Exploration of oxidative stress-related molecular signature for clear cell renal cell carcinoma

Abstract

Tumorigenesis and progression can be promoted by oxidative stress. However, there is a lack of evidences regarding the expression profiles of oxidative stress-associated genes (OSRGs) in the tumorigenesis and development of clear cell renal cell carcinoma (ccRCC). In the present study, we performed transcriptome analyses on 611 samples from TCGA database and 645 samples from GEO database. Using unsupervised consensus clustering method, we identified two oxidative stress patten (OS_A and OS_B) according to the 278 differentially expressed OSRGs in both TCGA and GEO databases. We further evaluated the prognostic significance, tumor immune microenvironment, and related pathways between two oxidative stress patterns. Subsequently, we identified 7 prognostic OSRGs (CCL7, CDCA3, CRABP2, IRF6, MAGEA4, PLG, SAA1) from both two patterns to incorporate into an OSRG-based prognostic model, and developed a prognostic risk scoring system (OS_score) to distinguish patients with high-risk and poor prognosis. We then develop a miRNA-OSRG regulatory network based on differentially expressed miRNAs and prognostic OSRGs genes, and identify the correlation between OS_score and tumor mutation burden (TMB). Patients from high-risk groups performed an increased expression of immune checkpoint inhibitor (ICI) genes, including PD-1, CTLA-4, B7H3, B7H4, and were more likely to respond to anti-PD-1 immunotherapy. Correlation analysis revealed that AKT inhibitor VIII, EHT-1864 and AS601245 might provide benefits for patients with high OS_score. In conclusion, we conducted a thorough analysis of the expression profiles of OSRGs in ccRCC, culminating in the development of a robust prognostic model and scoring system aimed at accurately predicting survival outcomes for ccRCC patients. This endeavor has the potential to yield novel insights into redox biology and to advance the current treatment strategies for ccRCC.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-04029-2.

Introduction

Renal cell carcinoma (RCC) stands as one of the most common cancers within the urinary system, constituting approximately 2% of all adult malignancies [1]. Clear cell renal cell carcinoma (ccRCC), the most common and aggressive types of RCC [2], was considered high heterogeneity of molecular characteristics [3]. Despite notable advancements have been made in the treatment of ccRCC in recent years, these methods often fail to achieve satisfactory therapeutic effects since the drug resistance [4]. 35% of patients occur metastasis upon initial diagnosis and 30% of early-stage patients still suffer from distant metastasis after surgery intervention [5]. Therefore, it is of great importance to elucidate the molecular mechanisms of the tumorigenesis and development of ccRCC and recognize reliable biomarkers to accurately predict patients’ prognosis.

Oxidative stress, characterized by the destruction of intracellular redox balance and the excessive accumulation of reactive oxygen species (ROS) [6, 7], is responsible for the occurrence and development of various tumors and exerts profound influences on the prognosis [8, 9]. A suitable range of ROS is required for maintaining cell biology. However, during the evolution of tumors, oxidative stress function as a double-edged sword to regulate the tumor progression. On one hand, ROS contributes to oxidative impairment of the normal cell structure, results in lipid peroxidation and accelerates cell death such as apoptosis and ferroptosis, so as to function as an inhibitory factor of tumor. On the other hand, ROS affects energy metabolism, destroys DNA repair mechanisms and thus leads to genomic instability and mutation, so that to function as an exacerbating factor of tumor [10–13]. Emerging evidences have indicated the significant role of oxidative stress in ccRCC progression, however, there is a lack of a comprehensively analysis of molecular characteristic and prognostic value of oxidative stress-related genes (OSRGs).

In this study, we employed data sets comprising 611 specimens sourced from The Cancer Genome Atlas (TCGA) - kidney renal clear cell carcinoma (KIRC) cohort, alongside 649 ccRCC samples gleaned from 6 distinct datasets within the Gene Expression Omnibus (GEO) database. By the unsupervised clustering method, two ccRCC patterns (OS_A and OS_B) were identified according to the expression profiles of OSRGs. Additional analysis revealed that the two patterns exhibited enrichment in distinct signaling pathways and manifested diverse immune microenvironments. A scoring system (OS_score) was established giving that the OSRGs extracted from these two patterns to assess the redox features of ccRCC. The influence of OS_score on survival outcome, clinical characteristic, drug sensitivity, and immunotherapy response were further elaborated.

Materials and methods

Data collection and processing

Transcriptome data containing 72 normal renal tissue specimens and 539 ccRCC specimens and corresponding clinical information were downloaded from the TCGA database (https://portal.gdc.cancer.gov/). Additionally, we acquired 6 datasets from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) that utilized the same microarray sequencing platform (Affymetrix GPL570-HGU133 plus 2.0), including GSE53757 (n = 144), GSE46699 (n = 130), GSE66270 (n = 28), GSE66272 (n = 54), GSE36895 (n = 76), GSE73731(n = 265). Background adjustment and quantile normalization of the included datasets were realized through the “RMA” algorithm using R package “Affy”. The batch effects of merging datasets were then removed by the “ComBat” algorithm using R package “SVA”. Moreover, clinical information and transcriptome data of patients with duplicate samples, censored data, transcriptome data expressing values less than 1, or patients with overall survival (OS) ≤ 90 days were removed.

Identification of CcRCC subclasses based on differentially expressed OSRGs

Using the search term “oxidative stress,” we retrieved 9469 human genes associated with oxidative stress from the OMIM database (https://www.oncomine.org/resource/), the NCBI gene function module (https://www.ncbi.nlm.nih.gov/gene/), and the GeneCard database (https://www.genecards.org/). Next, the differential expressed OSRGs were obtained via the R package “DESeq2” by the predefined criteria |log2 fold change (FC)| ≥2.0, P < 0.05 for TCGA-KIRC cohort and |log2 FC| ≥1.2, P < 0.05 for GEO datasets, respectively. Subsequently, the overlapping OSRGs that differentially expressed between normal and ccRCC samples were subjected to further analysis. Two OS patterns were recognized by performing unsupervised consensus clustering for these OSRGs using the k-means algorithm from R package “ConsensusClusterplus”. The results were repeated 1000 times to ensure the stability of classification [14].

