Potential role of DKK3 and WIF1 in prostate cancer: bioinformatics and clinical analysis

Abstract

Background

Prostate Cancer (PCa) is one of the most common malignant tumors in men. Some patients may progress to metastatic and Castration-Resistant Prostate Cancer (CRPC), leading to increased treatment difficulty and poor prognosis. With advancements in bioinformatics and machine learning technologies, the integration of multiple databases can efficiently identify core genes associated with the occurrence of PCa. Additionally, the Tumor Microenvironment (TME) and immune cell infiltration play crucial roles in the development and therapeutic response of PCa. Investigating the interactions between core genes and the immune microenvironment will enhance the understanding of the molecular mechanisms underlying PCa and provide new avenues for precision treatment.

Methods

This study obtained gene expression data of prostate cancer and normal tissues from the GEO (GSE69223, GSE46602) and TCGA (TCGA-PRAD) databases. Initially, each dataset was standardized, batch effects were corrected, and differentially expressed genes (DEGs) were identified, followed by GO and KEGG enrichment analyses. Subsequently, the STRING database was utilized to construct a protein-protein interaction (PPI) network, and core modules were selected using Cytoscape software. Lasso regression and Random Forest (RF) algorithms were employed to further pinpoint core genes from the key genes. In the external validation dataset (GSE46602), the diagnostic value and expression level differences of these core genes were assessed. Combining data from TCGA-PRAD, their correlations with immune cell infiltration were analyzed. The application value in molecular pathways and potential therapeutic interventions was explored through GSEA and CTRP drug sensitivity data. Finally, immunohistochemistry and Western blot experiments were conducted to confirm the expression changes of core genes in Benign Prostatic Hyperplasia (BPH) and prostate cancer tissues, alongside correlation analyses with clinical pathological parameters.

Results

Multi-omics integrative analysis identified a total of 339 common DEGs. The PPI network and machine learning algorithms further identified four potential core genes: DKK3, SNAI2, WIF1, and FOXA1. External validation and diagnostic value assessments revealed that DKK3 and WIF1 were significantly downregulated in PCa tissues and positively correlated with higher PSA levels and Gleason scores. Immune cell infiltration analysis indicated that the downregulation of these genes was closely associated with impaired adaptive immune function and matrix remodeling, and synergistically interacted with abnormalities in the Wnt/TGF-β pathways. Further analysis using the CTRP database showed that high expression of DKK3 increased cellular sensitivity to various anti-tumor drugs, whereas high expression of WIF1 might reduce the efficacy of certain small molecule inhibitors. Immunohistochemistry and protein level detection in clinical samples confirmed that the expression levels of DKK3 and WIF1 were significantly decreased in PCa tissues and were closely associated with patients’ PSA levels and Gleason scores.

Conclusion

Evidence from integrated databases and clinical samples indicates that WIF1 and DKK3 are significantly downregulated in PCa. Their inactivation may accelerate tumor progression and drug resistance by relieving the negative regulation of the Wnt/TGF-β pathways and affecting the immune microenvironment. Targeting the restoration or enhancement of these two tumor suppressor genes could become a new direction for the early diagnosis and precision treatment of prostate cancer. When combined with personalized medication selection, this approach is expected to significantly improve clinical outcomes for high-risk PCa patients.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-03488-x.

Introduction

Prostate Cancer (PCa) is one of the most common malignant tumors among men worldwide, with its incidence and mortality rates continuously rising [1]. Although Prostate-Specific Antigen (PSA) testing and imaging technologies play significant roles in PCa screening and staging assessments, substantial biological differences among patients mean that some cases may still progress to metastatic and Castration-Resistant Prostate Cancer (CRPC), significantly increasing treatment difficulty and leading to poorer prognosis [2, 3].

In recent years, the rapid development of bioinformatics and machine learning technologies has provided powerful tools for processing and analyzing gene expression data, enabling in-depth exploration of core genes associated with diseases [4–7]. By integrating data from multiple public databases (such as GEO and TCGA), this study utilizes Differentially Expressed Genes (DEGs) screening, functional enrichment analysis, and protein-protein interaction network construction to identify key genes closely related to the occurrence of PCa. Additionally, machine learning algorithms like Lasso regression and Random Forest (RF) are employed to further optimize the selection of core genes, enhancing the accuracy and stability of predictive models.

The Tumor Microenvironment (TME) and its immune cell infiltration play critical roles in the occurrence, development, and therapeutic response of PCa. Studies have shown that the types and quantities of immune cells are closely related to tumor aggressiveness, metastatic potential, and patient prognosis. Specifically, the state of immune cell infiltration not only affects the tumor’s immune evasion mechanisms but also has a significant impact on the efficacy of immunotherapy. Therefore, exploring the relationship between PCa core genes and immune cell infiltration not only helps to uncover the immune evasion mechanisms of PCa but also provides theoretical support for the development of novel immunotherapeutic strategies [8].

