum-free medium (control), CM from THY1⁺ CAFs, or CM from THY1⁻ CAFs. The right panels show magnified views (scale bar: 200 μm). (I, J) Quantification of total tube length (I) and the number of branch points (J) from four independent experiments. The data are presented as the means ± SEMs. ***: p < 0.001; ns, not significant (Student’s t-test).

We next identified differentially expressed genes (DEGs) to explore functional distinctions between THY1-high and THY1-low clusters. In RNA-seq dataset1, 322 genes were significantly upregulated in the THY1-high clusters (clusters 1–3; Fig. 6C). In single-cell RNA-seq dataset2, 435 genes were likewise upregulated in the THY1-high clusters (clusters 1 and 4; Fig. 6D). Intersection of these gene sets yielded 121 commonly upregulated genes (Fig. 6E). Functional enrichment analysis of these shared genes, conducted via Metascape, demonstrated significant enrichment in biological processes including vasculature development and artery morphogenesis (Fig. 6F), suggesting a specialized role for THY1⁺ CAFs in promoting angiogenesis.

To functionally validate this prediction, we isolated primary THY1⁺ and THY1⁻ CAF1 cells from a clinical prostate cancer specimen (GS = 4 + 5, NX/M1a) using sequential MACS and FACS, achieving a purity exceeding 99 % (Fig. 6G). We then assessed the angiogenic capacity of these subsets via a HUVEC tube formation assay. Conditioned medium (CM) derived from THY1⁺ CAFs significantly enhanced HUVEC tube formation compared with CM from THY1⁻ CAFs, as evidenced by increased total tube length and branch point number (p < 0.001; Fig. 6H–J).

Collectively, these results demonstrate that THY1⁺ CAFs possess markedly elevated proangiogenic activity, a functional phenotype concordant with their distinct transcriptional profile.

THY1⁺ CAFs promote angiogenesis through increased secretion of CXCL6

To elucidate the mechanism by which THY1⁺ CAFs enhance HUVEC tube formation, we hypothesized that they secrete specific proangiogenic factors. Conditioned media were collected from two pairs of primary CAFs (THY1⁺ vs. THY1⁻, derived from two independent donors) and analyzed using an antibody array profiling 507 human secretory proteins (Fig. 7A). Multiple proteins were consistently upregulated in THY1⁺ CAF-conditioned media across both donors, with CXCL6, VEGFA, MMP2, and MMP19 displaying the most prominent increases (Fig. 7B, C).

Fig. 7.

Identification and functional validation of secretory proteins from CAFs. (A) Schematic of the antibody arrays used to detect 507 secretory proteins in CAF-conditioned medium. (B) Antibody array results highlighting selected proteins: CXCL6 (yellow), VEGFA (green), MMP2 (red), and MMP19 (blue). (C) Quantification of fluorescence signals for selected proteins from the antibody arrays. (D, E) ELISA validation of CXCL6 (D) and VEGFA (E) secretion levels in conditioned media from THY1⁺ and THY1⁻ CAFs. (F–K) Effects of CXCL6 on HUVEC tube formation: (F) Representative images of tube formation after treatment with different concentrations of recombinant CXCL6; the bottom row shows magnified views (scale bar: 200 μm, enlarged: 100 μm). (G, H) Quantification of tube length and branch points from the four independent experiments shown in (F). (I) Representative images of HUVECs treated with THY1⁺ CAF-1-conditioned medium combined with different concentrations of CXCL6; the bottom row shows magnified views. (J, K) Quantification of tube formation from the four independent experiments shown in (I) (scale bar: 200 μm, enlarged: 100 μm). *: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001; p values were calculated via paired Student’s t tests.

Quantitative analysis of array fluorescence signals confirmed significantly elevated secretion of CXCL6, VEGFA, MMP2, and MMP19 in THY1⁺ CAFs relative to THY1⁻ CAFs (Fig. 7C). These findings were further validated by ELISA, which demonstrated markedly increased secretion of CXCL6 and VEGFA in THY1⁺ CAF-conditioned media from both donors (Fig. 7D, E).

