The publicly available application TNM plot (https://tnmplot.com/analysis/) was used to evaluate the mRNA expression levels of the genes of interest in normal (n = 106), tumour (n = 283) and metastatic (n = 6) samples (Bartha and Gyorffy, 2021). For statistical significance, the Kruskal–Wallis test was used to compare the gene expression levels among normal, tumour and metastatic tissue with p < 0.01 used as a cut-off value for statistical significance. In addition, using cBioPortal (cbioportal.org), we retrieved and analysed mRNA data (RNA SeqV2) of 493 patients with prostate adenocarcinoma from the Cancer Genome Atlas (TGCA) and Pan-Cancer Atlas. Log-rank test p-values were used to evaluate the statistical significance between the low and the high expression level (classified according to above and below mean mRNA expression of each biomarker. The Chi-squared test was used to evaluate the association between each biomarker and clinicopathological parameters including tumour stage as well as patient outcome. Moreover, the DNA methylation profiles retrieved from the TCGA database were assessed in order to evaluate the DKK3 promoter DNA methylation levels using the UALCAN portal. DNA methylation analysis was carried out using normalized beta values from 502 PCa patients and 50 healthy controls considering beta value cut-offs of hyper-methylation [beta value: 0.7–0.5] or hypo-methylation [beta-value: 0.3–0.25].

To evaluate the prognostic capacity of each marker of interest, we used the CANCERTOOL database, (http://web.bioinformatics.cicbiogune.es/CANCERTOOL), which is a comprehensive portal that aims to investigate different genes and their association with clinical data including disease progression, pathologic, and molecular characteristics, to evaluate the expression levels of our markers in normal versus malignant tissues.

The relationship between DKK3, TGFB1 and ECM1 and immune cell infiltration in the in PCa microenvironment was explored using the Tumour IMmune Estimation Resource (TIMER) (https://cistrome.shinyapps.io/timer/), which consists of deconvolution and comprehensive analysis of tumour-infiltrating immune cells in various types of cancer. The association between DKK3, TGFB1 and ECM1 expression and the abundances of immune infiltrates including B cells, CD4⁺ T cells, CD8⁺ T cells, neutrophils, macrophages, and dendritic cells in the PCa tumour microenvironment was assessed, using Spearman’s test (p < 0.05).

To assess the functional association between the studied genes DKK3, TGFB1 and ECM1 and their related connected genes, the gene-gene interaction network was constructed based on a large data of functional-related data sets including genetic and protein interactions, physical interaction, co-expression, co-localization, and common pathway, using GeneMANIA bioinformatics tool with defaults parameters (Warde-Farley et al., 2010). Genes can be linked by the interacted network based on variable attributes. The network includes nodes that represent genes while edges represent connections. Besides, functional enrichment analysis on Gene Ontology terms including biological process (BP), cellular component (CC), and molecular function (MF) categories was conducted using the Enrichr database (Xie et al., 2021) to explore the biological significance and common pathways between the DKK3, TGFB1 and ECM1 genes. The hypergeometric distribution cut-off for the functional enrichment analyses was a p-value of <0.05.

To improve our understanding of the role of DKK3, TGFB1 and ECM-1 genes in tumourigenesis, we initially investigated the expression levels of DKK3 in PCa samples compared to their normal counterparts using the CANCERTOOL database. Different datasets were used including Glinksy et al., Grasso et al., Lapointe et al., Taylor et al., TCGA (RNA-seq), Tomlins et al., and Varambally et al. (Figure 1). Our results showed a significant lower level of DKK3 in PCa samples compared to samples obtained from their normal counterparts. Similarly, TGFB1 showed also significant downregulation in PCa samples compared to normal counterparts. In contrast, ECM-1 levels showed no significant differences between the normal and PCa samples (Supplementary Figure S1).

Moreover, the evaluation of DNA methylation levels in the promoter region of DKK3 using TCGA publicly available methylome data showed significant DNA methylation differences between primary PCa and normal samples (p < 1.62 E-12). Indeed, the DKK3 promoter was highly hypermethylated among PCa cases (n = 502), while it was hypomethylated in the normal samples (n = 50). Our findings indicate that there is a notable decrease in the expression of DKK3 in PCa samples from various datasets, indicating that the promoter region of the DKK3 gene may be affected by DNA methylation, leading to its silencing (Figure 2).

To improve our understanding of the role of DKK3, ECM-1 and TGF-β in PCa tumourigenesis, we evaluated their expression in samples representing different stages of PCa progression (Figure 3). To achieve this, we analysed DKK3, ECM-1 and TGFB1 expression in normal prostate, primary tumours and samples obtained from more advanced diseases presented as patients with tumour metastases. Interestingly, our results showed a gradual and significant downregulation of DKK3 expression during tumour progression. Indeed, the expression levels of DKK3 showed significant downregulation in tumour samples compared to normal. Further downregulation was observed between tumour samples and samples obtained from patients with metastatic tumours (p < 0.0001). In contrast, while TGFB1 expression was comparable between normal and tumour samples, metastatic lesions showed a significantly higher level of TGFB1, compared to both primary tumour and normal samples (p = 0.04). Moreover, while ECM1 expression levels were higher in tumour samples compared to normal tissue, ECM1 expression was significantly lower in metastatic samples compared to tumour, as well as normal samples (p < 0.0001). As a reference, we also investigated the expression levels of the classical marker PSA (KLK3) in the same samples. Our findings showed that while PSA levels initially increased in the tumour samples compared to normal, those levels were significantly lower in metastatic samples, compared to tumours (p < 0.0001). In summary, our results highlight the possible benefits of DKK3 expression as a marker of PCa progression, owing to its gradual loss during tumourigenesis.