This study conducts differential gene analysis and functional enrichment on prostate cancer and normal tissue samples using public databases such as GEO and TCGA, identifies key genes, and employs machine learning algorithms to pinpoint core genes. Subsequently, through external dataset validation as well as immunohistochemistry and protein level detection, the study further confirms the downregulation trends of WIF1 and DKK3 in PCa and their associations with clinicopathological indicators. Based on immune infiltration analysis, we also reveal the potential interaction features between these genes and the tumor microenvironment, and utilize Gene Set Enrichment Analysis (GSEA) to deeply analyze their regulatory roles in the Wnt/TGF-β, ECM remodeling, and various metabolic pathways. Finally, by correlating drug sensitivity using the Cancer Therapeutics Response Portal (CTRP) database, the study provides important clues for selecting more targeted therapeutic strategies. By integrating multi-dimensional evidence and conducting functional validation of core genes with clinical samples, this research strives to offer more promising research ideas and potential intervention targets for the early diagnosis and precision treatment of prostate cancer from the perspectives of molecular mechanisms, immune microenvironment, and targeted therapy.

Materials and methods

Bioinformatics

Data sources

The data for this study were primarily obtained from the GEO (Gene Expression Omnibus) and TCGA (The Cancer Genome Atlas) databases. We downloaded the GSE69223 dataset (including 15 prostate cancer samples and 15 normal samples) and the GSE46602 dataset (including 36 prostate cancer samples and 14 normal samples) from GEO (https://www.ncbi.nlm.nih.gov/geo/). Additionally, STAR-counts data for TCGA-PRAD (including 126 prostate cancer samples and 52 normal samples) were downloaded from TCGA (https://portal.gdc.cancer.gov/).

Acquisition of differentially expressed genes (DEGs) in PCa

The raw, unnormalized data from the GSE69223 and GSE46602 datasets were log2-transformed and subjected to quantile normalization using the normalize.quantiles function [9, 10]. Probe IDs were converted to gene symbols based on platform annotation files, probes corresponding to multiple genes were excluded, and the average expression value was calculated for genes with multiple probes. If batch effects were present within the same dataset, they were corrected using the removeBatchEffect function from the limma package [11]. The STAR-counts data from TCGA-PRAD were converted to TPM (Transcripts Per Million) and log2(TPM + 1) normalized. Only samples with both RNA sequencing data and clinical information were included in the analysis [12]. Through these steps, standardized, batch-corrected, and well-annotated expression datasets were obtained. DEGs were identified from the GSE69223 and TCGA-PRAD datasets using the criteria P < 0.05 and |log2(FC)| >1. A differential expression matrix was created based on sample grouping, missing values were removed, and multiple testing corrections were performed using the Bayesian method. Volcano plots of DEGs were generated using the ggplot2 package, and heatmaps of the top 50 upregulated and top 50 downregulated genes were created using the pheatmap package.

GO function and KEGG pathway enrichment analysis

GO and KEGG analyses were performed on the intersection of DEGs from the GSE69223 and TCGA-PRAD datasets using DAVID (https://david.ncifcrf.gov/) [13].

Screening of core genes in PCa

The intersecting DEGs were imported into the STRING database (https://string-db.org/) to construct a protein-protein interaction (PPI) network. In Cytoscape (v3.10.0), the top 30 genes ranked by the Maximum Neighborhood Component (MNC) algorithm were selected as key genes in prostate cancer. The MCODE plugin was then used to identify key functional modules (module score ≥ 3) [14]. Subsequently, Lasso regression was employed for variable selection using the glmnet package in R to identify core genes. Additionally, a Random Forest (RF) algorithm was utilized to construct a model for classifying and predicting key genes, from which core genes were selected. The intersection of core genes identified by both machine learning algorithms was taken as the final set of core genes.

Validation of core gene expression and immune correlation analysis

To further validate the diagnostic value of the core genes in PCa, the processed GSE46602 dataset was used as an external validation set. The expression profiles of characteristic genes in prostate cancer and healthy controls were analyzed, and ROC curves were plotted using the pROC package in R for validation. Based on key functional modules, core genes related to the occurrence and development of PCa were selected for subsequent analysis. Additionally, violin plots, dot plots, and box plots were generated to analyze the expression levels of core genes in the GSE69223 training set and the GSE46602 validation set. Subsequently, the TCGA-PRAD dataset was used to analyze the correlation between the expression levels of core genes and the infiltration levels of different immune cells.

Gene set enrichment analysis (GSEA) of core genes and drug sensitivity analysis

Gene Set Enrichment Analysis (GSEA) is a commonly used bioinformatics method for identifying gene sets or pathways that are significantly enriched under different biological conditions based on gene expression data. The identified core genes were subjected to GSEA to perform enrichment analyses of GO pathways and KEGG pathways. The analysis involved 10,000 permutations, with a minimum gene set size of 10 genes and a maximum gene set size of 200 genes. Significant enrichment results were determined using the criteria of p < 0.05 and FDR < 0.05. To explore potential targeted drugs for the core genes, the correlation between drug sensitivity and gene expression was analyzed using the CTRP drug sensitivity data available in the GSCA database (http://bioinfo.life.hust.edu.cn/GSCA/#/), thereby assessing the relationship between core genes and drug sensitivity.

