Construction and verification of a prognostic model of neutrophil-related genes in clear cell renal cell carcinoma

Abstract

Background

The tumor microenvironment of clear cell renal cell carcinoma (ccRCC) is heterogeneous, leading to diverse prognoses among patients. Neutrophils, as a key component of the tumor microenvironment, have predictive value for the prognosis of ccRCC. However, there are currently no predictive models based on neutrophil-related genes. This study aimed to construct and validate a prognostic model for ccRCC based on neutrophil-related genes to facilitate risk stratification and treatment guidance. This study aimed to construct and validate a prognostic model for ccRCC based on neutrophil-related genes to facilitate risk stratification and treatment guidance.

Methods

We analyzed the RNA sequencing (RNA-seq) data and clinical information of ccRCC, screened out 10 neutrophil-related prognostic genes using R software, and constructed a risk prediction model. Single/multivariate Cox regression and least absolute shrinkage and selection operator (LASSO) regression were used for gene screening.

Results

Model validation showed that the area under the curve (AUC) values of the model for predicting 1-, 2-, and 3-year overall survival (OS) were 0.704, 0.674, and 0.656 in the test set. Its performance in the training set was better, with AUC values of 0.796, 0.784, and 0.793, respectively. The calibration curve confirmed that the model had good consistency. Kaplan-Meier (KM) survival analysis showed that the survival rate of patients in the high-risk group was significantly lower than that in the low-risk group (P<0.05), and the risk score prediction efficiency was better than clinical indicators such as age. In summary, this model demonstrated strong predictive performance on both the training and multiple validation sets, effectively identifying high-risk patients with poor prognosis who required intensive treatment and close follow-up. Further analysis showed that four drugs, such as axitinib_1021, may have anti-tumor potential, and the immune infiltration characteristics showed that the infiltration levels of B cells (naive) and CD4 memory activated T cells in the high-risk group were significantly increased.

Conclusions

The proposed 10 neutrophil-related genes are promising biomarkers to predict survival and therapeutic responses in ccRCC patients.

Highlight box.

Key findings

• This study identified 10 neutrophil-related genes as novel prognostic markers for clear cell renal cell carcinoma (ccRCC) through multi-omics analysis (P<0.001). The area under the curve values of the constructed prognostic model for predicting 1-, 2-, and 3-year overall survival (OS) were 0.704, 0.652, and 0.656 in the test set and 0.796, 0.784, and 0.793 in the training set, respectively. Its predictive performance was significantly superior to other clinical factors (P<0.05). Four potential therapeutic drugs (axitinib_1021, cisplatin_1005, cyclophosphamide_1512, and paclitaxel_1080) were screened based on a drug database, providing new strategies for clinical treatment.

What is known and what is new?

• Neutrophils promote ccRCC progression by mediating an inflammatory environment and inducing immunosuppression, but existing targeted therapies have limited efficacy.

• For the first time, 10 neutrophil-related genes were screened out as novel prognostic markers for ccRCC. A prognostic model was constructed, and the model effectively predicted the survival of ccRCC patients (OS in the high-risk group was significantly lower than that in the low-risk group, P<0.05).

What is the implication, and what should change now?

• The ccRCC prognostic model constructed in this study has clinical predictive value. axitinib_1021, cisplatin_1005, cyclophosphamide_1512, and paclitaxel_1080 may be candidate drugs for the treatment of ccRCC patients.

Introduction

In the global cancer spectrum, renal cell carcinoma (RCC) accounts for some 3% of all malignancies, with morbidity and mortality ranking 16^th (1). Among the various histopathological types of renal carcinoma, clear cell RCC (ccRCC) originates from renal epithelium, prevails, representing 70–80% of renal carcinoma cases (2). Epidemiological data suggest that ccRCC incidence in men is about twice that of women (3). The prognosis of ccRCC is very poor, with high metastasis and recurrence rates. Approximately 45% of patients develop metastases at or after diagnosis, and it recurs in 30% of patients who undergo surgical treatment, including 10–25% of patients initially assessed as low-risk. This grim prognosis is primarily attributed to the high heterogeneity between and within tumors (4). Noteworthily, ccRCC tends to metastasize and leads to the worst prognosis among all RCCs (5). From a pathological perspective, both ccRCC and papillary RCC originate from epithelial cells in the proximal tubule and typically feature cytoplasmic hyalinization (3). The formation and progression of ccRCC are tied to von Hippel-Lindau (VHL) (6-9), phosphatase and tensin homolog (PTEN) (10,11), and other genes. It involves signal paths like phosphatidylinositol 3-kinases/protein kinase B (PI3K/AKT) (12), Hippo (13), and programmed cell death-1/programmed death ligand-1 (PD-1/PD-L1) (14). Immunotherapy has become the key therapy for treatment-naive metastatic ccRCC (15).

Targeted drugs like bevacizumab (16) and nivolumab (17) intervene on key signaling pathways in tumor growth and spread, and exert tumor immunotherapy effects. Nivolumab plus ipilimumab represents the new standard of first-line treatment for ccRCC patients (18). Exploring the biomarkers and molecular mechanisms of ccRCC is vital to improving patient outcomes. However, only a few predictive markers are available currently, and intensive research on the possible mechanisms of its occurrence and development is urgently required.

Neutrophils, a crucial white blood cell (WBC), mainly swallow and digest foreign pathogens, and function critically in the body’s immune system (19). Neutrophils are dominantly involved in carcinogenesis and carcinomatosis by releasing genotoxic substances [e.g., elastase (20), reactive oxygen/nitrogen species (ROS/RNS)], promoting the release of angiogenic factors [e.g., matrix metalloproteinase-9 (MMP-9) (21) and vascular endothelial growth factor (VEGF) (22)], and inducing functional reprogramming of immunosuppressive cells [e.g., myeloid-derived suppressor cells (MDSCs) (23)], thus exacerbating DNA damage, promoting tumor angiogenesis and immune escape. Granulin, encoded by neutrophil-related genes, executes multiple functions, including immune regulation and cancer progression, thereby accelerating cancer growth (24). In colon cancer (CC), adenoma, and adjacent normal tissues, the expression level of neutrophil defensin dielectric elastomer fiber actuators (DEFAs) is significantly higher in tumor tissues vs. normal tissues (25). CD66b, a specific marker for neutrophils, advances the malignant tumor progression, leading to a worse prognosis in colorectal cancer (CRC) (26). Mouse tumor models have proven that neutrophils form a pro-inflammatory microenvironment through activation of neutrophil elastase encoded by the ELANE (ELA2) gene and drive the development of pancreatic tumors (27). Owing to those findings, this paper aims to analyze principal regulatory genes and molecular mechanisms of tumor-related neutrophils in the progression of ccRCC and to build a prognostic model based on neutrophil-related genes.

