Prostate cancer cancer‐associated fibroblasts with stable markers post‐androgen deprivation therapy associated with tumor progression and castration resistant prostate cancer

Abstract

The specificity and clinical relevance of cancer‐associated fibroblasts (CAFs) in prostate cancer (PCa), as well as the effect of androgen deprivation therapy (ADT) on CAFs, remain to be fully elucidated. Using cell lineage diversity and weighted gene co‐expression network analysis (WGCNA), we pinpointed a unique CAF signature exclusive to PCa. The specificity of this CAF signature was validated through single‐cell RNA sequencing (scRNA‐seq), cell line RNA sequencing, and immunohistochemistry. This signature associates CAFs with tumor progression, elevated Gleason scores, and the emergence of castration resistant prostate cancer (CRPC). Using scRNA‐seq on collected samples, we demonstrated that the CAF‐specific signature is not altered by ADT, maintaining its peak signal output. Identifying a PCa‐specific CAF signature and observing signaling changes in CAFs after ADT lay essential groundwork for further PCa studies.

Our study successfully identified a specific signature for CAFs within prostate cancer (PCa). We established that CAFs are linked to PCa progression, higher Gleason scores, and CRPC. The significance of CAFs in PCa was further underscored by their maintenance of high signal output post‐ADT and the stable gene signature.

1. INTRODUCTION

In 2020, prostate cancer (PCa) is expected to rank as the second most common and the fifth deadliest cancer among men, with approximately 1.4 million new cases and 375,000 deaths globally. ¹ Androgen deprivation therapy (ADT) forms the foundation of treatment for locally advanced and metastatic PCa. ² , ³ Although initially effective, most patients eventually develop metastatic castration‐resistant PCa (mCRPC). ⁴ , ⁵ Recurrent PCa following anti‐androgen therapy often displays varied histology and changes in lineage marker expression. ⁶ , ⁷ , ⁸ Increasing evidence suggests a connection between the tumor microenvironment (TME), particularly cancer‐associated fibroblasts (CAFs), and the progression of PCa and castration‐resistant prostate cancer (CRPC). ⁹ , ¹⁰

The role of CAFs in tumors is complex and diverse, with their potential as therapeutic targets increasingly recognized. ¹¹ CAF heterogeneity and phenotypic plasticity involve complex signaling, evident in the interconversion among various states or phenotypes, such as quiescent, inflammatory, and myofibroblastic fibroblasts. ¹² Investigations into the role of CAFs within PCa are notably scant. It has been elucidated that SPP1 + myofibroblasts, differentiated from the subset of inflammatory CAFs, impart resistance to ADT in PCa through a paracrine signaling mechanism mediated by SPP1‐ERK. ¹¹ , ¹³ CAFs within the stroma of PCa might also promote tumor progression or metastasis by manipulating the tumor immune microenvironment. ¹⁴ However, specific markers for CAFs in PCa remain unknown.

Using the CAF tags from the Estimating the Proportion of Immune and Cancer cells (EPIC), Microenvironment Cell Populations counter (MCPCOUNTER), and xCell open platforms, our preliminary assessment suggested that CAFs linked to PCa progression. The Methods section provides a comprehensive description of the different CAFs assessment platforms. In our previous studies, we developed a method for identifying specific markers of CAFs within the cancer and successfully identified CAF‐specific gene signatures in bladder urothelial carcinoma and renal clear cell carcinoma. ¹⁵ , ¹⁶ We will persist in identifying PCa‐specific CAF markers through weighted gene co‐expression network analysis (WGCNA). We confirmed the expression specificity of the gene signature for CAFs using single‐cell RNA sequencing (scRNA‐seq), cell line RNA sequencing, and immunohistochemistry (IHC) data. Furthermore, we established the clinical significance of CAFs in PCa. Following ADT, we collected clinical samples and conducted scRNA‐seq. Our findings revealed that the gene signature of CAFs remained unchanged, maintaining a high signal output state.

2. MATERIALS AND METHODS

2.1. Data download and processing

RNA sequencing (RNA‐seq) data and associated clinical information for human prostate adenocarcinoma (PRAD) samples were sourced from The Cancer Genome Atlas (TCGA) database (portal.gdc.cancer.gov). This dataset, updated on October 10, 2021, includes 540 files across 485 cases. The RNA‐seq data, comprising 51 normal and 489 cancer samples, were compiled into matrix files.

To explore whether the gene signatures screened by WGCNA are consistent with the cell‐specific genes revealed by single‐cell RNA sequencing (scRNA‐seq), we downloaded and analyzed the scRNA‐seq data GSE157703 from two prostate cancer patients. ¹⁷ The data processing and cell clustering were conducted using the Seurat R package, resulting in the generation of FeaturePlots. These plots were then used for Uniform Manifold Approximation and Projection (UMAP) to achieve dimensionality reduction.

