A novel six-biomarker panel identified from male breast cancer-associated fibroblasts demonstrates prognostic power for prostate tumors

Abstract

Background: Male breast cancer (BC) is rare, accounting for only approximately 1% of all cases of BC, and poorly characterized. In contrast, prostate cancer is the most prevalent cancer in men and serves as a model for understanding male-specific tumor biology. The advent of high-throughput technologies has enabled the development of gene expression signatures for both breast and prostate tumors that could inform prognosis and guide treatment. In this respect, the role of the tumor microenvironment, particularly cancer-associated fibroblasts (CAFs), remains largely underexplored. Here, we sought to identify a CAF-related gene signature in male patients with BC and prostate cancer to reveal specific protumorigenic mechanisms and identify novel therapeutic targets for both malignancies.

Methods: RNA sequencing was performed to analyze and compare the transcriptomes of CAFs isolated from female and male BC patients. Differentially expressed genes (DEGs) between female and male breast CAFs were identified and subjected to enrichment analyses. Using a set of candidate upregulated genes in male breast CAFs, K-means clustering of prostate cancer patients was performed using multiple datasets to define a prognostic gene signature. Kaplan‒Meier curves and log-rank tests were conducted to assess differences in patient outcomes and other clinical variables between groups of patients with high or low prognostic gene expression. The clustering results were then validated using decision tree analysis, and boosted calculations were employed to increase the classifier performance.

Results: Transcriptomic profiling revealed 775 DEGs between female and male breast CAFs. Owing to the limited transcriptomic data from male BC patients, we leveraged large prostate cancer cohorts to investigate the relevance of the genes expressed by male breast CAFs. A six-gene signature (ASPN, COL4A1, COL4A2, COL5A3, COMP and FN1) that could predict patient outcomes in multiple independent cohorts of prostate cancer patients was identified.

Conclusions: We identified a novel gene signature with strong prognostic value in prostate cancer and potential relevance to male BC. This gene signature represents a complementary tool to standard clinical parameters for improving patient stratification and management.

Keywords: Cancer-associated fibroblasts (CAFs); Gene signature; K-means; Male breast cancer; Prostate cancer; RNA-sequencing.

Fig. 1

Gene expression landscapes of CAFs from surgically resected female and male breast carcinomas. Immunofluorescence staining for α smooth muscle actin (α-SMA, green signal) in female (a) and male (b) breast CAFs. Nuclei were stained with DAPI (blue signal). Scale bar: 75 μm. The images shown represent 10 random fields from three independent experiments. Venn diagram (c), volcano plots (d) and heatmaps (e) showing the differentially expressed genes between CAFs isolated from male and female breast cancer tissues, as identified by RNA-sequencing; log₂₋fold change (log2FC) ≥ |2|, p ≤ 0.01

Fig. 2

Reactome pathway and Gene Ontology (GO) enrichment analyses of the genes upregulated in female breast CAFs compared with male breast CAFs. Bar plots showing the top five Reactome pathways (a) and GO terms, such as biological process (b), molecular function (c) and cellular component (d), of the genes upregulated in female breast CAFs compared with male breast CAFs; p < 0.05

Fig. 3

Enrichment analyses of the genes upregulated in male breast CAFs compared with female breast CAFs. Bar plots showing the top five Reactome pathways (a) and GO terms associated with the biological process (b), molecular function (c) and cellular component (d) categories of the genes upregulated in male breast CAFs compared with female breast CAFs; p < 0.05

Fig. 4

Identification of genes enriched in overlapping pathways and Gene Ontology (GO) terms in male breast CAFs. (a) Venn diagram showing the number of genes upregulated in CAFs from male breast tumors enriched in overlapping Reactome pathways and GO categories. (b) Heatmap showing the enrichment of the 12 genes in diverse pathways and biological process (BP), molecular function (MF) and cellular component (CC) GO terms; blue denotes that a gene is enriched in a pathway or GO term, and gray denotes that it is not. (c) Heatmap showing the differential expression of the 12 shared genes between male and female breast CAFs. (d) Bar plot showing the log₂₋fold change (log2FC) in the expression of the 12 genes in male breast CAFs compared with female breast CAFs

Fig. 5

Clustering of prostate cancer patients included in the TCGA for identification of a prognostic gene signature. The optimal number of K-means clusters was determined and visualized using the within-cluster sums of squares (a) and average silhouette (b) methods. (c) Visualization of the K-means partitioning results; individual prostate cancer patients in the TCGA dataset are represented by points in the principal component plot. Each data point in the reduced-dimensional space is color coded according to according to the assigned cluster, and the number of patients belonging to each cluster is indicated. (d) Boxplots displaying the expression of the 9 genes in the two clusters of patients. (e) Disease-free interval (DFI) of prostate cancer patients according to their assigned cluster. The number of patients in each cluster is shown in the panels. **** indicates p < 0.0001