Although neutrophils play multiple crucial roles in the tumor microenvironment of ccRCC, prognostic models based on neutrophil-related genes remain relatively scarce compared to other immune cells, such as T cells and macrophages. Their functions are highly malleable, and they can both inhibit and promote tumors. This complex function suggests that their gene expression profiles may contain rich prognostic information. Therefore, deeply exploring neutrophil-related genes holds promise for overcoming the limitations of existing prognostic biomarkers and providing a more discriminative risk stratification tool.

The RNA sequencing (RNA-seq) dataset of ccRCC was from The Cancer Genome Atlas (TCGA) program, and combined with the single-cell dataset of Gene Expression Omnibus (GEO) to identify key genes linked to neutrophils in ccRCC. A ten-gene risk prediction model was established, expecting to render fresh references and tactics for diagnosing and treating ccRCC patients. Currently, the clinical management of ccRCC has entered a new stage. According to the 2025 European Association of Urology (EAU) guidelines for RCC, immune checkpoint inhibitor combination therapy has become the first-line standard treatment for metastatic patients (28). Hence, more efficient prognostic assessment tools are needed. The guidelines recommend the use of the International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) risk stratification, but the prognosis of patients is still significantly heterogeneous. Therefore, it is imperative to develop a prognostic risk model to accurately identify patients with highly aggressive biological behaviors that cannot be identified by traditional standards. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-558/rc).

Methods

Data collection

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Figure S1 depicts an overall workflow of data collection. Batch RNA-seq count data, gene probe maps, clinical information, and survival information for ccRCC were downloaded from the University of California Santa Cruz (UCSC) Xena Browser (https://xenabrowser.net/). Samples with incomplete survival and clinical information were excluded. Then, 72 normal tissue samples and 526 tumor samples were identified. Subsequently, among these 526 tumor samples, 4 samples were excluded due to missing data on the expression levels of genes involved in subsequent modeling analyses. Strict false discovery rate (FDR) correction (P<0.05) was applied to mitigate the potential impact of sample imbalance on differential expression analysis. GSE167573 monoplast data file was taken from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) to verify the model’s feasibility. Differentially expressed genes (DEGs) were identified from 72 normal samples and 526 ccRCC samples using the “edgeR” package, and volcano plots were generated. The screening criteria for DEGs were: |log₂fold change (FC)| >1.5 and FDR <0.05.

To construct a neutrophil-related gene set, genes highly associated with neutrophils were retrieved from the GeneCards database (https://www.genecards.org/) as candidate genes. To enhance the reliability of the evidence, the expression differences of the feature genes in ccRCC were further validated using the Human Protein Atlas (HPA) database (https://www.proteinatlas.org/). The cellular origin and distribution of feature genes in the tumor microenvironment were analyzed using the Tumor Immune Single-Cell Hub (TISCH) database (http://tisch.comp-genomics.org/home/). Based on the comprehensive multidimensional evidence, the genes ultimately used to construct the prognostic model all showed a clear association with neutrophils or the tumor immune microenvironment.

Construction of risk models or prognosis models through univariate or multivariate Cox regressions

A dataset of 522 ccRCC samples with complete survival information was randomly, equally grouped into a training set (n=261, with 81 death events) and a testing set (n=261, with 90 death events). Model construction and feature selection were performed based on the training set to maintain the independence of the test set. Univariate Cox regression analysis was performed on the training set to screen for prognosis-related feature genes. Feature genes with P<0.05 were selected. Feature genes with non-zero selection coefficients were included in the least absolute shrinkage and selection operator (LASSO) regression analysis. Ten-fold cross-validation was used to determine the optimal penalty parameter (λ) in the LASSO regression to optimize the model and prevent overfitting. The model that produced the minimum mean error (λ.min) in the cross-validation was then selected to obtain candidate prognostic genes. Finally, the candidate genes were included in a multivariate Cox regression analysis to construct a risk prognostic model, and the risk coefficient for each feature gene was calculated.

The risk score of patients was calculated using the following formula: risk score = CTSE × 0.9220 + SUSD4 × 1.2223 + SPC25 × 1.5711 + CXCL5 × 1.0689 + ZNF395 × 0.7576 + MUC5B × 1.1931 + KITLG × 0.6629 + TUBA3C × 1.1290 + FLT1 × 1.2300 + HUNK × 0.8399.

Patients were split into two groups (high- vs. low-risk) by median risk scores. Survival analysis and curve plotting were executed with the survival and survminer packages, respectively. Receiver operating characteristic (ROC) curves were made utilizing the timeROC package. The model’s sensitivity and accuracy in forecasting the 1-, 2-, and 3-year overall survival (OS) among ccRCC individuals were appraised through the area under the curve (AUC). The 95% confidence interval (CI) for the AUC values was calculated using the built-in asymptotic method provided by the timeROC package. Risk factors and survival intervals were visualized via the ggplot package. Gene heatmaps for both groups were generated through the ComplexHeatmap package.

External validation

The GEO query package was employed to download patient data with 55 survival cases from the GEO database (GSE167573). We computed patients’ risk scores in the dataset and placed patients into two groups (high- vs. low-risk) by median risk scores. Survival curves were finished via the survminer package, and ROC curves via the time ROC package. The AUC was computed. Moreover, the risk factors and survival intervals for both groups in the dataset were analyzed. The ComplexHeatmap package was utilized to draw gene heatmaps of both groups.

Nomogram

One- to 5-year survival rates were forecasted with the risk score as a clinical feature, combined with other clinical features, like gender, age, and tumor (T)/metastasis (M)/node (N) stage. A nomogram with different clinical features was plotted through the rms package. Simultaneously, univariate Cox regression of clinical features was analyzed, survival curves were drawn, and points of different clinical features were compared. Finally, the prediction model’s validity was verified utilizing the calibration curve.

Enrichment analysis

DEGs with P<0.05 were chosen by the clusterProfiler package in R software. Their functions and pathways were analyzed via enrichment analyses of Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG). GO covered the biological process (BP), cellular component (CC), and molecular function (MF).

