NOX4 suppresses malignant progression in prostate adenocarcinoma and functions as a potential diagnostic biomarker

Introduction

Prostate adenocarcinoma (PRAD) is one of the most prevalent malignancies in men, with a particularly high incidence among elderly populations (1,2). Driven by demographic aging and changing lifestyle factors, the global burden of PRAD continues to rise. Clinically, early-stage PRAD is often asymptomatic; however, as the disease progresses, patients may develop urinary symptoms such as frequency, urgency, and discomfort (3,4). Serum prostate-specific antigen (PSA) testing remains the cornerstone of clinical screening for prostate cancer. However, its inherent limitations severely hinder its ability to meet the demands of precise clinical screening and preclude its utility for effective patient stratification and prognostic prediction. Current treatment approaches, including surgery, radiation therapy, hormonal therapy, and chemotherapy, are selected based on tumor stage, patient age, overall health status, and individual preferences (5). Importantly, recent advances in understanding the molecular basis of PRAD have enabled the development of novel targeted therapies and immunotherapies, offering renewed hope for improved patient outcomes, In clinical practice, targeted therapy for prostate cancer primarily focuses on the key molecular pathways and specific targets involved in tumorigenesis and progression, with core mainstream therapeutic strategies and targeted directions including androgen receptor-targeted therapy, PARP inhibitors, and PI3K/AKT/mTOR pathway inhibitors, among others. Immunotherapy for PRAD centers on activating the body’s anti-tumor immunity and reversing the tumor immunosuppressive microenvironment; its mainstream regimens are based on immune checkpoint inhibitors, combined with therapeutic vaccines and adoptive cellular therapy (6).

Beyond its epidemiological significance, PRAD exhibits substantial heterogeneity at both pathological and molecular levels. Despite being classified as a single disease entity, it displays marked variability in biological behavior, clinical presentation, tumor progression kinetics, and response to therapy (7,8). This heterogeneity manifests in differences in histological grade, architectural patterns, and invasive potential, ranging from indolent, localized tumors to aggressive, metastatic variants (9). Furthermore, molecular subtyping has identified distinct subclasses of PRAD characterized by differential gene expression profiles, chromosomal rearrangements, and dysregulation of key signaling pathways (10-12). Such molecular diversity contributes to heterogeneous treatment responses across patients, directly influencing prognosis and overall survival.

Given this complexity, a comprehensive understanding of PRAD heterogeneity is essential for advancing precision medicine. Specifically, the identification of core biomarkers through multi-omics analysis offers significant potential to enhance diagnostic accuracy and guide personalized therapeutic decisions. Among candidate regulators, NADPH oxidase 4 (NOX4), a critical enzyme in reactive oxygen species (ROS) generation, has emerged as a molecule of interest due to its context-dependent roles in tumorigenesis. For example, in hepatocellular carcinoma, NOX4 exerts tumor-suppressive effects by maintaining redox homeostasis (13), whereas in acute lung injury, excessive NOX4 activation exacerbates oxidative stress and tissue damage (14). However, the expression pattern and functional role of NOX4 in PRAD remain poorly defined, limiting its translational relevance in prostate cancer research. Moreover, the downstream molecular pathways mediated by NOX4 in PRAD, as well as its value in predicting clinical prognosis and treatment response, have not been systematically explored to date.

This study aims to systematically identify core biomarkers in PRAD using integrated multi-omics analysis, investigate their associations with the tumor immune microenvironment, and specifically validate the regulatory function of NOX4 in driving malignant phenotypes of PRAD cells. Additionally, we further explore the downstream signaling axis of NOX4 in PRAD cells, detect the changes of key molecules in the pathway after NOX4 modulation, and analyze the correlation between NOX4 expression and clinical pathological parameters and prognosis of PRAD patients. By combining computational approaches with experimental validation, we aim to elucidate the mechanistic role of NOX4 in PRAD pathogenesis and provide foundational evidence for future targeted therapeutic strategies. In clinical practice, targeted therapy for PRAD primarily focuses on the key molecular pathways and specific targets implicated in tumorigenesis and progression, among which inhibitors of the PI3K/AKT pathway represent a pivotal therapeutic target for PRAD. We present this article in accordance with the TRIPOD and MDAR reporting checklists (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-943/rc).

Methods

Data source

RNA-seq data and clinical data for PRAD were obtained from the UCSC Xena database (https://portal.gdc.cancer.gov/). The dataset consisted of 551 samples, including 51 normal samples and 500 disease samples. Follow-up survival data and detailed clinical pathological parameters of PRAD patients were also extracted for prognostic analysis. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Identification of differentially expressed genes (DEGs)

In our RNA-seq dataset, genes with more than half of their values equal to zero were removed. Differential analysis was conducted using the limma package in R (15). DEGs were selected based on criteria of P value <0.05 and |log2 fold change (FC)| >2. Visualization of volcano plots and heatmaps was performed using the ggplot2 and heatmap packages in R (16).

Construction of gene co-expression networks

To explore interactions and functional modules among genes in PRAD, weighted gene co-expression network analysis (WGCNA) was performed on the TCGA-PRAD dataset. Initially, a soft threshold of 13 (scale-free R²=0.85) was determined based on gene correlations. The minimum module size was set to 50 to identify key module genes. Here, we focused on the deep green module, comprising 2,695 genes.