Gene set variation analysis (GSVA) and function enrichment analysis

The gene set “c2.cp.kegg.v7.2.symbols” and “h.all.v7.4.symbols” were obtained from the MSigDB database. Subsequently, GSVA analysis and differential analysis were realized through the R packages “GSVA” and “Limma” respectively, to identify the variations in biological processes between two OS patterns. Furthermore, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes database (KEGG) enrichment analyses were conducted to explore the biological functions and involved pathways of these differentially expressed OSRGs.

Comparison of infiltrated immune cells and immune microenvironment between distinct OS patterns

The cell populations of immune cells in distinct OS patterns were evaluated using CIBERSORT algorithm and attached LM22 gene set [15]. CIBERSORT is a deconvolution approach that evaluates the relative abundance of immune cell population in each ccRCC patient by analyzing the expression profiles of 22 tumor-infiltrating lymphocyte subsets. The permutation tests were run 1000 times to stabilized the results.

The immune microenvironment characteristics between two OS patterns were assessed using the “ESTIMATE” algorithm which employed the expression profiles of each tumor samples to infer the components of normal cells (stromal cells and immune cells) in tumor based on the single sample Gene Set Enrichment Analysis (GSEA).

Construction and validation of OSRG-based prognostic model

By employing least absolute shrinkage and selection operator (LASSO) regression analysis and multivariate Cox regression analysis, we discerned the differentially expressed OSRGs exhibiting independent prognostic significance.

The risk score related to oxidative stress (OS_score) for each patient was calculated using the following formula:

OS_score = .

Of this, Exp represents the expression value of each gene, β represents the corresponding regression coefficient. All patients from TCGA-KIRC cohort were allocated into low-risk and high-risk subgroups based on the median risk score. And the Kaplan–Meier method and log-rank test were used to compare the differences of the overall survival (OS) between two risk groups. In addition, we employed time-dependent receiver operating characteristic (ROC) method to analyze the predictive power of the OSRG-based prognostic model. Moreover, the E-MTAB-1980 cohort containing 101 samples was served as an external validation dataset to evaluate the stability of the model.

Correlation between the OSRG-based prognostic model and clinical parameters

We analyzed the correlation between OS_score and survival outcomes of ccRCC patients in the TCGA-KIRC cohort. Kaplan-Meier survival analysis was applied to evaluate the disparity in prognosis between the high OSRG-based risk group and the low OSRG-based risk group across various clinical parameters stratifications, such as age, clinical stage and tumor grade. Subsequently, multivariable Cox regression analysis was used to ascertain the independence of OS_score and clinical parameters in predicting patients’ prognosis of ccRCC from the TCGA-KIRC cohort and E-MTAB-1980 cohort.

miRNA-OSRG regulatory network and CcRCC mutation analysis

miRNA expression data of ccRCC was retrieved from the TCGA database. Differentially expressed miRNAs between normal and ccRCC samples were identified according to the criteria (|log2 FC| >1.5, P < 0.05). Co-expression analysis of the identified miRNAs and prognostic OSRGs, and regulatory pairs with the criteria of |Cor| >0.3 and P < 0.001 were utilized to construct the miRNA-OSRG regulatory network.

Based on the somatic mutation data, the tumor mutation burden (TMB) was computed using R package “maftools”. Further analysis involved examining the difference in TMB levels across different risk groups and assessing its correlation with OS_score. Additionally, the landscape of driver gene mutations within the low- and high-risk groups was delineated.

Comparison of immunotherapy responsiveness between two risk groups

The expression difference of immune checkpoint inhibitor (ICI)-targeted genes such as PD-1, CTLA-4, B7H3 and B7H4 between the two risk groups was quantified. Meanwhile, the influence of OS-based risk stratification on the prognosis of ccRCC patients was evaluated while the expression of above ICI-targeted genes were taken into consideration.

We then applied the algorithm of Tumor Immune Dysfunction and Exclusion (TIDE) to elucidate the relationship between immunotherapy responsiveness and OS_score. Using the SubMap approach, we also investigated the similarity of gene expression patterns between two risk stratifications based on the open-access melanoma receiving immunotherapy [16]. Furthermore, survival analysis was conducted between the two risk stratifications using data from a public metastatic melanoma cohort (Riaz), where patients had undergone treatment with anti-PD-1 immunotherapy. This analysis aimed to validate whether patients classified in the high-risk group exhibited a positive response to immunotherapy.

Identification of available drug for patients with high-risk

The transcriptional data of RCC cell lines, IC50 values of various drugs, and drug-involved targets/pathways were obtained from the Genomics of Drug Sensitivity in Cancer (GDSC) database (https://www.cancerrxgene.org/). Pearson correlation analysis between the OS_score and drug sensitivity were conducted to seek for available drugs for ccRCC patients classified as high-risk, with results reported based on the criteria of |Rs|>0.15 and P < 0.05. Eventually, the targets/pathways regulated by above identified drugs were presented.

Cell culture

The human renal tubular epithelial cell line HK-2 and the clear cell renal cell carcinoma (ccRCC) line 786-O were acquired from Procell Life Science & Technology Co., Ltd. (Wuhan, China). HK-2 cells were grown in DMEM/F12 medium (Boster, China) containing 10% fetal bovine serum (FBS; Gibco, USA) and 1% penicillin‒streptomycin (Servicebio, China). The 786-O cells were cultured in RPMI-1640 complete medium (Boster, China). Both cell lines were maintained at 37 °C in a 5% CO₂ humidified incubator.

RT-qPCR

Cellular total RNA was isolated with the SteadyPure Universal RNA Extraction Kit (Accurate Biology, AG21017, China), following the protocols provided by the manufacturer. For reverse transcription, 1 µg of purified RNA per sample was converted into cDNA with the Evo M-MLV RT Premix for qPCR (Accurate Biology, AG11706, China). Sequence-specific primers (see below) were employed in quantitative real-time PCR carried out using SYBR Green Premix Pro Taq HS qPCR Kit (Accurate Biology, AG11701, China) to quantify mRNA expression. β-Actin served as the internal reference gene, and the 2^−ΔΔCt method was applied to compute relative gene expression levels.