To validate the expression specificity of gene signatures in fibroblasts, we obtained gene expression data and sample information for cell lines from the depmap portal (depmap.org/portal), which includes 1393 cell lines and 19,177 genes, updated as of March 16, 2022. Focusing on the tumor environment in PCa, we selected cell lines labeled as blood, fibroblast, prostate, lymphocyte, and plasma cells for comparison. We extracted RNA‐seq data for 11 PCa and 39 fibroblast cell lines. Using R software version 3.6.2, we created heatmaps and volcano plots to display the differentially expressed genes (DEGs).

To ascertain the clinical relevance, we downloaded data from the Gene Expression Omnibus (GEO) database encompassing various series: GSE134051 with 255 Gleason score samples, GSE21034 comprising 185 samples with diverse TNM stages and Gleason scores, GSE116918 containing 248 tumor samples with varying TNM stages, Gleason scores, prostate‐specific antigen (PSA) levels, GSE33269 with 55 CRPC related samples, and GSE70770 including 219 CRPC related samples.

2.2. Fractions of cancer‐associated fibroblasts in the tumor microenvironment

To examine the changes in CAFs within PRAD tumor tissues from TCGA, we analyzed CAF infiltration using algorithms from EPIC, ¹⁸ MCP‐counter, ¹⁷ and xCell ¹⁹ in the Immune Estimation module of TIMER2.0 (timer.cistrome.org). ²⁰ EPIC generates reference gene expression profiles for key tumor‐infiltrating immune cell types (CD4+ T, CD8+ T, B cells, natural killer cells, and macrophages) and extrapolates reference spectrums for CAFs and endothelial cells. ¹⁸ The MCP‐counter method enables accurate quantification of the absolute abundance of eight immune and two stromal cell populations in heterogeneous tissues using transcriptomic data. ¹⁷ xCell, a novel method based on ssGSEA, calculates abundance scores for 64 immune cell types, including extracellular matrix cells. ¹⁹

2.3. Weighted gene co‐expression network analysis

The principle of WGCNA posits that gene modules identified as highly co‐expressed can represent specifically expressed genes of certain cell types within tumor tissues, particularly fibroblasts. ¹⁵ , ¹⁶ This methodology becomes more effective with an increased number of tissue samples and a wider range of cell composition ratios, making it ideal for pinpointing gene sets unique to specific cell types. In the multifaceted tumor environment, the identification of cell‐specific genes largely relies on the prevalence of certain cell types, reducing the influence of other cellular components.

2.3.1. Highly co‐expressed gene set–gene module

We used the WGCNA package in R for weighted correlation network analysis. ²¹ To filter out highly correlated genes with insignificant expression changes, we selected genes with a top 25% variance. ¹⁵ , ¹⁶ After refining the RNA‐seq data to eliminate outliers, we constructed a Pearson correlation matrix and developed a weighted adjacency matrix. This matrix accentuates strong correlations while downplaying weaker ones. Choosing an appropriate β value through power calculation, we generated a topological overlap matrix (TOM). Using TOM‐based dissimilarity measures, we conducted average‐linkage hierarchical clustering and created module dendrograms, establishing modules with a minimum gene dendrogram size of 30.

2.3.2. Identification of interested modules

Gene significance (GS) was calculated to assess the correlation between genes and cell fractions, thereby determining the significance of each module. The expression patterns of module eigengenes were summarized as a singular characteristic within each module. To maintain module specificity, we refrained from setting a cutoff threshold for merging modules.

2.3.3. Representative genes in a module

Gene significance and module membership (MM) are valuable in evaluating gene‐phenotype associations and their significance within modules. Consistent with prior research, we established high MM and GS (MM.cor and GS.cor, respectively) as criteria to identify key genes in a module. ¹⁵ , ¹⁶

2.4. Cell markers

Initially, we identified PRAD‐specific CAF markers, including fibroblasts and myofibroblasts, and collated a comprehensive list of CAF markers. ²² , ²³ , ²⁴ These markers are both specific and nonspecific to CAFs. To validate the CAF signature in PRAD, we performed correlation analysis using TCGA data. Second, for cell annotation in scRNA‐seq, we annotated endothelial cells, monocytes, B cells, T cells, mast cells, luminal cells, and basal/intermediate cells using established PCa markers. ²⁵ Additionally, within the monocyte subset, cells such as M1, M2, and DCs were annotated with recognized markers: CD86, TLR2, and others for M1; MRC1, MSR1, MARCO, NRP2, CD276, and others for M2 ²⁶ , ²⁷ ; and CD1C, CLEC10A, and others, for DCs. ²⁸ Finally, CellMarker 2.0 enhanced our scRNA‐seq data annotation, including B cells and neutrophils. ²⁹

2.5. Pathway and process functional enrichment analysis

To validate gene functions, we conducted a gene enrichment analysis using Metascape (metascape.org). This included resources like the Kyoto Encyclopedia of Genes and Genomes, Gene Ontology, Reactome gene sets, and CORUM for pathway and process enrichment analysis. ³⁰ In single‐cell analysis, we identified functional differences across cell types using the Seurat and fgsea R packages.