Next, we evaluated the mRNA expression of the three genes in association with the tumour stage using the prostate adenocarcinoma (TCGA, PanCancer Atlas) cohort (Figure 4). Our results showed a marginal significance (p = 0.082) between DKK3 mRNA expression and LN involvement (Figure 4A). Patients with higher DKK3 mRNA expression showed less chance of positive lymph node involvement (N1) compared to patients with lower DKK3 mRNA expression. In comparison, while ECM1 showed no association with tumour stage (p = 0.773), TGFB1 showed a significant association with tumour stage (Figure 4B). Patients with high TGFB1 mRNA expression presented with the more advanced T3A (tumour spread outside the prostate, but not invading seminal vesicles) and T3B (tumour spread outside prostate with seminal vesicles invasion) stages, compared to patients with low TGFB1 mRNA expression (p < 0.001).

Interestingly, while patients with high DKK3 mRNA expression showed marginally significant lower chance of having recurrence presented as new neoplastic events post initial therapy (p = 0.082), Patients with higher TGFB1 mRNA expression showed a significant higher chance of recurrence presented as new neoplastic events post initial therapy compared with patients with low TGFB1 mRNA expression (p < 0.001). In contrast, ECM1 mRNA expression showed no significant association with new neoplastic events post initial therapy (p = 0.932) (Figure 4).

The analysis between the mRNA expression of the three markers with tumor grade showed that while both DKK3 and ECM1 showed no significant variation in samples from different Gleason grade, TGFB1 mRNA expression was significantly higher in patients presented with higher gleason score (G8,9&10) in the TCGA dataset (p = 0.0001) (Supplementary Figure S2).

Further analysis was performed to investigate the correlation between DKK3 expression and TGFB1 and ECM1 in prostate cancer samples. Indeed, this might better demonstrate the interplay between DKK3 and these markers. While our analysis showed no significant correlation between ECM1 and DKK3 mRNA levels (R = 0.07, p = 0.23), our results showed a significant and negative correlation between DKK3 mRNA expression levels and TGFB1 (R = -0.17, p < 0.0001). Additional analysis was performed to investigate the association between DKK3 and KLK3, which is most relevant marker for PCa diagnosis and management, and its expression is regulated by the upstream markers, such as AR. Our analysis showed no significant correlation between DKK3 mRNA expression levels and KLK3 expression (R = -0.09, p = 0.14) (Supplementary Figure S3).

Next, we evaluated the association between our markers and patient outcome presented as progression-free survival (Figure 5). While ECM-1 showed no significant association with patient outcome progression-free survival (p = 0.986), disease specific (p = 0.776) and disease free survival (p = 0.167). Both DKK3 and TGFB1 showed a significant and contrasting association with patient outcomes. While patients with high DKK3 mRNA expression showed a significant association with favourable outcomes presented as prolonged progression-free survival (p = 0.0470), disease specific (p = 0.0266) and disease free survival (p = 0.050), high TGFB1 mRNA expression showed a significant association with poor patient outcome presented as shortened progression-free survival (p = 3.267e-4) and disease free survival (p = 0.0433).

We were also interested to analyze the relationship between DKK3, TGFB1 and ECM1 and immune cell infiltration in prostate cancer (Figure 6). The results showed that gene expression level against tumor purity is highly significant for the three biomarkers with negative Spearman’s rho values (−0.5, −0.329 and −0.41), respectively in DKK3, TGFB1 and ECM1 suggesting that these biomarkers are highly expressed in the microenvironment of PCa (Figure 6). Moreover, all biomarkers have a positive significant correlation with B cells, CD4⁺ T cells, CD8⁺ T cells, Neutrophils, Macrophages, and Dendritic cells. Importantly, DKK3 has the highest correlation with macrophage at 60%, while ECM1 showed the highest correlation at about 54% with dendritic cells and TGFB1 has the highest correlation with neutrophil and dendritic cells at 60% and 68% respectively. Importantly, all biomarkers had a better correlation with CD4 than CD8.

To explore the potential mechanisms that involve DKK3, TGFB1 and ECM1 in the carcinogenesis of PCa, we explored GeneMania to build up a gene–gene interaction network for these biomarkers. As shown in Figure 7, the network revealed that DKK3, TGFB1 and ECM1 share many genes within the same gene family or similar protein domain under a series of cooperators suggesting that there is a possibility of the interplay between the 3 markers through interaction with proteins from a similar family. For instance, MMP9 presents a key marker in the network that showed physical interconnection with TGFB1 and ECM1, and co-expression with DKK3 suggesting the potential direct/indirect connections between these particular biomarkers. Knowing that MMP9 codes for a matrix metalloproteinase protein that is reported to promote metastasis and angiogenesis through decomposition of the extracellular matrix in several tumours which further supports the possible involvement of DKK3, TGFB1 and ECM1 in the PCa progression.

Furthermore, gene ontologies including biological process and molecular function were performed to reveal the functional enrichment of DKK3, TGFB1 and ECM1 and their co-operators. The significant annotated pathways were represented in Figure 7. Interestingly, it identifies cytokine activity (GO:0005125, p-value = 0.02) and macrophage-related pathways such as negative regulation of macrophage cytokine production (GO:0010936, p-value = 7.5E-04) among the differentially regulated pathways involving mainly TGFB1 as a key player which further supports the findings of immune cell infiltration in PCa (Figure 8). Additionally, the regulation of canonical Wnt signaling pathway (GO:0060828, p-value = 4.74E-04) was identified among the relevant significantly regulated biological processes involving TGFB1 and DKK3, which could be of a pivotal role in PCa pathogenesis, proliferation and resistance to treatment. Taken together, these findings highlight the potential utility of DKK3, TGFB1 and ECM1as prognostic markers for PCa.