Western blotting

Protein extraction from cultured cells was carried out using RIPA lysis buffer (Beyotime, China) containing protease and phosphatase inhibitors. The protein samples were then denatured by boiling in 1× loading buffer and equal quantities of protein were resolved by SDS-PAGE. Subsequently, the separated proteins were electrophoretically transferred onto PVDF membranes. After blocking, the membranes were probed with specific primary antibodies and corresponding HRP-conjugated secondary antibodies. Signal detection was performed using an enhanced chemiluminescence substrate, and protein bands were visualized with a chemiluminescence imaging system.

CCK-8 assay

Cells were seeded into 96-well plates according to the experimental design. Each well was supplemented with 100 µL of serum-free medium containing 10 µL of CCK-8 reagent (Boster, China), and the plates were incubated for 2–4 h. Absorbance at 450 nm was measured using a microplate reader.

Colony formation assay

Cells were transfected with siCRABP2 or vector and incubated for 24 h. Subsequently, they were cultured for an additional 14 days to allow colony formation. After incubation, the medium was removed, and the cells were fixed with methanol followed by staining with 0.1% crystal violet (Servicebio, Wuhan, China).

Transwell assay

For the Transwell migration assay, 4 × 10⁴ transiently transfected cells were resuspended in serum-free medium and loaded into the upper chamber, while the lower chamber was filled with complete medium. After 48 h of incubation at 37 °C with 5% CO₂, the cells were washed with PBS, fixed with paraformaldehyde, and stained with 0.1% crystal violet. The stained cells were imaged under a microscope and counted. All experiments were performed in triplicate to ensure reproducibility.

Wound healing assay

When cells reached 80–90% confluence, a straight wound was created using a sterile pipette tip. The culture medium was then replaced with serum-free medium. Images of the wound were captured at predetermined time intervals, and the wound area was quantified using ImageJ software.

Immunohistochemistry data from HPA

Immunohistochemistry images depicting gene protein expression in clear cell renal cell carcinoma(ccRCC) and normal tissue samples were obtained from the Human Protein Atlas (HPA) database (https://www.proteinatlas.org).

Measurement of intracellular reactive oxygen species (ROS)

Intracellular ROS levels were measured using a DCFH-DA probe (Beyotime, China) according to the manufacturer’s instructions. Briefly, cells were seeded into 96-well plates and incubated overnight. The DCFH-DA solution was diluted with PBS (1:1000) to prepare a working solution, which was then added to each well. After incubation at 37 °C for 20 min in a humidified incubator, cells were washed gently with PBS to remove excess probe. Fluorescence intensity was measured using a microplate reader at an excitation wavelength of 488 nm and an emission wavelength of 525 nm, representing the intracellular ROS level.

Detection of intracellular superoxide anion (O₂⁻•)

Intracellular superoxide anion levels were detected using a Reactive Oxygen Species Assay Kit for Superoxide Anion with DHE (Beyotime, China) following the manufacturer’s protocol. After treatment, cells were incubated with DHE working solution at 37 °C for the recommended time. The fluorescence signal was recorded using a microplate reader with excitation/emission wavelengths set to 535/610 nm. The relative fluorescence intensity was used to reflect intracellular superoxide anion production.

Statistical analyses

All datasets were subjected to statistical analysis using GraphPad Prism. Continuous variables are reported as mean ± standard deviation. After verifying normality distribution, inter-group comparisons were performed with parametric tests: the unpaired Student’s t-test for two groups or one-way ANOVA for multiple groups. The threshold for statistical significance was established at P < 0.05.

Results

Identification of differentially expressed OSRGs in CcRCC

The study analysis procedure was presented in the flow chart (Fig. 1). Firstly, we obtained a total of 9469 OSRGs from the public databases, and conducted differentially expression analysis comparing ccRCC tumor to normal renal samples. According to the inclusion criteria mentioned above, we ultimately identified 278 overlapping OSRGs that abnormally expressed in both the TCGA and GEO databases. We next performed unsupervised consensus clustering analysis of these 278 differentially expressed OSRGs based on the TCGA-KIRC cohort (n = 539) and merged GEO cohort (n = 460). After normalization and batch correction, PCA analysis demonstrated that the batch effects among different GEO datasets were effectively removed (Figure S1a, b), ensuring the comparability of subsequent analyses.After integrative analysis of CDF curve and delta area, k = 2 was regarded as the optimum cluster number to categorize the ccRCC patients into two OS patterns, with the boundary between the consensus matrix heatmaps remains clear and sharp (OS_A and OS_B) (Figs. 2a–c, S1c–e). Ultimately, 416 patients from the TCGA cohort were allocated to OS_A, and 123 ones were allocated to OS_B. Regarding the GEO cohorts, 402 patients were assigned to OS_A, with the remaining 58 allocated to OS_B. To further validate the robustness of subclass grouping, t-SNE dimension reduction was conducted, and found that the t-SNA distribution was in accordance with the performance from k-means clustering (Figs. 2d, S1f). Survival analysis indicated that patients from OS_A group performed superior overall survival (OS) than those from OS_B group (P = 0.0068, Fig. 2e). However, the underlying mechanisms of oxidative stress in affecting the prognosis of ccRCC patients deserves further exploration.

Fig. 1.

The flowchart for analyzing OSRGs molecular patterns and calculating OS_score for ccRCC

Fig. 2.