2.6. Analysis of the clinical significance of each gene in the gene signature

To investigate the clinical relevance of each gene in the PRAD CAF signature, we used receiver operating characteristic (ROC) curve analysis. This analysis assessed the diagnostic effectiveness of the gene signature in relation to the Gleason score and stage.

2.7. Protein expression of the gene signature

To examine the expression of pivotal genes in PCa and stromal cells, we sourced IHC images of PCa from The Human Protein Atlas (http://www.proteinatlas.org), a project dedicated to mapping human proteins in various pathologies. ³¹ We specifically chose images featuring both cell types, with the guidance of expert pathologists.

2.8. Single sample gene set enrichment analysis

Single‐sample gene set enrichment analysis (ssGSEA) is effective in detecting and differentiating changes in specific cell classes across various samples, using gene signatures. ¹⁶ , ³² , ³³ We applied the GSEA program to extract absolute enrichment scores from the previously identified CAF gene signature using WGCNA. The infiltration level of each cell type was quantified using ssGSEA in the R package gsva, where ssGSEA used a deconvolution approach.

2.9. scRNA‐seq

Two post‐ADT, castration‐sensitive PCa samples were sequenced using the 10× Genomics platform. One patient is a 68‐year‐old man with a Gleason score of 8 and seminal vesicle invasion. Another patient is a 69‐year‐old man with a Gleason score of 7 and no metastasis. The clinical surgical samples were collected immediately upon excision and based on preoperative biopsy results and imaging examinations. The excised tissue was cut into small pieces measuring 0.5 cm × 0.5 cm × 0.5 cm. We collected two to three tissue pieces, totaling approximately 200–400 mg, and immediately placed them into pre‐chilled tissue preservation solution at 4°C. The samples were then stored on ice and subsequently delivered to Genechem (Shanghai, China) for dissociation, detection, and analysis. The subsequent sections provide a detailed description of the methodology. The scRNA‐seq data from this study can be accessed publicly in the GEO repository, under the accession number GSE250189. We strictly adhered to the ethical guidelines and legal requirements set by the hospital's ethics committee, ensuring informed consent was obtained from every patient.

2.9.1. Sequencing

Cell suspensions were loaded onto Chromium microfluidic chips using 3′ chemistry (v2 or v3, project‐dependent) and barcoded using a 10× Chromium Controller (10X Genomics). RNA from these barcoded cells was reverse‐transcribed, and sequencing libraries were constructed using the Chromium Single Cell 3′ v2 (v2 or v3, depending on the project) reagent kit (10X Genomics), following the manufacturer's instructions. Sequencing was conducted on an Illumina platform (NovaSeq 6000), in accordance with Illumina's guidelines. The quality of the raw reads was assessed using fastp.

2.9.2. Generation and analysis of single‐cell transcriptomes

Raw reads were demultiplexed and aligned to the reference genome using the 10× Genomics Cell Ranger pipeline (https://support.10xgenomics.com/single‐cell‐geneexpression/software/pipelines/latest/what‐is‐cell‐ranger), using default settings. All subsequent single‐cell analyses utilized Cell Ranger and Seurat, except where explicitly stated. Briefly, for every gene and cell barcode (filtered by CellRanger), unique molecule identifiers (UMIs) were tallied to create digital expression matrices. Seurat then conducted secondary filtration: a gene was deemed expressed if found in more than three cells, and each cell required a minimum of 200 expressed genes. Additionally, non‐target cells were filtered out.

2.10. Cell–cell communication analysis

We used CellChat, a comprehensive repository encompassing ligands, receptors, cofactors, and their interactions, for inferring and analyzing cell‐to‐cell communication. The user‐friendly toolkits, CellChat and its Web‐based Explorer (http://www.cellchat.org/), facilitate the discovery of novel cell‐to‐cell communication types and the construction of cell communication networks. ³⁴ For cell–cell interaction analysis, we calculated expression levels relative to the total read length mapping to the same gene set across all transcriptomes. These expression values were then averaged within each single‐cell cluster or cell sample.

2.11. Trajectory analysis

We used Monocle 2 (version 2.28.0), an R package, to arrange individual cells along “pseudo‐time” and infer their developmental trajectories. ³⁵ Following quality control, genes were input into Monocle's Reverse Graph Embedding algorithm to model the trajectory. Monocle then reduced the data dimensionality and organized the cells in pseudo‐time order.