Experimental validation

Clinical specimens, ethical approval, and antibodies

This study included tissue specimens from 40 patients with Prostate Cancer (PCa) and 40 patients with Benign Prostatic Hyperplasia (BPH) who underwent prostate surgery at our hospital from June 2023 to May 2024. Histological diagnoses were independently confirmed by two senior pathologists. PCa patients had not received radiotherapy, chemotherapy, or novel androgen deprivation therapy prior to surgery. This study was approved by the hospital’s Ethics Committee (Ethics Approval Number: NO.202435) and was conducted in accordance with the Declaration of Helsinki [15]. All patients provided informed consent, and personal information was anonymized. The primary antibodies used in this study included an anti-DKK3 polyclonal antibody (Cat. No. 10365-1-AP, Sangon Biotech, China), an anti-WIF1 rabbit polyclonal antibody (Cat. No. 53483-1, SAB Biotech, USA), and an anti-GAPDH monoclonal antibody (Cat. No. 60004-1-Ig, Sangon Biotech, China). The secondary antibodies were HRP-conjugated goat anti-rabbit IgG (Cat. No. 111-035-003, Jackson ImmunoResearch, USA) and HRP-conjugated goat anti-mouse IgG (Cat. No. SA00001-1, Sangon Biotech, China).

Immunohistochemistry

Immunohistochemistry: After specimen extraction, tissues were fixed in 10% formalin for 24 h, followed by paraffin embedding. Sections of 4 μm thickness were cut and placed on slides. The slides were baked at a constant temperature of 65 °C for 2 h, then sequentially immersed in xylene I and II for 5 min each, followed by immersion in absolute ethanol I and II, 95% ethanol, and 90% ethanol for 2 min each. After performing heat-induced antigen retrieval, antibody incubation was conducted (marking, adding primary antibodies with DKK3 antibody diluted at 1:50 and WIF1 antibody diluted at 1:150) and incubated in a 37 °C incubator for 60 min. Following three washes with PBS, secondary antibody solution was added and incubated at 37 °C for 30 min. DAB was used for color development, and the sections were counterstained with hematoxylin for 5 min, washed with distilled water, and blued in warm water for 1 min. After dehydration and clearing, slides were mounted with neutral resin and observed under a microscope. Finally, two independent pathologists evaluated the staining intensity and the proportion of positive cells without knowledge of the sample grouping. Results were interpreted based on predefined scoring criteria (a total score ≥ 4 was considered positive expression).

Western blot experiment

Western Blot: Tissues were thoroughly ground to extract total protein, and protein concentration was measured. Equal amounts of protein were mixed with loading buffer and denatured by boiling for 10 min. SDS-PAGE was performed using a 12% separating gel and a 4% stacking gel. Proteins were then transferred to a membrane at a constant voltage of 75 V for 120 min at 4 °C. The membrane was blocked with 5% skimmed milk for 2 h. Primary antibodies for DKK3 and WIF1 (diluted at 1:1000) and GAPDH antibody (diluted at 1:3000) were incubated overnight at 4 °C. After washing the membrane, secondary antibodies (anti-rabbit and anti-mouse diluted at 1:5000) were added and incubated at room temperature for 55 min. Following another wash, DAB was used for detection. Image J software was utilized for grayscale analysis of the images, and results were expressed as (DKK3/WIF1)/GAPDH.

Statistical analysis

Data analysis was performed using SPSS 27.0 and GraphPad Prism 10 software. The expression differences of DKK3 and WIF1 between BPH and PCa tissues, as well as their associations with PSA levels, Gleason scores, clinical staging, and age, were evaluated using the χ² test. A p-value of < 0.05 was considered statistically significant [16].

Results

Bioinformatics

Identification of differentially expressed genes (DEGs)

A total of 713 differentially expressed genes (DEGs) were identified in the GEO database (GSE69223) dataset, including 504 downregulated genes and 209 upregulated genes (Fig. 1A). The top 50 most significantly upregulated and the top 50 most significantly downregulated genes were selected to create a heatmap (Fig. 1B). In the TCGA-PRAD dataset, 552 DEGs were identified, comprising 326 downregulated genes and 226 upregulated genes (Fig. 2A). A corresponding heatmap displaying the top 50 upregulated and top 50 downregulated genes was also generated (Fig. 2B). Detailed DEG lists for the GSE69223 and TCGA-PRAD datasets are provided in Supplementary Files 1 and 2, respectively.

Fig. 1.

GSE69223Screening of differentially expressed genes. A Volcano plot; B Heatmap

Fig. 2.