Identification of OSRGs patterns for ccRCC using unsupervised consensus clustering approach based on TCGA-KIRC cohort. a Matrix heat map of k-means clustering based on 278 differentially expressed OSRGs. b Consensus clustering cumulative distribution function (CDF) for k = 2 to 9. c Delta area of k-means clustering for k = 2 to 9. d The two-dimensional distribution characteristic of t-SNE at k = 2. e Kaplan-Meier survival analysis of ccRCC patients from two oxidative stress pattern

Molecular and tumor microenvironment features of OS patterns in CcRCC

GSVA enrichment and pathway difference analyses were conducted to explore the molecular mechanism of OS patterns in ccRCC. We found that OS_A was predominantly enriched in vary energy metabolism signals, such as inositol phosphate metabolism, ascorbate and aldarate metabolism, beta-alanine metabolism, tryptophan metabolism, fatty acid metabolism and pyruvate metabolism, while the OS_B was predominantly enriched in tumor-promoting signals, such as epithelial-mesenchymal transition (EMT), DNA replication, and Myc targets (Fig. 3a). We further conducted GO and KEGG enrichment analysis in two subgroups. As a result, the enriched biological processes were response to peptide, small molecule catabolic process for OS_A, and response to oxidative stress, positive regulation of cell adhesion for OS_B. The cell components localize in apical part of cell, apical plasma membrane for OS_A, and vesicle lumen, cytoplasmic vesicle lumen for OS_B. In light of molecular functions, OS_A were enriched in protein serine/threonine/tyrosine kinase activity, protein serine/threonine kinase activity, and OS_B were enriched in electron transfer activity and oxidoreductase activity, acting on NAD(P)H (Fig. 3b, c). KEGG analysis revealed that OS_A was enriched in cAMP signaling, FoxO signaling, and MAPK signaling, while OS_B was enriched in cell cycle, p53 signaling, and cytokine and cytokine receptor interaction (Figure S2a-b).

Fig. 3.

Molecular and tumor microenvironment features of OS patterns in ccRCC patients. a Heatmap visualization of the GSVA enrichment analysis on oxidative stress patterns according to the Hallmark and KEGG pathway features from MSigDB. b GO enrichment analysis of OS_A pattern. c GO enrichment analysis of OS_B pattern. d The immune cell populations between OS_A and OS_B patterns based on CIBERSORT algorithm. e–g Boxplots to display the immune score, ESTIMATE score, and tumor purity between OS_A and OS_B patterns

Considering the immune microenvironment exerts great influences on tumorigenesis and progression, we further explore the difference of infiltrated immune cell populations between two subgroups. Applying CIBERSORT algorithm, we found that the predominant cell populations in OS_A pattern were resisting memory CD4⁺ T cells, monocytes, and M1 macrophages, while those in OS_B pattern were plasma-cells, M0 macrophage, and M2 macrophages (Fig. 3d).

Furthermore, the tumor microenvironment of each ccRCC samples from two OS subgroups was evaluated by ESTIMATE algorithm. The results suggested that the immune score and ESTIMATE score in OS_A pattern (278.474 ± 440.8, and 572.718 ± 681.961, respectively) were significantly lower than those in OS_B pattern (446.882 ± 437.962, 784.335 ± 703.133, respectively), while the tumor purity in OS_A pattern (0.768 ± 0.062) were significantly higher than those in OS_B pattern (0.748 ± 0.066) (Fig. 3e–g).

Construction of a prognostic model derived from the differentially expressed OSRGs

Furthermore, we explored the difference of expression profiles between these two patterns. Differentially expression analysis was conducted on the basis of the filtering criteria of |log2 FC| >1.2 and P < 0.05 to distinguish the diversity of the two patterns. Eventually, we identified 331 differentially expressed OSRGs, and then performed unsupervised consensus clustering analysis based on the above OSRGs. After integrative analyze the CDF curve and delta area, k = 2 was selected as the optimum cluster number (Fig. 4a–c). The results of t-SNE distribution confirm the robustness of the classification (Fig. 4d), and the all ccRCC patients were categorized into two individual OS gene subgroups (Genecluster_A and Genecluster_B). GO analysis indicated that the identified OSRGs were predominantly involved in copper stress response and ligand-receptor activation signaling (Fig. 4e). Survival analysis revealed that patients form Genecluster_A group had a worse prognosis than those from Genecluster_B group (Fig. 4f).

Fig. 4.

Identification of the OSRGs expression profiles in ccRCC using unsupervised consensus clustering based on TCGA cohort. a Matrix heat map of k-means clustering based on 331 differentially expressed OSRGs. b Consensus clustering CDF for k = 2 to 9. c Delta area of k-means clustering for k = 2 to 9. d The two-dimensional distribution characteristic of t-SNE at k = 2. e GO enrichment analysis of 331 differentially expressed OSRGs. f Kaplan-Meier survival analysis of ccRCC patients from two gene characteristic patterns. g Multivariable Cox regression analysis to identify the OSRGs with independent prognostic values. h, i Comparison of the OS_score in oxidative stress patterns and gene characteristic pattens

Subsequently, 221 OSRGs were identified by univariate Cox regression analysis on the basis of the filtering criteria of P < 0.01. These genes were subjected to further Kaplan-Meier survival analysis to research the influence of these gene expression on patients’ prognosis, and thus we acquired 177 OSRGs with prognostic values. Then, we screened out the 18 OSRGs with most significant prognostic values by LASSO regression analysis. The results of cross-validation and the trajectory changes of the 18 independent variables are shown in Figure S2(a) and S2(b). Eventually, we identified 7 genes as independent prognostic values for ccRCC by multivariable Cox regression analysis (Table S5 and Fig. 4g. The risk score for each ccRCC patient (OS_score) was calculated according to the following formula: OS_score = (0.0924 × Exp CCL7) + (0.1837 × Exp CDCA3) + (0.1174 × Exp CRABP2) + (−0.1122 × Exp IRF6) + (0.1516 × Exp MAGEA4) + (−0.0590 × Exp PLG) + (0.0557 × Exp SSA). Moreover, a significantly elevated OS_score was observed in the OS_B group and Genecluster_B group (all P < = 2e-16) (Fig. 4h and i), indicating that the above stratification approaches were able to distinguish patients with diverse oxidative stress status.

Evaluation and validation of the performance and stability of OS-based prognostic model

A median OS_score was used to stratify 539 ccRCC patients in the TCGA-KIRC cohort into high-risk and low-risk subgroups. The Kaplan-Meier survival analysis revealed that patients in the high-risk group had a worse prognosis than those in the low-risk group (P = 5.412e-13) (Fig. 5a). Then ROC analysis was performed to evaluate the predictive performance of the model, and the predicted area under the ROC curve (AUC) was 0.794 at 1-year, 0.713 at 3-years and 0.727 at five years (Fig. 5b). Moreover, external validation data acquired from the GSE46602 cohort was used to confirm the stability of the prognostic model. Survival analysis revealed a poor prognosis existed in patients with high-risk (Fig. 5c), and the predictive AUC value was 0.794 at 1-year, 0.813 at 3-years and 0.836 at 5-years (Fig. 5d).