Fig. 6

Association of the gene signature with pathological stage and Gleason score in prostate cancer patients. Stacked bar charts illustrating the percentages of TCGA prostate cancer patients in clusters 1 and 2 (C1 and C2) with particular pathological stages (a) and Gleason scores (c). Boxplots showing the expression levels of the 9 genes used for K-means analysis in TCGA prostate cancer patients according to pathological stage (b) and Gleason score (d). Significance levels are indicated as follows: **** p < 0.0001; *** p < 0.001; ** p < 0.01; * p < 0.05; “ns” denotes nonsignificant differences

Fig. 7

Correlation of the gene signature with disease recurrence and stromal infiltration in prostate cancer patients. Stacked bar charts showing the percentages of TCGA prostate cancer patients in clusters 1 and 2 (C1 and C2) with residual tumors (a) and new tumor events after treatment (c) and particular stromal scores (e). Boxplots depicting the expression levels of the 9 genes used for K-means analysis in prostate cancer patients with residual tumors (b), new tumor events after treatment (d) and particular stromal scores (f). Significance levels are indicated as follows: **** p < 0.0001; *** p < 0.001; ** p < 0.01; * p < 0.05; “ns” denotes nonsignificant results

Fig. 8

Assessment of K-means clustering results for prostate cancer patients included in the TCGA using decision tree classification analysis. (a) Representative decision tree for the training dataset generated through the boosting algorithm on the basis of K-means clustering. (b) Confusion matrix providing a quantitative evaluation of model performance for the test set. The rows of the matrix correspond to the actual classes, whereas the columns represent the predicted classes; each cell shows the number of instances falling into a particular category. (c) Bar plot showing the attribute usage percentage of the 9 genes according to the boosting algorithm for the training dataset (10 trials)

Fig. 9

K-means clustering of prostate cancer patients in the DKFZ dataset. The optimal number of K-means clusters was determined and visualized using within-cluster sums of squares (a) and average silhouette (b) methods. (c) Visualization of the K-means partitioning results; individual prostate cancer patients in the DKFZ dataset are represented by points in the principal component plot. Each data point in the reduced-dimensional space is color coded according to its assigned cluster, and the number of patients belonging to each cluster is shown. (d) Boxplots showing the differential expression of the 6 genes in the two clusters of patients. (e) Biochemical recurrence (BCR) of prostate cancer patients according to their assigned cluster. The number of patients in each cluster is shown in the panels. Stacked bar charts illustrating the percentages of patients belonging to cluster 1 (C1) and cluster 2 (C2) with particular Gleason scores (f) and stages (g). Significance levels are indicated as follows: **** p < 0.0001; *** p < 0.001; ** p < 0.01

Fig. 10

Unsupervised K-means clustering of prostate cancer patients in the GSE54460 dataset. The optimal number of K-means clusters was determined and visualized using within-cluster sums of squares (a) and average silhouette (b) methods. (c) Visualization of the K-means partitioning results; individual prostate cancer patients in the GSE54460 dataset are represented by points in the principal component plot. Each data point in the reduced-dimensional space is color coded according to its assigned cluster, and the number of patients belonging to each cluster is shown. (d) Boxplots showing the differential expression of the 6 genes in the two clusters of patients. (e) Biochemical recurrence (BCR) of prostate cancer patients according to their assigned cluster. The number of patients in each cluster is shown in the panels. Significance levels are indicated as follows: **** p < 0.0001; *** p < 0.001

Fig. 11

Identification of prostate cancer patient subgroups in the MSK cohort using K-means clustering The optimal number of K-means clusters was determined and visualized using within-cluster sums of squares (a) and average silhouette (b) methods. (c) Visualization of the K-means partitioning results; individual prostate cancer patients in the MSK dataset are represented by points in the principal component plot. Each data point in the reduced-dimensional space is color coded according to its assigned cluster, and the number of patients belonging to each cluster is shown. (d) Boxplots showing the differential expression of the 6 genes in the two clusters of patients. (e) Disease-free survival (DFS) of prostate cancer patients according to their assigned cluster. The number of patients in each cluster is shown in the panels. Significance levels are indicated as follows: **** p < 0.0001; ** p < 0.01; * p < 0.05