Pathway enrichment analysis

To further investigate the functional roles of key genes in PRAD and the pathways they are involved in, an intersection was performed between DEGs and genes from the deep green module identified through WGCNA. This resulted in 148 genes, which were subjected to Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses (https://www.geneontology.org/ and https://www.genome.jp/kegg/). The enrichment analysis assessed the potential biological functions of these genes and their involvement in critical biological pathways (17), with statistical significance defined as P<0.01. Subsequently, gene set enrichment analysis (GSEA) was conducted on these 148 genes using the gene set h.all.v7.5.1.symbols.GMT downloaded from the MSigDB database (https://data.broadinstitute.org/gsea-msigdb/msigdb/release/7.5.1/) to validate pathway enrichments at the gene set level (18). Benjamini-Hochberg false discovery rate (FDR) correction was applied for multiple testing adjustment during enrichment analysis to reduce the risk of false-positive results. An adjusted P value (FDR) <0.05 was considered statistically significant for enriched terms.

Machine learning for PRAD biomarker selection

Three distinct machine learning methods were employed for PRAD biomarker selection: least absolute shrinkage and selection operator (LASSO) regression, support vector machine with recursive feature elimination (SVM-RFE), and random forest (RF). LASSO regression introduces L1 regularization to select features, reducing the weights of unimportant features to zero while retaining critical features associated with the response variable. SVM-RFE progressively selects and cross-validates features based on their importance using a 10-fold cross-validation approach. RF assigns an importance score to each feature, and we retained genes with scores greater than 1. The gene importance score threshold of >1 was set based on the default ranking criteria and empirical cutoff commonly adopted in RF feature selection to retain genes with robust predictive significance. This threshold was empirically determined to prioritize genes with relatively high importance scores while minimizing the inclusion of low‑contribution features, ensuring the stability and reliability of the model. The intersection of results from these three methods was considered as the final set of PRAD biomarkers.

Expression profiling of PRAD biomarkers

Based on RNA-seq data, we visualized the expression profiles of the selected PRAD biomarkers (AOX1, APOBEC3C, C8orf88, GCNT4, NOX4, and PHYHIPL-a total of six targets) using box plots with the ggpubr package in R. Wilcoxon rank-sum tests were conducted to calculate inter-group expression differences (19).

Cell line

Human prostate cancer cell lines (DU145 and PC3) were used in this study. All cell lines were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Cell identity was verified by short tandem repeat (STR) profiling to ensure no cross-contamination. Cells were cultured in high-glucose DMEM medium (Cat. C11995500BT, Gibco, California, USA) supplemented with 10% fetal bovine serum (FBS, Cat. 04-001-1ACS, Bioind) and 1% penicillin-streptomycin double antibody (Cat. 60162ES76, YEASEN, Shanghai, China). They were maintained in a humidified incubator at 37 ℃ with 5% CO₂, and the medium was refreshed every 2–3 days. Cells in the logarithmic growth phase were used for subsequent experiments.

Plasmid construction

NOX4 overexpression plasmids (OE-NOX4): The full-length open reading frame of the human NOX4 gene was amplified by PCR and cloned into the pcDNA3.1-3×Flag vector (Invitrogen) to construct NOX4 overexpression plasmids (OE-NOX4 #1, OE-NOX4 #2). The empty pcDNA3.1 vector served as the negative control (Control). Plasmid sequencing was performed by Beijing Tsingke Biotechnology Co., Ltd. (Beijing, China) to confirm the correctness of the sequence. In addition, siRNA targeting the key downstream molecules of NOX4 and their corresponding negative control siRNA were synthesized by Guangzhou RiboBio Co., Ltd. (Guangzhou, China) for rescue experiments.

Cell transfection

Preparation before transfection: DU145 and PC3 cells were seeded into 6-well plates at a density of 3×10⁵ cells/well and cultured until the cell confluency reached 70–80% for transfection. Lipofectamine 2000 transfection reagent (Cat. 40802ES01, YEASEN) was used following the manufacturer’s instructions: 2 µg of plasmid, siRNA or si-NOX4 and 4 µL of Lipofectamine 2000 were separately diluted in 250 µL of serum-free DMEM, incubated at room temperature for 5 min, then mixed and allowed to stand for 20 min to form transfection complexes, which were then added to the cell culture wells. At 48 h post-transfection, cells were collected, and the expression level of NOX4 protein was detected by Western blot (WB) to verify the efficiency of interference, overexpression or NOX4 silencing. Effective sequences (si-NOX4) were selected for subsequent experiments.

PCR

PCR was used to amplify the full-length cDNA sequence of the NOX4 gene and verify the success of plasmid construction. High-fidelity DNA polymerase (e.g., Phusion High-Fidelity DNA Polymerase) was used, with cDNA from prostate cancer cells as the template, and specific primers were used for PCR amplification. The PCR reaction system included template DNA, primers, dNTPs, buffer, and DNA polymerase. The reaction conditions were set as follows: pre-denaturation at 95 ℃ for 3 min, followed by 30–35 cycles of denaturation, annealing, and extension, and a final extension at 72 ℃ for 5–10 min. PCR products were verified by agarose gel electrophoresis for size and then purified and recovered for subsequent experiments.

WB

WB was used to detect the expression level of NOX4 protein in prostate cancer cells. Prostate cancer cells in the logarithmic growth phase were collected, lysed to extract total protein, and the protein concentration was determined by the BCA method. Equal amounts of protein were subjected to SDS-PAGE electrophoresis for separation, then transferred onto PVDF membranes. After blocking, the membranes were incubated with anti-NOX4 antibody and internal reference antibody respectively at 4 ℃ overnight. On the next day, the membranes were incubated with HRP-labeled secondary antibody, and ECL chemiluminescence was used for development. The results were recorded using an image acquisition system.