Fig. 5.

Construction of a prognostic model based on the differentially expressed OSRGs. a Kaplan–Meier survival analysis between two OSRG-based risk stratifications from the TCGA-KIRC cohort. b ROC curve to evaluate the performed of prognostic model based on the TCGA-KIRC cohort. c Kaplan–Meier survival analysis between two OSRG-based risk stratifications from the GSE46602 cohort. d ROC curve to evaluate the performed of prognostic model based on the GSE46602 cohort. e Multivariate Cox regression analysis of age, stage, grade, and OS_score to identify prognostic factors for ccRCC based on the TCGA-KIRC cohort. f Multivariate Cox regression analysis of age, stage, grade, and OS_score to identify prognostic factors for ccRCC based on the GSE46602 cohort

In order to assess the independent predictive capability of the OS_score on patient outcomes, a multivariate Cox analysis was conducted, incorporating common clinical characteristics such as age, stage, grade, and OS_score. The findings revealed that the OS_score emerged as an independent and robust prognostic indicator for ccRCC patients (HR = 0.43, 95% CI 0.3–0.62, P < 0.001, Fig. 5e). Moreover, to ascertain the generalizability of the model across diverse ccRCC patient cohorts, the identical formula was applied to compute the risk score for patients within the GSE46602 cohort. These analyses demonstrated the favorable predictive performance and stability of the OSRGs-based prognostic model (Fig. 5f). To verify whether the prognostic significance of OSRG clusters was independent of clinical features, we conducted a multivariate Cox regression analysis (Figure S4) including age, stage, and grade. The results demonstrated that the OS_cluster classification remained an independent prognostic factor (HR = 1.4, 95% CI: 1.07−1.90, P = 0.015 *), suggesting that the predictive power of the OSRG-based classification was not driven by clinical covariates.

Assessment of the OS_score’s clinical correlation and drawing a miRNA-RRG regulatory networks

Firstly, we investigated the correlation between OS_score and ccRCC patients’ survival state. As a result, patients categorized within the high-risk group exhibited a greater mortality rate compared to those classified in the low-risk group. (47.21% verse 14.44%, P < 0.001) (Fig. 6a). Likewise, the dead patients with ccRCC demonstrated a significantly elevated OS_score than the surviving patients (1.834 ± 0.868 verse 1.031 ± 0.66, P < 0.001) (Fig. 6b). Then, ccRCC patients of the TCGA cohort were stratified based on common clinical variables such as clinical stage and tumor grade, and we found that the low-risk group from each clinical stratification performed a significantly survival advantage when compared with the high-risk group (Fig. 6c–f).

Fig. 6.

Assessment of the OS_score’s clinical correlation and drawing a miRNA-RRG regulatory networks. a Comparison of death rate between two risk groups. b Comparison of OS_score between alive and dead patients. c–f Kaplan-Meier survival analysis between two risk groups after stratifying by stage 1–2, stage 3–4, grade 1–2, and grade 3–4. g Heatmap displayed the differentially expressed miRNAs between normal renal and ccRCC tumor samples. h Sankey plot to display the miRNA-OSRG regulatory network

Interaction between ROS homeostasis and miRNA has been demonstrated in many disorders [17]. Therefore, it may bring great benefits to underlie the regulatory relationship between miRNAs and OSRGs. Herein, we performed differentially analysis of miRNA expression profiles from the TCGA-KIRC cohort, and 126 up-regulated miRNAs and 129 down-regulated miRNAs were acquired based on the criteria of |log2 FC| >1.5 and P < 0.05. The differentially expressed miRNAs were illustrated in the heat map (Fig. 6g). Co-expression analysis was implemented to explore the relationship between these miRNAs and the 7 prognostic OSRGs, resulting in the identification of 17 miRNA-OSRG regulatory pairs that met the requirements of cor >0.3 and P < 0.001(Table S4). The regulatory network was visualized in the form of Sanky drawing (Fig. 6h).

Relationship between the OS_score with mutation profile in CcRCC patients

We proceeded to assess the correlation between the 7 prognostic OSRGs and somatic mutation profiles. Patients categorized in the high-risk group exhibited significantly elevated TMB levels (P = 0.044) (Fig. 7a), with a positive association observed between OS_score and TMB levels (R = 0.23, P = 1.4e-05) (Fig. 7b). We next measured whether TMB level affect the prognosis of ccRCC patients, revealing that patients with high TMB levels experienced worse outcome when compared with those with low TMB levels (P = 0.002) (Fig. 7c). Additionally, patients with low-risk showed a significantly survival advantage in OS than those with high risk, without considering the influence of TMB levels (P < 0.0001) (Fig. 7d). Additionally, we identified driver genes mutated in at least 5% of tumor samples from both low-risk and high-risk groups. Further survival analyses focused on 3 driver genes (VHL, PBRM1, and TTN) with the highest mutation frequencies to explore whether OS-based risk stratification retained prognostic value when considering common driver genes mutation profiles. The results indicated that patients with a VHL-mutated phenotype and high-risk exhibited worse outcomes than those with a VHL-mutated phenotype and low-risk, while the patients with a VHL-wild phenotype and high-risk displayed worse outcomes compared to those with a VHL-wild phenotype and low-risk (P < 0.0001) (Fig. 7g). Similar observations were noted in the case of other driver genes (Figs. 7h, i). Overall, we found that OS_score still has a preferable prognostic value when considering the driver gene mutations.

Fig. 7.