Although VEGFA is a well-established proangiogenic factor in prostate and other malignancies, the role of CXCL6 is less extensively studied. Given prior evidence implicating CXCL6 in prostate cancer–associated angiogenesis, we investigated its functional contribution. Treatment of HUVECs with recombinant CXCL6 (1 ng/mL and 10 ng/mL) significantly enhanced tube formation in a dose-dependent manner (Fig. 7F–H). Importantly, the proangiogenic effect of CXCL6 persisted in the presence of CAF-conditioned media, indicating that CXCL6 can stimulate tube formation independently of other CAF-derived factors (Fig. 7I–K).

Together, these results demonstrate that THY1⁺ CAFs possess a distinct secretory profile characterized by elevated levels of multiple proangiogenic mediators. In particular, we identify CXCL6 as a novel effector through which THY1⁺ CAFs promote angiogenesis in prostate cancer.

THY1⁺ CAFs promote angiogenesis through CXCL6/CXCR2 signaling and THY1-dependent regulation of VEGFA secretion

To elucidate the mechanism by which CXCL6, a key proangiogenic factor secreted by THY1⁺ CAFs, enhances HUVEC tube formation, we considered prior studies showing that CXCL6 binds to and activates its receptors CXCR1 and CXCR2 [52,53], with a preferential affinity for CXCR2 [54], and can promote tumor progression via the PI3K/AKT pathway in osteosarcoma. Based on this evidence, we hypothesized that THY1⁺ CAF-derived CXCL6 promotes angiogenesis through binding to CXCR2 on endothelial cells.

To test this hypothesis, HUVEC tube formation assays were performed using conditioned medium from THY1⁺ CAF1 cells in the presence or absence of the CXCR2-specific inhibitor SB225002. As shown in Fig. 8A–C, recombinant CXCL6 (10 ng/mL) significantly enhanced tube formation; however, this effect was completely abolished by SB225002. Notably, SB225002 also substantially suppressed the proangiogenic activity of THY1⁺ CAF1-conditioned medium even in the absence of exogenous CXCL6, indicating that endogenous CXCL6 secreted by THY1⁺ CAFs exerts its effects primarily through CXCR2 signaling in endothelial cells.

Fig. 8.

Functional role of THY1 in regulating the proangiogenic CAF phenotype. (A) Representative images of HUVEC tube formation after treatment with THY1⁺ CAF-1-conditioned medium combined with CXCL6 and the CXCR2 inhibitor SB225002; the bottom row shows magnified views (scale bar: 200 μm, enlarged: 100 μm). (B, C) Quantification of tube formation from four independent experiments. (D) Validation of THY1 knockdown in THY1⁺ CAF1 cells via shRNA, as assessed by qPCR and Western blotting. (E) HUVEC tube formation assay after treatment with CM from THY1⁺ CAF1-shNC, THY1⁺ CAF1-shTHY1#1, or THY1⁺ CAF1-shTHY1#2 cells. (F, G) Quantification of tube formation from (E). (H, I) qPCR and ELISA analysis of CXCL6 expression and secretion in shRNA-treated CAFs. (J, K) qPCR and ELISA analysis of VEGFA expression and secretion in shRNA-treated CAFs. ns: not significant.

We next investigated whether THY1 itself contributes directly to the proangiogenic phenotype of THY1⁺ CAFs. Using shRNA-mediated knockdown, we achieved >70 % suppression of THY1 expression at both mRNA and protein levels (Fig. 8D). Conditioned medium from THY1-knockdown cells exhibited a markedly diminished capacity to promote HUVEC tube formation compared with control shRNA-treated cells (Fig. 8E–G), underscoring the functional importance of THY1 expression in driving the angiogenic activity of CAFs.

Given that THY1⁺ CAFs secrete elevated levels of CXCL6 and VEGFA, we assessed whether THY1 knockdown affected the production of these factors. Following THY1 silencing, CXCL6 levels remained unchanged at both the transcript and protein levels, whereas VEGFA expression and secretion were significantly reduced (Fig. 8H–K). These findings indicate that THY1 regulates VEGFA, but not CXCL6, production in CAFs. Thus, the impaired angiogenic capacity observed upon THY1 knockdown is likely attributable to diminished VEGFA secretion.Given our previous findings that THY1⁺ CAFs secrete elevated levels of CXCL6 and VEGFA, we examined whether THY1 knockdown affects the expression and secretion of these factors. Following THY1 knockdown, CXCL6 levels showed no change at either the mRNA or protein levels, whereas VEGFA expression and secretion were notably decreased. These results indicate that THY1 expression in CAFs regulates VEGFA but not CXCL6 production and that the impaired angiogenic capacity resulting from THY1 knockdown is likely mediated through reduced VEGFA secretion.