Cell Counting Kit‑8 (CCK-8) assay

Transfected DU145 and PC3 cells were seeded into 96-well plates at a density of 2×10⁴ cells/well, with 100 µL of medium added to each well. Three replicate wells were set for each group, and blank control wells were also included. Detection was performed on days 0, 1, 2, 3, and 4 of culture. At each detection time point, 10 µL of CCK-8 reagent (Cat. C0037, Beyotime, Shanghai, China) was added to each well, followed by incubation at 37 ℃ with 5% CO₂ for 2 h. A BioTek Synergy H1 microplate reader (BioTek) was used to measure the absorbance (OD value) at a wavelength of 450 nm. A cell proliferation curve was plotted with culture time as the abscissa and OD value as the ordinate to analyze the effect of NOX4 on the proliferation of prostate cancer cells.

Wound healing assay

Transfected DU145 and PC3 cells were seeded into 6-well plates at a density of 4×10⁵ cells/well and cultured until the cell confluency reached 90–100%. A uniform scratch was made vertically on the bottom of the well using a sterile 200 µL pipette tip. Cells were washed three times with PBS to remove detached cells, and serum-free DMEM medium was added. Images of the same field were captured under an optical microscope at 0, 24, and 48 h after scratching. ImageJ software was used to measure the scratch width, and the scratch healing rate was calculated as follows: Healing rate = (Scratch width at 0 h − Scratch width at each time point)/Scratch width at 0 h × 100%. The effect of NOX4 on the migration ability of prostate cancer cells was analyzed.

Detecting intracellular ROS levels via DCFH-DA probe loading

For the detection of intracellular ROS levels in NOX4-overexpressing adherent DU145 and PC3 PRAD cells, the adherent cell in-situ DCFH-DA probe loading method was adopted. Briefly, DCFH-DA was diluted 1:1,000 with extracellular fluid (C0216) to prepare a working solution with a final concentration of 10 µmol/L. The cell culture medium was discarded, and no less than 1 mL of the prepared DCFH-DA working solution was added to each well of a 6-well plate, with the volume sufficient to completely cover the cells. The cells were then incubated at 37 ℃ in a cell incubator for 20 minutes. After incubation, the cells were washed three times with extracellular fluid (C0216) to thoroughly remove uninternalized free probes. A ROS positive control group was set up simultaneously, and ROS levels in this group were significantly elevated 20–30 minutes after cell stimulation, which was used to verify the validity of the experimental system. After the above operations were completed for the NOX4-overexpressing DU145 and PC3 cell groups as well as the control group, the detection of intracellular ROS levels in each group was subsequently performed.

Statistical analysis

All experiments were independently repeated three times, and the data were expressed as mean ± standard deviation (SD). GraphPad Prism 9 software was used for statistical analysis: independent samples t-test was used for comparison between two groups, and one-way analysis of variance (one-way ANOVA) was used for comparison among multiple groups. A P<0.05 was considered statistically significant.

Results

Identification and visualization of DEGs

To delineate the transcriptomic landscape of PRAD, we first normalized RNA-seq data from the TCGA-PRAD cohort. Box plots depicting the distribution of gene expression levels before and after normalization are shown in Figure 1A,1B. Subsequently, differential expression analysis was performed using the limma package, identifying 826 DEGs based on the thresholds of |FC| >2 and adjusted P value <0.05. These DEGs were visualized in a volcano plot, and hierarchical clustering was applied to the top 30 upregulated and 30 downregulated genes, which were displayed as a heatmap (Figure 1C,1D).

Figure 1 Differential analysis of PRAD and normal samples. (A) Box plot of gene expression before standardization. (B) Box plot of gene expression after standardization using the Limma package. (C) Heatmap depicting the top 30 upregulated and top 30 downregulated genes prior to differential analysis. (D) Volcano plot illustrating differential analysis results. FC, fold change; PRAD, prostate adenocarcinoma.

Construction of co-expression gene networks

To further identify gene modules associated with clinical features of PRAD, samples were classified into normal and disease groups. Using a soft threshold power of 13 (scale-free R²=0.85) and a cut height of 0.25, a weighted gene co-expression network was constructed, yielding nine distinct modules (Figure 2A-2C). Among these, the deep green module exhibited the strongest association with clinical traits—particularly the Gleason score (correlation =0.49, P=8.6×10⁻¹⁶³)—and comprised 2,695 genes (Figure 2D). Overlapping these with the previously identified DEGs yielded 148 shared genes for downstream analysis (Figure 2E).

Figure 2 Gene co-expression network analysis. (A) Selection of β=13 as the soft threshold using a combination of scale independence and average connectivity analysis. (B) Hierarchical clustering dendrogram of PRAD and normal samples. (C) Hierarchical clustering dendrograms of source modules and dynamic cut modules. (D) Correlation of PRAD group, normal group Gleason score (GS), and clinical metastasis (MM) in the MEdarkgreen module. (E) Heatmap depicting the correlation between module genes and control and PRAD. GS, gene significance; MM, module membership; PRAD, prostate adenocarcinoma.