Relationship between the OS_score with mutation profile in ccRCC patients. a Comparison of TMB levels between two risk groups stratified by OS_score. b Correlation analysis between OS_score and TMB level. c Kaplan-Meier survival analysis to compare the prognosis difference between low- and high-TMB groups. d Kaplan-Meier survival analysis to compare the prognosis differences among four groups stratified by the OS_score and TMB. e Waterfall plot to illustrated the mutation profiles in ccRCC patients from the low-risk group. f Waterfall plot to illustrated the mutation profiles in ccRCC patients from the high-risk group. g–i Kaplan-Meier survival analysis to compare the prognosis differences among four groups stratified by the OS_score and driver genes (VHL, PBRM1, and TTN) mutation status

The association between OS_score and immunotherapy responsiveness

Immunotherapy targeting ICI genes have renovated the current treatment paradigm of cancer [18]. Currently, we explored the expression levels of the crucial ICI-targeted genes (PD-1, CTLA-4, B7-H3 and B7-H4) between two risk subgroups. It is shown that the expression of all of these crucial ICI-targeted genes were markedly up-regulated in the patients belonging to the high-risk group (Fig. 8a–d). In addition, survival analysis indicated that patients with high-risk had a worse prognosis than those with low-risk, even when the expression of these ICI-targeted genes was considered. These finding was consistence with the results of Sun et al.’s [19] and Varn et al.’s [20], suggesting that the overexpression of ICI-targeted was correlated with poorer prognosis (Fig. 8e–h).

Fig. 8.

Benefit of OS_Score in predicting immunotherapy responsiveness. a–d Comparison of expression levels of ICI-targeted genes (PD-1, CTLA-4, B7H3, and B7H4) between two risk groups. e–h Kaplan-Meier survival analysis to compare the prognosis differences among four groups stratified by the OS_score and ICI-targeted gene (PD-1, CTLA-4, B7H3, and B7H4) expression levels. i Estimation the immunotherapy responsiveness between two risk groups. j Comparison of the difference in TIDE prediction score between two risk groups. k SubMap analysis to estimate the immune responsiveness based on the similarity of expression profiles of oxidative stress patterns between TCGA-KIRC cohort and ICI-treated melanoma patients. l Kaplan-Meier survival analysis to compare the survival prognosis between two risk groups in the Riaz’s cohort

Due to the absence of publicly available ccRCC cohorts treated with immune checkpoint inhibitors, we applied the TIDE algorithm to infer potential differences in immunotherapy response between the two risk groups.The results revealed that the immunotherapy response rate of high-risk to low-risk was 34.6% to 47.5% (c² = 9.169, P = 0.002, Fig. 8i). Moreover, patients with high-risk exhibited an increased TIDE prediction score compared to those with low-risk (P = 5.9e-06) (Fig. 8j).

To further explore the potential immunotherapy responsiveness, we performed SubMap analysis comparing the expression profiles of the two risk groups with an open-access melanoma cohort treated with ICIs. The findings indicated that the transcriptional pattern of the high-risk group was associated with a profile resembling anti–PD-1–responsive tumors. (Fig. 8k). Consistently, in a metastatic melanoma cohort treated with anti–PD-1 therapy, patients classified in the high-risk group exhibited poorer survival compared with those in the low-risk group (P = 0.026) (Fig. 8l). Collectively, these associations suggest a potential predictive utility of the OS_score in reflecting immune activity and putative immunotherapy responsiveness rather than a definitive predictive role.

Correlation between OS_score and drug sensitivity

In order to explore the available drugs for ccRCC patients in different risk groups, we conducted person correlation analysis between ORRG-based OS_score and IC50-based drug sensitivity in the GDSC database. Ultimately, high OS_score was associated with the drug sensitivities of 26 agents, such as MK (Rs = − 0.41, P = 1.85e-23), Cisplatin (Rs = − 0.40, P = 6.95e-22), Docetaxel (Rs = − 0.37, P = 1.18e-18), while low OS_score was likely correlated with the drug sensitivities of 10 agents, such as AKT inhibitor VIII (Rs = 0.34, P = 2.15e-15), EHT-1864 (Rs = 0.31, P = 3.52e-13) and AS601245 (Rs = 0.28, P = 8.89e-11) (Fig. 9a). We next identified the associated signaling pathways of these drug targets, and found that the sensitive drugs were mainly enriched in protein stability and degradation, p53 pathway, JNK and p38 signaling, genome integrity, and DNA replication, while the resistance drugs were mainly implicated in EGFR signaling, apoptosis regulation and PI3K/MTOR signaling (Fig. 9b). These findings implied that the OS_score could predict the drug sensitivity, which might guide the personalized treatment for ccRCC patients.

Fig. 9.

Correlation between OS_score and drug sensitivity. a Correlation analysis between OS_score and drug sensitivity based on the GDSC database. b Identifying the involved pathways of the available drugs

CRABP2 knockdown impairs proliferation, migration and invasion in renal cancer cells

The results showed that CRABP2 was significantly upregulated in 786-O and Caki-1 cells compared with normal renal tubular epithelial HK2 cells in m RNA level(Figure S5 a&b). Combined with the multivariable Cox regression (Table S1) and biological relevance, CRABP2 was ultimately chosen for subsequent functional experiments.

Based on prior bioinformatics analysis identifying seven candidate genes, we validated their mRNA expression levels in the human renal tubular epithelial cell line HK-2 and the clear cell renal cell carcinoma (ccRCC) line 786-O using quantitative real-time PCR (qRT-PCR) (Fig. 10a). Immunohistochemistry (IHC) images of CRABP2 expression in both normal kidney and renal carcinoma tissues were obtained from the Human Protein Atlas (HPA) database (Fig. 10b).

Fig. 10.

CRABP2 is overexpressed in ccRCC and promotes proliferation, migration, and invasion in vitro. a qRT-PCR analysis of the mRNA expression levels of CRABP2 in the human renal tubular epithelial cell line HK-2 and the ccRCC cell line 786-O. b Representative immunohistochemistry (IHC) images of CRABP2 expression in normal kidney and renal carcinoma tissues from the Human Protein Atlas (HPA) database. c, d CRABP2 knockdown efficiency was confirmed at the protein (c) and mRNA (d) levels in renal cancer cell lines. e CCK-8 assays. f Colony formation assays demonstrated that CRABP2 knockdown impaired the colony-forming ability of 786-O cells. g, h Transwell invasion assays and their quantitative analysis. i, j Scratch wound healing assays (i) and their quantitative analysis (j). Data are presented as mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001 vs. control group

Subsequently, we performed CRABP2 knockdown in renal cancer cell lines for further in vitro validation. Successful knockdown was confirmed at both the protein and mRNA levels (Fig. 10c, d). CCK-8 assays revealed that CRABP2 knockdown significantly attenuated cell proliferation, indicating a crucial role for CRABP2 in renal cancer proliferation (Fig. 10e). This finding was further corroborated by colony formation assays, which showed that knockdown significantly impaired colony-forming ability (Fig. 10d). Collectively, these findings demonstrate that CRABP2 is a key regulator of cell proliferation and may serve as an important prognostic biomarker for patients.