In summary, THY1⁺ CAFs promote angiogenesis through two complementary mechanisms: CXCL6 secretion, which activates CXCR2 on endothelial cells; and THY1-dependent regulation of VEGFA expression and secretion. Although the precise molecular mechanism by which THY1 controls VEGFA production remains to be clarified, our results establish THY1 as a functional regulator of the proangiogenic phenotype in CAFs.

Discussion

CAFs play a crucial role in the TME by mediating interactions between tumor cells and stromal components via the paracrine release of growth factors, immune modulators, and signaling molecules [55,56]. These interactions facilitate tumor progression by promoting stromal remodeling [57], angiogenesis [57], tumor cell proliferation [58,59], invasion/metastasis [60], and therapeutic resistance [60] while simultaneously suppressing antitumor immunity [61,62]. Targeting CAFs has become a promising therapeutic approach for various cancers due to their strong tumor-promoting functions [63].

However, CAFs represent a highly heterogeneous population originating from diverse cellular sources, including resident fibroblasts [64], endothelial cells [64], epithelial cells [65], pericytes [66], and bone marrow-derived mesenchymal stem cells [67]. This cellular diversity underpins profound functional heterogeneity: while certain CAF subpopulations drive tumor progression, others may exert tumor-restraining effects [61,62,68]. Therefore, a major challenge in developing effective CAF-targeted therapies lies in resolving this heterogeneity—specifically, in identifying, characterizing, and targeting specific protumorigenic CAF subsets. Although these subtypes have been suggested to promote tumor angiogenesis, this hypothesis has not been experimentally validated. In parallel, our subsequent scRNA-seq study [21] further delineated five functionally specialized fibroblast subpopulations in prostate cancer, highlighting their diverse roles in ECM remodeling, immune modulation, and tumor activation. However, the functional validation and mechanistic underpinnings of these distinct CAF subsets remain largely unexplored, underscoring a critical gap in understanding their specific contributions to prostate cancer progression.

This study sought to address this gap by comprehensively characterizing CAF abundance, heterogeneity, and clinical significance in PCa. We first demonstrated that high CAF abundance is consistently associated with aggressive clinicopathological features (e.g., higher Gleason score, advanced T stage, nodal involvement, and metastasis) and poor prognosis across multiple cohorts. These findings align with and extend previous observations in PCa and other malignancies [69,70], underscoring the general prognostic value of stromal activation.

To move beyond bulk abundance measurements and capture the molecular footprint of CAF activity, we developed and validated an 8-gene CAF-derived prognostic signature. This signature not only robustly predicts BCR but also independently stratifies patient risk beyond standard clinical parameters. Notably, the risk score was significantly elevated in Gleason score 4 + 3 tumors compared with 3 + 4 tumors, reflecting the heightened stromal activation and worse prognosis associated with the primary pattern 4 architecture. This distinction is clinically pertinent, as Gleason 4 + 3 disease carries a significantly worse prognosis than 3 + 4 disease does—a critical nuance often overlooked since both are frequently grouped under Gleason score 7, representing a major limitation of the conventional Gleason system [71]. In fact, a revised grading system proposed on the basis of data from Johns Hopkins categorizes Gleason 3 + 4 as prognostic grade group II and Gleason 4 + 3 as grade III, thereby formally recognizing their differing prognostic implications [72]. Studies comparing these subgroups have shown inconsistent results and largely focus on recurrence-free, rather than overall or cancer specific, survival [73,74]. Our signature provides a molecular correlate for this aggressive phenotype, potentially aiding in refined risk stratification and informing treatment decisions.