GO, KEGG, and GSEA pathway enrichment analysis

Functional characterization of these 148 intersection genes was conducted through enrichment analysis (Figure 3A). GO analysis revealed significant enrichment in biological processes such as thyroid hormone metabolism and amino acid modification, cellular components including the extracellular matrix and NADPH oxidase complex, and molecular functions such as superoxide-generating NAD(P)H oxidase activity and peroxidase activity. KEGG pathway enrichment further linked these genes to thyroid hormone synthesis, pancreatic secretion, and renin secretion (Figure 3B-3D). A correlation heatmap was generated to illustrate the associations between these genes and representative pathways derived from GSEA (Figure 3E).

Figure 3 Pathway and functional enrichment analysis. (A) Venn diagram showing the intersection of differentially expressed genes and key module genes identified by WGCNA, yielding 148 shared genes for subsequent functional and pathway analysis. These genes represent robust candidate genes closely associated with PRAD pathogenesis. (B) GO enrichment analysis categorized into BP, CC, and MF, revealing that these genes are mainly involved in tumor-related biological processes, molecular binding, and membrane or extracellular matrix functions, indicating their potential roles in PRAD progression. (C) KEGG pathway enrichment scatter plot, in which dot size represents the number of enriched genes and the x-axis shows −log10(P value). Key cancer-related signaling pathways were significantly enriched, highlighting critical pathways underlying PRAD development. (D) Network visualization of KEGG enrichment results, illustrating the crosstalk and interaction among significantly enriched pathways, which helps to reveal the complex regulatory network in PRAD. (E) GSEA enrichment heatmap demonstrating pathway activation patterns across samples, providing further evidence of pathway dysregulation in PRAD. BP, biological process; CC, cellular component; DEGs, differentially expressed genes; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function; PRAD, prostate adenocarcinoma; WGCNA, weighted gene co-expression network analysis.

Machine learning selection of PRAD biomarkers

To refine these candidates into a robust diagnostic signature, three machine learning algorithms were applied using the expression profiles of the 148 genes and clinical group labels (Figure 4A). LASSO regression identified 21 PRAD-associated genes (Figure 4B), SVM-RFE selected 24 candidate biomarkers, and RF analysis retained 19 genes with importance scores exceeding 1 (Figure 4C). The intersection of these three gene sets identified six consensus genes—AOX1, APOBEC3C, C8orf88, GCNT4, NOX4, and PHYHIPL—which were designated as the final diagnostic biomarkers for PRAD (Figure 4D).

Figure 4 Machine learning selection of PRAD diagnostic biomarkers. (A) Error plot for the random forest algorithm in PRAD. (B) Biomarker selection in the LASSO regression model. The lowest point on the curve corresponds to the optimal number of genes for PRAD diagnosis (n=21). (C) Ranking of gene importance scores in the random forest results. (D) Feature selection for PRAD by the SVM-RFE model (n=24). LASSO, least absolute shrinkage and selection operator; PRAD, prostate adenocarcinoma; SVM-RFE, support vector machine with recursive feature elimination.

Expression characteristics and immune infiltration of PRAD biomarkers

We then assessed the expression patterns and clinical relevance of these six biomarkers. Boxplot analysis coupled with Wilcoxon rank-sum tests confirmed statistically significant differential expression between normal and PRAD samples. Notably, NOX4 was the only gene upregulated in PRAD, whereas the remaining five were consistently downregulated (Figure 5A). Crucially, Kaplan-Meier survival analysis revealed that lower NOX4 expression was significantly associated with a shorter Overall Survival in PRAD patients (Figure 5A), further underscoring its potential as a prognostic indicator. Correlation analysis revealed strong positive intergene correlations among the upregulated biomarkers, particularly between C8orf88 and both AOX1 and GCNT4, while a pronounced negative correlation was observed between NOX4 and PHYHIPL (Figure 5B,5C). Furthermore, immune infiltration analysis using CIBERSORT indicated that NOX4 expression was negatively correlated with M2 macrophage abundance but positively correlated with neutrophil infiltration—a pattern inversely observed for the other five biomarkers (Figure 5D). These findings suggest a potential immunomodulatory role of NOX4 within the PRAD tumor microenvironment.

Figure 5 Expression characteristics of PRAD core biomarkers and their correlation with immune infiltration. (A) Expression levels of six core biomarkers (AOX1, APOBEC3C, C8orf88, GCNT4, NOX4, PHYHIPL) in normal and PRAD tissues. NOX4 was significantly upregulated in tumor tissues, whereas the other five genes were markedly downregulated, suggesting their distinct expression patterns and potential roles in PRAD. (B) Correlation heatmap of the six core biomarkers. AOX1, APOBEC3C, C8orf88, GCNT4, and PHYHIPL were positively correlated with each other, while NOX4 showed a negative correlation with PHYHIPL, implying coordinated or antagonistic regulatory relationships among these genes in PRAD. (C) Correlation coefficients between key biomarkers and representative immune cell types, showing the strength and direction of associations, which helps to understand the potential immune regulatory roles of these biomarkers. (D) Correlation heatmap between six core biomarkers and immune cell infiltration in PRAD. NOX4 was negatively correlated with M2 macrophages but positively correlated with neutrophils, whereas the other five genes showed the opposite pattern. These results suggest that the six biomarkers may participate in shaping the immunosuppressive microenvironment and regulating immune cell polarization in PRAD. *, P<0.05; **, P<0.01; ***, P<0.001. HR, hazard ratio; PRAD, prostate adenocarcinoma.