Screening of differentially expressed genes in TCGA-PRAD. A Volcano plot; B Heatmap

Functional and pathway enrichment analysis of DEGs

The intersection of DEGs from both datasets resulted in 339 common differentially expressed genes, with 144 upregulated genes and 195 downregulated genes (Fig. 3A, B). GO and KEGG pathway enrichment analyses were performed on these 339 DEGs. GO enrichment analysis showed that the downregulated genes were mainly involved in biological processes such as cytoskeleton regulation, extracellular matrix composition, and cell migration, including molecular functions like actin binding, extracellular matrix structural constituent, and cell differentiation. Conversely, the upregulated genes were significantly enriched in functions related to metabolic regulation and signaling pathways, including fatty acid metabolism, unsaturated fatty acid biosynthetic processes, and epithelial cell maturation involved in prostate gland development (Fig. 3C). KEGG pathway enrichment analysis revealed that the downregulated genes were primarily concentrated in several critical signaling and disease-related pathways, including the Wnt signaling pathway, TGF-beta signaling pathway, drug metabolism-cytochrome P450, focal adhesion, and the complement and coagulation cascades. These pathways are involved in cell differentiation, tissue remodeling, and inflammation regulation, and are closely associated with cancer, cardiovascular diseases, and metabolic toxicity. Additionally, the downregulated genes participated in disease processes such as dilated cardiomyopathy, fluid shear stress and atherosclerosis. In contrast, the upregulated genes were mainly enriched in pathways related to fatty acid metabolism and unsaturated fatty acid biosynthesis (Fig. 3D).

Fig. 3.

Intersection DEGs analysis in the both datasets. A Venn diagram of upregulated DEGs in the GSE69223 and TCGA-PRAD datasets. B Venn diagram of downregulated DEGs in the GSE69223 and TCGA-PRAD datasets. C GO enrichment analysis of intersected DEGs. D KEGG pathway enrichment analysis of intersected DEGs

PPI network construction and module analysis

Using the STRING database, a protein-protein interaction (PPI) network was constructed for the 339 DEGs, comprising 339 nodes and 418 edges (Fig. 4A). In Cytoscape (v3.10.0), the top 30 genes ranked by the Maximum Neighborhood Component (MNC) algorithm were selected as key genes in prostate cancer, suggesting their potential significance in PCa (Fig. 4B). Through MCODE analysis, three highly dense functional modules (Module A, B, C) were identified. Module A (MCODE score = 4.5) consisted of 5 nodes and 9 edges; Module B (MCODE score = 4.0) comprised 4 nodes and 6 edges; and Module C (MCODE score = 3.6) included 6 nodes and 9 edges (Fig. 4C). The genes in Module A were primarily enriched in metabolism-related pathways such as drug metabolism-cytochrome P450, chemical carcinogen-DNA adducts, glutathione metabolism, and steroid hormone biosynthesis, as well as disease-related pathways like fluid shear stress, atherosclerosis, and hepatocellular carcinoma. The genes in Module B were mainly enriched in pathways related to neural signal transmission, metabolic regulation, and endocrine functions, including neuroactive ligand-receptor interaction, tyrosine metabolism, adipocyte lipolysis regulation, adipocyte factor signaling pathways, cAMP signaling pathway, and alcohol dependence pathways. The genes in Module C were predominantly enriched in the Wnt signaling pathway, mTOR signaling pathway, Hippo signaling pathway, ECM-receptor interaction, and focal adhesion pathways (Table 1).

Fig. 4.

PPI network construction and module analysis. A Protein-protein interaction (PPI) network of 339 DEGs. B Top 30 key genes selected by the Maximum Neighborhood Component (MNC) algorithm. C Three functional modules identified by MCODE analysis

Table 1.

Pathway enrichment analysis of the genes in the three modules

Module	ID	Description	p value	Count
Module A
	hsa05207	Chemical carcinogenesis - receptor activation	0.000135216544853892	3
	hsa05208	Chemical carcinogenesis - reactive oxygen species	0.000158934564223772	3
	hsa00360	Phenylalanine metabolism	0.00898265026841349	1
	hsa00340	Histidine metabolism	0.0123344534246125	1
	hsa00410	beta-Alanine metabolism	0.0173451477855486	1
Module B
	hsa04080	Neuroactive ligand-receptor interaction	0.0050557775534517	2
	hsa00350	Tyrosine metabolism	0.0121197192947747	1
	hsa04923	Regulation of lipolysis in adipocytes	0.0198113727207683	1
	hsa04920	Adipocytokine signaling pathway	0.0234758330613873	1
	hsa05034	Alcoholism	0.0622126306553925	1
Module C
	hsa04310	Wnt signaling pathway	2.91981695807675e-05	3
	hsa05165	Human papillomavirus infection	0.00800774632780826	2
	hsa05144	Malaria	0.022346735791319	1
	hsa05217	Basal cell carcinoma	0.0280950208256855	1
	hsa04512	ECM-receptor interaction	0.039515508102703	1

Further screening of core genes using machine learning algorithms

Using both Lasso regression and Random Forest (RF) machine learning algorithms to further screen the 30 key genes identified, core genes were obtained. The coefficient path plot from the Lasso regression algorithm showed that as the Log(Lambda) value gradually decreased, the model coefficients converged. In the cross-validation bias plot, the optimal Lambda value (λ = 0.05191264) was selected, resulting in the identification of nine core feature genes: FOXA1, SLC45A3, KCNK3, KRT14, TPM2, WIF1, SNAI2, and DKK3 (Fig. 5A, B). The RF algorithm selected genes with MeanDecreaseGini > 1, resulting in four core genes: DKK3, SNAI2, WIF1, and FOXA1 (Fig. 5C). The intersection of the core genes identified by both machine learning algorithms yielded four common core genes: DKK3, SNAI2, WIF1, and FOXA1, indicating that these four genes are likely to be pivotal in prostate cancer.