Interrogation of the scRNA-seq data revealed that six of the eight signature genes (COL1A2, COL1A1, COL3A1, SULF1, COL5A1, and THY1) were predominantly expressed in CAFs. THY1, a cell surface glycoprotein, serves as a widely recognized marker for certain cell types [27,29]. Prior research indicated elevated levels of CD31⁺THY1⁺ endothelial cells in castration-resistant prostate cancer (CRPC) patients, implying their possible involvement in disease progression [20]. Thus, we focused our subsequent validation on THY1 and COL1A1, major collagen components highly specific to CAFs. Multiplex immunofluorescence analysis of a prostate cancer tissue microarray identified a THY1⁺ CAF subpopulation, linking its presence to clinical outcomes; a greater proportion of THY1⁺ CAFs correlated with elevated Gleason scores and significantly reduced recurrence-free survival. These findings identify THY1⁺ CAFs as a distinct, prognostically adverse stromal subpopulation in PCa, which is consistent with findings in lung cancer, where THY1⁺ CAFs are correlated with poor outcomes [75].

THY1 expression in the PCa microenvironment has been documented across multiple studies [23,[29], [30], [31]]. However, the functional properties of THY1⁺ CAFs and their distinction from THY1⁻ CAFs have not been experimentally characterized. In this study, we fill this critical gap by demonstrating the potent proangiogenic capacity of THY1⁺ CAFs, providing direct mechanistic support for their contribution to PCa progression. Building upon this foundation, we analyzed two independent scRNA-seq datasets—both generated in our team’s previous studies—and confirmed that THY1 expression demarcates specific CAF clusters. Functional enrichment analysis of genes upregulated in these THY1-high clusters strongly implicated their involvement in angiogenesis and vasculature development. This bioinformatic prediction was functionally validated: highly purified primary THY1⁺ CAFs significantly enhanced endothelial tube formation compared with THY1⁻ CAFs.

Mechanistic profiling revealed that THY1⁺ CAFs exhibit a distinct secretory phenotype enriched with proangiogenic factors such as CXCL6, an ELR⁺CXC chemokine signaling pathway via CXCR2 [54]. While CXCL6 is known to promote angiogenesis in various cancers, its specific role and source in prostate cancer—particularly in CAFs—remain underexplored. Our study demonstrated that both recombinant CXCL6 and THY1⁺ CAF-conditioned medium stimulate endothelial tube formation in a CXCR2-dependent manner, as inhibition by SB225002 abrogated this effect. These findings suggest that CAF-derived CXCL6 promotes angiogenesis in PCa via CXCL6–CXCR2 signaling, highlighting the importance of chemokine-mediated crosstalk in the TME. Further validation using CXCR2-knockdown models is warranted to confirm this mechanism. Our results suggest a paracrine axis whereby THY1⁺ CAF-derived CXCL6 activates endothelial CXCR2 to promote angiogenesis in PCa. These findings provide functional evidence for CXCL6-mediated angiogenesis and identify a specific stromal source of this chemokine, highlighting the CXCL6/CXCR2 pathway as a potential therapeutic target.

Interestingly, shRNA-mediated knockdown of THY1 itself impaired the proangiogenic function of CAFs but did not affect CXCL6 expression or secretion. However, THY1 knockdown significantly reduced the expression and secretion of VEGFA, a canonical angiogenic factor. This finding reveals a second, THY1-dependent mechanism regulating angiogenesis, potentially through the modulation of VEGFA production. The precise molecular link between THY1 and VEGFA regulation warrants further investigation. Emerging evidence further supports the concept that stromal–immune crosstalk shapes distinct immunobiological states of prostate cancer. Notably, Ge et al. recently reported that prostate cancer can be stratified into two immune phenotypes, TregR and TregP, of which driven by regulatory T-cell (Treg) activity [41]. The TregR phenotype is characterized by heightened immunosuppression, poor clinical outcomes, and resistance to standard therapies. Although our study focuses on CAF heterogeneity, the functional properties of THY1⁺ CAFs observed here, including enhanced secretion of CXCL6 and VEGFA, may contribute to a microenvironment permissive to Treg recruitment and expansion. Prior studies have shown that ELR⁺ CXC chemokines and VEGFA can indirectly promote Treg accumulation by impairing dendritic cell maturation or fostering endothelial barriers that restrict effector immune infiltration [76,77]. Thus, tumors enriched in THY1⁺ CAFs may overlap biologically with the TregR phenotype, exhibiting both angiogenic activation and reinforced immunosuppression.