Construction and validation of NOX4 knockdown and overexpression models

To investigate the function of NOX4 in PRAD, we first constructed NOX4 knockdown (si-NOX4) and overexpression (OE-NOX4) models in DU145 and PC3 cells, respectively. qRT-PCR and WB analyses revealed that, compared with the control group, transfection with si-NOX4 significantly reduced both mRNA and protein expression levels of NOX4 in DU145 and PC3 cells (Figure 6A-6D). Conversely, transfection with OE-NOX4 markedly increased NOX4 expression at both the mRNA and protein levels in these cell lines (Figure 6E-6H).

Figure 6 Construction and validation of NOX4 knockdown and overexpression models in DU145 and PC3 cells. To investigate the function of NOX4 in PRAD, we constructed NOX4 knockdown (si-NOX4) and overexpression (OE-NOX4) models in DU145 and PC3 cells, respectively. (A,B) qRT-PCR analysis of NOX4 mRNA expression in DU145 and PC3 cells after si-NOX4 transfection. (C,D) Western blot analysis of NOX4 protein expression in DU145 and PC3 cells after si-NOX4 transfection. (E,F) qRT-PCR analysis of NOX4 mRNA expression in DU145 and PC3 cells after OE-NOX4 transfection. (G,H) Western blot analysis of NOX4 protein expression in DU145 and PC3 cells after OE-NOX4 transfection. **, P<0.01; ***, P<0.001. OE, overexpression; PRAD, prostate adenocarcinoma; qRT-PCR, quantitative reverse-transcription polymerase chain reaction.

NOX4 regulates the proliferation and migration of PRAD cells

CCK-8 assays were performed to evaluate the effect of altered NOX4 expression on cell proliferation. As shown in Figure 7A,7B, in both DU145 and PC3 cells, the proliferation rate in the NOX4-siRNA group was significantly higher than that in the control group, indicating that NOX4 knockdown promotes cell proliferation. In contrast, NOX4 overexpression markedly inhibited cell proliferation (Figure 7C,7D). Taken together, these results suggest that, at the cellular level, NOX4 functions as a tumor suppressor and significantly inhibits the progression of PRAD.

Figure 7 Regulatory effects of NOX4 on the proliferation and migration of PRAD cells. CCK-8 and wound healing assays were performed to evaluate the impact of altered NOX4 expression on the proliferation and migration of PRAD cells. (A,B) CCK-8 assay showing the proliferation of DU145 and PC3 cells after NOX4 knockdown. (C,D) CCK-8 assay showing the proliferation of DU145 and PC3 cells after NOX4 overexpression. (E,F) Wound healing assay and quantitative analysis demonstrating that NOX4 knockdown significantly promotes the migration of DU145 cells. (G,H) Wound healing assay and quantitative analysis demonstrating that NOX4 knockdown significantly promotes the migration of PC3 cells. (I,J) Wound healing assay and quantitative analysis showing that NOX4 overexpression significantly inhibits the migration of DU145 cells. (K,L) Wound healing assay and quantitative analysis showing that NOX4 overexpression significantly inhibits the migration of PC3 cells. **, P<0.01. CCK-8, Cell Counting Kit‑8; NC, negative control; OE, overexpression; PRAD, prostate adenocarcinoma.

Wound healing assays were conducted to evaluate the impact of NOX4 expression on cell migration. In DU145 and PC3 cells, the migration rate of the si-NOX4 group was significantly higher than that of the control group at both 24 h and 48 h (Figure 7E-7H), indicating that NOX4 knockdown enhances cell migration. Conversely, the migration rate of the OE-NOX4 group was significantly lower than that of the control group at 24 h and 48 h (Figure 7I-7L), suggesting that NOX4 overexpression inhibits cell migration.

NOX4 modulates redox homeostasis and regulates the PI3K/AKT signaling pathway in PRAD cells

Intracellular ROS levels were detected using the DCFH-DA probe method. In DU145 and PC3 cells, the mean integrated optical density (IOD) values in the OE-NOX4 group were significantly higher than those in the control and vector groups (Figure 8A,8B), indicating that NOX4 overexpression markedly elevates intracellular ROS levels.

Figure 8 Effect of NOX4 on intracellular ROS levels and activation of the PI3K/AKT signaling pathway in PRAD cells. (A,B) DCFH-DA probe assay showing the mean IOD values in DU145 and PC3 cells after OE-NOX4 transfection, compared with the control and vector groups. (C) Western blot analysis of p-PI3K and p-AKT protein expression in DU145 and PC3 cells after NOX4 knockdown. (D) Western blot analysis of p-PI3K and p-AKT protein expression in DU145 and PC3 cells after NOX4 overexpression. ***, P<0.001. DCFH-DA, 2',7'-dichlorodihydrofluorescein diacetate; IOD, integrated optical density; OE, overexpression; PRAD, prostate adenocarcinoma; ROS, reactive oxygen species.

WB analysis was used to detect the expression of key molecules in the PI3K/AKT pathway. In DU145 and PC3 cells, the protein expression levels of p-PI3K and p-AKT were significantly reduced in the si-NOX4 group, while the total expression of PI3K and AKT remained unchanged (Figure 8C). In contrast, the expression levels of p-PI3K and p-AKT were significantly increased in the OE-NOX4 group, with no notable changes in total PI3K and AKT expression (Figure 8D). These results suggest that NOX4 may influence the activation of the PI3K/AKT signaling pathway by regulating intracellular ROS levels.