We next assessed the impact of CRABP2 knockdown on cell migration and invasion using scratch wound healing assays and Transwell assays (Fig. 10g, i). Figure 10h, i represent the statistical results of the Transwell assay and scratch assay, respectively.In the scratch assay, CRABP2 knockdown significantly reduced the wound closure capacity of 786-O cells compared to control cells, indicating markedly impaired cell migration. Similarly, in the Transwell assay, the number of 786-O cells migrating through the membrane to the lower chamber was significantly decreased in the knockdown group compared to the control group. This reduction in cell migration through the membrane pores indicates a significant inhibition of both migratory and invasive capabilities. In addition, DCFH-DA and DHE were performed to evaluate oxidative stress levels(Figure S5 c&d),. CRABP2 knockdown markedly reduced intracellular ROS and superoxide accumulation suggesting that CRABP2 may modulate oxidative stress in ccRCC cells.Together, these results strongly suggest that CRABP2 plays a critical role in promoting cell migration and invasion in renal cancer.

Discussion

Accumulating studies have confirmed the important role of oxidative stress in the occurrence and progression of ccRCC [21, 22]. Oxidative stress is defined as the dysfunction of redox homeostasis, with a bias toward oxidation, contributing to the generation of active free radicals and metabolic by-products. Oxidative stress plays a dual role in the ccRCC progression. For one thing, accumulation of oxidative stress products, mainly the ROS, it responsible for the induction of cell survival, proliferation and migration of ccRCC cells [23]. For another, ROS-driven lipid peroxidation can induce ferroptosis and thus inhibit tumor growth of ccRCC [24]. Most previous studies have just considered the function of a single gene, however, a comprehensive analysis of the oxidative stress status based on the expression profiles might provides novel insights into the host redox homeostasis and molecular features of ccRCC.

In the present study, we have identified two oxidative stress patterns based on the differentially expressed OSRGs between normal renal and ccRCC tumor samples. Survival analysis revealed that patients from OS_A group performed better prognosis than those from OS_B group. Mechanically, the OS_A subclass was primarily enriched in various energy metabolism signals, including inositol phosphate metabolism and beta-alanine metabolism, while the OS_B subclass was predominantly enriched in tumor-promoting signals, such as epithelial-mesenchymal transition (EMT) and Myc targets. EMT is a complex process that promotes epithelial cell to transit into a mesenchymal phenotype, with tumor-promoting properties such as evading death, migration, metastasis and drug resistance [25]. Myc is an important oncogene and the activation of Myc pathway were responsible for the cell proliferation of ccRCC [26]. Meanwhile, HIF and c-Myc could serve as the biomarkers for the sunitinib responsiveness in metastatic ccRCC [27]. Different immune cell populations occurred in these two groups, with a predominant accumulation of memory CD4⁺ T cells and monocytes in OS_A group, while an abundance of plasma-cells and M0 macrophage in OS_B group. CD4 + T cells plays a crucial role in anti-tumor activities to enhance CD8 + cell activation and secrete various cytokines. Applying a CD4⁺ T cell-based adoptive immunotherapy approach resulted in the GSH depletion and ROS accumulation, and thus promoted oxidative stress and cell death in a TNF-α-dependent manner [28]. In contrast, elevated infiltration of immunoglobulin-secreting plasma cell, the predominant immune cell components of OS_B groups, were correlated with poor outcomes of patients with ccRCC or glioblastoma [29].

Subsequently, the differentially expressed OSRGs were subjected to multivariable Cox regression analysis to obtain prognostic-related genes. Ultimately, 7 genes (CCL7, CDCA3, CRABP2, IRF6, MAGEA4, PLG, SAA1) with important prognostic values for ccRCC were identified. Among them, IRF6 and PLG were served as the protective factors, while CDCA3, CRABP2, MAGEA4 and SAA1 were served as the exacerbated factors. Downregulation of IRF6 due to DNA hypermethylation was observed in ccRCC tissues, which is positive correlated with the poor clinicopathological characteristics and survival outcome [30]. Suppression of transcriptional factor IRF6 decreased the expression of KIF20A and thus prohibited cell proliferation, invasion and migration of ccRCC tumor cells [31]. Upregulation of CDCA3 by lncRNA SNHG12 accelerated the cell cycle and resulted in tumor progression and sunitinib resistance of RCC [32], and it was considered to be a favorable indicator of poor prognosis in RCC [33]. CRABP2 is involved in the retinoid partitioning and widely accepted to promote tumor progression by anoikis resistance, migration, and invasion [34]. Upregulation of CRABP2 contributed to a poor outcome and advanced clinal and tumor grade in endometrial cancer [35]. MAGE-A4 is a cancer testis antigen that widely overexpressed in various tumor types, such as synovial sarcoma [36] and serous ovarian cancer [37]. T cell immunotherapy targeting MAGE-A4 have been applied in multiple clinical trials for treating synovial sarcoma, and the results revealed a preferable and durable response rate [36]. Owing to the hypomethylation of promoter region, upregulation of SAA1 were observed in tumor samples from advanced ccRCC, and was positively correlated with a poor prognosis. It promoted sunitinib resistance via activating STAT3 pathways and inducing drug efflux [38].