This study offers detailed evidence identifying THY1⁺ CAFs as a proangiogenic subgroup in prostate cancer, though it is important to recognize certain limitations. First, the prognostic model and CAF abundance analysis were derived from retrospective cohorts. Despite validation in various independent datasets, prospective multicenter studies are necessary to further verify the clinical utility of the 8-gene signature and THY1⁺ CAF quantification for risk stratification. Second, the precise molecular mechanism by which THY1 regulates VEGFA expression and secretion remains unclear. Future investigations employing chromatin immunoprecipitation (ChIP) and promoter-reporter assays and exploration of potential signaling pathways (e.g., TGF-β or PDGF signaling) are needed to elucidate how THY1, a GPI-anchored surface protein, modulates transcriptional activity or secretory processes. Third, the functional role of the CXCL6/CXCR2 axis was demonstrated primarily via in vitro endothelial cell models. The proangiogenic and protumorigenic roles of CXCL6 from THY1⁺ CAFs necessitate in vivo validation using patient-derived xenograft models or transgenic mouse models of prostate cancer, alongside CXCR2 knockout or pharmacological inhibition. Finally, the biological roles of other upregulated secretory proteins (e.g., MMP2) in THY1⁺ CAFs were not explored in this study. Their contributions to extracellular matrix remodeling and tumor invasion represent promising avenues for future research. Overcoming these limitations will enhance our conclusions and aid in developing new stromal-targeting therapies for advanced prostate cancer.

In conclusion, our study delineates the multifaceted role of cancer-associated fibroblasts in prostate cancer progression. We first established that high stromal abundance is correlated with aggressive disease and poor prognosis. To refine this observation, we constructed an 8-gene prognostic signature that includes THY1 as a key constituent and demonstrated its robust utility in predicting biochemical recurrence across multiple cohorts. In addition to this prognostic model, we functionally characterized THY1⁺ CAFs as a distinct proangiogenic subpopulation that drives tumor vascularization through two mechanisms: CXCL6/CXCR2-mediated paracrine signaling and THY1-dependent regulation of VEGFA secretion. These findings significantly advance our understanding of CAF heterogeneity in PCa and reveal that THY1 and the CXCL6/CXCR2 axis are promising stromal targets for therapeutic intervention. Future studies should prioritize in vivo validation of these mechanisms and explore strategies to selectively inhibit the protumorigenic functions of THY1⁺ CAFs.

Ethics approval and consent to participate

Our study was approved by the Ethics Committee of FUSCC.

Funding

This research received funding from the Shanghai Cancer Development Center (SHDC2025CCS007) and the National Natural Science Foundation of China (81202003).

Data availability statement

The single-cell RNA sequencing datasets utilized in this work are publicly available in the Gene Expression Omnibus (GEO) under accession numbers GSE141445 and GSE176031. These datasets can also be accessed through their respective original publications. Should download issues arise, interested researchers may contact the corresponding authors via email to request the data. Bulk RNA-seq and associated clinical data for TCGA-PRAD were obtained from the NCI Genomic Data Commons (GDC) Portal (https://portal.gdc.cancer.gov/). Additionally, the raw multiplex immunofluorescence images generated and analyzed during this study are available from the corresponding authors upon reasonable request, for non-commercial academic use only, subject to compliance with ethical provisions and patient confidentiality requirements.

CRediT authorship contribution statement

Yongqiang Huang: Writing – original draft, Software, Methodology, Formal analysis, Data curation. Chunping Xiang: Writing – review & editing, Validation, Methodology, Formal analysis. Yu Wang: Methodology, Investigation, Formal analysis, Data curation. Wei Zhang: Validation, Software, Resources, Data curation. Leilei Du: Writing – original draft, Visualization, Resources, Investigation, Conceptualization. Wenfeng Wang: Writing – original draft, Resources, Project administration, Investigation, Conceptualization. Guohai Shi: Writing – review & editing, Supervision, Methodology, Investigation, Funding acquisition. Jianhua Wang: Writing – review & editing, Supervision, Resources, Project administration, Investigation.

Declaration of competing interest

The authors declare no competing interests. Prof. Jianhua Wang has full access to all study data and assumes responsibility for data integrity and analytical accuracy.