Discussion

Through an integrated bioinformatics approach combining differential expression analysis, WGCNA, and three machine learning algorithms, LASSO, SVM-RFE, and RF (20-22), we identified six key genes (AOX1, APOBEC3C, C8orf88, GCNT4, NOX4, and PHYHIPL) strongly associated with PRAD. These genes represent promising diagnostic biomarkers and potential therapeutic targets for PRAD.

Immune infiltration analysis revealed distinct correlation patterns: NOX4 expression was negatively correlated with M2 macrophages but positively correlated with neutrophils, whereas the other five genes exhibited the opposite trend. Given that M2 macrophages promote immune suppression through cytokines such as IL-10 and TGF-β (23,24), and neutrophils display dual roles in tumor progression (25), we propose that NOX4 may modulate PRAD progression by influencing immune cell polarization. Specifically, reduced NOX4 expression may attenuate ROS-mediated inhibition of M2 polarization, thereby contributing to an immunosuppressive tumor microenvironment. This hypothesis warrants validation through functional immune co-culture assays.

The tumor-suppressive function of NOX4 may be mediated through oxidative stress signaling pathways. Prior studies have demonstrated that moderate levels of ROS (26), or inhibit oncogenic cascades including PI3K/AKT and MAPK (27). In our study, NOX4 overexpression likely suppressed PRAD cell proliferation, invasion, and migration via analogous mechanisms. Nevertheless, the precise “NOX4-ROS-signaling axis” requires further experimental confirmation, including measurements of intracellular ROS levels and phosphorylation status of key signaling proteins following NOX4 modulation.

Accumulating evidence from previous studies supports the existence of a bidirectional regulatory relationship between NOX4 and PI3K/AKT. Huang et al. demonstrated that lactate upregulates NADPH-dependent NOX4 expression via the HCAR1/PI3K pathway, and the subsequent NOX4-derived ROS exacerbates chondrocyte damage in osteoarthritis (28). Conversely, bisphenol A promotes proliferation and tumorigenesis in papillary thyroid carcinoma by activating NOX4, which induces ROS generation to further stimulate downstream pathways, including PI3K/AKT, thereby positioning NOX4 as an upstream trigger of oncogenic signaling (29). Additionally, Wang et al. reported that the MYH7 R453C mutation drives cardiac remodeling by activating the NOX4/ROS/NF-κB pathway, which acts in synergy with the PI3K/AKT pathway, where NOX4-derived ROS serves as a critical signal for activating PI3K/AKT (30). In another study, Wang et al. showed that aquaporin 3 (AQP3) promotes cervical cancer invasion and metastasis by regulating NOX4-derived H₂O₂, which in turn activates the Syk/PI3K/AKT signaling axis (31). Collectively, these studies confirm a tight upstream-downstream regulatory association between NOX4 and the PI3K/AKT pathway, with ROS serving as the central mediator of signal transduction. This interaction exhibits a bidirectional regulatory characteristic: PI3K/AKT can function as an upstream pathway to modulate NOX4 expression, or NOX4 can act upstream to activate PI3K/AKT via ROS generation. The specific regulatory pattern is highly context-dependent, varying with disease type and external stimuli. Ultimately, this crosstalk contributes to diverse pathological outcomes, including cellular damage, malignant tumor progression, and tissue remodeling.

Notably, a recent pan-cancer review by Sun et al. (32) characterized NOX4 as a potential oncogene and a poor prognostic biomarker in multiple malignancies, including certain prostate cancer datasets. This appears to contrast with our findings, which identify NOX4 as a tumor suppressor in PRAD. This discrepancy can be attributed to the well-documented ‘double-edged sword’ effect of NOX4-derived ROS. While moderate levels of ROS can activate survival signaling, high levels or specific temporal induction of ROS can lead to oxidative stress, DNA damage, and the inhibition of oncogenic pathways such as PI3K/AKT, as demonstrated in our mechanistic assays. Our data, supported by bidirectional functional experiments and independent cohort validation (GSE70768), suggest that in the specific context of the TCGA-PRAD cohort, the downregulation of NOX4 contributes to malignant progression, possibly by relieving the ROS-mediated inhibition of the PI3K/AKT axis. These results emphasize the necessity of considering tissue-specific contexts and ROS thresholds when evaluating the functional role of NOX family members.

Beyond the ROS-PI3K/AKT axis, other molecular layers may influence NOX4’s role in PRAD. For instance, Sun et al. (32) reported that NOX4 is positively associated with EMT, m6A methylation, and immune checkpoints across various cancers. While these mechanisms often drive tumor progression, our data in PRAD consistently point toward a suppressive role. This highlights the complex, context-specific interplay between oxidative stress and epigenetic regulation. As Sun et al. noted the limitations of their pan-cancer review and emphasized the need for validation in real-world patient cohorts using standardized assays, our study provides such a rigorous assessment specifically for PRAD. The downregulation of NOX4 in PRAD may reflect a unique metabolic or immune environment where its typical oncogenic pathways are superseded by its ability to maintain redox-mediated growth inhibition.