Besides, by referring to the Reactome gene sets in the gene set enrichment analysis (GSEA) website (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp), the gene set variation analysis (GSVA) package was adopted for GSEA.

Analysis of immune infiltration and assessments of immune cell infiltration (ICI) and immune microenvironment

Among 22 WBC subtypes (LM22) of patients in both groups, differences in ICI were appraised by the CIBERSORT algorithm. LM22 was from a known reference set on the CIBERSORTx website (https://cibersortx.stanford.edu/). The potential of tumor immune evasion was appraised through Tumor Immune Dysfunction and Exclusion (TIDE) scores. Scores for relevant indicators were gained through the TIDE website (http://tide.dfci.harvard.edu). Better TIDE scores implied stronger potentials for tumor immune escape and worse immunotherapy benefits in patients. Meanwhile, intergroup differences in immune infiltration were appraised through the ESTIMATE algorithm of the Immuno-Oncology Biological Research (IOBR)-master package. Stromal, immune, and ESTIMATE scores were compared.

By adopting the single-sample GSEA (ssGSEA) algorithm in the GSVA package, associations of risk scores with 28 types of infiltrated immunocytes were detected via Pearson’s correlation coefficient, and immune infiltrations in both groups were reevaluated.

Drug sensitivity screening

Links between drugs and prognostic genes were analyzed utilizing the Psych package. Sensitivity data on 198 drugs were extracted from the Genomics of Drug Sensitivity in Cancer (GDSC). Links of conventional targeted drugs, chemotherapy drugs, with the half-maximal inhibitory concentration (IC₅₀) in both groups were tested using the OncoPredict package.

Validation of protein expression and single-cell expression map of feature genes

The HPA database (https://www.proteinatlas.org/) was utilized to further verify protein expressions of prognostic genes among the model genes in cancerous tissue and adjacent non-cancerous tissue. Expressions of prognostic genes in various subcellular components were appraised utilizing the TISCH database (http://tisch.comp-genomics.org/home/).

Statistical analysis

Data processing and analysis were executed via R 4.4.1. Links between prognostic genes and drugs were appraised through Pearson’s correlation analysis. Survival curves were yielded by Kaplan-Meier (KM) analysis and contrasted via the log-rank test. The Wilcoxon test was employed for quantitative data comparison between two groups, and one-way analysis of variance (ANOVA) for three or more groups. P<0.05 denoted statistical significance.

Results

Screening of differential hotspot genes

To explore the potential role of neutrophils in ccRCC, this study systematically screened neutrophil-related genes associated with the prognosis of ccRCC. The overall study protocol is shown in Figure S1. First, raw RNA-seq count data, gene annotations, clinical survival information, and phenotypic data of 607 ccRCC samples were obtained from the UCSC Xena database (https://xenabrowser.net/datapages/). After quality control and deduplication, 598 samples with complete clinical data (72 normal tissues and 526 tumor tissues) were selected for further analysis. By comparing tumor and normal tissues, a total of 5,839 DEGs were identified (Figure 1A), of which 4,145 genes were upregulated in tumors and 1,362 genes were downregulated (|log₂FC| >1.5, FDR <0.05). To further explore neutrophil-related mechanisms, 2,309 neutrophil-related genes were obtained from the Genecards database (https://www.genecards.org/). After intersecting these genes with DEGs, 525 candidate neutrophil-related genes associated with ccRCC were successfully screened (Figure 1B). These genes may represent core molecules of neutrophils in the tumor microenvironment of ccRCC.

Figure 1.

Differential expression of neutrophil-related genes in ccRCC, and the construction and validation of the prognostic model. (A) The Venn diagram of differentially coexpressed genes. (B) Volcano map of 525 key differentially expressed hotspot genes. (C) Univariate Cox regression analysis. Forest plot displaying significant neutrophil-related genes associated with OS (P<0.05). (D) LASSO regression with 10-fold cross-validation. (E) LASSO coefficient distribution of neutrophil-related genes. (F) The results of 10 genes in multivariate Cox regression model. (G) KM survival curves for OS in high- and low-risk groups in training set. (H) Time-dependent ROC curves (training set, TCGA). The model demonstrated stable predictive accuracy for 1-, 2-, and 3-year OS, with AUCs (95% CI) of 0.796 (0.693–0.899), 0.784 (0.706–0.861), and 0.793 (0.721–0.865). (I) Risk factor distribution in TCGA training set. (J) Survival status and risk score distribution (TCGA training set). (K) Expression levels of 10 neutrophil-related genes (high- vs. low-risk groups). The P value was calculated using the Wilcoxon test. The characteristic gene set was identified using LASSO penalized Cox regression analysis. (L) Heatmap of neutrophil-related gene expression and clinical features. *, P<0.05; **, P<0.01; ***, P<0.001. AIC, Akaike’s information criterion; AUC, area under the curve; ccRCC, clear cell renal cell carcinoma; CI, confidence interval; DEG, differentially expressed gene; FC, fold change; FDR, false discovery rate; KM, Kaplan-Meier; LASSO, least absolute shrinkage and selection operator; OS, overall survival; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Prognostic model construction

To develop molecular tools for clinical prognostic assessment, a prognostic model was constructed based on the aforementioned 525 candidate genes. The 522 ccRCC samples were randomly divided into a training set (n=261) and a test set (n=261). Model construction was performed on the training set.

Univariate Cox regression analysis initially identified 161 genes significantly associated with survival (Figure 1C). Further LASSO regression and 10-fold cross-validation identified 20 genes with the highest predictive value (Figures 1D,1E). Final multivariate Cox regression analysis identified 10 key genes (CTSE, SUSD4, SPC25, CXCL5, ZNF395, MUC5B, KITLG, TUBA3C, FLT1, and HUNK), and a prognostic model was constructed based on these genes (Figure 1F).

The prognostic model demonstrated favorable performance in risk stratification in the training set. The samples in the training set were divided into high- and low-risk groups based on the median risk score. The results showed that OS of patients in the high-risk group was significantly shorter (P<0.001, Figure 1G), providing a reliable basis for clinical identification of patients with poor prognosis. The AUC values of the model for predicting 1-, 2-, and 3-year OS were 0.796 (95% CI: 0.693–0.899), 0.784 (95% CI: 0.706–0.861), and 0.793 (95% CI: 0.721–0.8649), respectively, suggesting that the model had predictive value (Figure 1H). The plots for the distribution of risk factors and survival status further confirmed that the mortality rate significantly increased with increasing risk score (Figure 1I,1J), supporting that this scoring system may be used in risk assessment.