We then formulated a prognostic model utilizing the identified OSRGs and established a risk scoring system (OS_score) to distinguish patients with high-risk. Survival analysis revealed that patients from high-risk group suffered from a worse outcome than those from low-risk group in both the TCGA-KIRC and GSE46602 cohorts. The risk stratification also showed considerable predictive capacities and stabilities, and the OS_score could function as an independent prognostic factor for ccRCC patients. The prognostic value of OS-based risk stratification was steady even when the clinical variables were consideration. And we then identified 17 miRNA-OSRG regulatory pairs, which might elucidate the underlying mechanism of ccRCC progression.

ccRCC is a malignancy with high heterogeneity and mutation susceptibility. We found that patients in the high-risk group exhibited higher level of TMB compared to those in low-risk group, and the most frequent mutation genes were VHL, PBRM1, and TTN in both groups. Loss mutation of VHL is considered as an initial step in the tumorigenesis of ccRCC, resulting in the activation of Akt pathway [39], and NF-κB pathway [40]. The loss of PBRM1 could upregulate the HIF signaling, modulate the replication stress, and impair p53-mediated cell cycle regulation, and mediate genomic instability [41]. TTN mutations could also predict poor prognosis in many cancers, such as gastric cancer [42], thyroid cancer [43], and colon cancer [44]. The present study also found the mutation frequency of VHL, PBRM1, and TTN is the highest in both subgroups, which may provide therapeutical targets for individualized treatment strategy.

Overall, our study had developed a considerable prognostic model based on the expression features of OSRGs, which performed accurately predive powers and robustness in assessing prognosis of patients with ccRCC.

To further validate the robustness and clinical value of our 7-gene OS_score, we compared its predictive performance with several recently published oxidative stress– or ROS-related prognostic signatures in ccRCC. Ma et al., Liu et al., and Wang et al. reported models with AUCs ranging from 0.67 to 0.72 [45–47], whereas our OS_score achieved higher AUCs of 0.794, 0.813, and 0.836 for 1-, 3-, and 5-year overall survival in the TCGA-KIRC cohort. The OS_score also exhibited significant prognostic discrimination (log-rank P = 5.41 × 10⁻¹³), indicating enhanced predictive precision compared with previous oxidative stress–based signatures. Furthermore, OS_score-associated drug sensitivity analysis suggested that high- and low-risk groups might respond differently to specific therapeutic agents, underscoring the model’s potential utility in prognostic evaluation and individualized treatment for ccRCC.

Despite these benefits, limitations still existed in the present study. First, all the genomic data were acquired from the retrospective TCGA and GEO cohorts, a prospective research should be conducted to confirm the prognostic value of OS-based predictive model. Second, the molecular function and specific mechanism of the identified OSRGs were measured by bioinformatic analysis.

Although our study revealed that high OS_score was associated with elevated expression of immune checkpoint molecules (PD-1, CTLA-4, B7-H3, and B7-H4) and potential immunotherapy responsiveness inferred by computational models, these findings should be interpreted with caution. The predictive value of tumor mutation burden (TMB) in ccRCC remains controversial and generally weaker than that observed in melanoma or non–small cell lung cancer. Therefore, our results indicate an association rather than a definitive causal relationship between OS_score and immunotherapy benefit. Furthermore, due to the lack of publicly available ccRCC cohorts treated with immune checkpoint inhibitors, our conclusions rely on indirect in silico estimations such as TIDE and SubMap analyses. Future studies integrating prospective ICI-treated ccRCC cohorts and experimental validation of immune modulation will be essential to confirm the clinical predictive value of the OS_score.

To verify the accuracy of our predictive model, we selected CRABP2 for targeted in vitro validation. siRNA-mediated knockdown of CRABP2 in 786-O cells confirmed its functional role in tumorigenicity and progression. Silencing the CRABP2 gene significantly inhibited the viability, proliferation, migration, and invasion of 786-O cells. As these processes are critical for cancer progression, these results support the reliability of our model for predicting patient prognosis and its potential for suggesting targeted therapeutic strategies.

Although our functional experiments confirmed that CRABP2 promotes proliferation, migration, and invasion in ccRCC cells and and that its silencing led to decreased intracellular ROS levels, we acknowledge that the current validation remains limited. Additional assays exploring apoptosis, ferroptosis, and downstream oxidative stress–related signaling pathways are warranted to provide deeper mechanistic insights. Future studies will therefore focus on systematically elucidating how CRABP2 modulates oxidative stress and contributes to ccRCC progression both in vitro and in vivo.

Conclusion

Collectively, we conducted comprehensively analysis on the OS patterns in ccRCC, and explored the enriched pathways and immune cell populations between two OS patterns. We then identified 7 genes as independent prognostic factors and integrated them into the prognostic model. We then developed a risk scoring system (OS_score) to distinguish patients at high-risk, and explored the influence of OS_score on survival prognosis when the clinical parameters were considered. The miRNA-OSRG regulatory network, tumor mutation features, and drug sensitivity were evaluated upon risk stratification.

Abbreviations

RCC

Renal cell carcinoma

ccRCC

Clear cell renal cell carcinoma

TCGA

The Cancer Genome Atlas

GEO

Gene Expression Omnibus

OSRGs

Oxidative stress related genes

ROS

Reactive oxygen species

ssGSEA

Single-sample gene sets enrichment analysis

ESTIMATE

Estimation of stromal and immune cells in tumor tissues

CIBERSORT

Cell type identification by estimating relative subsets of RNA transcripts

OS

Overall survival

FC

Fold change

LASSO

Least absolute shrinkage and selection operator

ROC

Receiver operating characteristic

GO

Gene ontology

AUC

Area under the receiver operating characteristic curve

TMB

Tumor mutation burden

ICI

Immune checkpoint inhibitor

Author contributions

Y.W. designed the study and performed the data analysis. Z.J. and B.H. performed the data analysis, result visualization and manuscript writing. S.Z., S.F.,Z.D. and C.W. collected the data. J.L. and T.W. performed the manuscript revising. All authors read and approved the final manuscript.

Funding

This work was supported by a grant from the National Natural Science Foundation of China (No. 81874165).

Data availability

The datasets analysed during the current study are available in the TCGA database (https://portal.gdc.cancer.gov/), Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/),GSE53757, GSE46699, GSE66270, GSE66272, GSE36895, GSE73731. The further analytical data is provided within the supplementary information files and the original data and code can be obtained by contacting the corresponding author.

Declarations

Ethics approval and consent to participate

All datasets in the present study were downloaded from public databases, including TCGA and GEO database. These public databases allowed researchers to download and analyze public datasets for scientific purposes and thus ethics approval was not required. Clinical trial number: Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.