It is important to recognize that NOX4 exhibits context-dependent functions across cancer types. While it acts as a tumor suppressor in hepatocellular carcinoma by maintaining redox homeostasis (33), it promotes metastasis in pancreatic cancer by inducing epithelial-mesenchymal transition (34). This functional heterogeneity may arise from differences in ROS thresholds, cellular environments, and downstream signaling networks. This study offers several novel contributions beyond existing literature on NOX4 in cancer (35,36). First, we provide the first systematic identification of NOX4 as a core diagnostic biomarker for PRAD through integrated WGCNA and machine learning approaches. Second, we report previously unrecognized immune correlation patterns for NOX4 in PRAD—specifically, its negative association with M2 macrophages and positive correlation with neutrophils. Third, we present the first functional evidence that NOX4 acts as a tumor suppressor in PRAD cells, demonstrating that its overexpression inhibits proliferation, migration, and invasion. These findings establish a foundation for investigating the “NOX4–ROS–immune” regulatory axis in PRAD pathogenesis. Nonetheless, a notable limitation of this study is the reliance on the TCGA-PRAD dataset for biomarker discovery. Although our experimental results provide functional support for NOX4, the diagnostic and prognostic performance of the six-gene panel requires further rigorous validation in independent external cohorts (e.g., GEO datasets) and large-scale, multi-center real-world studies.

In conclusion, our study provides novel insights into the molecular underpinnings of PRAD and identifies a robust set of candidate biomarkers for early diagnosis and targeted therapy. The unique immune correlation profile and functional significance of NOX4 underscore its potential as a central regulator in PRAD pathogenesis. Future investigations should aim to dissect the exact molecular mechanisms of NOX4 using both in vitro and in vivo models, paving the way for translational applications.

Conclusions

In summary, this integrative study identifies a six-gene signature with high diagnostic potential for PRAD and functionally characterizes NOX4 as a pivotal tumor suppressor in this malignancy. Our findings demonstrate that NOX4 modulates PRAD cell proliferation and migration, likely through the ROS-dependent regulation of the PI3K/AKT signaling axis, while also exhibiting a distinct correlation with the TME. These results establish a strong scientific foundation for the clinical translation of NOX4 and the identified gene panel as novel biomarkers and therapeutic targets for PRAD.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD and MDAR reporting checklists. Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-943/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-943/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-943/prf