2.12. Statistical analysis

Statistical analyses were conducted using GraphPad Prism 8.4 (GraphPad Software, San Diego, CA, USA). We determined statistical significance using Student's t‐test (two‐tailed) for comparisons between two groups and one‐way analysis of variance (ANOVA) and/or the Tukey test for multiple groups. The Cox proportional hazards regression model was used to assess the association between gene expression levels and overall survival. Data are presented as mean ± SD. In bar graphs, asterisks denote significance levels: *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. A p‐value of less than 0.05 was considered statistically significant.

3. RESULTS

3.1. Changing trends in the stroma and cancer‐associated fibroblasts

Initially, we used MCP‐counter, xCell, and EPIC to estimate CAF infiltration in PRAD. Results from MCP‐counter and xCell indicated no clinical significance associated with CAF infiltration (Figure S1A). However, EPIC revealed an increasing trend in CAF proportion with the advancement of tumor stage and Gleason score (Figure 1A), graded according to the International Society of Urological Pathology (ISUP) 2014 criteria. ³⁶ The EPIC results suggested an escalation in CAF proportion correlating with tumor progression and rising Gleason score (Figure 1B). Given that EPIC's CAF estimates were based on genetic signatures from other tissues, we deemed the assessment of CAFs to be imprecise. Combined with findings from MCP‐counter and xCell, this led us to question the clinical relevance of CAFs in PRAD.

FIGURE 1.

Cancer‐associated fibroblasts (CAFs) in prostate cancer. (A) Correlation of the Estimating the Proportion of Immune and Cancer cell (EPIC)‐derived CAFs with stage and Gleason score. (B) Association between CAF proportion and increased tumor stage and Gleason score. (C) Correlation of EPIC‐CAFs in the cyan module with pathological T stage and Gleason score. Correlation coefficients and p‐values are presented in each cell. (D) Cluster tree analysis of CAF‐correlated modules. (E) Weighted gene co‐expression network analysis (WGCNA) identifies cell‐specific genes as a highly co‐expressed gene set. ****p < 0.0001.

3.2. Identification of gene modules and gene signature

We effectively and objectively identified cell‐specific gene sets associated with CAFs using WGCNA, applied to TCGA‐PRAD sequencing samples. A soft threshold of β = 5 (scale‐free R ² = 0.894) was chosen to construct a scale‐free network. Three modules showed correlation with the CAF fraction. We selected the one or two modules most relevant to CAFs as our modules of interest: the Cyan, Green, and Salmon modules (Figure 1C). Notably, the Cyan module, related to EPIC‐CAFs, and the Green and Salmon modules, associated with MCP‐counter and xCell, exhibited distinct separations in the cluster tree (Figure 1D). This suggested that Cyan module‐related EPIC‐CAFs were heterogeneous compared to MCP‐counter and xCell‐related CAFs. The association of EPIC‐related CAFs with Gleason score progression and tumor stage (Figure 1C) indicated that EPIC‐related CAFs represent a clinically significant CAF subtype, thereby becoming the focus of our study.

Based on the EPIC‐correlated Cyan module, we established the gene signature or key genes for PRAD CAFs using criteria of GS.cor >0.85 and MM.cor >0.85 (Figure 2A). The EPIC‐related CAF gene signature comprises 14 genes: ADAM metallopeptidase with thrombospondin type 1 motif 2 (ADAMTS2), asporin (ASPN), cadherin 11 (CDH11), collagen type I alpha 1 chain (COL1A1), collagen type I alpha 2 chain (COL1A2), collagen type III alpha 1 chain (COL3A1), collagen type V alpha 1 chain (COL5A1), cathepsin K (CTSK), periostin (POSTN), ras‐related protein rab‐31 (RAB31), secreted frizzled related protein 2 (SFRP2), secreted protein acidic and cysteine rich (SPARC), thrombospondin 2 (THBS2), and versican (VCAN). Considering the similarity of the Green and Salmon modules, we defined GS.cor >0.85 and MM.cor >0.85 as the criteria for MCP‐counter and xCell‐related CAF gene signatures, with the requirement that key genes in both modules satisfy MM.cor.green >0.85 and MM.cor.salmon >0.85 (Figure S1B). This signature includes nine key genes, featuring the myofibroblast marker actin alpha 2 smooth muscle (ACTA2). ³⁷

FIGURE 2.

Prostate cancer cancer‐associated fibroblast (CAF) signature via weighted gene co‐expression network analysis (WGCNA): (A) Genes with MM.cor >0.85 and GS.cor >0.85 designated as CAF‐specific markers in the cyan module. (B) Of the various cells in the prostate cancer microenvironment, 14 key genes show the highest expression in fibroblasts. (C) Volcano plot depicting differential gene expression between prostate cancer and fibroblast cell lines, highlighting 14 key genes with black circles. (D) Elevated expression of key genes in stromal regions, indicated by a blue arrow for stromal parts. (E) Construction of a cell cluster feature plot from prostate cancer scRNA‐seq data. (F) Heatmap representation of 14 fibroblast and 9 myofibroblast markers in scRNA‐seq. (G) Annotation of fibroblasts and myofibroblasts in the cell cluster feature plot.