Fig. 5.

Screening of core genes using machine learning algorithms. A Lasso regression coefficient path plot. B Lasso regression cross-validation plot. C Random Forest selection of core genes

Validation of core gene expression

Using the GSE46602 dataset as an external validation set, ROC curve analysis was performed. The pROC and ggplot2 packages in R were utilized for statistical analysis and to plot the ROC curves of the core genes, calculating the Area Under the Curve (AUC). Core genes with an AUC value greater than 0.9—namely DKK3, SNAI2, and WIF1—were selected (Fig. 6). Combined with the analysis of key functional modules C from MCODE, DKK3 and WIF1 were identified as core genes related to the occurrence and development of PCa for further study.

Fig. 6.

The ROC curve of core genes in the validation set

Expression analysis of core genes

Using the ggviolin package, violin plots were generated to illustrate the expression trends of the two core genes, DKK3 and WIF1, in prostate cancer patients versus healthy controls. The results indicated that both DKK3 and WIF1 were downregulated in prostate cancer patients in the GSE69223 dataset. Similarly, the GSE46602 dataset also showed significant downregulation of DKK3 and WIF1, further validating the importance of these two core genes (Fig. 7).

Fig. 7.

The expression level of core genes. A DKK3 expression levels of the GSE69223 dataset and GSE46602 dataset; B WIF1 expression levels of the GSE69223 dataset and GSE46602 dataset; Compared with BPH group, *P < 0.05, **P < 0.01, ***P < 0.001

Immune correlation analysis of core genes

The WIF1 gene significantly affected adaptive immune cells, such as CD4⁺ T cells and CD8⁺ T cells, with a weaker impact on B cells and almost no significant effect on NK cells and endothelial cells. Additionally, WIF1 was significantly negatively correlated with fibroblasts, suggesting its potential role in regulating the tumor microenvironment by inhibiting matrix remodeling. In contrast, the DKK3 gene showed a strong positive correlation with B cells, CD4⁺ T cells, and CD8⁺ T cells, and a significant promoting effect on endothelial cells, but no significant correlation with NK cells. Both genes were negatively correlated with fibroblasts, with DKK3 exhibiting a more pronounced inhibitory trend (Figs. 8 and 9). Overall, WIF1 appears to be more inclined towards regulating adaptive immunity, whereas DKK3 has a broader range of functions, involving immune cells, endothelial cells, and matrix remodeling. These two genes may regulate the tumor microenvironment through different mechanisms.

Fig. 8.

Immune correlation analysis of DKK3

Fig. 9.

Immune correlation analysis of WIF1

Gene set enrichment analysis (GSEA) of core genes

Based on GSEA analysis, high expression of DKK3 at the GO-BP level is primarily enriched in functions related to steroids (such as aldosterone and mineralocorticoids), polyamine synthesis metabolism, and elastic fiber assembly. Conversely, low expression is associated with the differentiation of GABAergic neurons in the cerebral cortex and protein depalmitoylation (Fig. 10A). In the GO-CC category, high expression is concentrated in endoplasmic reticulum chaperone complexes, the Golgi apparatus, low-density lipoprotein particles, and perisynaptic/perineuronal nets (ECM), while low expression is inclined towards structures like the inner kinetochore and outer dynein arm (Fig. 10B). GO-MF analysis indicates that bile acid/branched-chain amino acid transmembrane transport, BMP receptor binding, and ECM elasticity functions are enhanced with upregulated DKK3, whereas U3 snoRNA binding and myosin V binding are more prominent in low expression (Fig. 10C). KEGG analysis reveals that aminoacyl-tRNA biosynthesis, glycosaminoglycan biosynthesis, TGF-β signaling, and cardiomyopathy pathways (dilated/hypertrophic, etc.) are positively correlated with high DKK3 expression, while viral infections (amoebiasis, hepatitis) and certain metabolic pathways are more inclined towards low expression (Fig. 10D).

Fig. 10.