Funding: This work was supported by the National Natural Science Foundation of China (No. 81802569).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-943/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Xu Z, Chen S, Zhang Y, et al. Roles of m5C RNA Modification Patterns in Biochemical Recurrence and Tumor Microenvironment Characterization of Prostate Adenocarcinoma. Front Immunol 2022;13:869759. [Crossref] [PubMed]
2. The Molecular Taxonomy of Primary Prostate Cancer. Cell 2015;163:1011-25.
3. Zhu S, Zhao J, Nie L, et al. Homologous recombination deficiency (HRD) score in aggressive prostatic adenocarcinoma with or without intraductal carcinoma of the prostate (IDC-P). BMC Med 2022;20:237. [Crossref] [PubMed]
4. Gu CY, Dai B, Zhu Y, et al. The novel transcriptomic signature of angiogenesis predicts clinical outcome, tumor microenvironment and treatment response for prostate adenocarcinoma. Mol Med 2022;28:78. [Crossref] [PubMed]
5. Bell PD, Teramoto Y, Gurung PMS, et al. Limited Adenocarcinoma of the Prostate on Needle Core Biopsy. Arch Pathol Lab Med 2022;146:469-77. [Crossref] [PubMed]
6. Cai Q, Meng X, Sun J, et al. Prostate Ductal Adenocarcinoma Revisited: Clinicopathological and Genomic Characterization Identifies Heterogenous Group of Diseases with Implications for Patient Management. Mod Pathol 2025;38:100743. [Crossref] [PubMed]
7. Berglund E, Maaskola J, Schultz N, et al. Spatial maps of prostate cancer transcriptomes reveal an unexplored landscape of heterogeneity. Nat Commun 2018;9:2419. [Crossref] [PubMed]
8. Kido LA, de Almeida Lamas C, Maróstica MR Jr, et al. Transgenic Adenocarcinoma of the Mouse Prostate (TRAMP) model: A good alternative to study PCa progression and chemoprevention approaches. Life Sci 2019;217:141-7. [Crossref] [PubMed]
9. Kanas AF, Fenelon SS, Santos JMMM, et al. Case 297: Mucinous Adenocarcinoma of the Prostate. Radiology 2022;302:238-42. [Crossref] [PubMed]
10. Montironi R, Cimadamore A, Lopez-Beltran A, et al. Morphologic, Molecular and Clinical Features of Aggressive Variant Prostate Cancer. Cells 2020;9:1073. [Crossref] [PubMed]
11. Tolkach Y, Kristiansen G. Cribriform and glomeruloid acinar adenocarcinoma of the prostate-an intratumoural intraductal carcinoma? Histopathology 2019;74:804-8. [Crossref] [PubMed]
12. Nakatani Y, Ohnishi H, Terada J, et al. Multiple Cavitary Lung Metastases from Prostate Adenocarcinoma. Am J Respir Crit Care Med 2023;207:1092-4. [Crossref] [PubMed]
13. Ford K, Hanley CJ, Mellone M, et al. NOX4 Inhibition Potentiates Immunotherapy by Overcoming Cancer-Associated Fibroblast-Mediated CD8 T-cell Exclusion from Tumors. Cancer Res 2020;80:1846-60. [Crossref] [PubMed]
14. Pecchillo Cimmino T, Ammendola R, Cattaneo F, et al. NOX Dependent ROS Generation and Cell Metabolism. Int J Mol Sci 2023;24:2086. [Crossref] [PubMed]
15. Barrios D, Prieto C. Rvisdiff: An R package for interactive visualization of differential expression. Front Bioinform 2024;4:1349205. [Crossref] [PubMed]
16. Chen S, Ben X, Guo L, et al. Identification of lncRNAs based on different patterns of immune infiltration in gastric cancer. J Gastrointest Oncol 2022;13:102-16. [Crossref] [PubMed]
17. Li M, Li X, Chen S, et al. IPO5 Mediates EMT and Promotes Esophageal Cancer Development through the RAS-ERK Pathway. Oxid Med Cell Longev 2022;2022:6570879. [Crossref] [PubMed]
18. Li X, Chen S, Zhu Y, et al. Comprehensive bioinformatics analyses identified Homeobox B9 as a potential prognostic biomarker and therapeutic target for gastric cancer. J Gastrointest Oncol 2021;12:2132-49. [Crossref] [PubMed]
19. Chen S, Li X, Guo L, et al. Characterization of the m6A-related lncRNA signature in predicting prognosis and immune response in patients with colon cancer. J BUON 2021;26:1931-41.
20. Peñuelas-Haro I, Espinosa-Sotelo R, Crosas-Molist E, et al. The NADPH oxidase NOX4 regulates redox and metabolic homeostasis preventing HCC progression. Hepatology 2023;78:416-33. [Crossref] [PubMed]
21. Jiang J, Huang K, Xu S, et al. Targeting NOX4 alleviates sepsis-induced acute lung injury via attenuation of redox-sensitive activation of CaMKII/ERK1/2/MLCK and endothelial cell barrier dysfunction. Redox Biol 2020;36:101638. [Crossref] [PubMed]
22. Fan X, Dong T, Yan K, et al. PM2.5 increases susceptibility to acute exacerbation of COPD via NOX4/Nrf2 redox imbalance-mediated mitophagy. Redox Biol 2023;59:102587. [Crossref] [PubMed]
23. Zhang J, Li H, Wu Q, et al. Tumoral NOX4 recruits M2 tumor-associated macrophages via ROS/PI3K signaling-dependent various cytokine production to promote NSCLC growth. Redox Biol 2019;22:101116. [Crossref] [PubMed]
24. Chung EJ, Reedy JL, Kwon S, et al. 12-Lipoxygenase is a Critical Mediator of Type II Pneumocyte Senescence, Macrophage Polarization and Pulmonary Fibrosis after Irradiation. Radiat Res 2019;192:367-79. [Crossref] [PubMed]
25. Greatorex S, Kaur S, Xirouchaki CE, et al. Mitochondria- and NOX4-dependent antioxidant defense mitigates progression to nonalcoholic steatohepatitis in obesity. J Clin Invest 2023;134:e162533. [Crossref] [PubMed]
26. Song C, Shi D, Chang K, et al. Sodium fluoride activates the extrinsic apoptosis via regulating NOX4/ROS-mediated p53/DR5 signaling pathway in lung cells both in vitro and in vivo. Free Radic Biol Med 2021;169:137-48. [Crossref] [PubMed]
27. Ton TT, Kovi RC, Peddada TN, et al. Cobalt-induced oxidative stress contributes to alveolar/bronchiolar carcinogenesis in B6C3F1/N mice. Arch Toxicol 2021;95:3171-90. [Crossref] [PubMed]
28. Huang YF, Wang G, Ding L, et al. Lactate-upregulated NADPH-dependent NOX4 expression via HCAR1/PI3K pathway contributes to ROS-induced osteoarthritis chondrocyte damage. Redox Biol 2023;67:102867. [Crossref] [PubMed]
29. Wang Y, Su X, Wang Q, et al. Bisphenol A exposure enhances proliferation and tumorigenesis of papillary thyroid carcinoma through ROS generation and activation of NOX4 signaling pathways. Ecotoxicol Environ Saf 2025;292:117946. [Crossref] [PubMed]
30. Wang L, Li L, Zhao D, et al. MYH7 R453C induced cardiac remodelling via activating TGF-β/Smad2/3, ERK1/2 and Nox4/ROS/NF-κB signalling pathways. Open Biol 2024;14:230427. [Crossref] [PubMed]
31. Wang Q, Lin B, Wei H, et al. AQP3 Promotes the Invasion and Metastasis in Cervical Cancer by Regulating NOX4-derived H(2)O(2) Activation of Syk/PI3K/Akt Signaling Axis. J Cancer 2024;15:1124-37. [Crossref] [PubMed]
32. Sun J, Gong Z, Zheng L, et al. NOX4 serves as a pan-cancer prognostic biomarker and therapeutic target in tumorigenesis. Sci Rep 2025;15:24612. [Crossref] [PubMed]
33. Wu Z, Lin X, Ying Y, et al. A dual-targeting strategy to inhibit colorectal cancer liver metastasis via tumor cell ferroptosis and cancer-associated fibroblast reprogramming. Bioact Mater 2025;52:73-91. [Crossref] [PubMed]
34. Jain P, Dvorkin-Gheva A, Mollen E, et al. NOX4 links metabolic regulation in pancreatic cancer to endoplasmic reticulum redox vulnerability and dependence on PRDX4. Sci Adv 2021;7:eabf7114. [Crossref] [PubMed]
35. Kim HK, Lee HY, Riaz TA, et al. Chalcone suppresses tumor growth through NOX4-IRE1α sulfonation-RIDD-miR-23b axis. Redox Biol 2021;40:101853. [Crossref] [PubMed]
36. Meitzler JL, Makhlouf HR, Antony S, et al. Decoding NADPH oxidase 4 expression in human tumors. Redox Biol 2017;13:182-95. [Crossref] [PubMed]