3.3. WGCNA can sort cancer‐associated fibroblast‐specific expressed genes

Our previous research established that WGCNA can pinpoint genes specifically expressed by CAFs in bladder and kidney cancers. ¹⁵ , ¹⁶ Similarly, using the same analytical approach, WGCNA successfully identified a set of genes specific to CAFs in PRAD (Figure 1E). CAFs predominantly consist of two cell types: myofibroblasts and fibroblasts. Considering the module clustering in Figure 1D, the marker‐module correlation in Figure 1C, and the established myofibroblast marker ACTA2, ³⁷ we inferred that EPIC‐related CAFs correspond to fibroblasts, while MCP‐counter and xCell‐related CAFs align with myofibroblasts. The varying proportions of these cells affect the expression profiles of their specific gene sets. Although potential interference from other cell types exists, it is limited in affecting their robust co‐expression patterns. Schematic diagrams were created to elucidate the role of WGCNA in this process (Figure 1E).

3.4. Expression of the key genes in cell lines

To confirm the specific expression of key genes identified in the modules, we examined their expression levels in cell lines associated with the prostate cancer microenvironment. Using the cell line classifications from the Cancer Cell Line Encyclopedia (CCLE), we selected blood, fibroblast, prostate, lymphocyte, and plasma cell lines corresponding to cells in the PRAD TME for comparison. We found that the expression levels of Cyan‐EPIC key genes in fibroblasts were significantly higher compared to the other cell types (Figure 2B). A similar trend was observed for key genes in myofibroblasts (Figure S1C). Furthermore, we analyzed PCa and fibroblast cell line data from the CCLE, setting |fold change| >1 and p < 0.05 as criteria to identify differentially expressed genes (DEGs) between these two cell types. The volcano plots demonstrated that these key genes were more highly expressed in fibroblasts and myofibroblasts compared to prostate cancer cells (Figure 2C, Figure S1D).

3.5. Proteins of the key genes in pathology

The initial findings were derived from transcriptome data. To delve into the protein expression locations of the key genes, we conducted preliminary analyses using IHC images from The Human Protein Atlas. With the exception of the fibroblast markers CTSK and SFRP2, which lacked pathological IHC data for PCa, the protein expressions of the other key genes exhibited consistent high expression characteristics in the CAFs‐containing stromal tissue in IHC images of PCa (Figure 2D, Figure S1E). Detailed information regarding prostate cancer IHC, including age and pathological grade, can be found in Table S1.

3.6. Distribution and expression of the key genes in the scRNA‐seq

To verify if WGCNA can identify PRAD CAF‐specific genes, we examined the distribution and expression of key genes within the cell landscape using scRNA‐seq data from GSE157703. Using the UMAP dimensionality reduction algorithm, we visualized the expression distribution of cells in two‐dimensional plots, labeling cells within the same cluster from two patients with identical colors (Figure 2E). For the 9 or 10 cell clusters identified, we depicted the expression of 14 and 9 markers, respectively, using heatmaps in the scRNA‐seq data (Figure 2F). The results aligned with our expectations, showing these markers predominantly expressed in fibroblasts and myofibroblasts. Having validated the marker specificity, we then annotated fibroblasts and myofibroblasts in the cell landscape (Figure 2G).

3.7. Correlation between the key genes and cancer‐associated fibroblast markers

Recognizing the clinical importance of fibroblasts in PRAD CAFs, we identified these fibroblasts as the characteristic CAFs in PRAD. Among the general and specific CAF markers, five and four, respectively, intersected with the PRAD CAF signature. The 14 key genes exhibited very high co‐expression (Figure 3A). Furthermore, when comparing the CAF gene signatures identified in our previous studies on Kidney Renal Clear Cell Carcinoma (KIRC) and Bladder Urothelial Carcinoma (BLCA), ¹⁵ , ¹⁶ we found that PRAD have only three common markers with them (Figure 3B). This also indicated that the CAFs gene signature of genitourinary cancers are different and specific.

FIGURE 3.

Comparison of PRAD‐specific cancer‐associated fibroblast (CAF) gene signature with other CAF Signatures: (A) Correlation analysis between PRAD‐specific CAF signature and established CAF markers; “X” indicates FDR ≥0.05. (B) Comparative analysis of distinct CAF signatures in PRAD, KIRC, and BLCA. (C) Functional enrichment analysis for the cyan module in fibroblasts. (D) Functional enrichment analysis for the green and salmon modules in myofibroblasts. (E) Comparative functional enrichment analysis between fibroblasts and myofibroblasts using scRNA‐seq data. (F) Differential functional enrichment analysis between fibroblasts and other cell types in scRNA‐seq data.