Gene expression analysis revealed the biological basis behind the model. SUSD4, SPC25, CXCL5, MUC5B, and TUBA3C were highly expressed in the high-risk group, while CTSE, ZNF395, KITLG, FLT1, and HUNK were highly expressed in the low-risk group (Figure 1K,1L). This expression pattern was consistent with the risk stratification of the model, suggesting that these genes may play a role in the prognosis of ccRCC.

To confirm the stability of the model, its predictive performance was evaluated on the test set (Figure S2). The results showed that the 10 genes effectively distinguished patients at different risks in the test set. Patients in the high-risk group had poorer survival outcomes, indicating that the model has a certain degree of general applicability. Finally, we identified 10 neutrophil-related genes and constructed a risk scoring model based on their expression levels and LASSO regression coefficients. The complete calculation formula for this model and detailed statistical information for each gene are provided in Table S1.

To verify the robustness of the model, an ablation study was conducted on the TCGA test set. The results (Figure S2G) showed that the AUC values of the model for 1-, 2-, and 3-year OS on the test set were 0.652, 0.631, and 0.622, respectively. After excluding the SPC25 gene, the corresponding AUC values were 0.653, 0.604, and 0.639. After excluding both SPC25 and KITLG, the AUC values were 0.644, 0.583, and 0.582, respectively. Performance variation analysis showed that after excluding the two most important genes, the model achieved a performance retention rate of 98.9%, 92.3%, and 93.5% for predicting 1-, 2-, and 3-year OS, respectively, demonstrating the good robustness of this polygenic feature.

Nomogram

To assess the clinical relevance of the risk score, this study first analyzed its relationship with clinicopathological parameters. The results showed no significant differences in risk score between different sexes and ages (Figure S3A,S3B), but it was significantly positively correlated with tumor progression. Specifically, the risk score was significantly elevated in patients with advanced disease (T3–4), distant metastasis (M1), and lymph node metastasis (N1) (Figure S3C-S3E). The overall differences among different T, N, and M stages were statistically significant (P<0.05), suggesting that the risk score may reflect the aggressive biological behavior of the tumor.

Its prognostic value was systematically evaluated using survival analysis and a Cox regression model. Survival analysis confirmed that the OS of patients in the low-risk group was significantly higher than that in the high-risk group (Figure 2A). Univariate Cox analysis showed that the risk score and T, M, and N stages were all significantly associated with patient prognosis (Figure 2B). More importantly, after multivariate adjustment, the risk score remained an independent predictor of OS (Figure 2C,2D), which preliminarily confirmed its prognostic value independent of TNM stages.

Figure 2.

The clinical predictive values of the model. (A) KM survival curves for nomogram model for OS. (B) Forest plots of univariate Cox analysis of risk scores and clinical characteristics. (C) Forest plots of multivariate Cox analysis of risk scores and clinical characteristics. (D) The histogram depicting the significant difference in the risk scores in ccRCC patients stratified by survival. (E) The nomogram model constructed based on risk score combined with other clinical features. (F) The calibration curve for evaluating the accuracy of the nomogram model. Gender: 0, male; 1, female. ***, P<0.001. ccRCC, clear cell renal cell carcinoma; CI, confidence interval; KM, Kaplan-Meier; M, metastasis; N, node; OS, overall survival; T, tumor.

Then, we combined the risk score with key clinical characteristics (gender, age, TNM stage) to construct a nomogram (Figure 2E). Model analysis showed that the risk score was an important contributing factor in this predictive model. The calibration curve further indicated that the predicted results of the nomogram were highly consistent with the actual observed values (Figure 2F).

Finally, to quantitatively compare the predictive performance of this model and traditional staging systems, time-dependent ROC curves were plotted (Figure S3F). The results show that while TNM staging performed best in predicting short-term prognosis (1 year: AUC =0.810), its predictive accuracy significantly decreased over time (5 years: AUC =0.691). In contrast, the risk signature based on 10 genes exhibited consistent and stable predictive performance. Its predictive accuracy for long-term prognosis (5 years: AUC =0.797) was significantly superior to TNM staging. Multivariate Cox regression confirmed that even after adjusting for T, N, and M stages, the risk score remained an independent prognostic factor.

External validation

To assess the generalizability of the prognostic model, preliminary validation was performed using an external independent cohort. Data on ccRCC patients (including gender, age, height, weight, tissue differentiation, and TNM stage) were downloaded from the GEO public database (GSE167573). After screening, 55 samples with complete survival information were obtained for external validation. Patients in this cohort were divided into high-risk and low-risk groups using the median risk score, and the survival difference between the two groups was assessed using KM analysis. The results showed that in this dataset, the OS was shorter in the high-risk group than the low-risk group, and the difference in OS was statistically significant (P<0.001) (Figure 3A). However, the discriminative power of the model in this small validation set was limited, with AUC values of 0.853, 0.803, and 0.762 at 1, 2, and 3 years, respectively (Figure 3B). The risk factor and survival status distribution plots showed that the mortality rate increased with increasing risk score (Figure 3C,3D), indicating that a high-risk score was associated with a worse prognosis. Gene expression analysis indicated that the expression patterns of some genes (such as SPC25) in the high- and low-risk groups were consistent with the predicted direction of the model (Figure 3E), but the expression differences of most genes did not reach statistical significance. The gene expression heatmap visually illustrates the overall expression patterns of all feature genes in the high- and low-risk groups (Figure 3F).

Figure 3.

Clinical predictive value of the model in external validation cohort (GSE167573). (A) KM survival curves for OS in high and low risk groups. (B) Time dependent ROC curves for predicting 1-, 2-, and 3-year survival rates. (C) The risk factor of GSE167573. (D) The survival interval of GSE167573. (E) Box plot visualizing the expression levels of 10 neutrophil-related genes between high and low-risk groups based on log₂(normalized count + 1) values. (F) Heatmap of differential expression of 10 neutrophil-related genes. The expression values were derived from normalized count data after log₂(x + 1) transformation. ***, P<0.001; ns, not significant (P≥0.05). AUC, area under the curve; KM, Kaplan-Meier; OS, overall survival; ROC, receiver operating characteristic.