GSEA enrichment analysis of DKK3. A GO-BP functional annotation analysis was conducted for DKK3. B GO-CC functional annotation analysis was conducted for DKK3. C GO-MF functional annotation analysis was conducted for DKK3. D KEGG pathway analysis was performed for DKK3

In contrast, for WIF1, high expression in GO-BP is mainly enriched in steroid and polyamine metabolism, elastic fiber assembly, and serine metabolism, whereas spindle midzone assembly and protein depalmitoylation are associated with low expression (Fig. 11A). Regarding GO-CC, endoplasmic reticulum chaperone complexes, the Golgi apparatus, LDL particles, and various ECMs (perineuronal net, synapse-associated ECM, etc.) are significantly aggregated at the high expression end, while the inner kinetochore and outer dynein arm are associated with low expression (Fig. 11B). At the GO-MF level, bile acid/branched-chain amino acid transmembrane transport, BMP receptor binding, ECM elasticity, and certain neurotransmitter receptor activities are enhanced with upregulated WIF1, whereas U3 snoRNA binding and myosin V binding tend towards low expression (Fig. 11C). KEGG analysis shows that high expression is mainly significantly enriched in pathways related to dilated/hypertrophic and arrhythmogenic right ventricular cardiomyopathy, TGF-β signaling, terpenoid backbone biosynthesis, and vitamin digestion and absorption, whereas viral myocarditis, fructose/mannose metabolism, and pentose and glucuronate interconversions pathways are more prominent in low expression (Fig. 11D).

Fig. 11.

GSEA enrichment analysis of WIF1. A GO-BP functional annotation analysis was conducted for WIF1. B GO-CC functional annotation analysis was conducted for WIF1. C GO-MF functional annotation analysis was conducted for WIF1. D KEGG pathway analysis was performed for WIF1

Drug-related correlation analysis of core genes

Using the CTRP drug sensitivity and expression correlation analysis from the GSCA database, DKK3 was found to be significantly positively correlated with a considerable number of anti-tumor drugs, including histone deacetylase (HDAC) inhibitors (such as belinostat and apicidin), nucleoside analogs (such as decitabine and cytarabine), and multifunctional agents like ciclopirox. These drugs typically exhibit higher sensitivity or lower half-maximal inhibitory concentration (IC₅₀) in cells with high DKK3 expression. Mechanistically, DKK3 may enhance the cytotoxicity induced by these drugs by regulating apoptosis, proliferation, or repair pathways in tumor cells, particularly those related to the Wnt/TGF-β pathways (Fig. 12A).

Fig. 12.

Correlation analysis of DKK3 and WIF1 expression with CTRP drug sensitivity and mRNA expression. A Correlation analysis of DKK3 expression with CTRP drug sensitivity and mRNA expression. B Correlation analysis of WIF1 expression with CTRP drug sensitivity and mRNA expression

In contrast, WIF1 is negatively correlated with various small molecule inhibitors, such as the ALK inhibitor ceritinib, the antifungal and anti-tumor agent ciclopirox, and CHM-1. These inhibitors demonstrate weaker drug sensitivity in cells with high WIF1 expression, suggesting that upregulation of WIF1 may reduce the efficacy of certain inhibitors in killing tumor cells. This reduction in efficacy may be related to WIF1’s regulation of the Wnt pathway or the induction of other compensatory survival signals (Fig. 12B). These findings provide important clues for subsequent combined drug treatments to elucidate the interactions between DKK3, WIF1, and biological processes such as cell proliferation, apoptosis, DNA damage repair, and immune regulation. This approach aims to identify more targeted intervention targets for drug sensitivity and resistance mechanisms.

Experimental validation

Expression of DKK3 and WIF1 in benign prostatic hyperplasia and prostate cancer tissues

Immunohistochemistry revealed that DKK3 protein was mainly localized in the cytoplasm and cell membrane, whereas WIF1 protein was specifically expressed at the cell membrane. In benign prostatic hyperplasia (BPH) tissues, both DKK3 and WIF1 exhibited strong and diffuse positive staining, while in prostate cancer (PCa) tissues, staining intensity was markedly decreased and presented as weaker, more heterogeneous signals with reduced membrane localization. Consistently, the positive expression rates of DKK3 and WIF1 in BPH tissues were 75% (30/40) and 70% (28/40), respectively. In contrast, the positive expression rates of DKK3 and WIF1 in PCa tissues were significantly reduced to 45% (18/40) and 30% (12/40), respectively, with statistically significant differences (DKK3: χ² = 6.30, P = 0.012; WIF1: χ² = 11.25, P < 0.01) (Table 2). Western blot analysis further confirmed that the protein levels of DKK3 and WIF1 were significantly lower in PCa tissues compared with BPH tissues (P < 0.01) (Fig. 13).

Table 2.

Expression of DKK3 and WIF1 in BPH and PCa

Groups	DKK3			WIF1
	Positive rate [n (%)]	χ²	P	Positive rate [n (%)]	χ²	P
BPH (n = 40)	30 (75.00)	6.30	0.012	28 (70.00)	11.25	< 0.01
PCa (n = 40)	18 (45.00)			12 (30.00)

Fig. 13.