3.8. Functions of the PRAD cancer‐associated fibroblast gene signature

Following the identification of the CAF gene signature, we used Metascape (metascape.org) to validate the functions of these specific genes. Initially, we analyzed the fibroblast‐associated Cyan module, finding it closely linked to the extracellular matrix (Figure 3C). The Green and Salmon modules, associated with myofibroblasts, showed relevance to actin filament‐based processes and muscle structure development (Figure 3D).

Subsequently, in the scRNA‐seq analysis of PCa, we examined the functional differences between fibroblasts and myofibroblasts, as well as between fibroblasts and other cell types. In both cases from GSE157703, fibroblasts demonstrated a notable advantage over myofibroblasts in functions like antigen processing and presentation and the nod‐like receptor signaling pathway (Figure 3E). In comparing fibroblasts with other cell types, the ECM receptor interaction emerged as the only consistently enriched function across both patients (Figure 3F).

3.9. Clinical significance of the key genes

Our previous findings indicated that the proportion of CAFs in PRAD pathological tissues increases with tumor progression (Figure 1A). Additionally, the Cyan module, representing CAFs, showed significant correlation with both stage and Gleason score grade in PRAD (Figure 1C). Furthermore, the 14 genes comprising the PRAD CAF gene signature demonstrated notable associations with clinicopathological parameters and prognosis. In the ROC analysis of these key genes, the area under the curve (AUC) for progressive tumors (Gleason score grade 4/5 or stage 3/4) compared to non‐progressive tumors (Gleason score grade 1/2/3 or stage 2) was predominantly greater than 0.6 (Figure S2A).

3.10. Cancer‐associated fibroblasts and related clinicopathological parameters

To assess the clinical relevance of CAFs in PRAD, we used the 14‐gene signature of the PRAD CAF to calculate CAF infiltration scores for each sample via ssGSEA. We then determined the CAF infiltration levels in 486 TCGA PRAD samples using these ssGSEA scores (Figure S2B). The ssGSEA CAF scores were categorized based on tumor stage and Gleason score grade, revealing a correlation between increased CAF infiltration, higher tumor stages, and elevated Gleason score grades (Figure 4A).

FIGURE 4.

Clinical relevance of 14 key genes in cancer‐associated fibroblast (CAF) signature: (A) Association of CAF infiltration with advanced stage and higher Gleason score in TCGA‐PRAD. (B) Correlation of CAF infiltration with higher Gleason score in GSE134051. (C) Association of CAF infiltration with advanced stage and higher Gleason score in GSE21034. (D) Correlation of CAF infiltration with Gleason score in GSE116918. (E) Relationship between CAF infiltration and tumor stage in GSE116918. (F) Association of CAF infiltration with Gleason scores, and prostate specific antigen (PSA) levels in GSE116918. (G) Analysis of CAF infiltration in relation to risk grading in GSE116918. (H) Link between CAF infiltration and castration‐resistant prostate cancer in GSE32269 and GSE70770. ns, not significance, *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

For data validation in the GSE134051 and GSE21034 series, we applied the same gene signature to score CAF infiltration (Figure S2C). In GSE134051, high CAF infiltration was associated with Gleason scores 8/9/10 (Figure 4B). In GSE21034, high CAF infiltration correlated with Gleason scores 8/9 and T3/4 stages (Figure 4C). Additionally, the CAF infiltration level was linked to the prognosis of PCa and progression to CRPC. This gene signature was also used to score CAF infiltration in GSE116918, GSE32269, and GSE70770 (Figure S2D). Clinicopathologically, a higher fibroblast count in GSE116918 was associated with Gleason scores 8/9/10, T3/4 stages, and high PSA levels (Figure 4D–F). According to the Contemporary Prostate Cancer Grading System, ³⁸ the high‐risk group (GS8/9/10 or T3/4 or PSA >20 ng/mL) had a higher CAF infiltration level than the low‐risk group (GS6 and T1 and PSA <10 ng/mL) (Figure 4G). As T2 stage was not subdivided into T2a, T2b, and T2c, these cases were excluded from the comparison. In GSE32269 and GSE70770, higher fibroblast counts were associated with bone metastases in CRPC and CRPC groups, respectively (Figure 4H).

3.11. Cancer‐associated fibroblast‐marker unaffected by androgen deprivation therapy

In the study, we analyzed two PCa scRNA‐seq cases (PCa1 and PCa2) from GSE157703, which were samples not treated with ADT. Additionally, we obtained two PCa samples (PCa3 and PCa4) treated with ADT for 1 month, aiming to compare CAFs and investigate whether ADT influenced them. We merged and clustered the cells from these two datasets separately (Figure 5A). Using established cell subset markers (Figure 5B), we identified and defined the clustered cell subsets (Figure 5C). In the scRNA‐seq data (PCa1 and PCa2) from PCa without ADT treatment, the undefined cluster 3 cell subset was hypothesized to be inactive or dead cells (Figure 5C). Enrichment analysis further validated the cell subtype, indicating that fibroblasts retained their enrichment function in epithelial‐mesenchymal transition (EMT) both before and after ADT treatment (Figure 5D). However, following ADT, there was a significant increase in fibroblast enrichment within the hallmark pancreas beta cells and hallmark myogenesis gene. This suggests that ADT might selectively influence specific functions and gene expressions of fibroblasts, particularly in pathways related to pancreatic beta cell function and myogenesis.