Enrichment analysis

To clarify the differences in functional and signaling pathways among the 10 neutrophil-related genes, GO enrichment analysis and KEGG pathway analysis were performed. The results (Figure 4A) showed that in ccRCC, the 10 genes were mainly enriched in the regulation of cell-cell adhesion, chemotaxis, and taxis processes in the BP. In the CC, they were mainly located in the external side of the plasma membrane and the collagen-containing extracellular matrix. In the MF, they were mainly associated with cytokine activity, glycosaminoglycan binding, and cytokine receptor binding. KEGG analysis further showed (Figure 4B) that these genes were significantly enriched in pathways such as cytokine-cytokine receptor interaction, the PI3K-AKT signaling pathway, and viral protein interaction with cytokine and cytokine receptor. These results suggested that these genes may participate in the regulation of the tumor immune microenvironment.

Figure 4.

Functional enrichment analysis of neutrophil-related genes in ccRCC. (A) GO enrichment analysis. (B) KEGG pathway enrichment analysis. (C) GSVA-based pathway activity analysis. BP, biological process; CC, cellular component; ccRCC, clear cell renal cell carcinoma; GO, Gene Ontology; GSVA, gene set variation analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function.

Further GSVA analysis identified 19 significantly enriched signaling pathways across different risk groups (P<0.05, FDR <0.05). Among these, there were three immune-related pathways, six cell cycle-related pathways, three inflammation-related pathways, and seven signaling pathways closely related to cancer development (Figure 4C). These findings suggested that these genes may participate in the progression of ccRCC by regulating cell proliferation and immune responses, providing potential targets for subsequent clinical interventions.

Immune infiltration analysis, and the assessment of immune cell infiltration and immune microenvironment

To explore the potential association between the 10 genes and the immune microenvironment of ccRCC, we compared immune characteristics between high- and low-risk groups. Immune cell infiltration analysis showed an increased proportion of various immune cells, including naive B cells, CD4 memory activated T cells, and macrophages M0, in the high-risk group (Figure S4A). Further functional enrichment analysis revealed differences in the activation status of various immune cell subsets between the two groups (Figure 5A), suggesting a possible tendency towards a pro-tumor immune microenvironment.

Figure 5.

Immune infiltration analysis and tumor microenvironment characterization between high- and low-risk groups. (A) Bar plot for proportional distribution of immune cell types. (B) Comparison of TIDE scores, where higher scores indicate a greater likelihood of immune therapy resistance. (C) Infiltration levels of MDSCs, associated with immunosuppressive tumor microenvironment. (D) TIDE sub-scores, reflecting potential immune evasion mechanisms. (E) Differential expression of nine immune checkpoint molecules between groups, shown as log₂(normalized count + 1) values. Statistical significance is denoted as ns, not significant (P≥0.05); *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. Gene expression values were obtained from normalized count data after log₂(x + 1) transformation. MDSC, myeloid-derived suppressor cell; NK, natural killer; NKT, natural killer T; TFH, T follicular helper; TGD, T gamma delta; Th, T helper; TIDE, Tumor Immune Dysfunction and Exclusion; Treg, regulatory T.

Regarding immunotherapy response, there was no significant difference in the total TIDE score between the high- and low-risk groups (Figure 5B), and no significant changes in Dysfunction score (Figure S4B) or CAF, CD8, and CD274 indicators (Figure S4C-S4E), the high-risk group showed an upregulated expression of MDSC-related genes (Figure 5C) and exhibited a higher Exclusion score (Figure 5D). These findings may suggest that high-risk patients may be more susceptible to immune escape.

Furthermore, the expression levels of several immune checkpoint molecules (such as CD274, CD152, and LAG3) were higher in the high-risk group (Figure 5E), which may indicate that these patients may have different response characteristics to corresponding immune checkpoint inhibitor treatments.

To further assess the overall characteristics of the tumor microenvironment, this study further analyzed ESTIMATE score, immune score, stromal score, and tumor purity. The results showed no significant differences in these scores between the two groups (Figure S4F-S4I). This result was in contrast to the aforementioned differences in immune cell subsets and functional states, suggesting that the high-risk group may drive immunosuppression through the activation of specific immune cell subsets. These results suggested that the risk model based on 10 neutrophil-related genes may provide some references for differentiating the characteristics of the immune microenvironment in ccRCC patients, but its performance in predicting immunotherapy responses still needs to be further validated.

Drug sensitivity and feature gene correlation

To preliminarily explore potential candidate drugs for high-risk patients, the IC₅₀ levels of 198 chemotherapy drugs or inhibitors in high- and low-risk groups were analyzed using data from the Genomics of Drug Sensibility in Cancer (GDSC) database (GDSC2_Res and GDSC2_Expr files). Pearson correlation analysis initially identified four drugs that were positively correlated with the risk score, and their IC₅₀ showed significant differences between the high- and low-risk groups (Figure 6A-6D). This preliminary result suggested that high-risk patients may be more sensitive to these drugs. To further explore the underlying mechanisms, the expression correlation between characteristic genes and sensitivity to the most relevant drug (axitinib) was analyzed. The results showed that the expression of TUBA3C, CTSE, MUC5B, and SPC25 was positively correlated with the sensitivity to axitinib (Figure 6E-6H), while the expression of SUSD4, CXCL5, HUNK, and FLT1 was negatively correlated with the sensitivity to axitinib (Figure 6I-6L). These results suggested that these genes may be related to the response pathway of axitinib, providing preliminary insights and hypotheses for further exploration of the possible mechanisms of drug response.

Figure 6.

Correlation between drug sensitivity and signature genes in ccRCC. (A-D) Drug sensitivity analysis of high- and low-risk groups for (A) axitinib (GDSC ID: 1021), (B) cisplatin (ID: 1005), (C) cyclophosphamide (ID: 1512), and (D) paclitaxel (ID: 1080), with lower IC₅₀ values indicating higher drug sensitivity. (E-L) Scatter plot for the correlation of axitinib sensitivity (axitinib_1021) with (E) TUBA3C, (F) CTSE, (G) MUC5B, (H) SPC25, (I) SUSD4, (J) CXCL5, (K) HUNK, and (L) FLT1. Statistical significance is denoted as ***, P<0.001; ****, P<0.0001. ccRCC, clear cell renal cell carcinoma; IC₅₀, half-maximal inhibitory concentration.