Expression of DKK3 and WIF1 in BPH and PCa. A Expression of DKK3 in BPH. B Expression of DKK3 in PCa. C Expression of WIF1 in BPH. D Expression of WIF1 in PCa. E The protein expression of DKK3 and WIF 1 in BPH and PCa. Compared with BPH group, *P < 0.05, **P < 0.01, ***P < 0.001

Relationship between DKK3 and WIF1 and pathological parameters of prostate cancer

Statistical analysis revealed that the positive expression rates of DKK3 and WIF1 in PCa patients were significantly associated with preoperative PSA levels (≥ 10 ng/ml group vs. <10 ng/ml group) and Gleason scores (> 7 group vs. ≤7 group) (P < 0.05). In groups with high PSA levels and high Gleason scores, the expression of DKK3 and WIF1 was lower. However, the expression levels of these two genes were not significantly associated with patient age or clinical staging (P > 0.05) (Table 3).

Table 3.

Relationship between DKK3 and WIF1 and pathological parameters of PCa

Variable	DKK3			WIF1
	Positive rate [n (%)]	χ²	P	Positive rate [n (%)]	χ²	P
Age (year)
≤ 70 (n = 14)	9(64.28)	1.34	0.247	7(50.00)	0.27	0.605
> 70 (n = 26)	9(34.61)			5(19.23)
PSA (ng/ml)
< 10 (n = 15)	10(66.66)	5.02	0.025	7(46.66)	4.88	0.027
≥ 10 (n = 25)	8(32.00)			5(20.00)
TNM
T1–T2 (n = 22)	10(45.45)	0.98	0.322	6(27.27)	0.19	0.662
T3–T4 (n = 18)	8(44.44)			6(33.33)
Gleason Score
≤ 7 (n = 18)	11(61.11)	5.21	0.022	8 (44.44)	5.12	0.024
> 7 (n = 22)	7(31.81)			4 (18.18)

Discussion

This study systematically identified differentially expressed genes (DEGs) in Prostate Cancer (PCa) by integrating multi-omics databases (GEO and TCGA) and validating with clinical samples. The focus was on the downregulated expression of Wnt signaling pathway inhibitors DKK3 and WIF1 and their potential biological significance. Initially, significant downregulation of DKK3 and WIF1 was repeatedly observed in the TCGA-PRAD and GEO (GSE69223, GSE46602) datasets. This downregulation was positively correlated with higher PSA levels and Gleason scores, indicating that the loss or functional inhibition of these genes may be a key event in the malignant progression of PCa [17–20]. Aberrant activation of the Wnt/β-catenin signaling pathway in PCa can promote tumor proliferation, metastasis, and drug resistance, primarily involving the accumulation of β-catenin in the nucleus and upregulation of cell cycle genes such as Cyclin D1 [21, 22]. This study found that the significant downregulation of DKK3 and WIF1 effectively removes their negative regulation of Wnt signaling, thereby amplifying the malignant phenotype of tumor cells [23–25]. DKK3 can competitively bind with the LRP receptor family to prevent β-catenin from entering the nucleus, while WIF1 directly binds to Wnt ligands to block their interaction with Frizzled receptors. These two inhibitors play important tumor-suppressive roles in various solid tumors, including prostate, gastric, breast, and bladder cancers [26–29]. Therefore, their “inactivation” not only provides sustained activation of Wnt signaling for tumor cells but may also serve as molecular markers for identifying metastatic tendencies and resistance risks in prostate cancer [30, 31].

The study also discovered potential synergistic abnormalities between TGF-β signaling and the Wnt pathway, as evidenced by differential gene enrichment analysis and exploration of the immune microenvironment. TGF-β typically suppresses excessive proliferation of tumor cells in the early stages but can induce epithelial-mesenchymal transition (EMT) and activate cancer-associated fibroblasts (CAFs) in advanced stages, ultimately accelerating angiogenesis and matrix remodeling [32–35]. When synergizing with Wnt signaling, TGF-β further promotes the secretion of matrix metalloproteinases (MMPs) and the formation of new blood vessels, thereby enhancing tumor cell infiltration and metastasis [36, 37]. Thus, in the complex microenvironment of PCa, the loss of WIF1 and DKK3 leads to the cumulative abnormal activation of these two pathways, significantly accelerating the emergence of malignant phenotypes [38, 39].

At the immune microenvironment level, the expression levels of WIF1 and DKK3 are closely related to immune cell infiltration, particularly showing significant positive correlations with adaptive immune populations such as CD8⁺ T cells, CD4⁺ T cells, and B cells. Additionally, DKK3 significantly promotes endothelial cell infiltration, while both genes are negatively correlated with fibroblast populations. This supports their dual impact on tumor immune regulation and matrix remodeling from another perspective [40, 41]. On one hand, DKK3 may weaken tumor immune evasion by regulating immune cell activation and polarization processes (e.g., transitioning macrophages from the M2 to the M1 phenotype). On the other hand, the loss of WIF1 may enhance the tumor cells’ resistance to host immunity [42–45]. The study’s results suggest that both genes may act synergistically within complex networks such as “immune regulation—Wnt signaling—cytoskeletal remodeling.” Based on this, targeting the restoration of DKK3 or WIF1 expression in subsequent studies, combined with novel immunotherapies like immune checkpoint inhibitors (PD-1/PD-L1, CTLA-4) or CAR-T cells, could potentially enhance efficacy in overcoming resistance and delaying recurrence [46–48].