FIGURE 5.

Cancer‐associated fibroblast (CAF) signature unaffected by androgen deprivation therapy (ADT). (A) Cell cluster feature plot comparing PCa1&PCa2 non‐ADT group with PCa3&PCa4 ADT group. (B) Bubble plot illustrating cell clusters and their respective markers. (C) Cell cluster feature plot of specified cell types. (D) Enrichment analysis of the defined cell types.

3.12. High signal outgoing cancer‐associated fibroblasts

To explore the signaling communication of CAFs, we used Cellchat to calculate potential ligand‐receptor pairs. Initially, we used a circle plot to quantify and assess the interactions among different cell subsets. This analysis revealed increased interactions between fibroblasts and other cell types in PCa. In non‐ADT samples, cancer cells and endothelial cells, alongside fibroblasts, were notably active in cellular interactions, as illustrated in Figure 6A. Conversely, in ADT samples, fibroblasts significantly surpassed other cells in terms of interaction, as shown in Figure 6B. A scatterplot analysis further indicated that fibroblasts exhibited the highest signal output in cellular interactions pre‐ADT (Figure 6C). Post‐ADT, despite fibroblasts maintaining high signal output, other cells previously characterized by high‐output signals were markedly affected (Figure 6D). Additionally, we observed a substantial increase in the input signal of monocytes post‐ADT (Figure 6C,D). Correspondingly, the notable reduction in the output signal of PCa cells post‐ADT aligns with the observed decline or demise of these cells following treatment (Figure 6C,D).

FIGURE 6.

Cancer‐associated fibroblasts (CAFs) exhibit high signal output. (A) CellChat circle plot showing signal exchange among diverse cells in non‐androgen deprivation therapy (ADT) samples. (B) CellChat circle plot of signal exchange in diverse cells within ADT‐treated samples. (C) Scatterplot comparing incoming and outgoing signals in various cells in non‐ADT samples. (D) Scatterplot of incoming and outgoing signals in cells from ADT‐treated samples. (E) Bubble plot illustrating communication pathways in secreting and target cells in non‐ADT samples. The orange circle highlights unique outgoing communication patterns in fibroblasts/CAFs in non‐ADT samples. (F) Bubble plot showing communication pathways in secreting and target cells in ADT samples. An orange circle marks the distinctive outgoing communication patterns in fibroblasts/CAFs in ADT samples.

3.13. Dominant outgoing signals of cancer‐associated fibroblasts: Significant changes

Fibroblasts, also known as CAFs, consistently exhibited the highest signal output post‐ADT. Despite non‐ADT PCa samples presenting fewer cell numbers and enrichment pathways, we identified notable changes in CAF enrichment pathways in ADT samples, focusing on their outgoing signals. In non‐ADT samples, the ANGPT, CCL, PDGF, PROS, and TWEAK pathways were exclusive to CAFs’ outgoing signals (Figure 6E). These pathways predominantly enriched myofibroblasts and endothelial cells in ADT samples (Figure 6F). Notably, while CAFs in ADT samples showed an increased number of cells and more enrichment pathways, their enrichment pathways differed significantly from those in non‐ADT samples. Specifically, the output signals of CAFs showed significant enrichment in the ACTIVIN, ANGPTL, CALCR, CXCL, GDF, HGF, IL6, LIFR, MIF, ncWNT, NRG, PERIOSTIN, SEMA3, and VEGF pathways in ADT samples (Figure 6F). Contrastingly, these pathways were mainly associated with signal output by basal/intermediate cells, luminal cells, and monocytes in non‐ADT samples (Figure 6E). In summary, ADT significantly altered the output signals of CAFs, which maintained high cellular activity. It appears that CAFs might play a crucial role in stabilizing the cellular environment, compensating for the loss of cancer cells and monocytes to ensure sustained signal output.

4. DISCUSSION

Our study identified PCa‐specific gene signature of CAFs using WGCNA and confirmed the specificity of the gene signature based on evidence at the RNA and protein levels. Similarly, our findings reaffirmed the advantage of WGCNA in identifying tumor‐specific CAF markers. Using ssGSEA for precise quantification of CAF infiltration levels via the gene signature, our analysis revealed a significant association between elevated CAF infiltration and tumor progression and CRPC. ADT did not alter the CAF signature, with CAFs continuing to exhibit the highest signal output post‐ADT. These insights establish a significant foundation for future studies of CAFs in PCa.