Validation of protein expression and single-cell expression map of feature genes

Immunohistochemical tests were performed utilizing the HPA database to further verify protein expressions of up-regulated genes in model genes in cancer and para-cancer tissues. Results uncovered that expressions of CXCL5, SUSD4, and TUBA3C were drastically up-regulated (Figure 7A-7C). The protein expression of ZNF395, FLT1, and HUNK was markedly down-regulated (Figure 7D-7F) in tumor vs. normal tissues. This result was consistent with the expression level trend of RNA-seq in the TCGA dataset, confirming the potentially important role of these genes in the occurrence and development of ccRCC.

Figure 7.

Representative immunohistochemistry images of (A) CXCL5, (B) SUSD4, (C) TUBA3C, (D) ZNF395, (E) FLT1, and (F) HUNK in renal carcinoma tissues and adjacent normal tissues.Images are sourced from the HPA database. Image credit belongs to the HPA. Links to the individual normal and tumor tissues for each protein are provided for CXCL5 (https://www.proteinatlas.org/ENSG00000163735-CXCL5/cancer/renal+cancer#img; https://www.proteinatlas.org/ENSG00000163735-CXCL5/tissue/kidney#img), SUSD4 (https://www.proteinatlas.org/ENSG00000143502-SUSD4/cancer/renal+cancer#img; https://www.proteinatlas.org/ENSG00000143502-SUSD4/tissue/kidney#img), TUBA3C (https://www.proteinatlas.org/ENSG00000198033-TUBA3C/cancer/renal+cancer#img; https://www.proteinatlas.org/ENSG00000198033-TUBA3C/tissue/kidney#img), ZNF395 (https://www.proteinatlas.org/ENSG00000186918-ZNF395/cancer/renal+cancer#img; https://www.proteinatlas.org/ENSG00000186918-ZNF395/tissue/kidney#img), FLT1 (https://www.proteinatlas.org/ENSG00000102755-FLT1/cancer/renal+cancer#img; https://www.proteinatlas.org/ENSG00000102755-FLT1/tissue/kidney#img), and HUNK (https://www.proteinatlas.org/ENSG00000142149-HUNK/cancer/renal+cancer#img), respectively. Scale bar, 200 µm. HPA, Human Protein Atlas.

To directly validate the cellular origin of feature genes in the tumor microenvironment, their expression in various subcells was analyzed using the TISCH single-cell database, revealing the specificity and heterogeneity of the expression of these genes (Figure 8). The results showed that SUSD4 and SPC25 were expressed in erythroblasts (Figure 8A-8D), while CTSE and KITLG were mildly expressed in erythroblasts and pericytes, respectively (Figure 8E-8H), suggesting that this prognostic model may have detected key immune or stromal cell infiltration signals in the tumor microenvironment. ZNF395 was mainly highly expressed in epithelial-derived malignant cells (Figure 8I-8J), suggesting that it may be directly produced by tumor cells and promote the development of tumors. FLT1 was highly expressed in endothelial cells (Figure 8K-8L), which was consistent with its role as a VEGF receptor 1 (VEGFR1). This result suggested that it may affect patient prognosis by regulating tumor angiogenesis. Furthermore, the expression of ZNF395 and FLT1 in batch RNA-seq data was associated with a favorable prognosis. This result may be attributed to the high infiltration of immune and stromal cells in the high-risk group, which dilutes the signal of malignant cells, thus reducing the expression of ZNF395. Furthermore, the disordered and dysregulated vascular structure in the high-risk group cannot maintain the high expression of FLT1. In addition, ZNF395 may play a dual role in ccRCC. Hence, the complex tumor biology needs to be further investigated. Additionally, the typical neutrophil chemokine CXCL5 and the tumor progression-associated kinase HUNK were not significantly expressed in specific cell clusters in this dataset (Figure 8M-8P). This may be due to technical differences or cell capture biases in different single-cell datasets. Nonetheless, their high protein expression was confirmed in HPA, and their prognostic value was also demonstrated in batch RNA-seq data.

Figure 8.

Subcellular expression patterns and distribution analysis of signature genes in ccRCC. Left panels (A,C,E,G,I,K,M,O): subcellular localization patterns demonstrated by immunofluorescence staining. Right panels (B,D,F,H,J,L,N,P): corresponding violin plots quantifying expression levels across cellular compartments. (A,B) SUSD4; (C,D) SPC25; (E,F) CTSE; (G,H) KITLG; (I,J) ZNF395; (K,L) FLT1; (M,N) CXCL5; and (O,P) HUNK. ccRCC, clear cell renal cell carcinoma.

Discussion

This study analyzed RNA-seq data and selected 10 neutrophil-related genes (including CTSE, SUSD4, SPC25, CXCL5, ZNF395, MUC5B, KITLG, TUBA3C, FLT1, and HUNK). These genes were used to build a prognostic model for ccRCC. The risk score, as a core prognostic indicator in this model, demonstrates favorable discriminative performance and may be used in clinical practice. Multivariate Cox regression analysis confirmed that it was superior to TNM-based American Joint Committee on Cancer (AJCC) clinical staging. In recent years, various prognostic models based on different molecular mechanisms have been developed for risk prediction in ccRCC. For example, the study by Tian et al. on specific immune checkpoint molecules reveals that CD72 may serve as a biomarker for disease diagnosis and prognosis (29). Xu et al. focus on emerging cell death mechanisms (cuproptosis) and identified the key role of FDX1 in ccRCC through systematic pan-cancer analysis (30). Saout et al. use single-cell sequencing technique to deeply analyze heterogeneity within tumors and identify specific malignant cell populations associated with poor prognosis in low-risk patients (4). These models have confirmed the accuracy of molecular subtyping in the prognosis of ccRCC from different dimensions. In our study, a prognostic model was constructed based on 10 neutrophil-related genes. This is the first time that neutrophil-related genes have been used to construct a prognostic model for ccRCC. Methodologically, it integrates the advantages of batch and single-cell data, and the selected feature genes can comprehensively reflect the overall state of the tumor microenvironment. The models demonstrated favorable predictive performance. Furthermore, the drug sensitivity analysis showed that the high-risk group was potentially sensitive to axitinib, a standard VEGFR inhibitor, highlighting its potential link with clinical targeted therapy.