The downregulation of DKK3 and WIF1, when combined with indicators such as PSA and Gleason scores, can be used to identify high-risk PCa and optimize personalized treatment strategies. Immunohistochemistry and protein level detections in this study confirmed the association of these two genes with PCa invasiveness and metastatic tendencies, providing preliminary support for constructing multi-gene predictive models that include WIF1 and DKK3 [49–52]. Additionally, analysis using the Cancer Therapeutics Response Portal (CTRP) revealed that DKK3 is positively correlated with the sensitivity to various chemotherapy and targeted drugs, while WIF1 shows a negative correlation with several small molecule inhibitors. This suggests that incorporating the expression levels of DKK3 and WIF1 in personalized treatment strategies could enhance therapeutic outcomes and reduce unnecessary drug side effects by selecting more targeted drug combinations. In the future, integrating these genes into comprehensive diagnostic and treatment decision-making systems, along with in vitro and in vivo functional experiments to confirm their direct impact on drug resistance and disease progression, could precisely identify key downstream targets and facilitate drug screening. This approach holds promise for providing more personalized diagnostic and therapeutic options for PCa patients.

However, this study has certain limitations. First, the cross-sectional design and limited overall sample size make it difficult to comprehensively assess dynamic changes in gene expression and drug response. Second, although the roles of DKK3 and WIF1 in Wnt/TGF-β signaling and the immune microenvironment were identified, in vitro and in vivo functional experiments were not performed to confirm their direct regulatory effects on prostate cancer biology. Third, the Western blot analysis was conducted on a relatively small number of tissue samples; however, these samples were carefully selected based on strict clinical and pathological criteria to ensure that they were representative of the larger cohort. Finally, additional validation in multicenter and large-scale prospective cohorts will be needed to further strengthen the generalizability of our findings. Overall, this study reveals the comprehensive regulatory roles and potential application values of WIF1 and DKK3 in prostate cancer from genomic, immune, and clinicopathological perspectives, laying the foundation for further elucidating the molecular mechanisms of prostate cancer and optimizing its diagnostic and therapeutic strategies. Future studies that combine multi-omics integrative analyses with functional research to explore the cross-pathway regulation of these two genes in epigenetic modifications, metabolic reprogramming, and immune homeostasis will not only deepen the understanding of the complex pathological processes of prostate cancer but also provide more precise theoretical support for improving its diagnosis and treatment levels and extending patient overall survival [53–55].

Conclusion

This study identified differentially expressed genes (DEGs) in prostate cancer (PCa) through multi-omics integration of GEO and TCGA databases, with clinical validation. It highlighted the significant downregulation of WIF1 and DKK3, which are associated with higher PSA levels and Gleason scores, suggesting their role in tumor progression by relieving negative regulation of the Wnt pathway. Functional and immunological analyses revealed complex interactions with TGF-β signaling, immune cell infiltration, and extracellular matrix remodeling. The loss of these inhibitors exacerbates Wnt/TGF-β pathway activation, impairing immune recognition and promoting metastasis. Drug sensitivity analysis showed that DKK3 expression is positively correlated with chemotherapy sensitivity, while WIF1 correlates with reduced efficacy of small molecule inhibitors. These findings suggest that targeting Wnt/TGF-β signaling and immune modulation could improve therapeutic strategies for high-risk PCa. Future research integrating functional experiments and large-scale studies will help clarify the mechanisms of WIF1 and DKK3 in tumorigenesis and drug resistance, potentially guiding more personalized treatment approaches.

Author contributions

Zhiliang Xia and Zhonggui Hu: Conceptualization (lead); data curation (lead); formal analysis (lead); investigation (equal); project administration (equal); writing – original draft (lead).Dan Du and Zhi Zhang: Data curation (equal); formal analysis (equal); methodology (equal); project administration (equal); writing – original draft (equal).Zonglai Liu : Investigation (equal); methodology (equal); data curation (supporting); writing – review and editing (supporting).Xinyu Li and Xiong Guo: : Investigation (equal); formal analysis (supporting); writing – review and editing (supporting).Ziqiu He: Conceptualization (lead); funding acquisition (lead); resources (lead); supervision (lead); validation (lead); writing – review and editing (lead).

Funding

This research was supported by the Yichang Science and Technology Innovation Project (No. A23-1-049 to Ziqiu He).

Data availability

The datasets analyzed during this study are publicly available. The GEO datasets can be accessed at GSE69223 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE69223) and GSE46602 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE46602). TCGA-PRAD RNA-seq data and associated clinical information are available through The Cancer Genome Atlas (TCGA) portal (https://portal.gdc.cancer.gov/). Additional data supporting the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the ethical standards of the Second People’s Hospital of China Three Gorges University Ethics Committee, which approved the research protocol(Ethics Approval Number: NO.202435). Written informed consent was obtained from all participants prior to sample collection.

Competing interests

The authors declare no competing interests.