Identifying the specific signature of CAFs in PCa is crucial due to the high heterogeneity among fibroblasts. For instance, scRNA‐seq analysis of fibroblasts from various tissues (heart, skeletal muscle, intestine, and bladder) in normal adult mice revealed less than a 20% overlap in their transcriptomes. ³⁹ To date, there has been limited progress in pinpointing specific markers for PCa fibroblasts, which constrains our understanding of their roles, localization, and distribution. ⁴⁰ Consequently, it is essential for future research to thoroughly investigate fibroblast populations in PCa and enhance our molecular understanding of their regulation and function in the disease.

The efficacy of WGCNA in identifying CAF‐specific markers has been reaffirmed. We previously used WGCNA to discern CAF‐specific tags in bladder and kidney cancers. ¹⁵ , ¹⁶ WGCNA identifies clusters (modules) of closely related genes and characterizes these clusters through module trait genes or intra‐module hub genes. By correlating modules with each other and external sample traits, and calculating module membership, WGCNA facilitates the identification of potential biomarkers or therapeutic targets. ²¹ The gene clusters identified by WGCNA can be interpreted as sets of genes highly expressed in specific cell types. Their expression is predominantly influenced by the proportion of such cells and is minimally affected by other cell types. With a substantial number of samples, the clearer the co‐expression significance, the more likely it is that these gene clusters represent specific expressions in certain cell types. Integrating this with gene membership degree enables the screening of markers specifically expressed by certain cells (Figure 1E).

The effects and status of CAFs following ADT treatment in PCa remained uncertain. We compared non‐ADT samples with those 1‐month post‐ADT and observed that the specific signature and highest signal output state of CAFs were unchanged. However, most studies on PCa and CAFs have been limited to addressing treatment resistance or CRPC. ¹¹ , ⁴¹ The fundamental treatment for advanced PCa involves gonadal testosterone deprivation, which also encompasses inhibiting androgen production by the adrenal glands and other extragonadal sources, and directly targeting androgen receptors (ARs). ⁴² Hormone‐sensitive PCa typically results in rapid cancer cell death, and changes in CAFs within the TME exhibit multiple subsets.

Our study successfully identified markers for PCa‐specific CAFs, providing a foundation for studying CAFs in PCa and drug resistance following ADT. Nevertheless, our research has certain limitations. First, the identification of CAF‐specific markers relies on a biocomputational method previously validated only for kidney and bladder cancers, lacking verification in other cancer types. Second, the scRNA‐seq samples were limited. This was partly due to the scarcity of post‐ADT samples and some scRNA‐seq samples being compromised by cold chain transport delays during the pandemic. Third, this study did not validate variational signals in CAFs. Since this aspect requires extensive investigation, it will be addressed in our future work.

In summary, our study successfully identified a specific signature for CAFs within PCa. We demonstrated that CAFs are associated with PCa progression, higher Gleason scores, and CRPC. The significance of CAFs in PCa was further highlighted by their sustained high signal output following ADT and the stable gene signature. Our findings established a foundation for CAF research in PCa and direct future investigations.

AUTHOR CONTRIBUTIONS

Shen Pan: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; project administration; resources; software; supervision; validation; visualization; writing – original draft; writing – review and editing. Rui Yin: Data curation; formal analysis; writing – original draft. Hehe Zhu: Data curation; formal analysis; writing – original draft. Siang Shen: Data curation; formal analysis; writing – original draft. Zhenhua Li: Writing – review and editing. Bitian Liu: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; project administration; resources; software; supervision; validation; visualization; writing – original draft; writing – review and editing.

FUNDING INFORMATION

This work was supported by grants from Japan China Sasakawa Medical Fellowship, Shengjing Hospital 345 Talent Project (Grant No. M1357 and M1358) and Liaoning Provincial Science and Technology Joint Applied Basic Research Project (Grant No. 2022JH2/101300073 and 2023JH2/101700141).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

ETHICS STATEMENTS

Approval of the research protocol by an Institutional Review Board: The study protocol received approval from the Investigation Ethical Committee of Shengjing Hospital of China Medical University.

Informed Consent: Informed consent was obtained from all participants or their legal guardians.

Registry and Registration No. of the Study/trial: The study was officially registered and approved by the Investigation Ethical Committee of Shengjing Hospital of China Medical University, bearing the registration number 2023PS1419K.

Animal Studies: N/A.

Supporting information

Figure S1.

Figure S2.

Table S1.

Pan S, Yin R, Zhu H, Shen S, Li Z, Liu B. Prostate cancer cancer‐associated fibroblasts with stable markers post‐androgen deprivation therapy associated with tumor progression and castration resistant prostate cancer. Cancer Sci. 2024;115:2893‐2907. doi: 10.1111/cas.16267

DATA AVAILABILITY STATEMENT

Data from this study we collected are publicly accessible in the GEO repository: GSE250189 for single‐cell data.