SUSD4 is a cell surface protein (31). Given the poor outcomes of CRC individuals with highly expressed SUSD4, subcutaneous tumor implantation experiments in nude mice showed that subcutaneous tumors in mice after SUSD4 knockdown were smaller and lighter than those in the control group (32). These results suggest that SUSD4 may play a cancer-promoting role in a variety of solid tumors, consistent with its high expression in ccRCC. Protein SPC25 constitutes the 1/4 NDC80 complex. Highly expressed SPC25 in hepatocellular carcinoma (HCC) promotes HCC progression, while silenced SPC25 greatly inhibits the invasion and migration capabilities of HCC cells (33). SPC25, a key factor in cell division, plays a crucial role in promoting various cancers, suggesting that it may play a similar role in the progression of ccRCC. Chemokines consist of a diverse array of small proteins with multifunctional properties, which bind G protein-coupled receptors on the target cell surface. Cellularly, overexpression of CXCL5 significantly drives CRC cells to migrate and invade. Its increased expression in a nude mouse intrasplenic injection model further verified that CXCL5 can greatly promote CRC cells to metastasize to the liver (34). As a representative of chemokines, CXCL5 is able to promote metastasis and may also be involved in the malignant progression of ccRCC. MUC5B is a secreted mucin (MUC). Highly expressed MUC5B is closely linked with an unfavorable prognosis among individuals suffering from lung adenocarcinoma (LUAD) (35). MUC5B mutations are linked with bladder cancer susceptibility (36) and are upregulated in multiple cancers like CRC (37), gastric cancer (38), and breast cancer (39). MUC5B is up-regulated in multiple cancer types, indicating that it is a potential pan-cancer oncogenic molecule and is important in the prognostic model for ccRCC.

Notably, this model also included several genes with complex functional backgrounds or dual roles in cancer. CTSE-encoded cathepsin E has aspartic acid endopeptidase activity and belongs to the lysosomal peptidases family. CTSE is involved in processes of protein degradation, antigen presentation, and apoptosis, and is vital to maintain cell homeostasis and initiate and enhance T cell-mediated immune responses (40). Mendelian randomization (MR) analysis proved that elevated cathepsin E levels were linked with higher risks of malignant breast neoplasm (41). This suggests that the immune-related functions of CTSE may be regulated by specific tumor microenvironments, thereby promoting the development of cancers. As a C2H2 ZNF, ZNF395 initiates numerous cancers and expedites cancer progression by promoting cell proliferation (42). Overexpression of ZNF395 promotes the proliferation of human glioma cell lines SHG-44 and U251 (43). Besides, highly expressed ZNF395 can drastically prevent hepatoma cells from migrating and invading, while down-regulated ZNF395 can reverse the effect (44). The function of ZNF395 is dependent on cell types, highlighting that its role is complex in cancer. KITLG, or stem cell factor (SCF), is a ligand for the c-KIT receptor. KITLG overexpression is a novel feature of World Health Organization (WHO) type A and AB thymomas and may play a key role by activating the MAPK pathway (45). Increased KITLG levels in desmocyte and endotheliocyte within CRC samples possibly activate MCs through the KITLG-KIT interaction, possibly suppressing tumor growth (46). KITLG may inhibit or promote tumors, based on specific cell types, tumor microenvironment, and interactions with other signaling pathways. Under different conditions, KITLG may exhibit dual functions of suppressing or promoting cancer, which may be linked to specific cell types, tumor microenvironment, and interactions with other signaling pathways. FLT1, or VEGFR1, is one of the major receptors of the VEGF family and usually participates in angiogenesis and cell survival signaling pathways. FLT1 promotes the growth, attack, and spread of LUAD cells, and is inversely related to individual survival time and recurrence-free survival. The variant C-allele of the FLT1 single-nucleotide polymorphism rs9582036 decreases FLT1 expression and may accelerate the relapse of non-small cell lung cancer via angiogenesis (47). FLT1 promotes poly (ADP-ribose) polymerase (PARP) inhibitors (PARPis) resistance in BRCA1/2-mutant breast cancer by activating the AKT pro-survival pathway and inhibiting CD8⁺ T cell infiltration. Its activation caused a decline in progression-free survival in patients. Blocking FLT1 can reverse drug resistance and restore PARPis sensitivity (48). FLT1 plays a central role in angiogenesis and immune regulation, indicating that it is a key molecule influencing the prognosis of ccRCC. HUNK, a mitogen-activated protein kinase, participates in multiple cellular processes and regulates the lifecycle, multiplication, and spread of tumor cells (49). HUNK activates the RhoA signaling pathway by phosphorylating GEF-H1 and inhibits epithelial-mesenchymal transformation (EMT) and metastasis of CRC. Its low expression is linked with CRC metastasis, proving that HUNK has cancer suppression and protective functions in CRC (50). Given the role of HUNK in inhibiting tumors, its low expression in this prognostic model may indicate a poor outcome, reflecting the inactivation of certain protective pathways during the progression of ccRCC. Although the role of TUBA3C in tumors is poorly studied, it may be a potential oncogene. Its inclusion in the model suggests that cytoskeleton remodeling may play an under-recognized role in ccRCC. In summary, the above studies confirm that these genes are involved in regulating key BPs such as cell proliferation, invasion, metastasis, immune microenvironment remodeling, and cell metabolism in various cancers. Although their roles in some cancer types are controversial or context-dependent, they all show a potential association with ccRCC. This suggests a potential link between this prognostic model and key tumor processes at the biological mechanism level. To demonstrate that the mode was neutrophil-related, functional enrichment analysis revealed that the genes used in the model were significantly enriched overall in core neutrophil BPs such as chemotaxis and cytokine signaling pathways. Crucially, the model included several hub genes with direct experimentally validated associations with neutrophils. For instance, CXCL5 (51) participates in immunosuppression by mediating neutrophil chemotaxis. FLT1 (52) can target neutrophils for in vivo imaging. CTSE (53) is expressed in neutrophils and participates in pain generation. MUC5B can induce the formation of neutrophil extracellular traps (NETs) (54). This evidence collectively suggests that the signature constitutes a synergistic program to promote the recruitment of neutrophils and activation of functional pathways.

In GSVA analysis, many immune-related pathways were enriched. PI3K/AKT signaling pathway was highly enriched in papillary thyroid carcinoma (PTC) samples, and cytokine-cytokine receptor interaction constituted the critical enrichment pathway (55). As a PI3K/AKT-negative regulator, PTEN inhibits PI3K/AKT activation and induces autophagosome formation, thereby inhibiting the growth of lung tumors (56). In the immunosuppressive tumor microenvironment, PD-L1⁺ naive B cells facilitate malignant proliferation by inhibiting effector T cells, while RCC cells recruit CD4⁺ T cells more efficiently than normal renal epithelial cells (57). A massive ICI due to activated CD4 memory T cells and M0 macrophages is highly linked with poor prognosis in CC patients (58). In RCC, resting mast cells secrete pro-angiogenic cytokines through c-KIT receptor activation, and then interact with cancerous and endothelial cells to promote tumor formation and progression (59). Naive B cells, activated CD4 memory T cells, T follicular helper cells, TGD, M0 macrophages, activated MCs, and neutrophils were highly infiltrated in the high-risk group in this paper. This is often closely linked to the growth, invasiveness, and unfavorable outcomes of tumors. ESTIMATE results discovered no obvious difference between the groups, but a rising tendency in the high-risk group. This suggests a higher chance of immune escape among patients in the high-risk group, thus resulting in less effective immunotherapy. Research by Tang et al. also showed that TIDE scores were apparently higher in RCC patients who passed away (60). Many immune checkpoints (e.g., CD274 and CTLA-4) had higher expression levels within the high-risk group vs. the low-risk group. This demonstrates that multiple immune checkpoint blockages benefit these patients (61), and the prognosis of patients is improved due to their enhanced immune response.

Based on the developed prognostic model, potential drug targets for patients with ccRCC were identified from the GDSC database. It was found that axitinib_1021, cisplatin_1005, cyclophosphamide_1512, and paclitaxel_1080 were possibly drug candidates for high-risk patients. It is important to emphasize that the above findings are entirely based on computational predictions and have not been experimentally verified. Hence, the results should be interpreted with caution. The prediction result for axitinib was the most clinically relevant. Axitinib, as a standard VEGFR tyrosine kinase inhibitor (TKI) (62,63), is widely used in the treatment of advanced RCC. This model predicted that the high-risk group would be more sensitive to axitinib. This result is not only highly consistent with clinical guidelines but also suggests that the high-risk group may be more sensitive to anti-angiogenic therapies. Furthermore, this model also screened out cisplatin_1005, cyclophosphamide_1512, and paclitaxel_1080 as non-first-line drugs. Although they are not currently the first-line treatment for RCC, they provide potential alternative treatment measures for patients with limited treatment options or who are insensitive to standard therapies. Among chemotherapy drugs for CC, cisplatin and axitinib also had low IC₅₀ values (64) and may be candidates for high-risk patients. Cyclophosphamide is a widely used anti-cancer drug in cancer treatment. It is efficient alone but usually combined with other anti-tumor drugs with significant efficacy against a variety of malignancies (65). Paclitaxel plus sorafenib with radiotherapy collaboratively reduces the survival rate of RCC cells and greatly expedites their apoptosis (66).

These studies provide a preliminary theoretical basis for the potential application of related non-first-line drugs. In summary, the drug sensitivity analysis in this study is essentially a hypothetical inference. The predictive results for axitinib are based on a relatively robust biological basis and are worthy of validation in subsequent clinical studies. In contrast, other candidate drugs, as exploratory discoveries, expand the scope of treatment strategies for ccRCC.

This study deeply analyzed the gene expression, enrichment pathways, and drug resistance of neutrophils in ccRCC through bioinformatics methods. However, some limitations in this study should be acknowledged. The validation of feature genes mainly relied on bioinformatics analysis and public databases. Although multi-level cross-validation was performed, there were no independent clinical samples to quantify messenger RNA (mRNA) levels using quantitative polymerase chain reaction (qPCR). The protein expression levels were examined using IHC data from HPA. Quantitative validation at the protein level was not performed using techniques such as Western blotting. Furthermore, due to the lack of key clinical information in public databases, this study could not directly compare the constructed prognostic model with established IMDC or Memorial Sloan Kettering Cancer Center (MSKCC) risk stratification systems in ccRCC. We acknowledge that these experiments would further enhance the reliability of the research conclusions. However, due to current research conditions and resources, the aforementioned wet experiments cannot yet be carried out. In the future, relevant experiments should be conducted to validate our findings. Furthermore, the data in this article were mainly collected from TCGA. Although TCGA collects samples from hundreds of ccRCC patients, the sample dataset of hundreds of ccRCC patients is relatively small and may not truly reflect the genetic and transcriptomic characteristics of ccRCC. Therefore, the results of this study may have certain deviations and limitations. Additionally, external validation was conducted in a relatively small cohort (GSE167573, n=55), which may limit the generalizability of the model. Future validation in larger multi-center cohorts is needed. Furthermore, the normal samples used in this analysis were adjacent tumor tissues, which may differ from truly healthy kidney tissue, posing a potential limitation to this study. As a complex tumor type, ccRCC is heterogeneous in its immune microenvironment characteristics.

In this study, it was discovered for the first time that 10 neutrophil-related genes contributed a lot to ccRCC. This finding is expected to assist clinicians in developing more specific prognostic assessments and personalized immunotherapy strategies for ccRCC patients, thereby improving their quality of life and clinical outcomes.

Conclusions

This study successfully constructed and validated a novel prognostic model for ccRCC based on 10 neutrophil-related genes. The model demonstrated stable performance for risk stratification across multiple datasets, and its risk score was confirmed as a strong prognostic indicator independent of TNM staging. Functional analysis showed that these 10 feature genes were significantly enriched in chemokine and cytokine signaling pathways, validating the biological basis of this prognostic label. Further analysis revealed high infiltration of naive B cells and activated CD4 memory T cells in the tumor microenvironment of high-risk patients, along with upregulated expression of immune checkpoint molecules. These results suggested that these patients may have unique immunotherapy response characteristics. Drug sensitivity analysis showed greater sensitivity to the anti-angiogenic drug axitinib, providing potential clues for personalized treatment in high-risk patients.

However, this study also has certain limitations, mainly due to its retrospective design and the lack of prospective clinical cohort and experimental validation. Future multi-center, large-sample prospective studies are necessary to validate the model’s generalizability and to further elucidate the functions of the hub genes, ultimately providing new diagnostic biomarkers and theoretical basis for personalized treatment of ccRCC.

Supplementary

The article’s supplementary files as

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, which utilized publicly available data, was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.